]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/btrfs/inode.c
806708dd7e3847c10207403505180367c8feb472
[linux-2.6-omap-h63xx.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/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.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 "compat.h"
51 #include "tree-log.h"
52 #include "ref-cache.h"
53 #include "compression.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89
90 /*
91  * a very lame attempt at stopping writes when the FS is 85% full.  There
92  * are countless ways this is incorrect, but it is better than nothing.
93  */
94 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
95                            int for_del)
96 {
97         u64 total;
98         u64 used;
99         u64 thresh;
100         unsigned long flags;
101         int ret = 0;
102
103         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
104         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
105         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
106         if (for_del)
107                 thresh = total * 90;
108         else
109                 thresh = total * 85;
110
111         do_div(thresh, 100);
112
113         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
114                 ret = -ENOSPC;
115         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
116         return ret;
117 }
118
119 /*
120  * this does all the hard work for inserting an inline extent into
121  * the btree.  The caller should have done a btrfs_drop_extents so that
122  * no overlapping inline items exist in the btree
123  */
124 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
125                                 struct btrfs_root *root, struct inode *inode,
126                                 u64 start, size_t size, size_t compressed_size,
127                                 struct page **compressed_pages)
128 {
129         struct btrfs_key key;
130         struct btrfs_path *path;
131         struct extent_buffer *leaf;
132         struct page *page = NULL;
133         char *kaddr;
134         unsigned long ptr;
135         struct btrfs_file_extent_item *ei;
136         int err = 0;
137         int ret;
138         size_t cur_size = size;
139         size_t datasize;
140         unsigned long offset;
141         int use_compress = 0;
142
143         if (compressed_size && compressed_pages) {
144                 use_compress = 1;
145                 cur_size = compressed_size;
146         }
147
148         path = btrfs_alloc_path(); if (!path)
149                 return -ENOMEM;
150
151         btrfs_set_trans_block_group(trans, inode);
152
153         key.objectid = inode->i_ino;
154         key.offset = start;
155         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
156         inode_add_bytes(inode, size);
157         datasize = btrfs_file_extent_calc_inline_size(cur_size);
158
159         inode_add_bytes(inode, size);
160         ret = btrfs_insert_empty_item(trans, root, path, &key,
161                                       datasize);
162         BUG_ON(ret);
163         if (ret) {
164                 err = ret;
165                 printk("got bad ret %d\n", ret);
166                 goto fail;
167         }
168         leaf = path->nodes[0];
169         ei = btrfs_item_ptr(leaf, path->slots[0],
170                             struct btrfs_file_extent_item);
171         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
172         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
173         btrfs_set_file_extent_encryption(leaf, ei, 0);
174         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
175         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
176         ptr = btrfs_file_extent_inline_start(ei);
177
178         if (use_compress) {
179                 struct page *cpage;
180                 int i = 0;
181                 while(compressed_size > 0) {
182                         cpage = compressed_pages[i];
183                         cur_size = min(compressed_size,
184                                        PAGE_CACHE_SIZE);
185
186                         kaddr = kmap(cpage);
187                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
188                         kunmap(cpage);
189
190                         i++;
191                         ptr += cur_size;
192                         compressed_size -= cur_size;
193                 }
194                 btrfs_set_file_extent_compression(leaf, ei,
195                                                   BTRFS_COMPRESS_ZLIB);
196         } else {
197                 page = find_get_page(inode->i_mapping,
198                                      start >> PAGE_CACHE_SHIFT);
199                 btrfs_set_file_extent_compression(leaf, ei, 0);
200                 kaddr = kmap_atomic(page, KM_USER0);
201                 offset = start & (PAGE_CACHE_SIZE - 1);
202                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
203                 kunmap_atomic(kaddr, KM_USER0);
204                 page_cache_release(page);
205         }
206         btrfs_mark_buffer_dirty(leaf);
207         btrfs_free_path(path);
208
209         BTRFS_I(inode)->disk_i_size = inode->i_size;
210         btrfs_update_inode(trans, root, inode);
211         return 0;
212 fail:
213         btrfs_free_path(path);
214         return err;
215 }
216
217
218 /*
219  * conditionally insert an inline extent into the file.  This
220  * does the checks required to make sure the data is small enough
221  * to fit as an inline extent.
222  */
223 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
224                                  struct btrfs_root *root,
225                                  struct inode *inode, u64 start, u64 end,
226                                  size_t compressed_size,
227                                  struct page **compressed_pages)
228 {
229         u64 isize = i_size_read(inode);
230         u64 actual_end = min(end + 1, isize);
231         u64 inline_len = actual_end - start;
232         u64 aligned_end = (end + root->sectorsize - 1) &
233                         ~((u64)root->sectorsize - 1);
234         u64 hint_byte;
235         u64 data_len = inline_len;
236         int ret;
237
238         if (compressed_size)
239                 data_len = compressed_size;
240
241         if (start > 0 ||
242             actual_end >= PAGE_CACHE_SIZE ||
243             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244             (!compressed_size &&
245             (actual_end & (root->sectorsize - 1)) == 0) ||
246             end + 1 < isize ||
247             data_len > root->fs_info->max_inline) {
248                 return 1;
249         }
250
251         ret = btrfs_drop_extents(trans, root, inode, start,
252                                  aligned_end, start, &hint_byte);
253         BUG_ON(ret);
254
255         if (isize > actual_end)
256                 inline_len = min_t(u64, isize, actual_end);
257         ret = insert_inline_extent(trans, root, inode, start,
258                                    inline_len, compressed_size,
259                                    compressed_pages);
260         BUG_ON(ret);
261         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
262         return 0;
263 }
264
265 /*
266  * when extent_io.c finds a delayed allocation range in the file,
267  * the call backs end up in this code.  The basic idea is to
268  * allocate extents on disk for the range, and create ordered data structs
269  * in ram to track those extents.
270  *
271  * locked_page is the page that writepage had locked already.  We use
272  * it to make sure we don't do extra locks or unlocks.
273  *
274  * *page_started is set to one if we unlock locked_page and do everything
275  * required to start IO on it.  It may be clean and already done with
276  * IO when we return.
277  */
278 static int cow_file_range(struct inode *inode, struct page *locked_page,
279                           u64 start, u64 end, int *page_started)
280 {
281         struct btrfs_root *root = BTRFS_I(inode)->root;
282         struct btrfs_trans_handle *trans;
283         u64 alloc_hint = 0;
284         u64 num_bytes;
285         unsigned long ram_size;
286         u64 orig_start;
287         u64 disk_num_bytes;
288         u64 cur_alloc_size;
289         u64 blocksize = root->sectorsize;
290         u64 actual_end;
291         struct btrfs_key ins;
292         struct extent_map *em;
293         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
294         int ret = 0;
295         struct page **pages = NULL;
296         unsigned long nr_pages;
297         unsigned long nr_pages_ret = 0;
298         unsigned long total_compressed = 0;
299         unsigned long total_in = 0;
300         unsigned long max_compressed = 128 * 1024;
301         unsigned long max_uncompressed = 256 * 1024;
302         int i;
303         int ordered_type;
304         int will_compress;
305
306         trans = btrfs_join_transaction(root, 1);
307         BUG_ON(!trans);
308         btrfs_set_trans_block_group(trans, inode);
309         orig_start = start;
310
311         /*
312          * compression made this loop a bit ugly, but the basic idea is to
313          * compress some pages but keep the total size of the compressed
314          * extent relatively small.  If compression is off, this goto target
315          * is never used.
316          */
317 again:
318         will_compress = 0;
319         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
320         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
321
322         actual_end = min_t(u64, i_size_read(inode), end + 1);
323         total_compressed = actual_end - start;
324
325         /* we want to make sure that amount of ram required to uncompress
326          * an extent is reasonable, so we limit the total size in ram
327          * of a compressed extent to 256k
328          */
329         total_compressed = min(total_compressed, max_uncompressed);
330         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
331         num_bytes = max(blocksize,  num_bytes);
332         disk_num_bytes = num_bytes;
333         total_in = 0;
334         ret = 0;
335
336         /* we do compression for mount -o compress and when the
337          * inode has not been flagged as nocompress
338          */
339         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
340             btrfs_test_opt(root, COMPRESS)) {
341                 WARN_ON(pages);
342                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
343
344                 /* we want to make sure the amount of IO required to satisfy
345                  * a random read is reasonably small, so we limit the size
346                  * of a compressed extent to 128k
347                  */
348                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
349                                                 total_compressed, pages,
350                                                 nr_pages, &nr_pages_ret,
351                                                 &total_in,
352                                                 &total_compressed,
353                                                 max_compressed);
354
355                 if (!ret) {
356                         unsigned long offset = total_compressed &
357                                 (PAGE_CACHE_SIZE - 1);
358                         struct page *page = pages[nr_pages_ret - 1];
359                         char *kaddr;
360
361                         /* zero the tail end of the last page, we might be
362                          * sending it down to disk
363                          */
364                         if (offset) {
365                                 kaddr = kmap_atomic(page, KM_USER0);
366                                 memset(kaddr + offset, 0,
367                                        PAGE_CACHE_SIZE - offset);
368                                 kunmap_atomic(kaddr, KM_USER0);
369                         }
370                         will_compress = 1;
371                 }
372         }
373         if (start == 0) {
374                 /* lets try to make an inline extent */
375                 if (ret || total_in < (end - start + 1)) {
376                         /* we didn't compress the entire range, try
377                          * to make an uncompressed inline extent.  This
378                          * is almost sure to fail, but maybe inline sizes
379                          * will get bigger later
380                          */
381                         ret = cow_file_range_inline(trans, root, inode,
382                                                     start, end, 0, NULL);
383                 } else {
384                         ret = cow_file_range_inline(trans, root, inode,
385                                                     start, end,
386                                                     total_compressed, pages);
387                 }
388                 if (ret == 0) {
389                         extent_clear_unlock_delalloc(inode,
390                                                      &BTRFS_I(inode)->io_tree,
391                                                      start, end, NULL,
392                                                      1, 1, 1);
393                         *page_started = 1;
394                         ret = 0;
395                         goto free_pages_out;
396                 }
397         }
398
399         if (will_compress) {
400                 /*
401                  * we aren't doing an inline extent round the compressed size
402                  * up to a block size boundary so the allocator does sane
403                  * things
404                  */
405                 total_compressed = (total_compressed + blocksize - 1) &
406                         ~(blocksize - 1);
407
408                 /*
409                  * one last check to make sure the compression is really a
410                  * win, compare the page count read with the blocks on disk
411                  */
412                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
413                         ~(PAGE_CACHE_SIZE - 1);
414                 if (total_compressed >= total_in) {
415                         will_compress = 0;
416                 } else {
417                         disk_num_bytes = total_compressed;
418                         num_bytes = total_in;
419                 }
420         }
421         if (!will_compress && pages) {
422                 /*
423                  * the compression code ran but failed to make things smaller,
424                  * free any pages it allocated and our page pointer array
425                  */
426                 for (i = 0; i < nr_pages_ret; i++) {
427                         WARN_ON(pages[i]->mapping);
428                         page_cache_release(pages[i]);
429                 }
430                 kfree(pages);
431                 pages = NULL;
432                 total_compressed = 0;
433                 nr_pages_ret = 0;
434
435                 /* flag the file so we don't compress in the future */
436                 btrfs_set_flag(inode, NOCOMPRESS);
437         }
438
439         BUG_ON(disk_num_bytes >
440                btrfs_super_total_bytes(&root->fs_info->super_copy));
441
442         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
443
444         while(disk_num_bytes > 0) {
445                 unsigned long min_bytes;
446
447                 /*
448                  * the max size of a compressed extent is pretty small,
449                  * make the code a little less complex by forcing
450                  * the allocator to find a whole compressed extent at once
451                  */
452                 if (will_compress)
453                         min_bytes = disk_num_bytes;
454                 else
455                         min_bytes = root->sectorsize;
456
457                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
458                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
459                                            min_bytes, 0, alloc_hint,
460                                            (u64)-1, &ins, 1);
461                 if (ret) {
462                         WARN_ON(1);
463                         goto free_pages_out_fail;
464                 }
465                 em = alloc_extent_map(GFP_NOFS);
466                 em->start = start;
467
468                 if (will_compress) {
469                         ram_size = num_bytes;
470                         em->len = num_bytes;
471                 } else {
472                         /* ramsize == disk size */
473                         ram_size = ins.offset;
474                         em->len = ins.offset;
475                 }
476
477                 em->block_start = ins.objectid;
478                 em->block_len = ins.offset;
479                 em->bdev = root->fs_info->fs_devices->latest_bdev;
480                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
481
482                 if (will_compress)
483                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
484
485                 while(1) {
486                         spin_lock(&em_tree->lock);
487                         ret = add_extent_mapping(em_tree, em);
488                         spin_unlock(&em_tree->lock);
489                         if (ret != -EEXIST) {
490                                 free_extent_map(em);
491                                 break;
492                         }
493                         btrfs_drop_extent_cache(inode, start,
494                                                 start + ram_size - 1, 0);
495                 }
496
497                 cur_alloc_size = ins.offset;
498                 ordered_type = will_compress ? BTRFS_ORDERED_COMPRESSED : 0;
499                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
500                                                ram_size, cur_alloc_size,
501                                                ordered_type);
502                 BUG_ON(ret);
503
504                 if (disk_num_bytes < cur_alloc_size) {
505                         printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
506                                cur_alloc_size);
507                         break;
508                 }
509
510                 if (will_compress) {
511                         /*
512                          * we're doing compression, we and we need to
513                          * submit the compressed extents down to the device.
514                          *
515                          * We lock down all the file pages, clearing their
516                          * dirty bits and setting them writeback.  Everyone
517                          * that wants to modify the page will wait on the
518                          * ordered extent above.
519                          *
520                          * The writeback bits on the file pages are
521                          * cleared when the compressed pages are on disk
522                          */
523                         btrfs_end_transaction(trans, root);
524
525                         if (start <= page_offset(locked_page) &&
526                             page_offset(locked_page) < start + ram_size) {
527                                 *page_started = 1;
528                         }
529
530                         extent_clear_unlock_delalloc(inode,
531                                                      &BTRFS_I(inode)->io_tree,
532                                                      start,
533                                                      start + ram_size - 1,
534                                                      NULL, 1, 1, 0);
535
536                         ret = btrfs_submit_compressed_write(inode, start,
537                                                  ram_size, ins.objectid,
538                                                  cur_alloc_size, pages,
539                                                  nr_pages_ret);
540
541                         BUG_ON(ret);
542                         trans = btrfs_join_transaction(root, 1);
543                         if (start + ram_size < end) {
544                                 start += ram_size;
545                                 alloc_hint = ins.objectid + ins.offset;
546                                 /* pages will be freed at end_bio time */
547                                 pages = NULL;
548                                 goto again;
549                         } else {
550                                 /* we've written everything, time to go */
551                                 break;
552                         }
553                 }
554                 /* we're not doing compressed IO, don't unlock the first
555                  * page (which the caller expects to stay locked), don't
556                  * clear any dirty bits and don't set any writeback bits
557                  */
558                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
559                                              start, start + ram_size - 1,
560                                              locked_page, 0, 0, 0);
561                 disk_num_bytes -= cur_alloc_size;
562                 num_bytes -= cur_alloc_size;
563                 alloc_hint = ins.objectid + ins.offset;
564                 start += cur_alloc_size;
565         }
566
567         ret = 0;
568 out:
569         btrfs_end_transaction(trans, root);
570
571         return ret;
572
573 free_pages_out_fail:
574         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
575                                      start, end, locked_page, 0, 0, 0);
576 free_pages_out:
577         for (i = 0; i < nr_pages_ret; i++) {
578                 WARN_ON(pages[i]->mapping);
579                 page_cache_release(pages[i]);
580         }
581         if (pages)
582                 kfree(pages);
583
584         goto out;
585 }
586
587 /*
588  * when nowcow writeback call back.  This checks for snapshots or COW copies
589  * of the extents that exist in the file, and COWs the file as required.
