Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[sfrench/cifs-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "ref-cache.h"
52 #include "compression.h"
53 #include "locking.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90                                    struct page *locked_page,
91                                    u64 start, u64 end, int *page_started,
92                                    unsigned long *nr_written, int unlock);
93
94 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
95 {
96         int err;
97
98         err = btrfs_init_acl(inode, dir);
99         if (!err)
100                 err = btrfs_xattr_security_init(inode, dir);
101         return err;
102 }
103
104 /*
105  * this does all the hard work for inserting an inline extent into
106  * the btree.  The caller should have done a btrfs_drop_extents so that
107  * no overlapping inline items exist in the btree
108  */
109 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
110                                 struct btrfs_root *root, struct inode *inode,
111                                 u64 start, size_t size, size_t compressed_size,
112                                 struct page **compressed_pages)
113 {
114         struct btrfs_key key;
115         struct btrfs_path *path;
116         struct extent_buffer *leaf;
117         struct page *page = NULL;
118         char *kaddr;
119         unsigned long ptr;
120         struct btrfs_file_extent_item *ei;
121         int err = 0;
122         int ret;
123         size_t cur_size = size;
124         size_t datasize;
125         unsigned long offset;
126         int use_compress = 0;
127
128         if (compressed_size && compressed_pages) {
129                 use_compress = 1;
130                 cur_size = compressed_size;
131         }
132
133         path = btrfs_alloc_path();
134         if (!path)
135                 return -ENOMEM;
136
137         path->leave_spinning = 1;
138         btrfs_set_trans_block_group(trans, inode);
139
140         key.objectid = inode->i_ino;
141         key.offset = start;
142         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143         datasize = btrfs_file_extent_calc_inline_size(cur_size);
144
145         inode_add_bytes(inode, size);
146         ret = btrfs_insert_empty_item(trans, root, path, &key,
147                                       datasize);
148         BUG_ON(ret);
149         if (ret) {
150                 err = ret;
151                 goto fail;
152         }
153         leaf = path->nodes[0];
154         ei = btrfs_item_ptr(leaf, path->slots[0],
155                             struct btrfs_file_extent_item);
156         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158         btrfs_set_file_extent_encryption(leaf, ei, 0);
159         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161         ptr = btrfs_file_extent_inline_start(ei);
162
163         if (use_compress) {
164                 struct page *cpage;
165                 int i = 0;
166                 while (compressed_size > 0) {
167                         cpage = compressed_pages[i];
168                         cur_size = min_t(unsigned long, compressed_size,
169                                        PAGE_CACHE_SIZE);
170
171                         kaddr = kmap_atomic(cpage, KM_USER0);
172                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
173                         kunmap_atomic(kaddr, KM_USER0);
174
175                         i++;
176                         ptr += cur_size;
177                         compressed_size -= cur_size;
178                 }
179                 btrfs_set_file_extent_compression(leaf, ei,
180                                                   BTRFS_COMPRESS_ZLIB);
181         } else {
182                 page = find_get_page(inode->i_mapping,
183                                      start >> PAGE_CACHE_SHIFT);
184                 btrfs_set_file_extent_compression(leaf, ei, 0);
185                 kaddr = kmap_atomic(page, KM_USER0);
186                 offset = start & (PAGE_CACHE_SIZE - 1);
187                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188                 kunmap_atomic(kaddr, KM_USER0);
189                 page_cache_release(page);
190         }
191         btrfs_mark_buffer_dirty(leaf);
192         btrfs_free_path(path);
193
194         BTRFS_I(inode)->disk_i_size = inode->i_size;
195         btrfs_update_inode(trans, root, inode);
196         return 0;
197 fail:
198         btrfs_free_path(path);
199         return err;
200 }
201
202
203 /*
204  * conditionally insert an inline extent into the file.  This
205  * does the checks required to make sure the data is small enough
206  * to fit as an inline extent.
207  */
208 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
209                                  struct btrfs_root *root,
210                                  struct inode *inode, u64 start, u64 end,
211                                  size_t compressed_size,
212                                  struct page **compressed_pages)
213 {
214         u64 isize = i_size_read(inode);
215         u64 actual_end = min(end + 1, isize);
216         u64 inline_len = actual_end - start;
217         u64 aligned_end = (end + root->sectorsize - 1) &
218                         ~((u64)root->sectorsize - 1);
219         u64 hint_byte;
220         u64 data_len = inline_len;
221         int ret;
222
223         if (compressed_size)
224                 data_len = compressed_size;
225
226         if (start > 0 ||
227             actual_end >= PAGE_CACHE_SIZE ||
228             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
229             (!compressed_size &&
230             (actual_end & (root->sectorsize - 1)) == 0) ||
231             end + 1 < isize ||
232             data_len > root->fs_info->max_inline) {
233                 return 1;
234         }
235
236         ret = btrfs_drop_extents(trans, root, inode, start,
237                                  aligned_end, start, &hint_byte);
238         BUG_ON(ret);
239
240         if (isize > actual_end)
241                 inline_len = min_t(u64, isize, actual_end);
242         ret = insert_inline_extent(trans, root, inode, start,
243                                    inline_len, compressed_size,
244                                    compressed_pages);
245         BUG_ON(ret);
246         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
247         return 0;
248 }
249
250 struct async_extent {
251         u64 start;
252         u64 ram_size;
253         u64 compressed_size;
254         struct page **pages;
255         unsigned long nr_pages;
256         struct list_head list;
257 };
258
259 struct async_cow {
260         struct inode *inode;
261         struct btrfs_root *root;
262         struct page *locked_page;
263         u64 start;
264         u64 end;
265         struct list_head extents;
266         struct btrfs_work work;
267 };
268
269 static noinline int add_async_extent(struct async_cow *cow,
270                                      u64 start, u64 ram_size,
271                                      u64 compressed_size,
272                                      struct page **pages,
273                                      unsigned long nr_pages)
274 {
275         struct async_extent *async_extent;
276
277         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
278         async_extent->start = start;
279         async_extent->ram_size = ram_size;
280         async_extent->compressed_size = compressed_size;
281         async_extent->pages = pages;
282         async_extent->nr_pages = nr_pages;
283         list_add_tail(&async_extent->list, &cow->extents);
284         return 0;
285 }
286
287 /*
288  * we create compressed extents in two phases.  The first
289  * phase compresses a range of pages that have already been
290  * locked (both pages and state bits are locked).
291  *
292  * This is done inside an ordered work queue, and the compression
293  * is spread across many cpus.  The actual IO submission is step
294  * two, and the ordered work queue takes care of making sure that
295  * happens in the same order things were put onto the queue by
296  * writepages and friends.
297  *
298  * If this code finds it can't get good compression, it puts an
299  * entry onto the work queue to write the uncompressed bytes.  This
300  * makes sure that both compressed inodes and uncompressed inodes
301  * are written in the same order that pdflush sent them down.
302  */
303 static noinline int compress_file_range(struct inode *inode,
304                                         struct page *locked_page,
305                                         u64 start, u64 end,
306                                         struct async_cow *async_cow,
307                                         int *num_added)
308 {
309         struct btrfs_root *root = BTRFS_I(inode)->root;
310         struct btrfs_trans_handle *trans;
311         u64 num_bytes;
312         u64 orig_start;
313         u64 disk_num_bytes;
314         u64 blocksize = root->sectorsize;
315         u64 actual_end;
316         u64 isize = i_size_read(inode);
317         int ret = 0;
318         struct page **pages = NULL;
319         unsigned long nr_pages;
320         unsigned long nr_pages_ret = 0;
321         unsigned long total_compressed = 0;
322         unsigned long total_in = 0;
323         unsigned long max_compressed = 128 * 1024;
324         unsigned long max_uncompressed = 128 * 1024;
325         int i;
326         int will_compress;
327
328         orig_start = start;
329
330         actual_end = min_t(u64, isize, end + 1);
331 again:
332         will_compress = 0;
333         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
334         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
335
336         /*
337          * we don't want to send crud past the end of i_size through
338          * compression, that's just a waste of CPU time.  So, if the
339          * end of the file is before the start of our current
340          * requested range of bytes, we bail out to the uncompressed
341          * cleanup code that can deal with all of this.
342          *
343          * It isn't really the fastest way to fix things, but this is a
344          * very uncommon corner.
345          */
346         if (actual_end <= start)
347                 goto cleanup_and_bail_uncompressed;
348
349         total_compressed = actual_end - start;
350
351         /* we want to make sure that amount of ram required to uncompress
352          * an extent is reasonable, so we limit the total size in ram
353          * of a compressed extent to 128k.  This is a crucial number
354          * because it also controls how easily we can spread reads across
355          * cpus for decompression.
356          *
357          * We also want to make sure the amount of IO required to do
358          * a random read is reasonably small, so we limit the size of
359          * a compressed extent to 128k.
360          */
361         total_compressed = min(total_compressed, max_uncompressed);
362         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
363         num_bytes = max(blocksize,  num_bytes);
364         disk_num_bytes = num_bytes;
365         total_in = 0;
366         ret = 0;
367
368         /*
369          * we do compression for mount -o compress and when the
370          * inode has not been flagged as nocompress.  This flag can
371          * change at any time if we discover bad compression ratios.
372          */
373         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
374             btrfs_test_opt(root, COMPRESS)) {
375                 WARN_ON(pages);
376                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
377
378                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
379                                                 total_compressed, pages,
380                                                 nr_pages, &nr_pages_ret,
381                                                 &total_in,
382                                                 &total_compressed,
383                                                 max_compressed);
384
385                 if (!ret) {
386                         unsigned long offset = total_compressed &
387                                 (PAGE_CACHE_SIZE - 1);
388                         struct page *page = pages[nr_pages_ret - 1];
389                         char *kaddr;
390
391                         /* zero the tail end of the last page, we might be
392                          * sending it down to disk
393                          */
394                         if (offset) {
395                                 kaddr = kmap_atomic(page, KM_USER0);
396                                 memset(kaddr + offset, 0,
397                                        PAGE_CACHE_SIZE - offset);
398                                 kunmap_atomic(kaddr, KM_USER0);
399                         }
400                         will_compress = 1;
401                 }
402         }
403         if (start == 0) {
404                 trans = btrfs_join_transaction(root, 1);
405                 BUG_ON(!trans);
406                 btrfs_set_trans_block_group(trans, inode);
407
408                 /* lets try to make an inline extent */
409                 if (ret || total_in < (actual_end - start)) {
410                         /* we didn't compress the entire range, try
411                          * to make an uncompressed inline extent.
412                          */
413                         ret = cow_file_range_inline(trans, root, inode,
414                                                     start, end, 0, NULL);
415                 } else {
416                         /* try making a compressed inline extent */
417                         ret = cow_file_range_inline(trans, root, inode,
418                                                     start, end,
419                                                     total_compressed, pages);
420                 }
421                 btrfs_end_transaction(trans, root);
422                 if (ret == 0) {
423                         /*
424                          * inline extent creation worked, we don't need
425                          * to create any more async work items.  Unlock
426                          * and free up our temp pages.
