Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
41 #include "compat.h"
42 #include "ctree.h"
43 #include "disk-io.h"
44 #include "transaction.h"
45 #include "btrfs_inode.h"
46 #include "ioctl.h"
47 #include "print-tree.h"
48 #include "volumes.h"
49 #include "ordered-data.h"
50 #include "xattr.h"
51 #include "tree-log.h"
52 #include "ref-cache.h"
53 #include "compression.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90                                    struct page *locked_page,
91                                    u64 start, u64 end, int *page_started,
92                                    unsigned long *nr_written, int unlock);
93
94 /*
95  * a very lame attempt at stopping writes when the FS is 85% full.  There
96  * are countless ways this is incorrect, but it is better than nothing.
97  */
98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
99                            int for_del)
100 {
101         u64 total;
102         u64 used;
103         u64 thresh;
104         int ret = 0;
105
106         spin_lock(&root->fs_info->delalloc_lock);
107         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
108         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
109         if (for_del)
110                 thresh = total * 90;
111         else
112                 thresh = total * 85;
113
114         do_div(thresh, 100);
115
116         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
117                 ret = -ENOSPC;
118         spin_unlock(&root->fs_info->delalloc_lock);
119         return ret;
120 }
121
122 /*
123  * this does all the hard work for inserting an inline extent into
124  * the btree.  The caller should have done a btrfs_drop_extents so that
125  * no overlapping inline items exist in the btree
126  */
127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
128                                 struct btrfs_root *root, struct inode *inode,
129                                 u64 start, size_t size, size_t compressed_size,
130                                 struct page **compressed_pages)
131 {
132         struct btrfs_key key;
133         struct btrfs_path *path;
134         struct extent_buffer *leaf;
135         struct page *page = NULL;
136         char *kaddr;
137         unsigned long ptr;
138         struct btrfs_file_extent_item *ei;
139         int err = 0;
140         int ret;
141         size_t cur_size = size;
142         size_t datasize;
143         unsigned long offset;
144         int use_compress = 0;
145
146         if (compressed_size && compressed_pages) {
147                 use_compress = 1;
148                 cur_size = compressed_size;
149         }
150
151         path = btrfs_alloc_path();
152         if (!path)
153                 return -ENOMEM;
154
155         btrfs_set_trans_block_group(trans, inode);
156
157         key.objectid = inode->i_ino;
158         key.offset = start;
159         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160         datasize = btrfs_file_extent_calc_inline_size(cur_size);
161
162         inode_add_bytes(inode, size);
163         ret = btrfs_insert_empty_item(trans, root, path, &key,
164                                       datasize);
165         BUG_ON(ret);
166         if (ret) {
167                 err = ret;
168                 goto fail;
169         }
170         leaf = path->nodes[0];
171         ei = btrfs_item_ptr(leaf, path->slots[0],
172                             struct btrfs_file_extent_item);
173         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
174         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
175         btrfs_set_file_extent_encryption(leaf, ei, 0);
176         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
177         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
178         ptr = btrfs_file_extent_inline_start(ei);
179
180         if (use_compress) {
181                 struct page *cpage;
182                 int i = 0;
183                 while (compressed_size > 0) {
184                         cpage = compressed_pages[i];
185                         cur_size = min_t(unsigned long, compressed_size,
186                                        PAGE_CACHE_SIZE);
187
188                         kaddr = kmap(cpage);
189                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
190                         kunmap(cpage);
191
192                         i++;
193                         ptr += cur_size;
194                         compressed_size -= cur_size;
195                 }
196                 btrfs_set_file_extent_compression(leaf, ei,
197                                                   BTRFS_COMPRESS_ZLIB);
198         } else {
199                 page = find_get_page(inode->i_mapping,
200                                      start >> PAGE_CACHE_SHIFT);
201                 btrfs_set_file_extent_compression(leaf, ei, 0);
202                 kaddr = kmap_atomic(page, KM_USER0);
203                 offset = start & (PAGE_CACHE_SIZE - 1);
204                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
205                 kunmap_atomic(kaddr, KM_USER0);
206                 page_cache_release(page);
207         }
208         btrfs_mark_buffer_dirty(leaf);
209         btrfs_free_path(path);
210
211         BTRFS_I(inode)->disk_i_size = inode->i_size;
212         btrfs_update_inode(trans, root, inode);
213         return 0;
214 fail:
215         btrfs_free_path(path);
216         return err;
217 }
218
219
220 /*
221  * conditionally insert an inline extent into the file.  This
222  * does the checks required to make sure the data is small enough
223  * to fit as an inline extent.
224  */
225 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
226                                  struct btrfs_root *root,
227                                  struct inode *inode, u64 start, u64 end,
228                                  size_t compressed_size,
229                                  struct page **compressed_pages)
230 {
231         u64 isize = i_size_read(inode);
232         u64 actual_end = min(end + 1, isize);
233         u64 inline_len = actual_end - start;
234         u64 aligned_end = (end + root->sectorsize - 1) &
235                         ~((u64)root->sectorsize - 1);
236         u64 hint_byte;
237         u64 data_len = inline_len;
238         int ret;
239
240         if (compressed_size)
241                 data_len = compressed_size;
242
243         if (start > 0 ||
244             actual_end >= PAGE_CACHE_SIZE ||
245             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
246             (!compressed_size &&
247             (actual_end & (root->sectorsize - 1)) == 0) ||
248             end + 1 < isize ||
249             data_len > root->fs_info->max_inline) {
250                 return 1;
251         }
252
253         ret = btrfs_drop_extents(trans, root, inode, start,
254                                  aligned_end, start, &hint_byte);
255         BUG_ON(ret);
256
257         if (isize > actual_end)
258                 inline_len = min_t(u64, isize, actual_end);
259         ret = insert_inline_extent(trans, root, inode, start,
260                                    inline_len, compressed_size,
261                                    compressed_pages);
262         BUG_ON(ret);
263         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
264         return 0;
265 }
266
267 struct async_extent {
268         u64 start;
269         u64 ram_size;
270         u64 compressed_size;
271         struct page **pages;
272         unsigned long nr_pages;
273         struct list_head list;
274 };
275
276 struct async_cow {
277         struct inode *inode;
278         struct btrfs_root *root;
279         struct page *locked_page;
280         u64 start;
281         u64 end;
282         struct list_head extents;
283         struct btrfs_work work;
284 };
285
286 static noinline int add_async_extent(struct async_cow *cow,
287                                      u64 start, u64 ram_size,
288                                      u64 compressed_size,
289                                      struct page **pages,
290                                      unsigned long nr_pages)
291 {
292         struct async_extent *async_extent;
293
294         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
295         async_extent->start = start;
296         async_extent->ram_size = ram_size;
297         async_extent->compressed_size = compressed_size;
298         async_extent->pages = pages;
299         async_extent->nr_pages = nr_pages;
300         list_add_tail(&async_extent->list, &cow->extents);
301         return 0;
302 }
303
304 /*
305  * we create compressed extents in two phases.  The first
306  * phase compresses a range of pages that have already been
307  * locked (both pages and state bits are locked).
308  *
309  * This is done inside an ordered work queue, and the compression
310  * is spread across many cpus.  The actual IO submission is step
311  * two, and the ordered work queue takes care of making sure that
312  * happens in the same order things were put onto the queue by
313  * writepages and friends.
314  *
315  * If this code finds it can't get good compression, it puts an
316  * entry onto the work queue to write the uncompressed bytes.  This
317  * makes sure that both compressed inodes and uncompressed inodes
318  * are written in the same order that pdflush sent them down.
319  */
320 static noinline int compress_file_range(struct inode *inode,
321                                         struct page *locked_page,
322                                         u64 start, u64 end,
323                                         struct async_cow *async_cow,
324                                         int *num_added)
325 {
326         struct btrfs_root *root = BTRFS_I(inode)->root;
327         struct btrfs_trans_handle *trans;
328         u64 num_bytes;
329         u64 orig_start;
330         u64 disk_num_bytes;
331         u64 blocksize = root->sectorsize;
332         u64 actual_end;
333         u64 isize = i_size_read(inode);
334         int ret = 0;
335         struct page **pages = NULL;
336         unsigned long nr_pages;
337         unsigned long nr_pages_ret = 0;
338         unsigned long total_compressed = 0;
339         unsigned long total_in = 0;
340         unsigned long max_compressed = 128 * 1024;
341         unsigned long max_uncompressed = 128 * 1024;
342         int i;
343         int will_compress;
344
345         orig_start = start;
346
347         actual_end = min_t(u64, isize, end + 1);
348 again:
349         will_compress = 0;
350         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
352
353         total_compressed = actual_end - start;
354
355         /* we want to make sure that amount of ram required to uncompress
356          * an extent is reasonable, so we limit the total size in ram
357          * of a compressed extent to 128k.  This is a crucial number
358          * because it also controls how easily we can spread reads across
359          * cpus for decompression.
360          *
361          * We also want to make sure the amount of IO required to do
362          * a random read is reasonably small, so we limit the size of
363          * a compressed extent to 128k.
364          */
365         total_compressed = min(total_compressed, max_uncompressed);
366         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
367         num_bytes = max(blocksize,  num_bytes);
368         disk_num_bytes = num_bytes;
369         total_in = 0;
370         ret = 0;
371
372         /*
373          * we do compression for mount -o compress and when the
374          * inode has not been flagged as nocompress.  This flag can
375          * change at any time if we discover bad compression ratios.
376          */
377         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
378             btrfs_test_opt(root, COMPRESS)) {
379                 WARN_ON(pages);
380                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
381
382                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
383                                                 total_compressed, pages,
384                                                 nr_pages, &nr_pages_ret,
385                                                 &total_in,
386                                                 &total_compressed,
387                                                 max_compressed);
388
389                 if (!ret) {
390                         unsigned long offset = total_compressed &
391                                 (PAGE_CACHE_SIZE - 1);
392                         struct page *page = pages[nr_pages_ret - 1];
393                         char *kaddr;
394
395                         /* zero the tail end of the last page, we might be
396                          * sending it down to disk
397                          */
398                         if (offset) {
399                                 kaddr = kmap_atomic(page, KM_USER0);
400                                 memset(kaddr + offset, 0,
401                                        PAGE_CACHE_SIZE - offset);
402                                 kunmap_atomic(kaddr, KM_USER0);
403                         }
404                         will_compress = 1;
405                 }
406         }
407         if (start == 0) {
408                 trans = btrfs_join_transaction(root, 1);
409                 BUG_ON(!trans);
410                 btrfs_set_trans_block_group(trans, inode);
411
412                 /* lets try to make an inline extent */
413                 if (ret || total_in < (actual_end - start)) {
414                         /* we didn't compress the entire range, try
415                          * to make an uncompressed inline extent.
416                          */
417                         ret = cow_file_range_inline(trans, root, inode,
418                                                     start, end, 0, NULL);
419                 } else {
420                         /* try making a compressed inline extent */
421                         ret = cow_file_range_inline(trans, root, inode,
422                                                     start, end,
423                                                     total_compressed, pages);
424                 }
425                 btrfs_end_transaction(trans, root);
426                 if (ret == 0) {
427                         /*
428                          * inline extent creation worked, we don't need
429                          * to create any more async work items.  Unlock
430                          * and free up our temp pages.
431                          */
432                         extent_clear_unlock_delalloc(inode,
433                                                      &BTRFS_I(inode)->io_tree,
434                                                      start, end, NULL, 1, 0,
435                                                      0, 1, 1, 1);
436                         ret = 0;
437                         goto free_pages_out;
438                 }
439         }
440
441         if (will_compress) {
442                 /*
443                  * we aren't doing an inline extent round the compressed size
444                  * up to a block size boundary so the allocator does sane
445                  * things
446                  */
447                 total_compressed = (total_compressed + blocksize - 1) &
448                         ~(blocksize - 1);
449
450                 /*
451                  * one last check to make sure the compression is really a
452                  * win, compare the page count read with the blocks on disk
453                  */
454                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
455                         ~(PAGE_CACHE_SIZE - 1);
456                 if (total_compressed >= total_in) {
457                         will_compress = 0;
458                 } else {
459                         disk_num_bytes = total_compressed;
460                         num_bytes = total_in;
461                 }
462         }
463         if (!will_compress && pages) {
464                 /*
465                  * the compression code ran but failed to make things smaller,
466                  * free any pages it allocated and our page pointer array
467                  */
468                 for (i = 0; i < nr_pages_ret; i++) {
469                         WARN_ON(pages[i]->mapping);
470                         page_cache_release(pages[i]);
471                 }
472                 kfree(pages);
473                 pages = NULL;
474                 total_compressed = 0;
475                 nr_pages_ret = 0;
476
477                 /* flag the file so we don't compress in the future */
478                 btrfs_set_flag(inode, NOCOMPRESS);
479         }
480         if (will_compress) {
481                 *num_added += 1;
482
483                 /* the async work queues will take care of doing actual
484                  * allocation on disk for these compressed pages,
485                  * and will submit them to the elevator.
486                  */
487                 add_async_extent(async_cow, start, num_bytes,
488                                  total_compressed, pages, nr_pages_ret);
489
490                 if (start + num_bytes < end && start + num_bytes < actual_end) {
491                         start += num_bytes;
492                         pages = NULL;
493                         cond_resched();
494                         goto again;
495                 }
496         } else {
497                 /*
498                  * No compression, but we still need to write the pages in
499                  * the file we've been given so far.  redirty the locked
500                  * page if it corresponds to our extent and set things up
501                  * for the async work queue to run cow_file_range to do
502                  * the normal delalloc dance
503                  */
504                 if (page_offset(locked_page) >= start &&
505                     page_offset(locked_page) <= end) {
506                         __set_page_dirty_nobuffers(locked_page);
507                         /* unlocked later on in the async handlers */
508                 }
509                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
510                 *num_added += 1;
511         }
512
513 out:
514         return 0;
515
516 free_pages_out:
517         for (i = 0; i < nr_pages_ret; i++) {
518                 WARN_ON(pages[i]->mapping);
519                 page_cache_release(pages[i]);
520         }
521         kfree(pages);
522
523         goto out;
524 }
525
526 /*
527  * phase two of compressed writeback.  This is the ordered portion
528  * of the code, which only gets called in the order the work was
529  * queued.  We walk all the async extents created by compress_file_range
530  * and send them down to the disk.
531  */
532 static noinline int submit_compressed_extents(struct inode *inode,
533                                               struct async_cow *async_cow)
534 {
535         struct async_extent *async_extent;
536         u64 alloc_hint = 0;
537         struct btrfs_trans_handle *trans;
538         struct btrfs_key ins;
539         struct extent_map *em;
540         struct btrfs_root *root = BTRFS_I(inode)->root;
541         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
542         struct extent_io_tree *io_tree;
543         int ret;
544
545         if (list_empty(&async_cow->extents))
546                 return 0;
547
548         trans = btrfs_join_transaction(root, 1);
549
550         while (!list_empty(&async_cow->extents)) {
551                 async_extent = list_entry(async_cow->extents.next,
552                                           struct async_extent, list);
553                 list_del(&async_extent->list);
554
555                 io_tree = &BTRFS_I(inode)->io_tree;
556
557                 /* did the compression code fall back to uncompressed IO? */
558                 if (!async_extent->pages) {
559                         int page_started = 0;
560                         unsigned long nr_written = 0;
561
562                         lock_extent(io_tree, async_extent->start,
563                                     async_extent->start +
564                                     async_extent->ram_size - 1, GFP_NOFS);
565
566                         /* allocate blocks */
567                         cow_file_range(inode, async_cow->locked_page,
568                                        async_extent->start,
569                                        async_extent->start +
570                                        async_extent->ram_size - 1,
571                                        &page_started, &nr_written, 0);
572
573                         /*
574                          * if page_started, cow_file_range inserted an
575                          * inline extent and took care of all the unlocking
576                          * and IO for us.  Otherwise, we need to submit
577                          * all those pages down to the drive.
578                          */
579                         if (!page_started)
580                                 extent_write_locked_range(io_tree,
581                                                   inode, async_extent->start,
582                                                   async_extent->start +
583                                                   async_extent->ram_size - 1,
584                                                   btrfs_get_extent,
585                                                   WB_SYNC_ALL);
586                         kfree(async_extent);
587                         cond_resched();
588                         continue;
589                 }
590
591                 lock_extent(io_tree, async_extent->start,
592                             async_extent->start + async_extent->ram_size - 1,
593                             GFP_NOFS);
594                 /*
595                  * here we're doing allocation and writeback of the
596                  * compressed pages
597                  */
598                 btrfs_drop_extent_cache(inode, async_extent->start,
599                                         async_extent->start +
600                                         async_extent->ram_size - 1, 0);
601
602                 ret = btrfs_reserve_extent(trans, root,
603                                            async_extent->compressed_size,
604                                            async_extent->compressed_size,
605                                            0, alloc_hint,
606                                            (u64)-1, &ins, 1);
607                 BUG_ON(ret);
608                 em = alloc_extent_map(GFP_NOFS);
609                 em->start = async_extent->start;
610                 em->len = async_extent->ram_size;
611                 em->orig_start = em->start;
612
613                 em->block_start = ins.objectid;
614                 em->block_len = ins.offset;
615                 em->bdev = root->fs_info->fs_devices->latest_bdev;
616                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
617                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
618
619                 while (1) {
620                         spin_lock(&em_tree->lock);
621                         ret = add_extent_mapping(em_tree, em);
622                         spin_unlock(&em_tree->lock);
623                         if (ret != -EEXIST) {
624                                 free_extent_map(em);
625                                 break;
626                         }
627                         btrfs_drop_extent_cache(inode, async_extent->start,
628                                                 async_extent->start +
629                                                 async_extent->ram_size - 1, 0);
630                 }
631
632                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
633                                                ins.objectid,
634                                                async_extent->ram_size,
635                                                ins.offset,
636                                                BTRFS_ORDERED_COMPRESSED);
637                 BUG_ON(ret);
638
639                 btrfs_end_transaction(trans, root);
640
641                 /*
642                  * clear dirty, set writeback and unlock the pages.
643                  */
644                 extent_clear_unlock_delalloc(inode,
645                                              &BTRFS_I(inode)->io_tree,
646                                              async_extent->start,
647                                              async_extent->start +
648                                              async_extent->ram_size - 1,
649                                              NULL, 1, 1, 0, 1, 1, 0);
650
651                 ret = btrfs_submit_compressed_write(inode,
652                                     async_extent->start,
653                                     async_extent->ram_size,
654                                     ins.objectid,
655                                     ins.offset, async_extent->pages,
656                                     async_extent->nr_pages);
657
658                 BUG_ON(ret);
659                 trans = btrfs_join_transaction(root, 1);
660                 alloc_hint = ins.objectid + ins.offset;
661                 kfree(async_extent);
662                 cond_resched();
663         }
664
665         btrfs_end_transaction(trans, root);
666         return 0;
667 }
668
669 /*
670  * when extent_io.c finds a delayed allocation range in the file,
671  * the call backs end up in this code.  The basic idea is to
672  * allocate extents on disk for the range, and create ordered data structs
673  * in ram to track those extents.