590  *
591  * If no cow copies or snapshots exist, we write directly to the existing
592  * blocks on disk
593  */
594 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
595                               u64 start, u64 end, int *page_started, int force)
596 {
597         struct btrfs_root *root = BTRFS_I(inode)->root;
598         struct btrfs_trans_handle *trans;
599         struct extent_buffer *leaf;
600         struct btrfs_path *path;
601         struct btrfs_file_extent_item *fi;
602         struct btrfs_key found_key;
603         u64 cow_start;
604         u64 cur_offset;
605         u64 extent_end;
606         u64 disk_bytenr;
607         u64 num_bytes;
608         int extent_type;
609         int ret;
610         int type;
611         int nocow;
612         int check_prev = 1;
613
614         path = btrfs_alloc_path();
615         BUG_ON(!path);
616         trans = btrfs_join_transaction(root, 1);
617         BUG_ON(!trans);
618
619         cow_start = (u64)-1;
620         cur_offset = start;
621         while (1) {
622                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
623                                                cur_offset, 0);
624                 BUG_ON(ret < 0);
625                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
626                         leaf = path->nodes[0];
627                         btrfs_item_key_to_cpu(leaf, &found_key,
628                                               path->slots[0] - 1);
629                         if (found_key.objectid == inode->i_ino &&
630                             found_key.type == BTRFS_EXTENT_DATA_KEY)
631                                 path->slots[0]--;
632                 }
633                 check_prev = 0;
634 next_slot:
635                 leaf = path->nodes[0];
636                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
637                         ret = btrfs_next_leaf(root, path);
638                         if (ret < 0)
639                                 BUG_ON(1);
640                         if (ret > 0)
641                                 break;
642                         leaf = path->nodes[0];
643                 }
644
645                 nocow = 0;
646                 disk_bytenr = 0;
647                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
648
649                 if (found_key.objectid > inode->i_ino ||
650                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
651                     found_key.offset > end)
652                         break;
653
654                 if (found_key.offset > cur_offset) {
655                         extent_end = found_key.offset;
656                         goto out_check;
657                 }
658
659                 fi = btrfs_item_ptr(leaf, path->slots[0],
660                                     struct btrfs_file_extent_item);
661                 extent_type = btrfs_file_extent_type(leaf, fi);
662
663                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
664                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
665                         struct btrfs_block_group_cache *block_group;
666                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
667                         extent_end = found_key.offset +
668                                 btrfs_file_extent_num_bytes(leaf, fi);
669                         if (extent_end <= start) {
670                                 path->slots[0]++;
671                                 goto next_slot;
672                         }
673                         if (btrfs_file_extent_compression(leaf, fi) ||
674                             btrfs_file_extent_encryption(leaf, fi) ||
675                             btrfs_file_extent_other_encoding(leaf, fi))
676                                 goto out_check;
677                         if (disk_bytenr == 0)
678                                 goto out_check;
679                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
680                                 goto out_check;
681                         if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
682                                 goto out_check;
683                         block_group = btrfs_lookup_block_group(root->fs_info,
684                                                                disk_bytenr);
685                         if (!block_group || block_group->ro)
686                                 goto out_check;
687                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
688                         nocow = 1;
689                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
690                         extent_end = found_key.offset +
691                                 btrfs_file_extent_inline_len(leaf, fi);
692                         extent_end = ALIGN(extent_end, root->sectorsize);
693                 } else {
694                         BUG_ON(1);
695                 }
696 out_check:
697                 if (extent_end <= start) {
698                         path->slots[0]++;
699                         goto next_slot;
700                 }
701                 if (!nocow) {
702                         if (cow_start == (u64)-1)
703                                 cow_start = cur_offset;
704                         cur_offset = extent_end;
705                         if (cur_offset > end)
706                                 break;
707                         path->slots[0]++;
708                         goto next_slot;
709                 }
710
711                 btrfs_release_path(root, path);
712                 if (cow_start != (u64)-1) {
713                         ret = cow_file_range(inode, locked_page, cow_start,
714                                         found_key.offset - 1, page_started);
715                         BUG_ON(ret);
716                         cow_start = (u64)-1;
717                 }
718
719                 disk_bytenr += cur_offset - found_key.offset;
720                 num_bytes = min(end + 1, extent_end) - cur_offset;
721                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
722                         struct extent_map *em;
723                         struct extent_map_tree *em_tree;
724                         em_tree = &BTRFS_I(inode)->extent_tree;
725                         em = alloc_extent_map(GFP_NOFS);
726                         em->start = cur_offset;
727                         em->len = num_bytes;
728                         em->block_len = num_bytes;
729                         em->block_start = disk_bytenr;
730                         em->bdev = root->fs_info->fs_devices->latest_bdev;
731                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
732                         while (1) {
733                                 spin_lock(&em_tree->lock);
734                                 ret = add_extent_mapping(em_tree, em);
735                                 spin_unlock(&em_tree->lock);
736                                 if (ret != -EEXIST) {
737                                         free_extent_map(em);
738                                         break;
739                                 }
740                                 btrfs_drop_extent_cache(inode, em->start,
741                                                 em->start + em->len - 1, 0);
742                         }
743                         type = BTRFS_ORDERED_PREALLOC;
744                 } else {
745                         type = BTRFS_ORDERED_NOCOW;
746                 }
747
748                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
749                                                num_bytes, num_bytes, type);
750                 BUG_ON(ret);
751                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
752                                         cur_offset, cur_offset + num_bytes - 1,
753                                         locked_page, 0, 0, 0);
754                 cur_offset = extent_end;
755                 if (cur_offset > end)
756                         break;
757         }
758         btrfs_release_path(root, path);
759
760         if (cur_offset <= end && cow_start == (u64)-1)
761                 cow_start = cur_offset;
762         if (cow_start != (u64)-1) {
763                 ret = cow_file_range(inode, locked_page, cow_start, end,
764                                      page_started);
765                 BUG_ON(ret);
766         }
767
768         ret = btrfs_end_transaction(trans, root);
769         BUG_ON(ret);
770         btrfs_free_path(path);
771         return 0;
772 }
773
774 /*
775  * extent_io.c call back to do delayed allocation processing
776  */
777 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
778                               u64 start, u64 end, int *page_started)
779 {
780         struct btrfs_root *root = BTRFS_I(inode)->root;
781         int ret;
782
783         if (btrfs_test_opt(root, NODATACOW) ||
784             btrfs_test_flag(inode, NODATACOW))
785                 ret = run_delalloc_nocow(inode, locked_page, start, end,
786                                          page_started, 0);
787         else if (btrfs_test_flag(inode, PREALLOC))
788                 ret = run_delalloc_nocow(inode, locked_page, start, end,
789                                          page_started, 1);
790         else
791                 ret = cow_file_range(inode, locked_page, start, end,
792                                      page_started);
793
794         return ret;
795 }
796
797 /*
798  * extent_io.c set_bit_hook, used to track delayed allocation
799  * bytes in this file, and to maintain the list of inodes that
800  * have pending delalloc work to be done.
801  */
802 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
803                        unsigned long old, unsigned long bits)
804 {
805         unsigned long flags;
806         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
807                 struct btrfs_root *root = BTRFS_I(inode)->root;
808                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
809                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
810                 root->fs_info->delalloc_bytes += end - start + 1;
811                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
812                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
813                                       &root->fs_info->delalloc_inodes);
814                 }
815                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
816         }
817         return 0;
818 }
819
820 /*
821  * extent_io.c clear_bit_hook, see set_bit_hook for why
822  */
823 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
824                          unsigned long old, unsigned long bits)
825 {
826         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
827                 struct btrfs_root *root = BTRFS_I(inode)->root;
828                 unsigned long flags;
829
830                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
831                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
832                         printk("warning: delalloc account %Lu %Lu\n",
833                                end - start + 1, root->fs_info->delalloc_bytes);
834                         root->fs_info->delalloc_bytes = 0;
835                         BTRFS_I(inode)->delalloc_bytes = 0;
836                 } else {
837                         root->fs_info->delalloc_bytes -= end - start + 1;
838                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
839                 }
840                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
841                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
842                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
843                 }
844                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
845         }
846         return 0;
847 }
848
849 /*
850  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
851  * we don't create bios that span stripes or chunks
852  */
853 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
854                          size_t size, struct bio *bio,
855                          unsigned long bio_flags)
856 {
857         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
858         struct btrfs_mapping_tree *map_tree;
859         u64 logical = (u64)bio->bi_sector << 9;
860         u64 length = 0;
861         u64 map_length;
862         int ret;
863
864         length = bio->bi_size;
865         map_tree = &root->fs_info->mapping_tree;
866         map_length = length;
867         ret = btrfs_map_block(map_tree, READ, logical,
868                               &map_length, NULL, 0);
869
870         if (map_length < length + size) {
871                 return 1;
872         }
873         return 0;
874 }
875
876 /*
877  * in order to insert checksums into the metadata in large chunks,
878  * we wait until bio submission time.   All the pages in the bio are
879  * checksummed and sums are attached onto the ordered extent record.
880  *
881  * At IO completion time the cums attached on the ordered extent record
882  * are inserted into the btree
883  */
884 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
885                           int mirror_num, unsigned long bio_flags)
886 {
887         struct btrfs_root *root = BTRFS_I(inode)->root;
888         int ret = 0;
889
890         ret = btrfs_csum_one_bio(root, inode, bio);
891         BUG_ON(ret);
892
893         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
894 }
895
896 /*
897  * extent_io.c submission hook. This does the right thing for csum calculation on write,
898  * or reading the csums from the tree before a read
899  */
900 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
901                           int mirror_num, unsigned long bio_flags)
902 {
903         struct btrfs_root *root = BTRFS_I(inode)->root;
904         int ret = 0;
905         int skip_sum;
906
907         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
908         BUG_ON(ret);
909
910         skip_sum = btrfs_test_opt(root, NODATASUM) ||
911                 btrfs_test_flag(inode, NODATASUM);
912
913         if (!(rw & (1 << BIO_RW))) {
914                 if (!skip_sum)
915                         btrfs_lookup_bio_sums(root, inode, bio);
916
917                 if (bio_flags & EXTENT_BIO_COMPRESSED)
918                         return btrfs_submit_compressed_read(inode, bio,
919                                                     mirror_num, bio_flags);
920                 goto mapit;
921         } else if (!skip_sum) {
922                 /* we're doing a write, do the async checksumming */
923                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
924                                    inode, rw, bio, mirror_num,
925                                    bio_flags, __btrfs_submit_bio_hook);
926         }
927
928 mapit:
929         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
930 }
931
932 /*
933  * given a list of ordered sums record them in the inode.  This happens
934  * at IO completion time based on sums calculated at bio submission time.
935  */
936 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
937                              struct inode *inode, u64 file_offset,
938                              struct list_head *list)
939 {
940         struct list_head *cur;
941         struct btrfs_ordered_sum *sum;
942
943         btrfs_set_trans_block_group(trans, inode);
944         list_for_each(cur, list) {
945                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
946                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
947                                        inode, sum);
948         }
949         return 0;
950 }
951
952 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
953 {
954         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
955                                    GFP_NOFS);
956 }
957
958 /* see btrfs_writepage_start_hook for details on why this is required */
959 struct btrfs_writepage_fixup {
960         struct page *page;
961         struct btrfs_work work;
962 };
963
964 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
965 {
966         struct btrfs_writepage_fixup *fixup;
967         struct btrfs_ordered_extent *ordered;
968         struct page *page;
969         struct inode *inode;
970         u64 page_start;
971         u64 page_end;
972
973         fixup = container_of(work, struct btrfs_writepage_fixup, work);
974         page = fixup->page;
975 again:
976         lock_page(page);
977         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
978                 ClearPageChecked(page);
979                 goto out_page;
980         }
981
982         inode = page->mapping->host;
983         page_start = page_offset(page);
984         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
985
986         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
987
988         /* already ordered? We're done */
989         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
990                              EXTENT_ORDERED, 0)) {
991                 goto out;
992         }
993
994         ordered = btrfs_lookup_ordered_extent(inode, page_start);
995         if (ordered) {
996                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
997                               page_end, GFP_NOFS);
998                 unlock_page(page);
999                 btrfs_start_ordered_extent(inode, ordered, 1);
1000                 goto again;
1001         }
1002
1003         btrfs_set_extent_delalloc(inode, page_start, page_end);
1004         ClearPageChecked(page);
1005 out:
1006         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1007 out_page:
1008         unlock_page(page);
1009         page_cache_release(page);
1010 }
1011
1012 /*
1013  * There are a few paths in the higher layers of the kernel that directly
1014  * set the page dirty bit without asking the filesystem if it is a
1015  * good idea.  This causes problems because we want to make sure COW
1016  * properly happens and the data=ordered rules are followed.
1017  *
1018  * In our case any range that doesn't have the ORDERED bit set
1019  * hasn't been properly setup for IO.  We kick off an async process
1020  * to fix it up.  The async helper will wait for ordered extents, set
1021  * the delalloc bit and make it safe to write the page.
1022  */
1023 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1024 {
1025         struct inode *inode = page->mapping->host;
1026         struct btrfs_writepage_fixup *fixup;
1027         struct btrfs_root *root = BTRFS_I(inode)->root;
1028         int ret;
1029
1030         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1031                              EXTENT_ORDERED, 0);
1032         if (ret)
1033                 return 0;
1034
1035         if (PageChecked(page))
1036                 return -EAGAIN;
1037
1038         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1039         if (!fixup)
1040                 return -EAGAIN;
1041
1042         SetPageChecked(page);
1043         page_cache_get(page);
1044         fixup->work.func = btrfs_writepage_fixup_worker;
1045         fixup->page = page;
1046         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1047         return -EAGAIN;
1048 }
1049
1050 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1051                                        struct inode *inode, u64 file_pos,
1052                                        u64 disk_bytenr, u64 disk_num_bytes,
1053                                        u64 num_bytes, u64 ram_bytes,
1054                                        u8 compression, u8 encryption,
1055                                        u16 other_encoding, int extent_type)
1056 {
1057         struct btrfs_root *root = BTRFS_I(inode)->root;
1058         struct btrfs_file_extent_item *fi;
1059         struct btrfs_path *path;
1060         struct extent_buffer *leaf;
1061         struct btrfs_key ins;
1062         u64 hint;
1063         int ret;
1064
1065         path = btrfs_alloc_path();
1066         BUG_ON(!path);
1067
1068         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1069                                  file_pos + num_bytes, file_pos, &hint);
1070         BUG_ON(ret);
1071
1072         ins.objectid = inode->i_ino;
1073         ins.offset = file_pos;
1074         ins.type = BTRFS_EXTENT_DATA_KEY;
1075         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1076         BUG_ON(ret);
1077         leaf = path->nodes[0];
1078         fi = btrfs_item_ptr(leaf, path->slots[0],
1079                             struct btrfs_file_extent_item);
1080         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1081         btrfs_set_file_extent_type(leaf, fi, extent_type);
1082         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1083         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1084         btrfs_set_file_extent_offset(leaf, fi, 0);
1085         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1086         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1087         btrfs_set_file_extent_compression(leaf, fi, compression);
1088         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1089         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1090         btrfs_mark_buffer_dirty(leaf);
1091
1092         inode_add_bytes(inode, num_bytes);
1093         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1094
1095         ins.objectid = disk_bytenr;
1096         ins.offset = disk_num_bytes;
1097         ins.type = BTRFS_EXTENT_ITEM_KEY;
1098         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1099                                           root->root_key.objectid,
1100                                           trans->transid, inode->i_ino, &ins);
1101         BUG_ON(ret);
1102
1103         btrfs_free_path(path);
1104         return 0;
1105 }
1106
1107 /* as ordered data IO finishes, this gets called so we can finish
1108  * an ordered extent if the range of bytes in the file it covers are
1109  * fully written.
1110  */
1111 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1112 {
1113         struct btrfs_root *root = BTRFS_I(inode)->root;
1114         struct btrfs_trans_handle *trans;
1115         struct btrfs_ordered_extent *ordered_extent;
1116         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1117         int compressed = 0;
1118         int ret;
1119
1120         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1121         if (!ret)
1122                 return 0;
1123
1124         trans = btrfs_join_transaction(root, 1);
1125
1126         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1127         BUG_ON(!ordered_extent);
1128         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1129                 goto nocow;
1130
1131         lock_extent(io_tree, ordered_extent->file_offset,
1132                     ordered_extent->file_offset + ordered_extent->len - 1,
1133                     GFP_NOFS);
1134
1135         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1136                 compressed = 1;
1137         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1138                 BUG_ON(compressed);
1139                 ret = btrfs_mark_extent_written(trans, root, inode,
1140                                                 ordered_extent->file_offset,
1141                                                 ordered_extent->file_offset +
1142                                                 ordered_extent->len);
1143                 BUG_ON(ret);
1144         } else {
1145                 ret = insert_reserved_file_extent(trans, inode,
1146                                                 ordered_extent->file_offset,
1147                                                 ordered_extent->start,
1148                                                 ordered_extent->disk_len,
1149                                                 ordered_extent->len,
1150                                                 ordered_extent->len,
1151                                                 compressed, 0, 0,
1152                                                 BTRFS_FILE_EXTENT_REG);
1153                 BUG_ON(ret);
1154         }
1155         unlock_extent(io_tree, ordered_extent->file_offset,
1156                     ordered_extent->file_offset + ordered_extent->len - 1,
1157                     GFP_NOFS);
1158 nocow:
1159         add_pending_csums(trans, inode, ordered_extent->file_offset,
1160                           &ordered_extent->list);
1161
1162         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1163         btrfs_ordered_update_i_size(inode, ordered_extent);
1164         btrfs_update_inode(trans, root, inode);
1165         btrfs_remove_ordered_extent(inode, ordered_extent);
1166         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1167
1168         /* once for us */
1169         btrfs_put_ordered_extent(ordered_extent);
1170         /* once for the tree */
1171         btrfs_put_ordered_extent(ordered_extent);
1172
1173         btrfs_end_transaction(trans, root);
1174         return 0;
1175 }
1176
1177 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1178                                 struct extent_state *state, int uptodate)
1179 {
1180         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1181 }
1182
1183 /*
1184  * When IO fails, either with EIO or csum verification fails, we
1185  * try other mirrors that might have a good copy of the data.  This
1186  * io_failure_record is used to record state as we go through all the
1187  * mirrors.  If another mirror has good data, the page is set up to date
1188  * and things continue.  If a good mirror can't be found, the original
1189  * bio end_io callback is called to indicate things have failed.
