4a79e1c5ebd07c5ffbb4c580ccf31fde3a9b9f32
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "ref-cache.h"
52 #include "compression.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static struct inode_operations btrfs_dir_inode_operations;
60 static struct inode_operations btrfs_symlink_inode_operations;
61 static struct inode_operations btrfs_dir_ro_inode_operations;
62 static struct inode_operations btrfs_special_inode_operations;
63 static struct inode_operations btrfs_file_inode_operations;
64 static struct address_space_operations btrfs_aops;
65 static struct address_space_operations btrfs_symlink_aops;
66 static struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_bit_radix_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74
75 #define S_SHIFT 12
76 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
78         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
79         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
80         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
81         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
82         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
83         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
84 };
85
86 static void btrfs_truncate(struct inode *inode);
87 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
88 static noinline int cow_file_range(struct inode *inode,
89                                    struct page *locked_page,
90                                    u64 start, u64 end, int *page_started,
91                                    unsigned long *nr_written, int unlock);
92
93 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
94 {
95         int err;
96
97         err = btrfs_init_acl(inode, dir);
98         if (!err)
99                 err = btrfs_xattr_security_init(inode, dir);
100         return err;
101 }
102
103 /*
104  * a very lame attempt at stopping writes when the FS is 85% full.  There
105  * are countless ways this is incorrect, but it is better than nothing.
106  */
107 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
108                            int for_del)
109 {
110         u64 total;
111         u64 used;
112         u64 thresh;
113         int ret = 0;
114
115         spin_lock(&root->fs_info->delalloc_lock);
116         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
117         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
118         if (for_del)
119                 thresh = total * 90;
120         else
121                 thresh = total * 85;
122
123         do_div(thresh, 100);
124
125         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
126                 ret = -ENOSPC;
127         spin_unlock(&root->fs_info->delalloc_lock);
128         return ret;
129 }
130
131 /*
132  * this does all the hard work for inserting an inline extent into
133  * the btree.  The caller should have done a btrfs_drop_extents so that
134  * no overlapping inline items exist in the btree
135  */
136 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
137                                 struct btrfs_root *root, struct inode *inode,
138                                 u64 start, size_t size, size_t compressed_size,
139                                 struct page **compressed_pages)
140 {
141         struct btrfs_key key;
142         struct btrfs_path *path;
143         struct extent_buffer *leaf;
144         struct page *page = NULL;
145         char *kaddr;
146         unsigned long ptr;
147         struct btrfs_file_extent_item *ei;
148         int err = 0;
149         int ret;
150         size_t cur_size = size;
151         size_t datasize;
152         unsigned long offset;
153         int use_compress = 0;
154
155         if (compressed_size && compressed_pages) {
156                 use_compress = 1;
157                 cur_size = compressed_size;
158         }
159
160         path = btrfs_alloc_path();
161         if (!path)
162                 return -ENOMEM;
163
164         btrfs_set_trans_block_group(trans, inode);
165
166         key.objectid = inode->i_ino;
167         key.offset = start;
168         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
169         datasize = btrfs_file_extent_calc_inline_size(cur_size);
170
171         inode_add_bytes(inode, size);
172         ret = btrfs_insert_empty_item(trans, root, path, &key,
173                                       datasize);
174         BUG_ON(ret);
175         if (ret) {
176                 err = ret;
177                 goto fail;
178         }
179         leaf = path->nodes[0];
180         ei = btrfs_item_ptr(leaf, path->slots[0],
181                             struct btrfs_file_extent_item);
182         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
183         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
184         btrfs_set_file_extent_encryption(leaf, ei, 0);
185         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
186         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
187         ptr = btrfs_file_extent_inline_start(ei);
188
189         if (use_compress) {
190                 struct page *cpage;
191                 int i = 0;
192                 while (compressed_size > 0) {
193                         cpage = compressed_pages[i];
194                         cur_size = min_t(unsigned long, compressed_size,
195                                        PAGE_CACHE_SIZE);
196
197                         kaddr = kmap(cpage);
198                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
199                         kunmap(cpage);
200
201                         i++;
202                         ptr += cur_size;
203                         compressed_size -= cur_size;
204                 }
205                 btrfs_set_file_extent_compression(leaf, ei,
206                                                   BTRFS_COMPRESS_ZLIB);
207         } else {
208                 page = find_get_page(inode->i_mapping,
209                                      start >> PAGE_CACHE_SHIFT);
210                 btrfs_set_file_extent_compression(leaf, ei, 0);
211                 kaddr = kmap_atomic(page, KM_USER0);
212                 offset = start & (PAGE_CACHE_SIZE - 1);
213                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
214                 kunmap_atomic(kaddr, KM_USER0);
215                 page_cache_release(page);
216         }
217         btrfs_mark_buffer_dirty(leaf);
218         btrfs_free_path(path);
219
220         BTRFS_I(inode)->disk_i_size = inode->i_size;
221         btrfs_update_inode(trans, root, inode);
222         return 0;
223 fail:
224         btrfs_free_path(path);
225         return err;
226 }
227
228
229 /*
230  * conditionally insert an inline extent into the file.  This
231  * does the checks required to make sure the data is small enough
232  * to fit as an inline extent.
233  */
234 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
235                                  struct btrfs_root *root,
236                                  struct inode *inode, u64 start, u64 end,
237                                  size_t compressed_size,
238                                  struct page **compressed_pages)
239 {
240         u64 isize = i_size_read(inode);
241         u64 actual_end = min(end + 1, isize);
242         u64 inline_len = actual_end - start;
243         u64 aligned_end = (end + root->sectorsize - 1) &
244                         ~((u64)root->sectorsize - 1);
245         u64 hint_byte;
246         u64 data_len = inline_len;
247         int ret;
248
249         if (compressed_size)
250                 data_len = compressed_size;
251
252         if (start > 0 ||
253             actual_end >= PAGE_CACHE_SIZE ||
254             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
255             (!compressed_size &&
256             (actual_end & (root->sectorsize - 1)) == 0) ||
257             end + 1 < isize ||
258             data_len > root->fs_info->max_inline) {
259                 return 1;
260         }
261
262         ret = btrfs_drop_extents(trans, root, inode, start,
263                                  aligned_end, start, &hint_byte);
264         BUG_ON(ret);
265
266         if (isize > actual_end)
267                 inline_len = min_t(u64, isize, actual_end);
268         ret = insert_inline_extent(trans, root, inode, start,
269                                    inline_len, compressed_size,
270                                    compressed_pages);
271         BUG_ON(ret);
272         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
273         return 0;
274 }
275
276 struct async_extent {
277         u64 start;
278         u64 ram_size;
279         u64 compressed_size;
280         struct page **pages;
281         unsigned long nr_pages;
282         struct list_head list;
283 };
284
285 struct async_cow {
286         struct inode *inode;
287         struct btrfs_root *root;
288         struct page *locked_page;
289         u64 start;
290         u64 end;
291         struct list_head extents;
292         struct btrfs_work work;
293 };
294
295 static noinline int add_async_extent(struct async_cow *cow,
296                                      u64 start, u64 ram_size,
297                                      u64 compressed_size,
298                                      struct page **pages,
299                                      unsigned long nr_pages)
300 {
301         struct async_extent *async_extent;
302
303         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
304         async_extent->start = start;
305         async_extent->ram_size = ram_size;
306         async_extent->compressed_size = compressed_size;
307         async_extent->pages = pages;
308         async_extent->nr_pages = nr_pages;
309         list_add_tail(&async_extent->list, &cow->extents);
310         return 0;
311 }
312
313 /*
314  * we create compressed extents in two phases.  The first
315  * phase compresses a range of pages that have already been
316  * locked (both pages and state bits are locked).
317  *
318  * This is done inside an ordered work queue, and the compression
319  * is spread across many cpus.  The actual IO submission is step
320  * two, and the ordered work queue takes care of making sure that
321  * happens in the same order things were put onto the queue by
322  * writepages and friends.
323  *
324  * If this code finds it can't get good compression, it puts an
325  * entry onto the work queue to write the uncompressed bytes.  This
326  * makes sure that both compressed inodes and uncompressed inodes
327  * are written in the same order that pdflush sent them down.
328  */
329 static noinline int compress_file_range(struct inode *inode,
330                                         struct page *locked_page,
331                                         u64 start, u64 end,
332                                         struct async_cow *async_cow,
333                                         int *num_added)
334 {
335         struct btrfs_root *root = BTRFS_I(inode)->root;
336         struct btrfs_trans_handle *trans;
337         u64 num_bytes;
338         u64 orig_start;
339         u64 disk_num_bytes;
340         u64 blocksize = root->sectorsize;
341         u64 actual_end;
342         u64 isize = i_size_read(inode);
343         int ret = 0;
344         struct page **pages = NULL;
345         unsigned long nr_pages;
346         unsigned long nr_pages_ret = 0;
347         unsigned long total_compressed = 0;
348         unsigned long total_in = 0;
349         unsigned long max_compressed = 128 * 1024;
350         unsigned long max_uncompressed = 128 * 1024;
351         int i;
352         int will_compress;
353
354         orig_start = start;
355
356         actual_end = min_t(u64, isize, end + 1);
357 again:
358         will_compress = 0;
359         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
360         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
361
362         total_compressed = actual_end - start;
363
364         /* we want to make sure that amount of ram required to uncompress
365          * an extent is reasonable, so we limit the total size in ram
366          * of a compressed extent to 128k.  This is a crucial number
367          * because it also controls how easily we can spread reads across
368          * cpus for decompression.
369          *
370          * We also want to make sure the amount of IO required to do
371          * a random read is reasonably small, so we limit the size of
372          * a compressed extent to 128k.
373          */
374         total_compressed = min(total_compressed, max_uncompressed);
375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376         num_bytes = max(blocksize,  num_bytes);
377         disk_num_bytes = num_bytes;
378         total_in = 0;
379         ret = 0;
380
381         /*
382          * we do compression for mount -o compress and when the
383          * inode has not been flagged as nocompress.  This flag can
384          * change at any time if we discover bad compression ratios.
385          */
386         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
387             btrfs_test_opt(root, COMPRESS)) {
388                 WARN_ON(pages);
389                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
390
391                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
392                                                 total_compressed, pages,
393                                                 nr_pages, &nr_pages_ret,
394                                                 &total_in,
395                                                 &total_compressed,
396                                                 max_compressed);
397
398                 if (!ret) {
399                         unsigned long offset = total_compressed &
400                                 (PAGE_CACHE_SIZE - 1);
401                         struct page *page = pages[nr_pages_ret - 1];
402                         char *kaddr;
403
404                         /* zero the tail end of the last page, we might be
405                          * sending it down to disk
406                          */
407                         if (offset) {
408                                 kaddr = kmap_atomic(page, KM_USER0);
409                                 memset(kaddr + offset, 0,
410                                        PAGE_CACHE_SIZE - offset);
411                                 kunmap_atomic(kaddr, KM_USER0);
412                         }
413                         will_compress = 1;
414                 }
415         }
416         if (start == 0) {
417                 trans = btrfs_join_transaction(root, 1);
418                 BUG_ON(!trans);
419                 btrfs_set_trans_block_group(trans, inode);
420
421                 /* lets try to make an inline extent */
422                 if (ret || total_in < (actual_end - start)) {
423                         /* we didn't compress the entire range, try
424                          * to make an uncompressed inline extent.
425                          */
426                         ret = cow_file_range_inline(trans, root, inode,
427                                                     start, end, 0, NULL);
428                 } else {
429                         /* try making a compressed inline extent */
430                         ret = cow_file_range_inline(trans, root, inode,
431                                                     start, end,
432                                                     total_compressed, pages);
433                 }
434                 btrfs_end_transaction(trans, root);
435                 if (ret == 0) {
436                         /*
437                          * inline extent creation worked, we don't need
438                          * to create any more async work items.  Unlock
439                          * and free up our temp pages.
440                          */
441                         extent_clear_unlock_delalloc(inode,
442                                                      &BTRFS_I(inode)->io_tree,
443                                                      start, end, NULL, 1, 0,
444                                                      0, 1, 1, 1);
445                         ret = 0;
446                         goto free_pages_out;
447                 }
448         }
449
450         if (will_compress) {
451                 /*
452                  * we aren't doing an inline extent round the compressed size
453                  * up to a block size boundary so the allocator does sane
454                  * things
455                  */
456                 total_compressed = (total_compressed + blocksize - 1) &
457                         ~(blocksize - 1);
458
459                 /*
460                  * one last check to make sure the compression is really a
461                  * win, compare the page count read with the blocks on disk
462                  */
463                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
464                         ~(PAGE_CACHE_SIZE - 1);
465                 if (total_compressed >= total_in) {
466                         will_compress = 0;
467                 } else {
468                         disk_num_bytes = total_compressed;
469                         num_bytes = total_in;
470                 }
471         }
472         if (!will_compress && pages) {
473                 /*
474                  * the compression code ran but failed to make things smaller,
475                  * free any pages it allocated and our page pointer array
476                  */
477                 for (i = 0; i < nr_pages_ret; i++) {
478                         WARN_ON(pages[i]->mapping);
479                         page_cache_release(pages[i]);
480                 }
481                 kfree(pages);
482                 pages = NULL;
483                 total_compressed = 0;
484                 nr_pages_ret = 0;
485
486                 /* flag the file so we don't compress in the future */
487                 btrfs_set_flag(inode, NOCOMPRESS);
488         }
489         if (will_compress) {
490                 *num_added += 1;
491
492                 /* the async work queues will take care of doing actual
493                  * allocation on disk for these compressed pages,
494                  * and will submit them to the elevator.
495                  */
496                 add_async_extent(async_cow, start, num_bytes,
497                                  total_compressed, pages, nr_pages_ret);
498
499                 if (start + num_bytes < end && start + num_bytes < actual_end) {
500                         start += num_bytes;
501                         pages = NULL;
502                         cond_resched();
503                         goto again;
504                 }
505         } else {
506                 /*
507                  * No compression, but we still need to write the pages in
508                  * the file we've been given so far.  redirty the locked
509                  * page if it corresponds to our extent and set things up
510                  * for the async work queue to run cow_file_range to do
511                  * the normal delalloc dance
512                  */
513                 if (page_offset(locked_page) >= start &&
514                     page_offset(locked_page) <= end) {
515                         __set_page_dirty_nobuffers(locked_page);
516                         /* unlocked later on in the async handlers */
517                 }
518                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
519                 *num_added += 1;
520         }
521
522 out:
523         return 0;
524
525 free_pages_out:
526         for (i = 0; i < nr_pages_ret; i++) {
527                 WARN_ON(pages[i]->mapping);
528                 page_cache_release(pages[i]);
529         }
530         kfree(pages);
531
532         goto out;
533 }
534
535 /*
536  * phase two of compressed writeback.  This is the ordered portion
537  * of the code, which only gets called in the order the work was
538  * queued.  We walk all the async extents created by compress_file_range
539  * and send them down to the disk.
540  */
541 static noinline int submit_compressed_extents(struct inode *inode,
542                                               struct async_cow *async_cow)
543 {
544         struct async_extent *async_extent;
545         u64 alloc_hint = 0;
546         struct btrfs_trans_handle *trans;
547         struct btrfs_key ins;
548         struct extent_map *em;
549         struct btrfs_root *root = BTRFS_I(inode)->root;
550         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
551         struct extent_io_tree *io_tree;
552         int ret;
553
554         if (list_empty(&async_cow->extents))
555                 return 0;
556
557         trans = btrfs_join_transaction(root, 1);
558
559         while (!list_empty(&async_cow->extents)) {
560                 async_extent = list_entry(async_cow->extents.next,
561                                           struct async_extent, list);
562                 list_del(&async_extent->list);
563
564                 io_tree = &BTRFS_I(inode)->io_tree;
565
566                 /* did the compression code fall back to uncompressed IO? */
567                 if (!async_extent->pages) {
568                         int page_started = 0;
569                         unsigned long nr_written = 0;
570
571                         lock_extent(io_tree, async_extent->start,
572                                     async_extent->start +
573                                     async_extent->ram_size - 1, GFP_NOFS);
574
575                         /* allocate blocks */
576                         cow_file_range(inode, async_cow->locked_page,
577                                        async_extent->start,
578                                        async_extent->start +
579                                        async_extent->ram_size - 1,
580                                        &page_started, &nr_written, 0);
581
582                         /*
583                          * if page_started, cow_file_range inserted an
584                          * inline extent and took care of all the unlocking
585                          * and IO for us.  Otherwise, we need to submit
586                          * all those pages down to the drive.
587                          */
588                         if (!page_started)
589                                 extent_write_locked_range(io_tree,
590                                                   inode, async_extent->start,
591                                                   async_extent->start +
592                                                   async_extent->ram_size - 1,
593                                                   btrfs_get_extent,
594                                                   WB_SYNC_ALL);
595                         kfree(async_extent);
596                         cond_resched();
597                         continue;
598                 }
599
600                 lock_extent(io_tree, async_extent->start,
601                             async_extent->start + async_extent->ram_size - 1,
602                             GFP_NOFS);
603                 /*
604                  * here we're doing allocation and writeback of the
605                  * compressed pages
606                  */
607                 btrfs_drop_extent_cache(inode, async_extent->start,
608                                         async_extent->start +
609                                         async_extent->ram_size - 1, 0);
610
611                 ret = btrfs_reserve_extent(trans, root,
612                                            async_extent->compressed_size,
613                                            async_extent->compressed_size,
614                                            0, alloc_hint,
615                                            (u64)-1, &ins, 1);
616                 BUG_ON(ret);
617                 em = alloc_extent_map(GFP_NOFS);
618                 em->start = async_extent->start;
619                 em->len = async_extent->ram_size;
620                 em->orig_start = em->start;
621
622                 em->block_start = ins.objectid;
623                 em->block_len = ins.offset;
624                 em->bdev = root->fs_info->fs_devices->latest_bdev;
625                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
626                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
627
628                 while (1) {
629                         spin_lock(&em_tree->lock);
630                         ret = add_extent_mapping(em_tree, em);
631                         spin_unlock(&em_tree->lock);
632                         if (ret != -EEXIST) {
633                                 free_extent_map(em);
634                                 break;
635                         }
636                         btrfs_drop_extent_cache(inode, async_extent->start,
637                                                 async_extent->start +
638                                                 async_extent->ram_size - 1, 0);
639                 }
640
641                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
642                                                ins.objectid,
643                                                async_extent->ram_size,
644                                                ins.offset,
645                                                BTRFS_ORDERED_COMPRESSED);
646                 BUG_ON(ret);
647
648                 btrfs_end_transaction(trans, root);
649
650                 /*
651                  * clear dirty, set writeback and unlock the pages.
652                  */
653                 extent_clear_unlock_delalloc(inode,
654                                              &BTRFS_I(inode)->io_tree,
655                                              async_extent->start,
656                                              async_extent->start +
657                                              async_extent->ram_size - 1,
658                                              NULL, 1, 1, 0, 1, 1, 0);
659
660                 ret = btrfs_submit_compressed_write(inode,
661                                     async_extent->start,
662                                     async_extent->ram_size,
663                                     ins.objectid,
664                                     ins.offset, async_extent->pages,
665                                     async_extent->nr_pages);
666
667                 BUG_ON(ret);
668                 trans = btrfs_join_transaction(root, 1);
669                 alloc_hint = ins.objectid + ins.offset;
670                 kfree(async_extent);
671                 cond_resched();
672         }
673
674         btrfs_end_transaction(trans, root);
675         return 0;
676 }
677
678 /*
679  * when extent_io.c finds a delayed allocation range in the file,
680  * the call backs end up in this code.  The basic idea is to
681  * allocate extents on disk for the range, and create ordered data structs
682  * in ram to track those extents.