427                          */
428                         extent_clear_unlock_delalloc(inode,
429                                                      &BTRFS_I(inode)->io_tree,
430                                                      start, end, NULL, 1, 0,
431                                                      0, 1, 1, 1);
432                         ret = 0;
433                         goto free_pages_out;
434                 }
435         }
436
437         if (will_compress) {
438                 /*
439                  * we aren't doing an inline extent round the compressed size
440                  * up to a block size boundary so the allocator does sane
441                  * things
442                  */
443                 total_compressed = (total_compressed + blocksize - 1) &
444                         ~(blocksize - 1);
445
446                 /*
447                  * one last check to make sure the compression is really a
448                  * win, compare the page count read with the blocks on disk
449                  */
450                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
451                         ~(PAGE_CACHE_SIZE - 1);
452                 if (total_compressed >= total_in) {
453                         will_compress = 0;
454                 } else {
455                         disk_num_bytes = total_compressed;
456                         num_bytes = total_in;
457                 }
458         }
459         if (!will_compress && pages) {
460                 /*
461                  * the compression code ran but failed to make things smaller,
462                  * free any pages it allocated and our page pointer array
463                  */
464                 for (i = 0; i < nr_pages_ret; i++) {
465                         WARN_ON(pages[i]->mapping);
466                         page_cache_release(pages[i]);
467                 }
468                 kfree(pages);
469                 pages = NULL;
470                 total_compressed = 0;
471                 nr_pages_ret = 0;
472
473                 /* flag the file so we don't compress in the future */
474                 btrfs_set_flag(inode, NOCOMPRESS);
475         }
476         if (will_compress) {
477                 *num_added += 1;
478
479                 /* the async work queues will take care of doing actual
480                  * allocation on disk for these compressed pages,
481                  * and will submit them to the elevator.
482                  */
483                 add_async_extent(async_cow, start, num_bytes,
484                                  total_compressed, pages, nr_pages_ret);
485
486                 if (start + num_bytes < end && start + num_bytes < actual_end) {
487                         start += num_bytes;
488                         pages = NULL;
489                         cond_resched();
490                         goto again;
491                 }
492         } else {
493 cleanup_and_bail_uncompressed:
494                 /*
495                  * No compression, but we still need to write the pages in
496                  * the file we've been given so far.  redirty the locked
497                  * page if it corresponds to our extent and set things up
498                  * for the async work queue to run cow_file_range to do
499                  * the normal delalloc dance
500                  */
501                 if (page_offset(locked_page) >= start &&
502                     page_offset(locked_page) <= end) {
503                         __set_page_dirty_nobuffers(locked_page);
504                         /* unlocked later on in the async handlers */
505                 }
506                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
507                 *num_added += 1;
508         }
509
510 out:
511         return 0;
512
513 free_pages_out:
514         for (i = 0; i < nr_pages_ret; i++) {
515                 WARN_ON(pages[i]->mapping);
516                 page_cache_release(pages[i]);
517         }
518         kfree(pages);
519
520         goto out;
521 }
522
523 /*
524  * phase two of compressed writeback.  This is the ordered portion
525  * of the code, which only gets called in the order the work was
526  * queued.  We walk all the async extents created by compress_file_range
527  * and send them down to the disk.
528  */
529 static noinline int submit_compressed_extents(struct inode *inode,
530                                               struct async_cow *async_cow)
531 {
532         struct async_extent *async_extent;
533         u64 alloc_hint = 0;
534         struct btrfs_trans_handle *trans;
535         struct btrfs_key ins;
536         struct extent_map *em;
537         struct btrfs_root *root = BTRFS_I(inode)->root;
538         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
539         struct extent_io_tree *io_tree;
540         int ret;
541
542         if (list_empty(&async_cow->extents))
543                 return 0;
544
545         trans = btrfs_join_transaction(root, 1);
546
547         while (!list_empty(&async_cow->extents)) {
548                 async_extent = list_entry(async_cow->extents.next,
549                                           struct async_extent, list);
550                 list_del(&async_extent->list);
551
552                 io_tree = &BTRFS_I(inode)->io_tree;
553
554                 /* did the compression code fall back to uncompressed IO? */
555                 if (!async_extent->pages) {
556                         int page_started = 0;
557                         unsigned long nr_written = 0;
558
559                         lock_extent(io_tree, async_extent->start,
560                                     async_extent->start +
561                                     async_extent->ram_size - 1, GFP_NOFS);
562
563                         /* allocate blocks */
564                         cow_file_range(inode, async_cow->locked_page,
565                                        async_extent->start,
566                                        async_extent->start +
567                                        async_extent->ram_size - 1,
568                                        &page_started, &nr_written, 0);
569
570                         /*
571                          * if page_started, cow_file_range inserted an
572                          * inline extent and took care of all the unlocking
573                          * and IO for us.  Otherwise, we need to submit
574                          * all those pages down to the drive.
575                          */
576                         if (!page_started)
577                                 extent_write_locked_range(io_tree,
578                                                   inode, async_extent->start,
579                                                   async_extent->start +
580                                                   async_extent->ram_size - 1,
581                                                   btrfs_get_extent,
582                                                   WB_SYNC_ALL);
583                         kfree(async_extent);
584                         cond_resched();
585                         continue;
586                 }
587
588                 lock_extent(io_tree, async_extent->start,
589                             async_extent->start + async_extent->ram_size - 1,
590                             GFP_NOFS);
591                 /*
592                  * here we're doing allocation and writeback of the
593                  * compressed pages
594                  */
595                 btrfs_drop_extent_cache(inode, async_extent->start,
596                                         async_extent->start +
597                                         async_extent->ram_size - 1, 0);
598
599                 ret = btrfs_reserve_extent(trans, root,
600                                            async_extent->compressed_size,
601                                            async_extent->compressed_size,
602                                            0, alloc_hint,
603                                            (u64)-1, &ins, 1);
604                 BUG_ON(ret);
605                 em = alloc_extent_map(GFP_NOFS);
606                 em->start = async_extent->start;
607                 em->len = async_extent->ram_size;
608                 em->orig_start = em->start;
609
610                 em->block_start = ins.objectid;
611                 em->block_len = ins.offset;
612                 em->bdev = root->fs_info->fs_devices->latest_bdev;
613                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
614                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
615
616                 while (1) {
617                         spin_lock(&em_tree->lock);
618                         ret = add_extent_mapping(em_tree, em);
619                         spin_unlock(&em_tree->lock);
620                         if (ret != -EEXIST) {
621                                 free_extent_map(em);
622                                 break;
623                         }
624                         btrfs_drop_extent_cache(inode, async_extent->start,
625                                                 async_extent->start +
626                                                 async_extent->ram_size - 1, 0);
627                 }
628
629                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
630                                                ins.objectid,
631                                                async_extent->ram_size,
632                                                ins.offset,
633                                                BTRFS_ORDERED_COMPRESSED);
634                 BUG_ON(ret);
635
636                 btrfs_end_transaction(trans, root);
637
638                 /*
639                  * clear dirty, set writeback and unlock the pages.
640                  */
641                 extent_clear_unlock_delalloc(inode,
642                                              &BTRFS_I(inode)->io_tree,
643                                              async_extent->start,
644                                              async_extent->start +
645                                              async_extent->ram_size - 1,
646                                              NULL, 1, 1, 0, 1, 1, 0);
647
648                 ret = btrfs_submit_compressed_write(inode,
649                                     async_extent->start,
650                                     async_extent->ram_size,
651                                     ins.objectid,
652                                     ins.offset, async_extent->pages,
653                                     async_extent->nr_pages);
654
655                 BUG_ON(ret);
656                 trans = btrfs_join_transaction(root, 1);
657                 alloc_hint = ins.objectid + ins.offset;
658                 kfree(async_extent);
659                 cond_resched();
660         }
661
662         btrfs_end_transaction(trans, root);
663         return 0;
664 }
665
666 /*
667  * when extent_io.c finds a delayed allocation range in the file,
668  * the call backs end up in this code.  The basic idea is to
669  * allocate extents on disk for the range, and create ordered data structs
670  * in ram to track those extents.
671  *
672  * locked_page is the page that writepage had locked already.  We use
673  * it to make sure we don't do extra locks or unlocks.
674  *
675  * *page_started is set to one if we unlock locked_page and do everything
676  * required to start IO on it.  It may be clean and already done with
677  * IO when we return.