674  *
675  * locked_page is the page that writepage had locked already.  We use
676  * it to make sure we don't do extra locks or unlocks.
677  *
678  * *page_started is set to one if we unlock locked_page and do everything
679  * required to start IO on it.  It may be clean and already done with
680  * IO when we return.
681  */
682 static noinline int cow_file_range(struct inode *inode,
683                                    struct page *locked_page,
684                                    u64 start, u64 end, int *page_started,
685                                    unsigned long *nr_written,
686                                    int unlock)
687 {
688         struct btrfs_root *root = BTRFS_I(inode)->root;
689         struct btrfs_trans_handle *trans;
690         u64 alloc_hint = 0;
691         u64 num_bytes;
692         unsigned long ram_size;
693         u64 disk_num_bytes;
694         u64 cur_alloc_size;
695         u64 blocksize = root->sectorsize;
696         u64 actual_end;
697         u64 isize = i_size_read(inode);
698         struct btrfs_key ins;
699         struct extent_map *em;
700         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
701         int ret = 0;
702
703         trans = btrfs_join_transaction(root, 1);
704         BUG_ON(!trans);
705         btrfs_set_trans_block_group(trans, inode);
706
707         actual_end = min_t(u64, isize, end + 1);
708
709         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
710         num_bytes = max(blocksize,  num_bytes);
711         disk_num_bytes = num_bytes;
712         ret = 0;
713
714         if (start == 0) {
715                 /* lets try to make an inline extent */
716                 ret = cow_file_range_inline(trans, root, inode,
717                                             start, end, 0, NULL);
718                 if (ret == 0) {
719                         extent_clear_unlock_delalloc(inode,
720                                                      &BTRFS_I(inode)->io_tree,
721                                                      start, end, NULL, 1, 1,
722                                                      1, 1, 1, 1);
723                         *nr_written = *nr_written +
724                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
725                         *page_started = 1;
726                         ret = 0;
727                         goto out;
728                 }
729         }
730
731         BUG_ON(disk_num_bytes >
732                btrfs_super_total_bytes(&root->fs_info->super_copy));
733
734         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
735
736         while (disk_num_bytes > 0) {
737                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
738                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
739                                            root->sectorsize, 0, alloc_hint,
740                                            (u64)-1, &ins, 1);
741                 BUG_ON(ret);
742
743                 em = alloc_extent_map(GFP_NOFS);
744                 em->start = start;
745                 em->orig_start = em->start;
746
747                 ram_size = ins.offset;
748                 em->len = ins.offset;
749
750                 em->block_start = ins.objectid;
751                 em->block_len = ins.offset;
752                 em->bdev = root->fs_info->fs_devices->latest_bdev;
753                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
754
755                 while (1) {
756                         spin_lock(&em_tree->lock);
757                         ret = add_extent_mapping(em_tree, em);
758                         spin_unlock(&em_tree->lock);
759                         if (ret != -EEXIST) {
760                                 free_extent_map(em);
761                                 break;
762                         }
763                         btrfs_drop_extent_cache(inode, start,
764                                                 start + ram_size - 1, 0);
765                 }
766
767                 cur_alloc_size = ins.offset;
768                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
769                                                ram_size, cur_alloc_size, 0);
770                 BUG_ON(ret);
771
772                 if (root->root_key.objectid ==
773                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
774                         ret = btrfs_reloc_clone_csums(inode, start,
775                                                       cur_alloc_size);
776                         BUG_ON(ret);
777                 }
778
779                 if (disk_num_bytes < cur_alloc_size)
780                         break;
781
782                 /* we're not doing compressed IO, don't unlock the first
783                  * page (which the caller expects to stay locked), don't
784                  * clear any dirty bits and don't set any writeback bits
785                  */
786                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
787                                              start, start + ram_size - 1,
788                                              locked_page, unlock, 1,
789                                              1, 0, 0, 0);
790                 disk_num_bytes -= cur_alloc_size;
791                 num_bytes -= cur_alloc_size;
792                 alloc_hint = ins.objectid + ins.offset;
793                 start += cur_alloc_size;
794         }
795 out:
796         ret = 0;
797         btrfs_end_transaction(trans, root);
798
799         return ret;
800 }
801
802 /*
803  * work queue call back to started compression on a file and pages
804  */
805 static noinline void async_cow_start(struct btrfs_work *work)
806 {
807         struct async_cow *async_cow;
808         int num_added = 0;
809         async_cow = container_of(work, struct async_cow, work);
810
811         compress_file_range(async_cow->inode, async_cow->locked_page,
812                             async_cow->start, async_cow->end, async_cow,
813                             &num_added);
814         if (num_added == 0)
815                 async_cow->inode = NULL;
816 }
817
818 /*
819  * work queue call back to submit previously compressed pages
820  */
821 static noinline void async_cow_submit(struct btrfs_work *work)
822 {
823         struct async_cow *async_cow;
824         struct btrfs_root *root;
825         unsigned long nr_pages;
826
827         async_cow = container_of(work, struct async_cow, work);
828
829         root = async_cow->root;
830         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
831                 PAGE_CACHE_SHIFT;
832
833         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
834
835         if (atomic_read(&root->fs_info->async_delalloc_pages) <
836             5 * 1042 * 1024 &&
837             waitqueue_active(&root->fs_info->async_submit_wait))
838                 wake_up(&root->fs_info->async_submit_wait);
839
840         if (async_cow->inode)
841                 submit_compressed_extents(async_cow->inode, async_cow);
842 }
843
844 static noinline void async_cow_free(struct btrfs_work *work)
845 {
846         struct async_cow *async_cow;
847         async_cow = container_of(work, struct async_cow, work);
848         kfree(async_cow);
849 }
850
851 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
852                                 u64 start, u64 end, int *page_started,
853                                 unsigned long *nr_written)
854 {
855         struct async_cow *async_cow;
856         struct btrfs_root *root = BTRFS_I(inode)->root;
857         unsigned long nr_pages;
858         u64 cur_end;
859         int limit = 10 * 1024 * 1042;
860
861         if (!btrfs_test_opt(root, COMPRESS)) {
862                 return cow_file_range(inode, locked_page, start, end,
863                                       page_started, nr_written, 1);
864         }
865
866         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
867                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
868         while (start < end) {
869                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
870                 async_cow->inode = inode;
871                 async_cow->root = root;
872                 async_cow->locked_page = locked_page;
873                 async_cow->start = start;
874
875                 if (btrfs_test_flag(inode, NOCOMPRESS))
876                         cur_end = end;
877                 else
878                         cur_end = min(end, start + 512 * 1024 - 1);
879
880                 async_cow->end = cur_end;
881                 INIT_LIST_HEAD(&async_cow->extents);
882
883                 async_cow->work.func = async_cow_start;
884                 async_cow->work.ordered_func = async_cow_submit;
885                 async_cow->work.ordered_free = async_cow_free;
886                 async_cow->work.flags = 0;
887
888                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
889                         PAGE_CACHE_SHIFT;
890                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
891
892                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
893                                    &async_cow->work);
894
895                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
896                         wait_event(root->fs_info->async_submit_wait,
897                            (atomic_read(&root->fs_info->async_delalloc_pages) <
898                             limit));
899                 }
900
901                 while (atomic_read(&root->fs_info->async_submit_draining) &&
902                       atomic_read(&root->fs_info->async_delalloc_pages)) {
903                         wait_event(root->fs_info->async_submit_wait,
904                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
905                            0));
906                 }
907
908                 *nr_written += nr_pages;
909                 start = cur_end + 1;
910         }
911         *page_started = 1;
912         return 0;
913 }
914
915 static noinline int csum_exist_in_range(struct btrfs_root *root,
916                                         u64 bytenr, u64 num_bytes)
917 {
918         int ret;
919         struct btrfs_ordered_sum *sums;
920         LIST_HEAD(list);
921
922         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
923                                        bytenr + num_bytes - 1, &list);
924         if (ret == 0 && list_empty(&list))
925                 return 0;
926
927         while (!list_empty(&list)) {
928                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
929                 list_del(&sums->list);
930                 kfree(sums);
931         }
932         return 1;
933 }
934
935 /*
936  * when nowcow writeback call back.  This checks for snapshots or COW copies
937  * of the extents that exist in the file, and COWs the file as required.
938  *
939  * If no cow copies or snapshots exist, we write directly to the existing
940  * blocks on disk
941  */
942 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
943                               u64 start, u64 end, int *page_started, int force,
944                               unsigned long *nr_written)
945 {
946         struct btrfs_root *root = BTRFS_I(inode)->root;
947         struct btrfs_trans_handle *trans;
948         struct extent_buffer *leaf;
949         struct btrfs_path *path;
950         struct btrfs_file_extent_item *fi;
951         struct btrfs_key found_key;
952         u64 cow_start;
953         u64 cur_offset;
954         u64 extent_end;
955         u64 disk_bytenr;
956         u64 num_bytes;
957         int extent_type;
958         int ret;
959         int type;
960         int nocow;
961         int check_prev = 1;
962
963         path = btrfs_alloc_path();
964         BUG_ON(!path);
965         trans = btrfs_join_transaction(root, 1);
966         BUG_ON(!trans);
967
968         cow_start = (u64)-1;
969         cur_offset = start;
970         while (1) {
971                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
972                                                cur_offset, 0);
973                 BUG_ON(ret < 0);
974                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
975                         leaf = path->nodes[0];
976                         btrfs_item_key_to_cpu(leaf, &found_key,
977                                               path->slots[0] - 1);
978                         if (found_key.objectid == inode->i_ino &&
979                             found_key.type == BTRFS_EXTENT_DATA_KEY)
980                                 path->slots[0]--;
981                 }
982                 check_prev = 0;
983 next_slot:
984                 leaf = path->nodes[0];
985                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
986                         ret = btrfs_next_leaf(root, path);
987                         if (ret < 0)
988                                 BUG_ON(1);
989                         if (ret > 0)
990                                 break;
991                         leaf = path->nodes[0];
992                 }
993
994                 nocow = 0;
995                 disk_bytenr = 0;
996                 num_bytes = 0;
997                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
998
999                 if (found_key.objectid > inode->i_ino ||
1000                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1001                     found_key.offset > end)
1002                         break;
1003
1004                 if (found_key.offset > cur_offset) {
1005                         extent_end = found_key.offset;
1006                         goto out_check;
1007                 }
1008
1009                 fi = btrfs_item_ptr(leaf, path->slots[0],
1010                                     struct btrfs_file_extent_item);
1011                 extent_type = btrfs_file_extent_type(leaf, fi);
1012
1013                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1014                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1015                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1016                         extent_end = found_key.offset +
1017                                 btrfs_file_extent_num_bytes(leaf, fi);
1018                         if (extent_end <= start) {
1019                                 path->slots[0]++;
1020                                 goto next_slot;
1021                         }
1022                         if (disk_bytenr == 0)
1023                                 goto out_check;
1024                         if (btrfs_file_extent_compression(leaf, fi) ||
1025                             btrfs_file_extent_encryption(leaf, fi) ||
1026                             btrfs_file_extent_other_encoding(leaf, fi))
1027                                 goto out_check;
1028                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1029                                 goto out_check;
1030                         if (btrfs_extent_readonly(root, disk_bytenr))
1031                                 goto out_check;
1032                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1033                                                   disk_bytenr))
1034                                 goto out_check;
1035                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1036                         disk_bytenr += cur_offset - found_key.offset;
1037                         num_bytes = min(end + 1, extent_end) - cur_offset;
1038                         /*
1039                          * force cow if csum exists in the range.
1040                          * this ensure that csum for a given extent are
1041                          * either valid or do not exist.
1042                          */
1043                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1044                                 goto out_check;
1045                         nocow = 1;
1046                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1047                         extent_end = found_key.offset +
1048                                 btrfs_file_extent_inline_len(leaf, fi);
1049                         extent_end = ALIGN(extent_end, root->sectorsize);
1050                 } else {
1051                         BUG_ON(1);
1052                 }
1053 out_check:
1054                 if (extent_end <= start) {
1055                         path->slots[0]++;
1056                         goto next_slot;
1057                 }
1058                 if (!nocow) {
1059                         if (cow_start == (u64)-1)
1060                                 cow_start = cur_offset;
1061                         cur_offset = extent_end;
1062                         if (cur_offset > end)
1063                                 break;
1064                         path->slots[0]++;
1065                         goto next_slot;
1066                 }
1067
1068                 btrfs_release_path(root, path);
1069                 if (cow_start != (u64)-1) {
1070                         ret = cow_file_range(inode, locked_page, cow_start,
1071                                         found_key.offset - 1, page_started,
1072                                         nr_written, 1);
1073                         BUG_ON(ret);
1074                         cow_start = (u64)-1;
1075                 }
1076
1077                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1078                         struct extent_map *em;
1079                         struct extent_map_tree *em_tree;
1080                         em_tree = &BTRFS_I(inode)->extent_tree;
1081                         em = alloc_extent_map(GFP_NOFS);
1082                         em->start = cur_offset;
1083                         em->orig_start = em->start;
1084                         em->len = num_bytes;
1085                         em->block_len = num_bytes;
1086                         em->block_start = disk_bytenr;
1087                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1088                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1089                         while (1) {
1090                                 spin_lock(&em_tree->lock);
1091                                 ret = add_extent_mapping(em_tree, em);
1092                                 spin_unlock(&em_tree->lock);
1093                                 if (ret != -EEXIST) {
1094                                         free_extent_map(em);
1095                                         break;
1096                                 }
1097                                 btrfs_drop_extent_cache(inode, em->start,
1098                                                 em->start + em->len - 1, 0);
1099                         }
1100                         type = BTRFS_ORDERED_PREALLOC;
1101                 } else {
1102                         type = BTRFS_ORDERED_NOCOW;
1103                 }
1104
1105                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1106                                                num_bytes, num_bytes, type);
1107                 BUG_ON(ret);
1108
1109                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1110                                         cur_offset, cur_offset + num_bytes - 1,
1111                                         locked_page, 1, 1, 1, 0, 0, 0);
1112                 cur_offset = extent_end;
1113                 if (cur_offset > end)
1114                         break;
1115         }
1116         btrfs_release_path(root, path);
1117
1118         if (cur_offset <= end && cow_start == (u64)-1)
1119                 cow_start = cur_offset;
1120         if (cow_start != (u64)-1) {
1121                 ret = cow_file_range(inode, locked_page, cow_start, end,
1122                                      page_started, nr_written, 1);
1123                 BUG_ON(ret);
1124         }
1125
1126         ret = btrfs_end_transaction(trans, root);
1127         BUG_ON(ret);
1128         btrfs_free_path(path);
1129         return 0;
1130 }
1131
1132 /*
1133  * extent_io.c call back to do delayed allocation processing
1134  */
1135 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1136                               u64 start, u64 end, int *page_started,
1137                               unsigned long *nr_written)
1138 {
1139         int ret;
1140
1141         if (btrfs_test_flag(inode, NODATACOW))
1142                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1143                                          page_started, 1, nr_written);
1144         else if (btrfs_test_flag(inode, PREALLOC))
1145                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1146                                          page_started, 0, nr_written);
1147         else
1148                 ret = cow_file_range_async(inode, locked_page, start, end,
1149                                            page_started, nr_written);
1150
1151         return ret;
1152 }
1153
1154 /*
1155  * extent_io.c set_bit_hook, used to track delayed allocation
1156  * bytes in this file, and to maintain the list of inodes that
1157  * have pending delalloc work to be done.
1158  */
1159 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1160                        unsigned long old, unsigned long bits)
1161 {
1162         /*
1163          * set_bit and clear bit hooks normally require _irqsave/restore
1164          * but in this case, we are only testeing for the DELALLOC
1165          * bit, which is only set or cleared with irqs on
1166          */
1167         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1168                 struct btrfs_root *root = BTRFS_I(inode)->root;
1169                 spin_lock(&root->fs_info->delalloc_lock);
1170                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1171                 root->fs_info->delalloc_bytes += end - start + 1;
1172                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1173                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1174                                       &root->fs_info->delalloc_inodes);
1175                 }
1176                 spin_unlock(&root->fs_info->delalloc_lock);
1177         }
1178         return 0;
1179 }
1180
1181 /*
1182  * extent_io.c clear_bit_hook, see set_bit_hook for why
1183  */
1184 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1185                          unsigned long old, unsigned long bits)
1186 {
1187         /*
1188          * set_bit and clear bit hooks normally require _irqsave/restore
1189          * but in this case, we are only testeing for the DELALLOC
1190          * bit, which is only set or cleared with irqs on
1191          */
1192         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1193                 struct btrfs_root *root = BTRFS_I(inode)->root;
1194
1195                 spin_lock(&root->fs_info->delalloc_lock);
1196                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1197                         printk(KERN_INFO "btrfs warning: delalloc account "
1198                                "%llu %llu\n",
1199                                (unsigned long long)end - start + 1,
1200                                (unsigned long long)
1201                                root->fs_info->delalloc_bytes);
1202                         root->fs_info->delalloc_bytes = 0;
1203                         BTRFS_I(inode)->delalloc_bytes = 0;
1204                 } else {
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct list_head *cur;
1328         struct btrfs_ordered_sum *sum;
1329
1330         btrfs_set_trans_block_group(trans, inode);
1331         list_for_each(cur, list) {
1332                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1333                 btrfs_csum_file_blocks(trans,
1334                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1335         }
1336         return 0;
1337 }
1338
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340 {
1341         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1342                 WARN_ON(1);
1343         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344                                    GFP_NOFS);
1345 }
1346
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1349         struct page *page;
1350         struct btrfs_work work;
1351 };
1352
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1354 {
1355         struct btrfs_writepage_fixup *fixup;
1356         struct btrfs_ordered_extent *ordered;
1357         struct page *page;
1358         struct inode *inode;
1359         u64 page_start;
1360         u64 page_end;
1361
1362         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363         page = fixup->page;
1364 again:
1365         lock_page(page);
1366         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367                 ClearPageChecked(page);
1368                 goto out_page;
1369         }
1370
1371         inode = page->mapping->host;
1372         page_start = page_offset(page);
1373         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1376
1377         /* already ordered? We're done */
1378         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379                              EXTENT_ORDERED, 0)) {
1380                 goto out;
1381         }
1382
1383         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384         if (ordered) {
1385                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386                               page_end, GFP_NOFS);
1387                 unlock_page(page);
1388                 btrfs_start_ordered_extent(inode, ordered, 1);
1389                 goto again;
1390         }
1391
1392         btrfs_set_extent_delalloc(inode, page_start, page_end);
1393         ClearPageChecked(page);
1394 out:
1395         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396 out_page:
1397         unlock_page(page);
1398         page_cache_release(page);
1399 }
1400
1401 /*
1402  * There are a few paths in the higher layers of the kernel that directly
1403  * set the page dirty bit without asking the filesystem if it is a
1404  * good idea.  This causes problems because we want to make sure COW
1405  * properly happens and the data=ordered rules are followed.
