Btrfs: fix balancing oops when invalidate_inode_pages2 returns EBUSY
[pandora-kernel.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static struct inode_operations btrfs_dir_inode_operations;
60 static struct inode_operations btrfs_symlink_inode_operations;
61 static struct inode_operations btrfs_dir_ro_inode_operations;
62 static struct inode_operations btrfs_special_inode_operations;
63 static struct inode_operations btrfs_file_inode_operations;
64 static struct address_space_operations btrfs_aops;
65 static struct address_space_operations btrfs_symlink_aops;
66 static struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
93 {
94         int err;
95
96         err = btrfs_init_acl(inode, dir);
97         if (!err)
98                 err = btrfs_xattr_security_init(inode, dir);
99         return err;
100 }
101
102 /*
103  * this does all the hard work for inserting an inline extent into
104  * the btree.  The caller should have done a btrfs_drop_extents so that
105  * no overlapping inline items exist in the btree
106  */
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108                                 struct btrfs_root *root, struct inode *inode,
109                                 u64 start, size_t size, size_t compressed_size,
110                                 struct page **compressed_pages)
111 {
112         struct btrfs_key key;
113         struct btrfs_path *path;
114         struct extent_buffer *leaf;
115         struct page *page = NULL;
116         char *kaddr;
117         unsigned long ptr;
118         struct btrfs_file_extent_item *ei;
119         int err = 0;
120         int ret;
121         size_t cur_size = size;
122         size_t datasize;
123         unsigned long offset;
124         int use_compress = 0;
125
126         if (compressed_size && compressed_pages) {
127                 use_compress = 1;
128                 cur_size = compressed_size;
129         }
130
131         path = btrfs_alloc_path();
132         if (!path)
133                 return -ENOMEM;
134
135         path->leave_spinning = 1;
136         btrfs_set_trans_block_group(trans, inode);
137
138         key.objectid = inode->i_ino;
139         key.offset = start;
140         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141         datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143         inode_add_bytes(inode, size);
144         ret = btrfs_insert_empty_item(trans, root, path, &key,
145                                       datasize);
146         BUG_ON(ret);
147         if (ret) {
148                 err = ret;
149                 goto fail;
150         }
151         leaf = path->nodes[0];
152         ei = btrfs_item_ptr(leaf, path->slots[0],
153                             struct btrfs_file_extent_item);
154         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156         btrfs_set_file_extent_encryption(leaf, ei, 0);
157         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159         ptr = btrfs_file_extent_inline_start(ei);
160
161         if (use_compress) {
162                 struct page *cpage;
163                 int i = 0;
164                 while (compressed_size > 0) {
165                         cpage = compressed_pages[i];
166                         cur_size = min_t(unsigned long, compressed_size,
167                                        PAGE_CACHE_SIZE);
168
169                         kaddr = kmap_atomic(cpage, KM_USER0);
170                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
171                         kunmap_atomic(kaddr, KM_USER0);
172
173                         i++;
174                         ptr += cur_size;
175                         compressed_size -= cur_size;
176                 }
177                 btrfs_set_file_extent_compression(leaf, ei,
178                                                   BTRFS_COMPRESS_ZLIB);
179         } else {
180                 page = find_get_page(inode->i_mapping,
181                                      start >> PAGE_CACHE_SHIFT);
182                 btrfs_set_file_extent_compression(leaf, ei, 0);
183                 kaddr = kmap_atomic(page, KM_USER0);
184                 offset = start & (PAGE_CACHE_SIZE - 1);
185                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186                 kunmap_atomic(kaddr, KM_USER0);
187                 page_cache_release(page);
188         }
189         btrfs_mark_buffer_dirty(leaf);
190         btrfs_free_path(path);
191
192         BTRFS_I(inode)->disk_i_size = inode->i_size;
193         btrfs_update_inode(trans, root, inode);
194         return 0;
195 fail:
196         btrfs_free_path(path);
197         return err;
198 }
199
200
201 /*
202  * conditionally insert an inline extent into the file.  This
203  * does the checks required to make sure the data is small enough
204  * to fit as an inline extent.
205  */
206 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
207                                  struct btrfs_root *root,
208                                  struct inode *inode, u64 start, u64 end,
209                                  size_t compressed_size,
210                                  struct page **compressed_pages)
211 {
212         u64 isize = i_size_read(inode);
213         u64 actual_end = min(end + 1, isize);
214         u64 inline_len = actual_end - start;
215         u64 aligned_end = (end + root->sectorsize - 1) &
216                         ~((u64)root->sectorsize - 1);
217         u64 hint_byte;
218         u64 data_len = inline_len;
219         int ret;
220
221         if (compressed_size)
222                 data_len = compressed_size;
223
224         if (start > 0 ||
225             actual_end >= PAGE_CACHE_SIZE ||
226             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227             (!compressed_size &&
228             (actual_end & (root->sectorsize - 1)) == 0) ||
229             end + 1 < isize ||
230             data_len > root->fs_info->max_inline) {
231                 return 1;
232         }
233
234         ret = btrfs_drop_extents(trans, root, inode, start,
235                                  aligned_end, aligned_end, start, &hint_byte);
236         BUG_ON(ret);
237
238         if (isize > actual_end)
239                 inline_len = min_t(u64, isize, actual_end);
240         ret = insert_inline_extent(trans, root, inode, start,
241                                    inline_len, compressed_size,
242                                    compressed_pages);
243         BUG_ON(ret);
244         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
245         return 0;
246 }
247
248 struct async_extent {
249         u64 start;
250         u64 ram_size;
251         u64 compressed_size;
252         struct page **pages;
253         unsigned long nr_pages;
254         struct list_head list;
255 };
256
257 struct async_cow {
258         struct inode *inode;
259         struct btrfs_root *root;
260         struct page *locked_page;
261         u64 start;
262         u64 end;
263         struct list_head extents;
264         struct btrfs_work work;
265 };
266
267 static noinline int add_async_extent(struct async_cow *cow,
268                                      u64 start, u64 ram_size,
269                                      u64 compressed_size,
270                                      struct page **pages,
271                                      unsigned long nr_pages)
272 {
273         struct async_extent *async_extent;
274
275         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276         async_extent->start = start;
277         async_extent->ram_size = ram_size;
278         async_extent->compressed_size = compressed_size;
279         async_extent->pages = pages;
280         async_extent->nr_pages = nr_pages;
281         list_add_tail(&async_extent->list, &cow->extents);
282         return 0;
283 }
284
285 /*
286  * we create compressed extents in two phases.  The first
287  * phase compresses a range of pages that have already been
288  * locked (both pages and state bits are locked).
289  *
290  * This is done inside an ordered work queue, and the compression
291  * is spread across many cpus.  The actual IO submission is step
292  * two, and the ordered work queue takes care of making sure that
293  * happens in the same order things were put onto the queue by
294  * writepages and friends.
295  *
296  * If this code finds it can't get good compression, it puts an
297  * entry onto the work queue to write the uncompressed bytes.  This
298  * makes sure that both compressed inodes and uncompressed inodes
299  * are written in the same order that pdflush sent them down.
300  */
301 static noinline int compress_file_range(struct inode *inode,
302                                         struct page *locked_page,
303                                         u64 start, u64 end,
304                                         struct async_cow *async_cow,
305                                         int *num_added)
306 {
307         struct btrfs_root *root = BTRFS_I(inode)->root;
308         struct btrfs_trans_handle *trans;
309         u64 num_bytes;
310         u64 orig_start;
311         u64 disk_num_bytes;
312         u64 blocksize = root->sectorsize;
313         u64 actual_end;
314         u64 isize = i_size_read(inode);
315         int ret = 0;
316         struct page **pages = NULL;
317         unsigned long nr_pages;
318         unsigned long nr_pages_ret = 0;
319         unsigned long total_compressed = 0;
320         unsigned long total_in = 0;
321         unsigned long max_compressed = 128 * 1024;
322         unsigned long max_uncompressed = 128 * 1024;
323         int i;
324         int will_compress;
325
326         orig_start = start;
327
328         actual_end = min_t(u64, isize, end + 1);
329 again:
330         will_compress = 0;
331         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
333
334         /*
335          * we don't want to send crud past the end of i_size through
336          * compression, that's just a waste of CPU time.  So, if the
337          * end of the file is before the start of our current
338          * requested range of bytes, we bail out to the uncompressed
339          * cleanup code that can deal with all of this.
340          *
341          * It isn't really the fastest way to fix things, but this is a
342          * very uncommon corner.
343          */
344         if (actual_end <= start)
345                 goto cleanup_and_bail_uncompressed;
346
347         total_compressed = actual_end - start;
348
349         /* we want to make sure that amount of ram required to uncompress
350          * an extent is reasonable, so we limit the total size in ram
351          * of a compressed extent to 128k.  This is a crucial number
352          * because it also controls how easily we can spread reads across
353          * cpus for decompression.
354          *
355          * We also want to make sure the amount of IO required to do
356          * a random read is reasonably small, so we limit the size of
357          * a compressed extent to 128k.
358          */
359         total_compressed = min(total_compressed, max_uncompressed);
360         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361         num_bytes = max(blocksize,  num_bytes);
362         disk_num_bytes = num_bytes;
363         total_in = 0;
364         ret = 0;
365
366         /*
367          * we do compression for mount -o compress and when the
368          * inode has not been flagged as nocompress.  This flag can
369          * change at any time if we discover bad compression ratios.
370          */
371         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
372             btrfs_test_opt(root, COMPRESS)) {
373                 WARN_ON(pages);
374                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
375
376                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377                                                 total_compressed, pages,
378                                                 nr_pages, &nr_pages_ret,
379                                                 &total_in,
380                                                 &total_compressed,
381                                                 max_compressed);
382
383                 if (!ret) {
384                         unsigned long offset = total_compressed &
385                                 (PAGE_CACHE_SIZE - 1);
386                         struct page *page = pages[nr_pages_ret - 1];
387                         char *kaddr;
388
389                         /* zero the tail end of the last page, we might be
390                          * sending it down to disk
391                          */
392                         if (offset) {
393                                 kaddr = kmap_atomic(page, KM_USER0);
394                                 memset(kaddr + offset, 0,
395                                        PAGE_CACHE_SIZE - offset);
396                                 kunmap_atomic(kaddr, KM_USER0);
397                         }
398                         will_compress = 1;
399                 }
400         }
401         if (start == 0) {
402                 trans = btrfs_join_transaction(root, 1);
403                 BUG_ON(!trans);
404                 btrfs_set_trans_block_group(trans, inode);
405
406                 /* lets try to make an inline extent */
407                 if (ret || total_in < (actual_end - start)) {
408                         /* we didn't compress the entire range, try
409                          * to make an uncompressed inline extent.