1190  */
1191 struct io_failure_record {
1192         struct page *page;
1193         u64 start;
1194         u64 len;
1195         u64 logical;
1196         int last_mirror;
1197 };
1198
1199 int btrfs_io_failed_hook(struct bio *failed_bio,
1200                          struct page *page, u64 start, u64 end,
1201                          struct extent_state *state)
1202 {
1203         struct io_failure_record *failrec = NULL;
1204         u64 private;
1205         struct extent_map *em;
1206         struct inode *inode = page->mapping->host;
1207         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1208         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1209         struct bio *bio;
1210         int num_copies;
1211         int ret;
1212         int rw;
1213         u64 logical;
1214         unsigned long bio_flags = 0;
1215
1216         ret = get_state_private(failure_tree, start, &private);
1217         if (ret) {
1218                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1219                 if (!failrec)
1220                         return -ENOMEM;
1221                 failrec->start = start;
1222                 failrec->len = end - start + 1;
1223                 failrec->last_mirror = 0;
1224
1225                 spin_lock(&em_tree->lock);
1226                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1227                 if (em->start > start || em->start + em->len < start) {
1228                         free_extent_map(em);
1229                         em = NULL;
1230                 }
1231                 spin_unlock(&em_tree->lock);
1232
1233                 if (!em || IS_ERR(em)) {
1234                         kfree(failrec);
1235                         return -EIO;
1236                 }
1237                 logical = start - em->start;
1238                 logical = em->block_start + logical;
1239                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1240                         bio_flags = EXTENT_BIO_COMPRESSED;
1241                 failrec->logical = logical;
1242                 free_extent_map(em);
1243                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1244                                 EXTENT_DIRTY, GFP_NOFS);
1245                 set_state_private(failure_tree, start,
1246                                  (u64)(unsigned long)failrec);
1247         } else {
1248                 failrec = (struct io_failure_record *)(unsigned long)private;
1249         }
1250         num_copies = btrfs_num_copies(
1251                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1252                               failrec->logical, failrec->len);
1253         failrec->last_mirror++;
1254         if (!state) {
1255                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1256                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1257                                                     failrec->start,
1258                                                     EXTENT_LOCKED);
1259                 if (state && state->start != failrec->start)
1260                         state = NULL;
1261                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1262         }
1263         if (!state || failrec->last_mirror > num_copies) {
1264                 set_state_private(failure_tree, failrec->start, 0);
1265                 clear_extent_bits(failure_tree, failrec->start,
1266                                   failrec->start + failrec->len - 1,
1267                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1268                 kfree(failrec);
1269                 return -EIO;
1270         }
1271         bio = bio_alloc(GFP_NOFS, 1);
1272         bio->bi_private = state;
1273         bio->bi_end_io = failed_bio->bi_end_io;
1274         bio->bi_sector = failrec->logical >> 9;
1275         bio->bi_bdev = failed_bio->bi_bdev;
1276         bio->bi_size = 0;
1277         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1278         if (failed_bio->bi_rw & (1 << BIO_RW))
1279                 rw = WRITE;
1280         else
1281                 rw = READ;
1282
1283         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1284                                                       failrec->last_mirror,
1285                                                       bio_flags);
1286         return 0;
1287 }
1288
1289 /*
1290  * each time an IO finishes, we do a fast check in the IO failure tree
1291  * to see if we need to process or clean up an io_failure_record
1292  */
1293 int btrfs_clean_io_failures(struct inode *inode, u64 start)
1294 {
1295         u64 private;
1296         u64 private_failure;
1297         struct io_failure_record *failure;
1298         int ret;
1299
1300         private = 0;
1301         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1302                              (u64)-1, 1, EXTENT_DIRTY)) {
1303                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1304                                         start, &private_failure);
1305                 if (ret == 0) {
1306                         failure = (struct io_failure_record *)(unsigned long)
1307                                    private_failure;
1308                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1309                                           failure->start, 0);
1310                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1311                                           failure->start,
1312                                           failure->start + failure->len - 1,
1313                                           EXTENT_DIRTY | EXTENT_LOCKED,
1314                                           GFP_NOFS);
1315                         kfree(failure);
1316                 }
1317         }
1318         return 0;
1319 }
1320
1321 /*
1322  * when reads are done, we need to check csums to verify the data is correct
1323  * if there's a match, we allow the bio to finish.  If not, we go through
1324  * the io_failure_record routines to find good copies
1325  */
1326 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1327                                struct extent_state *state)
1328 {
1329         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1330         struct inode *inode = page->mapping->host;
1331         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1332         char *kaddr;
1333         u64 private = ~(u32)0;
1334         int ret;
1335         struct btrfs_root *root = BTRFS_I(inode)->root;
1336         u32 csum = ~(u32)0;
1337         unsigned long flags;
1338
1339         if (btrfs_test_opt(root, NODATASUM) ||
1340             btrfs_test_flag(inode, NODATASUM))
1341                 return 0;
1342         if (state && state->start == start) {
1343                 private = state->private;
1344                 ret = 0;
1345         } else {
1346                 ret = get_state_private(io_tree, start, &private);
1347         }
1348         local_irq_save(flags);
1349         kaddr = kmap_atomic(page, KM_IRQ0);
1350         if (ret) {
1351                 goto zeroit;
1352         }
1353         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1354         btrfs_csum_final(csum, (char *)&csum);
1355         if (csum != private) {
1356                 goto zeroit;
1357         }
1358         kunmap_atomic(kaddr, KM_IRQ0);
1359         local_irq_restore(flags);
1360
1361         /* if the io failure tree for this inode is non-empty,
1362          * check to see if we've recovered from a failed IO
1363          */
1364         btrfs_clean_io_failures(inode, start);
1365         return 0;
1366
1367 zeroit:
1368         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1369                page->mapping->host->i_ino, (unsigned long long)start, csum,
1370                private);
1371         memset(kaddr + offset, 1, end - start + 1);
1372         flush_dcache_page(page);
1373         kunmap_atomic(kaddr, KM_IRQ0);
1374         local_irq_restore(flags);
1375         if (private == 0)
1376                 return 0;
1377         return -EIO;
1378 }
1379
1380 /*
1381  * This creates an orphan entry for the given inode in case something goes
1382  * wrong in the middle of an unlink/truncate.
1383  */
1384 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1385 {
1386         struct btrfs_root *root = BTRFS_I(inode)->root;
1387         int ret = 0;
1388
1389         spin_lock(&root->list_lock);
1390
1391         /* already on the orphan list, we're good */
1392         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1393                 spin_unlock(&root->list_lock);
1394                 return 0;
1395         }
1396
1397         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1398
1399         spin_unlock(&root->list_lock);
1400
1401         /*
1402          * insert an orphan item to track this unlinked/truncated file
1403          */
1404         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1405
1406         return ret;
1407 }
1408
1409 /*
1410  * We have done the truncate/delete so we can go ahead and remove the orphan
1411  * item for this particular inode.
1412  */
1413 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1414 {
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret = 0;
1417
1418         spin_lock(&root->list_lock);
1419
1420         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1421                 spin_unlock(&root->list_lock);
1422                 return 0;
1423         }
1424
1425         list_del_init(&BTRFS_I(inode)->i_orphan);
1426         if (!trans) {
1427                 spin_unlock(&root->list_lock);
1428                 return 0;
1429         }
1430
1431         spin_unlock(&root->list_lock);
1432
1433         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1434
1435         return ret;
1436 }
1437
1438 /*
1439  * this cleans up any orphans that may be left on the list from the last use
1440  * of this root.
1441  */
1442 void btrfs_orphan_cleanup(struct btrfs_root *root)
1443 {
1444         struct btrfs_path *path;
1445         struct extent_buffer *leaf;
1446         struct btrfs_item *item;
1447         struct btrfs_key key, found_key;
1448         struct btrfs_trans_handle *trans;
1449         struct inode *inode;
1450         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1451
1452         /* don't do orphan cleanup if the fs is readonly. */
1453         if (root->fs_info->sb->s_flags & MS_RDONLY)
1454                 return;
1455
1456         path = btrfs_alloc_path();
1457         if (!path)
1458                 return;
1459         path->reada = -1;
1460
1461         key.objectid = BTRFS_ORPHAN_OBJECTID;
1462         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1463         key.offset = (u64)-1;
1464
1465
1466         while (1) {
1467                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1468                 if (ret < 0) {
1469                         printk(KERN_ERR "Error searching slot for orphan: %d"
1470                                "\n", ret);
1471                         break;
1472                 }
1473
1474                 /*
1475                  * if ret == 0 means we found what we were searching for, which
1476                  * is weird, but possible, so only screw with path if we didnt
1477                  * find the key and see if we have stuff that matches
1478                  */
1479                 if (ret > 0) {
1480                         if (path->slots[0] == 0)
1481                                 break;
1482                         path->slots[0]--;
1483                 }
1484
1485                 /* pull out the item */
1486                 leaf = path->nodes[0];
1487                 item = btrfs_item_nr(leaf, path->slots[0]);
1488                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1489
1490                 /* make sure the item matches what we want */
1491                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1492                         break;
1493                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1494                         break;
1495
1496                 /* release the path since we're done with it */
1497                 btrfs_release_path(root, path);
1498
1499                 /*
1500                  * this is where we are basically btrfs_lookup, without the
1501                  * crossing root thing.  we store the inode number in the
1502                  * offset of the orphan item.
1503                  */
1504                 inode = btrfs_iget_locked(root->fs_info->sb,
1505                                           found_key.offset, root);
1506                 if (!inode)
1507                         break;
1508
1509                 if (inode->i_state & I_NEW) {
1510                         BTRFS_I(inode)->root = root;
1511
1512                         /* have to set the location manually */
1513                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1514                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1515                         BTRFS_I(inode)->location.offset = 0;
1516
1517                         btrfs_read_locked_inode(inode);
1518                         unlock_new_inode(inode);
1519                 }
1520
1521                 /*
1522                  * add this inode to the orphan list so btrfs_orphan_del does
1523                  * the proper thing when we hit it
1524                  */
1525                 spin_lock(&root->list_lock);
1526                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1527                 spin_unlock(&root->list_lock);
1528
1529                 /*
1530                  * if this is a bad inode, means we actually succeeded in
1531                  * removing the inode, but not the orphan record, which means
1532                  * we need to manually delete the orphan since iput will just
1533                  * do a destroy_inode
1534                  */
1535                 if (is_bad_inode(inode)) {
1536                         trans = btrfs_start_transaction(root, 1);
1537                         btrfs_orphan_del(trans, inode);
1538                         btrfs_end_transaction(trans, root);
1539                         iput(inode);
1540                         continue;
1541                 }
1542
1543                 /* if we have links, this was a truncate, lets do that */
1544                 if (inode->i_nlink) {
1545                         nr_truncate++;
1546                         btrfs_truncate(inode);
1547                 } else {
1548                         nr_unlink++;
1549                 }
1550
1551                 /* this will do delete_inode and everything for us */
1552                 iput(inode);
1553         }
1554
1555         if (nr_unlink)
1556                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1557         if (nr_truncate)
1558                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1559
1560         btrfs_free_path(path);
1561 }
1562
1563 /*
1564  * read an inode from the btree into the in-memory inode
1565  */
1566 void btrfs_read_locked_inode(struct inode *inode)
1567 {
1568         struct btrfs_path *path;
1569         struct extent_buffer *leaf;
1570         struct btrfs_inode_item *inode_item;
1571         struct btrfs_timespec *tspec;
1572         struct btrfs_root *root = BTRFS_I(inode)->root;
1573         struct btrfs_key location;
1574         u64 alloc_group_block;
1575         u32 rdev;
1576         int ret;
1577
1578         path = btrfs_alloc_path();
1579         BUG_ON(!path);
1580         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1581
1582         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1583         if (ret)
1584                 goto make_bad;
1585
1586         leaf = path->nodes[0];
1587         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1588                                     struct btrfs_inode_item);
1589
1590         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1591         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1592         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1593         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1594         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1595
1596         tspec = btrfs_inode_atime(inode_item);
1597         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1598         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1599
1600         tspec = btrfs_inode_mtime(inode_item);
1601         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1602         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1603
1604         tspec = btrfs_inode_ctime(inode_item);
1605         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1606         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1607
1608         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1609         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1610         inode->i_generation = BTRFS_I(inode)->generation;
1611         inode->i_rdev = 0;
1612         rdev = btrfs_inode_rdev(leaf, inode_item);
1613
1614         BTRFS_I(inode)->index_cnt = (u64)-1;
1615
1616         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1617         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1618                                                        alloc_group_block);
1619         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1620         if (!BTRFS_I(inode)->block_group) {
1621                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1622                                                  NULL, 0,
1623                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1624         }
1625         btrfs_free_path(path);
1626         inode_item = NULL;
1627
1628         switch (inode->i_mode & S_IFMT) {
1629         case S_IFREG:
1630                 inode->i_mapping->a_ops = &btrfs_aops;
1631                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1632                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1633                 inode->i_fop = &btrfs_file_operations;
1634                 inode->i_op = &btrfs_file_inode_operations;
1635                 break;
1636         case S_IFDIR:
1637                 inode->i_fop = &btrfs_dir_file_operations;
1638                 if (root == root->fs_info->tree_root)
1639                         inode->i_op = &btrfs_dir_ro_inode_operations;
1640                 else
1641                         inode->i_op = &btrfs_dir_inode_operations;
1642                 break;
1643         case S_IFLNK:
1644                 inode->i_op = &btrfs_symlink_inode_operations;
1645                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1646                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1647                 break;
1648         default:
1649                 init_special_inode(inode, inode->i_mode, rdev);
1650                 break;
1651         }
1652         return;
1653
1654 make_bad:
1655         btrfs_free_path(path);
1656         make_bad_inode(inode);
1657 }
1658
1659 /*
1660  * given a leaf and an inode, copy the inode fields into the leaf
1661  */
1662 static void fill_inode_item(struct btrfs_trans_handle *trans,
1663                             struct extent_buffer *leaf,
1664                             struct btrfs_inode_item *item,
1665                             struct inode *inode)
1666 {
1667         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1668         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1669         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1670         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1671         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1672
1673         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1674                                inode->i_atime.tv_sec);
1675         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1676                                 inode->i_atime.tv_nsec);
1677
1678         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1679                                inode->i_mtime.tv_sec);
1680         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1681                                 inode->i_mtime.tv_nsec);
1682
1683         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1684                                inode->i_ctime.tv_sec);
1685         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1686                                 inode->i_ctime.tv_nsec);
1687
1688         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
1689         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1690         btrfs_set_inode_transid(leaf, item, trans->transid);
1691         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1692         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1693         btrfs_set_inode_block_group(leaf, item,
1694                                     BTRFS_I(inode)->block_group->key.objectid);
1695 }
1696
1697 /*
1698  * copy everything in the in-memory inode into the btree.
1699  */
1700 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1701                               struct btrfs_root *root,
1702                               struct inode *inode)
1703 {
1704         struct btrfs_inode_item *inode_item;
1705         struct btrfs_path *path;
1706         struct extent_buffer *leaf;
1707         int ret;
1708
1709         path = btrfs_alloc_path();
1710         BUG_ON(!path);
1711         ret = btrfs_lookup_inode(trans, root, path,
1712                                  &BTRFS_I(inode)->location, 1);
1713         if (ret) {
1714                 if (ret > 0)
1715                         ret = -ENOENT;
1716                 goto failed;
1717         }
1718
1719         leaf = path->nodes[0];
1720         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1721                                   struct btrfs_inode_item);
1722
1723         fill_inode_item(trans, leaf, inode_item, inode);
1724         btrfs_mark_buffer_dirty(leaf);
1725         btrfs_set_inode_last_trans(trans, inode);
1726         ret = 0;
1727 failed:
1728         btrfs_free_path(path);
1729         return ret;
1730 }
1731
1732
1733 /*
1734  * unlink helper that gets used here in inode.c and in the tree logging
1735  * recovery code.  It remove a link in a directory with a given name, and
1736  * also drops the back refs in the inode to the directory
1737  */
1738 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1739                        struct btrfs_root *root,
1740                        struct inode *dir, struct inode *inode,
1741                        const char *name, int name_len)
1742 {
1743         struct btrfs_path *path;
1744         int ret = 0;
1745         struct extent_buffer *leaf;
1746         struct btrfs_dir_item *di;
1747         struct btrfs_key key;
1748         u64 index;
1749
1750         path = btrfs_alloc_path();
1751         if (!path) {
1752                 ret = -ENOMEM;
1753                 goto err;
1754         }
1755
1756         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1757                                     name, name_len, -1);
1758         if (IS_ERR(di)) {
1759                 ret = PTR_ERR(di);
1760                 goto err;
1761         }
1762         if (!di) {
1763                 ret = -ENOENT;
1764                 goto err;
1765         }
1766         leaf = path->nodes[0];
1767         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1768         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1769         if (ret)
1770                 goto err;
1771         btrfs_release_path(root, path);
1772
1773         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1774                                   inode->i_ino,
1775                                   dir->i_ino, &index);
1776         if (ret) {
1777                 printk("failed to delete reference to %.*s, "
1778                        "inode %lu parent %lu\n", name_len, name,
1779                        inode->i_ino, dir->i_ino);
1780                 goto err;
1781         }
1782
1783         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1784                                          index, name, name_len, -1);
1785         if (IS_ERR(di)) {
1786                 ret = PTR_ERR(di);
1787                 goto err;
1788         }
1789         if (!di) {
1790                 ret = -ENOENT;
1791                 goto err;
1792         }
1793         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1794         btrfs_release_path(root, path);
1795
1796         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1797                                          inode, dir->i_ino);
1798         BUG_ON(ret != 0 && ret != -ENOENT);
1799         if (ret != -ENOENT)
1800                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
1801
1802         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1803                                            dir, index);
1804         BUG_ON(ret);
1805 err:
1806         btrfs_free_path(path);
1807         if (ret)
1808                 goto out;
1809
1810         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1811         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1812         btrfs_update_inode(trans, root, dir);
1813         btrfs_drop_nlink(inode);
1814         ret = btrfs_update_inode(trans, root, inode);
1815         dir->i_sb->s_dirt = 1;
1816 out:
1817         return ret;
1818 }
1819
1820 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1821 {
1822         struct btrfs_root *root;
1823         struct btrfs_trans_handle *trans;
1824         struct inode *inode = dentry->d_inode;
1825         int ret;
1826         unsigned long nr = 0;
1827
1828         root = BTRFS_I(dir)->root;
1829
1830         ret = btrfs_check_free_space(root, 1, 1);
1831         if (ret)
1832                 goto fail;
1833
1834         trans = btrfs_start_transaction(root, 1);
1835
1836         btrfs_set_trans_block_group(trans, dir);
1837         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1838                                  dentry->d_name.name, dentry->d_name.len);
1839
1840         if (inode->i_nlink == 0)
1841                 ret = btrfs_orphan_add(trans, inode);
1842
1843         nr = trans->blocks_used;
1844
1845         btrfs_end_transaction_throttle(trans, root);
1846 fail:
1847         btrfs_btree_balance_dirty(root, nr);
1848         return ret;
1849 }
1850
1851 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1852 {
1853         struct inode *inode = dentry->d_inode;
1854         int err = 0;
1855         int ret;
1856         struct btrfs_root *root = BTRFS_I(dir)->root;
1857         struct btrfs_trans_handle *trans;
1858         unsigned long nr = 0;
1859
1860         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1861                 return -ENOTEMPTY;
1862         }
1863
1864         ret = btrfs_check_free_space(root, 1, 1);
1865         if (ret)
1866                 goto fail;
1867
1868         trans = btrfs_start_transaction(root, 1);
1869         btrfs_set_trans_block_group(trans, dir);
1870
1871         err = btrfs_orphan_add(trans, inode);
1872         if (err)
1873                 goto fail_trans;
1874
1875         /* now the directory is empty */
1876         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1877                                  dentry->d_name.name, dentry->d_name.len);
1878         if (!err) {
1879                 btrfs_i_size_write(inode, 0);
1880         }
1881
1882 fail_trans:
1883         nr = trans->blocks_used;
1884         ret = btrfs_end_transaction_throttle(trans, root);
1885 fail:
1886         btrfs_btree_balance_dirty(root, nr);
1887
1888         if (ret && !err)
1889                 err = ret;
1890         return err;
1891 }
1892
1893 /*
1894  * when truncating bytes in a file, it is possible to avoid reading
1895  * the leaves that contain only checksum items.  This can be the
1896  * majority of the IO required to delete a large file, but it must
1897  * be done carefully.