683  *
684  * locked_page is the page that writepage had locked already.  We use
685  * it to make sure we don't do extra locks or unlocks.
686  *
687  * *page_started is set to one if we unlock locked_page and do everything
688  * required to start IO on it.  It may be clean and already done with
689  * IO when we return.
690  */
691 static noinline int cow_file_range(struct inode *inode,
692                                    struct page *locked_page,
693                                    u64 start, u64 end, int *page_started,
694                                    unsigned long *nr_written,
695                                    int unlock)
696 {
697         struct btrfs_root *root = BTRFS_I(inode)->root;
698         struct btrfs_trans_handle *trans;
699         u64 alloc_hint = 0;
700         u64 num_bytes;
701         unsigned long ram_size;
702         u64 disk_num_bytes;
703         u64 cur_alloc_size;
704         u64 blocksize = root->sectorsize;
705         u64 actual_end;
706         u64 isize = i_size_read(inode);
707         struct btrfs_key ins;
708         struct extent_map *em;
709         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
710         int ret = 0;
711
712         trans = btrfs_join_transaction(root, 1);
713         BUG_ON(!trans);
714         btrfs_set_trans_block_group(trans, inode);
715
716         actual_end = min_t(u64, isize, end + 1);
717
718         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
719         num_bytes = max(blocksize,  num_bytes);
720         disk_num_bytes = num_bytes;
721         ret = 0;
722
723         if (start == 0) {
724                 /* lets try to make an inline extent */
725                 ret = cow_file_range_inline(trans, root, inode,
726                                             start, end, 0, NULL);
727                 if (ret == 0) {
728                         extent_clear_unlock_delalloc(inode,
729                                                      &BTRFS_I(inode)->io_tree,
730                                                      start, end, NULL, 1, 1,
731                                                      1, 1, 1, 1);
732                         *nr_written = *nr_written +
733                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
734                         *page_started = 1;
735                         ret = 0;
736                         goto out;
737                 }
738         }
739
740         BUG_ON(disk_num_bytes >
741                btrfs_super_total_bytes(&root->fs_info->super_copy));
742
743         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
744
745         while (disk_num_bytes > 0) {
746                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
747                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
748                                            root->sectorsize, 0, alloc_hint,
749                                            (u64)-1, &ins, 1);
750                 BUG_ON(ret);
751
752                 em = alloc_extent_map(GFP_NOFS);
753                 em->start = start;
754                 em->orig_start = em->start;
755
756                 ram_size = ins.offset;
757                 em->len = ins.offset;
758
759                 em->block_start = ins.objectid;
760                 em->block_len = ins.offset;
761                 em->bdev = root->fs_info->fs_devices->latest_bdev;
762                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
763
764                 while (1) {
765                         spin_lock(&em_tree->lock);
766                         ret = add_extent_mapping(em_tree, em);
767                         spin_unlock(&em_tree->lock);
768                         if (ret != -EEXIST) {
769                                 free_extent_map(em);
770                                 break;
771                         }
772                         btrfs_drop_extent_cache(inode, start,
773                                                 start + ram_size - 1, 0);
774                 }
775
776                 cur_alloc_size = ins.offset;
777                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
778                                                ram_size, cur_alloc_size, 0);
779                 BUG_ON(ret);
780
781                 if (root->root_key.objectid ==
782                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
783                         ret = btrfs_reloc_clone_csums(inode, start,
784                                                       cur_alloc_size);
785                         BUG_ON(ret);
786                 }
787
788                 if (disk_num_bytes < cur_alloc_size)
789                         break;
790
791                 /* we're not doing compressed IO, don't unlock the first
792                  * page (which the caller expects to stay locked), don't
793                  * clear any dirty bits and don't set any writeback bits
794                  */
795                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
796                                              start, start + ram_size - 1,
797                                              locked_page, unlock, 1,
798                                              1, 0, 0, 0);
799                 disk_num_bytes -= cur_alloc_size;
800                 num_bytes -= cur_alloc_size;
801                 alloc_hint = ins.objectid + ins.offset;
802                 start += cur_alloc_size;
803         }
804 out:
805         ret = 0;
806         btrfs_end_transaction(trans, root);
807
808         return ret;
809 }
810
811 /*
812  * work queue call back to started compression on a file and pages
813  */
814 static noinline void async_cow_start(struct btrfs_work *work)
815 {
816         struct async_cow *async_cow;
817         int num_added = 0;
818         async_cow = container_of(work, struct async_cow, work);
819
820         compress_file_range(async_cow->inode, async_cow->locked_page,
821                             async_cow->start, async_cow->end, async_cow,
822                             &num_added);
823         if (num_added == 0)
824                 async_cow->inode = NULL;
825 }
826
827 /*
828  * work queue call back to submit previously compressed pages
829  */
830 static noinline void async_cow_submit(struct btrfs_work *work)
831 {
832         struct async_cow *async_cow;
833         struct btrfs_root *root;
834         unsigned long nr_pages;
835
836         async_cow = container_of(work, struct async_cow, work);
837
838         root = async_cow->root;
839         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
840                 PAGE_CACHE_SHIFT;
841
842         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
843
844         if (atomic_read(&root->fs_info->async_delalloc_pages) <
845             5 * 1042 * 1024 &&
846             waitqueue_active(&root->fs_info->async_submit_wait))
847                 wake_up(&root->fs_info->async_submit_wait);
848
849         if (async_cow->inode)
850                 submit_compressed_extents(async_cow->inode, async_cow);
851 }
852
853 static noinline void async_cow_free(struct btrfs_work *work)
854 {
855         struct async_cow *async_cow;
856         async_cow = container_of(work, struct async_cow, work);
857         kfree(async_cow);
858 }
859
860 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
861                                 u64 start, u64 end, int *page_started,
862                                 unsigned long *nr_written)
863 {
864         struct async_cow *async_cow;
865         struct btrfs_root *root = BTRFS_I(inode)->root;
866         unsigned long nr_pages;
867         u64 cur_end;
868         int limit = 10 * 1024 * 1042;
869
870         if (!btrfs_test_opt(root, COMPRESS)) {
871                 return cow_file_range(inode, locked_page, start, end,
872                                       page_started, nr_written, 1);
873         }
874
875         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
876                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
877         while (start < end) {
878                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
879                 async_cow->inode = inode;
880                 async_cow->root = root;
881                 async_cow->locked_page = locked_page;
882                 async_cow->start = start;
883
884                 if (btrfs_test_flag(inode, NOCOMPRESS))
885                         cur_end = end;
886                 else
887                         cur_end = min(end, start + 512 * 1024 - 1);
888
889                 async_cow->end = cur_end;
890                 INIT_LIST_HEAD(&async_cow->extents);
891
892                 async_cow->work.func = async_cow_start;
893                 async_cow->work.ordered_func = async_cow_submit;
894                 async_cow->work.ordered_free = async_cow_free;
895                 async_cow->work.flags = 0;
896
897                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
898                         PAGE_CACHE_SHIFT;
899                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
900
901                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
902                                    &async_cow->work);
903
904                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
905                         wait_event(root->fs_info->async_submit_wait,
906                            (atomic_read(&root->fs_info->async_delalloc_pages) <
907                             limit));
908                 }
909
910                 while (atomic_read(&root->fs_info->async_submit_draining) &&
911                       atomic_read(&root->fs_info->async_delalloc_pages)) {
912                         wait_event(root->fs_info->async_submit_wait,
913                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
914                            0));
915                 }
916
917                 *nr_written += nr_pages;
918                 start = cur_end + 1;
919         }
920         *page_started = 1;
921         return 0;
922 }
923
924 static noinline int csum_exist_in_range(struct btrfs_root *root,
925                                         u64 bytenr, u64 num_bytes)
926 {
927         int ret;
928         struct btrfs_ordered_sum *sums;
929         LIST_HEAD(list);
930
931         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
932                                        bytenr + num_bytes - 1, &list);
933         if (ret == 0 && list_empty(&list))
934                 return 0;
935
936         while (!list_empty(&list)) {
937                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
938                 list_del(&sums->list);
939                 kfree(sums);
940         }
941         return 1;
942 }
943
944 /*
945  * when nowcow writeback call back.  This checks for snapshots or COW copies
946  * of the extents that exist in the file, and COWs the file as required.
947  *
948  * If no cow copies or snapshots exist, we write directly to the existing
949  * blocks on disk
950  */
951 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
952                               u64 start, u64 end, int *page_started, int force,
953                               unsigned long *nr_written)
954 {
955         struct btrfs_root *root = BTRFS_I(inode)->root;
956         struct btrfs_trans_handle *trans;
957         struct extent_buffer *leaf;
958         struct btrfs_path *path;
959         struct btrfs_file_extent_item *fi;
960         struct btrfs_key found_key;
961         u64 cow_start;
962         u64 cur_offset;
963         u64 extent_end;
964         u64 disk_bytenr;
965         u64 num_bytes;
966         int extent_type;
967         int ret;
968         int type;
969         int nocow;
970         int check_prev = 1;
971
972         path = btrfs_alloc_path();
973         BUG_ON(!path);
974         trans = btrfs_join_transaction(root, 1);
975         BUG_ON(!trans);
976
977         cow_start = (u64)-1;
978         cur_offset = start;
979         while (1) {
980                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
981                                                cur_offset, 0);
982                 BUG_ON(ret < 0);
983                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
984                         leaf = path->nodes[0];
985                         btrfs_item_key_to_cpu(leaf, &found_key,
986                                               path->slots[0] - 1);
987                         if (found_key.objectid == inode->i_ino &&
988                             found_key.type == BTRFS_EXTENT_DATA_KEY)
989                                 path->slots[0]--;
990                 }
991                 check_prev = 0;
992 next_slot:
993                 leaf = path->nodes[0];
994                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
995                         ret = btrfs_next_leaf(root, path);
996                         if (ret < 0)
997                                 BUG_ON(1);
998                         if (ret > 0)
999                                 break;
1000                         leaf = path->nodes[0];
1001                 }
1002
1003                 nocow = 0;
1004                 disk_bytenr = 0;
1005                 num_bytes = 0;
1006                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1007
1008                 if (found_key.objectid > inode->i_ino ||
1009                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1010                     found_key.offset > end)
1011                         break;
1012
1013                 if (found_key.offset > cur_offset) {
1014                         extent_end = found_key.offset;
1015                         goto out_check;
1016                 }
1017
1018                 fi = btrfs_item_ptr(leaf, path->slots[0],
1019                                     struct btrfs_file_extent_item);
1020                 extent_type = btrfs_file_extent_type(leaf, fi);
1021
1022                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1023                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1024                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1025                         extent_end = found_key.offset +
1026                                 btrfs_file_extent_num_bytes(leaf, fi);
1027                         if (extent_end <= start) {
1028                                 path->slots[0]++;
1029                                 goto next_slot;
1030                         }
1031                         if (disk_bytenr == 0)
1032                                 goto out_check;
1033                         if (btrfs_file_extent_compression(leaf, fi) ||
1034                             btrfs_file_extent_encryption(leaf, fi) ||
1035                             btrfs_file_extent_other_encoding(leaf, fi))
1036                                 goto out_check;
1037                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1038                                 goto out_check;
1039                         if (btrfs_extent_readonly(root, disk_bytenr))
1040                                 goto out_check;
1041                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1042                                                   disk_bytenr))
1043                                 goto out_check;
1044                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1045                         disk_bytenr += cur_offset - found_key.offset;
1046                         num_bytes = min(end + 1, extent_end) - cur_offset;
1047                         /*
1048                          * force cow if csum exists in the range.
1049                          * this ensure that csum for a given extent are
1050                          * either valid or do not exist.
1051                          */
1052                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1053                                 goto out_check;
1054                         nocow = 1;
1055                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1056                         extent_end = found_key.offset +
1057                                 btrfs_file_extent_inline_len(leaf, fi);
1058                         extent_end = ALIGN(extent_end, root->sectorsize);
1059                 } else {
1060                         BUG_ON(1);
1061                 }
1062 out_check:
1063                 if (extent_end <= start) {
1064                         path->slots[0]++;
1065                         goto next_slot;
1066                 }
1067                 if (!nocow) {
1068                         if (cow_start == (u64)-1)
1069                                 cow_start = cur_offset;
1070                         cur_offset = extent_end;
1071                         if (cur_offset > end)
1072                                 break;
1073                         path->slots[0]++;
1074                         goto next_slot;
1075                 }
1076
1077                 btrfs_release_path(root, path);
1078                 if (cow_start != (u64)-1) {
1079                         ret = cow_file_range(inode, locked_page, cow_start,
1080                                         found_key.offset - 1, page_started,
1081                                         nr_written, 1);
1082                         BUG_ON(ret);
1083                         cow_start = (u64)-1;
1084                 }
1085
1086                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1087                         struct extent_map *em;
1088                         struct extent_map_tree *em_tree;
1089                         em_tree = &BTRFS_I(inode)->extent_tree;
1090                         em = alloc_extent_map(GFP_NOFS);
1091                         em->start = cur_offset;
1092                         em->orig_start = em->start;
1093                         em->len = num_bytes;
1094                         em->block_len = num_bytes;
1095                         em->block_start = disk_bytenr;
1096                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1097                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1098                         while (1) {
1099                                 spin_lock(&em_tree->lock);
1100                                 ret = add_extent_mapping(em_tree, em);
1101                                 spin_unlock(&em_tree->lock);
1102                                 if (ret != -EEXIST) {
1103                                         free_extent_map(em);
1104                                         break;
1105                                 }
1106                                 btrfs_drop_extent_cache(inode, em->start,
1107                                                 em->start + em->len - 1, 0);
1108                         }
1109                         type = BTRFS_ORDERED_PREALLOC;
1110                 } else {
1111                         type = BTRFS_ORDERED_NOCOW;
1112                 }
1113
1114                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1115                                                num_bytes, num_bytes, type);
1116                 BUG_ON(ret);
1117
1118                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1119                                         cur_offset, cur_offset + num_bytes - 1,
1120                                         locked_page, 1, 1, 1, 0, 0, 0);
1121                 cur_offset = extent_end;
1122                 if (cur_offset > end)
1123                         break;
1124         }
1125         btrfs_release_path(root, path);
1126
1127         if (cur_offset <= end && cow_start == (u64)-1)
1128                 cow_start = cur_offset;
1129         if (cow_start != (u64)-1) {
1130                 ret = cow_file_range(inode, locked_page, cow_start, end,
1131                                      page_started, nr_written, 1);
1132                 BUG_ON(ret);
1133         }
1134
1135         ret = btrfs_end_transaction(trans, root);
1136         BUG_ON(ret);
1137         btrfs_free_path(path);
1138         return 0;
1139 }
1140
1141 /*
1142  * extent_io.c call back to do delayed allocation processing
1143  */
1144 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1145                               u64 start, u64 end, int *page_started,
1146                               unsigned long *nr_written)
1147 {
1148         int ret;
1149
1150         if (btrfs_test_flag(inode, NODATACOW))
1151                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1152                                          page_started, 1, nr_written);
1153         else if (btrfs_test_flag(inode, PREALLOC))
1154                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1155                                          page_started, 0, nr_written);
1156         else
1157                 ret = cow_file_range_async(inode, locked_page, start, end,
1158                                            page_started, nr_written);
1159
1160         return ret;
1161 }
1162
1163 /*
1164  * extent_io.c set_bit_hook, used to track delayed allocation
1165  * bytes in this file, and to maintain the list of inodes that
1166  * have pending delalloc work to be done.
1167  */
1168 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1169                        unsigned long old, unsigned long bits)
1170 {
1171         /*
1172          * set_bit and clear bit hooks normally require _irqsave/restore
1173          * but in this case, we are only testeing for the DELALLOC
1174          * bit, which is only set or cleared with irqs on
1175          */
1176         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1177                 struct btrfs_root *root = BTRFS_I(inode)->root;
1178                 spin_lock(&root->fs_info->delalloc_lock);
1179                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1180                 root->fs_info->delalloc_bytes += end - start + 1;
1181                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1182                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1183                                       &root->fs_info->delalloc_inodes);
1184                 }
1185                 spin_unlock(&root->fs_info->delalloc_lock);
1186         }
1187         return 0;
1188 }
1189
1190 /*
1191  * extent_io.c clear_bit_hook, see set_bit_hook for why
1192  */
1193 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1194                          unsigned long old, unsigned long bits)
1195 {
1196         /*
1197          * set_bit and clear bit hooks normally require _irqsave/restore
1198          * but in this case, we are only testeing for the DELALLOC
1199          * bit, which is only set or cleared with irqs on
1200          */
1201         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1202                 struct btrfs_root *root = BTRFS_I(inode)->root;
1203
1204                 spin_lock(&root->fs_info->delalloc_lock);
1205                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1206                         printk(KERN_INFO "btrfs warning: delalloc account "
1207                                "%llu %llu\n",
1208                                (unsigned long long)end - start + 1,
1209                                (unsigned long long)
1210                                root->fs_info->delalloc_bytes);
1211                         root->fs_info->delalloc_bytes = 0;
1212                         BTRFS_I(inode)->delalloc_bytes = 0;
1213                 } else {
1214                         root->fs_info->delalloc_bytes -= end - start + 1;
1215                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1216                 }
1217                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1218                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1219                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1220                 }
1221                 spin_unlock(&root->fs_info->delalloc_lock);
1222         }
1223         return 0;
1224 }
1225
1226 /*
1227  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1228  * we don't create bios that span stripes or chunks
1229  */
1230 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1231                          size_t size, struct bio *bio,
1232                          unsigned long bio_flags)
1233 {
1234         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1235         struct btrfs_mapping_tree *map_tree;
1236         u64 logical = (u64)bio->bi_sector << 9;
1237         u64 length = 0;
1238         u64 map_length;
1239         int ret;
1240
1241         if (bio_flags & EXTENT_BIO_COMPRESSED)
1242                 return 0;
1243
1244         length = bio->bi_size;
1245         map_tree = &root->fs_info->mapping_tree;
1246         map_length = length;
1247         ret = btrfs_map_block(map_tree, READ, logical,
1248                               &map_length, NULL, 0);
1249
1250         if (map_length < length + size)
1251                 return 1;
1252         return 0;
1253 }
1254
1255 /*
1256  * in order to insert checksums into the metadata in large chunks,
1257  * we wait until bio submission time.   All the pages in the bio are
1258  * checksummed and sums are attached onto the ordered extent record.
1259  *
1260  * At IO completion time the cums attached on the ordered extent record
1261  * are inserted into the btree
1262  */
1263 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1264                                     struct bio *bio, int mirror_num,
1265                                     unsigned long bio_flags)
1266 {
1267         struct btrfs_root *root = BTRFS_I(inode)->root;
1268         int ret = 0;
1269
1270         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1271         BUG_ON(ret);
1272         return 0;
1273 }
1274
1275 /*
1276  * in order to insert checksums into the metadata in large chunks,
1277  * we wait until bio submission time.   All the pages in the bio are
1278  * checksummed and sums are attached onto the ordered extent record.