410                          */
411                         ret = cow_file_range_inline(trans, root, inode,
412                                                     start, end, 0, NULL);
413                 } else {
414                         /* try making a compressed inline extent */
415                         ret = cow_file_range_inline(trans, root, inode,
416                                                     start, end,
417                                                     total_compressed, pages);
418                 }
419                 btrfs_end_transaction(trans, root);
420                 if (ret == 0) {
421                         /*
422                          * inline extent creation worked, we don't need
423                          * to create any more async work items.  Unlock
424                          * and free up our temp pages.
425                          */
426                         extent_clear_unlock_delalloc(inode,
427                                                      &BTRFS_I(inode)->io_tree,
428                                                      start, end, NULL, 1, 0,
429                                                      0, 1, 1, 1);
430                         ret = 0;
431                         goto free_pages_out;
432                 }
433         }
434
435         if (will_compress) {
436                 /*
437                  * we aren't doing an inline extent round the compressed size
438                  * up to a block size boundary so the allocator does sane
439                  * things
440                  */
441                 total_compressed = (total_compressed + blocksize - 1) &
442                         ~(blocksize - 1);
443
444                 /*
445                  * one last check to make sure the compression is really a
446                  * win, compare the page count read with the blocks on disk
447                  */
448                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
449                         ~(PAGE_CACHE_SIZE - 1);
450                 if (total_compressed >= total_in) {
451                         will_compress = 0;
452                 } else {
453                         disk_num_bytes = total_compressed;
454                         num_bytes = total_in;
455                 }
456         }
457         if (!will_compress && pages) {
458                 /*
459                  * the compression code ran but failed to make things smaller,
460                  * free any pages it allocated and our page pointer array
461                  */
462                 for (i = 0; i < nr_pages_ret; i++) {
463                         WARN_ON(pages[i]->mapping);
464                         page_cache_release(pages[i]);
465                 }
466                 kfree(pages);
467                 pages = NULL;
468                 total_compressed = 0;
469                 nr_pages_ret = 0;
470
471                 /* flag the file so we don't compress in the future */
472                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
473         }
474         if (will_compress) {
475                 *num_added += 1;
476
477                 /* the async work queues will take care of doing actual
478                  * allocation on disk for these compressed pages,
479                  * and will submit them to the elevator.
480                  */
481                 add_async_extent(async_cow, start, num_bytes,
482                                  total_compressed, pages, nr_pages_ret);
483
484                 if (start + num_bytes < end && start + num_bytes < actual_end) {
485                         start += num_bytes;
486                         pages = NULL;
487                         cond_resched();
488                         goto again;
489                 }
490         } else {
491 cleanup_and_bail_uncompressed:
492                 /*
493                  * No compression, but we still need to write the pages in
494                  * the file we've been given so far.  redirty the locked
495                  * page if it corresponds to our extent and set things up
496                  * for the async work queue to run cow_file_range to do
497                  * the normal delalloc dance
498                  */
499                 if (page_offset(locked_page) >= start &&
500                     page_offset(locked_page) <= end) {
501                         __set_page_dirty_nobuffers(locked_page);
502                         /* unlocked later on in the async handlers */
503                 }
504                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
505                 *num_added += 1;
506         }
507
508 out:
509         return 0;
510
511 free_pages_out:
512         for (i = 0; i < nr_pages_ret; i++) {
513                 WARN_ON(pages[i]->mapping);
514                 page_cache_release(pages[i]);
515         }
516         kfree(pages);
517
518         goto out;
519 }
520
521 /*
522  * phase two of compressed writeback.  This is the ordered portion
523  * of the code, which only gets called in the order the work was
524  * queued.  We walk all the async extents created by compress_file_range
525  * and send them down to the disk.
526  */
527 static noinline int submit_compressed_extents(struct inode *inode,
528                                               struct async_cow *async_cow)
529 {
530         struct async_extent *async_extent;
531         u64 alloc_hint = 0;
532         struct btrfs_trans_handle *trans;
533         struct btrfs_key ins;
534         struct extent_map *em;
535         struct btrfs_root *root = BTRFS_I(inode)->root;
536         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537         struct extent_io_tree *io_tree;
538         int ret;
539
540         if (list_empty(&async_cow->extents))
541                 return 0;
542
543         trans = btrfs_join_transaction(root, 1);
544
545         while (!list_empty(&async_cow->extents)) {
546                 async_extent = list_entry(async_cow->extents.next,
547                                           struct async_extent, list);
548                 list_del(&async_extent->list);
549
550                 io_tree = &BTRFS_I(inode)->io_tree;
551
552                 /* did the compression code fall back to uncompressed IO? */
553                 if (!async_extent->pages) {
554                         int page_started = 0;
555                         unsigned long nr_written = 0;
556
557                         lock_extent(io_tree, async_extent->start,
558                                     async_extent->start +
559                                     async_extent->ram_size - 1, GFP_NOFS);
560
561                         /* allocate blocks */
562                         cow_file_range(inode, async_cow->locked_page,
563                                        async_extent->start,
564                                        async_extent->start +
565                                        async_extent->ram_size - 1,
566                                        &page_started, &nr_written, 0);
567
568                         /*
569                          * if page_started, cow_file_range inserted an
570                          * inline extent and took care of all the unlocking
571                          * and IO for us.  Otherwise, we need to submit
572                          * all those pages down to the drive.
573                          */
574                         if (!page_started)
575                                 extent_write_locked_range(io_tree,
576                                                   inode, async_extent->start,
577                                                   async_extent->start +
578                                                   async_extent->ram_size - 1,
579                                                   btrfs_get_extent,
580                                                   WB_SYNC_ALL);
581                         kfree(async_extent);
582                         cond_resched();
583                         continue;
584                 }
585
586                 lock_extent(io_tree, async_extent->start,
587                             async_extent->start + async_extent->ram_size - 1,
588                             GFP_NOFS);
589                 /*
590                  * here we're doing allocation and writeback of the
591                  * compressed pages
592                  */
593                 btrfs_drop_extent_cache(inode, async_extent->start,
594                                         async_extent->start +
595                                         async_extent->ram_size - 1, 0);
596
597                 ret = btrfs_reserve_extent(trans, root,
598                                            async_extent->compressed_size,
599                                            async_extent->compressed_size,
600                                            0, alloc_hint,
601                                            (u64)-1, &ins, 1);
602                 BUG_ON(ret);
603                 em = alloc_extent_map(GFP_NOFS);
604                 em->start = async_extent->start;
605                 em->len = async_extent->ram_size;
606                 em->orig_start = em->start;
607
608                 em->block_start = ins.objectid;
609                 em->block_len = ins.offset;
610                 em->bdev = root->fs_info->fs_devices->latest_bdev;
611                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
612                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
613
614                 while (1) {
615                         spin_lock(&em_tree->lock);
616                         ret = add_extent_mapping(em_tree, em);
617                         spin_unlock(&em_tree->lock);
618                         if (ret != -EEXIST) {
619                                 free_extent_map(em);
620                                 break;
621                         }
622                         btrfs_drop_extent_cache(inode, async_extent->start,
623                                                 async_extent->start +
624                                                 async_extent->ram_size - 1, 0);
625                 }
626
627                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
628                                                ins.objectid,
629                                                async_extent->ram_size,
630                                                ins.offset,
631                                                BTRFS_ORDERED_COMPRESSED);
632                 BUG_ON(ret);
633
634                 btrfs_end_transaction(trans, root);
635
636                 /*
637                  * clear dirty, set writeback and unlock the pages.
638                  */
639                 extent_clear_unlock_delalloc(inode,
640                                              &BTRFS_I(inode)->io_tree,
641                                              async_extent->start,
642                                              async_extent->start +
643                                              async_extent->ram_size - 1,
644                                              NULL, 1, 1, 0, 1, 1, 0);
645
646                 ret = btrfs_submit_compressed_write(inode,
647                                     async_extent->start,
648                                     async_extent->ram_size,
649                                     ins.objectid,
650                                     ins.offset, async_extent->pages,
651                                     async_extent->nr_pages);
652
653                 BUG_ON(ret);
654                 trans = btrfs_join_transaction(root, 1);
655                 alloc_hint = ins.objectid + ins.offset;
656                 kfree(async_extent);
657                 cond_resched();
658         }
659
660         btrfs_end_transaction(trans, root);
661         return 0;
662 }
663
664 /*
665  * when extent_io.c finds a delayed allocation range in the file,
666  * the call backs end up in this code.  The basic idea is to
667  * allocate extents on disk for the range, and create ordered data structs
668  * in ram to track those extents.
669  *
670  * locked_page is the page that writepage had locked already.  We use
671  * it to make sure we don't do extra locks or unlocks.
672  *
673  * *page_started is set to one if we unlock locked_page and do everything
674  * required to start IO on it.  It may be clean and already done with
675  * IO when we return.