1898  *
1899  * The keys in the level just above the leaves are checked to make sure
1900  * the lowest key in a given leaf is a csum key, and starts at an offset
1901  * after the new  size.
1902  *
1903  * Then the key for the next leaf is checked to make sure it also has
1904  * a checksum item for the same file.  If it does, we know our target leaf
1905  * contains only checksum items, and it can be safely freed without reading
1906  * it.
1907  *
1908  * This is just an optimization targeted at large files.  It may do
1909  * nothing.  It will return 0 unless things went badly.
1910  */
1911 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
1912                                      struct btrfs_root *root,
1913                                      struct btrfs_path *path,
1914                                      struct inode *inode, u64 new_size)
1915 {
1916         struct btrfs_key key;
1917         int ret;
1918         int nritems;
1919         struct btrfs_key found_key;
1920         struct btrfs_key other_key;
1921         struct btrfs_leaf_ref *ref;
1922         u64 leaf_gen;
1923         u64 leaf_start;
1924
1925         path->lowest_level = 1;
1926         key.objectid = inode->i_ino;
1927         key.type = BTRFS_CSUM_ITEM_KEY;
1928         key.offset = new_size;
1929 again:
1930         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1931         if (ret < 0)
1932                 goto out;
1933
1934         if (path->nodes[1] == NULL) {
1935                 ret = 0;
1936                 goto out;
1937         }
1938         ret = 0;
1939         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
1940         nritems = btrfs_header_nritems(path->nodes[1]);
1941
1942         if (!nritems)
1943                 goto out;
1944
1945         if (path->slots[1] >= nritems)
1946                 goto next_node;
1947
1948         /* did we find a key greater than anything we want to delete? */
1949         if (found_key.objectid > inode->i_ino ||
1950            (found_key.objectid == inode->i_ino && found_key.type > key.type))
1951                 goto out;
1952
1953         /* we check the next key in the node to make sure the leave contains
1954          * only checksum items.  This comparison doesn't work if our
1955          * leaf is the last one in the node
1956          */
1957         if (path->slots[1] + 1 >= nritems) {
1958 next_node:
1959                 /* search forward from the last key in the node, this
1960                  * will bring us into the next node in the tree
1961                  */
1962                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
1963
1964                 /* unlikely, but we inc below, so check to be safe */
1965                 if (found_key.offset == (u64)-1)
1966                         goto out;
1967
1968                 /* search_forward needs a path with locks held, do the
1969                  * search again for the original key.  It is possible
1970                  * this will race with a balance and return a path that
1971                  * we could modify, but this drop is just an optimization
1972                  * and is allowed to miss some leaves.
1973                  */
1974                 btrfs_release_path(root, path);
1975                 found_key.offset++;
1976
1977                 /* setup a max key for search_forward */
1978                 other_key.offset = (u64)-1;
1979                 other_key.type = key.type;
1980                 other_key.objectid = key.objectid;
1981
1982                 path->keep_locks = 1;
1983                 ret = btrfs_search_forward(root, &found_key, &other_key,
1984                                            path, 0, 0);
1985                 path->keep_locks = 0;
1986                 if (ret || found_key.objectid != key.objectid ||
1987                     found_key.type != key.type) {
1988                         ret = 0;
1989                         goto out;
1990                 }
1991
1992                 key.offset = found_key.offset;
1993                 btrfs_release_path(root, path);
1994                 cond_resched();
1995                 goto again;
1996         }
1997
1998         /* we know there's one more slot after us in the tree,
1999          * read that key so we can verify it is also a checksum item
2000          */
2001         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2002
2003         if (found_key.objectid < inode->i_ino)
2004                 goto next_key;
2005
2006         if (found_key.type != key.type || found_key.offset < new_size)
2007                 goto next_key;
2008
2009         /*
2010          * if the key for the next leaf isn't a csum key from this objectid,
2011          * we can't be sure there aren't good items inside this leaf.
2012          * Bail out
2013          */
2014         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2015                 goto out;
2016
2017         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2018         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2019         /*
2020          * it is safe to delete this leaf, it contains only
2021          * csum items from this inode at an offset >= new_size
2022          */
2023         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2024         BUG_ON(ret);
2025
2026         if (root->ref_cows && leaf_gen < trans->transid) {
2027                 ref = btrfs_alloc_leaf_ref(root, 0);
2028                 if (ref) {
2029                         ref->root_gen = root->root_key.offset;
2030                         ref->bytenr = leaf_start;
2031                         ref->owner = 0;
2032                         ref->generation = leaf_gen;
2033                         ref->nritems = 0;
2034
2035                         ret = btrfs_add_leaf_ref(root, ref, 0);
2036                         WARN_ON(ret);
2037                         btrfs_free_leaf_ref(root, ref);
2038                 } else {
2039                         WARN_ON(1);
2040                 }
2041         }
2042 next_key:
2043         btrfs_release_path(root, path);
2044
2045         if (other_key.objectid == inode->i_ino &&
2046             other_key.type == key.type && other_key.offset > key.offset) {
2047                 key.offset = other_key.offset;
2048                 cond_resched();
2049                 goto again;
2050         }
2051         ret = 0;
2052 out:
2053         /* fixup any changes we've made to the path */
2054         path->lowest_level = 0;
2055         path->keep_locks = 0;
2056         btrfs_release_path(root, path);
2057         return ret;
2058 }
2059
2060 /*
2061  * this can truncate away extent items, csum items and directory items.
2062  * It starts at a high offset and removes keys until it can't find
2063  * any higher than new_size
2064  *
2065  * csum items that cross the new i_size are truncated to the new size
2066  * as well.
2067  *
2068  * min_type is the minimum key type to truncate down to.  If set to 0, this
2069  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2070  */
2071 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2072                                         struct btrfs_root *root,
2073                                         struct inode *inode,
2074                                         u64 new_size, u32 min_type)
2075 {
2076         int ret;
2077         struct btrfs_path *path;
2078         struct btrfs_key key;
2079         struct btrfs_key found_key;
2080         u32 found_type;
2081         struct extent_buffer *leaf;
2082         struct btrfs_file_extent_item *fi;
2083         u64 extent_start = 0;
2084         u64 extent_num_bytes = 0;
2085         u64 item_end = 0;
2086         u64 root_gen = 0;
2087         u64 root_owner = 0;
2088         int found_extent;
2089         int del_item;
2090         int pending_del_nr = 0;
2091         int pending_del_slot = 0;
2092         int extent_type = -1;
2093         u64 mask = root->sectorsize - 1;
2094
2095         if (root->ref_cows)
2096                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2097         path = btrfs_alloc_path();
2098         path->reada = -1;
2099         BUG_ON(!path);
2100
2101         /* FIXME, add redo link to tree so we don't leak on crash */
2102         key.objectid = inode->i_ino;
2103         key.offset = (u64)-1;
2104         key.type = (u8)-1;
2105
2106         btrfs_init_path(path);
2107
2108         ret = drop_csum_leaves(trans, root, path, inode, new_size);
2109         BUG_ON(ret);
2110
2111 search_again:
2112         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2113         if (ret < 0) {
2114                 goto error;
2115         }
2116         if (ret > 0) {
2117                 /* there are no items in the tree for us to truncate, we're
2118                  * done
2119                  */
2120                 if (path->slots[0] == 0) {
2121                         ret = 0;
2122                         goto error;
2123                 }
2124                 path->slots[0]--;
2125         }
2126
2127         while(1) {
2128                 fi = NULL;
2129                 leaf = path->nodes[0];
2130                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2131                 found_type = btrfs_key_type(&found_key);
2132
2133                 if (found_key.objectid != inode->i_ino)
2134                         break;
2135
2136                 if (found_type < min_type)
2137                         break;
2138
2139                 item_end = found_key.offset;
2140                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2141                         fi = btrfs_item_ptr(leaf, path->slots[0],
2142                                             struct btrfs_file_extent_item);
2143                         extent_type = btrfs_file_extent_type(leaf, fi);
2144                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2145                                 item_end +=
2146                                     btrfs_file_extent_num_bytes(leaf, fi);
2147                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2148                                 item_end += btrfs_file_extent_inline_len(leaf,
2149                                                                          fi);
2150                         }
2151                         item_end--;
2152                 }
2153                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2154                         ret = btrfs_csum_truncate(trans, root, path,
2155                                                   new_size);
2156                         BUG_ON(ret);
2157                 }
2158                 if (item_end < new_size) {
2159                         if (found_type == BTRFS_DIR_ITEM_KEY) {
2160                                 found_type = BTRFS_INODE_ITEM_KEY;
2161                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2162                                 found_type = BTRFS_CSUM_ITEM_KEY;
2163                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2164                                 found_type = BTRFS_XATTR_ITEM_KEY;
2165                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2166                                 found_type = BTRFS_INODE_REF_KEY;
2167                         } else if (found_type) {
2168                                 found_type--;
2169                         } else {
2170                                 break;
2171                         }
2172                         btrfs_set_key_type(&key, found_type);
2173                         goto next;
2174                 }
2175                 if (found_key.offset >= new_size)
2176                         del_item = 1;
2177                 else
2178                         del_item = 0;
2179                 found_extent = 0;
2180
2181                 /* FIXME, shrink the extent if the ref count is only 1 */
2182                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2183                         goto delete;
2184
2185                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2186                         u64 num_dec;
2187                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2188                         if (!del_item) {
2189                                 u64 orig_num_bytes =
2190                                         btrfs_file_extent_num_bytes(leaf, fi);
2191                                 extent_num_bytes = new_size -
2192                                         found_key.offset + root->sectorsize - 1;
2193                                 extent_num_bytes = extent_num_bytes &
2194                                         ~((u64)root->sectorsize - 1);
2195                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2196                                                          extent_num_bytes);
2197                                 num_dec = (orig_num_bytes -
2198                                            extent_num_bytes);
2199                                 if (root->ref_cows && extent_start != 0)
2200                                         inode_sub_bytes(inode, num_dec);
2201                                 btrfs_mark_buffer_dirty(leaf);
2202                         } else {
2203                                 extent_num_bytes =
2204                                         btrfs_file_extent_disk_num_bytes(leaf,
2205                                                                          fi);
2206                                 /* FIXME blocksize != 4096 */
2207                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2208                                 if (extent_start != 0) {
2209                                         found_extent = 1;
2210                                         if (root->ref_cows)
2211                                                 inode_sub_bytes(inode, num_dec);
2212                                 }
2213                                 root_gen = btrfs_header_generation(leaf);
2214                                 root_owner = btrfs_header_owner(leaf);
2215                         }
2216                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2217                         /*
2218                          * we can't truncate inline items that have had
2219                          * special encodings
2220                          */
2221                         if (!del_item &&
2222                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2223                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2224                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2225                                 u32 size = new_size - found_key.offset;
2226
2227                                 if (root->ref_cows) {
2228                                         inode_sub_bytes(inode, item_end + 1 -
2229                                                         new_size);
2230                                 }
2231                                 size =
2232                                     btrfs_file_extent_calc_inline_size(size);
2233                                 ret = btrfs_truncate_item(trans, root, path,
2234                                                           size, 1);
2235                                 BUG_ON(ret);
2236                         } else if (root->ref_cows) {
2237                                 inode_sub_bytes(inode, item_end + 1 -
2238                                                 found_key.offset);
2239                         }
2240                 }
2241 delete:
2242                 if (del_item) {
2243                         if (!pending_del_nr) {
2244                                 /* no pending yet, add ourselves */
2245                                 pending_del_slot = path->slots[0];
2246                                 pending_del_nr = 1;
2247                         } else if (pending_del_nr &&
2248                                    path->slots[0] + 1 == pending_del_slot) {
2249                                 /* hop on the pending chunk */
2250                                 pending_del_nr++;
2251                                 pending_del_slot = path->slots[0];
2252                         } else {
2253                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2254                         }
2255                 } else {
2256                         break;
2257                 }
2258                 if (found_extent) {
2259                         ret = btrfs_free_extent(trans, root, extent_start,
2260                                                 extent_num_bytes,
2261                                                 leaf->start, root_owner,
2262                                                 root_gen, inode->i_ino, 0);
2263                         BUG_ON(ret);
2264                 }
2265 next:
2266                 if (path->slots[0] == 0) {
2267                         if (pending_del_nr)
2268                                 goto del_pending;
2269                         btrfs_release_path(root, path);
2270                         goto search_again;
2271                 }
2272
2273                 path->slots[0]--;
2274                 if (pending_del_nr &&
2275                     path->slots[0] + 1 != pending_del_slot) {
2276                         struct btrfs_key debug;
2277 del_pending:
2278                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2279                                               pending_del_slot);
2280                         ret = btrfs_del_items(trans, root, path,
2281                                               pending_del_slot,
2282                                               pending_del_nr);
2283                         BUG_ON(ret);
2284                         pending_del_nr = 0;
2285                         btrfs_release_path(root, path);
2286                         goto search_again;
2287                 }
2288         }
2289         ret = 0;
2290 error:
2291         if (pending_del_nr) {
2292                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2293                                       pending_del_nr);
2294         }
2295         btrfs_free_path(path);
2296         inode->i_sb->s_dirt = 1;
2297         return ret;
2298 }
2299
2300 /*
2301  * taken from block_truncate_page, but does cow as it zeros out
2302  * any bytes left in the last page in the file.
2303  */
2304 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2305 {
2306         struct inode *inode = mapping->host;
2307         struct btrfs_root *root = BTRFS_I(inode)->root;
2308         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2309         struct btrfs_ordered_extent *ordered;
2310         char *kaddr;
2311         u32 blocksize = root->sectorsize;
2312         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2313         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2314         struct page *page;
2315         int ret = 0;
2316         u64 page_start;
2317         u64 page_end;
2318
2319         if ((offset & (blocksize - 1)) == 0)
2320                 goto out;
2321
2322         ret = -ENOMEM;
2323 again:
2324         page = grab_cache_page(mapping, index);
2325         if (!page)
2326                 goto out;
2327
2328         page_start = page_offset(page);
2329         page_end = page_start + PAGE_CACHE_SIZE - 1;
2330
2331         if (!PageUptodate(page)) {
2332                 ret = btrfs_readpage(NULL, page);
2333                 lock_page(page);
2334                 if (page->mapping != mapping) {
2335                         unlock_page(page);
2336                         page_cache_release(page);
2337                         goto again;
2338                 }
2339                 if (!PageUptodate(page)) {
2340                         ret = -EIO;
2341                         goto out_unlock;
2342                 }
2343         }
2344         wait_on_page_writeback(page);
2345
2346         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2347         set_page_extent_mapped(page);
2348
2349         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2350         if (ordered) {
2351                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2352                 unlock_page(page);
2353                 page_cache_release(page);
2354                 btrfs_start_ordered_extent(inode, ordered, 1);
2355                 btrfs_put_ordered_extent(ordered);
2356                 goto again;
2357         }
2358
2359         btrfs_set_extent_delalloc(inode, page_start, page_end);
2360         ret = 0;
2361         if (offset != PAGE_CACHE_SIZE) {
2362                 kaddr = kmap(page);
2363                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2364                 flush_dcache_page(page);
2365                 kunmap(page);
2366         }
2367         ClearPageChecked(page);
2368         set_page_dirty(page);
2369         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2370
2371 out_unlock:
2372         unlock_page(page);
2373         page_cache_release(page);
2374 out:
2375         return ret;
2376 }
2377
2378 int btrfs_cont_expand(struct inode *inode, loff_t size)
2379 {
2380         struct btrfs_trans_handle *trans;
2381         struct btrfs_root *root = BTRFS_I(inode)->root;
2382         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2383         struct extent_map *em;
2384         u64 mask = root->sectorsize - 1;
2385         u64 hole_start = (inode->i_size + mask) & ~mask;
2386         u64 block_end = (size + mask) & ~mask;
2387         u64 last_byte;
2388         u64 cur_offset;
2389         u64 hole_size;
2390         int err;
2391
2392         if (size <= hole_start)
2393                 return 0;
2394
2395         err = btrfs_check_free_space(root, 1, 0);
2396         if (err)
2397                 return err;
2398
2399         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2400
2401         while (1) {
2402                 struct btrfs_ordered_extent *ordered;
2403                 btrfs_wait_ordered_range(inode, hole_start,
2404                                          block_end - hole_start);
2405                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2406                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2407                 if (!ordered)
2408                         break;
2409                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2410                 btrfs_put_ordered_extent(ordered);
2411         }
2412
2413         trans = btrfs_start_transaction(root, 1);
2414         btrfs_set_trans_block_group(trans, inode);
2415
2416         cur_offset = hole_start;
2417         while (1) {
2418                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2419                                 block_end - cur_offset, 0);
2420                 BUG_ON(IS_ERR(em) || !em);
2421                 last_byte = min(extent_map_end(em), block_end);
2422                 last_byte = (last_byte + mask) & ~mask;
2423                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2424                         hole_size = last_byte - cur_offset;
2425                         err = btrfs_insert_file_extent(trans, root,
2426                                         inode->i_ino, cur_offset, 0,
2427                                         0, hole_size, 0, hole_size,
2428                                         0, 0, 0);
2429                         btrfs_drop_extent_cache(inode, hole_start,
2430                                         last_byte - 1, 0);
2431                 }
2432                 free_extent_map(em);
2433                 cur_offset = last_byte;
2434                 if (err || cur_offset >= block_end)
2435                         break;
2436         }
2437
2438         btrfs_end_transaction(trans, root);
2439         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2440         return err;
2441 }
2442
2443 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2444 {
2445         struct inode *inode = dentry->d_inode;
2446         int err;
2447
2448         err = inode_change_ok(inode, attr);
2449         if (err)
2450                 return err;
2451
2452         if (S_ISREG(inode->i_mode) &&
2453             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2454                 err = btrfs_cont_expand(inode, attr->ia_size);
2455                 if (err)
2456                         return err;
2457         }
2458
2459         err = inode_setattr(inode, attr);
2460
2461         if (!err && ((attr->ia_valid & ATTR_MODE)))
2462                 err = btrfs_acl_chmod(inode);
2463         return err;
2464 }
2465
2466 void btrfs_delete_inode(struct inode *inode)
2467 {
2468         struct btrfs_trans_handle *trans;
2469         struct btrfs_root *root = BTRFS_I(inode)->root;
2470         unsigned long nr;
2471         int ret;
2472
2473         truncate_inode_pages(&inode->i_data, 0);
2474         if (is_bad_inode(inode)) {
2475                 btrfs_orphan_del(NULL, inode);
2476                 goto no_delete;
2477         }
2478         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2479
2480         btrfs_i_size_write(inode, 0);
2481         trans = btrfs_start_transaction(root, 1);
2482
2483         btrfs_set_trans_block_group(trans, inode);
2484         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2485         if (ret) {
2486                 btrfs_orphan_del(NULL, inode);
2487                 goto no_delete_lock;
2488         }
2489
2490         btrfs_orphan_del(trans, inode);
2491
2492         nr = trans->blocks_used;
2493         clear_inode(inode);
2494
2495         btrfs_end_transaction(trans, root);
2496         btrfs_btree_balance_dirty(root, nr);
2497         return;
2498
2499 no_delete_lock:
2500         nr = trans->blocks_used;
2501         btrfs_end_transaction(trans, root);
2502         btrfs_btree_balance_dirty(root, nr);
2503 no_delete:
2504         clear_inode(inode);
2505 }
2506
2507 /*
2508  * this returns the key found in the dir entry in the location pointer.