678  */
679 static noinline int cow_file_range(struct inode *inode,
680                                    struct page *locked_page,
681                                    u64 start, u64 end, int *page_started,
682                                    unsigned long *nr_written,
683                                    int unlock)
684 {
685         struct btrfs_root *root = BTRFS_I(inode)->root;
686         struct btrfs_trans_handle *trans;
687         u64 alloc_hint = 0;
688         u64 num_bytes;
689         unsigned long ram_size;
690         u64 disk_num_bytes;
691         u64 cur_alloc_size;
692         u64 blocksize = root->sectorsize;
693         u64 actual_end;
694         u64 isize = i_size_read(inode);
695         struct btrfs_key ins;
696         struct extent_map *em;
697         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
698         int ret = 0;
699
700         trans = btrfs_join_transaction(root, 1);
701         BUG_ON(!trans);
702         btrfs_set_trans_block_group(trans, inode);
703
704         actual_end = min_t(u64, isize, end + 1);
705
706         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
707         num_bytes = max(blocksize,  num_bytes);
708         disk_num_bytes = num_bytes;
709         ret = 0;
710
711         if (start == 0) {
712                 /* lets try to make an inline extent */
713                 ret = cow_file_range_inline(trans, root, inode,
714                                             start, end, 0, NULL);
715                 if (ret == 0) {
716                         extent_clear_unlock_delalloc(inode,
717                                                      &BTRFS_I(inode)->io_tree,
718                                                      start, end, NULL, 1, 1,
719                                                      1, 1, 1, 1);
720                         *nr_written = *nr_written +
721                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
722                         *page_started = 1;
723                         ret = 0;
724                         goto out;
725                 }
726         }
727
728         BUG_ON(disk_num_bytes >
729                btrfs_super_total_bytes(&root->fs_info->super_copy));
730
731         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
732
733         while (disk_num_bytes > 0) {
734                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
735                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
736                                            root->sectorsize, 0, alloc_hint,
737                                            (u64)-1, &ins, 1);
738                 BUG_ON(ret);
739
740                 em = alloc_extent_map(GFP_NOFS);
741                 em->start = start;
742                 em->orig_start = em->start;
743
744                 ram_size = ins.offset;
745                 em->len = ins.offset;
746
747                 em->block_start = ins.objectid;
748                 em->block_len = ins.offset;
749                 em->bdev = root->fs_info->fs_devices->latest_bdev;
750                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
751
752                 while (1) {
753                         spin_lock(&em_tree->lock);
754                         ret = add_extent_mapping(em_tree, em);
755                         spin_unlock(&em_tree->lock);
756                         if (ret != -EEXIST) {
757                                 free_extent_map(em);
758                                 break;
759                         }
760                         btrfs_drop_extent_cache(inode, start,
761                                                 start + ram_size - 1, 0);
762                 }
763
764                 cur_alloc_size = ins.offset;
765                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
766                                                ram_size, cur_alloc_size, 0);
767                 BUG_ON(ret);
768
769                 if (root->root_key.objectid ==
770                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
771                         ret = btrfs_reloc_clone_csums(inode, start,
772                                                       cur_alloc_size);
773                         BUG_ON(ret);
774                 }
775
776                 if (disk_num_bytes < cur_alloc_size)
777                         break;
778
779                 /* we're not doing compressed IO, don't unlock the first
780                  * page (which the caller expects to stay locked), don't
781                  * clear any dirty bits and don't set any writeback bits
782                  */
783                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784                                              start, start + ram_size - 1,
785                                              locked_page, unlock, 1,
786                                              1, 0, 0, 0);
787                 disk_num_bytes -= cur_alloc_size;
788                 num_bytes -= cur_alloc_size;
789                 alloc_hint = ins.objectid + ins.offset;
790                 start += cur_alloc_size;
791         }
792 out:
793         ret = 0;
794         btrfs_end_transaction(trans, root);
795
796         return ret;
797 }
798
799 /*
800  * work queue call back to started compression on a file and pages
801  */
802 static noinline void async_cow_start(struct btrfs_work *work)
803 {
804         struct async_cow *async_cow;
805         int num_added = 0;
806         async_cow = container_of(work, struct async_cow, work);
807
808         compress_file_range(async_cow->inode, async_cow->locked_page,
809                             async_cow->start, async_cow->end, async_cow,
810                             &num_added);
811         if (num_added == 0)
812                 async_cow->inode = NULL;
813 }
814
815 /*
816  * work queue call back to submit previously compressed pages
817  */
818 static noinline void async_cow_submit(struct btrfs_work *work)
819 {
820         struct async_cow *async_cow;
821         struct btrfs_root *root;
822         unsigned long nr_pages;
823
824         async_cow = container_of(work, struct async_cow, work);
825
826         root = async_cow->root;
827         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828                 PAGE_CACHE_SHIFT;
829
830         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
831
832         if (atomic_read(&root->fs_info->async_delalloc_pages) <
833             5 * 1042 * 1024 &&
834             waitqueue_active(&root->fs_info->async_submit_wait))
835                 wake_up(&root->fs_info->async_submit_wait);
836
837         if (async_cow->inode)
838                 submit_compressed_extents(async_cow->inode, async_cow);
839 }
840
841 static noinline void async_cow_free(struct btrfs_work *work)
842 {
843         struct async_cow *async_cow;
844         async_cow = container_of(work, struct async_cow, work);
845         kfree(async_cow);
846 }
847
848 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
849                                 u64 start, u64 end, int *page_started,
850                                 unsigned long *nr_written)
851 {
852         struct async_cow *async_cow;
853         struct btrfs_root *root = BTRFS_I(inode)->root;
854         unsigned long nr_pages;
855         u64 cur_end;
856         int limit = 10 * 1024 * 1042;
857
858         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
859                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
860         while (start < end) {
861                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
862                 async_cow->inode = inode;
863                 async_cow->root = root;
864                 async_cow->locked_page = locked_page;
865                 async_cow->start = start;
866
867                 if (btrfs_test_flag(inode, NOCOMPRESS))
868                         cur_end = end;
869                 else
870                         cur_end = min(end, start + 512 * 1024 - 1);
871
872                 async_cow->end = cur_end;
873                 INIT_LIST_HEAD(&async_cow->extents);
874
875                 async_cow->work.func = async_cow_start;
876                 async_cow->work.ordered_func = async_cow_submit;
877                 async_cow->work.ordered_free = async_cow_free;
878                 async_cow->work.flags = 0;
879
880                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
881                         PAGE_CACHE_SHIFT;
882                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
883
884                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
885                                    &async_cow->work);
886
887                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
888                         wait_event(root->fs_info->async_submit_wait,
889                            (atomic_read(&root->fs_info->async_delalloc_pages) <
890                             limit));
891                 }
892
893                 while (atomic_read(&root->fs_info->async_submit_draining) &&
894                       atomic_read(&root->fs_info->async_delalloc_pages)) {
895                         wait_event(root->fs_info->async_submit_wait,
896                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
897                            0));
898                 }
899
900                 *nr_written += nr_pages;
901                 start = cur_end + 1;
902         }
903         *page_started = 1;
904         return 0;
905 }
906
907 static noinline int csum_exist_in_range(struct btrfs_root *root,
908                                         u64 bytenr, u64 num_bytes)
909 {
910         int ret;
911         struct btrfs_ordered_sum *sums;
912         LIST_HEAD(list);
913
914         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
915                                        bytenr + num_bytes - 1, &list);
916         if (ret == 0 && list_empty(&list))
917                 return 0;
918
919         while (!list_empty(&list)) {
920                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
921                 list_del(&sums->list);
922                 kfree(sums);
923         }
924         return 1;
925 }
926
927 /*
928  * when nowcow writeback call back.  This checks for snapshots or COW copies
929  * of the extents that exist in the file, and COWs the file as required.
930  *
931  * If no cow copies or snapshots exist, we write directly to the existing
932  * blocks on disk
933  */
934 static noinline int run_delalloc_nocow(struct inode *inode,
935                                        struct page *locked_page,
936                               u64 start, u64 end, int *page_started, int force,
937                               unsigned long *nr_written)
938 {
939         struct btrfs_root *root = BTRFS_I(inode)->root;
940         struct btrfs_trans_handle *trans;
941         struct extent_buffer *leaf;
942         struct btrfs_path *path;
943         struct btrfs_file_extent_item *fi;
944         struct btrfs_key found_key;
945         u64 cow_start;
946         u64 cur_offset;
947         u64 extent_end;
948         u64 disk_bytenr;
949         u64 num_bytes;
950         int extent_type;
951         int ret;
952         int type;
953         int nocow;
954         int check_prev = 1;
955
956         path = btrfs_alloc_path();
957         BUG_ON(!path);
958         trans = btrfs_join_transaction(root, 1);
959         BUG_ON(!trans);
960
961         cow_start = (u64)-1;
962         cur_offset = start;
963         while (1) {
964                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
965                                                cur_offset, 0);
966                 BUG_ON(ret < 0);
967                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
968                         leaf = path->nodes[0];
969                         btrfs_item_key_to_cpu(leaf, &found_key,
970                                               path->slots[0] - 1);
971                         if (found_key.objectid == inode->i_ino &&
972                             found_key.type == BTRFS_EXTENT_DATA_KEY)
973                                 path->slots[0]--;
974                 }
975                 check_prev = 0;
976 next_slot:
977                 leaf = path->nodes[0];
978                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
979                         ret = btrfs_next_leaf(root, path);
980                         if (ret < 0)
981                                 BUG_ON(1);
982                         if (ret > 0)
983                                 break;
984                         leaf = path->nodes[0];
985                 }
986
987                 nocow = 0;
988                 disk_bytenr = 0;
989                 num_bytes = 0;
990                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
991
992                 if (found_key.objectid > inode->i_ino ||
993                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
994                     found_key.offset > end)
995                         break;
996
997                 if (found_key.offset > cur_offset) {
998                         extent_end = found_key.offset;
999                         goto out_check;
1000                 }
1001
1002                 fi = btrfs_item_ptr(leaf, path->slots[0],
1003                                     struct btrfs_file_extent_item);
1004                 extent_type = btrfs_file_extent_type(leaf, fi);
1005
1006                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1007                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1008                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1009                         extent_end = found_key.offset +
1010                                 btrfs_file_extent_num_bytes(leaf, fi);
1011                         if (extent_end <= start) {
1012                                 path->slots[0]++;
1013                                 goto next_slot;
1014                         }
1015                         if (disk_bytenr == 0)
1016                                 goto out_check;
1017                         if (btrfs_file_extent_compression(leaf, fi) ||
1018                             btrfs_file_extent_encryption(leaf, fi) ||
1019                             btrfs_file_extent_other_encoding(leaf, fi))
1020                                 goto out_check;
1021                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022                                 goto out_check;
1023                         if (btrfs_extent_readonly(root, disk_bytenr))
1024                                 goto out_check;
1025                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026                                                   disk_bytenr))
1027                                 goto out_check;
1028                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1029                         disk_bytenr += cur_offset - found_key.offset;
1030                         num_bytes = min(end + 1, extent_end) - cur_offset;
1031                         /*
1032                          * force cow if csum exists in the range.
1033                          * this ensure that csum for a given extent are
1034                          * either valid or do not exist.
1035                          */
1036                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1037                                 goto out_check;
1038                         nocow = 1;
1039                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1040                         extent_end = found_key.offset +
1041                                 btrfs_file_extent_inline_len(leaf, fi);
1042                         extent_end = ALIGN(extent_end, root->sectorsize);
1043                 } else {
1044                         BUG_ON(1);
1045                 }
1046 out_check:
1047                 if (extent_end <= start) {
1048                         path->slots[0]++;
1049                         goto next_slot;
1050                 }
1051                 if (!nocow) {
1052                         if (cow_start == (u64)-1)
1053                                 cow_start = cur_offset;
1054                         cur_offset = extent_end;
1055                         if (cur_offset > end)
1056                                 break;
1057                         path->slots[0]++;
1058                         goto next_slot;
1059                 }
1060
1061                 btrfs_release_path(root, path);
1062                 if (cow_start != (u64)-1) {
1063                         ret = cow_file_range(inode, locked_page, cow_start,
1064                                         found_key.offset - 1, page_started,
1065                                         nr_written, 1);
1066                         BUG_ON(ret);
1067                         cow_start = (u64)-1;
1068                 }
1069
1070                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1071                         struct extent_map *em;
1072                         struct extent_map_tree *em_tree;
1073                         em_tree = &BTRFS_I(inode)->extent_tree;
1074                         em = alloc_extent_map(GFP_NOFS);
1075                         em->start = cur_offset;
1076                         em->orig_start = em->start;
1077                         em->len = num_bytes;
1078                         em->block_len = num_bytes;
1079                         em->block_start = disk_bytenr;
1080                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1081                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082                         while (1) {
1083                                 spin_lock(&em_tree->lock);
1084                                 ret = add_extent_mapping(em_tree, em);
1085                                 spin_unlock(&em_tree->lock);
1086                                 if (ret != -EEXIST) {
1087                                         free_extent_map(em);
1088                                         break;
1089                                 }
1090                                 btrfs_drop_extent_cache(inode, em->start,
1091                                                 em->start + em->len - 1, 0);
1092                         }
1093                         type = BTRFS_ORDERED_PREALLOC;
1094                 } else {
1095                         type = BTRFS_ORDERED_NOCOW;
1096                 }
1097
1098                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1099                                                num_bytes, num_bytes, type);
1100                 BUG_ON(ret);
1101
1102                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103                                         cur_offset, cur_offset + num_bytes - 1,
1104                                         locked_page, 1, 1, 1, 0, 0, 0);
1105                 cur_offset = extent_end;
1106                 if (cur_offset > end)
1107                         break;
1108         }
1109         btrfs_release_path(root, path);
1110
1111         if (cur_offset <= end && cow_start == (u64)-1)
1112                 cow_start = cur_offset;
1113         if (cow_start != (u64)-1) {
1114                 ret = cow_file_range(inode, locked_page, cow_start, end,
1115                                      page_started, nr_written, 1);
1116                 BUG_ON(ret);
1117         }
1118
1119         ret = btrfs_end_transaction(trans, root);
1120         BUG_ON(ret);
1121         btrfs_free_path(path);
1122         return 0;
1123 }
1124
1125 /*
1126  * extent_io.c call back to do delayed allocation processing
1127  */
1128 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1129                               u64 start, u64 end, int *page_started,
1130                               unsigned long *nr_written)
1131 {
1132         int ret;
1133         struct btrfs_root *root = BTRFS_I(inode)->root;
1134
1135         if (btrfs_test_flag(inode, NODATACOW))
1136                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1137                                          page_started, 1, nr_written);
1138         else if (btrfs_test_flag(inode, PREALLOC))
1139                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1140                                          page_started, 0, nr_written);
1141         else if (!btrfs_test_opt(root, COMPRESS))
1142                 ret = cow_file_range(inode, locked_page, start, end,
1143                                       page_started, nr_written, 1);
1144         else
1145                 ret = cow_file_range_async(inode, locked_page, start, end,
1146                                            page_started, nr_written);
1147         return ret;
1148 }
1149
1150 /*
1151  * extent_io.c set_bit_hook, used to track delayed allocation
1152  * bytes in this file, and to maintain the list of inodes that
1153  * have pending delalloc work to be done.