676  */
677 static noinline int cow_file_range(struct inode *inode,
678                                    struct page *locked_page,
679                                    u64 start, u64 end, int *page_started,
680                                    unsigned long *nr_written,
681                                    int unlock)
682 {
683         struct btrfs_root *root = BTRFS_I(inode)->root;
684         struct btrfs_trans_handle *trans;
685         u64 alloc_hint = 0;
686         u64 num_bytes;
687         unsigned long ram_size;
688         u64 disk_num_bytes;
689         u64 cur_alloc_size;
690         u64 blocksize = root->sectorsize;
691         u64 actual_end;
692         u64 isize = i_size_read(inode);
693         struct btrfs_key ins;
694         struct extent_map *em;
695         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
696         int ret = 0;
697
698         trans = btrfs_join_transaction(root, 1);
699         BUG_ON(!trans);
700         btrfs_set_trans_block_group(trans, inode);
701
702         actual_end = min_t(u64, isize, end + 1);
703
704         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
705         num_bytes = max(blocksize,  num_bytes);
706         disk_num_bytes = num_bytes;
707         ret = 0;
708
709         if (start == 0) {
710                 /* lets try to make an inline extent */
711                 ret = cow_file_range_inline(trans, root, inode,
712                                             start, end, 0, NULL);
713                 if (ret == 0) {
714                         extent_clear_unlock_delalloc(inode,
715                                                      &BTRFS_I(inode)->io_tree,
716                                                      start, end, NULL, 1, 1,
717                                                      1, 1, 1, 1);
718                         *nr_written = *nr_written +
719                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
720                         *page_started = 1;
721                         ret = 0;
722                         goto out;
723                 }
724         }
725
726         BUG_ON(disk_num_bytes >
727                btrfs_super_total_bytes(&root->fs_info->super_copy));
728
729         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
730
731         while (disk_num_bytes > 0) {
732                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
733                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
734                                            root->sectorsize, 0, alloc_hint,
735                                            (u64)-1, &ins, 1);
736                 BUG_ON(ret);
737
738                 em = alloc_extent_map(GFP_NOFS);
739                 em->start = start;
740                 em->orig_start = em->start;
741
742                 ram_size = ins.offset;
743                 em->len = ins.offset;
744
745                 em->block_start = ins.objectid;
746                 em->block_len = ins.offset;
747                 em->bdev = root->fs_info->fs_devices->latest_bdev;
748                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
749
750                 while (1) {
751                         spin_lock(&em_tree->lock);
752                         ret = add_extent_mapping(em_tree, em);
753                         spin_unlock(&em_tree->lock);
754                         if (ret != -EEXIST) {
755                                 free_extent_map(em);
756                                 break;
757                         }
758                         btrfs_drop_extent_cache(inode, start,
759                                                 start + ram_size - 1, 0);
760                 }
761
762                 cur_alloc_size = ins.offset;
763                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
764                                                ram_size, cur_alloc_size, 0);
765                 BUG_ON(ret);
766
767                 if (root->root_key.objectid ==
768                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
769                         ret = btrfs_reloc_clone_csums(inode, start,
770                                                       cur_alloc_size);
771                         BUG_ON(ret);
772                 }
773
774                 if (disk_num_bytes < cur_alloc_size)
775                         break;
776
777                 /* we're not doing compressed IO, don't unlock the first
778                  * page (which the caller expects to stay locked), don't
779                  * clear any dirty bits and don't set any writeback bits
780                  */
781                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
782                                              start, start + ram_size - 1,
783                                              locked_page, unlock, 1,
784                                              1, 0, 0, 0);
785                 disk_num_bytes -= cur_alloc_size;
786                 num_bytes -= cur_alloc_size;
787                 alloc_hint = ins.objectid + ins.offset;
788                 start += cur_alloc_size;
789         }
790 out:
791         ret = 0;
792         btrfs_end_transaction(trans, root);
793
794         return ret;
795 }
796
797 /*
798  * work queue call back to started compression on a file and pages
799  */
800 static noinline void async_cow_start(struct btrfs_work *work)
801 {
802         struct async_cow *async_cow;
803         int num_added = 0;
804         async_cow = container_of(work, struct async_cow, work);
805
806         compress_file_range(async_cow->inode, async_cow->locked_page,
807                             async_cow->start, async_cow->end, async_cow,
808                             &num_added);
809         if (num_added == 0)
810                 async_cow->inode = NULL;
811 }
812
813 /*
814  * work queue call back to submit previously compressed pages
815  */
816 static noinline void async_cow_submit(struct btrfs_work *work)
817 {
818         struct async_cow *async_cow;
819         struct btrfs_root *root;
820         unsigned long nr_pages;
821
822         async_cow = container_of(work, struct async_cow, work);
823
824         root = async_cow->root;
825         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
826                 PAGE_CACHE_SHIFT;
827
828         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
829
830         if (atomic_read(&root->fs_info->async_delalloc_pages) <
831             5 * 1042 * 1024 &&
832             waitqueue_active(&root->fs_info->async_submit_wait))
833                 wake_up(&root->fs_info->async_submit_wait);
834
835         if (async_cow->inode)
836                 submit_compressed_extents(async_cow->inode, async_cow);
837 }
838
839 static noinline void async_cow_free(struct btrfs_work *work)
840 {
841         struct async_cow *async_cow;
842         async_cow = container_of(work, struct async_cow, work);
843         kfree(async_cow);
844 }
845
846 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
847                                 u64 start, u64 end, int *page_started,
848                                 unsigned long *nr_written)
849 {
850         struct async_cow *async_cow;
851         struct btrfs_root *root = BTRFS_I(inode)->root;
852         unsigned long nr_pages;
853         u64 cur_end;
854         int limit = 10 * 1024 * 1042;
855
856         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
857                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
858         while (start < end) {
859                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
860                 async_cow->inode = inode;
861                 async_cow->root = root;
862                 async_cow->locked_page = locked_page;
863                 async_cow->start = start;
864
865                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
866                         cur_end = end;
867                 else
868                         cur_end = min(end, start + 512 * 1024 - 1);
869
870                 async_cow->end = cur_end;
871                 INIT_LIST_HEAD(&async_cow->extents);
872
873                 async_cow->work.func = async_cow_start;
874                 async_cow->work.ordered_func = async_cow_submit;
875                 async_cow->work.ordered_free = async_cow_free;
876                 async_cow->work.flags = 0;
877
878                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
879                         PAGE_CACHE_SHIFT;
880                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
881
882                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
883                                    &async_cow->work);
884
885                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
886                         wait_event(root->fs_info->async_submit_wait,
887                            (atomic_read(&root->fs_info->async_delalloc_pages) <
888                             limit));
889                 }
890
891                 while (atomic_read(&root->fs_info->async_submit_draining) &&
892                       atomic_read(&root->fs_info->async_delalloc_pages)) {
893                         wait_event(root->fs_info->async_submit_wait,
894                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
895                            0));
896                 }
897
898                 *nr_written += nr_pages;
899                 start = cur_end + 1;
900         }
901         *page_started = 1;
902         return 0;
903 }
904
905 static noinline int csum_exist_in_range(struct btrfs_root *root,
906                                         u64 bytenr, u64 num_bytes)
907 {
908         int ret;
909         struct btrfs_ordered_sum *sums;
910         LIST_HEAD(list);
911
912         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
913                                        bytenr + num_bytes - 1, &list);
914         if (ret == 0 && list_empty(&list))
915                 return 0;
916
917         while (!list_empty(&list)) {
918                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
919                 list_del(&sums->list);
920                 kfree(sums);
921         }
922         return 1;
923 }
924
925 /*
926  * when nowcow writeback call back.  This checks for snapshots or COW copies
927  * of the extents that exist in the file, and COWs the file as required.
928  *
929  * If no cow copies or snapshots exist, we write directly to the existing
930  * blocks on disk
931  */
932 static noinline int run_delalloc_nocow(struct inode *inode,
933                                        struct page *locked_page,
934                               u64 start, u64 end, int *page_started, int force,
935                               unsigned long *nr_written)
936 {
937         struct btrfs_root *root = BTRFS_I(inode)->root;
938         struct btrfs_trans_handle *trans;
939         struct extent_buffer *leaf;
940         struct btrfs_path *path;
941         struct btrfs_file_extent_item *fi;
942         struct btrfs_key found_key;
943         u64 cow_start;
944         u64 cur_offset;
945         u64 extent_end;
946         u64 extent_offset;
947         u64 disk_bytenr;
948         u64 num_bytes;
949         int extent_type;
950         int ret;
951         int type;
952         int nocow;
953         int check_prev = 1;
954
955         path = btrfs_alloc_path();
956         BUG_ON(!path);
957         trans = btrfs_join_transaction(root, 1);
958         BUG_ON(!trans);
959
960         cow_start = (u64)-1;
961         cur_offset = start;
962         while (1) {
963                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
964                                                cur_offset, 0);
965                 BUG_ON(ret < 0);
966                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
967                         leaf = path->nodes[0];
968                         btrfs_item_key_to_cpu(leaf, &found_key,
969                                               path->slots[0] - 1);
970                         if (found_key.objectid == inode->i_ino &&
971                             found_key.type == BTRFS_EXTENT_DATA_KEY)
972                                 path->slots[0]--;
973                 }
974                 check_prev = 0;
975 next_slot:
976                 leaf = path->nodes[0];
977                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
978                         ret = btrfs_next_leaf(root, path);
979                         if (ret < 0)
980                                 BUG_ON(1);
981                         if (ret > 0)
982                                 break;
983                         leaf = path->nodes[0];
984                 }
985
986                 nocow = 0;
987                 disk_bytenr = 0;
988                 num_bytes = 0;
989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
990
991                 if (found_key.objectid > inode->i_ino ||
992                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
993                     found_key.offset > end)
994                         break;
995
996                 if (found_key.offset > cur_offset) {
997                         extent_end = found_key.offset;
998                         goto out_check;
999                 }
1000
1001                 fi = btrfs_item_ptr(leaf, path->slots[0],
1002                                     struct btrfs_file_extent_item);
1003                 extent_type = btrfs_file_extent_type(leaf, fi);
1004
1005                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1006                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1007                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1008                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1009                         extent_end = found_key.offset +
1010                                 btrfs_file_extent_num_bytes(leaf, fi);
1011                         if (extent_end <= start) {
1012                                 path->slots[0]++;
1013                                 goto next_slot;
1014                         }
1015                         if (disk_bytenr == 0)
1016                                 goto out_check;
1017                         if (btrfs_file_extent_compression(leaf, fi) ||
1018                             btrfs_file_extent_encryption(leaf, fi) ||
1019                             btrfs_file_extent_other_encoding(leaf, fi))
1020                                 goto out_check;
1021                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022                                 goto out_check;
1023                         if (btrfs_extent_readonly(root, disk_bytenr))
1024                                 goto out_check;
1025                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026                                                   found_key.offset -
1027                                                   extent_offset, disk_bytenr))
1028                                 goto out_check;
1029                         disk_bytenr += extent_offset;
1030                         disk_bytenr += cur_offset - found_key.offset;
1031                         num_bytes = min(end + 1, extent_end) - cur_offset;
1032                         /*
1033                          * force cow if csum exists in the range.
1034                          * this ensure that csum for a given extent are
1035                          * either valid or do not exist.
1036                          */
1037                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1038                                 goto out_check;
1039                         nocow = 1;
1040                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1041                         extent_end = found_key.offset +
1042                                 btrfs_file_extent_inline_len(leaf, fi);
1043                         extent_end = ALIGN(extent_end, root->sectorsize);
1044                 } else {
1045                         BUG_ON(1);
1046                 }
1047 out_check:
1048                 if (extent_end <= start) {
1049                         path->slots[0]++;
1050                         goto next_slot;
1051                 }
1052                 if (!nocow) {
1053                         if (cow_start == (u64)-1)
1054                                 cow_start = cur_offset;
1055                         cur_offset = extent_end;
1056                         if (cur_offset > end)
1057                                 break;
1058                         path->slots[0]++;
1059                         goto next_slot;
1060                 }
1061
1062                 btrfs_release_path(root, path);
1063                 if (cow_start != (u64)-1) {
1064                         ret = cow_file_range(inode, locked_page, cow_start,
1065                                         found_key.offset - 1, page_started,
1066                                         nr_written, 1);
1067                         BUG_ON(ret);
1068                         cow_start = (u64)-1;
1069                 }
1070
1071                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1072                         struct extent_map *em;
1073                         struct extent_map_tree *em_tree;
1074                         em_tree = &BTRFS_I(inode)->extent_tree;
1075                         em = alloc_extent_map(GFP_NOFS);
1076                         em->start = cur_offset;
1077                         em->orig_start = em->start;
1078                         em->len = num_bytes;
1079                         em->block_len = num_bytes;
1080                         em->block_start = disk_bytenr;
1081                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1082                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1083                         while (1) {
1084                                 spin_lock(&em_tree->lock);
1085                                 ret = add_extent_mapping(em_tree, em);
1086                                 spin_unlock(&em_tree->lock);
1087                                 if (ret != -EEXIST) {
1088                                         free_extent_map(em);
1089                                         break;
1090                                 }
1091                                 btrfs_drop_extent_cache(inode, em->start,
1092                                                 em->start + em->len - 1, 0);
1093                         }
1094                         type = BTRFS_ORDERED_PREALLOC;
1095                 } else {
1096                         type = BTRFS_ORDERED_NOCOW;
1097                 }
1098
1099                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1100                                                num_bytes, num_bytes, type);
1101                 BUG_ON(ret);
1102
1103                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1104                                         cur_offset, cur_offset + num_bytes - 1,
1105                                         locked_page, 1, 1, 1, 0, 0, 0);
1106                 cur_offset = extent_end;
1107                 if (cur_offset > end)
1108                         break;
1109         }
1110         btrfs_release_path(root, path);
1111
1112         if (cur_offset <= end && cow_start == (u64)-1)
1113                 cow_start = cur_offset;
1114         if (cow_start != (u64)-1) {
1115                 ret = cow_file_range(inode, locked_page, cow_start, end,
1116                                      page_started, nr_written, 1);
1117                 BUG_ON(ret);
1118         }
1119
1120         ret = btrfs_end_transaction(trans, root);
1121         BUG_ON(ret);
1122         btrfs_free_path(path);
1123         return 0;
1124 }
1125
1126 /*
1127  * extent_io.c call back to do delayed allocation processing
1128  */
1129 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1130                               u64 start, u64 end, int *page_started,
1131                               unsigned long *nr_written)
1132 {
1133         int ret;
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135
1136         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1137                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1138                                          page_started, 1, nr_written);
1139         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1140                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1141                                          page_started, 0, nr_written);
1142         else if (!btrfs_test_opt(root, COMPRESS))
1143                 ret = cow_file_range(inode, locked_page, start, end,
1144                                       page_started, nr_written, 1);
1145         else
1146                 ret = cow_file_range_async(inode, locked_page, start, end,
1147                                            page_started, nr_written);
1148         return ret;
1149 }
1150
1151 /*
1152  * extent_io.c set_bit_hook, used to track delayed allocation
1153  * bytes in this file, and to maintain the list of inodes that
1154  * have pending delalloc work to be done.