2509  * If no dir entries were found, location->objectid is 0.
2510  */
2511 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2512                                struct btrfs_key *location)
2513 {
2514         const char *name = dentry->d_name.name;
2515         int namelen = dentry->d_name.len;
2516         struct btrfs_dir_item *di;
2517         struct btrfs_path *path;
2518         struct btrfs_root *root = BTRFS_I(dir)->root;
2519         int ret = 0;
2520
2521         path = btrfs_alloc_path();
2522         BUG_ON(!path);
2523
2524         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2525                                     namelen, 0);
2526         if (IS_ERR(di))
2527                 ret = PTR_ERR(di);
2528         if (!di || IS_ERR(di)) {
2529                 goto out_err;
2530         }
2531         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2532 out:
2533         btrfs_free_path(path);
2534         return ret;
2535 out_err:
2536         location->objectid = 0;
2537         goto out;
2538 }
2539
2540 /*
2541  * when we hit a tree root in a directory, the btrfs part of the inode
2542  * needs to be changed to reflect the root directory of the tree root.  This
2543  * is kind of like crossing a mount point.
2544  */
2545 static int fixup_tree_root_location(struct btrfs_root *root,
2546                              struct btrfs_key *location,
2547                              struct btrfs_root **sub_root,
2548                              struct dentry *dentry)
2549 {
2550         struct btrfs_root_item *ri;
2551
2552         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2553                 return 0;
2554         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2555                 return 0;
2556
2557         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2558                                         dentry->d_name.name,
2559                                         dentry->d_name.len);
2560         if (IS_ERR(*sub_root))
2561                 return PTR_ERR(*sub_root);
2562
2563         ri = &(*sub_root)->root_item;
2564         location->objectid = btrfs_root_dirid(ri);
2565         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2566         location->offset = 0;
2567
2568         return 0;
2569 }
2570
2571 static noinline void init_btrfs_i(struct inode *inode)
2572 {
2573         struct btrfs_inode *bi = BTRFS_I(inode);
2574
2575         bi->i_acl = NULL;
2576         bi->i_default_acl = NULL;
2577
2578         bi->generation = 0;
2579         bi->last_trans = 0;
2580         bi->logged_trans = 0;
2581         bi->delalloc_bytes = 0;
2582         bi->disk_i_size = 0;
2583         bi->flags = 0;
2584         bi->index_cnt = (u64)-1;
2585         bi->log_dirty_trans = 0;
2586         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2587         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2588                              inode->i_mapping, GFP_NOFS);
2589         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2590                              inode->i_mapping, GFP_NOFS);
2591         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2592         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2593         mutex_init(&BTRFS_I(inode)->csum_mutex);
2594         mutex_init(&BTRFS_I(inode)->extent_mutex);
2595         mutex_init(&BTRFS_I(inode)->log_mutex);
2596 }
2597
2598 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2599 {
2600         struct btrfs_iget_args *args = p;
2601         inode->i_ino = args->ino;
2602         init_btrfs_i(inode);
2603         BTRFS_I(inode)->root = args->root;
2604         return 0;
2605 }
2606
2607 static int btrfs_find_actor(struct inode *inode, void *opaque)
2608 {
2609         struct btrfs_iget_args *args = opaque;
2610         return (args->ino == inode->i_ino &&
2611                 args->root == BTRFS_I(inode)->root);
2612 }
2613
2614 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2615                             struct btrfs_root *root, int wait)
2616 {
2617         struct inode *inode;
2618         struct btrfs_iget_args args;
2619         args.ino = objectid;
2620         args.root = root;
2621
2622         if (wait) {
2623                 inode = ilookup5(s, objectid, btrfs_find_actor,
2624                                  (void *)&args);
2625         } else {
2626                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2627                                         (void *)&args);
2628         }
2629         return inode;
2630 }
2631
2632 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
2633                                 struct btrfs_root *root)
2634 {
2635         struct inode *inode;
2636         struct btrfs_iget_args args;
2637         args.ino = objectid;
2638         args.root = root;
2639
2640         inode = iget5_locked(s, objectid, btrfs_find_actor,
2641                              btrfs_init_locked_inode,
2642                              (void *)&args);
2643         return inode;
2644 }
2645
2646 /* Get an inode object given its location and corresponding root.
2647  * Returns in *is_new if the inode was read from disk
2648  */
2649 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
2650                          struct btrfs_root *root, int *is_new)
2651 {
2652         struct inode *inode;
2653
2654         inode = btrfs_iget_locked(s, location->objectid, root);
2655         if (!inode)
2656                 return ERR_PTR(-EACCES);
2657
2658         if (inode->i_state & I_NEW) {
2659                 BTRFS_I(inode)->root = root;
2660                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
2661                 btrfs_read_locked_inode(inode);
2662                 unlock_new_inode(inode);
2663                 if (is_new)
2664                         *is_new = 1;
2665         } else {
2666                 if (is_new)
2667                         *is_new = 0;
2668         }
2669
2670         return inode;
2671 }
2672
2673 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
2674                                    struct nameidata *nd)
2675 {
2676         struct inode * inode;
2677         struct btrfs_inode *bi = BTRFS_I(dir);
2678         struct btrfs_root *root = bi->root;
2679         struct btrfs_root *sub_root = root;
2680         struct btrfs_key location;
2681         int ret, new, do_orphan = 0;
2682
2683         if (dentry->d_name.len > BTRFS_NAME_LEN)
2684                 return ERR_PTR(-ENAMETOOLONG);
2685
2686         ret = btrfs_inode_by_name(dir, dentry, &location);
2687
2688         if (ret < 0)
2689                 return ERR_PTR(ret);
2690
2691         inode = NULL;
2692         if (location.objectid) {
2693                 ret = fixup_tree_root_location(root, &location, &sub_root,
2694                                                 dentry);
2695                 if (ret < 0)
2696                         return ERR_PTR(ret);
2697                 if (ret > 0)
2698                         return ERR_PTR(-ENOENT);
2699                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
2700                 if (IS_ERR(inode))
2701                         return ERR_CAST(inode);
2702
2703                 /* the inode and parent dir are two different roots */
2704                 if (new && root != sub_root) {
2705                         igrab(inode);
2706                         sub_root->inode = inode;
2707                         do_orphan = 1;
2708                 }
2709         }
2710
2711         if (unlikely(do_orphan))
2712                 btrfs_orphan_cleanup(sub_root);
2713
2714         return d_splice_alias(inode, dentry);
2715 }
2716
2717 static unsigned char btrfs_filetype_table[] = {
2718         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
2719 };
2720
2721 static int btrfs_real_readdir(struct file *filp, void *dirent,
2722                               filldir_t filldir)
2723 {
2724         struct inode *inode = filp->f_dentry->d_inode;
2725         struct btrfs_root *root = BTRFS_I(inode)->root;
2726         struct btrfs_item *item;
2727         struct btrfs_dir_item *di;
2728         struct btrfs_key key;
2729         struct btrfs_key found_key;
2730         struct btrfs_path *path;
2731         int ret;
2732         u32 nritems;
2733         struct extent_buffer *leaf;
2734         int slot;
2735         int advance;
2736         unsigned char d_type;
2737         int over = 0;
2738         u32 di_cur;
2739         u32 di_total;
2740         u32 di_len;
2741         int key_type = BTRFS_DIR_INDEX_KEY;
2742         char tmp_name[32];
2743         char *name_ptr;
2744         int name_len;
2745
2746         /* FIXME, use a real flag for deciding about the key type */
2747         if (root->fs_info->tree_root == root)
2748                 key_type = BTRFS_DIR_ITEM_KEY;
2749
2750         /* special case for "." */
2751         if (filp->f_pos == 0) {
2752                 over = filldir(dirent, ".", 1,
2753                                1, inode->i_ino,
2754                                DT_DIR);
2755                 if (over)
2756                         return 0;
2757                 filp->f_pos = 1;
2758         }
2759         /* special case for .., just use the back ref */
2760         if (filp->f_pos == 1) {
2761                 u64 pino = parent_ino(filp->f_path.dentry);
2762                 over = filldir(dirent, "..", 2,
2763                                2, pino, DT_DIR);
2764                 if (over)
2765                         return 0;
2766                 filp->f_pos = 2;
2767         }
2768
2769         path = btrfs_alloc_path();
2770         path->reada = 2;
2771
2772         btrfs_set_key_type(&key, key_type);
2773         key.offset = filp->f_pos;
2774         key.objectid = inode->i_ino;
2775
2776         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2777         if (ret < 0)
2778                 goto err;
2779         advance = 0;
2780
2781         while (1) {
2782                 leaf = path->nodes[0];
2783                 nritems = btrfs_header_nritems(leaf);
2784                 slot = path->slots[0];
2785                 if (advance || slot >= nritems) {
2786                         if (slot >= nritems - 1) {
2787                                 ret = btrfs_next_leaf(root, path);
2788                                 if (ret)
2789                                         break;
2790                                 leaf = path->nodes[0];
2791                                 nritems = btrfs_header_nritems(leaf);
2792                                 slot = path->slots[0];
2793                         } else {
2794                                 slot++;
2795                                 path->slots[0]++;
2796                         }
2797                 }
2798                 advance = 1;
2799                 item = btrfs_item_nr(leaf, slot);
2800                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2801
2802                 if (found_key.objectid != key.objectid)
2803                         break;
2804                 if (btrfs_key_type(&found_key) != key_type)
2805                         break;
2806                 if (found_key.offset < filp->f_pos)
2807                         continue;
2808
2809                 filp->f_pos = found_key.offset;
2810
2811                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2812                 di_cur = 0;
2813                 di_total = btrfs_item_size(leaf, item);
2814
2815                 while (di_cur < di_total) {
2816                         struct btrfs_key location;
2817
2818                         name_len = btrfs_dir_name_len(leaf, di);
2819                         if (name_len <= sizeof(tmp_name)) {
2820                                 name_ptr = tmp_name;
2821                         } else {
2822                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2823                                 if (!name_ptr) {
2824                                         ret = -ENOMEM;
2825                                         goto err;
2826                                 }
2827                         }
2828                         read_extent_buffer(leaf, name_ptr,
2829                                            (unsigned long)(di + 1), name_len);
2830
2831                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2832                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2833                         over = filldir(dirent, name_ptr, name_len,
2834                                        found_key.offset, location.objectid,
2835                                        d_type);
2836
2837                         if (name_ptr != tmp_name)
2838                                 kfree(name_ptr);
2839
2840                         if (over)
2841                                 goto nopos;
2842
2843                         di_len = btrfs_dir_name_len(leaf, di) +
2844                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2845                         di_cur += di_len;
2846                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2847                 }
2848         }
2849
2850         /* Reached end of directory/root. Bump pos past the last item. */
2851         if (key_type == BTRFS_DIR_INDEX_KEY)
2852                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2853         else
2854                 filp->f_pos++;
2855 nopos:
2856         ret = 0;
2857 err:
2858         btrfs_free_path(path);
2859         return ret;
2860 }
2861
2862 int btrfs_write_inode(struct inode *inode, int wait)
2863 {
2864         struct btrfs_root *root = BTRFS_I(inode)->root;
2865         struct btrfs_trans_handle *trans;
2866         int ret = 0;
2867
2868         if (root->fs_info->closing > 1)
2869                 return 0;
2870
2871         if (wait) {
2872                 trans = btrfs_join_transaction(root, 1);
2873                 btrfs_set_trans_block_group(trans, inode);
2874                 ret = btrfs_commit_transaction(trans, root);
2875         }
2876         return ret;
2877 }
2878
2879 /*
2880  * This is somewhat expensive, updating the tree every time the
2881  * inode changes.  But, it is most likely to find the inode in cache.
2882  * FIXME, needs more benchmarking...there are no reasons other than performance
2883  * to keep or drop this code.
2884  */
2885 void btrfs_dirty_inode(struct inode *inode)
2886 {
2887         struct btrfs_root *root = BTRFS_I(inode)->root;
2888         struct btrfs_trans_handle *trans;
2889
2890         trans = btrfs_join_transaction(root, 1);
2891         btrfs_set_trans_block_group(trans, inode);
2892         btrfs_update_inode(trans, root, inode);
2893         btrfs_end_transaction(trans, root);
2894 }
2895
2896 /*
2897  * find the highest existing sequence number in a directory
2898  * and then set the in-memory index_cnt variable to reflect
2899  * free sequence numbers
2900  */
2901 static int btrfs_set_inode_index_count(struct inode *inode)
2902 {
2903         struct btrfs_root *root = BTRFS_I(inode)->root;
2904         struct btrfs_key key, found_key;
2905         struct btrfs_path *path;
2906         struct extent_buffer *leaf;
2907         int ret;
2908
2909         key.objectid = inode->i_ino;
2910         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2911         key.offset = (u64)-1;
2912
2913         path = btrfs_alloc_path();
2914         if (!path)
2915                 return -ENOMEM;
2916
2917         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2918         if (ret < 0)
2919                 goto out;
2920         /* FIXME: we should be able to handle this */
2921         if (ret == 0)
2922                 goto out;
2923         ret = 0;
2924
2925         /*
2926          * MAGIC NUMBER EXPLANATION:
2927          * since we search a directory based on f_pos we have to start at 2
2928          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2929          * else has to start at 2
2930          */
2931         if (path->slots[0] == 0) {
2932                 BTRFS_I(inode)->index_cnt = 2;
2933                 goto out;
2934         }
2935
2936         path->slots[0]--;
2937
2938         leaf = path->nodes[0];
2939         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2940
2941         if (found_key.objectid != inode->i_ino ||
2942             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2943                 BTRFS_I(inode)->index_cnt = 2;
2944                 goto out;
2945         }
2946
2947         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2948 out:
2949         btrfs_free_path(path);
2950         return ret;
2951 }
2952
2953 /*
2954  * helper to find a free sequence number in a given directory.  This current
2955  * code is very simple, later versions will do smarter things in the btree
2956  */
2957 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2958                                  u64 *index)
2959 {
2960         int ret = 0;
2961
2962         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2963                 ret = btrfs_set_inode_index_count(dir);
2964                 if (ret) {
2965                         return ret;
2966                 }
2967         }
2968
2969         *index = BTRFS_I(dir)->index_cnt;
2970         BTRFS_I(dir)->index_cnt++;
2971
2972         return ret;
2973 }
2974
2975 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2976                                      struct btrfs_root *root,
2977                                      struct inode *dir,
2978                                      const char *name, int name_len,
2979                                      u64 ref_objectid,
2980                                      u64 objectid,
2981                                      struct btrfs_block_group_cache *group,
2982                                      int mode, u64 *index)
2983 {
2984         struct inode *inode;
2985         struct btrfs_inode_item *inode_item;
2986         struct btrfs_block_group_cache *new_inode_group;
2987         struct btrfs_key *location;
2988         struct btrfs_path *path;
2989         struct btrfs_inode_ref *ref;
2990         struct btrfs_key key[2];
2991         u32 sizes[2];
2992         unsigned long ptr;
2993         int ret;
2994         int owner;
2995
2996         path = btrfs_alloc_path();
2997         BUG_ON(!path);
2998
2999         inode = new_inode(root->fs_info->sb);
3000         if (!inode)
3001                 return ERR_PTR(-ENOMEM);
3002
3003         if (dir) {
3004                 ret = btrfs_set_inode_index(dir, inode, index);
3005                 if (ret)
3006                         return ERR_PTR(ret);
3007         }
3008         /*
3009          * index_cnt is ignored for everything but a dir,
3010          * btrfs_get_inode_index_count has an explanation for the magic
3011          * number
3012          */
3013         init_btrfs_i(inode);
3014         BTRFS_I(inode)->index_cnt = 2;
3015         BTRFS_I(inode)->root = root;
3016         BTRFS_I(inode)->generation = trans->transid;
3017
3018         if (mode & S_IFDIR)
3019                 owner = 0;
3020         else
3021                 owner = 1;
3022         new_inode_group = btrfs_find_block_group(root, group, 0,
3023                                        BTRFS_BLOCK_GROUP_METADATA, owner);
3024         if (!new_inode_group) {
3025                 printk("find_block group failed\n");
3026                 new_inode_group = group;
3027         }
3028         BTRFS_I(inode)->block_group = new_inode_group;
3029
3030         key[0].objectid = objectid;
3031         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3032         key[0].offset = 0;
3033
3034         key[1].objectid = objectid;
3035         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3036         key[1].offset = ref_objectid;
3037
3038         sizes[0] = sizeof(struct btrfs_inode_item);
3039         sizes[1] = name_len + sizeof(*ref);
3040
3041         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3042         if (ret != 0)
3043                 goto fail;
3044
3045         if (objectid > root->highest_inode)
3046                 root->highest_inode = objectid;
3047
3048         inode->i_uid = current->fsuid;
3049         inode->i_gid = current->fsgid;
3050         inode->i_mode = mode;
3051         inode->i_ino = objectid;
3052         inode_set_bytes(inode, 0);
3053         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3054         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3055                                   struct btrfs_inode_item);
3056         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3057
3058         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3059                              struct btrfs_inode_ref);
3060         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3061         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3062         ptr = (unsigned long)(ref + 1);
3063         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3064
3065         btrfs_mark_buffer_dirty(path->nodes[0]);
3066         btrfs_free_path(path);
3067
3068         location = &BTRFS_I(inode)->location;
3069         location->objectid = objectid;
3070         location->offset = 0;
3071         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3072
3073         insert_inode_hash(inode);
3074         return inode;
3075 fail:
3076         if (dir)
3077                 BTRFS_I(dir)->index_cnt--;
3078         btrfs_free_path(path);
3079         return ERR_PTR(ret);
3080 }
3081
3082 static inline u8 btrfs_inode_type(struct inode *inode)
3083 {
3084         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3085 }
3086
3087 /*
3088  * utility function to add 'inode' into 'parent_inode' with
3089  * a give name and a given sequence number.