1406  *
1407  * In our case any range that doesn't have the ORDERED bit set
1408  * hasn't been properly setup for IO.  We kick off an async process
1409  * to fix it up.  The async helper will wait for ordered extents, set
1410  * the delalloc bit and make it safe to write the page.
1411  */
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 {
1414         struct inode *inode = page->mapping->host;
1415         struct btrfs_writepage_fixup *fixup;
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret;
1418
1419         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420                              EXTENT_ORDERED, 0);
1421         if (ret)
1422                 return 0;
1423
1424         if (PageChecked(page))
1425                 return -EAGAIN;
1426
1427         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428         if (!fixup)
1429                 return -EAGAIN;
1430
1431         SetPageChecked(page);
1432         page_cache_get(page);
1433         fixup->work.func = btrfs_writepage_fixup_worker;
1434         fixup->page = page;
1435         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436         return -EAGAIN;
1437 }
1438
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440                                        struct inode *inode, u64 file_pos,
1441                                        u64 disk_bytenr, u64 disk_num_bytes,
1442                                        u64 num_bytes, u64 ram_bytes,
1443                                        u8 compression, u8 encryption,
1444                                        u16 other_encoding, int extent_type)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         struct btrfs_file_extent_item *fi;
1448         struct btrfs_path *path;
1449         struct extent_buffer *leaf;
1450         struct btrfs_key ins;
1451         u64 hint;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         BUG_ON(!path);
1456
1457         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458                                  file_pos + num_bytes, file_pos, &hint);
1459         BUG_ON(ret);
1460
1461         ins.objectid = inode->i_ino;
1462         ins.offset = file_pos;
1463         ins.type = BTRFS_EXTENT_DATA_KEY;
1464         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1465         BUG_ON(ret);
1466         leaf = path->nodes[0];
1467         fi = btrfs_item_ptr(leaf, path->slots[0],
1468                             struct btrfs_file_extent_item);
1469         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1470         btrfs_set_file_extent_type(leaf, fi, extent_type);
1471         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1472         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1473         btrfs_set_file_extent_offset(leaf, fi, 0);
1474         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1475         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1476         btrfs_set_file_extent_compression(leaf, fi, compression);
1477         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1478         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1479         btrfs_mark_buffer_dirty(leaf);
1480
1481         inode_add_bytes(inode, num_bytes);
1482         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1483
1484         ins.objectid = disk_bytenr;
1485         ins.offset = disk_num_bytes;
1486         ins.type = BTRFS_EXTENT_ITEM_KEY;
1487         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1488                                           root->root_key.objectid,
1489                                           trans->transid, inode->i_ino, &ins);
1490         BUG_ON(ret);
1491
1492         btrfs_free_path(path);
1493         return 0;
1494 }
1495
1496 /* as ordered data IO finishes, this gets called so we can finish
1497  * an ordered extent if the range of bytes in the file it covers are
1498  * fully written.
1499  */
1500 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1501 {
1502         struct btrfs_root *root = BTRFS_I(inode)->root;
1503         struct btrfs_trans_handle *trans;
1504         struct btrfs_ordered_extent *ordered_extent;
1505         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1506         int compressed = 0;
1507         int ret;
1508
1509         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1510         if (!ret)
1511                 return 0;
1512
1513         trans = btrfs_join_transaction(root, 1);
1514
1515         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1516         BUG_ON(!ordered_extent);
1517         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1518                 goto nocow;
1519
1520         lock_extent(io_tree, ordered_extent->file_offset,
1521                     ordered_extent->file_offset + ordered_extent->len - 1,
1522                     GFP_NOFS);
1523
1524         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1525                 compressed = 1;
1526         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1527                 BUG_ON(compressed);
1528                 ret = btrfs_mark_extent_written(trans, root, inode,
1529                                                 ordered_extent->file_offset,
1530                                                 ordered_extent->file_offset +
1531                                                 ordered_extent->len);
1532                 BUG_ON(ret);
1533         } else {
1534                 ret = insert_reserved_file_extent(trans, inode,
1535                                                 ordered_extent->file_offset,
1536                                                 ordered_extent->start,
1537                                                 ordered_extent->disk_len,
1538                                                 ordered_extent->len,
1539                                                 ordered_extent->len,
1540                                                 compressed, 0, 0,
1541                                                 BTRFS_FILE_EXTENT_REG);
1542                 BUG_ON(ret);
1543         }
1544         unlock_extent(io_tree, ordered_extent->file_offset,
1545                     ordered_extent->file_offset + ordered_extent->len - 1,
1546                     GFP_NOFS);
1547 nocow:
1548         add_pending_csums(trans, inode, ordered_extent->file_offset,
1549                           &ordered_extent->list);
1550
1551         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1552         btrfs_ordered_update_i_size(inode, ordered_extent);
1553         btrfs_update_inode(trans, root, inode);
1554         btrfs_remove_ordered_extent(inode, ordered_extent);
1555         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1556
1557         /* once for us */
1558         btrfs_put_ordered_extent(ordered_extent);
1559         /* once for the tree */
1560         btrfs_put_ordered_extent(ordered_extent);
1561
1562         btrfs_end_transaction(trans, root);
1563         return 0;
1564 }
1565
1566 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1567                                 struct extent_state *state, int uptodate)
1568 {
1569         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1570 }
1571
1572 /*
1573  * When IO fails, either with EIO or csum verification fails, we
1574  * try other mirrors that might have a good copy of the data.  This
1575  * io_failure_record is used to record state as we go through all the
1576  * mirrors.  If another mirror has good data, the page is set up to date
1577  * and things continue.  If a good mirror can't be found, the original
1578  * bio end_io callback is called to indicate things have failed.
1579  */
1580 struct io_failure_record {
1581         struct page *page;
1582         u64 start;
1583         u64 len;
1584         u64 logical;
1585         unsigned long bio_flags;
1586         int last_mirror;
1587 };
1588
1589 static int btrfs_io_failed_hook(struct bio *failed_bio,
1590                          struct page *page, u64 start, u64 end,
1591                          struct extent_state *state)
1592 {
1593         struct io_failure_record *failrec = NULL;
1594         u64 private;
1595         struct extent_map *em;
1596         struct inode *inode = page->mapping->host;
1597         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1598         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1599         struct bio *bio;
1600         int num_copies;
1601         int ret;
1602         int rw;
1603         u64 logical;
1604
1605         ret = get_state_private(failure_tree, start, &private);
1606         if (ret) {
1607                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1608                 if (!failrec)
1609                         return -ENOMEM;
1610                 failrec->start = start;
1611                 failrec->len = end - start + 1;
1612                 failrec->last_mirror = 0;
1613                 failrec->bio_flags = 0;
1614
1615                 spin_lock(&em_tree->lock);
1616                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1617                 if (em->start > start || em->start + em->len < start) {
1618                         free_extent_map(em);
1619                         em = NULL;
1620                 }
1621                 spin_unlock(&em_tree->lock);
1622
1623                 if (!em || IS_ERR(em)) {
1624                         kfree(failrec);
1625                         return -EIO;
1626                 }
1627                 logical = start - em->start;
1628                 logical = em->block_start + logical;
1629                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1630                         logical = em->block_start;
1631                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1632                 }
1633                 failrec->logical = logical;
1634                 free_extent_map(em);
1635                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1636                                 EXTENT_DIRTY, GFP_NOFS);
1637                 set_state_private(failure_tree, start,
1638                                  (u64)(unsigned long)failrec);
1639         } else {
1640                 failrec = (struct io_failure_record *)(unsigned long)private;
1641         }
1642         num_copies = btrfs_num_copies(
1643                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1644                               failrec->logical, failrec->len);
1645         failrec->last_mirror++;
1646         if (!state) {
1647                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1648                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1649                                                     failrec->start,
1650                                                     EXTENT_LOCKED);
1651                 if (state && state->start != failrec->start)
1652                         state = NULL;
1653                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1654         }
1655         if (!state || failrec->last_mirror > num_copies) {
1656                 set_state_private(failure_tree, failrec->start, 0);
1657                 clear_extent_bits(failure_tree, failrec->start,
1658                                   failrec->start + failrec->len - 1,
1659                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1660                 kfree(failrec);
1661                 return -EIO;
1662         }
1663         bio = bio_alloc(GFP_NOFS, 1);
1664         bio->bi_private = state;
1665         bio->bi_end_io = failed_bio->bi_end_io;
1666         bio->bi_sector = failrec->logical >> 9;
1667         bio->bi_bdev = failed_bio->bi_bdev;
1668         bio->bi_size = 0;
1669
1670         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1671         if (failed_bio->bi_rw & (1 << BIO_RW))
1672                 rw = WRITE;
1673         else
1674                 rw = READ;
1675
1676         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1677                                                       failrec->last_mirror,
1678                                                       failrec->bio_flags);
1679         return 0;
1680 }
1681
1682 /*
1683  * each time an IO finishes, we do a fast check in the IO failure tree
1684  * to see if we need to process or clean up an io_failure_record
1685  */
1686 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1687 {
1688         u64 private;
1689         u64 private_failure;
1690         struct io_failure_record *failure;
1691         int ret;
1692
1693         private = 0;
1694         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1695                              (u64)-1, 1, EXTENT_DIRTY)) {
1696                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1697                                         start, &private_failure);
1698                 if (ret == 0) {
1699                         failure = (struct io_failure_record *)(unsigned long)
1700                                    private_failure;
1701                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1702                                           failure->start, 0);
1703                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1704                                           failure->start,
1705                                           failure->start + failure->len - 1,
1706                                           EXTENT_DIRTY | EXTENT_LOCKED,
1707                                           GFP_NOFS);
1708                         kfree(failure);
1709                 }
1710         }
1711         return 0;
1712 }
1713
1714 /*
1715  * when reads are done, we need to check csums to verify the data is correct
1716  * if there's a match, we allow the bio to finish.  If not, we go through
1717  * the io_failure_record routines to find good copies
1718  */
1719 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1720                                struct extent_state *state)
1721 {
1722         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1723         struct inode *inode = page->mapping->host;
1724         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1725         char *kaddr;
1726         u64 private = ~(u32)0;
1727         int ret;
1728         struct btrfs_root *root = BTRFS_I(inode)->root;
1729         u32 csum = ~(u32)0;
1730
1731         if (PageChecked(page)) {
1732                 ClearPageChecked(page);
1733                 goto good;
1734         }
1735         if (btrfs_test_flag(inode, NODATASUM))
1736                 return 0;
1737
1738         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1739             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1740                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1741                                   GFP_NOFS);
1742                 return 0;
1743         }
1744
1745         if (state && state->start == start) {
1746                 private = state->private;
1747                 ret = 0;
1748         } else {
1749                 ret = get_state_private(io_tree, start, &private);
1750         }
1751         kaddr = kmap_atomic(page, KM_USER0);
1752         if (ret)
1753                 goto zeroit;
1754
1755         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1756         btrfs_csum_final(csum, (char *)&csum);
1757         if (csum != private)
1758                 goto zeroit;
1759
1760         kunmap_atomic(kaddr, KM_USER0);
1761 good:
1762         /* if the io failure tree for this inode is non-empty,
1763          * check to see if we've recovered from a failed IO
1764          */
1765         btrfs_clean_io_failures(inode, start);
1766         return 0;
1767
1768 zeroit:
1769         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1770                "private %llu\n", page->mapping->host->i_ino,
1771                (unsigned long long)start, csum,
1772                (unsigned long long)private);
1773         memset(kaddr + offset, 1, end - start + 1);
1774         flush_dcache_page(page);
1775         kunmap_atomic(kaddr, KM_USER0);
1776         if (private == 0)
1777                 return 0;
1778         return -EIO;
1779 }
1780
1781 /*
1782  * This creates an orphan entry for the given inode in case something goes
1783  * wrong in the middle of an unlink/truncate.
1784  */
1785 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1786 {
1787         struct btrfs_root *root = BTRFS_I(inode)->root;
1788         int ret = 0;
1789
1790         spin_lock(&root->list_lock);
1791
1792         /* already on the orphan list, we're good */
1793         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1794                 spin_unlock(&root->list_lock);
1795                 return 0;
1796         }
1797
1798         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1799
1800         spin_unlock(&root->list_lock);
1801
1802         /*
1803          * insert an orphan item to track this unlinked/truncated file
1804          */
1805         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1806
1807         return ret;
1808 }
1809
1810 /*
1811  * We have done the truncate/delete so we can go ahead and remove the orphan
1812  * item for this particular inode.
1813  */
1814 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1815 {
1816         struct btrfs_root *root = BTRFS_I(inode)->root;
1817         int ret = 0;
1818
1819         spin_lock(&root->list_lock);
1820
1821         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1822                 spin_unlock(&root->list_lock);
1823                 return 0;
1824         }
1825
1826         list_del_init(&BTRFS_I(inode)->i_orphan);
1827         if (!trans) {
1828                 spin_unlock(&root->list_lock);
1829                 return 0;
1830         }
1831
1832         spin_unlock(&root->list_lock);
1833
1834         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1835
1836         return ret;
1837 }
1838
1839 /*
1840  * this cleans up any orphans that may be left on the list from the last use
1841  * of this root.
1842  */
1843 void btrfs_orphan_cleanup(struct btrfs_root *root)
1844 {
1845         struct btrfs_path *path;
1846         struct extent_buffer *leaf;
1847         struct btrfs_item *item;
1848         struct btrfs_key key, found_key;
1849         struct btrfs_trans_handle *trans;
1850         struct inode *inode;
1851         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1852
1853         path = btrfs_alloc_path();
1854         if (!path)
1855                 return;
1856         path->reada = -1;
1857
1858         key.objectid = BTRFS_ORPHAN_OBJECTID;
1859         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1860         key.offset = (u64)-1;
1861
1862
1863         while (1) {
1864                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1865                 if (ret < 0) {
1866                         printk(KERN_ERR "Error searching slot for orphan: %d"
1867                                "\n", ret);
1868                         break;
1869                 }
1870
1871                 /*
1872                  * if ret == 0 means we found what we were searching for, which
1873                  * is weird, but possible, so only screw with path if we didnt
1874                  * find the key and see if we have stuff that matches
1875                  */
1876                 if (ret > 0) {
1877                         if (path->slots[0] == 0)
1878                                 break;
1879                         path->slots[0]--;
1880                 }
1881
1882                 /* pull out the item */
1883                 leaf = path->nodes[0];
1884                 item = btrfs_item_nr(leaf, path->slots[0]);
1885                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1886
1887                 /* make sure the item matches what we want */
1888                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1889                         break;
1890                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1891                         break;
1892
1893                 /* release the path since we're done with it */
1894                 btrfs_release_path(root, path);
1895
1896                 /*
1897                  * this is where we are basically btrfs_lookup, without the
1898                  * crossing root thing.  we store the inode number in the
1899                  * offset of the orphan item.
1900                  */
1901                 inode = btrfs_iget_locked(root->fs_info->sb,
1902                                           found_key.offset, root);
1903                 if (!inode)
1904                         break;
1905
1906                 if (inode->i_state & I_NEW) {
1907                         BTRFS_I(inode)->root = root;
1908
1909                         /* have to set the location manually */
1910                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1911                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1912                         BTRFS_I(inode)->location.offset = 0;
1913
1914                         btrfs_read_locked_inode(inode);
1915                         unlock_new_inode(inode);
1916                 }
1917
1918                 /*
1919                  * add this inode to the orphan list so btrfs_orphan_del does
1920                  * the proper thing when we hit it
1921                  */
1922                 spin_lock(&root->list_lock);
1923                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1924                 spin_unlock(&root->list_lock);
1925
1926                 /*
1927                  * if this is a bad inode, means we actually succeeded in
1928                  * removing the inode, but not the orphan record, which means
1929                  * we need to manually delete the orphan since iput will just
1930                  * do a destroy_inode
1931                  */
1932                 if (is_bad_inode(inode)) {
1933                         trans = btrfs_start_transaction(root, 1);
1934                         btrfs_orphan_del(trans, inode);
1935                         btrfs_end_transaction(trans, root);
1936                         iput(inode);
1937                         continue;
1938                 }
1939
1940                 /* if we have links, this was a truncate, lets do that */
1941                 if (inode->i_nlink) {
1942                         nr_truncate++;
1943                         btrfs_truncate(inode);
1944                 } else {
1945                         nr_unlink++;
1946                 }
1947
1948                 /* this will do delete_inode and everything for us */
1949                 iput(inode);
1950         }
1951
1952         if (nr_unlink)
1953                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1954         if (nr_truncate)
1955                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1956
1957         btrfs_free_path(path);
1958 }
1959
1960 /*
1961  * read an inode from the btree into the in-memory inode
1962  */
1963 void btrfs_read_locked_inode(struct inode *inode)
1964 {
1965         struct btrfs_path *path;
1966         struct extent_buffer *leaf;
1967         struct btrfs_inode_item *inode_item;
1968         struct btrfs_timespec *tspec;
1969         struct btrfs_root *root = BTRFS_I(inode)->root;
1970         struct btrfs_key location;
1971         u64 alloc_group_block;
1972         u32 rdev;
1973         int ret;
1974
1975         path = btrfs_alloc_path();
1976         BUG_ON(!path);
1977         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1978
1979         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1980         if (ret)
1981                 goto make_bad;
1982
1983         leaf = path->nodes[0];
1984         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1985                                     struct btrfs_inode_item);
1986
1987         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1988         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1989         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1990         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1991         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1992
1993         tspec = btrfs_inode_atime(inode_item);
1994         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1995         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1996
1997         tspec = btrfs_inode_mtime(inode_item);
1998         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1999         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2000
2001         tspec = btrfs_inode_ctime(inode_item);
2002         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2003         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2004
2005         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2006         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2007         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2008         inode->i_generation = BTRFS_I(inode)->generation;
2009         inode->i_rdev = 0;
2010         rdev = btrfs_inode_rdev(leaf, inode_item);
2011
2012         BTRFS_I(inode)->index_cnt = (u64)-1;
2013         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2014
2015         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2016         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2017                                                 alloc_group_block, 0);
2018         btrfs_free_path(path);
2019         inode_item = NULL;
2020
2021         switch (inode->i_mode & S_IFMT) {
2022         case S_IFREG:
2023                 inode->i_mapping->a_ops = &btrfs_aops;
2024                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2025                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2026                 inode->i_fop = &btrfs_file_operations;
2027                 inode->i_op = &btrfs_file_inode_operations;
2028                 break;
2029         case S_IFDIR:
2030                 inode->i_fop = &btrfs_dir_file_operations;
2031                 if (root == root->fs_info->tree_root)
2032                         inode->i_op = &btrfs_dir_ro_inode_operations;
2033                 else
2034                         inode->i_op = &btrfs_dir_inode_operations;
2035                 break;
2036         case S_IFLNK:
2037                 inode->i_op = &btrfs_symlink_inode_operations;
2038                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2039                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2040                 break;
2041         default:
2042                 init_special_inode(inode, inode->i_mode, rdev);
2043                 break;
2044         }
2045         return;
2046
2047 make_bad:
2048         btrfs_free_path(path);
2049         make_bad_inode(inode);
2050 }
2051
2052 /*
2053  * given a leaf and an inode, copy the inode fields into the leaf
2054  */
2055 static void fill_inode_item(struct btrfs_trans_handle *trans,
2056                             struct extent_buffer *leaf,
2057                             struct btrfs_inode_item *item,
2058                             struct inode *inode)
2059 {
2060         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2061         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2062         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2063         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2064         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2065
2066         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2067                                inode->i_atime.tv_sec);
2068         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2069                                 inode->i_atime.tv_nsec);
2070
2071         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2072                                inode->i_mtime.tv_sec);
2073         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2074                                 inode->i_mtime.tv_nsec);
2075
2076         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2077                                inode->i_ctime.tv_sec);
2078         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2079                                 inode->i_ctime.tv_nsec);
2080
2081         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2082         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2083         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2084         btrfs_set_inode_transid(leaf, item, trans->transid);
2085         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2086         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2087         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2088 }
2089
2090 /*
2091  * copy everything in the in-memory inode into the btree.