1154  */
1155 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156                        unsigned long old, unsigned long bits)
1157 {
1158         /*
1159          * set_bit and clear bit hooks normally require _irqsave/restore
1160          * but in this case, we are only testeing for the DELALLOC
1161          * bit, which is only set or cleared with irqs on
1162          */
1163         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164                 struct btrfs_root *root = BTRFS_I(inode)->root;
1165                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166                 spin_lock(&root->fs_info->delalloc_lock);
1167                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1168                 root->fs_info->delalloc_bytes += end - start + 1;
1169                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171                                       &root->fs_info->delalloc_inodes);
1172                 }
1173                 spin_unlock(&root->fs_info->delalloc_lock);
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * extent_io.c clear_bit_hook, see set_bit_hook for why
1180  */
1181 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1182                          unsigned long old, unsigned long bits)
1183 {
1184         /*
1185          * set_bit and clear bit hooks normally require _irqsave/restore
1186          * but in this case, we are only testeing for the DELALLOC
1187          * bit, which is only set or cleared with irqs on
1188          */
1189         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190                 struct btrfs_root *root = BTRFS_I(inode)->root;
1191
1192                 spin_lock(&root->fs_info->delalloc_lock);
1193                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1194                         printk(KERN_INFO "btrfs warning: delalloc account "
1195                                "%llu %llu\n",
1196                                (unsigned long long)end - start + 1,
1197                                (unsigned long long)
1198                                root->fs_info->delalloc_bytes);
1199                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1200                         root->fs_info->delalloc_bytes = 0;
1201                         BTRFS_I(inode)->delalloc_bytes = 0;
1202                 } else {
1203                         btrfs_delalloc_free_space(root, inode,
1204                                                   end - start + 1);
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct btrfs_ordered_sum *sum;
1328
1329         btrfs_set_trans_block_group(trans, inode);
1330
1331         list_for_each_entry(sum, list, list) {
1332                 btrfs_csum_file_blocks(trans,
1333                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1334         }
1335         return 0;
1336 }
1337
1338 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339 {
1340         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1341                 WARN_ON(1);
1342         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343                                    GFP_NOFS);
1344 }
1345
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup {
1348         struct page *page;
1349         struct btrfs_work work;
1350 };
1351
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1353 {
1354         struct btrfs_writepage_fixup *fixup;
1355         struct btrfs_ordered_extent *ordered;
1356         struct page *page;
1357         struct inode *inode;
1358         u64 page_start;
1359         u64 page_end;
1360
1361         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362         page = fixup->page;
1363 again:
1364         lock_page(page);
1365         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366                 ClearPageChecked(page);
1367                 goto out_page;
1368         }
1369
1370         inode = page->mapping->host;
1371         page_start = page_offset(page);
1372         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375
1376         /* already ordered? We're done */
1377         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378                              EXTENT_ORDERED, 0)) {
1379                 goto out;
1380         }
1381
1382         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383         if (ordered) {
1384                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385                               page_end, GFP_NOFS);
1386                 unlock_page(page);
1387                 btrfs_start_ordered_extent(inode, ordered, 1);
1388                 goto again;
1389         }
1390
1391         btrfs_set_extent_delalloc(inode, page_start, page_end);
1392         ClearPageChecked(page);
1393 out:
1394         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395 out_page:
1396         unlock_page(page);
1397         page_cache_release(page);
1398 }
1399
1400 /*
1401  * There are a few paths in the higher layers of the kernel that directly
1402  * set the page dirty bit without asking the filesystem if it is a
1403  * good idea.  This causes problems because we want to make sure COW
1404  * properly happens and the data=ordered rules are followed.
1405  *
1406  * In our case any range that doesn't have the ORDERED bit set
1407  * hasn't been properly setup for IO.  We kick off an async process
1408  * to fix it up.  The async helper will wait for ordered extents, set
1409  * the delalloc bit and make it safe to write the page.
1410  */
1411 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1412 {
1413         struct inode *inode = page->mapping->host;
1414         struct btrfs_writepage_fixup *fixup;
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret;
1417
1418         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419                              EXTENT_ORDERED, 0);
1420         if (ret)
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u8 compression, u8 encryption,
1443                                        u16 other_encoding, int extent_type)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         struct btrfs_file_extent_item *fi;
1447         struct btrfs_path *path;
1448         struct extent_buffer *leaf;
1449         struct btrfs_key ins;
1450         u64 hint;
1451         int ret;
1452
1453         path = btrfs_alloc_path();
1454         BUG_ON(!path);
1455
1456         path->leave_spinning = 1;
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
1480         btrfs_unlock_up_safe(path, 1);
1481         btrfs_set_lock_blocking(leaf);
1482
1483         btrfs_mark_buffer_dirty(leaf);
1484
1485         inode_add_bytes(inode, num_bytes);
1486         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1487
1488         ins.objectid = disk_bytenr;
1489         ins.offset = disk_num_bytes;
1490         ins.type = BTRFS_EXTENT_ITEM_KEY;
1491         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1492                                           root->root_key.objectid,
1493                                           trans->transid, inode->i_ino, &ins);
1494         BUG_ON(ret);
1495         btrfs_free_path(path);
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * helper function for btrfs_finish_ordered_io, this
1502  * just reads in some of the csum leaves to prime them into ram
1503  * before we start the transaction.  It limits the amount of btree
1504  * reads required while inside the transaction.
1505  */
1506 static noinline void reada_csum(struct btrfs_root *root,
1507                                 struct btrfs_path *path,
1508                                 struct btrfs_ordered_extent *ordered_extent)
1509 {
1510         struct btrfs_ordered_sum *sum;
1511         u64 bytenr;
1512
1513         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1514                          list);
1515         bytenr = sum->sums[0].bytenr;
1516
1517         /*
1518          * we don't care about the results, the point of this search is
1519          * just to get the btree leaves into ram
1520          */
1521         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1522 }
1523
1524 /* as ordered data IO finishes, this gets called so we can finish
1525  * an ordered extent if the range of bytes in the file it covers are
1526  * fully written.
1527  */
1528 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1529 {
1530         struct btrfs_root *root = BTRFS_I(inode)->root;
1531         struct btrfs_trans_handle *trans;
1532         struct btrfs_ordered_extent *ordered_extent = NULL;
1533         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1534         struct btrfs_path *path;
1535         int compressed = 0;
1536         int ret;
1537
1538         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1539         if (!ret)
1540                 return 0;
1541
1542         /*
1543          * before we join the transaction, try to do some of our IO.
1544          * This will limit the amount of IO that we have to do with
1545          * the transaction running.  We're unlikely to need to do any
1546          * IO if the file extents are new, the disk_i_size checks
1547          * covers the most common case.
1548          */
1549         if (start < BTRFS_I(inode)->disk_i_size) {
1550                 path = btrfs_alloc_path();
1551                 if (path) {
1552                         ret = btrfs_lookup_file_extent(NULL, root, path,
1553                                                        inode->i_ino,
1554                                                        start, 0);
1555                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1556                                                                      start);
1557                         if (!list_empty(&ordered_extent->list)) {
1558                                 btrfs_release_path(root, path);
1559                                 reada_csum(root, path, ordered_extent);
1560                         }
1561                         btrfs_free_path(path);
1562                 }
1563         }
1564
1565         trans = btrfs_join_transaction(root, 1);
1566
1567         if (!ordered_extent)
1568                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1569         BUG_ON(!ordered_extent);
1570         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1571                 goto nocow;
1572
1573         lock_extent(io_tree, ordered_extent->file_offset,
1574                     ordered_extent->file_offset + ordered_extent->len - 1,
1575                     GFP_NOFS);
1576
1577         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1578                 compressed = 1;
1579         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1580                 BUG_ON(compressed);
1581                 ret = btrfs_mark_extent_written(trans, root, inode,
1582                                                 ordered_extent->file_offset,
1583                                                 ordered_extent->file_offset +
1584                                                 ordered_extent->len);
1585                 BUG_ON(ret);
1586         } else {
1587                 ret = insert_reserved_file_extent(trans, inode,
1588                                                 ordered_extent->file_offset,
1589                                                 ordered_extent->start,
1590                                                 ordered_extent->disk_len,
1591                                                 ordered_extent->len,
1592                                                 ordered_extent->len,
1593                                                 compressed, 0, 0,
1594                                                 BTRFS_FILE_EXTENT_REG);
1595                 BUG_ON(ret);
1596         }
1597         unlock_extent(io_tree, ordered_extent->file_offset,
1598                     ordered_extent->file_offset + ordered_extent->len - 1,
1599                     GFP_NOFS);
1600 nocow:
1601         add_pending_csums(trans, inode, ordered_extent->file_offset,
1602                           &ordered_extent->list);
1603
1604         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1605         btrfs_ordered_update_i_size(inode, ordered_extent);
1606         btrfs_update_inode(trans, root, inode);
1607         btrfs_remove_ordered_extent(inode, ordered_extent);
1608         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1609
1610         /* once for us */
1611         btrfs_put_ordered_extent(ordered_extent);
1612         /* once for the tree */
1613         btrfs_put_ordered_extent(ordered_extent);
1614
1615         btrfs_end_transaction(trans, root);
1616         return 0;
1617 }
1618
1619 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1620                                 struct extent_state *state, int uptodate)
1621 {
1622         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1623 }
1624
1625 /*
1626  * When IO fails, either with EIO or csum verification fails, we
1627  * try other mirrors that might have a good copy of the data.  This
1628  * io_failure_record is used to record state as we go through all the
1629  * mirrors.  If another mirror has good data, the page is set up to date
1630  * and things continue.  If a good mirror can't be found, the original
1631  * bio end_io callback is called to indicate things have failed.