1279  *
1280  * At IO completion time the cums attached on the ordered extent record
1281  * are inserted into the btree
1282  */
1283 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1284                           int mirror_num, unsigned long bio_flags)
1285 {
1286         struct btrfs_root *root = BTRFS_I(inode)->root;
1287         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1288 }
1289
1290 /*
1291  * extent_io.c submission hook. This does the right thing for csum calculation
1292  * on write, or reading the csums from the tree before a read
1293  */
1294 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1295                           int mirror_num, unsigned long bio_flags)
1296 {
1297         struct btrfs_root *root = BTRFS_I(inode)->root;
1298         int ret = 0;
1299         int skip_sum;
1300
1301         skip_sum = btrfs_test_flag(inode, NODATASUM);
1302
1303         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1304         BUG_ON(ret);
1305
1306         if (!(rw & (1 << BIO_RW))) {
1307                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1308                         return btrfs_submit_compressed_read(inode, bio,
1309                                                     mirror_num, bio_flags);
1310                 } else if (!skip_sum)
1311                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1312                 goto mapit;
1313         } else if (!skip_sum) {
1314                 /* csum items have already been cloned */
1315                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1316                         goto mapit;
1317                 /* we're doing a write, do the async checksumming */
1318                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1319                                    inode, rw, bio, mirror_num,
1320                                    bio_flags, __btrfs_submit_bio_start,
1321                                    __btrfs_submit_bio_done);
1322         }
1323
1324 mapit:
1325         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1326 }
1327
1328 /*
1329  * given a list of ordered sums record them in the inode.  This happens
1330  * at IO completion time based on sums calculated at bio submission time.
1331  */
1332 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1333                              struct inode *inode, u64 file_offset,
1334                              struct list_head *list)
1335 {
1336         struct btrfs_ordered_sum *sum;
1337
1338         btrfs_set_trans_block_group(trans, inode);
1339
1340         list_for_each_entry(sum, list, list) {
1341                 btrfs_csum_file_blocks(trans,
1342                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1343         }
1344         return 0;
1345 }
1346
1347 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1348 {
1349         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1350                 WARN_ON(1);
1351         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1352                                    GFP_NOFS);
1353 }
1354
1355 /* see btrfs_writepage_start_hook for details on why this is required */
1356 struct btrfs_writepage_fixup {
1357         struct page *page;
1358         struct btrfs_work work;
1359 };
1360
1361 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1362 {
1363         struct btrfs_writepage_fixup *fixup;
1364         struct btrfs_ordered_extent *ordered;
1365         struct page *page;
1366         struct inode *inode;
1367         u64 page_start;
1368         u64 page_end;
1369
1370         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1371         page = fixup->page;
1372 again:
1373         lock_page(page);
1374         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1375                 ClearPageChecked(page);
1376                 goto out_page;
1377         }
1378
1379         inode = page->mapping->host;
1380         page_start = page_offset(page);
1381         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1382
1383         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1384
1385         /* already ordered? We're done */
1386         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1387                              EXTENT_ORDERED, 0)) {
1388                 goto out;
1389         }
1390
1391         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1392         if (ordered) {
1393                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1394                               page_end, GFP_NOFS);
1395                 unlock_page(page);
1396                 btrfs_start_ordered_extent(inode, ordered, 1);
1397                 goto again;
1398         }
1399
1400         btrfs_set_extent_delalloc(inode, page_start, page_end);
1401         ClearPageChecked(page);
1402 out:
1403         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1404 out_page:
1405         unlock_page(page);
1406         page_cache_release(page);
1407 }
1408
1409 /*
1410  * There are a few paths in the higher layers of the kernel that directly
1411  * set the page dirty bit without asking the filesystem if it is a
1412  * good idea.  This causes problems because we want to make sure COW
1413  * properly happens and the data=ordered rules are followed.
1414  *
1415  * In our case any range that doesn't have the ORDERED bit set
1416  * hasn't been properly setup for IO.  We kick off an async process
1417  * to fix it up.  The async helper will wait for ordered extents, set
1418  * the delalloc bit and make it safe to write the page.
1419  */
1420 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1421 {
1422         struct inode *inode = page->mapping->host;
1423         struct btrfs_writepage_fixup *fixup;
1424         struct btrfs_root *root = BTRFS_I(inode)->root;
1425         int ret;
1426
1427         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1428                              EXTENT_ORDERED, 0);
1429         if (ret)
1430                 return 0;
1431
1432         if (PageChecked(page))
1433                 return -EAGAIN;
1434
1435         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1436         if (!fixup)
1437                 return -EAGAIN;
1438
1439         SetPageChecked(page);
1440         page_cache_get(page);
1441         fixup->work.func = btrfs_writepage_fixup_worker;
1442         fixup->page = page;
1443         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1444         return -EAGAIN;
1445 }
1446
1447 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1448                                        struct inode *inode, u64 file_pos,
1449                                        u64 disk_bytenr, u64 disk_num_bytes,
1450                                        u64 num_bytes, u64 ram_bytes,
1451                                        u8 compression, u8 encryption,
1452                                        u16 other_encoding, int extent_type)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         struct btrfs_file_extent_item *fi;
1456         struct btrfs_path *path;
1457         struct extent_buffer *leaf;
1458         struct btrfs_key ins;
1459         u64 hint;
1460         int ret;
1461
1462         path = btrfs_alloc_path();
1463         BUG_ON(!path);
1464
1465         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1466                                  file_pos + num_bytes, file_pos, &hint);
1467         BUG_ON(ret);
1468
1469         ins.objectid = inode->i_ino;
1470         ins.offset = file_pos;
1471         ins.type = BTRFS_EXTENT_DATA_KEY;
1472         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1473         BUG_ON(ret);
1474         leaf = path->nodes[0];
1475         fi = btrfs_item_ptr(leaf, path->slots[0],
1476                             struct btrfs_file_extent_item);
1477         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1478         btrfs_set_file_extent_type(leaf, fi, extent_type);
1479         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1480         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1481         btrfs_set_file_extent_offset(leaf, fi, 0);
1482         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1483         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1484         btrfs_set_file_extent_compression(leaf, fi, compression);
1485         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1486         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1487         btrfs_mark_buffer_dirty(leaf);
1488
1489         inode_add_bytes(inode, num_bytes);
1490         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1491
1492         ins.objectid = disk_bytenr;
1493         ins.offset = disk_num_bytes;
1494         ins.type = BTRFS_EXTENT_ITEM_KEY;
1495         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1496                                           root->root_key.objectid,
1497                                           trans->transid, inode->i_ino, &ins);
1498         BUG_ON(ret);
1499
1500         btrfs_free_path(path);
1501         return 0;
1502 }
1503
1504 /* as ordered data IO finishes, this gets called so we can finish
1505  * an ordered extent if the range of bytes in the file it covers are
1506  * fully written.
1507  */
1508 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1509 {
1510         struct btrfs_root *root = BTRFS_I(inode)->root;
1511         struct btrfs_trans_handle *trans;
1512         struct btrfs_ordered_extent *ordered_extent;
1513         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1514         int compressed = 0;
1515         int ret;
1516
1517         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1518         if (!ret)
1519                 return 0;
1520
1521         trans = btrfs_join_transaction(root, 1);
1522
1523         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1524         BUG_ON(!ordered_extent);
1525         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1526                 goto nocow;
1527
1528         lock_extent(io_tree, ordered_extent->file_offset,
1529                     ordered_extent->file_offset + ordered_extent->len - 1,
1530                     GFP_NOFS);
1531
1532         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1533                 compressed = 1;
1534         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1535                 BUG_ON(compressed);
1536                 ret = btrfs_mark_extent_written(trans, root, inode,
1537                                                 ordered_extent->file_offset,
1538                                                 ordered_extent->file_offset +
1539                                                 ordered_extent->len);
1540                 BUG_ON(ret);
1541         } else {
1542                 ret = insert_reserved_file_extent(trans, inode,
1543                                                 ordered_extent->file_offset,
1544                                                 ordered_extent->start,
1545                                                 ordered_extent->disk_len,
1546                                                 ordered_extent->len,
1547                                                 ordered_extent->len,
1548                                                 compressed, 0, 0,
1549                                                 BTRFS_FILE_EXTENT_REG);
1550                 BUG_ON(ret);
1551         }
1552         unlock_extent(io_tree, ordered_extent->file_offset,
1553                     ordered_extent->file_offset + ordered_extent->len - 1,
1554                     GFP_NOFS);
1555 nocow:
1556         add_pending_csums(trans, inode, ordered_extent->file_offset,
1557                           &ordered_extent->list);
1558
1559         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1560         btrfs_ordered_update_i_size(inode, ordered_extent);
1561         btrfs_update_inode(trans, root, inode);
1562         btrfs_remove_ordered_extent(inode, ordered_extent);
1563         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1564
1565         /* once for us */
1566         btrfs_put_ordered_extent(ordered_extent);
1567         /* once for the tree */
1568         btrfs_put_ordered_extent(ordered_extent);
1569
1570         btrfs_end_transaction(trans, root);
1571         return 0;
1572 }
1573
1574 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1575                                 struct extent_state *state, int uptodate)
1576 {
1577         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1578 }
1579
1580 /*
1581  * When IO fails, either with EIO or csum verification fails, we
1582  * try other mirrors that might have a good copy of the data.  This
1583  * io_failure_record is used to record state as we go through all the
1584  * mirrors.  If another mirror has good data, the page is set up to date
1585  * and things continue.  If a good mirror can't be found, the original
1586  * bio end_io callback is called to indicate things have failed.
1587  */
1588 struct io_failure_record {
1589         struct page *page;
1590         u64 start;
1591         u64 len;
1592         u64 logical;
1593         unsigned long bio_flags;
1594         int last_mirror;
1595 };
1596
1597 static int btrfs_io_failed_hook(struct bio *failed_bio,
1598                          struct page *page, u64 start, u64 end,
1599                          struct extent_state *state)
1600 {
1601         struct io_failure_record *failrec = NULL;
1602         u64 private;
1603         struct extent_map *em;
1604         struct inode *inode = page->mapping->host;
1605         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1606         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1607         struct bio *bio;
1608         int num_copies;
1609         int ret;
1610         int rw;
1611         u64 logical;
1612
1613         ret = get_state_private(failure_tree, start, &private);
1614         if (ret) {
1615                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1616                 if (!failrec)
1617                         return -ENOMEM;
1618                 failrec->start = start;
1619                 failrec->len = end - start + 1;
1620                 failrec->last_mirror = 0;
1621                 failrec->bio_flags = 0;
1622
1623                 spin_lock(&em_tree->lock);
1624                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1625                 if (em->start > start || em->start + em->len < start) {
1626                         free_extent_map(em);
1627                         em = NULL;
1628                 }
1629                 spin_unlock(&em_tree->lock);
1630
1631                 if (!em || IS_ERR(em)) {
1632                         kfree(failrec);
1633                         return -EIO;
1634                 }
1635                 logical = start - em->start;
1636                 logical = em->block_start + logical;
1637                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1638                         logical = em->block_start;
1639                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1640                 }
1641                 failrec->logical = logical;
1642                 free_extent_map(em);
1643                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1644                                 EXTENT_DIRTY, GFP_NOFS);
1645                 set_state_private(failure_tree, start,
1646                                  (u64)(unsigned long)failrec);
1647         } else {
1648                 failrec = (struct io_failure_record *)(unsigned long)private;
1649         }
1650         num_copies = btrfs_num_copies(
1651                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1652                               failrec->logical, failrec->len);
1653         failrec->last_mirror++;
1654         if (!state) {
1655                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1656                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1657                                                     failrec->start,
1658                                                     EXTENT_LOCKED);
1659                 if (state && state->start != failrec->start)
1660                         state = NULL;
1661                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1662         }
1663         if (!state || failrec->last_mirror > num_copies) {
1664                 set_state_private(failure_tree, failrec->start, 0);
1665                 clear_extent_bits(failure_tree, failrec->start,
1666                                   failrec->start + failrec->len - 1,
1667                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1668                 kfree(failrec);
1669                 return -EIO;
1670         }
1671         bio = bio_alloc(GFP_NOFS, 1);
1672         bio->bi_private = state;
1673         bio->bi_end_io = failed_bio->bi_end_io;
1674         bio->bi_sector = failrec->logical >> 9;
1675         bio->bi_bdev = failed_bio->bi_bdev;
1676         bio->bi_size = 0;
1677
1678         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1679         if (failed_bio->bi_rw & (1 << BIO_RW))
1680                 rw = WRITE;
1681         else
1682                 rw = READ;
1683
1684         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1685                                                       failrec->last_mirror,
1686                                                       failrec->bio_flags);
1687         return 0;
1688 }
1689
1690 /*
1691  * each time an IO finishes, we do a fast check in the IO failure tree
1692  * to see if we need to process or clean up an io_failure_record
1693  */
1694 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1695 {
1696         u64 private;
1697         u64 private_failure;
1698         struct io_failure_record *failure;
1699         int ret;
1700
1701         private = 0;
1702         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1703                              (u64)-1, 1, EXTENT_DIRTY)) {
1704                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1705                                         start, &private_failure);
1706                 if (ret == 0) {
1707                         failure = (struct io_failure_record *)(unsigned long)
1708                                    private_failure;
1709                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1710                                           failure->start, 0);
1711                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1712                                           failure->start,
1713                                           failure->start + failure->len - 1,
1714                                           EXTENT_DIRTY | EXTENT_LOCKED,
1715                                           GFP_NOFS);
1716                         kfree(failure);
1717                 }
1718         }
1719         return 0;
1720 }
1721
1722 /*
1723  * when reads are done, we need to check csums to verify the data is correct
1724  * if there's a match, we allow the bio to finish.  If not, we go through
1725  * the io_failure_record routines to find good copies
1726  */
1727 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1728                                struct extent_state *state)
1729 {
1730         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1731         struct inode *inode = page->mapping->host;
1732         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1733         char *kaddr;
1734         u64 private = ~(u32)0;
1735         int ret;
1736         struct btrfs_root *root = BTRFS_I(inode)->root;
1737         u32 csum = ~(u32)0;
1738
1739         if (PageChecked(page)) {
1740                 ClearPageChecked(page);
1741                 goto good;
1742         }
1743         if (btrfs_test_flag(inode, NODATASUM))
1744                 return 0;
1745
1746         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1747             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1748                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1749                                   GFP_NOFS);
1750                 return 0;
1751         }
1752
1753         if (state && state->start == start) {
1754                 private = state->private;
1755                 ret = 0;
1756         } else {
1757                 ret = get_state_private(io_tree, start, &private);
1758         }
1759         kaddr = kmap_atomic(page, KM_USER0);
1760         if (ret)
1761                 goto zeroit;
1762
1763         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1764         btrfs_csum_final(csum, (char *)&csum);
1765         if (csum != private)
1766                 goto zeroit;
1767
1768         kunmap_atomic(kaddr, KM_USER0);
1769 good:
1770         /* if the io failure tree for this inode is non-empty,
1771          * check to see if we've recovered from a failed IO
1772          */
1773         btrfs_clean_io_failures(inode, start);
1774         return 0;
1775
1776 zeroit:
1777         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1778                "private %llu\n", page->mapping->host->i_ino,
1779                (unsigned long long)start, csum,
1780                (unsigned long long)private);
1781         memset(kaddr + offset, 1, end - start + 1);
1782         flush_dcache_page(page);
1783         kunmap_atomic(kaddr, KM_USER0);
1784         if (private == 0)
1785                 return 0;
1786         return -EIO;
1787 }
1788
1789 /*
1790  * This creates an orphan entry for the given inode in case something goes
1791  * wrong in the middle of an unlink/truncate.
1792  */
1793 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1794 {
1795         struct btrfs_root *root = BTRFS_I(inode)->root;
1796         int ret = 0;
1797
1798         spin_lock(&root->list_lock);
1799
1800         /* already on the orphan list, we're good */
1801         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1802                 spin_unlock(&root->list_lock);
1803                 return 0;
1804         }
1805
1806         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1807
1808         spin_unlock(&root->list_lock);
1809
1810         /*
1811          * insert an orphan item to track this unlinked/truncated file
1812          */
1813         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1814
1815         return ret;
1816 }
1817
1818 /*
1819  * We have done the truncate/delete so we can go ahead and remove the orphan
1820  * item for this particular inode.
1821  */
1822 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1823 {
1824         struct btrfs_root *root = BTRFS_I(inode)->root;
1825         int ret = 0;
1826
1827         spin_lock(&root->list_lock);
1828
1829         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1830                 spin_unlock(&root->list_lock);
1831                 return 0;
1832         }
1833
1834         list_del_init(&BTRFS_I(inode)->i_orphan);
1835         if (!trans) {
1836                 spin_unlock(&root->list_lock);
1837                 return 0;
1838         }
1839
1840         spin_unlock(&root->list_lock);
1841
1842         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1843
1844         return ret;
1845 }
1846
1847 /*
1848  * this cleans up any orphans that may be left on the list from the last use
1849  * of this root.
1850  */
1851 void btrfs_orphan_cleanup(struct btrfs_root *root)
1852 {
1853         struct btrfs_path *path;
1854         struct extent_buffer *leaf;
1855         struct btrfs_item *item;
1856         struct btrfs_key key, found_key;
1857         struct btrfs_trans_handle *trans;
1858         struct inode *inode;
1859         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1860
1861         path = btrfs_alloc_path();
1862         if (!path)
1863                 return;
1864         path->reada = -1;
1865
1866         key.objectid = BTRFS_ORPHAN_OBJECTID;
1867         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1868         key.offset = (u64)-1;
1869
1870
1871         while (1) {
1872                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1873                 if (ret < 0) {
1874                         printk(KERN_ERR "Error searching slot for orphan: %d"
1875                                "\n", ret);
1876                         break;
1877                 }
1878
1879                 /*
1880                  * if ret == 0 means we found what we were searching for, which
1881                  * is weird, but possible, so only screw with path if we didnt
1882                  * find the key and see if we have stuff that matches
1883                  */
1884                 if (ret > 0) {
1885                         if (path->slots[0] == 0)
1886                                 break;
1887                         path->slots[0]--;
1888                 }
1889
1890                 /* pull out the item */
1891                 leaf = path->nodes[0];
1892                 item = btrfs_item_nr(leaf, path->slots[0]);
1893                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1894
1895                 /* make sure the item matches what we want */
1896                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1897                         break;
1898                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1899                         break;
1900
1901                 /* release the path since we're done with it */
1902                 btrfs_release_path(root, path);
1903
1904                 /*
1905                  * this is where we are basically btrfs_lookup, without the
1906                  * crossing root thing.  we store the inode number in the
1907                  * offset of the orphan item.