2092  */
2093 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2094                                 struct btrfs_root *root, struct inode *inode)
2095 {
2096         struct btrfs_inode_item *inode_item;
2097         struct btrfs_path *path;
2098         struct extent_buffer *leaf;
2099         int ret;
2100
2101         path = btrfs_alloc_path();
2102         BUG_ON(!path);
2103         ret = btrfs_lookup_inode(trans, root, path,
2104                                  &BTRFS_I(inode)->location, 1);
2105         if (ret) {
2106                 if (ret > 0)
2107                         ret = -ENOENT;
2108                 goto failed;
2109         }
2110
2111         leaf = path->nodes[0];
2112         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2113                                   struct btrfs_inode_item);
2114
2115         fill_inode_item(trans, leaf, inode_item, inode);
2116         btrfs_mark_buffer_dirty(leaf);
2117         btrfs_set_inode_last_trans(trans, inode);
2118         ret = 0;
2119 failed:
2120         btrfs_free_path(path);
2121         return ret;
2122 }
2123
2124
2125 /*
2126  * unlink helper that gets used here in inode.c and in the tree logging
2127  * recovery code.  It remove a link in a directory with a given name, and
2128  * also drops the back refs in the inode to the directory
2129  */
2130 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2131                        struct btrfs_root *root,
2132                        struct inode *dir, struct inode *inode,
2133                        const char *name, int name_len)
2134 {
2135         struct btrfs_path *path;
2136         int ret = 0;
2137         struct extent_buffer *leaf;
2138         struct btrfs_dir_item *di;
2139         struct btrfs_key key;
2140         u64 index;
2141
2142         path = btrfs_alloc_path();
2143         if (!path) {
2144                 ret = -ENOMEM;
2145                 goto err;
2146         }
2147
2148         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2149                                     name, name_len, -1);
2150         if (IS_ERR(di)) {
2151                 ret = PTR_ERR(di);
2152                 goto err;
2153         }
2154         if (!di) {
2155                 ret = -ENOENT;
2156                 goto err;
2157         }
2158         leaf = path->nodes[0];
2159         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2160         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2161         if (ret)
2162                 goto err;
2163         btrfs_release_path(root, path);
2164
2165         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2166                                   inode->i_ino,
2167                                   dir->i_ino, &index);
2168         if (ret) {
2169                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2170                        "inode %lu parent %lu\n", name_len, name,
2171                        inode->i_ino, dir->i_ino);
2172                 goto err;
2173         }
2174
2175         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2176                                          index, name, name_len, -1);
2177         if (IS_ERR(di)) {
2178                 ret = PTR_ERR(di);
2179                 goto err;
2180         }
2181         if (!di) {
2182                 ret = -ENOENT;
2183                 goto err;
2184         }
2185         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2186         btrfs_release_path(root, path);
2187
2188         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2189                                          inode, dir->i_ino);
2190         BUG_ON(ret != 0 && ret != -ENOENT);
2191         if (ret != -ENOENT)
2192                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2193
2194         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2195                                            dir, index);
2196         BUG_ON(ret);
2197 err:
2198         btrfs_free_path(path);
2199         if (ret)
2200                 goto out;
2201
2202         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2203         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2204         btrfs_update_inode(trans, root, dir);
2205         btrfs_drop_nlink(inode);
2206         ret = btrfs_update_inode(trans, root, inode);
2207         dir->i_sb->s_dirt = 1;
2208 out:
2209         return ret;
2210 }
2211
2212 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2213 {
2214         struct btrfs_root *root;
2215         struct btrfs_trans_handle *trans;
2216         struct inode *inode = dentry->d_inode;
2217         int ret;
2218         unsigned long nr = 0;
2219
2220         root = BTRFS_I(dir)->root;
2221
2222         ret = btrfs_check_free_space(root, 1, 1);
2223         if (ret)
2224                 goto fail;
2225
2226         trans = btrfs_start_transaction(root, 1);
2227
2228         btrfs_set_trans_block_group(trans, dir);
2229         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2230                                  dentry->d_name.name, dentry->d_name.len);
2231
2232         if (inode->i_nlink == 0)
2233                 ret = btrfs_orphan_add(trans, inode);
2234
2235         nr = trans->blocks_used;
2236
2237         btrfs_end_transaction_throttle(trans, root);
2238 fail:
2239         btrfs_btree_balance_dirty(root, nr);
2240         return ret;
2241 }
2242
2243 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2244 {
2245         struct inode *inode = dentry->d_inode;
2246         int err = 0;
2247         int ret;
2248         struct btrfs_root *root = BTRFS_I(dir)->root;
2249         struct btrfs_trans_handle *trans;
2250         unsigned long nr = 0;
2251
2252         /*
2253          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2254          * the root of a subvolume or snapshot
2255          */
2256         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2257             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2258                 return -ENOTEMPTY;
2259         }
2260
2261         ret = btrfs_check_free_space(root, 1, 1);
2262         if (ret)
2263                 goto fail;
2264
2265         trans = btrfs_start_transaction(root, 1);
2266         btrfs_set_trans_block_group(trans, dir);
2267
2268         err = btrfs_orphan_add(trans, inode);
2269         if (err)
2270                 goto fail_trans;
2271
2272         /* now the directory is empty */
2273         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2274                                  dentry->d_name.name, dentry->d_name.len);
2275         if (!err)
2276                 btrfs_i_size_write(inode, 0);
2277
2278 fail_trans:
2279         nr = trans->blocks_used;
2280         ret = btrfs_end_transaction_throttle(trans, root);
2281 fail:
2282         btrfs_btree_balance_dirty(root, nr);
2283
2284         if (ret && !err)
2285                 err = ret;
2286         return err;
2287 }
2288
2289 #if 0
2290 /*
2291  * when truncating bytes in a file, it is possible to avoid reading
2292  * the leaves that contain only checksum items.  This can be the
2293  * majority of the IO required to delete a large file, but it must
2294  * be done carefully.
2295  *
2296  * The keys in the level just above the leaves are checked to make sure
2297  * the lowest key in a given leaf is a csum key, and starts at an offset
2298  * after the new  size.
2299  *
2300  * Then the key for the next leaf is checked to make sure it also has
2301  * a checksum item for the same file.  If it does, we know our target leaf
2302  * contains only checksum items, and it can be safely freed without reading
2303  * it.
2304  *
2305  * This is just an optimization targeted at large files.  It may do
2306  * nothing.  It will return 0 unless things went badly.
2307  */
2308 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2309                                      struct btrfs_root *root,
2310                                      struct btrfs_path *path,
2311                                      struct inode *inode, u64 new_size)
2312 {
2313         struct btrfs_key key;
2314         int ret;
2315         int nritems;
2316         struct btrfs_key found_key;
2317         struct btrfs_key other_key;
2318         struct btrfs_leaf_ref *ref;
2319         u64 leaf_gen;
2320         u64 leaf_start;
2321
2322         path->lowest_level = 1;
2323         key.objectid = inode->i_ino;
2324         key.type = BTRFS_CSUM_ITEM_KEY;
2325         key.offset = new_size;
2326 again:
2327         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2328         if (ret < 0)
2329                 goto out;
2330
2331         if (path->nodes[1] == NULL) {
2332                 ret = 0;
2333                 goto out;
2334         }
2335         ret = 0;
2336         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2337         nritems = btrfs_header_nritems(path->nodes[1]);
2338
2339         if (!nritems)
2340                 goto out;
2341
2342         if (path->slots[1] >= nritems)
2343                 goto next_node;
2344
2345         /* did we find a key greater than anything we want to delete? */
2346         if (found_key.objectid > inode->i_ino ||
2347            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2348                 goto out;
2349
2350         /* we check the next key in the node to make sure the leave contains
2351          * only checksum items.  This comparison doesn't work if our
2352          * leaf is the last one in the node
2353          */
2354         if (path->slots[1] + 1 >= nritems) {
2355 next_node:
2356                 /* search forward from the last key in the node, this
2357                  * will bring us into the next node in the tree
2358                  */
2359                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2360
2361                 /* unlikely, but we inc below, so check to be safe */
2362                 if (found_key.offset == (u64)-1)
2363                         goto out;
2364
2365                 /* search_forward needs a path with locks held, do the
2366                  * search again for the original key.  It is possible
2367                  * this will race with a balance and return a path that
2368                  * we could modify, but this drop is just an optimization
2369                  * and is allowed to miss some leaves.
2370                  */
2371                 btrfs_release_path(root, path);
2372                 found_key.offset++;
2373
2374                 /* setup a max key for search_forward */
2375                 other_key.offset = (u64)-1;
2376                 other_key.type = key.type;
2377                 other_key.objectid = key.objectid;
2378
2379                 path->keep_locks = 1;
2380                 ret = btrfs_search_forward(root, &found_key, &other_key,
2381                                            path, 0, 0);
2382                 path->keep_locks = 0;
2383                 if (ret || found_key.objectid != key.objectid ||
2384                     found_key.type != key.type) {
2385                         ret = 0;
2386                         goto out;
2387                 }
2388
2389                 key.offset = found_key.offset;
2390                 btrfs_release_path(root, path);
2391                 cond_resched();
2392                 goto again;
2393         }
2394
2395         /* we know there's one more slot after us in the tree,
2396          * read that key so we can verify it is also a checksum item
2397          */
2398         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2399
2400         if (found_key.objectid < inode->i_ino)
2401                 goto next_key;
2402
2403         if (found_key.type != key.type || found_key.offset < new_size)
2404                 goto next_key;
2405
2406         /*
2407          * if the key for the next leaf isn't a csum key from this objectid,
2408          * we can't be sure there aren't good items inside this leaf.
2409          * Bail out
2410          */
2411         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2412                 goto out;
2413
2414         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2415         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2416         /*
2417          * it is safe to delete this leaf, it contains only
2418          * csum items from this inode at an offset >= new_size
2419          */
2420         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2421         BUG_ON(ret);
2422
2423         if (root->ref_cows && leaf_gen < trans->transid) {
2424                 ref = btrfs_alloc_leaf_ref(root, 0);
2425                 if (ref) {
2426                         ref->root_gen = root->root_key.offset;
2427                         ref->bytenr = leaf_start;
2428                         ref->owner = 0;
2429                         ref->generation = leaf_gen;
2430                         ref->nritems = 0;
2431
2432                         ret = btrfs_add_leaf_ref(root, ref, 0);
2433                         WARN_ON(ret);
2434                         btrfs_free_leaf_ref(root, ref);
2435                 } else {
2436                         WARN_ON(1);
2437                 }
2438         }
2439 next_key:
2440         btrfs_release_path(root, path);
2441
2442         if (other_key.objectid == inode->i_ino &&
2443             other_key.type == key.type && other_key.offset > key.offset) {
2444                 key.offset = other_key.offset;
2445                 cond_resched();
2446                 goto again;
2447         }
2448         ret = 0;
2449 out:
2450         /* fixup any changes we've made to the path */
2451         path->lowest_level = 0;
2452         path->keep_locks = 0;
2453         btrfs_release_path(root, path);
2454         return ret;
2455 }
2456
2457 #endif
2458
2459 /*
2460  * this can truncate away extent items, csum items and directory items.
2461  * It starts at a high offset and removes keys until it can't find
2462  * any higher than new_size
2463  *
2464  * csum items that cross the new i_size are truncated to the new size
2465  * as well.
2466  *
2467  * min_type is the minimum key type to truncate down to.  If set to 0, this
2468  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2469  */
2470 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2471                                         struct btrfs_root *root,
2472                                         struct inode *inode,
2473                                         u64 new_size, u32 min_type)
2474 {
2475         int ret;
2476         struct btrfs_path *path;
2477         struct btrfs_key key;
2478         struct btrfs_key found_key;
2479         u32 found_type;
2480         struct extent_buffer *leaf;
2481         struct btrfs_file_extent_item *fi;
2482         u64 extent_start = 0;
2483         u64 extent_num_bytes = 0;
2484         u64 item_end = 0;
2485         u64 root_gen = 0;
2486         u64 root_owner = 0;
2487         int found_extent;
2488         int del_item;
2489         int pending_del_nr = 0;
2490         int pending_del_slot = 0;
2491         int extent_type = -1;
2492         int encoding;
2493         u64 mask = root->sectorsize - 1;
2494
2495         if (root->ref_cows)
2496                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2497         path = btrfs_alloc_path();
2498         path->reada = -1;
2499         BUG_ON(!path);
2500
2501         /* FIXME, add redo link to tree so we don't leak on crash */
2502         key.objectid = inode->i_ino;
2503         key.offset = (u64)-1;
2504         key.type = (u8)-1;
2505
2506         btrfs_init_path(path);
2507
2508 search_again:
2509         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2510         if (ret < 0)
2511                 goto error;
2512
2513         if (ret > 0) {
2514                 /* there are no items in the tree for us to truncate, we're
2515                  * done
2516                  */
2517                 if (path->slots[0] == 0) {
2518                         ret = 0;
2519                         goto error;
2520                 }
2521                 path->slots[0]--;
2522         }
2523
2524         while (1) {
2525                 fi = NULL;
2526                 leaf = path->nodes[0];
2527                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2528                 found_type = btrfs_key_type(&found_key);
2529                 encoding = 0;
2530
2531                 if (found_key.objectid != inode->i_ino)
2532                         break;
2533
2534                 if (found_type < min_type)
2535                         break;
2536
2537                 item_end = found_key.offset;
2538                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2539                         fi = btrfs_item_ptr(leaf, path->slots[0],
2540                                             struct btrfs_file_extent_item);
2541                         extent_type = btrfs_file_extent_type(leaf, fi);
2542                         encoding = btrfs_file_extent_compression(leaf, fi);
2543                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2544                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2545
2546                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2547                                 item_end +=
2548                                     btrfs_file_extent_num_bytes(leaf, fi);
2549                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2550                                 item_end += btrfs_file_extent_inline_len(leaf,
2551                                                                          fi);
2552                         }
2553                         item_end--;
2554                 }
2555                 if (item_end < new_size) {
2556                         if (found_type == BTRFS_DIR_ITEM_KEY)
2557                                 found_type = BTRFS_INODE_ITEM_KEY;
2558                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2559                                 found_type = BTRFS_EXTENT_DATA_KEY;
2560                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2561                                 found_type = BTRFS_XATTR_ITEM_KEY;
2562                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2563                                 found_type = BTRFS_INODE_REF_KEY;
2564                         else if (found_type)
2565                                 found_type--;
2566                         else
2567                                 break;
2568                         btrfs_set_key_type(&key, found_type);
2569                         goto next;
2570                 }
2571                 if (found_key.offset >= new_size)
2572                         del_item = 1;
2573                 else
2574                         del_item = 0;
2575                 found_extent = 0;
2576
2577                 /* FIXME, shrink the extent if the ref count is only 1 */
2578                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2579                         goto delete;
2580
2581                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2582                         u64 num_dec;
2583                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2584                         if (!del_item && !encoding) {
2585                                 u64 orig_num_bytes =
2586                                         btrfs_file_extent_num_bytes(leaf, fi);
2587                                 extent_num_bytes = new_size -
2588                                         found_key.offset + root->sectorsize - 1;
2589                                 extent_num_bytes = extent_num_bytes &
2590                                         ~((u64)root->sectorsize - 1);
2591                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2592                                                          extent_num_bytes);
2593                                 num_dec = (orig_num_bytes -
2594                                            extent_num_bytes);
2595                                 if (root->ref_cows && extent_start != 0)
2596                                         inode_sub_bytes(inode, num_dec);
2597                                 btrfs_mark_buffer_dirty(leaf);
2598                         } else {
2599                                 extent_num_bytes =
2600                                         btrfs_file_extent_disk_num_bytes(leaf,
2601                                                                          fi);
2602                                 /* FIXME blocksize != 4096 */
2603                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2604                                 if (extent_start != 0) {
2605                                         found_extent = 1;
2606                                         if (root->ref_cows)
2607                                                 inode_sub_bytes(inode, num_dec);
2608                                 }
2609                                 root_gen = btrfs_header_generation(leaf);
2610                                 root_owner = btrfs_header_owner(leaf);
2611                         }
2612                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2613                         /*
2614                          * we can't truncate inline items that have had
2615                          * special encodings
2616                          */
2617                         if (!del_item &&
2618                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2619                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2620                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2621                                 u32 size = new_size - found_key.offset;
2622
2623                                 if (root->ref_cows) {
2624                                         inode_sub_bytes(inode, item_end + 1 -
2625                                                         new_size);
2626                                 }
2627                                 size =
2628                                     btrfs_file_extent_calc_inline_size(size);
2629                                 ret = btrfs_truncate_item(trans, root, path,
2630                                                           size, 1);
2631                                 BUG_ON(ret);
2632                         } else if (root->ref_cows) {
2633                                 inode_sub_bytes(inode, item_end + 1 -
2634                                                 found_key.offset);
2635                         }
2636                 }
2637 delete:
2638                 if (del_item) {
2639                         if (!pending_del_nr) {
2640                                 /* no pending yet, add ourselves */
2641                                 pending_del_slot = path->slots[0];
2642                                 pending_del_nr = 1;
2643                         } else if (pending_del_nr &&
2644                                    path->slots[0] + 1 == pending_del_slot) {
2645                                 /* hop on the pending chunk */
2646                                 pending_del_nr++;
2647                                 pending_del_slot = path->slots[0];
2648                         } else {
2649                                 BUG();
2650                         }
2651                 } else {
2652                         break;
2653                 }
2654                 if (found_extent) {
2655                         ret = btrfs_free_extent(trans, root, extent_start,
2656                                                 extent_num_bytes,
2657                                                 leaf->start, root_owner,
2658                                                 root_gen, inode->i_ino, 0);
2659                         BUG_ON(ret);
2660                 }
2661 next:
2662                 if (path->slots[0] == 0) {
2663                         if (pending_del_nr)
2664                                 goto del_pending;
2665                         btrfs_release_path(root, path);
2666                         goto search_again;
2667                 }
2668
2669                 path->slots[0]--;
2670                 if (pending_del_nr &&
2671                     path->slots[0] + 1 != pending_del_slot) {
2672                         struct btrfs_key debug;
2673 del_pending:
2674                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2675                                               pending_del_slot);
2676                         ret = btrfs_del_items(trans, root, path,
2677                                               pending_del_slot,
2678                                               pending_del_nr);
2679                         BUG_ON(ret);
2680                         pending_del_nr = 0;
2681                         btrfs_release_path(root, path);
2682                         goto search_again;
2683                 }
2684         }
2685         ret = 0;
2686 error:
2687         if (pending_del_nr) {
2688                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2689                                       pending_del_nr);
2690         }
2691         btrfs_free_path(path);
2692         inode->i_sb->s_dirt = 1;
2693         return ret;
2694 }
2695
2696 /*
2697  * taken from block_truncate_page, but does cow as it zeros out
2698  * any bytes left in the last page in the file.