1632  */
1633 struct io_failure_record {
1634         struct page *page;
1635         u64 start;
1636         u64 len;
1637         u64 logical;
1638         unsigned long bio_flags;
1639         int last_mirror;
1640 };
1641
1642 static int btrfs_io_failed_hook(struct bio *failed_bio,
1643                          struct page *page, u64 start, u64 end,
1644                          struct extent_state *state)
1645 {
1646         struct io_failure_record *failrec = NULL;
1647         u64 private;
1648         struct extent_map *em;
1649         struct inode *inode = page->mapping->host;
1650         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1651         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1652         struct bio *bio;
1653         int num_copies;
1654         int ret;
1655         int rw;
1656         u64 logical;
1657
1658         ret = get_state_private(failure_tree, start, &private);
1659         if (ret) {
1660                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1661                 if (!failrec)
1662                         return -ENOMEM;
1663                 failrec->start = start;
1664                 failrec->len = end - start + 1;
1665                 failrec->last_mirror = 0;
1666                 failrec->bio_flags = 0;
1667
1668                 spin_lock(&em_tree->lock);
1669                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1670                 if (em->start > start || em->start + em->len < start) {
1671                         free_extent_map(em);
1672                         em = NULL;
1673                 }
1674                 spin_unlock(&em_tree->lock);
1675
1676                 if (!em || IS_ERR(em)) {
1677                         kfree(failrec);
1678                         return -EIO;
1679                 }
1680                 logical = start - em->start;
1681                 logical = em->block_start + logical;
1682                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1683                         logical = em->block_start;
1684                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1685                 }
1686                 failrec->logical = logical;
1687                 free_extent_map(em);
1688                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1689                                 EXTENT_DIRTY, GFP_NOFS);
1690                 set_state_private(failure_tree, start,
1691                                  (u64)(unsigned long)failrec);
1692         } else {
1693                 failrec = (struct io_failure_record *)(unsigned long)private;
1694         }
1695         num_copies = btrfs_num_copies(
1696                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1697                               failrec->logical, failrec->len);
1698         failrec->last_mirror++;
1699         if (!state) {
1700                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1701                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1702                                                     failrec->start,
1703                                                     EXTENT_LOCKED);
1704                 if (state && state->start != failrec->start)
1705                         state = NULL;
1706                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1707         }
1708         if (!state || failrec->last_mirror > num_copies) {
1709                 set_state_private(failure_tree, failrec->start, 0);
1710                 clear_extent_bits(failure_tree, failrec->start,
1711                                   failrec->start + failrec->len - 1,
1712                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1713                 kfree(failrec);
1714                 return -EIO;
1715         }
1716         bio = bio_alloc(GFP_NOFS, 1);
1717         bio->bi_private = state;
1718         bio->bi_end_io = failed_bio->bi_end_io;
1719         bio->bi_sector = failrec->logical >> 9;
1720         bio->bi_bdev = failed_bio->bi_bdev;
1721         bio->bi_size = 0;
1722
1723         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1724         if (failed_bio->bi_rw & (1 << BIO_RW))
1725                 rw = WRITE;
1726         else
1727                 rw = READ;
1728
1729         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1730                                                       failrec->last_mirror,
1731                                                       failrec->bio_flags);
1732         return 0;
1733 }
1734
1735 /*
1736  * each time an IO finishes, we do a fast check in the IO failure tree
1737  * to see if we need to process or clean up an io_failure_record
1738  */
1739 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1740 {
1741         u64 private;
1742         u64 private_failure;
1743         struct io_failure_record *failure;
1744         int ret;
1745
1746         private = 0;
1747         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1748                              (u64)-1, 1, EXTENT_DIRTY)) {
1749                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1750                                         start, &private_failure);
1751                 if (ret == 0) {
1752                         failure = (struct io_failure_record *)(unsigned long)
1753                                    private_failure;
1754                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1755                                           failure->start, 0);
1756                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1757                                           failure->start,
1758                                           failure->start + failure->len - 1,
1759                                           EXTENT_DIRTY | EXTENT_LOCKED,
1760                                           GFP_NOFS);
1761                         kfree(failure);
1762                 }
1763         }
1764         return 0;
1765 }
1766
1767 /*
1768  * when reads are done, we need to check csums to verify the data is correct
1769  * if there's a match, we allow the bio to finish.  If not, we go through
1770  * the io_failure_record routines to find good copies
1771  */
1772 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1773                                struct extent_state *state)
1774 {
1775         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1776         struct inode *inode = page->mapping->host;
1777         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1778         char *kaddr;
1779         u64 private = ~(u32)0;
1780         int ret;
1781         struct btrfs_root *root = BTRFS_I(inode)->root;
1782         u32 csum = ~(u32)0;
1783
1784         if (PageChecked(page)) {
1785                 ClearPageChecked(page);
1786                 goto good;
1787         }
1788         if (btrfs_test_flag(inode, NODATASUM))
1789                 return 0;
1790
1791         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1792             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1793                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1794                                   GFP_NOFS);
1795                 return 0;
1796         }
1797
1798         if (state && state->start == start) {
1799                 private = state->private;
1800                 ret = 0;
1801         } else {
1802                 ret = get_state_private(io_tree, start, &private);
1803         }
1804         kaddr = kmap_atomic(page, KM_USER0);
1805         if (ret)
1806                 goto zeroit;
1807
1808         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1809         btrfs_csum_final(csum, (char *)&csum);
1810         if (csum != private)
1811                 goto zeroit;
1812
1813         kunmap_atomic(kaddr, KM_USER0);
1814 good:
1815         /* if the io failure tree for this inode is non-empty,
1816          * check to see if we've recovered from a failed IO
1817          */
1818         btrfs_clean_io_failures(inode, start);
1819         return 0;
1820
1821 zeroit:
1822         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1823                "private %llu\n", page->mapping->host->i_ino,
1824                (unsigned long long)start, csum,
1825                (unsigned long long)private);
1826         memset(kaddr + offset, 1, end - start + 1);
1827         flush_dcache_page(page);
1828         kunmap_atomic(kaddr, KM_USER0);
1829         if (private == 0)
1830                 return 0;
1831         return -EIO;
1832 }
1833
1834 /*
1835  * This creates an orphan entry for the given inode in case something goes
1836  * wrong in the middle of an unlink/truncate.
1837  */
1838 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1839 {
1840         struct btrfs_root *root = BTRFS_I(inode)->root;
1841         int ret = 0;
1842
1843         spin_lock(&root->list_lock);
1844
1845         /* already on the orphan list, we're good */
1846         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1847                 spin_unlock(&root->list_lock);
1848                 return 0;
1849         }
1850
1851         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1852
1853         spin_unlock(&root->list_lock);
1854
1855         /*
1856          * insert an orphan item to track this unlinked/truncated file
1857          */
1858         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1859
1860         return ret;
1861 }
1862
1863 /*
1864  * We have done the truncate/delete so we can go ahead and remove the orphan
1865  * item for this particular inode.
1866  */
1867 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1868 {
1869         struct btrfs_root *root = BTRFS_I(inode)->root;
1870         int ret = 0;
1871
1872         spin_lock(&root->list_lock);
1873
1874         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1875                 spin_unlock(&root->list_lock);
1876                 return 0;
1877         }
1878
1879         list_del_init(&BTRFS_I(inode)->i_orphan);
1880         if (!trans) {
1881                 spin_unlock(&root->list_lock);
1882                 return 0;
1883         }
1884
1885         spin_unlock(&root->list_lock);
1886
1887         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1888
1889         return ret;
1890 }
1891
1892 /*
1893  * this cleans up any orphans that may be left on the list from the last use
1894  * of this root.
1895  */
1896 void btrfs_orphan_cleanup(struct btrfs_root *root)
1897 {
1898         struct btrfs_path *path;
1899         struct extent_buffer *leaf;
1900         struct btrfs_item *item;
1901         struct btrfs_key key, found_key;
1902         struct btrfs_trans_handle *trans;
1903         struct inode *inode;
1904         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1905
1906         path = btrfs_alloc_path();
1907         if (!path)
1908                 return;
1909         path->reada = -1;
1910
1911         key.objectid = BTRFS_ORPHAN_OBJECTID;
1912         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1913         key.offset = (u64)-1;
1914
1915
1916         while (1) {
1917                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1918                 if (ret < 0) {
1919                         printk(KERN_ERR "Error searching slot for orphan: %d"
1920                                "\n", ret);
1921                         break;
1922                 }
1923
1924                 /*
1925                  * if ret == 0 means we found what we were searching for, which
1926                  * is weird, but possible, so only screw with path if we didnt
1927                  * find the key and see if we have stuff that matches
1928                  */
1929                 if (ret > 0) {
1930                         if (path->slots[0] == 0)
1931                                 break;
1932                         path->slots[0]--;
1933                 }
1934
1935                 /* pull out the item */
1936                 leaf = path->nodes[0];
1937                 item = btrfs_item_nr(leaf, path->slots[0]);
1938                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1939
1940                 /* make sure the item matches what we want */
1941                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1942                         break;
1943                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1944                         break;
1945
1946                 /* release the path since we're done with it */
1947                 btrfs_release_path(root, path);
1948
1949                 /*
1950                  * this is where we are basically btrfs_lookup, without the
1951                  * crossing root thing.  we store the inode number in the
1952                  * offset of the orphan item.