1155  */
1156 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1157                        unsigned long old, unsigned long bits)
1158 {
1159         /*
1160          * set_bit and clear bit hooks normally require _irqsave/restore
1161          * but in this case, we are only testeing for the DELALLOC
1162          * bit, which is only set or cleared with irqs on
1163          */
1164         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165                 struct btrfs_root *root = BTRFS_I(inode)->root;
1166                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1167                 spin_lock(&root->fs_info->delalloc_lock);
1168                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1169                 root->fs_info->delalloc_bytes += end - start + 1;
1170                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1171                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1172                                       &root->fs_info->delalloc_inodes);
1173                 }
1174                 spin_unlock(&root->fs_info->delalloc_lock);
1175         }
1176         return 0;
1177 }
1178
1179 /*
1180  * extent_io.c clear_bit_hook, see set_bit_hook for why
1181  */
1182 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1183                          unsigned long old, unsigned long bits)
1184 {
1185         /*
1186          * set_bit and clear bit hooks normally require _irqsave/restore
1187          * but in this case, we are only testeing for the DELALLOC
1188          * bit, which is only set or cleared with irqs on
1189          */
1190         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1191                 struct btrfs_root *root = BTRFS_I(inode)->root;
1192
1193                 spin_lock(&root->fs_info->delalloc_lock);
1194                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1195                         printk(KERN_INFO "btrfs warning: delalloc account "
1196                                "%llu %llu\n",
1197                                (unsigned long long)end - start + 1,
1198                                (unsigned long long)
1199                                root->fs_info->delalloc_bytes);
1200                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1201                         root->fs_info->delalloc_bytes = 0;
1202                         BTRFS_I(inode)->delalloc_bytes = 0;
1203                 } else {
1204                         btrfs_delalloc_free_space(root, inode,
1205                                                   end - start + 1);
1206                         root->fs_info->delalloc_bytes -= end - start + 1;
1207                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1208                 }
1209                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1210                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1211                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1212                 }
1213                 spin_unlock(&root->fs_info->delalloc_lock);
1214         }
1215         return 0;
1216 }
1217
1218 /*
1219  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1220  * we don't create bios that span stripes or chunks
1221  */
1222 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1223                          size_t size, struct bio *bio,
1224                          unsigned long bio_flags)
1225 {
1226         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1227         struct btrfs_mapping_tree *map_tree;
1228         u64 logical = (u64)bio->bi_sector << 9;
1229         u64 length = 0;
1230         u64 map_length;
1231         int ret;
1232
1233         if (bio_flags & EXTENT_BIO_COMPRESSED)
1234                 return 0;
1235
1236         length = bio->bi_size;
1237         map_tree = &root->fs_info->mapping_tree;
1238         map_length = length;
1239         ret = btrfs_map_block(map_tree, READ, logical,
1240                               &map_length, NULL, 0);
1241
1242         if (map_length < length + size)
1243                 return 1;
1244         return 0;
1245 }
1246
1247 /*
1248  * in order to insert checksums into the metadata in large chunks,
1249  * we wait until bio submission time.   All the pages in the bio are
1250  * checksummed and sums are attached onto the ordered extent record.
1251  *
1252  * At IO completion time the cums attached on the ordered extent record
1253  * are inserted into the btree
1254  */
1255 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1256                                     struct bio *bio, int mirror_num,
1257                                     unsigned long bio_flags)
1258 {
1259         struct btrfs_root *root = BTRFS_I(inode)->root;
1260         int ret = 0;
1261
1262         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1263         BUG_ON(ret);
1264         return 0;
1265 }
1266
1267 /*
1268  * in order to insert checksums into the metadata in large chunks,
1269  * we wait until bio submission time.   All the pages in the bio are
1270  * checksummed and sums are attached onto the ordered extent record.
1271  *
1272  * At IO completion time the cums attached on the ordered extent record
1273  * are inserted into the btree
1274  */
1275 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1276                           int mirror_num, unsigned long bio_flags)
1277 {
1278         struct btrfs_root *root = BTRFS_I(inode)->root;
1279         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1280 }
1281
1282 /*
1283  * extent_io.c submission hook. This does the right thing for csum calculation
1284  * on write, or reading the csums from the tree before a read
1285  */
1286 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1287                           int mirror_num, unsigned long bio_flags)
1288 {
1289         struct btrfs_root *root = BTRFS_I(inode)->root;
1290         int ret = 0;
1291         int skip_sum;
1292
1293         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1294
1295         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1296         BUG_ON(ret);
1297
1298         if (!(rw & (1 << BIO_RW))) {
1299                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1300                         return btrfs_submit_compressed_read(inode, bio,
1301                                                     mirror_num, bio_flags);
1302                 } else if (!skip_sum)
1303                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1304                 goto mapit;
1305         } else if (!skip_sum) {
1306                 /* csum items have already been cloned */
1307                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1308                         goto mapit;
1309                 /* we're doing a write, do the async checksumming */
1310                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1311                                    inode, rw, bio, mirror_num,
1312                                    bio_flags, __btrfs_submit_bio_start,
1313                                    __btrfs_submit_bio_done);
1314         }
1315
1316 mapit:
1317         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1318 }
1319
1320 /*
1321  * given a list of ordered sums record them in the inode.  This happens
1322  * at IO completion time based on sums calculated at bio submission time.
1323  */
1324 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1325                              struct inode *inode, u64 file_offset,
1326                              struct list_head *list)
1327 {
1328         struct btrfs_ordered_sum *sum;
1329
1330         btrfs_set_trans_block_group(trans, inode);
1331
1332         list_for_each_entry(sum, list, list) {
1333                 btrfs_csum_file_blocks(trans,
1334                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1335         }
1336         return 0;
1337 }
1338
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340 {
1341         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1342                 WARN_ON(1);
1343         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344                                    GFP_NOFS);
1345 }
1346
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1349         struct page *page;
1350         struct btrfs_work work;
1351 };
1352
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1354 {
1355         struct btrfs_writepage_fixup *fixup;
1356         struct btrfs_ordered_extent *ordered;
1357         struct page *page;
1358         struct inode *inode;
1359         u64 page_start;
1360         u64 page_end;
1361
1362         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363         page = fixup->page;
1364 again:
1365         lock_page(page);
1366         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367                 ClearPageChecked(page);
1368                 goto out_page;
1369         }
1370
1371         inode = page->mapping->host;
1372         page_start = page_offset(page);
1373         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1376
1377         /* already ordered? We're done */
1378         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379                              EXTENT_ORDERED, 0)) {
1380                 goto out;
1381         }
1382
1383         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384         if (ordered) {
1385                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386                               page_end, GFP_NOFS);
1387                 unlock_page(page);
1388                 btrfs_start_ordered_extent(inode, ordered, 1);
1389                 goto again;
1390         }
1391
1392         btrfs_set_extent_delalloc(inode, page_start, page_end);
1393         ClearPageChecked(page);
1394 out:
1395         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396 out_page:
1397         unlock_page(page);
1398         page_cache_release(page);
1399 }
1400
1401 /*
1402  * There are a few paths in the higher layers of the kernel that directly
1403  * set the page dirty bit without asking the filesystem if it is a
1404  * good idea.  This causes problems because we want to make sure COW
1405  * properly happens and the data=ordered rules are followed.
1406  *
1407  * In our case any range that doesn't have the ORDERED bit set
1408  * hasn't been properly setup for IO.  We kick off an async process
1409  * to fix it up.  The async helper will wait for ordered extents, set
1410  * the delalloc bit and make it safe to write the page.
1411  */
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 {
1414         struct inode *inode = page->mapping->host;
1415         struct btrfs_writepage_fixup *fixup;
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret;
1418
1419         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420                              EXTENT_ORDERED, 0);
1421         if (ret)
1422                 return 0;
1423
1424         if (PageChecked(page))
1425                 return -EAGAIN;
1426
1427         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428         if (!fixup)
1429                 return -EAGAIN;
1430
1431         SetPageChecked(page);
1432         page_cache_get(page);
1433         fixup->work.func = btrfs_writepage_fixup_worker;
1434         fixup->page = page;
1435         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436         return -EAGAIN;
1437 }
1438
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440                                        struct inode *inode, u64 file_pos,
1441                                        u64 disk_bytenr, u64 disk_num_bytes,
1442                                        u64 num_bytes, u64 ram_bytes,
1443                                        u64 locked_end,
1444                                        u8 compression, u8 encryption,
1445                                        u16 other_encoding, int extent_type)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         struct btrfs_file_extent_item *fi;
1449         struct btrfs_path *path;
1450         struct extent_buffer *leaf;
1451         struct btrfs_key ins;
1452         u64 hint;
1453         int ret;
1454
1455         path = btrfs_alloc_path();
1456         BUG_ON(!path);
1457
1458         path->leave_spinning = 1;
1459         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1460                                  file_pos + num_bytes, locked_end,
1461                                  file_pos, &hint);
1462         BUG_ON(ret);
1463
1464         ins.objectid = inode->i_ino;
1465         ins.offset = file_pos;
1466         ins.type = BTRFS_EXTENT_DATA_KEY;
1467         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1468         BUG_ON(ret);
1469         leaf = path->nodes[0];
1470         fi = btrfs_item_ptr(leaf, path->slots[0],
1471                             struct btrfs_file_extent_item);
1472         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1473         btrfs_set_file_extent_type(leaf, fi, extent_type);
1474         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1475         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1476         btrfs_set_file_extent_offset(leaf, fi, 0);
1477         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1478         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1479         btrfs_set_file_extent_compression(leaf, fi, compression);
1480         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1481         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1482
1483         btrfs_unlock_up_safe(path, 1);
1484         btrfs_set_lock_blocking(leaf);
1485
1486         btrfs_mark_buffer_dirty(leaf);
1487
1488         inode_add_bytes(inode, num_bytes);
1489         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1490
1491         ins.objectid = disk_bytenr;
1492         ins.offset = disk_num_bytes;
1493         ins.type = BTRFS_EXTENT_ITEM_KEY;
1494         ret = btrfs_alloc_reserved_file_extent(trans, root,
1495                                         root->root_key.objectid,
1496                                         inode->i_ino, file_pos, &ins);
1497         BUG_ON(ret);
1498         btrfs_free_path(path);
1499
1500         return 0;
1501 }
1502
1503 /*
1504  * helper function for btrfs_finish_ordered_io, this
1505  * just reads in some of the csum leaves to prime them into ram
1506  * before we start the transaction.  It limits the amount of btree
1507  * reads required while inside the transaction.