1908                  */
1909                 inode = btrfs_iget_locked(root->fs_info->sb,
1910                                           found_key.offset, root);
1911                 if (!inode)
1912                         break;
1913
1914                 if (inode->i_state & I_NEW) {
1915                         BTRFS_I(inode)->root = root;
1916
1917                         /* have to set the location manually */
1918                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1919                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1920                         BTRFS_I(inode)->location.offset = 0;
1921
1922                         btrfs_read_locked_inode(inode);
1923                         unlock_new_inode(inode);
1924                 }
1925
1926                 /*
1927                  * add this inode to the orphan list so btrfs_orphan_del does
1928                  * the proper thing when we hit it
1929                  */
1930                 spin_lock(&root->list_lock);
1931                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1932                 spin_unlock(&root->list_lock);
1933
1934                 /*
1935                  * if this is a bad inode, means we actually succeeded in
1936                  * removing the inode, but not the orphan record, which means
1937                  * we need to manually delete the orphan since iput will just
1938                  * do a destroy_inode
1939                  */
1940                 if (is_bad_inode(inode)) {
1941                         trans = btrfs_start_transaction(root, 1);
1942                         btrfs_orphan_del(trans, inode);
1943                         btrfs_end_transaction(trans, root);
1944                         iput(inode);
1945                         continue;
1946                 }
1947
1948                 /* if we have links, this was a truncate, lets do that */
1949                 if (inode->i_nlink) {
1950                         nr_truncate++;
1951                         btrfs_truncate(inode);
1952                 } else {
1953                         nr_unlink++;
1954                 }
1955
1956                 /* this will do delete_inode and everything for us */
1957                 iput(inode);
1958         }
1959
1960         if (nr_unlink)
1961                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1962         if (nr_truncate)
1963                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1964
1965         btrfs_free_path(path);
1966 }
1967
1968 /*
1969  * read an inode from the btree into the in-memory inode
1970  */
1971 void btrfs_read_locked_inode(struct inode *inode)
1972 {
1973         struct btrfs_path *path;
1974         struct extent_buffer *leaf;
1975         struct btrfs_inode_item *inode_item;
1976         struct btrfs_timespec *tspec;
1977         struct btrfs_root *root = BTRFS_I(inode)->root;
1978         struct btrfs_key location;
1979         u64 alloc_group_block;
1980         u32 rdev;
1981         int ret;
1982
1983         path = btrfs_alloc_path();
1984         BUG_ON(!path);
1985         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1986
1987         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1988         if (ret)
1989                 goto make_bad;
1990
1991         leaf = path->nodes[0];
1992         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1993                                     struct btrfs_inode_item);
1994
1995         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1996         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1997         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1998         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1999         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2000
2001         tspec = btrfs_inode_atime(inode_item);
2002         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2003         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2004
2005         tspec = btrfs_inode_mtime(inode_item);
2006         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2007         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2008
2009         tspec = btrfs_inode_ctime(inode_item);
2010         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2011         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2012
2013         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2014         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2015         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2016         inode->i_generation = BTRFS_I(inode)->generation;
2017         inode->i_rdev = 0;
2018         rdev = btrfs_inode_rdev(leaf, inode_item);
2019
2020         BTRFS_I(inode)->index_cnt = (u64)-1;
2021         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2022
2023         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2024         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2025                                                 alloc_group_block, 0);
2026         btrfs_free_path(path);
2027         inode_item = NULL;
2028
2029         switch (inode->i_mode & S_IFMT) {
2030         case S_IFREG:
2031                 inode->i_mapping->a_ops = &btrfs_aops;
2032                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2033                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2034                 inode->i_fop = &btrfs_file_operations;
2035                 inode->i_op = &btrfs_file_inode_operations;
2036                 break;
2037         case S_IFDIR:
2038                 inode->i_fop = &btrfs_dir_file_operations;
2039                 if (root == root->fs_info->tree_root)
2040                         inode->i_op = &btrfs_dir_ro_inode_operations;
2041                 else
2042                         inode->i_op = &btrfs_dir_inode_operations;
2043                 break;
2044         case S_IFLNK:
2045                 inode->i_op = &btrfs_symlink_inode_operations;
2046                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2047                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2048                 break;
2049         default:
2050                 inode->i_op = &btrfs_special_inode_operations;
2051                 init_special_inode(inode, inode->i_mode, rdev);
2052                 break;
2053         }
2054         return;
2055
2056 make_bad:
2057         btrfs_free_path(path);
2058         make_bad_inode(inode);
2059 }
2060
2061 /*
2062  * given a leaf and an inode, copy the inode fields into the leaf
2063  */
2064 static void fill_inode_item(struct btrfs_trans_handle *trans,
2065                             struct extent_buffer *leaf,
2066                             struct btrfs_inode_item *item,
2067                             struct inode *inode)
2068 {
2069         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2070         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2071         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2072         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2073         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2074
2075         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2076                                inode->i_atime.tv_sec);
2077         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2078                                 inode->i_atime.tv_nsec);
2079
2080         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2081                                inode->i_mtime.tv_sec);
2082         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2083                                 inode->i_mtime.tv_nsec);
2084
2085         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2086                                inode->i_ctime.tv_sec);
2087         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2088                                 inode->i_ctime.tv_nsec);
2089
2090         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2091         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2092         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2093         btrfs_set_inode_transid(leaf, item, trans->transid);
2094         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2095         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2096         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2097 }
2098
2099 /*
2100  * copy everything in the in-memory inode into the btree.
2101  */
2102 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2103                                 struct btrfs_root *root, struct inode *inode)
2104 {
2105         struct btrfs_inode_item *inode_item;
2106         struct btrfs_path *path;
2107         struct extent_buffer *leaf;
2108         int ret;
2109
2110         path = btrfs_alloc_path();
2111         BUG_ON(!path);
2112         ret = btrfs_lookup_inode(trans, root, path,
2113                                  &BTRFS_I(inode)->location, 1);
2114         if (ret) {
2115                 if (ret > 0)
2116                         ret = -ENOENT;
2117                 goto failed;
2118         }
2119
2120         leaf = path->nodes[0];
2121         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2122                                   struct btrfs_inode_item);
2123
2124         fill_inode_item(trans, leaf, inode_item, inode);
2125         btrfs_mark_buffer_dirty(leaf);
2126         btrfs_set_inode_last_trans(trans, inode);
2127         ret = 0;
2128 failed:
2129         btrfs_free_path(path);
2130         return ret;
2131 }
2132
2133
2134 /*
2135  * unlink helper that gets used here in inode.c and in the tree logging
2136  * recovery code.  It remove a link in a directory with a given name, and
2137  * also drops the back refs in the inode to the directory
2138  */
2139 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2140                        struct btrfs_root *root,
2141                        struct inode *dir, struct inode *inode,
2142                        const char *name, int name_len)
2143 {
2144         struct btrfs_path *path;
2145         int ret = 0;
2146         struct extent_buffer *leaf;
2147         struct btrfs_dir_item *di;
2148         struct btrfs_key key;
2149         u64 index;
2150
2151         path = btrfs_alloc_path();
2152         if (!path) {
2153                 ret = -ENOMEM;
2154                 goto err;
2155         }
2156
2157         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2158                                     name, name_len, -1);
2159         if (IS_ERR(di)) {
2160                 ret = PTR_ERR(di);
2161                 goto err;
2162         }
2163         if (!di) {
2164                 ret = -ENOENT;
2165                 goto err;
2166         }
2167         leaf = path->nodes[0];
2168         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2169         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2170         if (ret)
2171                 goto err;
2172         btrfs_release_path(root, path);
2173
2174         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2175                                   inode->i_ino,
2176                                   dir->i_ino, &index);
2177         if (ret) {
2178                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2179                        "inode %lu parent %lu\n", name_len, name,
2180                        inode->i_ino, dir->i_ino);
2181                 goto err;
2182         }
2183
2184         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2185                                          index, name, name_len, -1);
2186         if (IS_ERR(di)) {
2187                 ret = PTR_ERR(di);
2188                 goto err;
2189         }
2190         if (!di) {
2191                 ret = -ENOENT;
2192                 goto err;
2193         }
2194         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2195         btrfs_release_path(root, path);
2196
2197         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2198                                          inode, dir->i_ino);
2199         BUG_ON(ret != 0 && ret != -ENOENT);
2200         if (ret != -ENOENT)
2201                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2202
2203         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2204                                            dir, index);
2205         BUG_ON(ret);
2206 err:
2207         btrfs_free_path(path);
2208         if (ret)
2209                 goto out;
2210
2211         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2212         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2213         btrfs_update_inode(trans, root, dir);
2214         btrfs_drop_nlink(inode);
2215         ret = btrfs_update_inode(trans, root, inode);
2216         dir->i_sb->s_dirt = 1;
2217 out:
2218         return ret;
2219 }
2220
2221 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2222 {
2223         struct btrfs_root *root;
2224         struct btrfs_trans_handle *trans;
2225         struct inode *inode = dentry->d_inode;
2226         int ret;
2227         unsigned long nr = 0;
2228
2229         root = BTRFS_I(dir)->root;
2230
2231         ret = btrfs_check_free_space(root, 1, 1);
2232         if (ret)
2233                 goto fail;
2234
2235         trans = btrfs_start_transaction(root, 1);
2236
2237         btrfs_set_trans_block_group(trans, dir);
2238         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2239                                  dentry->d_name.name, dentry->d_name.len);
2240
2241         if (inode->i_nlink == 0)
2242                 ret = btrfs_orphan_add(trans, inode);
2243
2244         nr = trans->blocks_used;
2245
2246         btrfs_end_transaction_throttle(trans, root);
2247 fail:
2248         btrfs_btree_balance_dirty(root, nr);
2249         return ret;
2250 }
2251
2252 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2253 {
2254         struct inode *inode = dentry->d_inode;
2255         int err = 0;
2256         int ret;
2257         struct btrfs_root *root = BTRFS_I(dir)->root;
2258         struct btrfs_trans_handle *trans;
2259         unsigned long nr = 0;
2260
2261         /*
2262          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2263          * the root of a subvolume or snapshot
2264          */
2265         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2266             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2267                 return -ENOTEMPTY;
2268         }
2269
2270         ret = btrfs_check_free_space(root, 1, 1);
2271         if (ret)
2272                 goto fail;
2273
2274         trans = btrfs_start_transaction(root, 1);
2275         btrfs_set_trans_block_group(trans, dir);
2276
2277         err = btrfs_orphan_add(trans, inode);
2278         if (err)
2279                 goto fail_trans;
2280
2281         /* now the directory is empty */
2282         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2283                                  dentry->d_name.name, dentry->d_name.len);
2284         if (!err)
2285                 btrfs_i_size_write(inode, 0);
2286
2287 fail_trans:
2288         nr = trans->blocks_used;
2289         ret = btrfs_end_transaction_throttle(trans, root);
2290 fail:
2291         btrfs_btree_balance_dirty(root, nr);
2292
2293         if (ret && !err)
2294                 err = ret;
2295         return err;
2296 }
2297
2298 #if 0
2299 /*
2300  * when truncating bytes in a file, it is possible to avoid reading
2301  * the leaves that contain only checksum items.  This can be the
2302  * majority of the IO required to delete a large file, but it must
2303  * be done carefully.
2304  *
2305  * The keys in the level just above the leaves are checked to make sure
2306  * the lowest key in a given leaf is a csum key, and starts at an offset
2307  * after the new  size.
2308  *
2309  * Then the key for the next leaf is checked to make sure it also has
2310  * a checksum item for the same file.  If it does, we know our target leaf
2311  * contains only checksum items, and it can be safely freed without reading
2312  * it.
2313  *
2314  * This is just an optimization targeted at large files.  It may do
2315  * nothing.  It will return 0 unless things went badly.
2316  */
2317 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2318                                      struct btrfs_root *root,
2319                                      struct btrfs_path *path,
2320                                      struct inode *inode, u64 new_size)
2321 {
2322         struct btrfs_key key;
2323         int ret;
2324         int nritems;
2325         struct btrfs_key found_key;
2326         struct btrfs_key other_key;
2327         struct btrfs_leaf_ref *ref;
2328         u64 leaf_gen;
2329         u64 leaf_start;
2330
2331         path->lowest_level = 1;
2332         key.objectid = inode->i_ino;
2333         key.type = BTRFS_CSUM_ITEM_KEY;
2334         key.offset = new_size;
2335 again:
2336         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2337         if (ret < 0)
2338                 goto out;
2339
2340         if (path->nodes[1] == NULL) {
2341                 ret = 0;
2342                 goto out;
2343         }
2344         ret = 0;
2345         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2346         nritems = btrfs_header_nritems(path->nodes[1]);
2347
2348         if (!nritems)
2349                 goto out;
2350
2351         if (path->slots[1] >= nritems)
2352                 goto next_node;
2353
2354         /* did we find a key greater than anything we want to delete? */
2355         if (found_key.objectid > inode->i_ino ||
2356            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2357                 goto out;
2358
2359         /* we check the next key in the node to make sure the leave contains
2360          * only checksum items.  This comparison doesn't work if our
2361          * leaf is the last one in the node
2362          */
2363         if (path->slots[1] + 1 >= nritems) {
2364 next_node:
2365                 /* search forward from the last key in the node, this
2366                  * will bring us into the next node in the tree
2367                  */
2368                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2369
2370                 /* unlikely, but we inc below, so check to be safe */
2371                 if (found_key.offset == (u64)-1)
2372                         goto out;
2373
2374                 /* search_forward needs a path with locks held, do the
2375                  * search again for the original key.  It is possible
2376                  * this will race with a balance and return a path that
2377                  * we could modify, but this drop is just an optimization
2378                  * and is allowed to miss some leaves.
2379                  */
2380                 btrfs_release_path(root, path);
2381                 found_key.offset++;
2382
2383                 /* setup a max key for search_forward */
2384                 other_key.offset = (u64)-1;
2385                 other_key.type = key.type;
2386                 other_key.objectid = key.objectid;
2387
2388                 path->keep_locks = 1;
2389                 ret = btrfs_search_forward(root, &found_key, &other_key,
2390                                            path, 0, 0);
2391                 path->keep_locks = 0;
2392                 if (ret || found_key.objectid != key.objectid ||
2393                     found_key.type != key.type) {
2394                         ret = 0;
2395                         goto out;
2396                 }
2397
2398                 key.offset = found_key.offset;
2399                 btrfs_release_path(root, path);
2400                 cond_resched();
2401                 goto again;
2402         }
2403
2404         /* we know there's one more slot after us in the tree,
2405          * read that key so we can verify it is also a checksum item
2406          */
2407         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2408
2409         if (found_key.objectid < inode->i_ino)
2410                 goto next_key;
2411
2412         if (found_key.type != key.type || found_key.offset < new_size)
2413                 goto next_key;
2414
2415         /*
2416          * if the key for the next leaf isn't a csum key from this objectid,
2417          * we can't be sure there aren't good items inside this leaf.
2418          * Bail out
2419          */
2420         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2421                 goto out;
2422
2423         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2424         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2425         /*
2426          * it is safe to delete this leaf, it contains only
2427          * csum items from this inode at an offset >= new_size
2428          */
2429         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2430         BUG_ON(ret);
2431
2432         if (root->ref_cows && leaf_gen < trans->transid) {
2433                 ref = btrfs_alloc_leaf_ref(root, 0);
2434                 if (ref) {
2435                         ref->root_gen = root->root_key.offset;
2436                         ref->bytenr = leaf_start;
2437                         ref->owner = 0;
2438                         ref->generation = leaf_gen;
2439                         ref->nritems = 0;
2440
2441                         ret = btrfs_add_leaf_ref(root, ref, 0);
2442                         WARN_ON(ret);
2443                         btrfs_free_leaf_ref(root, ref);
2444                 } else {
2445                         WARN_ON(1);
2446                 }
2447         }
2448 next_key:
2449         btrfs_release_path(root, path);
2450
2451         if (other_key.objectid == inode->i_ino &&
2452             other_key.type == key.type && other_key.offset > key.offset) {
2453                 key.offset = other_key.offset;
2454                 cond_resched();
2455                 goto again;
2456         }
2457         ret = 0;
2458 out:
2459         /* fixup any changes we've made to the path */
2460         path->lowest_level = 0;
2461         path->keep_locks = 0;
2462         btrfs_release_path(root, path);
2463         return ret;
2464 }
2465
2466 #endif
2467
2468 /*
2469  * this can truncate away extent items, csum items and directory items.
2470  * It starts at a high offset and removes keys until it can't find
2471  * any higher than new_size
2472  *
2473  * csum items that cross the new i_size are truncated to the new size
2474  * as well.
2475  *
2476  * min_type is the minimum key type to truncate down to.  If set to 0, this
2477  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2478  */
2479 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2480                                         struct btrfs_root *root,
2481                                         struct inode *inode,
2482                                         u64 new_size, u32 min_type)
2483 {
2484         int ret;
2485         struct btrfs_path *path;
2486         struct btrfs_key key;
2487         struct btrfs_key found_key;
2488         u32 found_type;
2489         struct extent_buffer *leaf;
2490         struct btrfs_file_extent_item *fi;
2491         u64 extent_start = 0;
2492         u64 extent_num_bytes = 0;
2493         u64 item_end = 0;
2494         u64 root_gen = 0;
2495         u64 root_owner = 0;
2496         int found_extent;
2497         int del_item;
2498         int pending_del_nr = 0;
2499         int pending_del_slot = 0;
2500         int extent_type = -1;
2501         int encoding;
2502         u64 mask = root->sectorsize - 1;
2503
2504         if (root->ref_cows)
2505                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2506         path = btrfs_alloc_path();
2507         path->reada = -1;
2508         BUG_ON(!path);
2509
2510         /* FIXME, add redo link to tree so we don't leak on crash */
2511         key.objectid = inode->i_ino;
2512         key.offset = (u64)-1;
2513         key.type = (u8)-1;
2514
2515         btrfs_init_path(path);
2516
2517 search_again:
2518         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2519         if (ret < 0)
2520                 goto error;
2521
2522         if (ret > 0) {
2523                 /* there are no items in the tree for us to truncate, we're
2524                  * done
2525                  */
2526                 if (path->slots[0] == 0) {
2527                         ret = 0;
2528                         goto error;
2529                 }
2530                 path->slots[0]--;
2531         }
2532
2533         while (1) {
2534                 fi = NULL;
2535                 leaf = path->nodes[0];
2536                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2537                 found_type = btrfs_key_type(&found_key);
2538                 encoding = 0;
2539
2540                 if (found_key.objectid != inode->i_ino)
2541                         break;
2542
2543                 if (found_type < min_type)
2544                         break;
2545
2546                 item_end = found_key.offset;
2547                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2548                         fi = btrfs_item_ptr(leaf, path->slots[0],
2549                                             struct btrfs_file_extent_item);
2550                         extent_type = btrfs_file_extent_type(leaf, fi);
2551                         encoding = btrfs_file_extent_compression(leaf, fi);
2552                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2553                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2554
2555                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2556                                 item_end +=
2557                                     btrfs_file_extent_num_bytes(leaf, fi);
2558                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2559                                 item_end += btrfs_file_extent_inline_len(leaf,
2560                                                                          fi);
2561                         }
2562                         item_end--;
2563                 }
2564                 if (item_end < new_size) {
2565                         if (found_type == BTRFS_DIR_ITEM_KEY)
2566                                 found_type = BTRFS_INODE_ITEM_KEY;
2567                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2568                                 found_type = BTRFS_EXTENT_DATA_KEY;
2569                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2570                                 found_type = BTRFS_XATTR_ITEM_KEY;
2571                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2572                                 found_type = BTRFS_INODE_REF_KEY;
2573                         else if (found_type)
2574                                 found_type--;
2575                         else
2576                                 break;
2577                         btrfs_set_key_type(&key, found_type);
2578                         goto next;
2579                 }
2580                 if (found_key.offset >= new_size)
2581                         del_item = 1;
2582                 else
2583                         del_item = 0;
2584                 found_extent = 0;
2585
2586                 /* FIXME, shrink the extent if the ref count is only 1 */
2587                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2588                         goto delete;
2589
2590                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2591                         u64 num_dec;
2592                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2593                         if (!del_item && !encoding) {
2594                                 u64 orig_num_bytes =
2595                                         btrfs_file_extent_num_bytes(leaf, fi);
2596                                 extent_num_bytes = new_size -
2597                                         found_key.offset + root->sectorsize - 1;
2598                                 extent_num_bytes = extent_num_bytes &
2599                                         ~((u64)root->sectorsize - 1);
2600                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2601                                                          extent_num_bytes);
2602                                 num_dec = (orig_num_bytes -
2603                                            extent_num_bytes);
2604                                 if (root->ref_cows && extent_start != 0)
2605                                         inode_sub_bytes(inode, num_dec);
2606                                 btrfs_mark_buffer_dirty(leaf);
2607                         } else {
2608                                 extent_num_bytes =
2609                                         btrfs_file_extent_disk_num_bytes(leaf,
2610                                                                          fi);
2611                                 /* FIXME blocksize != 4096 */
2612                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2613                                 if (extent_start != 0) {
2614                                         found_extent = 1;
2615                                         if (root->ref_cows)
2616                                                 inode_sub_bytes(inode, num_dec);
2617                                 }
2618                                 root_gen = btrfs_header_generation(leaf);
2619                                 root_owner = btrfs_header_owner(leaf);
2620                         }
2621                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2622                         /*
2623                          * we can't truncate inline items that have had
2624                          * special encodings
2625                          */
2626                         if (!del_item &&
2627                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2628                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2629                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2630                                 u32 size = new_size - found_key.offset;
2631
2632                                 if (root->ref_cows) {
2633                                         inode_sub_bytes(inode, item_end + 1 -
2634                                                         new_size);
2635                                 }
2636                                 size =
2637                                     btrfs_file_extent_calc_inline_size(size);
2638                                 ret = btrfs_truncate_item(trans, root, path,
2639                                                           size, 1);
2640                                 BUG_ON(ret);
2641                         } else if (root->ref_cows) {
2642                                 inode_sub_bytes(inode, item_end + 1 -
2643                                                 found_key.offset);
2644                         }
2645                 }
2646 delete:
2647                 if (del_item) {
2648                         if (!pending_del_nr) {
2649                                 /* no pending yet, add ourselves */
2650                                 pending_del_slot = path->slots[0];
2651                                 pending_del_nr = 1;
2652                         } else if (pending_del_nr &&
2653                                    path->slots[0] + 1 == pending_del_slot) {
2654                                 /* hop on the pending chunk */
2655                                 pending_del_nr++;
2656                                 pending_del_slot = path->slots[0];
2657                         } else {
2658                                 BUG();
2659                         }
2660                 } else {
2661                         break;
2662                 }
2663                 if (found_extent) {
2664                         ret = btrfs_free_extent(trans, root, extent_start,
2665                                                 extent_num_bytes,
2666                                                 leaf->start, root_owner,
2667                                                 root_gen, inode->i_ino, 0);
2668                         BUG_ON(ret);
2669                 }
2670 next:
2671                 if (path->slots[0] == 0) {
2672                         if (pending_del_nr)
2673                                 goto del_pending;
2674                         btrfs_release_path(root, path);
2675                         goto search_again;
2676                 }
2677
2678                 path->slots[0]--;
2679                 if (pending_del_nr &&
2680                     path->slots[0] + 1 != pending_del_slot) {
2681                         struct btrfs_key debug;
2682 del_pending:
2683                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2684                                               pending_del_slot);
2685                         ret = btrfs_del_items(trans, root, path,
2686                                               pending_del_slot,
2687                                               pending_del_nr);
2688                         BUG_ON(ret);
2689                         pending_del_nr = 0;
2690                         btrfs_release_path(root, path);
2691                         goto search_again;
2692                 }
2693         }
2694         ret = 0;
2695 error:
2696         if (pending_del_nr) {
2697                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2698                                       pending_del_nr);
2699         }
2700         btrfs_free_path(path);
2701         inode->i_sb->s_dirt = 1;
2702         return ret;
2703 }
2704
2705 /*
2706  * taken from block_truncate_page, but does cow as it zeros out
2707  * any bytes left in the last page in the file.