3090  * if 'add_backref' is true, also insert a backref from the
3091  * inode to the parent directory.
3092  */
3093 int btrfs_add_link(struct btrfs_trans_handle *trans,
3094                    struct inode *parent_inode, struct inode *inode,
3095                    const char *name, int name_len, int add_backref, u64 index)
3096 {
3097         int ret;
3098         struct btrfs_key key;
3099         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3100
3101         key.objectid = inode->i_ino;
3102         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3103         key.offset = 0;
3104
3105         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3106                                     parent_inode->i_ino,
3107                                     &key, btrfs_inode_type(inode),
3108                                     index);
3109         if (ret == 0) {
3110                 if (add_backref) {
3111                         ret = btrfs_insert_inode_ref(trans, root,
3112                                                      name, name_len,
3113                                                      inode->i_ino,
3114                                                      parent_inode->i_ino,
3115                                                      index);
3116                 }
3117                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3118                                    name_len * 2);
3119                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3120                 ret = btrfs_update_inode(trans, root, parent_inode);
3121         }
3122         return ret;
3123 }
3124
3125 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3126                             struct dentry *dentry, struct inode *inode,
3127                             int backref, u64 index)
3128 {
3129         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3130                                  inode, dentry->d_name.name,
3131                                  dentry->d_name.len, backref, index);
3132         if (!err) {
3133                 d_instantiate(dentry, inode);
3134                 return 0;
3135         }
3136         if (err > 0)
3137                 err = -EEXIST;
3138         return err;
3139 }
3140
3141 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3142                         int mode, dev_t rdev)
3143 {
3144         struct btrfs_trans_handle *trans;
3145         struct btrfs_root *root = BTRFS_I(dir)->root;
3146         struct inode *inode = NULL;
3147         int err;
3148         int drop_inode = 0;
3149         u64 objectid;
3150         unsigned long nr = 0;
3151         u64 index = 0;
3152
3153         if (!new_valid_dev(rdev))
3154                 return -EINVAL;
3155
3156         err = btrfs_check_free_space(root, 1, 0);
3157         if (err)
3158                 goto fail;
3159
3160         trans = btrfs_start_transaction(root, 1);
3161         btrfs_set_trans_block_group(trans, dir);
3162
3163         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3164         if (err) {
3165                 err = -ENOSPC;
3166                 goto out_unlock;
3167         }
3168
3169         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3170                                 dentry->d_name.len,
3171                                 dentry->d_parent->d_inode->i_ino, objectid,
3172                                 BTRFS_I(dir)->block_group, mode, &index);
3173         err = PTR_ERR(inode);
3174         if (IS_ERR(inode))
3175                 goto out_unlock;
3176
3177         err = btrfs_init_acl(inode, dir);
3178         if (err) {
3179                 drop_inode = 1;
3180                 goto out_unlock;
3181         }
3182
3183         btrfs_set_trans_block_group(trans, inode);
3184         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3185         if (err)
3186                 drop_inode = 1;
3187         else {
3188                 inode->i_op = &btrfs_special_inode_operations;
3189                 init_special_inode(inode, inode->i_mode, rdev);
3190                 btrfs_update_inode(trans, root, inode);
3191         }
3192         dir->i_sb->s_dirt = 1;
3193         btrfs_update_inode_block_group(trans, inode);
3194         btrfs_update_inode_block_group(trans, dir);
3195 out_unlock:
3196         nr = trans->blocks_used;
3197         btrfs_end_transaction_throttle(trans, root);
3198 fail:
3199         if (drop_inode) {
3200                 inode_dec_link_count(inode);
3201                 iput(inode);
3202         }
3203         btrfs_btree_balance_dirty(root, nr);
3204         return err;
3205 }
3206
3207 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3208                         int mode, struct nameidata *nd)
3209 {
3210         struct btrfs_trans_handle *trans;
3211         struct btrfs_root *root = BTRFS_I(dir)->root;
3212         struct inode *inode = NULL;
3213         int err;
3214         int drop_inode = 0;
3215         unsigned long nr = 0;
3216         u64 objectid;
3217         u64 index = 0;
3218
3219         err = btrfs_check_free_space(root, 1, 0);
3220         if (err)
3221                 goto fail;
3222         trans = btrfs_start_transaction(root, 1);
3223         btrfs_set_trans_block_group(trans, dir);
3224
3225         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3226         if (err) {
3227                 err = -ENOSPC;
3228                 goto out_unlock;
3229         }
3230
3231         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3232                                 dentry->d_name.len,
3233                                 dentry->d_parent->d_inode->i_ino,
3234                                 objectid, BTRFS_I(dir)->block_group, mode,
3235                                 &index);
3236         err = PTR_ERR(inode);
3237         if (IS_ERR(inode))
3238                 goto out_unlock;
3239
3240         err = btrfs_init_acl(inode, dir);
3241         if (err) {
3242                 drop_inode = 1;
3243                 goto out_unlock;
3244         }
3245
3246         btrfs_set_trans_block_group(trans, inode);
3247         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3248         if (err)
3249                 drop_inode = 1;
3250         else {
3251                 inode->i_mapping->a_ops = &btrfs_aops;
3252                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3253                 inode->i_fop = &btrfs_file_operations;
3254                 inode->i_op = &btrfs_file_inode_operations;
3255                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3256         }
3257         dir->i_sb->s_dirt = 1;
3258         btrfs_update_inode_block_group(trans, inode);
3259         btrfs_update_inode_block_group(trans, dir);
3260 out_unlock:
3261         nr = trans->blocks_used;
3262         btrfs_end_transaction_throttle(trans, root);
3263 fail:
3264         if (drop_inode) {
3265                 inode_dec_link_count(inode);
3266                 iput(inode);
3267         }
3268         btrfs_btree_balance_dirty(root, nr);
3269         return err;
3270 }
3271
3272 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3273                       struct dentry *dentry)
3274 {
3275         struct btrfs_trans_handle *trans;
3276         struct btrfs_root *root = BTRFS_I(dir)->root;
3277         struct inode *inode = old_dentry->d_inode;
3278         u64 index;
3279         unsigned long nr = 0;
3280         int err;
3281         int drop_inode = 0;
3282
3283         if (inode->i_nlink == 0)
3284                 return -ENOENT;
3285
3286         btrfs_inc_nlink(inode);
3287         err = btrfs_check_free_space(root, 1, 0);
3288         if (err)
3289                 goto fail;
3290         err = btrfs_set_inode_index(dir, inode, &index);
3291         if (err)
3292                 goto fail;
3293
3294         trans = btrfs_start_transaction(root, 1);
3295
3296         btrfs_set_trans_block_group(trans, dir);
3297         atomic_inc(&inode->i_count);
3298
3299         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3300
3301         if (err)
3302                 drop_inode = 1;
3303
3304         dir->i_sb->s_dirt = 1;
3305         btrfs_update_inode_block_group(trans, dir);
3306         err = btrfs_update_inode(trans, root, inode);
3307
3308         if (err)
3309                 drop_inode = 1;
3310
3311         nr = trans->blocks_used;
3312         btrfs_end_transaction_throttle(trans, root);
3313 fail:
3314         if (drop_inode) {
3315                 inode_dec_link_count(inode);
3316                 iput(inode);
3317         }
3318         btrfs_btree_balance_dirty(root, nr);
3319         return err;
3320 }
3321
3322 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3323 {
3324         struct inode *inode = NULL;
3325         struct btrfs_trans_handle *trans;
3326         struct btrfs_root *root = BTRFS_I(dir)->root;
3327         int err = 0;
3328         int drop_on_err = 0;
3329         u64 objectid = 0;
3330         u64 index = 0;
3331         unsigned long nr = 1;
3332
3333         err = btrfs_check_free_space(root, 1, 0);
3334         if (err)
3335                 goto out_unlock;
3336
3337         trans = btrfs_start_transaction(root, 1);
3338         btrfs_set_trans_block_group(trans, dir);
3339
3340         if (IS_ERR(trans)) {
3341                 err = PTR_ERR(trans);
3342                 goto out_unlock;
3343         }
3344
3345         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3346         if (err) {
3347                 err = -ENOSPC;
3348                 goto out_unlock;
3349         }
3350
3351         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3352                                 dentry->d_name.len,
3353                                 dentry->d_parent->d_inode->i_ino, objectid,
3354                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3355                                 &index);
3356         if (IS_ERR(inode)) {
3357                 err = PTR_ERR(inode);
3358                 goto out_fail;
3359         }
3360
3361         drop_on_err = 1;
3362
3363         err = btrfs_init_acl(inode, dir);
3364         if (err)
3365                 goto out_fail;
3366
3367         inode->i_op = &btrfs_dir_inode_operations;
3368         inode->i_fop = &btrfs_dir_file_operations;
3369         btrfs_set_trans_block_group(trans, inode);
3370
3371         btrfs_i_size_write(inode, 0);
3372         err = btrfs_update_inode(trans, root, inode);
3373         if (err)
3374                 goto out_fail;
3375
3376         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3377                                  inode, dentry->d_name.name,
3378                                  dentry->d_name.len, 0, index);
3379         if (err)
3380                 goto out_fail;
3381
3382         d_instantiate(dentry, inode);
3383         drop_on_err = 0;
3384         dir->i_sb->s_dirt = 1;
3385         btrfs_update_inode_block_group(trans, inode);
3386         btrfs_update_inode_block_group(trans, dir);
3387
3388 out_fail:
3389         nr = trans->blocks_used;
3390         btrfs_end_transaction_throttle(trans, root);
3391
3392 out_unlock:
3393         if (drop_on_err)
3394                 iput(inode);
3395         btrfs_btree_balance_dirty(root, nr);
3396         return err;
3397 }
3398
3399 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3400  * and an extent that you want to insert, deal with overlap and insert
3401  * the new extent into the tree.
3402  */
3403 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3404                                 struct extent_map *existing,
3405                                 struct extent_map *em,
3406                                 u64 map_start, u64 map_len)
3407 {
3408         u64 start_diff;
3409
3410         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3411         start_diff = map_start - em->start;
3412         em->start = map_start;
3413         em->len = map_len;
3414         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3415             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3416                 em->block_start += start_diff;
3417                 em->block_len -= start_diff;
3418         }
3419         return add_extent_mapping(em_tree, em);
3420 }
3421
3422 static noinline int uncompress_inline(struct btrfs_path *path,
3423                                       struct inode *inode, struct page *page,
3424                                       size_t pg_offset, u64 extent_offset,
3425                                       struct btrfs_file_extent_item *item)
3426 {
3427         int ret;
3428         struct extent_buffer *leaf = path->nodes[0];
3429         char *tmp;
3430         size_t max_size;
3431         unsigned long inline_size;
3432         unsigned long ptr;
3433
3434         WARN_ON(pg_offset != 0);
3435         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3436         inline_size = btrfs_file_extent_inline_item_len(leaf,
3437                                         btrfs_item_nr(leaf, path->slots[0]));
3438         tmp = kmalloc(inline_size, GFP_NOFS);
3439         ptr = btrfs_file_extent_inline_start(item);
3440
3441         read_extent_buffer(leaf, tmp, ptr, inline_size);
3442
3443         max_size = min(PAGE_CACHE_SIZE, max_size);
3444         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3445                                     inline_size, max_size);
3446         if (ret) {
3447                 char *kaddr = kmap_atomic(page, KM_USER0);
3448                 unsigned long copy_size = min_t(u64,
3449                                   PAGE_CACHE_SIZE - pg_offset,
3450                                   max_size - extent_offset);
3451                 memset(kaddr + pg_offset, 0, copy_size);
3452                 kunmap_atomic(kaddr, KM_USER0);
3453         }
3454         kfree(tmp);
3455         return 0;
3456 }
3457
3458 /*
3459  * a bit scary, this does extent mapping from logical file offset to the disk.
3460  * the ugly parts come from merging extents from the disk with the
3461  * in-ram representation.  This gets more complex because of the data=ordered code,
3462  * where the in-ram extents might be locked pending data=ordered completion.
3463  *
3464  * This also copies inline extents directly into the page.