1508  */
1509 static noinline void reada_csum(struct btrfs_root *root,
1510                                 struct btrfs_path *path,
1511                                 struct btrfs_ordered_extent *ordered_extent)
1512 {
1513         struct btrfs_ordered_sum *sum;
1514         u64 bytenr;
1515
1516         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1517                          list);
1518         bytenr = sum->sums[0].bytenr;
1519
1520         /*
1521          * we don't care about the results, the point of this search is
1522          * just to get the btree leaves into ram
1523          */
1524         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1525 }
1526
1527 /* as ordered data IO finishes, this gets called so we can finish
1528  * an ordered extent if the range of bytes in the file it covers are
1529  * fully written.
1530  */
1531 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1532 {
1533         struct btrfs_root *root = BTRFS_I(inode)->root;
1534         struct btrfs_trans_handle *trans;
1535         struct btrfs_ordered_extent *ordered_extent = NULL;
1536         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1537         struct btrfs_path *path;
1538         int compressed = 0;
1539         int ret;
1540
1541         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1542         if (!ret)
1543                 return 0;
1544
1545         /*
1546          * before we join the transaction, try to do some of our IO.
1547          * This will limit the amount of IO that we have to do with
1548          * the transaction running.  We're unlikely to need to do any
1549          * IO if the file extents are new, the disk_i_size checks
1550          * covers the most common case.
1551          */
1552         if (start < BTRFS_I(inode)->disk_i_size) {
1553                 path = btrfs_alloc_path();
1554                 if (path) {
1555                         ret = btrfs_lookup_file_extent(NULL, root, path,
1556                                                        inode->i_ino,
1557                                                        start, 0);
1558                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1559                                                                      start);
1560                         if (!list_empty(&ordered_extent->list)) {
1561                                 btrfs_release_path(root, path);
1562                                 reada_csum(root, path, ordered_extent);
1563                         }
1564                         btrfs_free_path(path);
1565                 }
1566         }
1567
1568         trans = btrfs_join_transaction(root, 1);
1569
1570         if (!ordered_extent)
1571                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1572         BUG_ON(!ordered_extent);
1573         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1574                 goto nocow;
1575
1576         lock_extent(io_tree, ordered_extent->file_offset,
1577                     ordered_extent->file_offset + ordered_extent->len - 1,
1578                     GFP_NOFS);
1579
1580         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1581                 compressed = 1;
1582         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1583                 BUG_ON(compressed);
1584                 ret = btrfs_mark_extent_written(trans, root, inode,
1585                                                 ordered_extent->file_offset,
1586                                                 ordered_extent->file_offset +
1587                                                 ordered_extent->len);
1588                 BUG_ON(ret);
1589         } else {
1590                 ret = insert_reserved_file_extent(trans, inode,
1591                                                 ordered_extent->file_offset,
1592                                                 ordered_extent->start,
1593                                                 ordered_extent->disk_len,
1594                                                 ordered_extent->len,
1595                                                 ordered_extent->len,
1596                                                 ordered_extent->file_offset +
1597                                                 ordered_extent->len,
1598                                                 compressed, 0, 0,
1599                                                 BTRFS_FILE_EXTENT_REG);
1600                 BUG_ON(ret);
1601         }
1602         unlock_extent(io_tree, ordered_extent->file_offset,
1603                     ordered_extent->file_offset + ordered_extent->len - 1,
1604                     GFP_NOFS);
1605 nocow:
1606         add_pending_csums(trans, inode, ordered_extent->file_offset,
1607                           &ordered_extent->list);
1608
1609         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1610         btrfs_ordered_update_i_size(inode, ordered_extent);
1611         btrfs_update_inode(trans, root, inode);
1612         btrfs_remove_ordered_extent(inode, ordered_extent);
1613         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1614
1615         /* once for us */
1616         btrfs_put_ordered_extent(ordered_extent);
1617         /* once for the tree */
1618         btrfs_put_ordered_extent(ordered_extent);
1619
1620         btrfs_end_transaction(trans, root);
1621         return 0;
1622 }
1623
1624 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1625                                 struct extent_state *state, int uptodate)
1626 {
1627         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1628 }
1629
1630 /*
1631  * When IO fails, either with EIO or csum verification fails, we
1632  * try other mirrors that might have a good copy of the data.  This
1633  * io_failure_record is used to record state as we go through all the
1634  * mirrors.  If another mirror has good data, the page is set up to date
1635  * and things continue.  If a good mirror can't be found, the original
1636  * bio end_io callback is called to indicate things have failed.
1637  */
1638 struct io_failure_record {
1639         struct page *page;
1640         u64 start;
1641         u64 len;
1642         u64 logical;
1643         unsigned long bio_flags;
1644         int last_mirror;
1645 };
1646
1647 static int btrfs_io_failed_hook(struct bio *failed_bio,
1648                          struct page *page, u64 start, u64 end,
1649                          struct extent_state *state)
1650 {
1651         struct io_failure_record *failrec = NULL;
1652         u64 private;
1653         struct extent_map *em;
1654         struct inode *inode = page->mapping->host;
1655         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1656         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1657         struct bio *bio;
1658         int num_copies;
1659         int ret;
1660         int rw;
1661         u64 logical;
1662
1663         ret = get_state_private(failure_tree, start, &private);
1664         if (ret) {
1665                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1666                 if (!failrec)
1667                         return -ENOMEM;
1668                 failrec->start = start;
1669                 failrec->len = end - start + 1;
1670                 failrec->last_mirror = 0;
1671                 failrec->bio_flags = 0;
1672
1673                 spin_lock(&em_tree->lock);
1674                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1675                 if (em->start > start || em->start + em->len < start) {
1676                         free_extent_map(em);
1677                         em = NULL;
1678                 }
1679                 spin_unlock(&em_tree->lock);
1680
1681                 if (!em || IS_ERR(em)) {
1682                         kfree(failrec);
1683                         return -EIO;
1684                 }
1685                 logical = start - em->start;
1686                 logical = em->block_start + logical;
1687                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1688                         logical = em->block_start;
1689                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1690                 }
1691                 failrec->logical = logical;
1692                 free_extent_map(em);
1693                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1694                                 EXTENT_DIRTY, GFP_NOFS);
1695                 set_state_private(failure_tree, start,
1696                                  (u64)(unsigned long)failrec);
1697         } else {
1698                 failrec = (struct io_failure_record *)(unsigned long)private;
1699         }
1700         num_copies = btrfs_num_copies(
1701                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1702                               failrec->logical, failrec->len);
1703         failrec->last_mirror++;
1704         if (!state) {
1705                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1706                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1707                                                     failrec->start,
1708                                                     EXTENT_LOCKED);
1709                 if (state && state->start != failrec->start)
1710                         state = NULL;
1711                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1712         }
1713         if (!state || failrec->last_mirror > num_copies) {
1714                 set_state_private(failure_tree, failrec->start, 0);
1715                 clear_extent_bits(failure_tree, failrec->start,
1716                                   failrec->start + failrec->len - 1,
1717                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1718                 kfree(failrec);
1719                 return -EIO;
1720         }
1721         bio = bio_alloc(GFP_NOFS, 1);
1722         bio->bi_private = state;
1723         bio->bi_end_io = failed_bio->bi_end_io;
1724         bio->bi_sector = failrec->logical >> 9;
1725         bio->bi_bdev = failed_bio->bi_bdev;
1726         bio->bi_size = 0;
1727
1728         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1729         if (failed_bio->bi_rw & (1 << BIO_RW))
1730                 rw = WRITE;
1731         else
1732                 rw = READ;
1733
1734         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1735                                                       failrec->last_mirror,
1736                                                       failrec->bio_flags);
1737         return 0;
1738 }
1739
1740 /*
1741  * each time an IO finishes, we do a fast check in the IO failure tree
1742  * to see if we need to process or clean up an io_failure_record
1743  */
1744 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1745 {
1746         u64 private;
1747         u64 private_failure;
1748         struct io_failure_record *failure;
1749         int ret;
1750
1751         private = 0;
1752         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1753                              (u64)-1, 1, EXTENT_DIRTY)) {
1754                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1755                                         start, &private_failure);
1756                 if (ret == 0) {
1757                         failure = (struct io_failure_record *)(unsigned long)
1758                                    private_failure;
1759                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1760                                           failure->start, 0);
1761                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1762                                           failure->start,
1763                                           failure->start + failure->len - 1,
1764                                           EXTENT_DIRTY | EXTENT_LOCKED,
1765                                           GFP_NOFS);
1766                         kfree(failure);
1767                 }
1768         }
1769         return 0;
1770 }
1771
1772 /*
1773  * when reads are done, we need to check csums to verify the data is correct
1774  * if there's a match, we allow the bio to finish.  If not, we go through
1775  * the io_failure_record routines to find good copies
1776  */
1777 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1778                                struct extent_state *state)
1779 {
1780         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1781         struct inode *inode = page->mapping->host;
1782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1783         char *kaddr;
1784         u64 private = ~(u32)0;
1785         int ret;
1786         struct btrfs_root *root = BTRFS_I(inode)->root;
1787         u32 csum = ~(u32)0;
1788
1789         if (PageChecked(page)) {
1790                 ClearPageChecked(page);
1791                 goto good;
1792         }
1793
1794         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1795                 return 0;
1796
1797         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1798             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1799                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1800                                   GFP_NOFS);
1801                 return 0;
1802         }
1803
1804         if (state && state->start == start) {
1805                 private = state->private;
1806                 ret = 0;
1807         } else {
1808                 ret = get_state_private(io_tree, start, &private);
1809         }
1810         kaddr = kmap_atomic(page, KM_USER0);
1811         if (ret)
1812                 goto zeroit;
1813
1814         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1815         btrfs_csum_final(csum, (char *)&csum);
1816         if (csum != private)
1817                 goto zeroit;
1818
1819         kunmap_atomic(kaddr, KM_USER0);
1820 good:
1821         /* if the io failure tree for this inode is non-empty,
1822          * check to see if we've recovered from a failed IO
1823          */
1824         btrfs_clean_io_failures(inode, start);
1825         return 0;
1826
1827 zeroit:
1828         if (printk_ratelimit()) {
1829                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1830                        "private %llu\n", page->mapping->host->i_ino,
1831                        (unsigned long long)start, csum,
1832                        (unsigned long long)private);
1833         }
1834         memset(kaddr + offset, 1, end - start + 1);
1835         flush_dcache_page(page);
1836         kunmap_atomic(kaddr, KM_USER0);
1837         if (private == 0)
1838                 return 0;
1839         return -EIO;
1840 }
1841
1842 /*
1843  * This creates an orphan entry for the given inode in case something goes
1844  * wrong in the middle of an unlink/truncate.