2708  */
2709 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2710 {
2711         struct inode *inode = mapping->host;
2712         struct btrfs_root *root = BTRFS_I(inode)->root;
2713         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2714         struct btrfs_ordered_extent *ordered;
2715         char *kaddr;
2716         u32 blocksize = root->sectorsize;
2717         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2718         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2719         struct page *page;
2720         int ret = 0;
2721         u64 page_start;
2722         u64 page_end;
2723
2724         if ((offset & (blocksize - 1)) == 0)
2725                 goto out;
2726
2727         ret = -ENOMEM;
2728 again:
2729         page = grab_cache_page(mapping, index);
2730         if (!page)
2731                 goto out;
2732
2733         page_start = page_offset(page);
2734         page_end = page_start + PAGE_CACHE_SIZE - 1;
2735
2736         if (!PageUptodate(page)) {
2737                 ret = btrfs_readpage(NULL, page);
2738                 lock_page(page);
2739                 if (page->mapping != mapping) {
2740                         unlock_page(page);
2741                         page_cache_release(page);
2742                         goto again;
2743                 }
2744                 if (!PageUptodate(page)) {
2745                         ret = -EIO;
2746                         goto out_unlock;
2747                 }
2748         }
2749         wait_on_page_writeback(page);
2750
2751         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2752         set_page_extent_mapped(page);
2753
2754         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2755         if (ordered) {
2756                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2757                 unlock_page(page);
2758                 page_cache_release(page);
2759                 btrfs_start_ordered_extent(inode, ordered, 1);
2760                 btrfs_put_ordered_extent(ordered);
2761                 goto again;
2762         }
2763
2764         btrfs_set_extent_delalloc(inode, page_start, page_end);
2765         ret = 0;
2766         if (offset != PAGE_CACHE_SIZE) {
2767                 kaddr = kmap(page);
2768                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2769                 flush_dcache_page(page);
2770                 kunmap(page);
2771         }
2772         ClearPageChecked(page);
2773         set_page_dirty(page);
2774         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2775
2776 out_unlock:
2777         unlock_page(page);
2778         page_cache_release(page);
2779 out:
2780         return ret;
2781 }
2782
2783 int btrfs_cont_expand(struct inode *inode, loff_t size)
2784 {
2785         struct btrfs_trans_handle *trans;
2786         struct btrfs_root *root = BTRFS_I(inode)->root;
2787         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2788         struct extent_map *em;
2789         u64 mask = root->sectorsize - 1;
2790         u64 hole_start = (inode->i_size + mask) & ~mask;
2791         u64 block_end = (size + mask) & ~mask;
2792         u64 last_byte;
2793         u64 cur_offset;
2794         u64 hole_size;
2795         int err;
2796
2797         if (size <= hole_start)
2798                 return 0;
2799
2800         err = btrfs_check_free_space(root, 1, 0);
2801         if (err)
2802                 return err;
2803
2804         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2805
2806         while (1) {
2807                 struct btrfs_ordered_extent *ordered;
2808                 btrfs_wait_ordered_range(inode, hole_start,
2809                                          block_end - hole_start);
2810                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2811                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2812                 if (!ordered)
2813                         break;
2814                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2815                 btrfs_put_ordered_extent(ordered);
2816         }
2817
2818         trans = btrfs_start_transaction(root, 1);
2819         btrfs_set_trans_block_group(trans, inode);
2820
2821         cur_offset = hole_start;
2822         while (1) {
2823                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2824                                 block_end - cur_offset, 0);
2825                 BUG_ON(IS_ERR(em) || !em);
2826                 last_byte = min(extent_map_end(em), block_end);
2827                 last_byte = (last_byte + mask) & ~mask;
2828                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2829                         u64 hint_byte = 0;
2830                         hole_size = last_byte - cur_offset;
2831                         err = btrfs_drop_extents(trans, root, inode,
2832                                                  cur_offset,
2833                                                  cur_offset + hole_size,
2834                                                  cur_offset, &hint_byte);
2835                         if (err)
2836                                 break;
2837                         err = btrfs_insert_file_extent(trans, root,
2838                                         inode->i_ino, cur_offset, 0,
2839                                         0, hole_size, 0, hole_size,
2840                                         0, 0, 0);
2841                         btrfs_drop_extent_cache(inode, hole_start,
2842                                         last_byte - 1, 0);
2843                 }
2844                 free_extent_map(em);
2845                 cur_offset = last_byte;
2846                 if (err || cur_offset >= block_end)
2847                         break;
2848         }
2849
2850         btrfs_end_transaction(trans, root);
2851         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2852         return err;
2853 }
2854
2855 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2856 {
2857         struct inode *inode = dentry->d_inode;
2858         int err;
2859
2860         err = inode_change_ok(inode, attr);
2861         if (err)
2862                 return err;
2863
2864         if (S_ISREG(inode->i_mode) &&
2865             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2866                 err = btrfs_cont_expand(inode, attr->ia_size);
2867                 if (err)
2868                         return err;
2869         }
2870
2871         err = inode_setattr(inode, attr);
2872
2873         if (!err && ((attr->ia_valid & ATTR_MODE)))
2874                 err = btrfs_acl_chmod(inode);
2875         return err;
2876 }
2877
2878 void btrfs_delete_inode(struct inode *inode)
2879 {
2880         struct btrfs_trans_handle *trans;
2881         struct btrfs_root *root = BTRFS_I(inode)->root;
2882         unsigned long nr;
2883         int ret;
2884
2885         truncate_inode_pages(&inode->i_data, 0);
2886         if (is_bad_inode(inode)) {
2887                 btrfs_orphan_del(NULL, inode);
2888                 goto no_delete;
2889         }
2890         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2891
2892         btrfs_i_size_write(inode, 0);
2893         trans = btrfs_join_transaction(root, 1);
2894
2895         btrfs_set_trans_block_group(trans, inode);
2896         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2897         if (ret) {
2898                 btrfs_orphan_del(NULL, inode);
2899                 goto no_delete_lock;
2900         }
2901
2902         btrfs_orphan_del(trans, inode);
2903
2904         nr = trans->blocks_used;
2905         clear_inode(inode);
2906
2907         btrfs_end_transaction(trans, root);
2908         btrfs_btree_balance_dirty(root, nr);
2909         return;
2910
2911 no_delete_lock:
2912         nr = trans->blocks_used;
2913         btrfs_end_transaction(trans, root);
2914         btrfs_btree_balance_dirty(root, nr);
2915 no_delete:
2916         clear_inode(inode);
2917 }
2918
2919 /*
2920  * this returns the key found in the dir entry in the location pointer.
2921  * If no dir entries were found, location->objectid is 0.
2922  */
2923 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2924                                struct btrfs_key *location)
2925 {
2926         const char *name = dentry->d_name.name;
2927         int namelen = dentry->d_name.len;
2928         struct btrfs_dir_item *di;
2929         struct btrfs_path *path;
2930         struct btrfs_root *root = BTRFS_I(dir)->root;
2931         int ret = 0;
2932
2933         path = btrfs_alloc_path();
2934         BUG_ON(!path);
2935
2936         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2937                                     namelen, 0);
2938         if (IS_ERR(di))
2939                 ret = PTR_ERR(di);
2940
2941         if (!di || IS_ERR(di))
2942                 goto out_err;
2943
2944         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2945 out:
2946         btrfs_free_path(path);
2947         return ret;
2948 out_err:
2949         location->objectid = 0;
2950         goto out;
2951 }
2952
2953 /*
2954  * when we hit a tree root in a directory, the btrfs part of the inode
2955  * needs to be changed to reflect the root directory of the tree root.  This
2956  * is kind of like crossing a mount point.
2957  */
2958 static int fixup_tree_root_location(struct btrfs_root *root,
2959                              struct btrfs_key *location,
2960                              struct btrfs_root **sub_root,
2961                              struct dentry *dentry)
2962 {
2963         struct btrfs_root_item *ri;
2964
2965         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2966                 return 0;
2967         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2968                 return 0;
2969
2970         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2971                                         dentry->d_name.name,
2972                                         dentry->d_name.len);
2973         if (IS_ERR(*sub_root))
2974                 return PTR_ERR(*sub_root);
2975
2976         ri = &(*sub_root)->root_item;
2977         location->objectid = btrfs_root_dirid(ri);
2978         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2979         location->offset = 0;
2980
2981         return 0;
2982 }
2983
2984 static noinline void init_btrfs_i(struct inode *inode)
2985 {
2986         struct btrfs_inode *bi = BTRFS_I(inode);
2987
2988         bi->i_acl = NULL;
2989         bi->i_default_acl = NULL;
2990
2991         bi->generation = 0;
2992         bi->sequence = 0;
2993         bi->last_trans = 0;
2994         bi->logged_trans = 0;
2995         bi->delalloc_bytes = 0;
2996         bi->disk_i_size = 0;
2997         bi->flags = 0;
2998         bi->index_cnt = (u64)-1;
2999         bi->log_dirty_trans = 0;
3000         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3001         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3002                              inode->i_mapping, GFP_NOFS);
3003         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3004                              inode->i_mapping, GFP_NOFS);
3005         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3006         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3007         mutex_init(&BTRFS_I(inode)->extent_mutex);
3008         mutex_init(&BTRFS_I(inode)->log_mutex);
3009 }
3010
3011 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3012 {
3013         struct btrfs_iget_args *args = p;
3014         inode->i_ino = args->ino;
3015         init_btrfs_i(inode);
3016         BTRFS_I(inode)->root = args->root;
3017         return 0;
3018 }
3019
3020 static int btrfs_find_actor(struct inode *inode, void *opaque)
3021 {
3022         struct btrfs_iget_args *args = opaque;
3023         return args->ino == inode->i_ino &&
3024                 args->root == BTRFS_I(inode)->root;
3025 }
3026
3027 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3028                             struct btrfs_root *root, int wait)
3029 {
3030         struct inode *inode;
3031         struct btrfs_iget_args args;
3032         args.ino = objectid;
3033         args.root = root;
3034
3035         if (wait) {
3036                 inode = ilookup5(s, objectid, btrfs_find_actor,
3037                                  (void *)&args);
3038         } else {
3039                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3040                                         (void *)&args);
3041         }
3042         return inode;
3043 }
3044
3045 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3046                                 struct btrfs_root *root)
3047 {
3048         struct inode *inode;
3049         struct btrfs_iget_args args;
3050         args.ino = objectid;
3051         args.root = root;
3052
3053         inode = iget5_locked(s, objectid, btrfs_find_actor,
3054                              btrfs_init_locked_inode,
3055                              (void *)&args);
3056         return inode;
3057 }
3058
3059 /* Get an inode object given its location and corresponding root.
3060  * Returns in *is_new if the inode was read from disk
3061  */
3062 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3063                          struct btrfs_root *root, int *is_new)
3064 {
3065         struct inode *inode;
3066
3067         inode = btrfs_iget_locked(s, location->objectid, root);
3068         if (!inode)
3069                 return ERR_PTR(-EACCES);
3070
3071         if (inode->i_state & I_NEW) {
3072                 BTRFS_I(inode)->root = root;
3073                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3074                 btrfs_read_locked_inode(inode);
3075                 unlock_new_inode(inode);
3076                 if (is_new)
3077                         *is_new = 1;
3078         } else {
3079                 if (is_new)
3080                         *is_new = 0;
3081         }
3082
3083         return inode;
3084 }
3085
3086 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3087 {
3088         struct inode *inode;
3089         struct btrfs_inode *bi = BTRFS_I(dir);
3090         struct btrfs_root *root = bi->root;
3091         struct btrfs_root *sub_root = root;
3092         struct btrfs_key location;
3093         int ret, new;
3094
3095         if (dentry->d_name.len > BTRFS_NAME_LEN)
3096                 return ERR_PTR(-ENAMETOOLONG);
3097
3098         ret = btrfs_inode_by_name(dir, dentry, &location);
3099
3100         if (ret < 0)
3101                 return ERR_PTR(ret);
3102
3103         inode = NULL;
3104         if (location.objectid) {
3105                 ret = fixup_tree_root_location(root, &location, &sub_root,
3106                                                 dentry);
3107                 if (ret < 0)
3108                         return ERR_PTR(ret);
3109                 if (ret > 0)
3110                         return ERR_PTR(-ENOENT);
3111                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3112                 if (IS_ERR(inode))
3113                         return ERR_CAST(inode);
3114         }
3115         return inode;
3116 }
3117
3118 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3119                                    struct nameidata *nd)
3120 {
3121         struct inode *inode;
3122
3123         if (dentry->d_name.len > BTRFS_NAME_LEN)
3124                 return ERR_PTR(-ENAMETOOLONG);
3125
3126         inode = btrfs_lookup_dentry(dir, dentry);
3127         if (IS_ERR(inode))
3128                 return ERR_CAST(inode);
3129
3130         return d_splice_alias(inode, dentry);
3131 }
3132
3133 static unsigned char btrfs_filetype_table[] = {
3134         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3135 };
3136
3137 static int btrfs_real_readdir(struct file *filp, void *dirent,
3138                               filldir_t filldir)
3139 {
3140         struct inode *inode = filp->f_dentry->d_inode;
3141         struct btrfs_root *root = BTRFS_I(inode)->root;
3142         struct btrfs_item *item;
3143         struct btrfs_dir_item *di;
3144         struct btrfs_key key;
3145         struct btrfs_key found_key;
3146         struct btrfs_path *path;
3147         int ret;
3148         u32 nritems;
3149         struct extent_buffer *leaf;
3150         int slot;
3151         int advance;
3152         unsigned char d_type;
3153         int over = 0;
3154         u32 di_cur;
3155         u32 di_total;
3156         u32 di_len;
3157         int key_type = BTRFS_DIR_INDEX_KEY;
3158         char tmp_name[32];
3159         char *name_ptr;
3160         int name_len;
3161
3162         /* FIXME, use a real flag for deciding about the key type */
3163         if (root->fs_info->tree_root == root)
3164                 key_type = BTRFS_DIR_ITEM_KEY;
3165
3166         /* special case for "." */
3167         if (filp->f_pos == 0) {
3168                 over = filldir(dirent, ".", 1,
3169                                1, inode->i_ino,
3170                                DT_DIR);
3171                 if (over)
3172                         return 0;
3173                 filp->f_pos = 1;
3174         }
3175         /* special case for .., just use the back ref */
3176         if (filp->f_pos == 1) {
3177                 u64 pino = parent_ino(filp->f_path.dentry);
3178                 over = filldir(dirent, "..", 2,
3179                                2, pino, DT_DIR);
3180                 if (over)
3181                         return 0;
3182                 filp->f_pos = 2;
3183         }
3184         path = btrfs_alloc_path();
3185         path->reada = 2;
3186
3187         btrfs_set_key_type(&key, key_type);
3188         key.offset = filp->f_pos;
3189         key.objectid = inode->i_ino;
3190
3191         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3192         if (ret < 0)
3193                 goto err;
3194         advance = 0;
3195
3196         while (1) {
3197                 leaf = path->nodes[0];
3198                 nritems = btrfs_header_nritems(leaf);
3199                 slot = path->slots[0];
3200                 if (advance || slot >= nritems) {
3201                         if (slot >= nritems - 1) {
3202                                 ret = btrfs_next_leaf(root, path);
3203                                 if (ret)
3204                                         break;
3205                                 leaf = path->nodes[0];
3206                                 nritems = btrfs_header_nritems(leaf);
3207                                 slot = path->slots[0];
3208                         } else {
3209                                 slot++;
3210                                 path->slots[0]++;
3211                         }
3212                 }
3213
3214                 advance = 1;
3215                 item = btrfs_item_nr(leaf, slot);
3216                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3217
3218                 if (found_key.objectid != key.objectid)
3219                         break;
3220                 if (btrfs_key_type(&found_key) != key_type)
3221                         break;
3222                 if (found_key.offset < filp->f_pos)
3223                         continue;
3224
3225                 filp->f_pos = found_key.offset;
3226
3227                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3228                 di_cur = 0;
3229                 di_total = btrfs_item_size(leaf, item);
3230
3231                 while (di_cur < di_total) {
3232                         struct btrfs_key location;
3233
3234                         name_len = btrfs_dir_name_len(leaf, di);
3235                         if (name_len <= sizeof(tmp_name)) {
3236                                 name_ptr = tmp_name;
3237                         } else {
3238                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3239                                 if (!name_ptr) {
3240                                         ret = -ENOMEM;
3241                                         goto err;
3242                                 }
3243                         }
3244                         read_extent_buffer(leaf, name_ptr,
3245                                            (unsigned long)(di + 1), name_len);
3246
3247                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3248                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3249
3250                         /* is this a reference to our own snapshot? If so
3251                          * skip it
3252                          */
3253                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3254                             location.objectid == root->root_key.objectid) {
3255                                 over = 0;
3256                                 goto skip;
3257                         }
3258                         over = filldir(dirent, name_ptr, name_len,
3259                                        found_key.offset, location.objectid,
3260                                        d_type);
3261
3262 skip:
3263                         if (name_ptr != tmp_name)
3264                                 kfree(name_ptr);
3265
3266                         if (over)
3267                                 goto nopos;
3268                         di_len = btrfs_dir_name_len(leaf, di) +
3269                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3270                         di_cur += di_len;
3271                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3272                 }
3273         }
3274
3275         /* Reached end of directory/root. Bump pos past the last item. */
3276         if (key_type == BTRFS_DIR_INDEX_KEY)
3277                 filp->f_pos = INT_LIMIT(off_t);
3278         else
3279                 filp->f_pos++;
3280 nopos:
3281         ret = 0;
3282 err:
3283         btrfs_free_path(path);
3284         return ret;
3285 }
3286
3287 int btrfs_write_inode(struct inode *inode, int wait)
3288 {
3289         struct btrfs_root *root = BTRFS_I(inode)->root;
3290         struct btrfs_trans_handle *trans;
3291         int ret = 0;
3292
3293         if (root->fs_info->btree_inode == inode)
3294                 return 0;
3295
3296         if (wait) {
3297                 trans = btrfs_join_transaction(root, 1);
3298                 btrfs_set_trans_block_group(trans, inode);
3299                 ret = btrfs_commit_transaction(trans, root);
3300         }
3301         return ret;
3302 }
3303
3304 /*
3305  * This is somewhat expensive, updating the tree every time the
3306  * inode changes.  But, it is most likely to find the inode in cache.