3465  */
3466 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3467                                     size_t pg_offset, u64 start, u64 len,
3468                                     int create)
3469 {
3470         int ret;
3471         int err = 0;
3472         u64 bytenr;
3473         u64 extent_start = 0;
3474         u64 extent_end = 0;
3475         u64 objectid = inode->i_ino;
3476         u32 found_type;
3477         struct btrfs_path *path = NULL;
3478         struct btrfs_root *root = BTRFS_I(inode)->root;
3479         struct btrfs_file_extent_item *item;
3480         struct extent_buffer *leaf;
3481         struct btrfs_key found_key;
3482         struct extent_map *em = NULL;
3483         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3484         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3485         struct btrfs_trans_handle *trans = NULL;
3486         int compressed;
3487
3488 again:
3489         spin_lock(&em_tree->lock);
3490         em = lookup_extent_mapping(em_tree, start, len);
3491         if (em)
3492                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3493         spin_unlock(&em_tree->lock);
3494
3495         if (em) {
3496                 if (em->start > start || em->start + em->len <= start)
3497                         free_extent_map(em);
3498                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3499                         free_extent_map(em);
3500                 else
3501                         goto out;
3502         }
3503         em = alloc_extent_map(GFP_NOFS);
3504         if (!em) {
3505                 err = -ENOMEM;
3506                 goto out;
3507         }
3508         em->bdev = root->fs_info->fs_devices->latest_bdev;
3509         em->start = EXTENT_MAP_HOLE;
3510         em->len = (u64)-1;
3511         em->block_len = (u64)-1;
3512
3513         if (!path) {
3514                 path = btrfs_alloc_path();
3515                 BUG_ON(!path);
3516         }
3517
3518         ret = btrfs_lookup_file_extent(trans, root, path,
3519                                        objectid, start, trans != NULL);
3520         if (ret < 0) {
3521                 err = ret;
3522                 goto out;
3523         }
3524
3525         if (ret != 0) {
3526                 if (path->slots[0] == 0)
3527                         goto not_found;
3528                 path->slots[0]--;
3529         }
3530
3531         leaf = path->nodes[0];
3532         item = btrfs_item_ptr(leaf, path->slots[0],
3533                               struct btrfs_file_extent_item);
3534         /* are we inside the extent that was found? */
3535         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3536         found_type = btrfs_key_type(&found_key);
3537         if (found_key.objectid != objectid ||
3538             found_type != BTRFS_EXTENT_DATA_KEY) {
3539                 goto not_found;
3540         }
3541
3542         found_type = btrfs_file_extent_type(leaf, item);
3543         extent_start = found_key.offset;
3544         compressed = btrfs_file_extent_compression(leaf, item);
3545         if (found_type == BTRFS_FILE_EXTENT_REG ||
3546             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3547                 extent_end = extent_start +
3548                        btrfs_file_extent_num_bytes(leaf, item);
3549         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3550                 size_t size;
3551                 size = btrfs_file_extent_inline_len(leaf, item);
3552                 extent_end = (extent_start + size + root->sectorsize - 1) &
3553                         ~((u64)root->sectorsize - 1);
3554         }
3555
3556         if (start >= extent_end) {
3557                 path->slots[0]++;
3558                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3559                         ret = btrfs_next_leaf(root, path);
3560                         if (ret < 0) {
3561                                 err = ret;
3562                                 goto out;
3563                         }
3564                         if (ret > 0)
3565                                 goto not_found;
3566                         leaf = path->nodes[0];
3567                 }
3568                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3569                 if (found_key.objectid != objectid ||
3570                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3571                         goto not_found;
3572                 if (start + len <= found_key.offset)
3573                         goto not_found;
3574                 em->start = start;
3575                 em->len = found_key.offset - start;
3576                 goto not_found_em;
3577         }
3578
3579         if (found_type == BTRFS_FILE_EXTENT_REG ||
3580             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3581                 em->start = extent_start;
3582                 em->len = extent_end - extent_start;
3583                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3584                 if (bytenr == 0) {
3585                         em->block_start = EXTENT_MAP_HOLE;
3586                         goto insert;
3587                 }
3588                 if (compressed) {
3589                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3590                         em->block_start = bytenr;
3591                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3592                                                                          item);
3593                 } else {
3594                         bytenr += btrfs_file_extent_offset(leaf, item);
3595                         em->block_start = bytenr;
3596                         em->block_len = em->len;
3597                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3598                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3599                 }
3600                 goto insert;
3601         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3602                 unsigned long ptr;
3603                 char *map;
3604                 size_t size;
3605                 size_t extent_offset;
3606                 size_t copy_size;
3607
3608                 em->block_start = EXTENT_MAP_INLINE;
3609                 if (!page || create) {
3610                         em->start = extent_start;
3611                         em->len = extent_end - extent_start;
3612                         goto out;
3613                 }
3614
3615                 size = btrfs_file_extent_inline_len(leaf, item);
3616                 extent_offset = page_offset(page) + pg_offset - extent_start;
3617                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3618                                 size - extent_offset);
3619                 em->start = extent_start + extent_offset;
3620                 em->len = (copy_size + root->sectorsize - 1) &
3621                         ~((u64)root->sectorsize - 1);
3622                 if (compressed)
3623                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3624                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
3625                 if (create == 0 && !PageUptodate(page)) {
3626                         if (btrfs_file_extent_compression(leaf, item) ==
3627                             BTRFS_COMPRESS_ZLIB) {
3628                                 ret = uncompress_inline(path, inode, page,
3629                                                         pg_offset,
3630                                                         extent_offset, item);
3631                                 BUG_ON(ret);
3632                         } else {
3633                                 map = kmap(page);
3634                                 read_extent_buffer(leaf, map + pg_offset, ptr,
3635                                                    copy_size);
3636                                 kunmap(page);
3637                         }
3638                         flush_dcache_page(page);
3639                 } else if (create && PageUptodate(page)) {
3640                         if (!trans) {
3641                                 kunmap(page);
3642                                 free_extent_map(em);
3643                                 em = NULL;
3644                                 btrfs_release_path(root, path);
3645                                 trans = btrfs_join_transaction(root, 1);
3646                                 goto again;
3647                         }
3648                         map = kmap(page);
3649                         write_extent_buffer(leaf, map + pg_offset, ptr,
3650                                             copy_size);
3651                         kunmap(page);
3652                         btrfs_mark_buffer_dirty(leaf);
3653                 }
3654                 set_extent_uptodate(io_tree, em->start,
3655                                     extent_map_end(em) - 1, GFP_NOFS);
3656                 goto insert;
3657         } else {
3658                 printk("unkknown found_type %d\n", found_type);
3659                 WARN_ON(1);
3660         }
3661 not_found:
3662         em->start = start;
3663         em->len = len;
3664 not_found_em:
3665         em->block_start = EXTENT_MAP_HOLE;
3666         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
3667 insert:
3668         btrfs_release_path(root, path);
3669         if (em->start > start || extent_map_end(em) <= start) {
3670                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
3671                 err = -EIO;
3672                 goto out;
3673         }
3674
3675         err = 0;
3676         spin_lock(&em_tree->lock);
3677         ret = add_extent_mapping(em_tree, em);
3678         /* it is possible that someone inserted the extent into the tree
3679          * while we had the lock dropped.  It is also possible that
3680          * an overlapping map exists in the tree
3681          */
3682         if (ret == -EEXIST) {
3683                 struct extent_map *existing;
3684
3685                 ret = 0;
3686
3687                 existing = lookup_extent_mapping(em_tree, start, len);
3688                 if (existing && (existing->start > start ||
3689                     existing->start + existing->len <= start)) {
3690                         free_extent_map(existing);
3691                         existing = NULL;
3692                 }
3693                 if (!existing) {
3694                         existing = lookup_extent_mapping(em_tree, em->start,
3695                                                          em->len);
3696                         if (existing) {
3697                                 err = merge_extent_mapping(em_tree, existing,
3698                                                            em, start,
3699                                                            root->sectorsize);
3700                                 free_extent_map(existing);
3701                                 if (err) {
3702                                         free_extent_map(em);
3703                                         em = NULL;
3704                                 }
3705                         } else {
3706                                 err = -EIO;
3707                                 printk("failing to insert %Lu %Lu\n",
3708                                        start, len);
3709                                 free_extent_map(em);
3710                                 em = NULL;
3711                         }
3712                 } else {
3713                         free_extent_map(em);
3714                         em = existing;
3715                         err = 0;
3716                 }
3717         }
3718         spin_unlock(&em_tree->lock);
3719 out:
3720         if (path)
3721                 btrfs_free_path(path);
3722         if (trans) {
3723                 ret = btrfs_end_transaction(trans, root);
3724                 if (!err) {
3725                         err = ret;
3726                 }
3727         }
3728         if (err) {
3729                 free_extent_map(em);
3730                 WARN_ON(1);
3731                 return ERR_PTR(err);
3732         }
3733         return em;
3734 }
3735
3736 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
3737                         const struct iovec *iov, loff_t offset,
3738                         unsigned long nr_segs)
3739 {
3740         return -EINVAL;
3741 }
3742
3743 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3744 {
3745         return extent_bmap(mapping, iblock, btrfs_get_extent);
3746 }
3747
3748 int btrfs_readpage(struct file *file, struct page *page)
3749 {
3750         struct extent_io_tree *tree;
3751         tree = &BTRFS_I(page->mapping->host)->io_tree;
3752         return extent_read_full_page(tree, page, btrfs_get_extent);
3753 }
3754
3755 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3756 {
3757         struct extent_io_tree *tree;
3758
3759
3760         if (current->flags & PF_MEMALLOC) {
3761                 redirty_page_for_writepage(wbc, page);
3762                 unlock_page(page);
3763                 return 0;
3764         }
3765         tree = &BTRFS_I(page->mapping->host)->io_tree;
3766         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3767 }
3768
3769 int btrfs_writepages(struct address_space *mapping,
3770                      struct writeback_control *wbc)
3771 {
3772         struct extent_io_tree *tree;
3773         tree = &BTRFS_I(mapping->host)->io_tree;
3774         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3775 }
3776
3777 static int
3778 btrfs_readpages(struct file *file, struct address_space *mapping,
3779                 struct list_head *pages, unsigned nr_pages)
3780 {
3781         struct extent_io_tree *tree;
3782         tree = &BTRFS_I(mapping->host)->io_tree;
3783         return extent_readpages(tree, mapping, pages, nr_pages,
3784                                 btrfs_get_extent);
3785 }
3786 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3787 {
3788         struct extent_io_tree *tree;
3789         struct extent_map_tree *map;
3790         int ret;
3791
3792         tree = &BTRFS_I(page->mapping->host)->io_tree;
3793         map = &BTRFS_I(page->mapping->host)->extent_tree;
3794         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3795         if (ret == 1) {
3796                 ClearPagePrivate(page);
3797                 set_page_private(page, 0);
3798                 page_cache_release(page);
3799         }
3800         return ret;
3801 }
3802
3803 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3804 {
3805         if (PageWriteback(page) || PageDirty(page))
3806                 return 0;
3807         return __btrfs_releasepage(page, gfp_flags);
3808 }
3809
3810 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3811 {
3812         struct extent_io_tree *tree;
3813         struct btrfs_ordered_extent *ordered;
3814         u64 page_start = page_offset(page);
3815         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3816
3817         wait_on_page_writeback(page);
3818         tree = &BTRFS_I(page->mapping->host)->io_tree;
3819         if (offset) {
3820                 btrfs_releasepage(page, GFP_NOFS);
3821                 return;
3822         }
3823
3824         lock_extent(tree, page_start, page_end, GFP_NOFS);
3825         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3826                                            page_offset(page));
3827         if (ordered) {
3828                 /*
3829                  * IO on this page will never be started, so we need
3830                  * to account for any ordered extents now
3831                  */
3832                 clear_extent_bit(tree, page_start, page_end,
3833                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3834                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3835                 btrfs_finish_ordered_io(page->mapping->host,
3836                                         page_start, page_end);
3837                 btrfs_put_ordered_extent(ordered);
3838                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3839         }
3840         clear_extent_bit(tree, page_start, page_end,
3841                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3842                  EXTENT_ORDERED,
3843                  1, 1, GFP_NOFS);
3844         __btrfs_releasepage(page, GFP_NOFS);
3845
3846         ClearPageChecked(page);
3847         if (PagePrivate(page)) {
3848                 ClearPagePrivate(page);
3849                 set_page_private(page, 0);
3850                 page_cache_release(page);
3851         }
3852 }
3853
3854 /*
3855  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3856  * called from a page fault handler when a page is first dirtied. Hence we must
3857  * be careful to check for EOF conditions here. We set the page up correctly
3858  * for a written page which means we get ENOSPC checking when writing into
3859  * holes and correct delalloc and unwritten extent mapping on filesystems that
3860  * support these features.
3861  *
3862  * We are not allowed to take the i_mutex here so we have to play games to
3863  * protect against truncate races as the page could now be beyond EOF.  Because
3864  * vmtruncate() writes the inode size before removing pages, once we have the
3865  * page lock we can determine safely if the page is beyond EOF. If it is not
3866  * beyond EOF, then the page is guaranteed safe against truncation until we
3867  * unlock the page.
3868  */
3869 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3870 {
3871         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3872         struct btrfs_root *root = BTRFS_I(inode)->root;
3873         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3874         struct btrfs_ordered_extent *ordered;
3875         char *kaddr;
3876         unsigned long zero_start;
3877         loff_t size;
3878         int ret;
3879         u64 page_start;
3880         u64 page_end;
3881
3882         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3883         if (ret)
3884                 goto out;
3885
3886         ret = -EINVAL;
3887 again:
3888         lock_page(page);
3889         size = i_size_read(inode);
3890         page_start = page_offset(page);
3891         page_end = page_start + PAGE_CACHE_SIZE - 1;
3892
3893         if ((page->mapping != inode->i_mapping) ||
3894             (page_start >= size)) {
3895                 /* page got truncated out from underneath us */
3896                 goto out_unlock;
3897         }
3898         wait_on_page_writeback(page);
3899
3900         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3901         set_page_extent_mapped(page);
3902
3903         /*
3904          * we can't set the delalloc bits if there are pending ordered
3905          * extents.  Drop our locks and wait for them to finish
3906          */
3907         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3908         if (ordered) {
3909                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3910                 unlock_page(page);
3911                 btrfs_start_ordered_extent(inode, ordered, 1);
3912                 btrfs_put_ordered_extent(ordered);
3913                 goto again;
3914         }
3915
3916         btrfs_set_extent_delalloc(inode, page_start, page_end);
3917         ret = 0;
3918
3919         /* page is wholly or partially inside EOF */
3920         if (page_start + PAGE_CACHE_SIZE > size)
3921                 zero_start = size & ~PAGE_CACHE_MASK;
3922         else
3923                 zero_start = PAGE_CACHE_SIZE;
3924
3925         if (zero_start != PAGE_CACHE_SIZE) {
3926                 kaddr = kmap(page);
3927                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3928                 flush_dcache_page(page);
3929                 kunmap(page);
3930         }
3931         ClearPageChecked(page);
3932         set_page_dirty(page);
3933         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3934
3935 out_unlock:
3936         unlock_page(page);
3937 out:
3938         return ret;
3939 }
3940
3941 static void btrfs_truncate(struct inode *inode)
3942 {
3943         struct btrfs_root *root = BTRFS_I(inode)->root;
3944         int ret;
3945         struct btrfs_trans_handle *trans;
3946         unsigned long nr;
3947         u64 mask = root->sectorsize - 1;
3948
3949         if (!S_ISREG(inode->i_mode))
3950                 return;
3951         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3952                 return;
3953
3954         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3955         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3956
3957         trans = btrfs_start_transaction(root, 1);
3958         btrfs_set_trans_block_group(trans, inode);
3959         btrfs_i_size_write(inode, inode->i_size);
3960
3961         ret = btrfs_orphan_add(trans, inode);
3962         if (ret)
3963                 goto out;
3964         /* FIXME, add redo link to tree so we don't leak on crash */
3965         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3966                                       BTRFS_EXTENT_DATA_KEY);
3967         btrfs_update_inode(trans, root, inode);
3968
3969         ret = btrfs_orphan_del(trans, inode);
3970         BUG_ON(ret);
3971
3972 out:
3973         nr = trans->blocks_used;
3974         ret = btrfs_end_transaction_throttle(trans, root);
3975         BUG_ON(ret);
3976         btrfs_btree_balance_dirty(root, nr);
3977 }
3978
3979 /*
3980  * Invalidate a single dcache entry at the root of the filesystem.
3981  * Needed after creation of snapshot or subvolume.
3982  */
3983 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3984                                   int namelen)
3985 {
3986         struct dentry *alias, *entry;
3987         struct qstr qstr;
3988
3989         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3990         if (alias) {
3991                 qstr.name = name;
3992                 qstr.len = namelen;
3993                 /* change me if btrfs ever gets a d_hash operation */
3994                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3995                 entry = d_lookup(alias, &qstr);
3996                 dput(alias);
3997                 if (entry) {
3998                         d_invalidate(entry);
3999                         dput(entry);
4000                 }
4001         }
4002 }
4003
4004 /*
4005  * create a new subvolume directory/inode (helper for the ioctl).