2699  */
2700 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2701 {
2702         struct inode *inode = mapping->host;
2703         struct btrfs_root *root = BTRFS_I(inode)->root;
2704         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2705         struct btrfs_ordered_extent *ordered;
2706         char *kaddr;
2707         u32 blocksize = root->sectorsize;
2708         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2709         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2710         struct page *page;
2711         int ret = 0;
2712         u64 page_start;
2713         u64 page_end;
2714
2715         if ((offset & (blocksize - 1)) == 0)
2716                 goto out;
2717
2718         ret = -ENOMEM;
2719 again:
2720         page = grab_cache_page(mapping, index);
2721         if (!page)
2722                 goto out;
2723
2724         page_start = page_offset(page);
2725         page_end = page_start + PAGE_CACHE_SIZE - 1;
2726
2727         if (!PageUptodate(page)) {
2728                 ret = btrfs_readpage(NULL, page);
2729                 lock_page(page);
2730                 if (page->mapping != mapping) {
2731                         unlock_page(page);
2732                         page_cache_release(page);
2733                         goto again;
2734                 }
2735                 if (!PageUptodate(page)) {
2736                         ret = -EIO;
2737                         goto out_unlock;
2738                 }
2739         }
2740         wait_on_page_writeback(page);
2741
2742         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2743         set_page_extent_mapped(page);
2744
2745         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2746         if (ordered) {
2747                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2748                 unlock_page(page);
2749                 page_cache_release(page);
2750                 btrfs_start_ordered_extent(inode, ordered, 1);
2751                 btrfs_put_ordered_extent(ordered);
2752                 goto again;
2753         }
2754
2755         btrfs_set_extent_delalloc(inode, page_start, page_end);
2756         ret = 0;
2757         if (offset != PAGE_CACHE_SIZE) {
2758                 kaddr = kmap(page);
2759                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2760                 flush_dcache_page(page);
2761                 kunmap(page);
2762         }
2763         ClearPageChecked(page);
2764         set_page_dirty(page);
2765         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2766
2767 out_unlock:
2768         unlock_page(page);
2769         page_cache_release(page);
2770 out:
2771         return ret;
2772 }
2773
2774 int btrfs_cont_expand(struct inode *inode, loff_t size)
2775 {
2776         struct btrfs_trans_handle *trans;
2777         struct btrfs_root *root = BTRFS_I(inode)->root;
2778         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2779         struct extent_map *em;
2780         u64 mask = root->sectorsize - 1;
2781         u64 hole_start = (inode->i_size + mask) & ~mask;
2782         u64 block_end = (size + mask) & ~mask;
2783         u64 last_byte;
2784         u64 cur_offset;
2785         u64 hole_size;
2786         int err;
2787
2788         if (size <= hole_start)
2789                 return 0;
2790
2791         err = btrfs_check_free_space(root, 1, 0);
2792         if (err)
2793                 return err;
2794
2795         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2796
2797         while (1) {
2798                 struct btrfs_ordered_extent *ordered;
2799                 btrfs_wait_ordered_range(inode, hole_start,
2800                                          block_end - hole_start);
2801                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2802                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2803                 if (!ordered)
2804                         break;
2805                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2806                 btrfs_put_ordered_extent(ordered);
2807         }
2808
2809         trans = btrfs_start_transaction(root, 1);
2810         btrfs_set_trans_block_group(trans, inode);
2811
2812         cur_offset = hole_start;
2813         while (1) {
2814                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2815                                 block_end - cur_offset, 0);
2816                 BUG_ON(IS_ERR(em) || !em);
2817                 last_byte = min(extent_map_end(em), block_end);
2818                 last_byte = (last_byte + mask) & ~mask;
2819                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2820                         u64 hint_byte = 0;
2821                         hole_size = last_byte - cur_offset;
2822                         err = btrfs_drop_extents(trans, root, inode,
2823                                                  cur_offset,
2824                                                  cur_offset + hole_size,
2825                                                  cur_offset, &hint_byte);
2826                         if (err)
2827                                 break;
2828                         err = btrfs_insert_file_extent(trans, root,
2829                                         inode->i_ino, cur_offset, 0,
2830                                         0, hole_size, 0, hole_size,
2831                                         0, 0, 0);
2832                         btrfs_drop_extent_cache(inode, hole_start,
2833                                         last_byte - 1, 0);
2834                 }
2835                 free_extent_map(em);
2836                 cur_offset = last_byte;
2837                 if (err || cur_offset >= block_end)
2838                         break;
2839         }
2840
2841         btrfs_end_transaction(trans, root);
2842         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2843         return err;
2844 }
2845
2846 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2847 {
2848         struct inode *inode = dentry->d_inode;
2849         int err;
2850
2851         err = inode_change_ok(inode, attr);
2852         if (err)
2853                 return err;
2854
2855         if (S_ISREG(inode->i_mode) &&
2856             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2857                 err = btrfs_cont_expand(inode, attr->ia_size);
2858                 if (err)
2859                         return err;
2860         }
2861
2862         err = inode_setattr(inode, attr);
2863
2864         if (!err && ((attr->ia_valid & ATTR_MODE)))
2865                 err = btrfs_acl_chmod(inode);
2866         return err;
2867 }
2868
2869 void btrfs_delete_inode(struct inode *inode)
2870 {
2871         struct btrfs_trans_handle *trans;
2872         struct btrfs_root *root = BTRFS_I(inode)->root;
2873         unsigned long nr;
2874         int ret;
2875
2876         truncate_inode_pages(&inode->i_data, 0);
2877         if (is_bad_inode(inode)) {
2878                 btrfs_orphan_del(NULL, inode);
2879                 goto no_delete;
2880         }
2881         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2882
2883         btrfs_i_size_write(inode, 0);
2884         trans = btrfs_join_transaction(root, 1);
2885
2886         btrfs_set_trans_block_group(trans, inode);
2887         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2888         if (ret) {
2889                 btrfs_orphan_del(NULL, inode);
2890                 goto no_delete_lock;
2891         }
2892
2893         btrfs_orphan_del(trans, inode);
2894
2895         nr = trans->blocks_used;
2896         clear_inode(inode);
2897
2898         btrfs_end_transaction(trans, root);
2899         btrfs_btree_balance_dirty(root, nr);
2900         return;
2901
2902 no_delete_lock:
2903         nr = trans->blocks_used;
2904         btrfs_end_transaction(trans, root);
2905         btrfs_btree_balance_dirty(root, nr);
2906 no_delete:
2907         clear_inode(inode);
2908 }
2909
2910 /*
2911  * this returns the key found in the dir entry in the location pointer.
2912  * If no dir entries were found, location->objectid is 0.
2913  */
2914 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2915                                struct btrfs_key *location)
2916 {
2917         const char *name = dentry->d_name.name;
2918         int namelen = dentry->d_name.len;
2919         struct btrfs_dir_item *di;
2920         struct btrfs_path *path;
2921         struct btrfs_root *root = BTRFS_I(dir)->root;
2922         int ret = 0;
2923
2924         path = btrfs_alloc_path();
2925         BUG_ON(!path);
2926
2927         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2928                                     namelen, 0);
2929         if (IS_ERR(di))
2930                 ret = PTR_ERR(di);
2931
2932         if (!di || IS_ERR(di))
2933                 goto out_err;
2934
2935         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2936 out:
2937         btrfs_free_path(path);
2938         return ret;
2939 out_err:
2940         location->objectid = 0;
2941         goto out;
2942 }
2943
2944 /*
2945  * when we hit a tree root in a directory, the btrfs part of the inode
2946  * needs to be changed to reflect the root directory of the tree root.  This
2947  * is kind of like crossing a mount point.
2948  */
2949 static int fixup_tree_root_location(struct btrfs_root *root,
2950                              struct btrfs_key *location,
2951                              struct btrfs_root **sub_root,
2952                              struct dentry *dentry)
2953 {
2954         struct btrfs_root_item *ri;
2955
2956         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2957                 return 0;
2958         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2959                 return 0;
2960
2961         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2962                                         dentry->d_name.name,
2963                                         dentry->d_name.len);
2964         if (IS_ERR(*sub_root))
2965                 return PTR_ERR(*sub_root);
2966
2967         ri = &(*sub_root)->root_item;
2968         location->objectid = btrfs_root_dirid(ri);
2969         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2970         location->offset = 0;
2971
2972         return 0;
2973 }
2974
2975 static noinline void init_btrfs_i(struct inode *inode)
2976 {
2977         struct btrfs_inode *bi = BTRFS_I(inode);
2978
2979         bi->i_acl = NULL;
2980         bi->i_default_acl = NULL;
2981
2982         bi->generation = 0;
2983         bi->sequence = 0;
2984         bi->last_trans = 0;
2985         bi->logged_trans = 0;
2986         bi->delalloc_bytes = 0;
2987         bi->disk_i_size = 0;
2988         bi->flags = 0;
2989         bi->index_cnt = (u64)-1;
2990         bi->log_dirty_trans = 0;
2991         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2992         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2993                              inode->i_mapping, GFP_NOFS);
2994         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2995                              inode->i_mapping, GFP_NOFS);
2996         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2997         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2998         mutex_init(&BTRFS_I(inode)->extent_mutex);
2999         mutex_init(&BTRFS_I(inode)->log_mutex);
3000 }
3001
3002 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3003 {
3004         struct btrfs_iget_args *args = p;
3005         inode->i_ino = args->ino;
3006         init_btrfs_i(inode);
3007         BTRFS_I(inode)->root = args->root;
3008         return 0;
3009 }
3010
3011 static int btrfs_find_actor(struct inode *inode, void *opaque)
3012 {
3013         struct btrfs_iget_args *args = opaque;
3014         return args->ino == inode->i_ino &&
3015                 args->root == BTRFS_I(inode)->root;
3016 }
3017
3018 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3019                             struct btrfs_root *root, int wait)
3020 {
3021         struct inode *inode;
3022         struct btrfs_iget_args args;
3023         args.ino = objectid;
3024         args.root = root;
3025
3026         if (wait) {
3027                 inode = ilookup5(s, objectid, btrfs_find_actor,
3028                                  (void *)&args);
3029         } else {
3030                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3031                                         (void *)&args);
3032         }
3033         return inode;
3034 }
3035
3036 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3037                                 struct btrfs_root *root)
3038 {
3039         struct inode *inode;
3040         struct btrfs_iget_args args;
3041         args.ino = objectid;
3042         args.root = root;
3043
3044         inode = iget5_locked(s, objectid, btrfs_find_actor,
3045                              btrfs_init_locked_inode,
3046                              (void *)&args);
3047         return inode;
3048 }
3049
3050 /* Get an inode object given its location and corresponding root.
3051  * Returns in *is_new if the inode was read from disk
3052  */
3053 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3054                          struct btrfs_root *root, int *is_new)
3055 {
3056         struct inode *inode;
3057
3058         inode = btrfs_iget_locked(s, location->objectid, root);
3059         if (!inode)
3060                 return ERR_PTR(-EACCES);
3061
3062         if (inode->i_state & I_NEW) {
3063                 BTRFS_I(inode)->root = root;
3064                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3065                 btrfs_read_locked_inode(inode);
3066                 unlock_new_inode(inode);
3067                 if (is_new)
3068                         *is_new = 1;
3069         } else {
3070                 if (is_new)
3071                         *is_new = 0;
3072         }
3073
3074         return inode;
3075 }
3076
3077 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3078 {
3079         struct inode *inode;
3080         struct btrfs_inode *bi = BTRFS_I(dir);
3081         struct btrfs_root *root = bi->root;
3082         struct btrfs_root *sub_root = root;
3083         struct btrfs_key location;
3084         int ret, new;
3085
3086         if (dentry->d_name.len > BTRFS_NAME_LEN)
3087                 return ERR_PTR(-ENAMETOOLONG);
3088
3089         ret = btrfs_inode_by_name(dir, dentry, &location);
3090
3091         if (ret < 0)
3092                 return ERR_PTR(ret);
3093
3094         inode = NULL;
3095         if (location.objectid) {
3096                 ret = fixup_tree_root_location(root, &location, &sub_root,
3097                                                 dentry);
3098                 if (ret < 0)
3099                         return ERR_PTR(ret);
3100                 if (ret > 0)
3101                         return ERR_PTR(-ENOENT);
3102                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3103                 if (IS_ERR(inode))
3104                         return ERR_CAST(inode);
3105         }
3106         return inode;
3107 }
3108
3109 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3110                                    struct nameidata *nd)
3111 {
3112         struct inode *inode;
3113
3114         if (dentry->d_name.len > BTRFS_NAME_LEN)
3115                 return ERR_PTR(-ENAMETOOLONG);
3116
3117         inode = btrfs_lookup_dentry(dir, dentry);
3118         if (IS_ERR(inode))
3119                 return ERR_CAST(inode);
3120
3121         return d_splice_alias(inode, dentry);
3122 }
3123
3124 static unsigned char btrfs_filetype_table[] = {
3125         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3126 };
3127
3128 static int btrfs_real_readdir(struct file *filp, void *dirent,
3129                               filldir_t filldir)
3130 {
3131         struct inode *inode = filp->f_dentry->d_inode;
3132         struct btrfs_root *root = BTRFS_I(inode)->root;
3133         struct btrfs_item *item;
3134         struct btrfs_dir_item *di;
3135         struct btrfs_key key;
3136         struct btrfs_key found_key;
3137         struct btrfs_path *path;
3138         int ret;
3139         u32 nritems;
3140         struct extent_buffer *leaf;
3141         int slot;
3142         int advance;
3143         unsigned char d_type;
3144         int over = 0;
3145         u32 di_cur;
3146         u32 di_total;
3147         u32 di_len;
3148         int key_type = BTRFS_DIR_INDEX_KEY;
3149         char tmp_name[32];
3150         char *name_ptr;
3151         int name_len;
3152
3153         /* FIXME, use a real flag for deciding about the key type */
3154         if (root->fs_info->tree_root == root)
3155                 key_type = BTRFS_DIR_ITEM_KEY;
3156
3157         /* special case for "." */
3158         if (filp->f_pos == 0) {
3159                 over = filldir(dirent, ".", 1,
3160                                1, inode->i_ino,
3161                                DT_DIR);
3162                 if (over)
3163                         return 0;
3164                 filp->f_pos = 1;
3165         }
3166         /* special case for .., just use the back ref */
3167         if (filp->f_pos == 1) {
3168                 u64 pino = parent_ino(filp->f_path.dentry);
3169                 over = filldir(dirent, "..", 2,
3170                                2, pino, DT_DIR);
3171                 if (over)
3172                         return 0;
3173                 filp->f_pos = 2;
3174         }
3175         path = btrfs_alloc_path();
3176         path->reada = 2;
3177
3178         btrfs_set_key_type(&key, key_type);
3179         key.offset = filp->f_pos;
3180         key.objectid = inode->i_ino;
3181
3182         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3183         if (ret < 0)
3184                 goto err;
3185         advance = 0;
3186
3187         while (1) {
3188                 leaf = path->nodes[0];
3189                 nritems = btrfs_header_nritems(leaf);
3190                 slot = path->slots[0];
3191                 if (advance || slot >= nritems) {
3192                         if (slot >= nritems - 1) {
3193                                 ret = btrfs_next_leaf(root, path);
3194                                 if (ret)
3195                                         break;
3196                                 leaf = path->nodes[0];
3197                                 nritems = btrfs_header_nritems(leaf);
3198                                 slot = path->slots[0];
3199                         } else {
3200                                 slot++;
3201                                 path->slots[0]++;
3202                         }
3203                 }
3204
3205                 advance = 1;
3206                 item = btrfs_item_nr(leaf, slot);
3207                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3208
3209                 if (found_key.objectid != key.objectid)
3210                         break;
3211                 if (btrfs_key_type(&found_key) != key_type)
3212                         break;
3213                 if (found_key.offset < filp->f_pos)
3214                         continue;
3215
3216                 filp->f_pos = found_key.offset;
3217
3218                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3219                 di_cur = 0;
3220                 di_total = btrfs_item_size(leaf, item);
3221
3222                 while (di_cur < di_total) {
3223                         struct btrfs_key location;
3224
3225                         name_len = btrfs_dir_name_len(leaf, di);
3226                         if (name_len <= sizeof(tmp_name)) {
3227                                 name_ptr = tmp_name;
3228                         } else {
3229                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3230                                 if (!name_ptr) {
3231                                         ret = -ENOMEM;
3232                                         goto err;
3233                                 }
3234                         }
3235                         read_extent_buffer(leaf, name_ptr,
3236                                            (unsigned long)(di + 1), name_len);
3237
3238                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3239                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3240
3241                         /* is this a reference to our own snapshot? If so
3242                          * skip it
3243                          */
3244                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3245                             location.objectid == root->root_key.objectid) {
3246                                 over = 0;
3247                                 goto skip;
3248                         }
3249                         over = filldir(dirent, name_ptr, name_len,
3250                                        found_key.offset, location.objectid,
3251                                        d_type);
3252
3253 skip:
3254                         if (name_ptr != tmp_name)
3255                                 kfree(name_ptr);
3256
3257                         if (over)
3258                                 goto nopos;
3259                         di_len = btrfs_dir_name_len(leaf, di) +
3260                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3261                         di_cur += di_len;
3262                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3263                 }
3264         }
3265
3266         /* Reached end of directory/root. Bump pos past the last item. */
3267         if (key_type == BTRFS_DIR_INDEX_KEY)
3268                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3269         else
3270                 filp->f_pos++;
3271 nopos:
3272         ret = 0;
3273 err:
3274         btrfs_free_path(path);
3275         return ret;
3276 }
3277
3278 int btrfs_write_inode(struct inode *inode, int wait)
3279 {
3280         struct btrfs_root *root = BTRFS_I(inode)->root;
3281         struct btrfs_trans_handle *trans;
3282         int ret = 0;
3283
3284         if (root->fs_info->btree_inode == inode)
3285                 return 0;
3286
3287         if (wait) {
3288                 trans = btrfs_join_transaction(root, 1);
3289                 btrfs_set_trans_block_group(trans, inode);
3290                 ret = btrfs_commit_transaction(trans, root);
3291         }
3292         return ret;
3293 }
3294
3295 /*
3296  * This is somewhat expensive, updating the tree every time the
3297  * inode changes.  But, it is most likely to find the inode in cache.