3307  * FIXME, needs more benchmarking...there are no reasons other than performance
3308  * to keep or drop this code.
3309  */
3310 void btrfs_dirty_inode(struct inode *inode)
3311 {
3312         struct btrfs_root *root = BTRFS_I(inode)->root;
3313         struct btrfs_trans_handle *trans;
3314
3315         trans = btrfs_join_transaction(root, 1);
3316         btrfs_set_trans_block_group(trans, inode);
3317         btrfs_update_inode(trans, root, inode);
3318         btrfs_end_transaction(trans, root);
3319 }
3320
3321 /*
3322  * find the highest existing sequence number in a directory
3323  * and then set the in-memory index_cnt variable to reflect
3324  * free sequence numbers
3325  */
3326 static int btrfs_set_inode_index_count(struct inode *inode)
3327 {
3328         struct btrfs_root *root = BTRFS_I(inode)->root;
3329         struct btrfs_key key, found_key;
3330         struct btrfs_path *path;
3331         struct extent_buffer *leaf;
3332         int ret;
3333
3334         key.objectid = inode->i_ino;
3335         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3336         key.offset = (u64)-1;
3337
3338         path = btrfs_alloc_path();
3339         if (!path)
3340                 return -ENOMEM;
3341
3342         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3343         if (ret < 0)
3344                 goto out;
3345         /* FIXME: we should be able to handle this */
3346         if (ret == 0)
3347                 goto out;
3348         ret = 0;
3349
3350         /*
3351          * MAGIC NUMBER EXPLANATION:
3352          * since we search a directory based on f_pos we have to start at 2
3353          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3354          * else has to start at 2
3355          */
3356         if (path->slots[0] == 0) {
3357                 BTRFS_I(inode)->index_cnt = 2;
3358                 goto out;
3359         }
3360
3361         path->slots[0]--;
3362
3363         leaf = path->nodes[0];
3364         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3365
3366         if (found_key.objectid != inode->i_ino ||
3367             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3368                 BTRFS_I(inode)->index_cnt = 2;
3369                 goto out;
3370         }
3371
3372         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3373 out:
3374         btrfs_free_path(path);
3375         return ret;
3376 }
3377
3378 /*
3379  * helper to find a free sequence number in a given directory.  This current
3380  * code is very simple, later versions will do smarter things in the btree
3381  */
3382 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3383 {
3384         int ret = 0;
3385
3386         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3387                 ret = btrfs_set_inode_index_count(dir);
3388                 if (ret)
3389                         return ret;
3390         }
3391
3392         *index = BTRFS_I(dir)->index_cnt;
3393         BTRFS_I(dir)->index_cnt++;
3394
3395         return ret;
3396 }
3397
3398 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3399                                      struct btrfs_root *root,
3400                                      struct inode *dir,
3401                                      const char *name, int name_len,
3402                                      u64 ref_objectid, u64 objectid,
3403                                      u64 alloc_hint, int mode, u64 *index)
3404 {
3405         struct inode *inode;
3406         struct btrfs_inode_item *inode_item;
3407         struct btrfs_key *location;
3408         struct btrfs_path *path;
3409         struct btrfs_inode_ref *ref;
3410         struct btrfs_key key[2];
3411         u32 sizes[2];
3412         unsigned long ptr;
3413         int ret;
3414         int owner;
3415
3416         path = btrfs_alloc_path();
3417         BUG_ON(!path);
3418
3419         inode = new_inode(root->fs_info->sb);
3420         if (!inode)
3421                 return ERR_PTR(-ENOMEM);
3422
3423         if (dir) {
3424                 ret = btrfs_set_inode_index(dir, index);
3425                 if (ret)
3426                         return ERR_PTR(ret);
3427         }
3428         /*
3429          * index_cnt is ignored for everything but a dir,
3430          * btrfs_get_inode_index_count has an explanation for the magic
3431          * number
3432          */
3433         init_btrfs_i(inode);
3434         BTRFS_I(inode)->index_cnt = 2;
3435         BTRFS_I(inode)->root = root;
3436         BTRFS_I(inode)->generation = trans->transid;
3437
3438         if (mode & S_IFDIR)
3439                 owner = 0;
3440         else
3441                 owner = 1;
3442         BTRFS_I(inode)->block_group =
3443                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3444         if ((mode & S_IFREG)) {
3445                 if (btrfs_test_opt(root, NODATASUM))
3446                         btrfs_set_flag(inode, NODATASUM);
3447                 if (btrfs_test_opt(root, NODATACOW))
3448                         btrfs_set_flag(inode, NODATACOW);
3449         }
3450
3451         key[0].objectid = objectid;
3452         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3453         key[0].offset = 0;
3454
3455         key[1].objectid = objectid;
3456         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3457         key[1].offset = ref_objectid;
3458
3459         sizes[0] = sizeof(struct btrfs_inode_item);
3460         sizes[1] = name_len + sizeof(*ref);
3461
3462         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3463         if (ret != 0)
3464                 goto fail;
3465
3466         if (objectid > root->highest_inode)
3467                 root->highest_inode = objectid;
3468
3469         inode->i_uid = current_fsuid();
3470         inode->i_gid = current_fsgid();
3471         inode->i_mode = mode;
3472         inode->i_ino = objectid;
3473         inode_set_bytes(inode, 0);
3474         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3475         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3476                                   struct btrfs_inode_item);
3477         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3478
3479         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3480                              struct btrfs_inode_ref);
3481         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3482         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3483         ptr = (unsigned long)(ref + 1);
3484         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3485
3486         btrfs_mark_buffer_dirty(path->nodes[0]);
3487         btrfs_free_path(path);
3488
3489         location = &BTRFS_I(inode)->location;
3490         location->objectid = objectid;
3491         location->offset = 0;
3492         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3493
3494         insert_inode_hash(inode);
3495         return inode;
3496 fail:
3497         if (dir)
3498                 BTRFS_I(dir)->index_cnt--;
3499         btrfs_free_path(path);
3500         return ERR_PTR(ret);
3501 }
3502
3503 static inline u8 btrfs_inode_type(struct inode *inode)
3504 {
3505         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3506 }
3507
3508 /*
3509  * utility function to add 'inode' into 'parent_inode' with
3510  * a give name and a given sequence number.
3511  * if 'add_backref' is true, also insert a backref from the
3512  * inode to the parent directory.
3513  */
3514 int btrfs_add_link(struct btrfs_trans_handle *trans,
3515                    struct inode *parent_inode, struct inode *inode,
3516                    const char *name, int name_len, int add_backref, u64 index)
3517 {
3518         int ret;
3519         struct btrfs_key key;
3520         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3521
3522         key.objectid = inode->i_ino;
3523         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3524         key.offset = 0;
3525
3526         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3527                                     parent_inode->i_ino,
3528                                     &key, btrfs_inode_type(inode),
3529                                     index);
3530         if (ret == 0) {
3531                 if (add_backref) {
3532                         ret = btrfs_insert_inode_ref(trans, root,
3533                                                      name, name_len,
3534                                                      inode->i_ino,
3535                                                      parent_inode->i_ino,
3536                                                      index);
3537                 }
3538                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3539                                    name_len * 2);
3540                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3541                 ret = btrfs_update_inode(trans, root, parent_inode);
3542         }
3543         return ret;
3544 }
3545
3546 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3547                             struct dentry *dentry, struct inode *inode,
3548                             int backref, u64 index)
3549 {
3550         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3551                                  inode, dentry->d_name.name,
3552                                  dentry->d_name.len, backref, index);
3553         if (!err) {
3554                 d_instantiate(dentry, inode);
3555                 return 0;
3556         }
3557         if (err > 0)
3558                 err = -EEXIST;
3559         return err;
3560 }
3561
3562 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3563                         int mode, dev_t rdev)
3564 {
3565         struct btrfs_trans_handle *trans;
3566         struct btrfs_root *root = BTRFS_I(dir)->root;
3567         struct inode *inode = NULL;
3568         int err;
3569         int drop_inode = 0;
3570         u64 objectid;
3571         unsigned long nr = 0;
3572         u64 index = 0;
3573
3574         if (!new_valid_dev(rdev))
3575                 return -EINVAL;
3576
3577         err = btrfs_check_free_space(root, 1, 0);
3578         if (err)
3579                 goto fail;
3580
3581         trans = btrfs_start_transaction(root, 1);
3582         btrfs_set_trans_block_group(trans, dir);
3583
3584         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3585         if (err) {
3586                 err = -ENOSPC;
3587                 goto out_unlock;
3588         }
3589
3590         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3591                                 dentry->d_name.len,
3592                                 dentry->d_parent->d_inode->i_ino, objectid,
3593                                 BTRFS_I(dir)->block_group, mode, &index);
3594         err = PTR_ERR(inode);
3595         if (IS_ERR(inode))
3596                 goto out_unlock;
3597
3598         err = btrfs_init_inode_security(inode, dir);
3599         if (err) {
3600                 drop_inode = 1;
3601                 goto out_unlock;
3602         }
3603
3604         btrfs_set_trans_block_group(trans, inode);
3605         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3606         if (err)
3607                 drop_inode = 1;
3608         else {
3609                 inode->i_op = &btrfs_special_inode_operations;
3610                 init_special_inode(inode, inode->i_mode, rdev);
3611                 btrfs_update_inode(trans, root, inode);
3612         }
3613         dir->i_sb->s_dirt = 1;
3614         btrfs_update_inode_block_group(trans, inode);
3615         btrfs_update_inode_block_group(trans, dir);
3616 out_unlock:
3617         nr = trans->blocks_used;
3618         btrfs_end_transaction_throttle(trans, root);
3619 fail:
3620         if (drop_inode) {
3621                 inode_dec_link_count(inode);
3622                 iput(inode);
3623         }
3624         btrfs_btree_balance_dirty(root, nr);
3625         return err;
3626 }
3627
3628 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3629                         int mode, struct nameidata *nd)
3630 {
3631         struct btrfs_trans_handle *trans;
3632         struct btrfs_root *root = BTRFS_I(dir)->root;
3633         struct inode *inode = NULL;
3634         int err;
3635         int drop_inode = 0;
3636         unsigned long nr = 0;
3637         u64 objectid;
3638         u64 index = 0;
3639
3640         err = btrfs_check_free_space(root, 1, 0);
3641         if (err)
3642                 goto fail;
3643         trans = btrfs_start_transaction(root, 1);
3644         btrfs_set_trans_block_group(trans, dir);
3645
3646         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3647         if (err) {
3648                 err = -ENOSPC;
3649                 goto out_unlock;
3650         }
3651
3652         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3653                                 dentry->d_name.len,
3654                                 dentry->d_parent->d_inode->i_ino,
3655                                 objectid, BTRFS_I(dir)->block_group, mode,
3656                                 &index);
3657         err = PTR_ERR(inode);
3658         if (IS_ERR(inode))
3659                 goto out_unlock;
3660
3661         err = btrfs_init_inode_security(inode, dir);
3662         if (err) {
3663                 drop_inode = 1;
3664                 goto out_unlock;
3665         }
3666
3667         btrfs_set_trans_block_group(trans, inode);
3668         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3669         if (err)
3670                 drop_inode = 1;
3671         else {
3672                 inode->i_mapping->a_ops = &btrfs_aops;
3673                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3674                 inode->i_fop = &btrfs_file_operations;
3675                 inode->i_op = &btrfs_file_inode_operations;
3676                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3677         }
3678         dir->i_sb->s_dirt = 1;
3679         btrfs_update_inode_block_group(trans, inode);
3680         btrfs_update_inode_block_group(trans, dir);
3681 out_unlock:
3682         nr = trans->blocks_used;
3683         btrfs_end_transaction_throttle(trans, root);
3684 fail:
3685         if (drop_inode) {
3686                 inode_dec_link_count(inode);
3687                 iput(inode);
3688         }
3689         btrfs_btree_balance_dirty(root, nr);
3690         return err;
3691 }
3692
3693 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3694                       struct dentry *dentry)
3695 {
3696         struct btrfs_trans_handle *trans;
3697         struct btrfs_root *root = BTRFS_I(dir)->root;
3698         struct inode *inode = old_dentry->d_inode;
3699         u64 index;
3700         unsigned long nr = 0;
3701         int err;
3702         int drop_inode = 0;
3703
3704         if (inode->i_nlink == 0)
3705                 return -ENOENT;
3706
3707         btrfs_inc_nlink(inode);
3708         err = btrfs_check_free_space(root, 1, 0);
3709         if (err)
3710                 goto fail;
3711         err = btrfs_set_inode_index(dir, &index);
3712         if (err)
3713                 goto fail;
3714
3715         trans = btrfs_start_transaction(root, 1);
3716
3717         btrfs_set_trans_block_group(trans, dir);
3718         atomic_inc(&inode->i_count);
3719
3720         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3721
3722         if (err)
3723                 drop_inode = 1;
3724
3725         dir->i_sb->s_dirt = 1;
3726         btrfs_update_inode_block_group(trans, dir);
3727         err = btrfs_update_inode(trans, root, inode);
3728
3729         if (err)
3730                 drop_inode = 1;
3731
3732         nr = trans->blocks_used;
3733         btrfs_end_transaction_throttle(trans, root);
3734 fail:
3735         if (drop_inode) {
3736                 inode_dec_link_count(inode);
3737                 iput(inode);
3738         }
3739         btrfs_btree_balance_dirty(root, nr);
3740         return err;
3741 }
3742
3743 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3744 {
3745         struct inode *inode = NULL;
3746         struct btrfs_trans_handle *trans;
3747         struct btrfs_root *root = BTRFS_I(dir)->root;
3748         int err = 0;
3749         int drop_on_err = 0;
3750         u64 objectid = 0;
3751         u64 index = 0;
3752         unsigned long nr = 1;
3753
3754         err = btrfs_check_free_space(root, 1, 0);
3755         if (err)
3756                 goto out_unlock;
3757
3758         trans = btrfs_start_transaction(root, 1);
3759         btrfs_set_trans_block_group(trans, dir);
3760
3761         if (IS_ERR(trans)) {
3762                 err = PTR_ERR(trans);
3763                 goto out_unlock;
3764         }
3765
3766         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3767         if (err) {
3768                 err = -ENOSPC;
3769                 goto out_unlock;
3770         }
3771
3772         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3773                                 dentry->d_name.len,
3774                                 dentry->d_parent->d_inode->i_ino, objectid,
3775                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3776                                 &index);
3777         if (IS_ERR(inode)) {
3778                 err = PTR_ERR(inode);
3779                 goto out_fail;
3780         }
3781
3782         drop_on_err = 1;
3783
3784         err = btrfs_init_inode_security(inode, dir);
3785         if (err)
3786                 goto out_fail;
3787
3788         inode->i_op = &btrfs_dir_inode_operations;
3789         inode->i_fop = &btrfs_dir_file_operations;
3790         btrfs_set_trans_block_group(trans, inode);
3791
3792         btrfs_i_size_write(inode, 0);
3793         err = btrfs_update_inode(trans, root, inode);
3794         if (err)
3795                 goto out_fail;
3796
3797         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3798                                  inode, dentry->d_name.name,
3799                                  dentry->d_name.len, 0, index);
3800         if (err)
3801                 goto out_fail;
3802
3803         d_instantiate(dentry, inode);
3804         drop_on_err = 0;
3805         dir->i_sb->s_dirt = 1;
3806         btrfs_update_inode_block_group(trans, inode);
3807         btrfs_update_inode_block_group(trans, dir);
3808
3809 out_fail:
3810         nr = trans->blocks_used;
3811         btrfs_end_transaction_throttle(trans, root);
3812
3813 out_unlock:
3814         if (drop_on_err)
3815                 iput(inode);
3816         btrfs_btree_balance_dirty(root, nr);
3817         return err;
3818 }
3819
3820 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3821  * and an extent that you want to insert, deal with overlap and insert
3822  * the new extent into the tree.
3823  */
3824 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3825                                 struct extent_map *existing,
3826                                 struct extent_map *em,
3827                                 u64 map_start, u64 map_len)
3828 {
3829         u64 start_diff;
3830
3831         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3832         start_diff = map_start - em->start;
3833         em->start = map_start;
3834         em->len = map_len;
3835         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3836             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3837                 em->block_start += start_diff;
3838                 em->block_len -= start_diff;
3839         }
3840         return add_extent_mapping(em_tree, em);
3841 }
3842
3843 static noinline int uncompress_inline(struct btrfs_path *path,
3844                                       struct inode *inode, struct page *page,
3845                                       size_t pg_offset, u64 extent_offset,
3846                                       struct btrfs_file_extent_item *item)
3847 {
3848         int ret;
3849         struct extent_buffer *leaf = path->nodes[0];
3850         char *tmp;
3851         size_t max_size;
3852         unsigned long inline_size;
3853         unsigned long ptr;
3854
3855         WARN_ON(pg_offset != 0);
3856         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3857         inline_size = btrfs_file_extent_inline_item_len(leaf,
3858                                         btrfs_item_nr(leaf, path->slots[0]));
3859         tmp = kmalloc(inline_size, GFP_NOFS);
3860         ptr = btrfs_file_extent_inline_start(item);
3861
3862         read_extent_buffer(leaf, tmp, ptr, inline_size);
3863
3864         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3865         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3866                                     inline_size, max_size);
3867         if (ret) {
3868                 char *kaddr = kmap_atomic(page, KM_USER0);
3869                 unsigned long copy_size = min_t(u64,
3870                                   PAGE_CACHE_SIZE - pg_offset,
3871                                   max_size - extent_offset);
3872                 memset(kaddr + pg_offset, 0, copy_size);
3873                 kunmap_atomic(kaddr, KM_USER0);
3874         }
3875         kfree(tmp);
3876         return 0;
3877 }
3878
3879 /*
3880  * a bit scary, this does extent mapping from logical file offset to the disk.
3881  * the ugly parts come from merging extents from the disk with the in-ram
3882  * representation.  This gets more complex because of the data=ordered code,
3883  * where the in-ram extents might be locked pending data=ordered completion.
3884  *
3885  * This also copies inline extents directly into the page.