4006  */
4007 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4008                 struct btrfs_trans_handle *trans, u64 new_dirid,
4009                 struct btrfs_block_group_cache *block_group)
4010 {
4011         struct inode *inode;
4012         int error;
4013         u64 index = 0;
4014
4015         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4016                                 new_dirid, block_group, S_IFDIR | 0700, &index);
4017         if (IS_ERR(inode))
4018                 return PTR_ERR(inode);
4019         inode->i_op = &btrfs_dir_inode_operations;
4020         inode->i_fop = &btrfs_dir_file_operations;
4021         new_root->inode = inode;
4022
4023         inode->i_nlink = 1;
4024         btrfs_i_size_write(inode, 0);
4025
4026         error = btrfs_update_inode(trans, new_root, inode);
4027         if (error)
4028                 return error;
4029
4030         atomic_inc(&inode->i_count);
4031         d_instantiate(dentry, inode);
4032         return 0;
4033 }
4034
4035 /* helper function for file defrag and space balancing.  This
4036  * forces readahead on a given range of bytes in an inode
4037  */
4038 unsigned long btrfs_force_ra(struct address_space *mapping,
4039                               struct file_ra_state *ra, struct file *file,
4040                               pgoff_t offset, pgoff_t last_index)
4041 {
4042         pgoff_t req_size = last_index - offset + 1;
4043
4044         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4045         return offset + req_size;
4046 }
4047
4048 struct inode *btrfs_alloc_inode(struct super_block *sb)
4049 {
4050         struct btrfs_inode *ei;
4051
4052         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4053         if (!ei)
4054                 return NULL;
4055         ei->last_trans = 0;
4056         ei->logged_trans = 0;
4057         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4058         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4059         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4060         INIT_LIST_HEAD(&ei->i_orphan);
4061         return &ei->vfs_inode;
4062 }
4063
4064 void btrfs_destroy_inode(struct inode *inode)
4065 {
4066         struct btrfs_ordered_extent *ordered;
4067         WARN_ON(!list_empty(&inode->i_dentry));
4068         WARN_ON(inode->i_data.nrpages);
4069
4070         if (BTRFS_I(inode)->i_acl &&
4071             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4072                 posix_acl_release(BTRFS_I(inode)->i_acl);
4073         if (BTRFS_I(inode)->i_default_acl &&
4074             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4075                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4076
4077         spin_lock(&BTRFS_I(inode)->root->list_lock);
4078         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4079                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4080                        " list\n", inode->i_ino);
4081                 dump_stack();
4082         }
4083         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4084
4085         while(1) {
4086                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4087                 if (!ordered)
4088                         break;
4089                 else {
4090                         printk("found ordered extent %Lu %Lu\n",
4091                                ordered->file_offset, ordered->len);
4092                         btrfs_remove_ordered_extent(inode, ordered);
4093                         btrfs_put_ordered_extent(ordered);
4094                         btrfs_put_ordered_extent(ordered);
4095                 }
4096         }
4097         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4098         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4099 }
4100
4101 static void init_once(void *foo)
4102 {
4103         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4104
4105         inode_init_once(&ei->vfs_inode);
4106 }
4107
4108 void btrfs_destroy_cachep(void)
4109 {
4110         if (btrfs_inode_cachep)
4111                 kmem_cache_destroy(btrfs_inode_cachep);
4112         if (btrfs_trans_handle_cachep)
4113                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4114         if (btrfs_transaction_cachep)
4115                 kmem_cache_destroy(btrfs_transaction_cachep);
4116         if (btrfs_bit_radix_cachep)
4117                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4118         if (btrfs_path_cachep)
4119                 kmem_cache_destroy(btrfs_path_cachep);
4120 }
4121
4122 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4123                                        unsigned long extra_flags,
4124                                        void (*ctor)(void *))
4125 {
4126         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4127                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4128 }
4129
4130 int btrfs_init_cachep(void)
4131 {
4132         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4133                                           sizeof(struct btrfs_inode),
4134                                           0, init_once);
4135         if (!btrfs_inode_cachep)
4136                 goto fail;
4137         btrfs_trans_handle_cachep =
4138                         btrfs_cache_create("btrfs_trans_handle_cache",
4139                                            sizeof(struct btrfs_trans_handle),
4140                                            0, NULL);
4141         if (!btrfs_trans_handle_cachep)
4142                 goto fail;
4143         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4144                                              sizeof(struct btrfs_transaction),
4145                                              0, NULL);
4146         if (!btrfs_transaction_cachep)
4147                 goto fail;
4148         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4149                                          sizeof(struct btrfs_path),
4150                                          0, NULL);
4151         if (!btrfs_path_cachep)
4152                 goto fail;
4153         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4154                                               SLAB_DESTROY_BY_RCU, NULL);
4155         if (!btrfs_bit_radix_cachep)
4156                 goto fail;
4157         return 0;
4158 fail:
4159         btrfs_destroy_cachep();
4160         return -ENOMEM;
4161 }
4162
4163 static int btrfs_getattr(struct vfsmount *mnt,
4164                          struct dentry *dentry, struct kstat *stat)
4165 {
4166         struct inode *inode = dentry->d_inode;
4167         generic_fillattr(inode, stat);
4168         stat->blksize = PAGE_CACHE_SIZE;
4169         stat->blocks = (inode_get_bytes(inode) +
4170                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4171         return 0;
4172 }
4173
4174 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4175                            struct inode * new_dir,struct dentry *new_dentry)
4176 {
4177         struct btrfs_trans_handle *trans;
4178         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4179         struct inode *new_inode = new_dentry->d_inode;
4180         struct inode *old_inode = old_dentry->d_inode;
4181         struct timespec ctime = CURRENT_TIME;
4182         u64 index = 0;
4183         int ret;
4184
4185         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4186             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4187                 return -ENOTEMPTY;
4188         }
4189
4190         ret = btrfs_check_free_space(root, 1, 0);
4191         if (ret)
4192                 goto out_unlock;
4193
4194         trans = btrfs_start_transaction(root, 1);
4195
4196         btrfs_set_trans_block_group(trans, new_dir);
4197
4198         btrfs_inc_nlink(old_dentry->d_inode);
4199         old_dir->i_ctime = old_dir->i_mtime = ctime;
4200         new_dir->i_ctime = new_dir->i_mtime = ctime;
4201         old_inode->i_ctime = ctime;
4202
4203         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4204                                  old_dentry->d_name.name,
4205                                  old_dentry->d_name.len);
4206         if (ret)
4207                 goto out_fail;
4208
4209         if (new_inode) {
4210                 new_inode->i_ctime = CURRENT_TIME;
4211                 ret = btrfs_unlink_inode(trans, root, new_dir,
4212                                          new_dentry->d_inode,
4213                                          new_dentry->d_name.name,
4214                                          new_dentry->d_name.len);
4215                 if (ret)
4216                         goto out_fail;
4217                 if (new_inode->i_nlink == 0) {
4218                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4219                         if (ret)
4220                                 goto out_fail;
4221                 }
4222
4223         }
4224         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
4225         if (ret)
4226                 goto out_fail;
4227
4228         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4229                              old_inode, new_dentry->d_name.name,
4230                              new_dentry->d_name.len, 1, index);
4231         if (ret)
4232                 goto out_fail;
4233
4234 out_fail:
4235         btrfs_end_transaction_throttle(trans, root);
4236 out_unlock:
4237         return ret;
4238 }
4239
4240 /*
4241  * some fairly slow code that needs optimization. This walks the list
4242  * of all the inodes with pending delalloc and forces them to disk.
4243  */
4244 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4245 {
4246         struct list_head *head = &root->fs_info->delalloc_inodes;
4247         struct btrfs_inode *binode;
4248         struct inode *inode;
4249         unsigned long flags;
4250
4251         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4252         while(!list_empty(head)) {
4253                 binode = list_entry(head->next, struct btrfs_inode,
4254                                     delalloc_inodes);
4255                 inode = igrab(&binode->vfs_inode);
4256                 if (!inode)
4257                         list_del_init(&binode->delalloc_inodes);
4258                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4259                 if (inode) {
4260                         filemap_flush(inode->i_mapping);
4261                         iput(inode);
4262                 }
4263                 cond_resched();
4264                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4265         }
4266         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4267
4268         /* the filemap_flush will queue IO into the worker threads, but
4269          * we have to make sure the IO is actually started and that
4270          * ordered extents get created before we return
4271          */
4272         atomic_inc(&root->fs_info->async_submit_draining);
4273         while(atomic_read(&root->fs_info->nr_async_submits)) {
4274                 wait_event(root->fs_info->async_submit_wait,
4275                    (atomic_read(&root->fs_info->nr_async_submits) == 0));
4276         }
4277         atomic_dec(&root->fs_info->async_submit_draining);
4278         return 0;
4279 }
4280
4281 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4282                          const char *symname)
4283 {
4284         struct btrfs_trans_handle *trans;
4285         struct btrfs_root *root = BTRFS_I(dir)->root;
4286         struct btrfs_path *path;
4287         struct btrfs_key key;
4288         struct inode *inode = NULL;
4289         int err;
4290         int drop_inode = 0;
4291         u64 objectid;
4292         u64 index = 0 ;
4293         int name_len;
4294         int datasize;
4295         unsigned long ptr;
4296         struct btrfs_file_extent_item *ei;
4297         struct extent_buffer *leaf;
4298         unsigned long nr = 0;
4299
4300         name_len = strlen(symname) + 1;
4301         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4302                 return -ENAMETOOLONG;
4303
4304         err = btrfs_check_free_space(root, 1, 0);
4305         if (err)
4306                 goto out_fail;
4307
4308         trans = btrfs_start_transaction(root, 1);
4309         btrfs_set_trans_block_group(trans, dir);
4310
4311         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4312         if (err) {
4313                 err = -ENOSPC;
4314                 goto out_unlock;
4315         }
4316
4317         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4318                                 dentry->d_name.len,
4319                                 dentry->d_parent->d_inode->i_ino, objectid,
4320                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4321                                 &index);
4322         err = PTR_ERR(inode);
4323         if (IS_ERR(inode))
4324                 goto out_unlock;
4325
4326         err = btrfs_init_acl(inode, dir);
4327         if (err) {
4328                 drop_inode = 1;
4329                 goto out_unlock;
4330         }
4331
4332         btrfs_set_trans_block_group(trans, inode);
4333         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4334         if (err)
4335                 drop_inode = 1;
4336         else {
4337                 inode->i_mapping->a_ops = &btrfs_aops;
4338                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4339                 inode->i_fop = &btrfs_file_operations;
4340                 inode->i_op = &btrfs_file_inode_operations;
4341                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4342         }
4343         dir->i_sb->s_dirt = 1;
4344         btrfs_update_inode_block_group(trans, inode);
4345         btrfs_update_inode_block_group(trans, dir);
4346         if (drop_inode)
4347                 goto out_unlock;
4348
4349         path = btrfs_alloc_path();
4350         BUG_ON(!path);
4351         key.objectid = inode->i_ino;
4352         key.offset = 0;
4353         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4354         datasize = btrfs_file_extent_calc_inline_size(name_len);
4355         err = btrfs_insert_empty_item(trans, root, path, &key,
4356                                       datasize);
4357         if (err) {
4358                 drop_inode = 1;
4359                 goto out_unlock;
4360         }
4361         leaf = path->nodes[0];
4362         ei = btrfs_item_ptr(leaf, path->slots[0],
4363                             struct btrfs_file_extent_item);
4364         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4365         btrfs_set_file_extent_type(leaf, ei,
4366                                    BTRFS_FILE_EXTENT_INLINE);
4367         btrfs_set_file_extent_encryption(leaf, ei, 0);
4368         btrfs_set_file_extent_compression(leaf, ei, 0);
4369         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4370         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4371
4372         ptr = btrfs_file_extent_inline_start(ei);
4373         write_extent_buffer(leaf, symname, ptr, name_len);
4374         btrfs_mark_buffer_dirty(leaf);
4375         btrfs_free_path(path);
4376
4377         inode->i_op = &btrfs_symlink_inode_operations;
4378         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4379         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4380         inode_set_bytes(inode, name_len);
4381         btrfs_i_size_write(inode, name_len - 1);
4382         err = btrfs_update_inode(trans, root, inode);
4383         if (err)
4384                 drop_inode = 1;
4385
4386 out_unlock:
4387         nr = trans->blocks_used;
4388         btrfs_end_transaction_throttle(trans, root);
4389 out_fail:
4390         if (drop_inode) {
4391                 inode_dec_link_count(inode);
4392                 iput(inode);
4393         }
4394         btrfs_btree_balance_dirty(root, nr);
4395         return err;
4396 }
4397
4398 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4399                                u64 alloc_hint, int mode)
4400 {
4401         struct btrfs_trans_handle *trans;
4402         struct btrfs_root *root = BTRFS_I(inode)->root;
4403         struct btrfs_key ins;
4404         u64 alloc_size;
4405         u64 cur_offset = start;
4406         u64 num_bytes = end - start;
4407         int ret = 0;
4408
4409         trans = btrfs_join_transaction(root, 1);
4410         BUG_ON(!trans);
4411         btrfs_set_trans_block_group(trans, inode);
4412
4413         while (num_bytes > 0) {
4414                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4415                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4416                                            root->sectorsize, 0, alloc_hint,
4417                                            (u64)-1, &ins, 1);
4418                 if (ret) {
4419                         WARN_ON(1);
4420                         goto out;
4421                 }
4422                 ret = insert_reserved_file_extent(trans, inode,
4423                                                   cur_offset, ins.objectid,
4424                                                   ins.offset, ins.offset,
4425                                                   ins.offset, 0, 0, 0,
4426                                                   BTRFS_FILE_EXTENT_PREALLOC);
4427                 BUG_ON(ret);
4428                 num_bytes -= ins.offset;
4429                 cur_offset += ins.offset;
4430                 alloc_hint = ins.objectid + ins.offset;
4431         }
4432 out:
4433         if (cur_offset > start) {
4434                 inode->i_ctime = CURRENT_TIME;
4435                 btrfs_set_flag(inode, PREALLOC);
4436                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4437                     cur_offset > i_size_read(inode))
4438                         btrfs_i_size_write(inode, cur_offset);
4439                 ret = btrfs_update_inode(trans, root, inode);
4440                 BUG_ON(ret);
4441         }
4442
4443         btrfs_end_transaction(trans, root);
4444         return ret;
4445 }
4446
4447 static long btrfs_fallocate(struct inode *inode, int mode,
4448                             loff_t offset, loff_t len)
4449 {
4450         u64 cur_offset;
4451         u64 last_byte;
4452         u64 alloc_start;
4453         u64 alloc_end;
4454         u64 alloc_hint = 0;
4455         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4456         struct extent_map *em;
4457         int ret;
4458
4459         alloc_start = offset & ~mask;
4460         alloc_end =  (offset + len + mask) & ~mask;
4461
4462         mutex_lock(&inode->i_mutex);
4463         if (alloc_start > inode->i_size) {
4464                 ret = btrfs_cont_expand(inode, alloc_start);
4465                 if (ret)
4466                         goto out;
4467         }
4468
4469         while (1) {
4470                 struct btrfs_ordered_extent *ordered;
4471                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4472                             alloc_end - 1, GFP_NOFS);
4473                 ordered = btrfs_lookup_first_ordered_extent(inode,
4474                                                             alloc_end - 1);
4475                 if (ordered &&
4476                     ordered->file_offset + ordered->len > alloc_start &&
4477                     ordered->file_offset < alloc_end) {
4478                         btrfs_put_ordered_extent(ordered);
4479                         unlock_extent(&BTRFS_I(inode)->io_tree,
4480                                       alloc_start, alloc_end - 1, GFP_NOFS);
4481                         btrfs_wait_ordered_range(inode, alloc_start,
4482                                                  alloc_end - alloc_start);
4483                 } else {
4484                         if (ordered)
4485                                 btrfs_put_ordered_extent(ordered);
4486                         break;
4487                 }
4488         }
4489
4490         cur_offset = alloc_start;
4491         while (1) {
4492                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4493                                       alloc_end - cur_offset, 0);
4494                 BUG_ON(IS_ERR(em) || !em);
4495                 last_byte = min(extent_map_end(em), alloc_end);
4496                 last_byte = (last_byte + mask) & ~mask;
4497                 if (em->block_start == EXTENT_MAP_HOLE) {
4498                         ret = prealloc_file_range(inode, cur_offset,
4499                                         last_byte, alloc_hint, mode);
4500                         if (ret < 0) {
4501                                 free_extent_map(em);
4502                                 break;
4503                         }
4504                 }
4505                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4506                         alloc_hint = em->block_start;
4507                 free_extent_map(em);
4508
4509                 cur_offset = last_byte;
4510                 if (cur_offset >= alloc_end) {
4511                         ret = 0;
4512                         break;
4513                 }
4514         }
4515         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4516                       GFP_NOFS);
4517 out:
4518         mutex_unlock(&inode->i_mutex);
4519         return ret;
4520 }
4521
4522 static int btrfs_set_page_dirty(struct page *page)
4523 {
4524         return __set_page_dirty_nobuffers(page);
4525 }
4526
4527 static int btrfs_permission(struct inode *inode, int mask)
4528 {
4529         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4530                 return -EACCES;
4531         return generic_permission(inode, mask, btrfs_check_acl);
4532 }
4533
4534 static struct inode_operations btrfs_dir_inode_operations = {
4535         .lookup         = btrfs_lookup,
4536         .create         = btrfs_create,
4537         .unlink         = btrfs_unlink,
4538         .link           = btrfs_link,
4539         .mkdir          = btrfs_mkdir,
4540         .rmdir          = btrfs_rmdir,
4541         .rename         = btrfs_rename,
4542         .symlink        = btrfs_symlink,
4543         .setattr        = btrfs_setattr,
4544         .mknod          = btrfs_mknod,
4545         .setxattr       = btrfs_setxattr,
4546         .getxattr       = btrfs_getxattr,
4547         .listxattr      = btrfs_listxattr,
4548         .removexattr    = btrfs_removexattr,
4549         .permission     = btrfs_permission,
4550 };
4551 static struct inode_operations btrfs_dir_ro_inode_operations = {
4552         .lookup         = btrfs_lookup,
4553         .permission     = btrfs_permission,
4554 };
4555 static struct file_operations btrfs_dir_file_operations = {
4556         .llseek         = generic_file_llseek,
4557         .read           = generic_read_dir,
4558         .readdir        = btrfs_real_readdir,
4559         .unlocked_ioctl = btrfs_ioctl,
4560 #ifdef CONFIG_COMPAT
4561         .compat_ioctl   = btrfs_ioctl,
4562 #endif
4563         .release        = btrfs_release_file,
4564         .fsync          = btrfs_sync_file,
4565 };
4566
4567 static struct extent_io_ops btrfs_extent_io_ops = {
4568         .fill_delalloc = run_delalloc_range,
4569         .submit_bio_hook = btrfs_submit_bio_hook,
4570         .merge_bio_hook = btrfs_merge_bio_hook,
4571         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4572         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4573         .writepage_start_hook = btrfs_writepage_start_hook,
4574         .readpage_io_failed_hook = btrfs_io_failed_hook,
4575         .set_bit_hook = btrfs_set_bit_hook,
4576         .clear_bit_hook = btrfs_clear_bit_hook,
4577 };
4578
4579 static struct address_space_operations btrfs_aops = {
4580         .readpage       = btrfs_readpage,
4581         .writepage      = btrfs_writepage,
4582         .writepages     = btrfs_writepages,
4583         .readpages      = btrfs_readpages,
4584         .sync_page      = block_sync_page,
4585         .bmap           = btrfs_bmap,
4586         .direct_IO      = btrfs_direct_IO,
4587         .invalidatepage = btrfs_invalidatepage,
4588         .releasepage    = btrfs_releasepage,
4589         .set_page_dirty = btrfs_set_page_dirty,
4590 };
4591
4592 static struct address_space_operations btrfs_symlink_aops = {
4593         .readpage       = btrfs_readpage,
4594         .writepage      = btrfs_writepage,
4595         .invalidatepage = btrfs_invalidatepage,
4596         .releasepage    = btrfs_releasepage,
4597 };
4598
4599 static struct inode_operations btrfs_file_inode_operations = {
4600         .truncate       = btrfs_truncate,
4601         .getattr        = btrfs_getattr,
4602         .setattr        = btrfs_setattr,
4603         .setxattr       = btrfs_setxattr,
4604         .getxattr       = btrfs_getxattr,
4605         .listxattr      = btrfs_listxattr,
4606         .removexattr    = btrfs_removexattr,
4607         .permission     = btrfs_permission,
4608         .fallocate      = btrfs_fallocate,
4609 };
4610 static struct inode_operations btrfs_special_inode_operations = {
4611         .getattr        = btrfs_getattr,
4612         .setattr        = btrfs_setattr,
4613         .permission     = btrfs_permission,
4614         .setxattr       = btrfs_setxattr,
4615         .getxattr       = btrfs_getxattr,
4616         .listxattr      = btrfs_listxattr,
4617         .removexattr    = btrfs_removexattr,
4618 };
4619 static struct inode_operations btrfs_symlink_inode_operations = {
4620         .readlink       = generic_readlink,
4621         .follow_link    = page_follow_link_light,
4622         .put_link       = page_put_link,
4623         .permission     = btrfs_permission,
4624 };