3298  * FIXME, needs more benchmarking...there are no reasons other than performance
3299  * to keep or drop this code.
3300  */
3301 void btrfs_dirty_inode(struct inode *inode)
3302 {
3303         struct btrfs_root *root = BTRFS_I(inode)->root;
3304         struct btrfs_trans_handle *trans;
3305
3306         trans = btrfs_join_transaction(root, 1);
3307         btrfs_set_trans_block_group(trans, inode);
3308         btrfs_update_inode(trans, root, inode);
3309         btrfs_end_transaction(trans, root);
3310 }
3311
3312 /*
3313  * find the highest existing sequence number in a directory
3314  * and then set the in-memory index_cnt variable to reflect
3315  * free sequence numbers
3316  */
3317 static int btrfs_set_inode_index_count(struct inode *inode)
3318 {
3319         struct btrfs_root *root = BTRFS_I(inode)->root;
3320         struct btrfs_key key, found_key;
3321         struct btrfs_path *path;
3322         struct extent_buffer *leaf;
3323         int ret;
3324
3325         key.objectid = inode->i_ino;
3326         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3327         key.offset = (u64)-1;
3328
3329         path = btrfs_alloc_path();
3330         if (!path)
3331                 return -ENOMEM;
3332
3333         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3334         if (ret < 0)
3335                 goto out;
3336         /* FIXME: we should be able to handle this */
3337         if (ret == 0)
3338                 goto out;
3339         ret = 0;
3340
3341         /*
3342          * MAGIC NUMBER EXPLANATION:
3343          * since we search a directory based on f_pos we have to start at 2
3344          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3345          * else has to start at 2
3346          */
3347         if (path->slots[0] == 0) {
3348                 BTRFS_I(inode)->index_cnt = 2;
3349                 goto out;
3350         }
3351
3352         path->slots[0]--;
3353
3354         leaf = path->nodes[0];
3355         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3356
3357         if (found_key.objectid != inode->i_ino ||
3358             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3359                 BTRFS_I(inode)->index_cnt = 2;
3360                 goto out;
3361         }
3362
3363         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3364 out:
3365         btrfs_free_path(path);
3366         return ret;
3367 }
3368
3369 /*
3370  * helper to find a free sequence number in a given directory.  This current
3371  * code is very simple, later versions will do smarter things in the btree
3372  */
3373 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3374 {
3375         int ret = 0;
3376
3377         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3378                 ret = btrfs_set_inode_index_count(dir);
3379                 if (ret)
3380                         return ret;
3381         }
3382
3383         *index = BTRFS_I(dir)->index_cnt;
3384         BTRFS_I(dir)->index_cnt++;
3385
3386         return ret;
3387 }
3388
3389 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3390                                      struct btrfs_root *root,
3391                                      struct inode *dir,
3392                                      const char *name, int name_len,
3393                                      u64 ref_objectid, u64 objectid,
3394                                      u64 alloc_hint, int mode, u64 *index)
3395 {
3396         struct inode *inode;
3397         struct btrfs_inode_item *inode_item;
3398         struct btrfs_key *location;
3399         struct btrfs_path *path;
3400         struct btrfs_inode_ref *ref;
3401         struct btrfs_key key[2];
3402         u32 sizes[2];
3403         unsigned long ptr;
3404         int ret;
3405         int owner;
3406
3407         path = btrfs_alloc_path();
3408         BUG_ON(!path);
3409
3410         inode = new_inode(root->fs_info->sb);
3411         if (!inode)
3412                 return ERR_PTR(-ENOMEM);
3413
3414         if (dir) {
3415                 ret = btrfs_set_inode_index(dir, index);
3416                 if (ret)
3417                         return ERR_PTR(ret);
3418         }
3419         /*
3420          * index_cnt is ignored for everything but a dir,
3421          * btrfs_get_inode_index_count has an explanation for the magic
3422          * number
3423          */
3424         init_btrfs_i(inode);
3425         BTRFS_I(inode)->index_cnt = 2;
3426         BTRFS_I(inode)->root = root;
3427         BTRFS_I(inode)->generation = trans->transid;
3428
3429         if (mode & S_IFDIR)
3430                 owner = 0;
3431         else
3432                 owner = 1;
3433         BTRFS_I(inode)->block_group =
3434                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3435         if ((mode & S_IFREG)) {
3436                 if (btrfs_test_opt(root, NODATASUM))
3437                         btrfs_set_flag(inode, NODATASUM);
3438                 if (btrfs_test_opt(root, NODATACOW))
3439                         btrfs_set_flag(inode, NODATACOW);
3440         }
3441
3442         key[0].objectid = objectid;
3443         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3444         key[0].offset = 0;
3445
3446         key[1].objectid = objectid;
3447         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3448         key[1].offset = ref_objectid;
3449
3450         sizes[0] = sizeof(struct btrfs_inode_item);
3451         sizes[1] = name_len + sizeof(*ref);
3452
3453         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3454         if (ret != 0)
3455                 goto fail;
3456
3457         if (objectid > root->highest_inode)
3458                 root->highest_inode = objectid;
3459
3460         inode->i_uid = current_fsuid();
3461         inode->i_gid = current_fsgid();
3462         inode->i_mode = mode;
3463         inode->i_ino = objectid;
3464         inode_set_bytes(inode, 0);
3465         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3466         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3467                                   struct btrfs_inode_item);
3468         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3469
3470         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3471                              struct btrfs_inode_ref);
3472         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3473         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3474         ptr = (unsigned long)(ref + 1);
3475         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3476
3477         btrfs_mark_buffer_dirty(path->nodes[0]);
3478         btrfs_free_path(path);
3479
3480         location = &BTRFS_I(inode)->location;
3481         location->objectid = objectid;
3482         location->offset = 0;
3483         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3484
3485         insert_inode_hash(inode);
3486         return inode;
3487 fail:
3488         if (dir)
3489                 BTRFS_I(dir)->index_cnt--;
3490         btrfs_free_path(path);
3491         return ERR_PTR(ret);
3492 }
3493
3494 static inline u8 btrfs_inode_type(struct inode *inode)
3495 {
3496         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3497 }
3498
3499 /*
3500  * utility function to add 'inode' into 'parent_inode' with
3501  * a give name and a given sequence number.
3502  * if 'add_backref' is true, also insert a backref from the
3503  * inode to the parent directory.
3504  */
3505 int btrfs_add_link(struct btrfs_trans_handle *trans,
3506                    struct inode *parent_inode, struct inode *inode,
3507                    const char *name, int name_len, int add_backref, u64 index)
3508 {
3509         int ret;
3510         struct btrfs_key key;
3511         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3512
3513         key.objectid = inode->i_ino;
3514         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3515         key.offset = 0;
3516
3517         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3518                                     parent_inode->i_ino,
3519                                     &key, btrfs_inode_type(inode),
3520                                     index);
3521         if (ret == 0) {
3522                 if (add_backref) {
3523                         ret = btrfs_insert_inode_ref(trans, root,
3524                                                      name, name_len,
3525                                                      inode->i_ino,
3526                                                      parent_inode->i_ino,
3527                                                      index);
3528                 }
3529                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3530                                    name_len * 2);
3531                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3532                 ret = btrfs_update_inode(trans, root, parent_inode);
3533         }
3534         return ret;
3535 }
3536
3537 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3538                             struct dentry *dentry, struct inode *inode,
3539                             int backref, u64 index)
3540 {
3541         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3542                                  inode, dentry->d_name.name,
3543                                  dentry->d_name.len, backref, index);
3544         if (!err) {
3545                 d_instantiate(dentry, inode);
3546                 return 0;
3547         }
3548         if (err > 0)
3549                 err = -EEXIST;
3550         return err;
3551 }
3552
3553 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3554                         int mode, dev_t rdev)
3555 {
3556         struct btrfs_trans_handle *trans;
3557         struct btrfs_root *root = BTRFS_I(dir)->root;
3558         struct inode *inode = NULL;
3559         int err;
3560         int drop_inode = 0;
3561         u64 objectid;
3562         unsigned long nr = 0;
3563         u64 index = 0;
3564
3565         if (!new_valid_dev(rdev))
3566                 return -EINVAL;
3567
3568         err = btrfs_check_free_space(root, 1, 0);
3569         if (err)
3570                 goto fail;
3571
3572         trans = btrfs_start_transaction(root, 1);
3573         btrfs_set_trans_block_group(trans, dir);
3574
3575         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3576         if (err) {
3577                 err = -ENOSPC;
3578                 goto out_unlock;
3579         }
3580
3581         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3582                                 dentry->d_name.len,
3583                                 dentry->d_parent->d_inode->i_ino, objectid,
3584                                 BTRFS_I(dir)->block_group, mode, &index);
3585         err = PTR_ERR(inode);
3586         if (IS_ERR(inode))
3587                 goto out_unlock;
3588
3589         err = btrfs_init_acl(inode, dir);
3590         if (err) {
3591                 drop_inode = 1;
3592                 goto out_unlock;
3593         }
3594
3595         btrfs_set_trans_block_group(trans, inode);
3596         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3597         if (err)
3598                 drop_inode = 1;
3599         else {
3600                 inode->i_op = &btrfs_special_inode_operations;
3601                 init_special_inode(inode, inode->i_mode, rdev);
3602                 btrfs_update_inode(trans, root, inode);
3603         }
3604         dir->i_sb->s_dirt = 1;
3605         btrfs_update_inode_block_group(trans, inode);
3606         btrfs_update_inode_block_group(trans, dir);
3607 out_unlock:
3608         nr = trans->blocks_used;
3609         btrfs_end_transaction_throttle(trans, root);
3610 fail:
3611         if (drop_inode) {
3612                 inode_dec_link_count(inode);
3613                 iput(inode);
3614         }
3615         btrfs_btree_balance_dirty(root, nr);
3616         return err;
3617 }
3618
3619 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3620                         int mode, struct nameidata *nd)
3621 {
3622         struct btrfs_trans_handle *trans;
3623         struct btrfs_root *root = BTRFS_I(dir)->root;
3624         struct inode *inode = NULL;
3625         int err;
3626         int drop_inode = 0;
3627         unsigned long nr = 0;
3628         u64 objectid;
3629         u64 index = 0;
3630
3631         err = btrfs_check_free_space(root, 1, 0);
3632         if (err)
3633                 goto fail;
3634         trans = btrfs_start_transaction(root, 1);
3635         btrfs_set_trans_block_group(trans, dir);
3636
3637         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3638         if (err) {
3639                 err = -ENOSPC;
3640                 goto out_unlock;
3641         }
3642
3643         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3644                                 dentry->d_name.len,
3645                                 dentry->d_parent->d_inode->i_ino,
3646                                 objectid, BTRFS_I(dir)->block_group, mode,
3647                                 &index);
3648         err = PTR_ERR(inode);
3649         if (IS_ERR(inode))
3650                 goto out_unlock;
3651
3652         err = btrfs_init_acl(inode, dir);
3653         if (err) {
3654                 drop_inode = 1;
3655                 goto out_unlock;
3656         }
3657
3658         btrfs_set_trans_block_group(trans, inode);
3659         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3660         if (err)
3661                 drop_inode = 1;
3662         else {
3663                 inode->i_mapping->a_ops = &btrfs_aops;
3664                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3665                 inode->i_fop = &btrfs_file_operations;
3666                 inode->i_op = &btrfs_file_inode_operations;
3667                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3668         }
3669         dir->i_sb->s_dirt = 1;
3670         btrfs_update_inode_block_group(trans, inode);
3671         btrfs_update_inode_block_group(trans, dir);
3672 out_unlock:
3673         nr = trans->blocks_used;
3674         btrfs_end_transaction_throttle(trans, root);
3675 fail:
3676         if (drop_inode) {
3677                 inode_dec_link_count(inode);
3678                 iput(inode);
3679         }
3680         btrfs_btree_balance_dirty(root, nr);
3681         return err;
3682 }
3683
3684 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3685                       struct dentry *dentry)
3686 {
3687         struct btrfs_trans_handle *trans;
3688         struct btrfs_root *root = BTRFS_I(dir)->root;
3689         struct inode *inode = old_dentry->d_inode;
3690         u64 index;
3691         unsigned long nr = 0;
3692         int err;
3693         int drop_inode = 0;
3694
3695         if (inode->i_nlink == 0)
3696                 return -ENOENT;
3697
3698         btrfs_inc_nlink(inode);
3699         err = btrfs_check_free_space(root, 1, 0);
3700         if (err)
3701                 goto fail;
3702         err = btrfs_set_inode_index(dir, &index);
3703         if (err)
3704                 goto fail;
3705
3706         trans = btrfs_start_transaction(root, 1);
3707
3708         btrfs_set_trans_block_group(trans, dir);
3709         atomic_inc(&inode->i_count);
3710
3711         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3712
3713         if (err)
3714                 drop_inode = 1;
3715
3716         dir->i_sb->s_dirt = 1;
3717         btrfs_update_inode_block_group(trans, dir);
3718         err = btrfs_update_inode(trans, root, inode);
3719
3720         if (err)
3721                 drop_inode = 1;
3722
3723         nr = trans->blocks_used;
3724         btrfs_end_transaction_throttle(trans, root);
3725 fail:
3726         if (drop_inode) {
3727                 inode_dec_link_count(inode);
3728                 iput(inode);
3729         }
3730         btrfs_btree_balance_dirty(root, nr);
3731         return err;
3732 }
3733
3734 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3735 {
3736         struct inode *inode = NULL;
3737         struct btrfs_trans_handle *trans;
3738         struct btrfs_root *root = BTRFS_I(dir)->root;
3739         int err = 0;
3740         int drop_on_err = 0;
3741         u64 objectid = 0;
3742         u64 index = 0;
3743         unsigned long nr = 1;
3744
3745         err = btrfs_check_free_space(root, 1, 0);
3746         if (err)
3747                 goto out_unlock;
3748
3749         trans = btrfs_start_transaction(root, 1);
3750         btrfs_set_trans_block_group(trans, dir);
3751
3752         if (IS_ERR(trans)) {
3753                 err = PTR_ERR(trans);
3754                 goto out_unlock;
3755         }
3756
3757         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3758         if (err) {
3759                 err = -ENOSPC;
3760                 goto out_unlock;
3761         }
3762
3763         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3764                                 dentry->d_name.len,
3765                                 dentry->d_parent->d_inode->i_ino, objectid,
3766                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3767                                 &index);
3768         if (IS_ERR(inode)) {
3769                 err = PTR_ERR(inode);
3770                 goto out_fail;
3771         }
3772
3773         drop_on_err = 1;
3774
3775         err = btrfs_init_acl(inode, dir);
3776         if (err)
3777                 goto out_fail;
3778
3779         inode->i_op = &btrfs_dir_inode_operations;
3780         inode->i_fop = &btrfs_dir_file_operations;
3781         btrfs_set_trans_block_group(trans, inode);
3782
3783         btrfs_i_size_write(inode, 0);
3784         err = btrfs_update_inode(trans, root, inode);
3785         if (err)
3786                 goto out_fail;
3787
3788         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3789                                  inode, dentry->d_name.name,
3790                                  dentry->d_name.len, 0, index);
3791         if (err)
3792                 goto out_fail;
3793
3794         d_instantiate(dentry, inode);
3795         drop_on_err = 0;
3796         dir->i_sb->s_dirt = 1;
3797         btrfs_update_inode_block_group(trans, inode);
3798         btrfs_update_inode_block_group(trans, dir);
3799
3800 out_fail:
3801         nr = trans->blocks_used;
3802         btrfs_end_transaction_throttle(trans, root);
3803
3804 out_unlock:
3805         if (drop_on_err)
3806                 iput(inode);
3807         btrfs_btree_balance_dirty(root, nr);
3808         return err;
3809 }
3810
3811 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3812  * and an extent that you want to insert, deal with overlap and insert
3813  * the new extent into the tree.
3814  */
3815 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3816                                 struct extent_map *existing,
3817                                 struct extent_map *em,
3818                                 u64 map_start, u64 map_len)
3819 {
3820         u64 start_diff;
3821
3822         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3823         start_diff = map_start - em->start;
3824         em->start = map_start;
3825         em->len = map_len;
3826         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3827             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3828                 em->block_start += start_diff;
3829                 em->block_len -= start_diff;
3830         }
3831         return add_extent_mapping(em_tree, em);
3832 }
3833
3834 static noinline int uncompress_inline(struct btrfs_path *path,
3835                                       struct inode *inode, struct page *page,
3836                                       size_t pg_offset, u64 extent_offset,
3837                                       struct btrfs_file_extent_item *item)
3838 {
3839         int ret;
3840         struct extent_buffer *leaf = path->nodes[0];
3841         char *tmp;
3842         size_t max_size;
3843         unsigned long inline_size;
3844         unsigned long ptr;
3845
3846         WARN_ON(pg_offset != 0);
3847         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3848         inline_size = btrfs_file_extent_inline_item_len(leaf,
3849                                         btrfs_item_nr(leaf, path->slots[0]));
3850         tmp = kmalloc(inline_size, GFP_NOFS);
3851         ptr = btrfs_file_extent_inline_start(item);
3852
3853         read_extent_buffer(leaf, tmp, ptr, inline_size);
3854
3855         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3856         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3857                                     inline_size, max_size);
3858         if (ret) {
3859                 char *kaddr = kmap_atomic(page, KM_USER0);
3860                 unsigned long copy_size = min_t(u64,
3861                                   PAGE_CACHE_SIZE - pg_offset,
3862                                   max_size - extent_offset);
3863                 memset(kaddr + pg_offset, 0, copy_size);
3864                 kunmap_atomic(kaddr, KM_USER0);
3865         }
3866         kfree(tmp);
3867         return 0;
3868 }
3869
3870 /*
3871  * a bit scary, this does extent mapping from logical file offset to the disk.
3872  * the ugly parts come from merging extents from the disk with the in-ram
3873  * representation.  This gets more complex because of the data=ordered code,
3874  * where the in-ram extents might be locked pending data=ordered completion.
3875  *
3876  * This also copies inline extents directly into the page.