1953                  */
1954                 inode = btrfs_iget_locked(root->fs_info->sb,
1955                                           found_key.offset, root);
1956                 if (!inode)
1957                         break;
1958
1959                 if (inode->i_state & I_NEW) {
1960                         BTRFS_I(inode)->root = root;
1961
1962                         /* have to set the location manually */
1963                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1964                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1965                         BTRFS_I(inode)->location.offset = 0;
1966
1967                         btrfs_read_locked_inode(inode);
1968                         unlock_new_inode(inode);
1969                 }
1970
1971                 /*
1972                  * add this inode to the orphan list so btrfs_orphan_del does
1973                  * the proper thing when we hit it
1974                  */
1975                 spin_lock(&root->list_lock);
1976                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1977                 spin_unlock(&root->list_lock);
1978
1979                 /*
1980                  * if this is a bad inode, means we actually succeeded in
1981                  * removing the inode, but not the orphan record, which means
1982                  * we need to manually delete the orphan since iput will just
1983                  * do a destroy_inode
1984                  */
1985                 if (is_bad_inode(inode)) {
1986                         trans = btrfs_start_transaction(root, 1);
1987                         btrfs_orphan_del(trans, inode);
1988                         btrfs_end_transaction(trans, root);
1989                         iput(inode);
1990                         continue;
1991                 }
1992
1993                 /* if we have links, this was a truncate, lets do that */
1994                 if (inode->i_nlink) {
1995                         nr_truncate++;
1996                         btrfs_truncate(inode);
1997                 } else {
1998                         nr_unlink++;
1999                 }
2000
2001                 /* this will do delete_inode and everything for us */
2002                 iput(inode);
2003         }
2004
2005         if (nr_unlink)
2006                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2007         if (nr_truncate)
2008                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2009
2010         btrfs_free_path(path);
2011 }
2012
2013 /*
2014  * read an inode from the btree into the in-memory inode
2015  */
2016 void btrfs_read_locked_inode(struct inode *inode)
2017 {
2018         struct btrfs_path *path;
2019         struct extent_buffer *leaf;
2020         struct btrfs_inode_item *inode_item;
2021         struct btrfs_timespec *tspec;
2022         struct btrfs_root *root = BTRFS_I(inode)->root;
2023         struct btrfs_key location;
2024         u64 alloc_group_block;
2025         u32 rdev;
2026         int ret;
2027
2028         path = btrfs_alloc_path();
2029         BUG_ON(!path);
2030         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2031
2032         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2033         if (ret)
2034                 goto make_bad;
2035
2036         leaf = path->nodes[0];
2037         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2038                                     struct btrfs_inode_item);
2039
2040         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2041         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2042         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2043         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2044         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2045
2046         tspec = btrfs_inode_atime(inode_item);
2047         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2048         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2049
2050         tspec = btrfs_inode_mtime(inode_item);
2051         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2052         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2053
2054         tspec = btrfs_inode_ctime(inode_item);
2055         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2056         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2057
2058         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2059         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2060         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2061         inode->i_generation = BTRFS_I(inode)->generation;
2062         inode->i_rdev = 0;
2063         rdev = btrfs_inode_rdev(leaf, inode_item);
2064
2065         BTRFS_I(inode)->index_cnt = (u64)-1;
2066         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2067
2068         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2069
2070         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2071                                                 alloc_group_block, 0);
2072         btrfs_free_path(path);
2073         inode_item = NULL;
2074
2075         switch (inode->i_mode & S_IFMT) {
2076         case S_IFREG:
2077                 inode->i_mapping->a_ops = &btrfs_aops;
2078                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2079                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2080                 inode->i_fop = &btrfs_file_operations;
2081                 inode->i_op = &btrfs_file_inode_operations;
2082                 break;
2083         case S_IFDIR:
2084                 inode->i_fop = &btrfs_dir_file_operations;
2085                 if (root == root->fs_info->tree_root)
2086                         inode->i_op = &btrfs_dir_ro_inode_operations;
2087                 else
2088                         inode->i_op = &btrfs_dir_inode_operations;
2089                 break;
2090         case S_IFLNK:
2091                 inode->i_op = &btrfs_symlink_inode_operations;
2092                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2093                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2094                 break;
2095         default:
2096                 inode->i_op = &btrfs_special_inode_operations;
2097                 init_special_inode(inode, inode->i_mode, rdev);
2098                 break;
2099         }
2100         return;
2101
2102 make_bad:
2103         btrfs_free_path(path);
2104         make_bad_inode(inode);
2105 }
2106
2107 /*
2108  * given a leaf and an inode, copy the inode fields into the leaf
2109  */
2110 static void fill_inode_item(struct btrfs_trans_handle *trans,
2111                             struct extent_buffer *leaf,
2112                             struct btrfs_inode_item *item,
2113                             struct inode *inode)
2114 {
2115         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2116         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2117         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2118         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2119         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2120
2121         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2122                                inode->i_atime.tv_sec);
2123         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2124                                 inode->i_atime.tv_nsec);
2125
2126         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2127                                inode->i_mtime.tv_sec);
2128         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2129                                 inode->i_mtime.tv_nsec);
2130
2131         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2132                                inode->i_ctime.tv_sec);
2133         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2134                                 inode->i_ctime.tv_nsec);
2135
2136         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2137         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2138         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2139         btrfs_set_inode_transid(leaf, item, trans->transid);
2140         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2141         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2142         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2143 }
2144
2145 /*
2146  * copy everything in the in-memory inode into the btree.
2147  */
2148 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2149                                 struct btrfs_root *root, struct inode *inode)
2150 {
2151         struct btrfs_inode_item *inode_item;
2152         struct btrfs_path *path;
2153         struct extent_buffer *leaf;
2154         int ret;
2155
2156         path = btrfs_alloc_path();
2157         BUG_ON(!path);
2158         path->leave_spinning = 1;
2159         ret = btrfs_lookup_inode(trans, root, path,
2160                                  &BTRFS_I(inode)->location, 1);
2161         if (ret) {
2162                 if (ret > 0)
2163                         ret = -ENOENT;
2164                 goto failed;
2165         }
2166
2167         btrfs_unlock_up_safe(path, 1);
2168         leaf = path->nodes[0];
2169         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2170                                   struct btrfs_inode_item);
2171
2172         fill_inode_item(trans, leaf, inode_item, inode);
2173         btrfs_mark_buffer_dirty(leaf);
2174         btrfs_set_inode_last_trans(trans, inode);
2175         ret = 0;
2176 failed:
2177         btrfs_free_path(path);
2178         return ret;
2179 }
2180
2181
2182 /*
2183  * unlink helper that gets used here in inode.c and in the tree logging
2184  * recovery code.  It remove a link in a directory with a given name, and
2185  * also drops the back refs in the inode to the directory
2186  */
2187 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2188                        struct btrfs_root *root,
2189                        struct inode *dir, struct inode *inode,
2190                        const char *name, int name_len)
2191 {
2192         struct btrfs_path *path;
2193         int ret = 0;
2194         struct extent_buffer *leaf;
2195         struct btrfs_dir_item *di;
2196         struct btrfs_key key;
2197         u64 index;
2198
2199         path = btrfs_alloc_path();
2200         if (!path) {
2201                 ret = -ENOMEM;
2202                 goto err;
2203         }
2204
2205         path->leave_spinning = 1;
2206         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2207                                     name, name_len, -1);
2208         if (IS_ERR(di)) {
2209                 ret = PTR_ERR(di);
2210                 goto err;
2211         }
2212         if (!di) {
2213                 ret = -ENOENT;
2214                 goto err;
2215         }
2216         leaf = path->nodes[0];
2217         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2218         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2219         if (ret)
2220                 goto err;
2221         btrfs_release_path(root, path);
2222
2223         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2224                                   inode->i_ino,
2225                                   dir->i_ino, &index);
2226         if (ret) {
2227                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2228                        "inode %lu parent %lu\n", name_len, name,
2229                        inode->i_ino, dir->i_ino);
2230                 goto err;
2231         }
2232
2233         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2234                                          index, name, name_len, -1);
2235         if (IS_ERR(di)) {
2236                 ret = PTR_ERR(di);
2237                 goto err;
2238         }
2239         if (!di) {
2240                 ret = -ENOENT;
2241                 goto err;
2242         }
2243         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2244         btrfs_release_path(root, path);
2245
2246         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2247                                          inode, dir->i_ino);
2248         BUG_ON(ret != 0 && ret != -ENOENT);
2249
2250         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2251                                            dir, index);
2252         BUG_ON(ret);
2253 err:
2254         btrfs_free_path(path);
2255         if (ret)
2256                 goto out;
2257
2258         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2259         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2260         btrfs_update_inode(trans, root, dir);
2261         btrfs_drop_nlink(inode);
2262         ret = btrfs_update_inode(trans, root, inode);
2263         dir->i_sb->s_dirt = 1;
2264 out:
2265         return ret;
2266 }
2267
2268 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2269 {
2270         struct btrfs_root *root;
2271         struct btrfs_trans_handle *trans;
2272         struct inode *inode = dentry->d_inode;
2273         int ret;
2274         unsigned long nr = 0;
2275
2276         root = BTRFS_I(dir)->root;
2277
2278         trans = btrfs_start_transaction(root, 1);
2279
2280         btrfs_set_trans_block_group(trans, dir);
2281
2282         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2283
2284         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2285                                  dentry->d_name.name, dentry->d_name.len);
2286
2287         if (inode->i_nlink == 0)
2288                 ret = btrfs_orphan_add(trans, inode);
2289
2290         nr = trans->blocks_used;
2291
2292         btrfs_end_transaction_throttle(trans, root);
2293         btrfs_btree_balance_dirty(root, nr);
2294         return ret;
2295 }
2296
2297 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2298 {
2299         struct inode *inode = dentry->d_inode;
2300         int err = 0;
2301         int ret;
2302         struct btrfs_root *root = BTRFS_I(dir)->root;
2303         struct btrfs_trans_handle *trans;
2304         unsigned long nr = 0;
2305
2306         /*
2307          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2308          * the root of a subvolume or snapshot
2309          */
2310         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2311             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2312                 return -ENOTEMPTY;
2313         }
2314
2315         trans = btrfs_start_transaction(root, 1);
2316         btrfs_set_trans_block_group(trans, dir);
2317
2318         err = btrfs_orphan_add(trans, inode);
2319         if (err)
2320                 goto fail_trans;
2321
2322         /* now the directory is empty */
2323         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2324                                  dentry->d_name.name, dentry->d_name.len);
2325         if (!err)
2326                 btrfs_i_size_write(inode, 0);
2327
2328 fail_trans:
2329         nr = trans->blocks_used;
2330         ret = btrfs_end_transaction_throttle(trans, root);
2331         btrfs_btree_balance_dirty(root, nr);
2332
2333         if (ret && !err)
2334                 err = ret;
2335         return err;
2336 }
2337
2338 #if 0
2339 /*
2340  * when truncating bytes in a file, it is possible to avoid reading
2341  * the leaves that contain only checksum items.  This can be the
2342  * majority of the IO required to delete a large file, but it must
2343  * be done carefully.
2344  *
2345  * The keys in the level just above the leaves are checked to make sure
2346  * the lowest key in a given leaf is a csum key, and starts at an offset
2347  * after the new  size.
2348  *
2349  * Then the key for the next leaf is checked to make sure it also has
2350  * a checksum item for the same file.  If it does, we know our target leaf
2351  * contains only checksum items, and it can be safely freed without reading
2352  * it.
2353  *
2354  * This is just an optimization targeted at large files.  It may do
2355  * nothing.  It will return 0 unless things went badly.
2356  */
2357 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2358                                      struct btrfs_root *root,
2359                                      struct btrfs_path *path,
2360                                      struct inode *inode, u64 new_size)
2361 {
2362         struct btrfs_key key;
2363         int ret;
2364         int nritems;
2365         struct btrfs_key found_key;
2366         struct btrfs_key other_key;
2367         struct btrfs_leaf_ref *ref;
2368         u64 leaf_gen;
2369         u64 leaf_start;
2370
2371         path->lowest_level = 1;
2372         key.objectid = inode->i_ino;
2373         key.type = BTRFS_CSUM_ITEM_KEY;
2374         key.offset = new_size;
2375 again:
2376         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2377         if (ret < 0)
2378                 goto out;
2379
2380         if (path->nodes[1] == NULL) {
2381                 ret = 0;
2382                 goto out;
2383         }
2384         ret = 0;
2385         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2386         nritems = btrfs_header_nritems(path->nodes[1]);
2387
2388         if (!nritems)
2389                 goto out;
2390
2391         if (path->slots[1] >= nritems)
2392                 goto next_node;
2393
2394         /* did we find a key greater than anything we want to delete? */
2395         if (found_key.objectid > inode->i_ino ||
2396            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2397                 goto out;
2398
2399         /* we check the next key in the node to make sure the leave contains
2400          * only checksum items.  This comparison doesn't work if our
2401          * leaf is the last one in the node
2402          */
2403         if (path->slots[1] + 1 >= nritems) {
2404 next_node:
2405                 /* search forward from the last key in the node, this
2406                  * will bring us into the next node in the tree
2407                  */
2408                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2409
2410                 /* unlikely, but we inc below, so check to be safe */
2411                 if (found_key.offset == (u64)-1)
2412                         goto out;
2413
2414                 /* search_forward needs a path with locks held, do the
2415                  * search again for the original key.  It is possible
2416                  * this will race with a balance and return a path that
2417                  * we could modify, but this drop is just an optimization
2418                  * and is allowed to miss some leaves.