3886  */
3887
3888 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3889                                     size_t pg_offset, u64 start, u64 len,
3890                                     int create)
3891 {
3892         int ret;
3893         int err = 0;
3894         u64 bytenr;
3895         u64 extent_start = 0;
3896         u64 extent_end = 0;
3897         u64 objectid = inode->i_ino;
3898         u32 found_type;
3899         struct btrfs_path *path = NULL;
3900         struct btrfs_root *root = BTRFS_I(inode)->root;
3901         struct btrfs_file_extent_item *item;
3902         struct extent_buffer *leaf;
3903         struct btrfs_key found_key;
3904         struct extent_map *em = NULL;
3905         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3906         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3907         struct btrfs_trans_handle *trans = NULL;
3908         int compressed;
3909
3910 again:
3911         spin_lock(&em_tree->lock);
3912         em = lookup_extent_mapping(em_tree, start, len);
3913         if (em)
3914                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3915         spin_unlock(&em_tree->lock);
3916
3917         if (em) {
3918                 if (em->start > start || em->start + em->len <= start)
3919                         free_extent_map(em);
3920                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3921                         free_extent_map(em);
3922                 else
3923                         goto out;
3924         }
3925         em = alloc_extent_map(GFP_NOFS);
3926         if (!em) {
3927                 err = -ENOMEM;
3928                 goto out;
3929         }
3930         em->bdev = root->fs_info->fs_devices->latest_bdev;
3931         em->start = EXTENT_MAP_HOLE;
3932         em->orig_start = EXTENT_MAP_HOLE;
3933         em->len = (u64)-1;
3934         em->block_len = (u64)-1;
3935
3936         if (!path) {
3937                 path = btrfs_alloc_path();
3938                 BUG_ON(!path);
3939         }
3940
3941         ret = btrfs_lookup_file_extent(trans, root, path,
3942                                        objectid, start, trans != NULL);
3943         if (ret < 0) {
3944                 err = ret;
3945                 goto out;
3946         }
3947
3948         if (ret != 0) {
3949                 if (path->slots[0] == 0)
3950                         goto not_found;
3951                 path->slots[0]--;
3952         }
3953
3954         leaf = path->nodes[0];
3955         item = btrfs_item_ptr(leaf, path->slots[0],
3956                               struct btrfs_file_extent_item);
3957         /* are we inside the extent that was found? */
3958         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3959         found_type = btrfs_key_type(&found_key);
3960         if (found_key.objectid != objectid ||
3961             found_type != BTRFS_EXTENT_DATA_KEY) {
3962                 goto not_found;
3963         }
3964
3965         found_type = btrfs_file_extent_type(leaf, item);
3966         extent_start = found_key.offset;
3967         compressed = btrfs_file_extent_compression(leaf, item);
3968         if (found_type == BTRFS_FILE_EXTENT_REG ||
3969             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3970                 extent_end = extent_start +
3971                        btrfs_file_extent_num_bytes(leaf, item);
3972         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3973                 size_t size;
3974                 size = btrfs_file_extent_inline_len(leaf, item);
3975                 extent_end = (extent_start + size + root->sectorsize - 1) &
3976                         ~((u64)root->sectorsize - 1);
3977         }
3978
3979         if (start >= extent_end) {
3980                 path->slots[0]++;
3981                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3982                         ret = btrfs_next_leaf(root, path);
3983                         if (ret < 0) {
3984                                 err = ret;
3985                                 goto out;
3986                         }
3987                         if (ret > 0)
3988                                 goto not_found;
3989                         leaf = path->nodes[0];
3990                 }
3991                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3992                 if (found_key.objectid != objectid ||
3993                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3994                         goto not_found;
3995                 if (start + len <= found_key.offset)
3996                         goto not_found;
3997                 em->start = start;
3998                 em->len = found_key.offset - start;
3999                 goto not_found_em;
4000         }
4001
4002         if (found_type == BTRFS_FILE_EXTENT_REG ||
4003             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4004                 em->start = extent_start;
4005                 em->len = extent_end - extent_start;
4006                 em->orig_start = extent_start -
4007                                  btrfs_file_extent_offset(leaf, item);
4008                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4009                 if (bytenr == 0) {
4010                         em->block_start = EXTENT_MAP_HOLE;
4011                         goto insert;
4012                 }
4013                 if (compressed) {
4014                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4015                         em->block_start = bytenr;
4016                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4017                                                                          item);
4018                 } else {
4019                         bytenr += btrfs_file_extent_offset(leaf, item);
4020                         em->block_start = bytenr;
4021                         em->block_len = em->len;
4022                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4023                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4024                 }
4025                 goto insert;
4026         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4027                 unsigned long ptr;
4028                 char *map;
4029                 size_t size;
4030                 size_t extent_offset;
4031                 size_t copy_size;
4032
4033                 em->block_start = EXTENT_MAP_INLINE;
4034                 if (!page || create) {
4035                         em->start = extent_start;
4036                         em->len = extent_end - extent_start;
4037                         goto out;
4038                 }
4039
4040                 size = btrfs_file_extent_inline_len(leaf, item);
4041                 extent_offset = page_offset(page) + pg_offset - extent_start;
4042                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4043                                 size - extent_offset);
4044                 em->start = extent_start + extent_offset;
4045                 em->len = (copy_size + root->sectorsize - 1) &
4046                         ~((u64)root->sectorsize - 1);
4047                 em->orig_start = EXTENT_MAP_INLINE;
4048                 if (compressed)
4049                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4050                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4051                 if (create == 0 && !PageUptodate(page)) {
4052                         if (btrfs_file_extent_compression(leaf, item) ==
4053                             BTRFS_COMPRESS_ZLIB) {
4054                                 ret = uncompress_inline(path, inode, page,
4055                                                         pg_offset,
4056                                                         extent_offset, item);
4057                                 BUG_ON(ret);
4058                         } else {
4059                                 map = kmap(page);
4060                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4061                                                    copy_size);
4062                                 kunmap(page);
4063                         }
4064                         flush_dcache_page(page);
4065                 } else if (create && PageUptodate(page)) {
4066                         if (!trans) {
4067                                 kunmap(page);
4068                                 free_extent_map(em);
4069                                 em = NULL;
4070                                 btrfs_release_path(root, path);
4071                                 trans = btrfs_join_transaction(root, 1);
4072                                 goto again;
4073                         }
4074                         map = kmap(page);
4075                         write_extent_buffer(leaf, map + pg_offset, ptr,
4076                                             copy_size);
4077                         kunmap(page);
4078                         btrfs_mark_buffer_dirty(leaf);
4079                 }
4080                 set_extent_uptodate(io_tree, em->start,
4081                                     extent_map_end(em) - 1, GFP_NOFS);
4082                 goto insert;
4083         } else {
4084                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4085                 WARN_ON(1);
4086         }
4087 not_found:
4088         em->start = start;
4089         em->len = len;
4090 not_found_em:
4091         em->block_start = EXTENT_MAP_HOLE;
4092         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4093 insert:
4094         btrfs_release_path(root, path);
4095         if (em->start > start || extent_map_end(em) <= start) {
4096                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4097                        "[%llu %llu]\n", (unsigned long long)em->start,
4098                        (unsigned long long)em->len,
4099                        (unsigned long long)start,
4100                        (unsigned long long)len);
4101                 err = -EIO;
4102                 goto out;
4103         }
4104
4105         err = 0;
4106         spin_lock(&em_tree->lock);
4107         ret = add_extent_mapping(em_tree, em);
4108         /* it is possible that someone inserted the extent into the tree
4109          * while we had the lock dropped.  It is also possible that
4110          * an overlapping map exists in the tree
4111          */
4112         if (ret == -EEXIST) {
4113                 struct extent_map *existing;
4114
4115                 ret = 0;
4116
4117                 existing = lookup_extent_mapping(em_tree, start, len);
4118                 if (existing && (existing->start > start ||
4119                     existing->start + existing->len <= start)) {
4120                         free_extent_map(existing);
4121                         existing = NULL;
4122                 }
4123                 if (!existing) {
4124                         existing = lookup_extent_mapping(em_tree, em->start,
4125                                                          em->len);
4126                         if (existing) {
4127                                 err = merge_extent_mapping(em_tree, existing,
4128                                                            em, start,
4129                                                            root->sectorsize);
4130                                 free_extent_map(existing);
4131                                 if (err) {
4132                                         free_extent_map(em);
4133                                         em = NULL;
4134                                 }
4135                         } else {
4136                                 err = -EIO;
4137                                 free_extent_map(em);
4138                                 em = NULL;
4139                         }
4140                 } else {
4141                         free_extent_map(em);
4142                         em = existing;
4143                         err = 0;
4144                 }
4145         }
4146         spin_unlock(&em_tree->lock);
4147 out:
4148         if (path)
4149                 btrfs_free_path(path);
4150         if (trans) {
4151                 ret = btrfs_end_transaction(trans, root);
4152                 if (!err)
4153                         err = ret;
4154         }
4155         if (err) {
4156                 free_extent_map(em);
4157                 WARN_ON(1);
4158                 return ERR_PTR(err);
4159         }
4160         return em;
4161 }
4162
4163 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4164                         const struct iovec *iov, loff_t offset,
4165                         unsigned long nr_segs)
4166 {
4167         return -EINVAL;
4168 }
4169
4170 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4171                 __u64 start, __u64 len)
4172 {
4173         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4174 }
4175
4176 int btrfs_readpage(struct file *file, struct page *page)
4177 {
4178         struct extent_io_tree *tree;
4179         tree = &BTRFS_I(page->mapping->host)->io_tree;
4180         return extent_read_full_page(tree, page, btrfs_get_extent);
4181 }
4182
4183 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4184 {
4185         struct extent_io_tree *tree;
4186
4187
4188         if (current->flags & PF_MEMALLOC) {
4189                 redirty_page_for_writepage(wbc, page);
4190                 unlock_page(page);
4191                 return 0;
4192         }
4193         tree = &BTRFS_I(page->mapping->host)->io_tree;
4194         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4195 }
4196
4197 int btrfs_writepages(struct address_space *mapping,
4198                      struct writeback_control *wbc)
4199 {
4200         struct extent_io_tree *tree;
4201
4202         tree = &BTRFS_I(mapping->host)->io_tree;
4203         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4204 }
4205
4206 static int
4207 btrfs_readpages(struct file *file, struct address_space *mapping,
4208                 struct list_head *pages, unsigned nr_pages)
4209 {
4210         struct extent_io_tree *tree;
4211         tree = &BTRFS_I(mapping->host)->io_tree;
4212         return extent_readpages(tree, mapping, pages, nr_pages,
4213                                 btrfs_get_extent);
4214 }
4215 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4216 {
4217         struct extent_io_tree *tree;
4218         struct extent_map_tree *map;
4219         int ret;
4220
4221         tree = &BTRFS_I(page->mapping->host)->io_tree;
4222         map = &BTRFS_I(page->mapping->host)->extent_tree;
4223         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4224         if (ret == 1) {
4225                 ClearPagePrivate(page);
4226                 set_page_private(page, 0);
4227                 page_cache_release(page);
4228         }
4229         return ret;
4230 }
4231
4232 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4233 {
4234         if (PageWriteback(page) || PageDirty(page))
4235                 return 0;
4236         return __btrfs_releasepage(page, gfp_flags);
4237 }
4238
4239 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4240 {
4241         struct extent_io_tree *tree;
4242         struct btrfs_ordered_extent *ordered;
4243         u64 page_start = page_offset(page);
4244         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4245
4246         wait_on_page_writeback(page);
4247         tree = &BTRFS_I(page->mapping->host)->io_tree;
4248         if (offset) {
4249                 btrfs_releasepage(page, GFP_NOFS);
4250                 return;
4251         }
4252
4253         lock_extent(tree, page_start, page_end, GFP_NOFS);
4254         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4255                                            page_offset(page));
4256         if (ordered) {
4257                 /*
4258                  * IO on this page will never be started, so we need
4259                  * to account for any ordered extents now
4260                  */
4261                 clear_extent_bit(tree, page_start, page_end,
4262                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4263                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4264                 btrfs_finish_ordered_io(page->mapping->host,
4265                                         page_start, page_end);
4266                 btrfs_put_ordered_extent(ordered);
4267                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4268         }
4269         clear_extent_bit(tree, page_start, page_end,
4270                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4271                  EXTENT_ORDERED,
4272                  1, 1, GFP_NOFS);
4273         __btrfs_releasepage(page, GFP_NOFS);
4274
4275         ClearPageChecked(page);
4276         if (PagePrivate(page)) {
4277                 ClearPagePrivate(page);
4278                 set_page_private(page, 0);
4279                 page_cache_release(page);
4280         }
4281 }
4282
4283 /*
4284  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4285  * called from a page fault handler when a page is first dirtied. Hence we must
4286  * be careful to check for EOF conditions here. We set the page up correctly
4287  * for a written page which means we get ENOSPC checking when writing into
4288  * holes and correct delalloc and unwritten extent mapping on filesystems that
4289  * support these features.
4290  *
4291  * We are not allowed to take the i_mutex here so we have to play games to
4292  * protect against truncate races as the page could now be beyond EOF.  Because
4293  * vmtruncate() writes the inode size before removing pages, once we have the
4294  * page lock we can determine safely if the page is beyond EOF. If it is not
4295  * beyond EOF, then the page is guaranteed safe against truncation until we
4296  * unlock the page.
4297  */
4298 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4299 {
4300         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4301         struct btrfs_root *root = BTRFS_I(inode)->root;
4302         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4303         struct btrfs_ordered_extent *ordered;
4304         char *kaddr;
4305         unsigned long zero_start;
4306         loff_t size;
4307         int ret;
4308         u64 page_start;
4309         u64 page_end;
4310
4311         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4312         if (ret)
4313                 goto out;
4314
4315         ret = -EINVAL;
4316 again:
4317         lock_page(page);
4318         size = i_size_read(inode);
4319         page_start = page_offset(page);
4320         page_end = page_start + PAGE_CACHE_SIZE - 1;
4321
4322         if ((page->mapping != inode->i_mapping) ||
4323             (page_start >= size)) {
4324                 /* page got truncated out from underneath us */
4325                 goto out_unlock;
4326         }
4327         wait_on_page_writeback(page);
4328
4329         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4330         set_page_extent_mapped(page);
4331
4332         /*
4333          * we can't set the delalloc bits if there are pending ordered
4334          * extents.  Drop our locks and wait for them to finish
4335          */
4336         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4337         if (ordered) {
4338                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4339                 unlock_page(page);
4340                 btrfs_start_ordered_extent(inode, ordered, 1);
4341                 btrfs_put_ordered_extent(ordered);
4342                 goto again;
4343         }
4344
4345         btrfs_set_extent_delalloc(inode, page_start, page_end);
4346         ret = 0;
4347
4348         /* page is wholly or partially inside EOF */
4349         if (page_start + PAGE_CACHE_SIZE > size)
4350                 zero_start = size & ~PAGE_CACHE_MASK;
4351         else
4352                 zero_start = PAGE_CACHE_SIZE;
4353
4354         if (zero_start != PAGE_CACHE_SIZE) {
4355                 kaddr = kmap(page);
4356                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4357                 flush_dcache_page(page);
4358                 kunmap(page);
4359         }
4360         ClearPageChecked(page);
4361         set_page_dirty(page);
4362         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4363
4364 out_unlock:
4365         unlock_page(page);
4366 out:
4367         return ret;
4368 }
4369
4370 static void btrfs_truncate(struct inode *inode)
4371 {
4372         struct btrfs_root *root = BTRFS_I(inode)->root;
4373         int ret;
4374         struct btrfs_trans_handle *trans;
4375         unsigned long nr;
4376         u64 mask = root->sectorsize - 1;
4377
4378         if (!S_ISREG(inode->i_mode))
4379                 return;
4380         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4381                 return;
4382
4383         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4384         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4385
4386         trans = btrfs_start_transaction(root, 1);
4387         btrfs_set_trans_block_group(trans, inode);
4388         btrfs_i_size_write(inode, inode->i_size);
4389
4390         ret = btrfs_orphan_add(trans, inode);
4391         if (ret)
4392                 goto out;
4393         /* FIXME, add redo link to tree so we don't leak on crash */
4394         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4395                                       BTRFS_EXTENT_DATA_KEY);
4396         btrfs_update_inode(trans, root, inode);
4397
4398         ret = btrfs_orphan_del(trans, inode);
4399         BUG_ON(ret);
4400
4401 out:
4402         nr = trans->blocks_used;
4403         ret = btrfs_end_transaction_throttle(trans, root);
4404         BUG_ON(ret);
4405         btrfs_btree_balance_dirty(root, nr);
4406 }
4407
4408 /*
4409  * create a new subvolume directory/inode (helper for the ioctl).