3877  */
3878
3879 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3880                                     size_t pg_offset, u64 start, u64 len,
3881                                     int create)
3882 {
3883         int ret;
3884         int err = 0;
3885         u64 bytenr;
3886         u64 extent_start = 0;
3887         u64 extent_end = 0;
3888         u64 objectid = inode->i_ino;
3889         u32 found_type;
3890         struct btrfs_path *path = NULL;
3891         struct btrfs_root *root = BTRFS_I(inode)->root;
3892         struct btrfs_file_extent_item *item;
3893         struct extent_buffer *leaf;
3894         struct btrfs_key found_key;
3895         struct extent_map *em = NULL;
3896         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3898         struct btrfs_trans_handle *trans = NULL;
3899         int compressed;
3900
3901 again:
3902         spin_lock(&em_tree->lock);
3903         em = lookup_extent_mapping(em_tree, start, len);
3904         if (em)
3905                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3906         spin_unlock(&em_tree->lock);
3907
3908         if (em) {
3909                 if (em->start > start || em->start + em->len <= start)
3910                         free_extent_map(em);
3911                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3912                         free_extent_map(em);
3913                 else
3914                         goto out;
3915         }
3916         em = alloc_extent_map(GFP_NOFS);
3917         if (!em) {
3918                 err = -ENOMEM;
3919                 goto out;
3920         }
3921         em->bdev = root->fs_info->fs_devices->latest_bdev;
3922         em->start = EXTENT_MAP_HOLE;
3923         em->orig_start = EXTENT_MAP_HOLE;
3924         em->len = (u64)-1;
3925         em->block_len = (u64)-1;
3926
3927         if (!path) {
3928                 path = btrfs_alloc_path();
3929                 BUG_ON(!path);
3930         }
3931
3932         ret = btrfs_lookup_file_extent(trans, root, path,
3933                                        objectid, start, trans != NULL);
3934         if (ret < 0) {
3935                 err = ret;
3936                 goto out;
3937         }
3938
3939         if (ret != 0) {
3940                 if (path->slots[0] == 0)
3941                         goto not_found;
3942                 path->slots[0]--;
3943         }
3944
3945         leaf = path->nodes[0];
3946         item = btrfs_item_ptr(leaf, path->slots[0],
3947                               struct btrfs_file_extent_item);
3948         /* are we inside the extent that was found? */
3949         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3950         found_type = btrfs_key_type(&found_key);
3951         if (found_key.objectid != objectid ||
3952             found_type != BTRFS_EXTENT_DATA_KEY) {
3953                 goto not_found;
3954         }
3955
3956         found_type = btrfs_file_extent_type(leaf, item);
3957         extent_start = found_key.offset;
3958         compressed = btrfs_file_extent_compression(leaf, item);
3959         if (found_type == BTRFS_FILE_EXTENT_REG ||
3960             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3961                 extent_end = extent_start +
3962                        btrfs_file_extent_num_bytes(leaf, item);
3963         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3964                 size_t size;
3965                 size = btrfs_file_extent_inline_len(leaf, item);
3966                 extent_end = (extent_start + size + root->sectorsize - 1) &
3967                         ~((u64)root->sectorsize - 1);
3968         }
3969
3970         if (start >= extent_end) {
3971                 path->slots[0]++;
3972                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3973                         ret = btrfs_next_leaf(root, path);
3974                         if (ret < 0) {
3975                                 err = ret;
3976                                 goto out;
3977                         }
3978                         if (ret > 0)
3979                                 goto not_found;
3980                         leaf = path->nodes[0];
3981                 }
3982                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3983                 if (found_key.objectid != objectid ||
3984                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3985                         goto not_found;
3986                 if (start + len <= found_key.offset)
3987                         goto not_found;
3988                 em->start = start;
3989                 em->len = found_key.offset - start;
3990                 goto not_found_em;
3991         }
3992
3993         if (found_type == BTRFS_FILE_EXTENT_REG ||
3994             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3995                 em->start = extent_start;
3996                 em->len = extent_end - extent_start;
3997                 em->orig_start = extent_start -
3998                                  btrfs_file_extent_offset(leaf, item);
3999                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4000                 if (bytenr == 0) {
4001                         em->block_start = EXTENT_MAP_HOLE;
4002                         goto insert;
4003                 }
4004                 if (compressed) {
4005                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4006                         em->block_start = bytenr;
4007                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4008                                                                          item);
4009                 } else {
4010                         bytenr += btrfs_file_extent_offset(leaf, item);
4011                         em->block_start = bytenr;
4012                         em->block_len = em->len;
4013                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4014                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4015                 }
4016                 goto insert;
4017         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4018                 unsigned long ptr;
4019                 char *map;
4020                 size_t size;
4021                 size_t extent_offset;
4022                 size_t copy_size;
4023
4024                 em->block_start = EXTENT_MAP_INLINE;
4025                 if (!page || create) {
4026                         em->start = extent_start;
4027                         em->len = extent_end - extent_start;
4028                         goto out;
4029                 }
4030
4031                 size = btrfs_file_extent_inline_len(leaf, item);
4032                 extent_offset = page_offset(page) + pg_offset - extent_start;
4033                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4034                                 size - extent_offset);
4035                 em->start = extent_start + extent_offset;
4036                 em->len = (copy_size + root->sectorsize - 1) &
4037                         ~((u64)root->sectorsize - 1);
4038                 em->orig_start = EXTENT_MAP_INLINE;
4039                 if (compressed)
4040                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4041                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4042                 if (create == 0 && !PageUptodate(page)) {
4043                         if (btrfs_file_extent_compression(leaf, item) ==
4044                             BTRFS_COMPRESS_ZLIB) {
4045                                 ret = uncompress_inline(path, inode, page,
4046                                                         pg_offset,
4047                                                         extent_offset, item);
4048                                 BUG_ON(ret);
4049                         } else {
4050                                 map = kmap(page);
4051                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4052                                                    copy_size);
4053                                 kunmap(page);
4054                         }
4055                         flush_dcache_page(page);
4056                 } else if (create && PageUptodate(page)) {
4057                         if (!trans) {
4058                                 kunmap(page);
4059                                 free_extent_map(em);
4060                                 em = NULL;
4061                                 btrfs_release_path(root, path);
4062                                 trans = btrfs_join_transaction(root, 1);
4063                                 goto again;
4064                         }
4065                         map = kmap(page);
4066                         write_extent_buffer(leaf, map + pg_offset, ptr,
4067                                             copy_size);
4068                         kunmap(page);
4069                         btrfs_mark_buffer_dirty(leaf);
4070                 }
4071                 set_extent_uptodate(io_tree, em->start,
4072                                     extent_map_end(em) - 1, GFP_NOFS);
4073                 goto insert;
4074         } else {
4075                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4076                 WARN_ON(1);
4077         }
4078 not_found:
4079         em->start = start;
4080         em->len = len;
4081 not_found_em:
4082         em->block_start = EXTENT_MAP_HOLE;
4083         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4084 insert:
4085         btrfs_release_path(root, path);
4086         if (em->start > start || extent_map_end(em) <= start) {
4087                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4088                        "[%llu %llu]\n", (unsigned long long)em->start,
4089                        (unsigned long long)em->len,
4090                        (unsigned long long)start,
4091                        (unsigned long long)len);
4092                 err = -EIO;
4093                 goto out;
4094         }
4095
4096         err = 0;
4097         spin_lock(&em_tree->lock);
4098         ret = add_extent_mapping(em_tree, em);
4099         /* it is possible that someone inserted the extent into the tree
4100          * while we had the lock dropped.  It is also possible that
4101          * an overlapping map exists in the tree
4102          */
4103         if (ret == -EEXIST) {
4104                 struct extent_map *existing;
4105
4106                 ret = 0;
4107
4108                 existing = lookup_extent_mapping(em_tree, start, len);
4109                 if (existing && (existing->start > start ||
4110                     existing->start + existing->len <= start)) {
4111                         free_extent_map(existing);
4112                         existing = NULL;
4113                 }
4114                 if (!existing) {
4115                         existing = lookup_extent_mapping(em_tree, em->start,
4116                                                          em->len);
4117                         if (existing) {
4118                                 err = merge_extent_mapping(em_tree, existing,
4119                                                            em, start,
4120                                                            root->sectorsize);
4121                                 free_extent_map(existing);
4122                                 if (err) {
4123                                         free_extent_map(em);
4124                                         em = NULL;
4125                                 }
4126                         } else {
4127                                 err = -EIO;
4128                                 free_extent_map(em);
4129                                 em = NULL;
4130                         }
4131                 } else {
4132                         free_extent_map(em);
4133                         em = existing;
4134                         err = 0;
4135                 }
4136         }
4137         spin_unlock(&em_tree->lock);
4138 out:
4139         if (path)
4140                 btrfs_free_path(path);
4141         if (trans) {
4142                 ret = btrfs_end_transaction(trans, root);
4143                 if (!err)
4144                         err = ret;
4145         }
4146         if (err) {
4147                 free_extent_map(em);
4148                 WARN_ON(1);
4149                 return ERR_PTR(err);
4150         }
4151         return em;
4152 }
4153
4154 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4155                         const struct iovec *iov, loff_t offset,
4156                         unsigned long nr_segs)
4157 {
4158         return -EINVAL;
4159 }
4160
4161 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4162 {
4163         return extent_bmap(mapping, iblock, btrfs_get_extent);
4164 }
4165
4166 int btrfs_readpage(struct file *file, struct page *page)
4167 {
4168         struct extent_io_tree *tree;
4169         tree = &BTRFS_I(page->mapping->host)->io_tree;
4170         return extent_read_full_page(tree, page, btrfs_get_extent);
4171 }
4172
4173 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4174 {
4175         struct extent_io_tree *tree;
4176
4177
4178         if (current->flags & PF_MEMALLOC) {
4179                 redirty_page_for_writepage(wbc, page);
4180                 unlock_page(page);
4181                 return 0;
4182         }
4183         tree = &BTRFS_I(page->mapping->host)->io_tree;
4184         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4185 }
4186
4187 int btrfs_writepages(struct address_space *mapping,
4188                      struct writeback_control *wbc)
4189 {
4190         struct extent_io_tree *tree;
4191
4192         tree = &BTRFS_I(mapping->host)->io_tree;
4193         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4194 }
4195
4196 static int
4197 btrfs_readpages(struct file *file, struct address_space *mapping,
4198                 struct list_head *pages, unsigned nr_pages)
4199 {
4200         struct extent_io_tree *tree;
4201         tree = &BTRFS_I(mapping->host)->io_tree;
4202         return extent_readpages(tree, mapping, pages, nr_pages,
4203                                 btrfs_get_extent);
4204 }
4205 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4206 {
4207         struct extent_io_tree *tree;
4208         struct extent_map_tree *map;
4209         int ret;
4210
4211         tree = &BTRFS_I(page->mapping->host)->io_tree;
4212         map = &BTRFS_I(page->mapping->host)->extent_tree;
4213         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4214         if (ret == 1) {
4215                 ClearPagePrivate(page);
4216                 set_page_private(page, 0);
4217                 page_cache_release(page);
4218         }
4219         return ret;
4220 }
4221
4222 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4223 {
4224         if (PageWriteback(page) || PageDirty(page))
4225                 return 0;
4226         return __btrfs_releasepage(page, gfp_flags);
4227 }
4228
4229 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4230 {
4231         struct extent_io_tree *tree;
4232         struct btrfs_ordered_extent *ordered;
4233         u64 page_start = page_offset(page);
4234         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4235
4236         wait_on_page_writeback(page);
4237         tree = &BTRFS_I(page->mapping->host)->io_tree;
4238         if (offset) {
4239                 btrfs_releasepage(page, GFP_NOFS);
4240                 return;
4241         }
4242
4243         lock_extent(tree, page_start, page_end, GFP_NOFS);
4244         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4245                                            page_offset(page));
4246         if (ordered) {
4247                 /*
4248                  * IO on this page will never be started, so we need
4249                  * to account for any ordered extents now
4250                  */
4251                 clear_extent_bit(tree, page_start, page_end,
4252                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4253                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4254                 btrfs_finish_ordered_io(page->mapping->host,
4255                                         page_start, page_end);
4256                 btrfs_put_ordered_extent(ordered);
4257                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4258         }
4259         clear_extent_bit(tree, page_start, page_end,
4260                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4261                  EXTENT_ORDERED,
4262                  1, 1, GFP_NOFS);
4263         __btrfs_releasepage(page, GFP_NOFS);
4264
4265         ClearPageChecked(page);
4266         if (PagePrivate(page)) {
4267                 ClearPagePrivate(page);
4268                 set_page_private(page, 0);
4269                 page_cache_release(page);
4270         }
4271 }
4272
4273 /*
4274  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4275  * called from a page fault handler when a page is first dirtied. Hence we must
4276  * be careful to check for EOF conditions here. We set the page up correctly
4277  * for a written page which means we get ENOSPC checking when writing into
4278  * holes and correct delalloc and unwritten extent mapping on filesystems that
4279  * support these features.
4280  *
4281  * We are not allowed to take the i_mutex here so we have to play games to
4282  * protect against truncate races as the page could now be beyond EOF.  Because
4283  * vmtruncate() writes the inode size before removing pages, once we have the
4284  * page lock we can determine safely if the page is beyond EOF. If it is not
4285  * beyond EOF, then the page is guaranteed safe against truncation until we
4286  * unlock the page.
4287  */
4288 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4289 {
4290         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4291         struct btrfs_root *root = BTRFS_I(inode)->root;
4292         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4293         struct btrfs_ordered_extent *ordered;
4294         char *kaddr;
4295         unsigned long zero_start;
4296         loff_t size;
4297         int ret;
4298         u64 page_start;
4299         u64 page_end;
4300
4301         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4302         if (ret)
4303                 goto out;
4304
4305         ret = -EINVAL;
4306 again:
4307         lock_page(page);
4308         size = i_size_read(inode);
4309         page_start = page_offset(page);
4310         page_end = page_start + PAGE_CACHE_SIZE - 1;
4311
4312         if ((page->mapping != inode->i_mapping) ||
4313             (page_start >= size)) {
4314                 /* page got truncated out from underneath us */
4315                 goto out_unlock;
4316         }
4317         wait_on_page_writeback(page);
4318
4319         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4320         set_page_extent_mapped(page);
4321
4322         /*
4323          * we can't set the delalloc bits if there are pending ordered
4324          * extents.  Drop our locks and wait for them to finish
4325          */
4326         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4327         if (ordered) {
4328                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4329                 unlock_page(page);
4330                 btrfs_start_ordered_extent(inode, ordered, 1);
4331                 btrfs_put_ordered_extent(ordered);
4332                 goto again;
4333         }
4334
4335         btrfs_set_extent_delalloc(inode, page_start, page_end);
4336         ret = 0;
4337
4338         /* page is wholly or partially inside EOF */
4339         if (page_start + PAGE_CACHE_SIZE > size)
4340                 zero_start = size & ~PAGE_CACHE_MASK;
4341         else
4342                 zero_start = PAGE_CACHE_SIZE;
4343
4344         if (zero_start != PAGE_CACHE_SIZE) {
4345                 kaddr = kmap(page);
4346                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4347                 flush_dcache_page(page);
4348                 kunmap(page);
4349         }
4350         ClearPageChecked(page);
4351         set_page_dirty(page);
4352         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4353
4354 out_unlock:
4355         unlock_page(page);
4356 out:
4357         return ret;
4358 }
4359
4360 static void btrfs_truncate(struct inode *inode)
4361 {
4362         struct btrfs_root *root = BTRFS_I(inode)->root;
4363         int ret;
4364         struct btrfs_trans_handle *trans;
4365         unsigned long nr;
4366         u64 mask = root->sectorsize - 1;
4367
4368         if (!S_ISREG(inode->i_mode))
4369                 return;
4370         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4371                 return;
4372
4373         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4374         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4375
4376         trans = btrfs_start_transaction(root, 1);
4377         btrfs_set_trans_block_group(trans, inode);
4378         btrfs_i_size_write(inode, inode->i_size);
4379
4380         ret = btrfs_orphan_add(trans, inode);
4381         if (ret)
4382                 goto out;
4383         /* FIXME, add redo link to tree so we don't leak on crash */
4384         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4385                                       BTRFS_EXTENT_DATA_KEY);
4386         btrfs_update_inode(trans, root, inode);
4387
4388         ret = btrfs_orphan_del(trans, inode);
4389         BUG_ON(ret);
4390
4391 out:
4392         nr = trans->blocks_used;
4393         ret = btrfs_end_transaction_throttle(trans, root);
4394         BUG_ON(ret);
4395         btrfs_btree_balance_dirty(root, nr);
4396 }
4397
4398 /*
4399  * create a new subvolume directory/inode (helper for the ioctl).