1845  */
1846 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1847 {
1848         struct btrfs_root *root = BTRFS_I(inode)->root;
1849         int ret = 0;
1850
1851         spin_lock(&root->list_lock);
1852
1853         /* already on the orphan list, we're good */
1854         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1855                 spin_unlock(&root->list_lock);
1856                 return 0;
1857         }
1858
1859         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1860
1861         spin_unlock(&root->list_lock);
1862
1863         /*
1864          * insert an orphan item to track this unlinked/truncated file
1865          */
1866         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1867
1868         return ret;
1869 }
1870
1871 /*
1872  * We have done the truncate/delete so we can go ahead and remove the orphan
1873  * item for this particular inode.
1874  */
1875 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1876 {
1877         struct btrfs_root *root = BTRFS_I(inode)->root;
1878         int ret = 0;
1879
1880         spin_lock(&root->list_lock);
1881
1882         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1883                 spin_unlock(&root->list_lock);
1884                 return 0;
1885         }
1886
1887         list_del_init(&BTRFS_I(inode)->i_orphan);
1888         if (!trans) {
1889                 spin_unlock(&root->list_lock);
1890                 return 0;
1891         }
1892
1893         spin_unlock(&root->list_lock);
1894
1895         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1896
1897         return ret;
1898 }
1899
1900 /*
1901  * this cleans up any orphans that may be left on the list from the last use
1902  * of this root.
1903  */
1904 void btrfs_orphan_cleanup(struct btrfs_root *root)
1905 {
1906         struct btrfs_path *path;
1907         struct extent_buffer *leaf;
1908         struct btrfs_item *item;
1909         struct btrfs_key key, found_key;
1910         struct btrfs_trans_handle *trans;
1911         struct inode *inode;
1912         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1913
1914         path = btrfs_alloc_path();
1915         if (!path)
1916                 return;
1917         path->reada = -1;
1918
1919         key.objectid = BTRFS_ORPHAN_OBJECTID;
1920         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1921         key.offset = (u64)-1;
1922
1923
1924         while (1) {
1925                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1926                 if (ret < 0) {
1927                         printk(KERN_ERR "Error searching slot for orphan: %d"
1928                                "\n", ret);
1929                         break;
1930                 }
1931
1932                 /*
1933                  * if ret == 0 means we found what we were searching for, which
1934                  * is weird, but possible, so only screw with path if we didnt
1935                  * find the key and see if we have stuff that matches
1936                  */
1937                 if (ret > 0) {
1938                         if (path->slots[0] == 0)
1939                                 break;
1940                         path->slots[0]--;
1941                 }
1942
1943                 /* pull out the item */
1944                 leaf = path->nodes[0];
1945                 item = btrfs_item_nr(leaf, path->slots[0]);
1946                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1947
1948                 /* make sure the item matches what we want */
1949                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1950                         break;
1951                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1952                         break;
1953
1954                 /* release the path since we're done with it */
1955                 btrfs_release_path(root, path);
1956
1957                 /*
1958                  * this is where we are basically btrfs_lookup, without the
1959                  * crossing root thing.  we store the inode number in the
1960                  * offset of the orphan item.
1961                  */
1962                 found_key.objectid = found_key.offset;
1963                 found_key.type = BTRFS_INODE_ITEM_KEY;
1964                 found_key.offset = 0;
1965                 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1966                 if (IS_ERR(inode))
1967                         break;
1968
1969                 /*
1970                  * add this inode to the orphan list so btrfs_orphan_del does
1971                  * the proper thing when we hit it
1972                  */
1973                 spin_lock(&root->list_lock);
1974                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1975                 spin_unlock(&root->list_lock);
1976
1977                 /*
1978                  * if this is a bad inode, means we actually succeeded in
1979                  * removing the inode, but not the orphan record, which means
1980                  * we need to manually delete the orphan since iput will just
1981                  * do a destroy_inode
1982                  */
1983                 if (is_bad_inode(inode)) {
1984                         trans = btrfs_start_transaction(root, 1);
1985                         btrfs_orphan_del(trans, inode);
1986                         btrfs_end_transaction(trans, root);
1987                         iput(inode);
1988                         continue;
1989                 }
1990
1991                 /* if we have links, this was a truncate, lets do that */
1992                 if (inode->i_nlink) {
1993                         nr_truncate++;
1994                         btrfs_truncate(inode);
1995                 } else {
1996                         nr_unlink++;
1997                 }
1998
1999                 /* this will do delete_inode and everything for us */
2000                 iput(inode);
2001         }
2002
2003         if (nr_unlink)
2004                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2005         if (nr_truncate)
2006                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2007
2008         btrfs_free_path(path);
2009 }
2010
2011 /*
2012  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2013  * don't find any xattrs, we know there can't be any acls.
2014  *
2015  * slot is the slot the inode is in, objectid is the objectid of the inode
2016  */
2017 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2018                                           int slot, u64 objectid)
2019 {
2020         u32 nritems = btrfs_header_nritems(leaf);
2021         struct btrfs_key found_key;
2022         int scanned = 0;
2023
2024         slot++;
2025         while (slot < nritems) {
2026                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2027
2028                 /* we found a different objectid, there must not be acls */
2029                 if (found_key.objectid != objectid)
2030                         return 0;
2031
2032                 /* we found an xattr, assume we've got an acl */
2033                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2034                         return 1;
2035
2036                 /*
2037                  * we found a key greater than an xattr key, there can't
2038                  * be any acls later on
2039                  */
2040                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2041                         return 0;
2042
2043                 slot++;
2044                 scanned++;
2045
2046                 /*
2047                  * it goes inode, inode backrefs, xattrs, extents,
2048                  * so if there are a ton of hard links to an inode there can
2049                  * be a lot of backrefs.  Don't waste time searching too hard,
2050                  * this is just an optimization
2051                  */
2052                 if (scanned >= 8)
2053                         break;
2054         }
2055         /* we hit the end of the leaf before we found an xattr or
2056          * something larger than an xattr.  We have to assume the inode
2057          * has acls
2058          */
2059         return 1;
2060 }
2061
2062 /*
2063  * read an inode from the btree into the in-memory inode
2064  */
2065 static void btrfs_read_locked_inode(struct inode *inode)
2066 {
2067         struct btrfs_path *path;
2068         struct extent_buffer *leaf;
2069         struct btrfs_inode_item *inode_item;
2070         struct btrfs_timespec *tspec;
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         struct btrfs_key location;
2073         int maybe_acls;
2074         u64 alloc_group_block;
2075         u32 rdev;
2076         int ret;
2077
2078         path = btrfs_alloc_path();
2079         BUG_ON(!path);
2080         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2081
2082         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2083         if (ret)
2084                 goto make_bad;
2085
2086         leaf = path->nodes[0];
2087         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2088                                     struct btrfs_inode_item);
2089
2090         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2091         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2092         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2093         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2094         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2095
2096         tspec = btrfs_inode_atime(inode_item);
2097         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2098         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2099
2100         tspec = btrfs_inode_mtime(inode_item);
2101         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2102         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2103
2104         tspec = btrfs_inode_ctime(inode_item);
2105         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2106         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2107
2108         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2109         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2110         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2111         inode->i_generation = BTRFS_I(inode)->generation;
2112         inode->i_rdev = 0;
2113         rdev = btrfs_inode_rdev(leaf, inode_item);
2114
2115         BTRFS_I(inode)->index_cnt = (u64)-1;
2116         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2117
2118         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2119
2120         /*
2121          * try to precache a NULL acl entry for files that don't have
2122          * any xattrs or acls
2123          */
2124         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2125         if (!maybe_acls) {
2126                 BTRFS_I(inode)->i_acl = NULL;
2127                 BTRFS_I(inode)->i_default_acl = NULL;
2128         }
2129
2130         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2131                                                 alloc_group_block, 0);
2132         btrfs_free_path(path);
2133         inode_item = NULL;
2134
2135         switch (inode->i_mode & S_IFMT) {
2136         case S_IFREG:
2137                 inode->i_mapping->a_ops = &btrfs_aops;
2138                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2139                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2140                 inode->i_fop = &btrfs_file_operations;
2141                 inode->i_op = &btrfs_file_inode_operations;
2142                 break;
2143         case S_IFDIR:
2144                 inode->i_fop = &btrfs_dir_file_operations;
2145                 if (root == root->fs_info->tree_root)
2146                         inode->i_op = &btrfs_dir_ro_inode_operations;
2147                 else
2148                         inode->i_op = &btrfs_dir_inode_operations;
2149                 break;
2150         case S_IFLNK:
2151                 inode->i_op = &btrfs_symlink_inode_operations;
2152                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2153                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2154                 break;
2155         default:
2156                 inode->i_op = &btrfs_special_inode_operations;
2157                 init_special_inode(inode, inode->i_mode, rdev);
2158                 break;
2159         }
2160
2161         btrfs_update_iflags(inode);
2162         return;
2163
2164 make_bad:
2165         btrfs_free_path(path);
2166         make_bad_inode(inode);
2167 }
2168
2169 /*
2170  * given a leaf and an inode, copy the inode fields into the leaf
2171  */
2172 static void fill_inode_item(struct btrfs_trans_handle *trans,
2173                             struct extent_buffer *leaf,
2174                             struct btrfs_inode_item *item,
2175                             struct inode *inode)
2176 {
2177         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2178         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2179         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2180         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2181         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2182
2183         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2184                                inode->i_atime.tv_sec);
2185         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2186                                 inode->i_atime.tv_nsec);
2187
2188         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2189                                inode->i_mtime.tv_sec);
2190         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2191                                 inode->i_mtime.tv_nsec);
2192
2193         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2194                                inode->i_ctime.tv_sec);
2195         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2196                                 inode->i_ctime.tv_nsec);
2197
2198         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2199         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2200         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2201         btrfs_set_inode_transid(leaf, item, trans->transid);
2202         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2203         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2204         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2205 }
2206
2207 /*
2208  * copy everything in the in-memory inode into the btree.