4410  */
4411 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4412                              struct btrfs_root *new_root, struct dentry *dentry,
4413                              u64 new_dirid, u64 alloc_hint)
4414 {
4415         struct inode *inode;
4416         int error;
4417         u64 index = 0;
4418
4419         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4420                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4421         if (IS_ERR(inode))
4422                 return PTR_ERR(inode);
4423         inode->i_op = &btrfs_dir_inode_operations;
4424         inode->i_fop = &btrfs_dir_file_operations;
4425
4426         inode->i_nlink = 1;
4427         btrfs_i_size_write(inode, 0);
4428
4429         error = btrfs_update_inode(trans, new_root, inode);
4430         if (error)
4431                 return error;
4432
4433         d_instantiate(dentry, inode);
4434         return 0;
4435 }
4436
4437 /* helper function for file defrag and space balancing.  This
4438  * forces readahead on a given range of bytes in an inode
4439  */
4440 unsigned long btrfs_force_ra(struct address_space *mapping,
4441                               struct file_ra_state *ra, struct file *file,
4442                               pgoff_t offset, pgoff_t last_index)
4443 {
4444         pgoff_t req_size = last_index - offset + 1;
4445
4446         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4447         return offset + req_size;
4448 }
4449
4450 struct inode *btrfs_alloc_inode(struct super_block *sb)
4451 {
4452         struct btrfs_inode *ei;
4453
4454         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4455         if (!ei)
4456                 return NULL;
4457         ei->last_trans = 0;
4458         ei->logged_trans = 0;
4459         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4460         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4461         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4462         INIT_LIST_HEAD(&ei->i_orphan);
4463         return &ei->vfs_inode;
4464 }
4465
4466 void btrfs_destroy_inode(struct inode *inode)
4467 {
4468         struct btrfs_ordered_extent *ordered;
4469         WARN_ON(!list_empty(&inode->i_dentry));
4470         WARN_ON(inode->i_data.nrpages);
4471
4472         if (BTRFS_I(inode)->i_acl &&
4473             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4474                 posix_acl_release(BTRFS_I(inode)->i_acl);
4475         if (BTRFS_I(inode)->i_default_acl &&
4476             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4477                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4478
4479         spin_lock(&BTRFS_I(inode)->root->list_lock);
4480         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4481                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4482                        " list\n", inode->i_ino);
4483                 dump_stack();
4484         }
4485         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4486
4487         while (1) {
4488                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4489                 if (!ordered)
4490                         break;
4491                 else {
4492                         printk(KERN_ERR "btrfs found ordered "
4493                                "extent %llu %llu on inode cleanup\n",
4494                                (unsigned long long)ordered->file_offset,
4495                                (unsigned long long)ordered->len);
4496                         btrfs_remove_ordered_extent(inode, ordered);
4497                         btrfs_put_ordered_extent(ordered);
4498                         btrfs_put_ordered_extent(ordered);
4499                 }
4500         }
4501         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4502         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4503 }
4504
4505 static void init_once(void *foo)
4506 {
4507         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4508
4509         inode_init_once(&ei->vfs_inode);
4510 }
4511
4512 void btrfs_destroy_cachep(void)
4513 {
4514         if (btrfs_inode_cachep)
4515                 kmem_cache_destroy(btrfs_inode_cachep);
4516         if (btrfs_trans_handle_cachep)
4517                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4518         if (btrfs_transaction_cachep)
4519                 kmem_cache_destroy(btrfs_transaction_cachep);
4520         if (btrfs_bit_radix_cachep)
4521                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4522         if (btrfs_path_cachep)
4523                 kmem_cache_destroy(btrfs_path_cachep);
4524 }
4525
4526 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4527                                        unsigned long extra_flags,
4528                                        void (*ctor)(void *))
4529 {
4530         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4531                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4532 }
4533
4534 int btrfs_init_cachep(void)
4535 {
4536         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4537                                           sizeof(struct btrfs_inode),
4538                                           0, init_once);
4539         if (!btrfs_inode_cachep)
4540                 goto fail;
4541         btrfs_trans_handle_cachep =
4542                         btrfs_cache_create("btrfs_trans_handle_cache",
4543                                            sizeof(struct btrfs_trans_handle),
4544                                            0, NULL);
4545         if (!btrfs_trans_handle_cachep)
4546                 goto fail;
4547         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4548                                              sizeof(struct btrfs_transaction),
4549                                              0, NULL);
4550         if (!btrfs_transaction_cachep)
4551                 goto fail;
4552         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4553                                          sizeof(struct btrfs_path),
4554                                          0, NULL);
4555         if (!btrfs_path_cachep)
4556                 goto fail;
4557         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4558                                               SLAB_DESTROY_BY_RCU, NULL);
4559         if (!btrfs_bit_radix_cachep)
4560                 goto fail;
4561         return 0;
4562 fail:
4563         btrfs_destroy_cachep();
4564         return -ENOMEM;
4565 }
4566
4567 static int btrfs_getattr(struct vfsmount *mnt,
4568                          struct dentry *dentry, struct kstat *stat)
4569 {
4570         struct inode *inode = dentry->d_inode;
4571         generic_fillattr(inode, stat);
4572         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4573         stat->blksize = PAGE_CACHE_SIZE;
4574         stat->blocks = (inode_get_bytes(inode) +
4575                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4576         return 0;
4577 }
4578
4579 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4580                            struct inode *new_dir, struct dentry *new_dentry)
4581 {
4582         struct btrfs_trans_handle *trans;
4583         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4584         struct inode *new_inode = new_dentry->d_inode;
4585         struct inode *old_inode = old_dentry->d_inode;
4586         struct timespec ctime = CURRENT_TIME;
4587         u64 index = 0;
4588         int ret;
4589
4590         /* we're not allowed to rename between subvolumes */
4591         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4592             BTRFS_I(new_dir)->root->root_key.objectid)
4593                 return -EXDEV;
4594
4595         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4596             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4597                 return -ENOTEMPTY;
4598         }
4599
4600         /* to rename a snapshot or subvolume, we need to juggle the
4601          * backrefs.  This isn't coded yet
4602          */
4603         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4604                 return -EXDEV;
4605
4606         ret = btrfs_check_free_space(root, 1, 0);
4607         if (ret)
4608                 goto out_unlock;
4609
4610         trans = btrfs_start_transaction(root, 1);
4611
4612         btrfs_set_trans_block_group(trans, new_dir);
4613
4614         btrfs_inc_nlink(old_dentry->d_inode);
4615         old_dir->i_ctime = old_dir->i_mtime = ctime;
4616         new_dir->i_ctime = new_dir->i_mtime = ctime;
4617         old_inode->i_ctime = ctime;
4618
4619         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4620                                  old_dentry->d_name.name,
4621                                  old_dentry->d_name.len);
4622         if (ret)
4623                 goto out_fail;
4624
4625         if (new_inode) {
4626                 new_inode->i_ctime = CURRENT_TIME;
4627                 ret = btrfs_unlink_inode(trans, root, new_dir,
4628                                          new_dentry->d_inode,
4629                                          new_dentry->d_name.name,
4630                                          new_dentry->d_name.len);
4631                 if (ret)
4632                         goto out_fail;
4633                 if (new_inode->i_nlink == 0) {
4634                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4635                         if (ret)
4636                                 goto out_fail;
4637                 }
4638
4639         }
4640         ret = btrfs_set_inode_index(new_dir, &index);
4641         if (ret)
4642                 goto out_fail;
4643
4644         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4645                              old_inode, new_dentry->d_name.name,
4646                              new_dentry->d_name.len, 1, index);
4647         if (ret)
4648                 goto out_fail;
4649
4650 out_fail:
4651         btrfs_end_transaction_throttle(trans, root);
4652 out_unlock:
4653         return ret;
4654 }
4655
4656 /*
4657  * some fairly slow code that needs optimization. This walks the list
4658  * of all the inodes with pending delalloc and forces them to disk.
4659  */
4660 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4661 {
4662         struct list_head *head = &root->fs_info->delalloc_inodes;
4663         struct btrfs_inode *binode;
4664         struct inode *inode;
4665
4666         if (root->fs_info->sb->s_flags & MS_RDONLY)
4667                 return -EROFS;
4668
4669         spin_lock(&root->fs_info->delalloc_lock);
4670         while (!list_empty(head)) {
4671                 binode = list_entry(head->next, struct btrfs_inode,
4672                                     delalloc_inodes);
4673                 inode = igrab(&binode->vfs_inode);
4674                 if (!inode)
4675                         list_del_init(&binode->delalloc_inodes);
4676                 spin_unlock(&root->fs_info->delalloc_lock);
4677                 if (inode) {
4678                         filemap_flush(inode->i_mapping);
4679                         iput(inode);
4680                 }
4681                 cond_resched();
4682                 spin_lock(&root->fs_info->delalloc_lock);
4683         }
4684         spin_unlock(&root->fs_info->delalloc_lock);
4685
4686         /* the filemap_flush will queue IO into the worker threads, but
4687          * we have to make sure the IO is actually started and that
4688          * ordered extents get created before we return
4689          */
4690         atomic_inc(&root->fs_info->async_submit_draining);
4691         while (atomic_read(&root->fs_info->nr_async_submits) ||
4692               atomic_read(&root->fs_info->async_delalloc_pages)) {
4693                 wait_event(root->fs_info->async_submit_wait,
4694                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4695                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4696         }
4697         atomic_dec(&root->fs_info->async_submit_draining);
4698         return 0;
4699 }
4700
4701 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4702                          const char *symname)
4703 {
4704         struct btrfs_trans_handle *trans;
4705         struct btrfs_root *root = BTRFS_I(dir)->root;
4706         struct btrfs_path *path;
4707         struct btrfs_key key;
4708         struct inode *inode = NULL;
4709         int err;
4710         int drop_inode = 0;
4711         u64 objectid;
4712         u64 index = 0 ;
4713         int name_len;
4714         int datasize;
4715         unsigned long ptr;
4716         struct btrfs_file_extent_item *ei;
4717         struct extent_buffer *leaf;
4718         unsigned long nr = 0;
4719
4720         name_len = strlen(symname) + 1;
4721         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4722                 return -ENAMETOOLONG;
4723
4724         err = btrfs_check_free_space(root, 1, 0);
4725         if (err)
4726                 goto out_fail;
4727
4728         trans = btrfs_start_transaction(root, 1);
4729         btrfs_set_trans_block_group(trans, dir);
4730
4731         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4732         if (err) {
4733                 err = -ENOSPC;
4734                 goto out_unlock;
4735         }
4736
4737         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4738                                 dentry->d_name.len,
4739                                 dentry->d_parent->d_inode->i_ino, objectid,
4740                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4741                                 &index);
4742         err = PTR_ERR(inode);
4743         if (IS_ERR(inode))
4744                 goto out_unlock;
4745
4746         err = btrfs_init_inode_security(inode, dir);
4747         if (err) {
4748                 drop_inode = 1;
4749                 goto out_unlock;
4750         }
4751
4752         btrfs_set_trans_block_group(trans, inode);
4753         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4754         if (err)
4755                 drop_inode = 1;
4756         else {
4757                 inode->i_mapping->a_ops = &btrfs_aops;
4758                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4759                 inode->i_fop = &btrfs_file_operations;
4760                 inode->i_op = &btrfs_file_inode_operations;
4761                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4762         }
4763         dir->i_sb->s_dirt = 1;
4764         btrfs_update_inode_block_group(trans, inode);
4765         btrfs_update_inode_block_group(trans, dir);
4766         if (drop_inode)
4767                 goto out_unlock;
4768
4769         path = btrfs_alloc_path();
4770         BUG_ON(!path);
4771         key.objectid = inode->i_ino;
4772         key.offset = 0;
4773         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4774         datasize = btrfs_file_extent_calc_inline_size(name_len);
4775         err = btrfs_insert_empty_item(trans, root, path, &key,
4776                                       datasize);
4777         if (err) {
4778                 drop_inode = 1;
4779                 goto out_unlock;
4780         }
4781         leaf = path->nodes[0];
4782         ei = btrfs_item_ptr(leaf, path->slots[0],
4783                             struct btrfs_file_extent_item);
4784         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4785         btrfs_set_file_extent_type(leaf, ei,
4786                                    BTRFS_FILE_EXTENT_INLINE);
4787         btrfs_set_file_extent_encryption(leaf, ei, 0);
4788         btrfs_set_file_extent_compression(leaf, ei, 0);
4789         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4790         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4791
4792         ptr = btrfs_file_extent_inline_start(ei);
4793         write_extent_buffer(leaf, symname, ptr, name_len);
4794         btrfs_mark_buffer_dirty(leaf);
4795         btrfs_free_path(path);
4796
4797         inode->i_op = &btrfs_symlink_inode_operations;
4798         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4799         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4800         inode_set_bytes(inode, name_len);
4801         btrfs_i_size_write(inode, name_len - 1);
4802         err = btrfs_update_inode(trans, root, inode);
4803         if (err)
4804                 drop_inode = 1;
4805
4806 out_unlock:
4807         nr = trans->blocks_used;
4808         btrfs_end_transaction_throttle(trans, root);
4809 out_fail:
4810         if (drop_inode) {
4811                 inode_dec_link_count(inode);
4812                 iput(inode);
4813         }
4814         btrfs_btree_balance_dirty(root, nr);
4815         return err;
4816 }
4817
4818 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4819                                u64 alloc_hint, int mode)
4820 {
4821         struct btrfs_trans_handle *trans;
4822         struct btrfs_root *root = BTRFS_I(inode)->root;
4823         struct btrfs_key ins;
4824         u64 alloc_size;
4825         u64 cur_offset = start;
4826         u64 num_bytes = end - start;
4827         int ret = 0;
4828
4829         trans = btrfs_join_transaction(root, 1);
4830         BUG_ON(!trans);
4831         btrfs_set_trans_block_group(trans, inode);
4832
4833         while (num_bytes > 0) {
4834                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4835                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4836                                            root->sectorsize, 0, alloc_hint,
4837                                            (u64)-1, &ins, 1);
4838                 if (ret) {
4839                         WARN_ON(1);
4840                         goto out;
4841                 }
4842                 ret = insert_reserved_file_extent(trans, inode,
4843                                                   cur_offset, ins.objectid,
4844                                                   ins.offset, ins.offset,
4845                                                   ins.offset, 0, 0, 0,
4846                                                   BTRFS_FILE_EXTENT_PREALLOC);
4847                 BUG_ON(ret);
4848                 num_bytes -= ins.offset;
4849                 cur_offset += ins.offset;
4850                 alloc_hint = ins.objectid + ins.offset;
4851         }
4852 out:
4853         if (cur_offset > start) {
4854                 inode->i_ctime = CURRENT_TIME;
4855                 btrfs_set_flag(inode, PREALLOC);
4856                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4857                     cur_offset > i_size_read(inode))
4858                         btrfs_i_size_write(inode, cur_offset);
4859                 ret = btrfs_update_inode(trans, root, inode);
4860                 BUG_ON(ret);
4861         }
4862
4863         btrfs_end_transaction(trans, root);
4864         return ret;
4865 }
4866
4867 static long btrfs_fallocate(struct inode *inode, int mode,
4868                             loff_t offset, loff_t len)
4869 {
4870         u64 cur_offset;
4871         u64 last_byte;
4872         u64 alloc_start;
4873         u64 alloc_end;
4874         u64 alloc_hint = 0;
4875         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4876         struct extent_map *em;
4877         int ret;
4878
4879         alloc_start = offset & ~mask;
4880         alloc_end =  (offset + len + mask) & ~mask;
4881
4882         mutex_lock(&inode->i_mutex);
4883         if (alloc_start > inode->i_size) {
4884                 ret = btrfs_cont_expand(inode, alloc_start);
4885                 if (ret)
4886                         goto out;
4887         }
4888
4889         while (1) {
4890                 struct btrfs_ordered_extent *ordered;
4891                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4892                             alloc_end - 1, GFP_NOFS);
4893                 ordered = btrfs_lookup_first_ordered_extent(inode,
4894                                                             alloc_end - 1);
4895                 if (ordered &&
4896                     ordered->file_offset + ordered->len > alloc_start &&
4897                     ordered->file_offset < alloc_end) {
4898                         btrfs_put_ordered_extent(ordered);
4899                         unlock_extent(&BTRFS_I(inode)->io_tree,
4900                                       alloc_start, alloc_end - 1, GFP_NOFS);
4901                         btrfs_wait_ordered_range(inode, alloc_start,
4902                                                  alloc_end - alloc_start);
4903                 } else {
4904                         if (ordered)
4905                                 btrfs_put_ordered_extent(ordered);
4906                         break;
4907                 }
4908         }
4909
4910         cur_offset = alloc_start;
4911         while (1) {
4912                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4913                                       alloc_end - cur_offset, 0);
4914                 BUG_ON(IS_ERR(em) || !em);
4915                 last_byte = min(extent_map_end(em), alloc_end);
4916                 last_byte = (last_byte + mask) & ~mask;
4917                 if (em->block_start == EXTENT_MAP_HOLE) {
4918                         ret = prealloc_file_range(inode, cur_offset,
4919                                         last_byte, alloc_hint, mode);
4920                         if (ret < 0) {
4921                                 free_extent_map(em);
4922                                 break;
4923                         }
4924                 }
4925                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4926                         alloc_hint = em->block_start;
4927                 free_extent_map(em);
4928
4929                 cur_offset = last_byte;
4930                 if (cur_offset >= alloc_end) {
4931                         ret = 0;
4932                         break;
4933                 }
4934         }
4935         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4936                       GFP_NOFS);
4937 out:
4938         mutex_unlock(&inode->i_mutex);
4939         return ret;
4940 }
4941
4942 static int btrfs_set_page_dirty(struct page *page)
4943 {
4944         return __set_page_dirty_nobuffers(page);
4945 }
4946
4947 static int btrfs_permission(struct inode *inode, int mask)
4948 {
4949         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4950                 return -EACCES;
4951         return generic_permission(inode, mask, btrfs_check_acl);
4952 }
4953
4954 static struct inode_operations btrfs_dir_inode_operations = {
4955         .getattr        = btrfs_getattr,
4956         .lookup         = btrfs_lookup,
4957         .create         = btrfs_create,
4958         .unlink         = btrfs_unlink,
4959         .link           = btrfs_link,
4960         .mkdir          = btrfs_mkdir,
4961         .rmdir          = btrfs_rmdir,
4962         .rename         = btrfs_rename,
4963         .symlink        = btrfs_symlink,
4964         .setattr        = btrfs_setattr,
4965         .mknod          = btrfs_mknod,
4966         .setxattr       = btrfs_setxattr,
4967         .getxattr       = btrfs_getxattr,
4968         .listxattr      = btrfs_listxattr,
4969         .removexattr    = btrfs_removexattr,
4970         .permission     = btrfs_permission,
4971 };
4972 static struct inode_operations btrfs_dir_ro_inode_operations = {
4973         .lookup         = btrfs_lookup,
4974         .permission     = btrfs_permission,
4975 };
4976 static struct file_operations btrfs_dir_file_operations = {
4977         .llseek         = generic_file_llseek,
4978         .read           = generic_read_dir,
4979         .readdir        = btrfs_real_readdir,
4980         .unlocked_ioctl = btrfs_ioctl,
4981 #ifdef CONFIG_COMPAT
4982         .compat_ioctl   = btrfs_ioctl,
4983 #endif
4984         .release        = btrfs_release_file,
4985         .fsync          = btrfs_sync_file,
4986 };
4987
4988 static struct extent_io_ops btrfs_extent_io_ops = {
4989         .fill_delalloc = run_delalloc_range,
4990         .submit_bio_hook = btrfs_submit_bio_hook,
4991         .merge_bio_hook = btrfs_merge_bio_hook,
4992         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4993         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4994         .writepage_start_hook = btrfs_writepage_start_hook,
4995         .readpage_io_failed_hook = btrfs_io_failed_hook,
4996         .set_bit_hook = btrfs_set_bit_hook,
4997         .clear_bit_hook = btrfs_clear_bit_hook,
4998 };
4999
5000 /*
5001  * btrfs doesn't support the bmap operation because swapfiles
5002  * use bmap to make a mapping of extents in the file.  They assume
5003  * these extents won't change over the life of the file and they
5004  * use the bmap result to do IO directly to the drive.
5005  *
5006  * the btrfs bmap call would return logical addresses that aren't
5007  * suitable for IO and they also will change frequently as COW
5008  * operations happen.  So, swapfile + btrfs == corruption.
5009  *
5010  * For now we're avoiding this by dropping bmap.
5011  */
5012 static struct address_space_operations btrfs_aops = {
5013         .readpage       = btrfs_readpage,
5014         .writepage      = btrfs_writepage,
5015         .writepages     = btrfs_writepages,
5016         .readpages      = btrfs_readpages,
5017         .sync_page      = block_sync_page,
5018         .direct_IO      = btrfs_direct_IO,
5019         .invalidatepage = btrfs_invalidatepage,
5020         .releasepage    = btrfs_releasepage,
5021         .set_page_dirty = btrfs_set_page_dirty,
5022 };
5023
5024 static struct address_space_operations btrfs_symlink_aops = {
5025         .readpage       = btrfs_readpage,
5026         .writepage      = btrfs_writepage,
5027         .invalidatepage = btrfs_invalidatepage,
5028         .releasepage    = btrfs_releasepage,
5029 };
5030
5031 static struct inode_operations btrfs_file_inode_operations = {
5032         .truncate       = btrfs_truncate,
5033         .getattr        = btrfs_getattr,
5034         .setattr        = btrfs_setattr,
5035         .setxattr       = btrfs_setxattr,
5036         .getxattr       = btrfs_getxattr,
5037         .listxattr      = btrfs_listxattr,
5038         .removexattr    = btrfs_removexattr,
5039         .permission     = btrfs_permission,
5040         .fallocate      = btrfs_fallocate,
5041         .fiemap         = btrfs_fiemap,
5042 };
5043 static struct inode_operations btrfs_special_inode_operations = {
5044         .getattr        = btrfs_getattr,
5045         .setattr        = btrfs_setattr,
5046         .permission     = btrfs_permission,
5047         .setxattr       = btrfs_setxattr,
5048         .getxattr       = btrfs_getxattr,
5049         .listxattr      = btrfs_listxattr,
5050         .removexattr    = btrfs_removexattr,
5051 };
5052 static struct inode_operations btrfs_symlink_inode_operations = {
5053         .readlink       = generic_readlink,
5054         .follow_link    = page_follow_link_light,
5055         .put_link       = page_put_link,
5056         .permission     = btrfs_permission,
5057         .setxattr       = btrfs_setxattr,
5058         .getxattr       = btrfs_getxattr,
5059         .listxattr      = btrfs_listxattr,
5060         .removexattr    = btrfs_removexattr,
5061 };