4400  */
4401 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4402                              struct btrfs_root *new_root, struct dentry *dentry,
4403                              u64 new_dirid, u64 alloc_hint)
4404 {
4405         struct inode *inode;
4406         int error;
4407         u64 index = 0;
4408
4409         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4410                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4411         if (IS_ERR(inode))
4412                 return PTR_ERR(inode);
4413         inode->i_op = &btrfs_dir_inode_operations;
4414         inode->i_fop = &btrfs_dir_file_operations;
4415
4416         inode->i_nlink = 1;
4417         btrfs_i_size_write(inode, 0);
4418
4419         error = btrfs_update_inode(trans, new_root, inode);
4420         if (error)
4421                 return error;
4422
4423         d_instantiate(dentry, inode);
4424         return 0;
4425 }
4426
4427 /* helper function for file defrag and space balancing.  This
4428  * forces readahead on a given range of bytes in an inode
4429  */
4430 unsigned long btrfs_force_ra(struct address_space *mapping,
4431                               struct file_ra_state *ra, struct file *file,
4432                               pgoff_t offset, pgoff_t last_index)
4433 {
4434         pgoff_t req_size = last_index - offset + 1;
4435
4436         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4437         return offset + req_size;
4438 }
4439
4440 struct inode *btrfs_alloc_inode(struct super_block *sb)
4441 {
4442         struct btrfs_inode *ei;
4443
4444         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4445         if (!ei)
4446                 return NULL;
4447         ei->last_trans = 0;
4448         ei->logged_trans = 0;
4449         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4450         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4451         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4452         INIT_LIST_HEAD(&ei->i_orphan);
4453         return &ei->vfs_inode;
4454 }
4455
4456 void btrfs_destroy_inode(struct inode *inode)
4457 {
4458         struct btrfs_ordered_extent *ordered;
4459         WARN_ON(!list_empty(&inode->i_dentry));
4460         WARN_ON(inode->i_data.nrpages);
4461
4462         if (BTRFS_I(inode)->i_acl &&
4463             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4464                 posix_acl_release(BTRFS_I(inode)->i_acl);
4465         if (BTRFS_I(inode)->i_default_acl &&
4466             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4467                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4468
4469         spin_lock(&BTRFS_I(inode)->root->list_lock);
4470         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4471                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4472                        " list\n", inode->i_ino);
4473                 dump_stack();
4474         }
4475         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4476
4477         while (1) {
4478                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4479                 if (!ordered)
4480                         break;
4481                 else {
4482                         printk(KERN_ERR "btrfs found ordered "
4483                                "extent %llu %llu on inode cleanup\n",
4484                                (unsigned long long)ordered->file_offset,
4485                                (unsigned long long)ordered->len);
4486                         btrfs_remove_ordered_extent(inode, ordered);
4487                         btrfs_put_ordered_extent(ordered);
4488                         btrfs_put_ordered_extent(ordered);
4489                 }
4490         }
4491         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4492         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4493 }
4494
4495 static void init_once(void *foo)
4496 {
4497         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4498
4499         inode_init_once(&ei->vfs_inode);
4500 }
4501
4502 void btrfs_destroy_cachep(void)
4503 {
4504         if (btrfs_inode_cachep)
4505                 kmem_cache_destroy(btrfs_inode_cachep);
4506         if (btrfs_trans_handle_cachep)
4507                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4508         if (btrfs_transaction_cachep)
4509                 kmem_cache_destroy(btrfs_transaction_cachep);
4510         if (btrfs_bit_radix_cachep)
4511                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4512         if (btrfs_path_cachep)
4513                 kmem_cache_destroy(btrfs_path_cachep);
4514 }
4515
4516 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4517                                        unsigned long extra_flags,
4518                                        void (*ctor)(void *))
4519 {
4520         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4521                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4522 }
4523
4524 int btrfs_init_cachep(void)
4525 {
4526         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4527                                           sizeof(struct btrfs_inode),
4528                                           0, init_once);
4529         if (!btrfs_inode_cachep)
4530                 goto fail;
4531         btrfs_trans_handle_cachep =
4532                         btrfs_cache_create("btrfs_trans_handle_cache",
4533                                            sizeof(struct btrfs_trans_handle),
4534                                            0, NULL);
4535         if (!btrfs_trans_handle_cachep)
4536                 goto fail;
4537         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4538                                              sizeof(struct btrfs_transaction),
4539                                              0, NULL);
4540         if (!btrfs_transaction_cachep)
4541                 goto fail;
4542         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4543                                          sizeof(struct btrfs_path),
4544                                          0, NULL);
4545         if (!btrfs_path_cachep)
4546                 goto fail;
4547         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4548                                               SLAB_DESTROY_BY_RCU, NULL);
4549         if (!btrfs_bit_radix_cachep)
4550                 goto fail;
4551         return 0;
4552 fail:
4553         btrfs_destroy_cachep();
4554         return -ENOMEM;
4555 }
4556
4557 static int btrfs_getattr(struct vfsmount *mnt,
4558                          struct dentry *dentry, struct kstat *stat)
4559 {
4560         struct inode *inode = dentry->d_inode;
4561         generic_fillattr(inode, stat);
4562         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4563         stat->blksize = PAGE_CACHE_SIZE;
4564         stat->blocks = (inode_get_bytes(inode) +
4565                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4566         return 0;
4567 }
4568
4569 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4570                            struct inode *new_dir, struct dentry *new_dentry)
4571 {
4572         struct btrfs_trans_handle *trans;
4573         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4574         struct inode *new_inode = new_dentry->d_inode;
4575         struct inode *old_inode = old_dentry->d_inode;
4576         struct timespec ctime = CURRENT_TIME;
4577         u64 index = 0;
4578         int ret;
4579
4580         /* we're not allowed to rename between subvolumes */
4581         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4582             BTRFS_I(new_dir)->root->root_key.objectid)
4583                 return -EXDEV;
4584
4585         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4586             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4587                 return -ENOTEMPTY;
4588         }
4589
4590         /* to rename a snapshot or subvolume, we need to juggle the
4591          * backrefs.  This isn't coded yet
4592          */
4593         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4594                 return -EXDEV;
4595
4596         ret = btrfs_check_free_space(root, 1, 0);
4597         if (ret)
4598                 goto out_unlock;
4599
4600         trans = btrfs_start_transaction(root, 1);
4601
4602         btrfs_set_trans_block_group(trans, new_dir);
4603
4604         btrfs_inc_nlink(old_dentry->d_inode);
4605         old_dir->i_ctime = old_dir->i_mtime = ctime;
4606         new_dir->i_ctime = new_dir->i_mtime = ctime;
4607         old_inode->i_ctime = ctime;
4608
4609         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4610                                  old_dentry->d_name.name,
4611                                  old_dentry->d_name.len);
4612         if (ret)
4613                 goto out_fail;
4614
4615         if (new_inode) {
4616                 new_inode->i_ctime = CURRENT_TIME;
4617                 ret = btrfs_unlink_inode(trans, root, new_dir,
4618                                          new_dentry->d_inode,
4619                                          new_dentry->d_name.name,
4620                                          new_dentry->d_name.len);
4621                 if (ret)
4622                         goto out_fail;
4623                 if (new_inode->i_nlink == 0) {
4624                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4625                         if (ret)
4626                                 goto out_fail;
4627                 }
4628
4629         }
4630         ret = btrfs_set_inode_index(new_dir, &index);
4631         if (ret)
4632                 goto out_fail;
4633
4634         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4635                              old_inode, new_dentry->d_name.name,
4636                              new_dentry->d_name.len, 1, index);
4637         if (ret)
4638                 goto out_fail;
4639
4640 out_fail:
4641         btrfs_end_transaction_throttle(trans, root);
4642 out_unlock:
4643         return ret;
4644 }
4645
4646 /*
4647  * some fairly slow code that needs optimization. This walks the list
4648  * of all the inodes with pending delalloc and forces them to disk.
4649  */
4650 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4651 {
4652         struct list_head *head = &root->fs_info->delalloc_inodes;
4653         struct btrfs_inode *binode;
4654         struct inode *inode;
4655
4656         if (root->fs_info->sb->s_flags & MS_RDONLY)
4657                 return -EROFS;
4658
4659         spin_lock(&root->fs_info->delalloc_lock);
4660         while (!list_empty(head)) {
4661                 binode = list_entry(head->next, struct btrfs_inode,
4662                                     delalloc_inodes);
4663                 inode = igrab(&binode->vfs_inode);
4664                 if (!inode)
4665                         list_del_init(&binode->delalloc_inodes);
4666                 spin_unlock(&root->fs_info->delalloc_lock);
4667                 if (inode) {
4668                         filemap_flush(inode->i_mapping);
4669                         iput(inode);
4670                 }
4671                 cond_resched();
4672                 spin_lock(&root->fs_info->delalloc_lock);
4673         }
4674         spin_unlock(&root->fs_info->delalloc_lock);
4675
4676         /* the filemap_flush will queue IO into the worker threads, but
4677          * we have to make sure the IO is actually started and that
4678          * ordered extents get created before we return
4679          */
4680         atomic_inc(&root->fs_info->async_submit_draining);
4681         while (atomic_read(&root->fs_info->nr_async_submits) ||
4682               atomic_read(&root->fs_info->async_delalloc_pages)) {
4683                 wait_event(root->fs_info->async_submit_wait,
4684                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4685                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4686         }
4687         atomic_dec(&root->fs_info->async_submit_draining);
4688         return 0;
4689 }
4690
4691 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4692                          const char *symname)
4693 {
4694         struct btrfs_trans_handle *trans;
4695         struct btrfs_root *root = BTRFS_I(dir)->root;
4696         struct btrfs_path *path;
4697         struct btrfs_key key;
4698         struct inode *inode = NULL;
4699         int err;
4700         int drop_inode = 0;
4701         u64 objectid;
4702         u64 index = 0 ;
4703         int name_len;
4704         int datasize;
4705         unsigned long ptr;
4706         struct btrfs_file_extent_item *ei;
4707         struct extent_buffer *leaf;
4708         unsigned long nr = 0;
4709
4710         name_len = strlen(symname) + 1;
4711         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4712                 return -ENAMETOOLONG;
4713
4714         err = btrfs_check_free_space(root, 1, 0);
4715         if (err)
4716                 goto out_fail;
4717
4718         trans = btrfs_start_transaction(root, 1);
4719         btrfs_set_trans_block_group(trans, dir);
4720
4721         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4722         if (err) {
4723                 err = -ENOSPC;
4724                 goto out_unlock;
4725         }
4726
4727         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4728                                 dentry->d_name.len,
4729                                 dentry->d_parent->d_inode->i_ino, objectid,
4730                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4731                                 &index);
4732         err = PTR_ERR(inode);
4733         if (IS_ERR(inode))
4734                 goto out_unlock;
4735
4736         err = btrfs_init_acl(inode, dir);
4737         if (err) {
4738                 drop_inode = 1;
4739                 goto out_unlock;
4740         }
4741
4742         btrfs_set_trans_block_group(trans, inode);
4743         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4744         if (err)
4745                 drop_inode = 1;
4746         else {
4747                 inode->i_mapping->a_ops = &btrfs_aops;
4748                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4749                 inode->i_fop = &btrfs_file_operations;
4750                 inode->i_op = &btrfs_file_inode_operations;
4751                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4752         }
4753         dir->i_sb->s_dirt = 1;
4754         btrfs_update_inode_block_group(trans, inode);
4755         btrfs_update_inode_block_group(trans, dir);
4756         if (drop_inode)
4757                 goto out_unlock;
4758
4759         path = btrfs_alloc_path();
4760         BUG_ON(!path);
4761         key.objectid = inode->i_ino;
4762         key.offset = 0;
4763         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4764         datasize = btrfs_file_extent_calc_inline_size(name_len);
4765         err = btrfs_insert_empty_item(trans, root, path, &key,
4766                                       datasize);
4767         if (err) {
4768                 drop_inode = 1;
4769                 goto out_unlock;
4770         }
4771         leaf = path->nodes[0];
4772         ei = btrfs_item_ptr(leaf, path->slots[0],
4773                             struct btrfs_file_extent_item);
4774         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4775         btrfs_set_file_extent_type(leaf, ei,
4776                                    BTRFS_FILE_EXTENT_INLINE);
4777         btrfs_set_file_extent_encryption(leaf, ei, 0);
4778         btrfs_set_file_extent_compression(leaf, ei, 0);
4779         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4780         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4781
4782         ptr = btrfs_file_extent_inline_start(ei);
4783         write_extent_buffer(leaf, symname, ptr, name_len);
4784         btrfs_mark_buffer_dirty(leaf);
4785         btrfs_free_path(path);
4786
4787         inode->i_op = &btrfs_symlink_inode_operations;
4788         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4789         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4790         inode_set_bytes(inode, name_len);
4791         btrfs_i_size_write(inode, name_len - 1);
4792         err = btrfs_update_inode(trans, root, inode);
4793         if (err)
4794                 drop_inode = 1;
4795
4796 out_unlock:
4797         nr = trans->blocks_used;
4798         btrfs_end_transaction_throttle(trans, root);
4799 out_fail:
4800         if (drop_inode) {
4801                 inode_dec_link_count(inode);
4802                 iput(inode);
4803         }
4804         btrfs_btree_balance_dirty(root, nr);
4805         return err;
4806 }
4807
4808 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4809                                u64 alloc_hint, int mode)
4810 {
4811         struct btrfs_trans_handle *trans;
4812         struct btrfs_root *root = BTRFS_I(inode)->root;
4813         struct btrfs_key ins;
4814         u64 alloc_size;
4815         u64 cur_offset = start;
4816         u64 num_bytes = end - start;
4817         int ret = 0;
4818
4819         trans = btrfs_join_transaction(root, 1);
4820         BUG_ON(!trans);
4821         btrfs_set_trans_block_group(trans, inode);
4822
4823         while (num_bytes > 0) {
4824                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4825                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4826                                            root->sectorsize, 0, alloc_hint,
4827                                            (u64)-1, &ins, 1);
4828                 if (ret) {
4829                         WARN_ON(1);
4830                         goto out;
4831                 }
4832                 ret = insert_reserved_file_extent(trans, inode,
4833                                                   cur_offset, ins.objectid,
4834                                                   ins.offset, ins.offset,
4835                                                   ins.offset, 0, 0, 0,
4836                                                   BTRFS_FILE_EXTENT_PREALLOC);
4837                 BUG_ON(ret);
4838                 num_bytes -= ins.offset;
4839                 cur_offset += ins.offset;
4840                 alloc_hint = ins.objectid + ins.offset;
4841         }
4842 out:
4843         if (cur_offset > start) {
4844                 inode->i_ctime = CURRENT_TIME;
4845                 btrfs_set_flag(inode, PREALLOC);
4846                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4847                     cur_offset > i_size_read(inode))
4848                         btrfs_i_size_write(inode, cur_offset);
4849                 ret = btrfs_update_inode(trans, root, inode);
4850                 BUG_ON(ret);
4851         }
4852
4853         btrfs_end_transaction(trans, root);
4854         return ret;
4855 }
4856
4857 static long btrfs_fallocate(struct inode *inode, int mode,
4858                             loff_t offset, loff_t len)
4859 {
4860         u64 cur_offset;
4861         u64 last_byte;
4862         u64 alloc_start;
4863         u64 alloc_end;
4864         u64 alloc_hint = 0;
4865         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4866         struct extent_map *em;
4867         int ret;
4868
4869         alloc_start = offset & ~mask;
4870         alloc_end =  (offset + len + mask) & ~mask;
4871
4872         mutex_lock(&inode->i_mutex);
4873         if (alloc_start > inode->i_size) {
4874                 ret = btrfs_cont_expand(inode, alloc_start);
4875                 if (ret)
4876                         goto out;
4877         }
4878
4879         while (1) {
4880                 struct btrfs_ordered_extent *ordered;
4881                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4882                             alloc_end - 1, GFP_NOFS);
4883                 ordered = btrfs_lookup_first_ordered_extent(inode,
4884                                                             alloc_end - 1);
4885                 if (ordered &&
4886                     ordered->file_offset + ordered->len > alloc_start &&
4887                     ordered->file_offset < alloc_end) {
4888                         btrfs_put_ordered_extent(ordered);
4889                         unlock_extent(&BTRFS_I(inode)->io_tree,
4890                                       alloc_start, alloc_end - 1, GFP_NOFS);
4891                         btrfs_wait_ordered_range(inode, alloc_start,
4892                                                  alloc_end - alloc_start);
4893                 } else {
4894                         if (ordered)
4895                                 btrfs_put_ordered_extent(ordered);
4896                         break;
4897                 }
4898         }
4899
4900         cur_offset = alloc_start;
4901         while (1) {
4902                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4903                                       alloc_end - cur_offset, 0);
4904                 BUG_ON(IS_ERR(em) || !em);
4905                 last_byte = min(extent_map_end(em), alloc_end);
4906                 last_byte = (last_byte + mask) & ~mask;
4907                 if (em->block_start == EXTENT_MAP_HOLE) {
4908                         ret = prealloc_file_range(inode, cur_offset,
4909                                         last_byte, alloc_hint, mode);
4910                         if (ret < 0) {
4911                                 free_extent_map(em);
4912                                 break;
4913                         }
4914                 }
4915                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4916                         alloc_hint = em->block_start;
4917                 free_extent_map(em);
4918
4919                 cur_offset = last_byte;
4920                 if (cur_offset >= alloc_end) {
4921                         ret = 0;
4922                         break;
4923                 }
4924         }
4925         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4926                       GFP_NOFS);
4927 out:
4928         mutex_unlock(&inode->i_mutex);
4929         return ret;
4930 }
4931
4932 static int btrfs_set_page_dirty(struct page *page)
4933 {
4934         return __set_page_dirty_nobuffers(page);
4935 }
4936
4937 static int btrfs_permission(struct inode *inode, int mask)
4938 {
4939         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4940                 return -EACCES;
4941         return generic_permission(inode, mask, btrfs_check_acl);
4942 }
4943
4944 static struct inode_operations btrfs_dir_inode_operations = {
4945         .getattr        = btrfs_getattr,
4946         .lookup         = btrfs_lookup,
4947         .create         = btrfs_create,
4948         .unlink         = btrfs_unlink,
4949         .link           = btrfs_link,
4950         .mkdir          = btrfs_mkdir,
4951         .rmdir          = btrfs_rmdir,
4952         .rename         = btrfs_rename,
4953         .symlink        = btrfs_symlink,
4954         .setattr        = btrfs_setattr,
4955         .mknod          = btrfs_mknod,
4956         .setxattr       = btrfs_setxattr,
4957         .getxattr       = btrfs_getxattr,
4958         .listxattr      = btrfs_listxattr,
4959         .removexattr    = btrfs_removexattr,
4960         .permission     = btrfs_permission,
4961 };
4962 static struct inode_operations btrfs_dir_ro_inode_operations = {
4963         .lookup         = btrfs_lookup,
4964         .permission     = btrfs_permission,
4965 };
4966 static struct file_operations btrfs_dir_file_operations = {
4967         .llseek         = generic_file_llseek,
4968         .read           = generic_read_dir,
4969         .readdir        = btrfs_real_readdir,
4970         .unlocked_ioctl = btrfs_ioctl,
4971 #ifdef CONFIG_COMPAT
4972         .compat_ioctl   = btrfs_ioctl,
4973 #endif
4974         .release        = btrfs_release_file,
4975         .fsync          = btrfs_sync_file,
4976 };
4977
4978 static struct extent_io_ops btrfs_extent_io_ops = {
4979         .fill_delalloc = run_delalloc_range,
4980         .submit_bio_hook = btrfs_submit_bio_hook,
4981         .merge_bio_hook = btrfs_merge_bio_hook,
4982         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4983         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4984         .writepage_start_hook = btrfs_writepage_start_hook,
4985         .readpage_io_failed_hook = btrfs_io_failed_hook,
4986         .set_bit_hook = btrfs_set_bit_hook,
4987         .clear_bit_hook = btrfs_clear_bit_hook,
4988 };
4989
4990 static struct address_space_operations btrfs_aops = {
4991         .readpage       = btrfs_readpage,
4992         .writepage      = btrfs_writepage,
4993         .writepages     = btrfs_writepages,
4994         .readpages      = btrfs_readpages,
4995         .sync_page      = block_sync_page,
4996         .bmap           = btrfs_bmap,
4997         .direct_IO      = btrfs_direct_IO,
4998         .invalidatepage = btrfs_invalidatepage,
4999         .releasepage    = btrfs_releasepage,
5000         .set_page_dirty = btrfs_set_page_dirty,
5001 };
5002
5003 static struct address_space_operations btrfs_symlink_aops = {
5004         .readpage       = btrfs_readpage,
5005         .writepage      = btrfs_writepage,
5006         .invalidatepage = btrfs_invalidatepage,
5007         .releasepage    = btrfs_releasepage,
5008 };
5009
5010 static struct inode_operations btrfs_file_inode_operations = {
5011         .truncate       = btrfs_truncate,
5012         .getattr        = btrfs_getattr,
5013         .setattr        = btrfs_setattr,
5014         .setxattr       = btrfs_setxattr,
5015         .getxattr       = btrfs_getxattr,
5016         .listxattr      = btrfs_listxattr,
5017         .removexattr    = btrfs_removexattr,
5018         .permission     = btrfs_permission,
5019         .fallocate      = btrfs_fallocate,
5020 };
5021 static struct inode_operations btrfs_special_inode_operations = {
5022         .getattr        = btrfs_getattr,
5023         .setattr        = btrfs_setattr,
5024         .permission     = btrfs_permission,
5025         .setxattr       = btrfs_setxattr,
5026         .getxattr       = btrfs_getxattr,
5027         .listxattr      = btrfs_listxattr,
5028         .removexattr    = btrfs_removexattr,
5029 };
5030 static struct inode_operations btrfs_symlink_inode_operations = {
5031         .readlink       = generic_readlink,
5032         .follow_link    = page_follow_link_light,
5033         .put_link       = page_put_link,
5034         .permission     = btrfs_permission,
5035 };