2209  */
2210 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2211                                 struct btrfs_root *root, struct inode *inode)
2212 {
2213         struct btrfs_inode_item *inode_item;
2214         struct btrfs_path *path;
2215         struct extent_buffer *leaf;
2216         int ret;
2217
2218         path = btrfs_alloc_path();
2219         BUG_ON(!path);
2220         path->leave_spinning = 1;
2221         ret = btrfs_lookup_inode(trans, root, path,
2222                                  &BTRFS_I(inode)->location, 1);
2223         if (ret) {
2224                 if (ret > 0)
2225                         ret = -ENOENT;
2226                 goto failed;
2227         }
2228
2229         btrfs_unlock_up_safe(path, 1);
2230         leaf = path->nodes[0];
2231         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2232                                   struct btrfs_inode_item);
2233
2234         fill_inode_item(trans, leaf, inode_item, inode);
2235         btrfs_mark_buffer_dirty(leaf);
2236         btrfs_set_inode_last_trans(trans, inode);
2237         ret = 0;
2238 failed:
2239         btrfs_free_path(path);
2240         return ret;
2241 }
2242
2243
2244 /*
2245  * unlink helper that gets used here in inode.c and in the tree logging
2246  * recovery code.  It remove a link in a directory with a given name, and
2247  * also drops the back refs in the inode to the directory
2248  */
2249 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2250                        struct btrfs_root *root,
2251                        struct inode *dir, struct inode *inode,
2252                        const char *name, int name_len)
2253 {
2254         struct btrfs_path *path;
2255         int ret = 0;
2256         struct extent_buffer *leaf;
2257         struct btrfs_dir_item *di;
2258         struct btrfs_key key;
2259         u64 index;
2260
2261         path = btrfs_alloc_path();
2262         if (!path) {
2263                 ret = -ENOMEM;
2264                 goto err;
2265         }
2266
2267         path->leave_spinning = 1;
2268         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2269                                     name, name_len, -1);
2270         if (IS_ERR(di)) {
2271                 ret = PTR_ERR(di);
2272                 goto err;
2273         }
2274         if (!di) {
2275                 ret = -ENOENT;
2276                 goto err;
2277         }
2278         leaf = path->nodes[0];
2279         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2280         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2281         if (ret)
2282                 goto err;
2283         btrfs_release_path(root, path);
2284
2285         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2286                                   inode->i_ino,
2287                                   dir->i_ino, &index);
2288         if (ret) {
2289                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2290                        "inode %lu parent %lu\n", name_len, name,
2291                        inode->i_ino, dir->i_ino);
2292                 goto err;
2293         }
2294
2295         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2296                                          index, name, name_len, -1);
2297         if (IS_ERR(di)) {
2298                 ret = PTR_ERR(di);
2299                 goto err;
2300         }
2301         if (!di) {
2302                 ret = -ENOENT;
2303                 goto err;
2304         }
2305         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2306         btrfs_release_path(root, path);
2307
2308         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2309                                          inode, dir->i_ino);
2310         BUG_ON(ret != 0 && ret != -ENOENT);
2311
2312         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2313                                            dir, index);
2314         BUG_ON(ret);
2315 err:
2316         btrfs_free_path(path);
2317         if (ret)
2318                 goto out;
2319
2320         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2321         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2322         btrfs_update_inode(trans, root, dir);
2323         btrfs_drop_nlink(inode);
2324         ret = btrfs_update_inode(trans, root, inode);
2325         dir->i_sb->s_dirt = 1;
2326 out:
2327         return ret;
2328 }
2329
2330 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2331 {
2332         struct btrfs_root *root;
2333         struct btrfs_trans_handle *trans;
2334         struct inode *inode = dentry->d_inode;
2335         int ret;
2336         unsigned long nr = 0;
2337
2338         root = BTRFS_I(dir)->root;
2339
2340         trans = btrfs_start_transaction(root, 1);
2341
2342         btrfs_set_trans_block_group(trans, dir);
2343
2344         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2345
2346         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2347                                  dentry->d_name.name, dentry->d_name.len);
2348
2349         if (inode->i_nlink == 0)
2350                 ret = btrfs_orphan_add(trans, inode);
2351
2352         nr = trans->blocks_used;
2353
2354         btrfs_end_transaction_throttle(trans, root);
2355         btrfs_btree_balance_dirty(root, nr);
2356         return ret;
2357 }
2358
2359 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2360 {
2361         struct inode *inode = dentry->d_inode;
2362         int err = 0;
2363         int ret;
2364         struct btrfs_root *root = BTRFS_I(dir)->root;
2365         struct btrfs_trans_handle *trans;
2366         unsigned long nr = 0;
2367
2368         /*
2369          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2370          * the root of a subvolume or snapshot
2371          */
2372         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2373             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2374                 return -ENOTEMPTY;
2375         }
2376
2377         trans = btrfs_start_transaction(root, 1);
2378         btrfs_set_trans_block_group(trans, dir);
2379
2380         err = btrfs_orphan_add(trans, inode);
2381         if (err)
2382                 goto fail_trans;
2383
2384         /* now the directory is empty */
2385         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2386                                  dentry->d_name.name, dentry->d_name.len);
2387         if (!err)
2388                 btrfs_i_size_write(inode, 0);
2389
2390 fail_trans:
2391         nr = trans->blocks_used;
2392         ret = btrfs_end_transaction_throttle(trans, root);
2393         btrfs_btree_balance_dirty(root, nr);
2394
2395         if (ret && !err)
2396                 err = ret;
2397         return err;
2398 }
2399
2400 #if 0
2401 /*
2402  * when truncating bytes in a file, it is possible to avoid reading
2403  * the leaves that contain only checksum items.  This can be the
2404  * majority of the IO required to delete a large file, but it must
2405  * be done carefully.
2406  *
2407  * The keys in the level just above the leaves are checked to make sure
2408  * the lowest key in a given leaf is a csum key, and starts at an offset
2409  * after the new  size.
2410  *
2411  * Then the key for the next leaf is checked to make sure it also has
2412  * a checksum item for the same file.  If it does, we know our target leaf
2413  * contains only checksum items, and it can be safely freed without reading
2414  * it.
2415  *
2416  * This is just an optimization targeted at large files.  It may do
2417  * nothing.  It will return 0 unless things went badly.
2418  */
2419 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2420                                      struct btrfs_root *root,
2421                                      struct btrfs_path *path,
2422                                      struct inode *inode, u64 new_size)
2423 {
2424         struct btrfs_key key;
2425         int ret;
2426         int nritems;
2427         struct btrfs_key found_key;
2428         struct btrfs_key other_key;
2429         struct btrfs_leaf_ref *ref;
2430         u64 leaf_gen;
2431         u64 leaf_start;
2432
2433         path->lowest_level = 1;
2434         key.objectid = inode->i_ino;
2435         key.type = BTRFS_CSUM_ITEM_KEY;
2436         key.offset = new_size;
2437 again:
2438         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2439         if (ret < 0)
2440                 goto out;
2441
2442         if (path->nodes[1] == NULL) {
2443                 ret = 0;
2444                 goto out;
2445         }
2446         ret = 0;
2447         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2448         nritems = btrfs_header_nritems(path->nodes[1]);
2449
2450         if (!nritems)
2451                 goto out;
2452
2453         if (path->slots[1] >= nritems)
2454                 goto next_node;
2455
2456         /* did we find a key greater than anything we want to delete? */
2457         if (found_key.objectid > inode->i_ino ||
2458            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2459                 goto out;
2460
2461         /* we check the next key in the node to make sure the leave contains
2462          * only checksum items.  This comparison doesn't work if our
2463          * leaf is the last one in the node
2464          */
2465         if (path->slots[1] + 1 >= nritems) {
2466 next_node:
2467                 /* search forward from the last key in the node, this
2468                  * will bring us into the next node in the tree
2469                  */
2470                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2471
2472                 /* unlikely, but we inc below, so check to be safe */
2473                 if (found_key.offset == (u64)-1)
2474                         goto out;
2475
2476                 /* search_forward needs a path with locks held, do the
2477                  * search again for the original key.  It is possible
2478                  * this will race with a balance and return a path that
2479                  * we could modify, but this drop is just an optimization
2480                  * and is allowed to miss some leaves.
2481                  */
2482                 btrfs_release_path(root, path);
2483                 found_key.offset++;
2484
2485                 /* setup a max key for search_forward */
2486                 other_key.offset = (u64)-1;
2487                 other_key.type = key.type;
2488                 other_key.objectid = key.objectid;
2489
2490                 path->keep_locks = 1;
2491                 ret = btrfs_search_forward(root, &found_key, &other_key,
2492                                            path, 0, 0);
2493                 path->keep_locks = 0;
2494                 if (ret || found_key.objectid != key.objectid ||
2495                     found_key.type != key.type) {
2496                         ret = 0;
2497                         goto out;
2498                 }
2499
2500                 key.offset = found_key.offset;
2501                 btrfs_release_path(root, path);
2502                 cond_resched();
2503                 goto again;
2504         }
2505
2506         /* we know there's one more slot after us in the tree,
2507          * read that key so we can verify it is also a checksum item
2508          */
2509         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2510
2511         if (found_key.objectid < inode->i_ino)
2512                 goto next_key;
2513
2514         if (found_key.type != key.type || found_key.offset < new_size)
2515                 goto next_key;
2516
2517         /*
2518          * if the key for the next leaf isn't a csum key from this objectid,
2519          * we can't be sure there aren't good items inside this leaf.
2520          * Bail out
2521          */
2522         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2523                 goto out;
2524
2525         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2526         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2527         /*
2528          * it is safe to delete this leaf, it contains only
2529          * csum items from this inode at an offset >= new_size
2530          */
2531         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2532         BUG_ON(ret);
2533
2534         if (root->ref_cows && leaf_gen < trans->transid) {
2535                 ref = btrfs_alloc_leaf_ref(root, 0);
2536                 if (ref) {
2537                         ref->root_gen = root->root_key.offset;
2538                         ref->bytenr = leaf_start;
2539                         ref->owner = 0;
2540                         ref->generation = leaf_gen;
2541                         ref->nritems = 0;
2542
2543                         btrfs_sort_leaf_ref(ref);
2544
2545                         ret = btrfs_add_leaf_ref(root, ref, 0);
2546                         WARN_ON(ret);
2547                         btrfs_free_leaf_ref(root, ref);
2548                 } else {
2549                         WARN_ON(1);
2550                 }
2551         }
2552 next_key:
2553         btrfs_release_path(root, path);
2554
2555         if (other_key.objectid == inode->i_ino &&
2556             other_key.type == key.type && other_key.offset > key.offset) {
2557                 key.offset = other_key.offset;
2558                 cond_resched();
2559                 goto again;
2560         }
2561         ret = 0;
2562 out:
2563         /* fixup any changes we've made to the path */
2564         path->lowest_level = 0;
2565         path->keep_locks = 0;
2566         btrfs_release_path(root, path);
2567         return ret;
2568 }
2569
2570 #endif
2571
2572 /*
2573  * this can truncate away extent items, csum items and directory items.