2419                  */
2420                 btrfs_release_path(root, path);
2421                 found_key.offset++;
2422
2423                 /* setup a max key for search_forward */
2424                 other_key.offset = (u64)-1;
2425                 other_key.type = key.type;
2426                 other_key.objectid = key.objectid;
2427
2428                 path->keep_locks = 1;
2429                 ret = btrfs_search_forward(root, &found_key, &other_key,
2430                                            path, 0, 0);
2431                 path->keep_locks = 0;
2432                 if (ret || found_key.objectid != key.objectid ||
2433                     found_key.type != key.type) {
2434                         ret = 0;
2435                         goto out;
2436                 }
2437
2438                 key.offset = found_key.offset;
2439                 btrfs_release_path(root, path);
2440                 cond_resched();
2441                 goto again;
2442         }
2443
2444         /* we know there's one more slot after us in the tree,
2445          * read that key so we can verify it is also a checksum item
2446          */
2447         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2448
2449         if (found_key.objectid < inode->i_ino)
2450                 goto next_key;
2451
2452         if (found_key.type != key.type || found_key.offset < new_size)
2453                 goto next_key;
2454
2455         /*
2456          * if the key for the next leaf isn't a csum key from this objectid,
2457          * we can't be sure there aren't good items inside this leaf.
2458          * Bail out
2459          */
2460         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2461                 goto out;
2462
2463         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2464         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2465         /*
2466          * it is safe to delete this leaf, it contains only
2467          * csum items from this inode at an offset >= new_size
2468          */
2469         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2470         BUG_ON(ret);
2471
2472         if (root->ref_cows && leaf_gen < trans->transid) {
2473                 ref = btrfs_alloc_leaf_ref(root, 0);
2474                 if (ref) {
2475                         ref->root_gen = root->root_key.offset;
2476                         ref->bytenr = leaf_start;
2477                         ref->owner = 0;
2478                         ref->generation = leaf_gen;
2479                         ref->nritems = 0;
2480
2481                         btrfs_sort_leaf_ref(ref);
2482
2483                         ret = btrfs_add_leaf_ref(root, ref, 0);
2484                         WARN_ON(ret);
2485                         btrfs_free_leaf_ref(root, ref);
2486                 } else {
2487                         WARN_ON(1);
2488                 }
2489         }
2490 next_key:
2491         btrfs_release_path(root, path);
2492
2493         if (other_key.objectid == inode->i_ino &&
2494             other_key.type == key.type && other_key.offset > key.offset) {
2495                 key.offset = other_key.offset;
2496                 cond_resched();
2497                 goto again;
2498         }
2499         ret = 0;
2500 out:
2501         /* fixup any changes we've made to the path */
2502         path->lowest_level = 0;
2503         path->keep_locks = 0;
2504         btrfs_release_path(root, path);
2505         return ret;
2506 }
2507
2508 #endif
2509
2510 /*
2511  * this can truncate away extent items, csum items and directory items.
2512  * It starts at a high offset and removes keys until it can't find
2513  * any higher than new_size
2514  *
2515  * csum items that cross the new i_size are truncated to the new size
2516  * as well.
2517  *
2518  * min_type is the minimum key type to truncate down to.  If set to 0, this
2519  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2520  */
2521 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2522                                         struct btrfs_root *root,
2523                                         struct inode *inode,
2524                                         u64 new_size, u32 min_type)
2525 {
2526         int ret;
2527         struct btrfs_path *path;
2528         struct btrfs_key key;
2529         struct btrfs_key found_key;
2530         u32 found_type = (u8)-1;
2531         struct extent_buffer *leaf;
2532         struct btrfs_file_extent_item *fi;
2533         u64 extent_start = 0;
2534         u64 extent_num_bytes = 0;
2535         u64 item_end = 0;
2536         u64 root_gen = 0;
2537         u64 root_owner = 0;
2538         int found_extent;
2539         int del_item;
2540         int pending_del_nr = 0;
2541         int pending_del_slot = 0;
2542         int extent_type = -1;
2543         int encoding;
2544         u64 mask = root->sectorsize - 1;
2545
2546         if (root->ref_cows)
2547                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2548         path = btrfs_alloc_path();
2549         path->reada = -1;
2550         BUG_ON(!path);
2551
2552         /* FIXME, add redo link to tree so we don't leak on crash */
2553         key.objectid = inode->i_ino;
2554         key.offset = (u64)-1;
2555         key.type = (u8)-1;
2556
2557 search_again:
2558         path->leave_spinning = 1;
2559         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2560         if (ret < 0)
2561                 goto error;
2562
2563         if (ret > 0) {
2564                 /* there are no items in the tree for us to truncate, we're
2565                  * done
2566                  */
2567                 if (path->slots[0] == 0) {
2568                         ret = 0;
2569                         goto error;
2570                 }
2571                 path->slots[0]--;
2572         }
2573
2574         while (1) {
2575                 fi = NULL;
2576                 leaf = path->nodes[0];
2577                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2578                 found_type = btrfs_key_type(&found_key);
2579                 encoding = 0;
2580
2581                 if (found_key.objectid != inode->i_ino)
2582                         break;
2583
2584                 if (found_type < min_type)
2585                         break;
2586
2587                 item_end = found_key.offset;
2588                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2589                         fi = btrfs_item_ptr(leaf, path->slots[0],
2590                                             struct btrfs_file_extent_item);
2591                         extent_type = btrfs_file_extent_type(leaf, fi);
2592                         encoding = btrfs_file_extent_compression(leaf, fi);
2593                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2594                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2595
2596                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2597                                 item_end +=
2598                                     btrfs_file_extent_num_bytes(leaf, fi);
2599                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2600                                 item_end += btrfs_file_extent_inline_len(leaf,
2601                                                                          fi);
2602                         }
2603                         item_end--;
2604                 }
2605                 if (item_end < new_size) {
2606                         if (found_type == BTRFS_DIR_ITEM_KEY)
2607                                 found_type = BTRFS_INODE_ITEM_KEY;
2608                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2609                                 found_type = BTRFS_EXTENT_DATA_KEY;
2610                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2611                                 found_type = BTRFS_XATTR_ITEM_KEY;
2612                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2613                                 found_type = BTRFS_INODE_REF_KEY;
2614                         else if (found_type)
2615                                 found_type--;
2616                         else
2617                                 break;
2618                         btrfs_set_key_type(&key, found_type);
2619                         goto next;
2620                 }
2621                 if (found_key.offset >= new_size)
2622                         del_item = 1;
2623                 else
2624                         del_item = 0;
2625                 found_extent = 0;
2626
2627                 /* FIXME, shrink the extent if the ref count is only 1 */
2628                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2629                         goto delete;
2630
2631                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2632                         u64 num_dec;
2633                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2634                         if (!del_item && !encoding) {
2635                                 u64 orig_num_bytes =
2636                                         btrfs_file_extent_num_bytes(leaf, fi);
2637                                 extent_num_bytes = new_size -
2638                                         found_key.offset + root->sectorsize - 1;
2639                                 extent_num_bytes = extent_num_bytes &
2640                                         ~((u64)root->sectorsize - 1);
2641                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2642                                                          extent_num_bytes);
2643                                 num_dec = (orig_num_bytes -
2644                                            extent_num_bytes);
2645                                 if (root->ref_cows && extent_start != 0)
2646                                         inode_sub_bytes(inode, num_dec);
2647                                 btrfs_mark_buffer_dirty(leaf);
2648                         } else {
2649                                 extent_num_bytes =
2650                                         btrfs_file_extent_disk_num_bytes(leaf,
2651                                                                          fi);
2652                                 /* FIXME blocksize != 4096 */
2653                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2654                                 if (extent_start != 0) {
2655                                         found_extent = 1;
2656                                         if (root->ref_cows)
2657                                                 inode_sub_bytes(inode, num_dec);
2658                                 }
2659                                 root_gen = btrfs_header_generation(leaf);
2660                                 root_owner = btrfs_header_owner(leaf);
2661                         }
2662                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2663                         /*
2664                          * we can't truncate inline items that have had
2665                          * special encodings
2666                          */
2667                         if (!del_item &&
2668                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2669                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2670                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2671                                 u32 size = new_size - found_key.offset;
2672
2673                                 if (root->ref_cows) {
2674                                         inode_sub_bytes(inode, item_end + 1 -
2675                                                         new_size);
2676                                 }
2677                                 size =
2678                                     btrfs_file_extent_calc_inline_size(size);
2679                                 ret = btrfs_truncate_item(trans, root, path,
2680                                                           size, 1);
2681                                 BUG_ON(ret);
2682                         } else if (root->ref_cows) {
2683                                 inode_sub_bytes(inode, item_end + 1 -
2684                                                 found_key.offset);
2685                         }
2686                 }
2687 delete:
2688                 if (del_item) {
2689                         if (!pending_del_nr) {
2690                                 /* no pending yet, add ourselves */
2691                                 pending_del_slot = path->slots[0];
2692                                 pending_del_nr = 1;
2693                         } else if (pending_del_nr &&
2694                                    path->slots[0] + 1 == pending_del_slot) {
2695                                 /* hop on the pending chunk */
2696                                 pending_del_nr++;
2697                                 pending_del_slot = path->slots[0];
2698                         } else {
2699                                 BUG();
2700                         }
2701                 } else {
2702                         break;
2703                 }
2704                 if (found_extent) {
2705                         btrfs_set_path_blocking(path);
2706                         ret = btrfs_free_extent(trans, root, extent_start,
2707                                                 extent_num_bytes,
2708                                                 leaf->start, root_owner,
2709                                                 root_gen, inode->i_ino, 0);
2710                         BUG_ON(ret);
2711                 }
2712 next:
2713                 if (path->slots[0] == 0) {
2714                         if (pending_del_nr)
2715                                 goto del_pending;
2716                         btrfs_release_path(root, path);
2717                         if (found_type == BTRFS_INODE_ITEM_KEY)
2718                                 break;
2719                         goto search_again;
2720                 }
2721
2722                 path->slots[0]--;
2723                 if (pending_del_nr &&
2724                     path->slots[0] + 1 != pending_del_slot) {
2725                         struct btrfs_key debug;
2726 del_pending:
2727                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2728                                               pending_del_slot);
2729                         ret = btrfs_del_items(trans, root, path,
2730                                               pending_del_slot,
2731                                               pending_del_nr);
2732                         BUG_ON(ret);
2733                         pending_del_nr = 0;
2734                         btrfs_release_path(root, path);
2735                         if (found_type == BTRFS_INODE_ITEM_KEY)
2736                                 break;
2737                         goto search_again;
2738                 }
2739         }
2740         ret = 0;
2741 error:
2742         if (pending_del_nr) {
2743                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2744                                       pending_del_nr);
2745         }
2746         btrfs_free_path(path);
2747         inode->i_sb->s_dirt = 1;
2748         return ret;
2749 }
2750
2751 /*
2752  * taken from block_truncate_page, but does cow as it zeros out
2753  * any bytes left in the last page in the file.
2754  */
2755 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2756 {
2757         struct inode *inode = mapping->host;
2758         struct btrfs_root *root = BTRFS_I(inode)->root;
2759         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2760         struct btrfs_ordered_extent *ordered;
2761         char *kaddr;
2762         u32 blocksize = root->sectorsize;
2763         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2764         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2765         struct page *page;
2766         int ret = 0;
2767         u64 page_start;
2768         u64 page_end;
2769
2770         if ((offset & (blocksize - 1)) == 0)
2771                 goto out;
2772
2773         ret = -ENOMEM;
2774 again:
2775         page = grab_cache_page(mapping, index);
2776         if (!page)
2777                 goto out;
2778
2779         page_start = page_offset(page);
2780         page_end = page_start + PAGE_CACHE_SIZE - 1;
2781
2782         if (!PageUptodate(page)) {
2783                 ret = btrfs_readpage(NULL, page);
2784                 lock_page(page);
2785                 if (page->mapping != mapping) {
2786                         unlock_page(page);
2787                         page_cache_release(page);
2788                         goto again;
2789                 }
2790                 if (!PageUptodate(page)) {
2791                         ret = -EIO;
2792                         goto out_unlock;
2793                 }
2794         }
2795         wait_on_page_writeback(page);
2796
2797         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2798         set_page_extent_mapped(page);
2799
2800         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2801         if (ordered) {
2802                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2803                 unlock_page(page);
2804                 page_cache_release(page);
2805                 btrfs_start_ordered_extent(inode, ordered, 1);
2806                 btrfs_put_ordered_extent(ordered);
2807                 goto again;
2808         }
2809
2810         btrfs_set_extent_delalloc(inode, page_start, page_end);
2811         ret = 0;
2812         if (offset != PAGE_CACHE_SIZE) {
2813                 kaddr = kmap(page);
2814                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2815                 flush_dcache_page(page);
2816                 kunmap(page);
2817         }
2818         ClearPageChecked(page);
2819         set_page_dirty(page);
2820         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2821
2822 out_unlock:
2823         unlock_page(page);
2824         page_cache_release(page);
2825 out:
2826         return ret;
2827 }
2828
2829 int btrfs_cont_expand(struct inode *inode, loff_t size)
2830 {
2831         struct btrfs_trans_handle *trans;
2832         struct btrfs_root *root = BTRFS_I(inode)->root;
2833         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2834         struct extent_map *em;
2835         u64 mask = root->sectorsize - 1;
2836         u64 hole_start = (inode->i_size + mask) & ~mask;
2837         u64 block_end = (size + mask) & ~mask;
2838         u64 last_byte;
2839         u64 cur_offset;
2840         u64 hole_size;
2841         int err;
2842
2843         if (size <= hole_start)
2844                 return 0;
2845
2846         err = btrfs_check_metadata_free_space(root);
2847         if (err)
2848                 return err;
2849
2850         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2851
2852         while (1) {
2853                 struct btrfs_ordered_extent *ordered;
2854                 btrfs_wait_ordered_range(inode, hole_start,
2855                                          block_end - hole_start);
2856                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2857                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2858                 if (!ordered)
2859                         break;
2860                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2861                 btrfs_put_ordered_extent(ordered);
2862         }
2863
2864         trans = btrfs_start_transaction(root, 1);
2865         btrfs_set_trans_block_group(trans, inode);
2866
2867         cur_offset = hole_start;
2868         while (1) {
2869                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2870                                 block_end - cur_offset, 0);
2871                 BUG_ON(IS_ERR(em) || !em);
2872                 last_byte = min(extent_map_end(em), block_end);
2873                 last_byte = (last_byte + mask) & ~mask;
2874                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2875                         u64 hint_byte = 0;
2876                         hole_size = last_byte - cur_offset;
2877                         err = btrfs_drop_extents(trans, root, inode,
2878                                                  cur_offset,
2879                                                  cur_offset + hole_size,
2880                                                  cur_offset, &hint_byte);
2881                         if (err)
2882                                 break;
2883                         err = btrfs_insert_file_extent(trans, root,
2884                                         inode->i_ino, cur_offset, 0,
2885                                         0, hole_size, 0, hole_size,
2886                                         0, 0, 0);
2887                         btrfs_drop_extent_cache(inode, hole_start,
2888                                         last_byte - 1, 0);
2889                 }
2890                 free_extent_map(em);
2891                 cur_offset = last_byte;
2892                 if (err || cur_offset >= block_end)
2893                         break;
2894         }
2895
2896         btrfs_end_transaction(trans, root);
2897         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2898         return err;
2899 }
2900
2901 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2902 {
2903         struct inode *inode = dentry->d_inode;
2904         int err;
2905
2906         err = inode_change_ok(inode, attr);
2907         if (err)
2908                 return err;
2909
2910         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2911                 if (attr->ia_size > inode->i_size) {
2912                         err = btrfs_cont_expand(inode, attr->ia_size);
2913                         if (err)
2914                                 return err;
2915                 } else if (inode->i_size > 0 &&
2916                            attr->ia_size == 0) {
2917
2918                         /* we're truncating a file that used to have good
2919                          * data down to zero.  Make sure it gets into
2920                          * the ordered flush list so that any new writes
2921                          * get down to disk quickly.
2922                          */
2923                         BTRFS_I(inode)->ordered_data_close = 1;
2924                 }
2925         }
2926
2927         err = inode_setattr(inode, attr);
2928
2929         if (!err && ((attr->ia_valid & ATTR_MODE)))
2930                 err = btrfs_acl_chmod(inode);
2931         return err;
2932 }
2933
2934 void btrfs_delete_inode(struct inode *inode)
2935 {
2936         struct btrfs_trans_handle *trans;
2937         struct btrfs_root *root = BTRFS_I(inode)->root;
2938         unsigned long nr;
2939         int ret;
2940
2941         truncate_inode_pages(&inode->i_data, 0);
2942         if (is_bad_inode(inode)) {
2943                 btrfs_orphan_del(NULL, inode);
2944                 goto no_delete;
2945         }
2946         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2947
2948         btrfs_i_size_write(inode, 0);
2949         trans = btrfs_join_transaction(root, 1);
2950
2951         btrfs_set_trans_block_group(trans, inode);
2952         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2953         if (ret) {
2954                 btrfs_orphan_del(NULL, inode);
2955                 goto no_delete_lock;
2956         }
2957
2958         btrfs_orphan_del(trans, inode);
2959
2960         nr = trans->blocks_used;
2961         clear_inode(inode);
2962
2963         btrfs_end_transaction(trans, root);
2964         btrfs_btree_balance_dirty(root, nr);
2965         return;
2966
2967 no_delete_lock:
2968         nr = trans->blocks_used;
2969         btrfs_end_transaction(trans, root);
2970         btrfs_btree_balance_dirty(root, nr);
2971 no_delete:
2972         clear_inode(inode);
2973 }
2974
2975 /*
2976  * this returns the key found in the dir entry in the location pointer.
2977  * If no dir entries were found, location->objectid is 0.
2978  */
2979 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2980                                struct btrfs_key *location)
2981 {
2982         const char *name = dentry->d_name.name;
2983         int namelen = dentry->d_name.len;
2984         struct btrfs_dir_item *di;
2985         struct btrfs_path *path;
2986         struct btrfs_root *root = BTRFS_I(dir)->root;
2987         int ret = 0;
2988
2989         path = btrfs_alloc_path();
2990         BUG_ON(!path);
2991
2992         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2993                                     namelen, 0);
2994         if (IS_ERR(di))
2995                 ret = PTR_ERR(di);
2996
2997         if (!di || IS_ERR(di))
2998                 goto out_err;
2999
3000         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3001 out:
3002         btrfs_free_path(path);
3003         return ret;
3004 out_err:
3005         location->objectid = 0;
3006         goto out;
3007 }
3008
3009 /*
3010  * when we hit a tree root in a directory, the btrfs part of the inode
3011  * needs to be changed to reflect the root directory of the tree root.  This
3012  * is kind of like crossing a mount point.
3013  */
3014 static int fixup_tree_root_location(struct btrfs_root *root,
3015                              struct btrfs_key *location,
3016                              struct btrfs_root **sub_root,
3017                              struct dentry *dentry)
3018 {
3019         struct btrfs_root_item *ri;
3020
3021         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3022                 return 0;
3023         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3024                 return 0;
3025
3026         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3027                                         dentry->d_name.name,
3028                                         dentry->d_name.len);
3029         if (IS_ERR(*sub_root))
3030                 return PTR_ERR(*sub_root);
3031
3032         ri = &(*sub_root)->root_item;
3033         location->objectid = btrfs_root_dirid(ri);
3034         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3035         location->offset = 0;
3036
3037         return 0;
3038 }
3039
3040 static noinline void init_btrfs_i(struct inode *inode)
3041 {
3042         struct btrfs_inode *bi = BTRFS_I(inode);
3043
3044         bi->i_acl = NULL;
3045         bi->i_default_acl = NULL;
3046
3047         bi->generation = 0;
3048         bi->sequence = 0;
3049         bi->last_trans = 0;
3050         bi->logged_trans = 0;
3051         bi->delalloc_bytes = 0;
3052         bi->reserved_bytes = 0;
3053         bi->disk_i_size = 0;
3054         bi->flags = 0;
3055         bi->index_cnt = (u64)-1;
3056         bi->last_unlink_trans = 0;
3057         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3058         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3059                              inode->i_mapping, GFP_NOFS);
3060         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3061                              inode->i_mapping, GFP_NOFS);
3062         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3063         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3064         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3065         mutex_init(&BTRFS_I(inode)->extent_mutex);
3066         mutex_init(&BTRFS_I(inode)->log_mutex);
3067 }
3068
3069 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3070 {
3071         struct btrfs_iget_args *args = p;
3072         inode->i_ino = args->ino;
3073         init_btrfs_i(inode);
3074         BTRFS_I(inode)->root = args->root;
3075         btrfs_set_inode_space_info(args->root, inode);
3076         return 0;
3077 }
3078
3079 static int btrfs_find_actor(struct inode *inode, void *opaque)
3080 {
3081         struct btrfs_iget_args *args = opaque;
3082         return args->ino == inode->i_ino &&
3083                 args->root == BTRFS_I(inode)->root;
3084 }
3085
3086 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3087                             struct btrfs_root *root, int wait)
3088 {
3089         struct inode *inode;
3090         struct btrfs_iget_args args;
3091         args.ino = objectid;
3092         args.root = root;
3093
3094         if (wait) {
3095                 inode = ilookup5(s, objectid, btrfs_find_actor,
3096                                  (void *)&args);
3097         } else {
3098                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3099                                         (void *)&args);
3100         }
3101         return inode;
3102 }
3103
3104 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3105                                 struct btrfs_root *root)
3106 {
3107         struct inode *inode;
3108         struct btrfs_iget_args args;
3109         args.ino = objectid;
3110         args.root = root;
3111
3112         inode = iget5_locked(s, objectid, btrfs_find_actor,
3113                              btrfs_init_locked_inode,
3114                              (void *)&args);
3115         return inode;
3116 }
3117
3118 /* Get an inode object given its location and corresponding root.
3119  * Returns in *is_new if the inode was read from disk
3120  */
3121 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,