2574  * It starts at a high offset and removes keys until it can't find
2575  * any higher than new_size
2576  *
2577  * csum items that cross the new i_size are truncated to the new size
2578  * as well.
2579  *
2580  * min_type is the minimum key type to truncate down to.  If set to 0, this
2581  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2582  */
2583 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2584                                         struct btrfs_root *root,
2585                                         struct inode *inode,
2586                                         u64 new_size, u32 min_type)
2587 {
2588         int ret;
2589         struct btrfs_path *path;
2590         struct btrfs_key key;
2591         struct btrfs_key found_key;
2592         u32 found_type = (u8)-1;
2593         struct extent_buffer *leaf;
2594         struct btrfs_file_extent_item *fi;
2595         u64 extent_start = 0;
2596         u64 extent_num_bytes = 0;
2597         u64 extent_offset = 0;
2598         u64 item_end = 0;
2599         int found_extent;
2600         int del_item;
2601         int pending_del_nr = 0;
2602         int pending_del_slot = 0;
2603         int extent_type = -1;
2604         int encoding;
2605         u64 mask = root->sectorsize - 1;
2606
2607         if (root->ref_cows)
2608                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2609         path = btrfs_alloc_path();
2610         BUG_ON(!path);
2611         path->reada = -1;
2612
2613         /* FIXME, add redo link to tree so we don't leak on crash */
2614         key.objectid = inode->i_ino;
2615         key.offset = (u64)-1;
2616         key.type = (u8)-1;
2617
2618 search_again:
2619         path->leave_spinning = 1;
2620         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2621         if (ret < 0)
2622                 goto error;
2623
2624         if (ret > 0) {
2625                 /* there are no items in the tree for us to truncate, we're
2626                  * done
2627                  */
2628                 if (path->slots[0] == 0) {
2629                         ret = 0;
2630                         goto error;
2631                 }
2632                 path->slots[0]--;
2633         }
2634
2635         while (1) {
2636                 fi = NULL;
2637                 leaf = path->nodes[0];
2638                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2639                 found_type = btrfs_key_type(&found_key);
2640                 encoding = 0;
2641
2642                 if (found_key.objectid != inode->i_ino)
2643                         break;
2644
2645                 if (found_type < min_type)
2646                         break;
2647
2648                 item_end = found_key.offset;
2649                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2650                         fi = btrfs_item_ptr(leaf, path->slots[0],
2651                                             struct btrfs_file_extent_item);
2652                         extent_type = btrfs_file_extent_type(leaf, fi);
2653                         encoding = btrfs_file_extent_compression(leaf, fi);
2654                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2655                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2656
2657                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2658                                 item_end +=
2659                                     btrfs_file_extent_num_bytes(leaf, fi);
2660                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2661                                 item_end += btrfs_file_extent_inline_len(leaf,
2662                                                                          fi);
2663                         }
2664                         item_end--;
2665                 }
2666                 if (item_end < new_size) {
2667                         if (found_type == BTRFS_DIR_ITEM_KEY)
2668                                 found_type = BTRFS_INODE_ITEM_KEY;
2669                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2670                                 found_type = BTRFS_EXTENT_DATA_KEY;
2671                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2672                                 found_type = BTRFS_XATTR_ITEM_KEY;
2673                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2674                                 found_type = BTRFS_INODE_REF_KEY;
2675                         else if (found_type)
2676                                 found_type--;
2677                         else
2678                                 break;
2679                         btrfs_set_key_type(&key, found_type);
2680                         goto next;
2681                 }
2682                 if (found_key.offset >= new_size)
2683                         del_item = 1;
2684                 else
2685                         del_item = 0;
2686                 found_extent = 0;
2687
2688                 /* FIXME, shrink the extent if the ref count is only 1 */
2689                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2690                         goto delete;
2691
2692                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2693                         u64 num_dec;
2694                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2695                         if (!del_item && !encoding) {
2696                                 u64 orig_num_bytes =
2697                                         btrfs_file_extent_num_bytes(leaf, fi);
2698                                 extent_num_bytes = new_size -
2699                                         found_key.offset + root->sectorsize - 1;
2700                                 extent_num_bytes = extent_num_bytes &
2701                                         ~((u64)root->sectorsize - 1);
2702                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2703                                                          extent_num_bytes);
2704                                 num_dec = (orig_num_bytes -
2705                                            extent_num_bytes);
2706                                 if (root->ref_cows && extent_start != 0)
2707                                         inode_sub_bytes(inode, num_dec);
2708                                 btrfs_mark_buffer_dirty(leaf);
2709                         } else {
2710                                 extent_num_bytes =
2711                                         btrfs_file_extent_disk_num_bytes(leaf,
2712                                                                          fi);
2713                                 extent_offset = found_key.offset -
2714                                         btrfs_file_extent_offset(leaf, fi);
2715
2716                                 /* FIXME blocksize != 4096 */
2717                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2718                                 if (extent_start != 0) {
2719                                         found_extent = 1;
2720                                         if (root->ref_cows)
2721                                                 inode_sub_bytes(inode, num_dec);
2722                                 }
2723                         }
2724                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2725                         /*
2726                          * we can't truncate inline items that have had
2727                          * special encodings
2728                          */
2729                         if (!del_item &&
2730                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2731                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2732                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2733                                 u32 size = new_size - found_key.offset;
2734
2735                                 if (root->ref_cows) {
2736                                         inode_sub_bytes(inode, item_end + 1 -
2737                                                         new_size);
2738                                 }
2739                                 size =
2740                                     btrfs_file_extent_calc_inline_size(size);
2741                                 ret = btrfs_truncate_item(trans, root, path,
2742                                                           size, 1);
2743                                 BUG_ON(ret);
2744                         } else if (root->ref_cows) {
2745                                 inode_sub_bytes(inode, item_end + 1 -
2746                                                 found_key.offset);
2747                         }
2748                 }
2749 delete:
2750                 if (del_item) {
2751                         if (!pending_del_nr) {
2752                                 /* no pending yet, add ourselves */
2753                                 pending_del_slot = path->slots[0];
2754                                 pending_del_nr = 1;
2755                         } else if (pending_del_nr &&
2756                                    path->slots[0] + 1 == pending_del_slot) {
2757                                 /* hop on the pending chunk */
2758                                 pending_del_nr++;
2759                                 pending_del_slot = path->slots[0];
2760                         } else {
2761                                 BUG();
2762                         }
2763                 } else {
2764                         break;
2765                 }
2766                 if (found_extent && root->ref_cows) {
2767                         btrfs_set_path_blocking(path);
2768                         ret = btrfs_free_extent(trans, root, extent_start,
2769                                                 extent_num_bytes, 0,
2770                                                 btrfs_header_owner(leaf),
2771                                                 inode->i_ino, extent_offset);
2772                         BUG_ON(ret);
2773                 }
2774 next:
2775                 if (path->slots[0] == 0) {
2776                         if (pending_del_nr)
2777                                 goto del_pending;
2778                         btrfs_release_path(root, path);
2779                         if (found_type == BTRFS_INODE_ITEM_KEY)
2780                                 break;
2781                         goto search_again;
2782                 }
2783
2784                 path->slots[0]--;
2785                 if (pending_del_nr &&
2786                     path->slots[0] + 1 != pending_del_slot) {
2787                         struct btrfs_key debug;
2788 del_pending:
2789                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2790                                               pending_del_slot);
2791                         ret = btrfs_del_items(trans, root, path,
2792                                               pending_del_slot,
2793                                               pending_del_nr);
2794                         BUG_ON(ret);
2795                         pending_del_nr = 0;
2796                         btrfs_release_path(root, path);
2797                         if (found_type == BTRFS_INODE_ITEM_KEY)
2798                                 break;
2799                         goto search_again;
2800                 }
2801         }
2802         ret = 0;
2803 error:
2804         if (pending_del_nr) {
2805                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2806                                       pending_del_nr);
2807         }
2808         btrfs_free_path(path);
2809         inode->i_sb->s_dirt = 1;
2810         return ret;
2811 }
2812
2813 /*
2814  * taken from block_truncate_page, but does cow as it zeros out
2815  * any bytes left in the last page in the file.
2816  */
2817 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2818 {
2819         struct inode *inode = mapping->host;
2820         struct btrfs_root *root = BTRFS_I(inode)->root;
2821         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2822         struct btrfs_ordered_extent *ordered;
2823         char *kaddr;
2824         u32 blocksize = root->sectorsize;
2825         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2826         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2827         struct page *page;
2828         int ret = 0;
2829         u64 page_start;
2830         u64 page_end;
2831
2832         if ((offset & (blocksize - 1)) == 0)
2833                 goto out;
2834
2835         ret = -ENOMEM;
2836 again:
2837         page = grab_cache_page(mapping, index);
2838         if (!page)
2839                 goto out;
2840
2841         page_start = page_offset(page);
2842         page_end = page_start + PAGE_CACHE_SIZE - 1;
2843
2844         if (!PageUptodate(page)) {
2845                 ret = btrfs_readpage(NULL, page);
2846                 lock_page(page);
2847                 if (page->mapping != mapping) {
2848                         unlock_page(page);
2849                         page_cache_release(page);
2850                         goto again;
2851                 }
2852                 if (!PageUptodate(page)) {
2853                         ret = -EIO;
2854                         goto out_unlock;
2855                 }
2856         }
2857         wait_on_page_writeback(page);
2858
2859         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2860         set_page_extent_mapped(page);
2861
2862         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2863         if (ordered) {
2864                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2865                 unlock_page(page);
2866                 page_cache_release(page);
2867                 btrfs_start_ordered_extent(inode, ordered, 1);
2868                 btrfs_put_ordered_extent(ordered);
2869                 goto again;
2870         }
2871
2872         btrfs_set_extent_delalloc(inode, page_start, page_end);
2873         ret = 0;
2874         if (offset != PAGE_CACHE_SIZE) {
2875                 kaddr = kmap(page);
2876                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2877                 flush_dcache_page(page);
2878                 kunmap(page);
2879         }
2880         ClearPageChecked(page);
2881         set_page_dirty(page);
2882         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2883
2884 out_unlock:
2885         unlock_page(page);
2886         page_cache_release(page);
2887 out:
2888         return ret;
2889 }
2890
2891 int btrfs_cont_expand(struct inode *inode, loff_t size)
2892 {
2893         struct btrfs_trans_handle *trans;
2894         struct btrfs_root *root = BTRFS_I(inode)->root;
2895         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2896         struct extent_map *em;
2897         u64 mask = root->sectorsize - 1;
2898         u64 hole_start = (inode->i_size + mask) & ~mask;
2899         u64 block_end = (size + mask) & ~mask;
2900         u64 last_byte;
2901         u64 cur_offset;
2902         u64 hole_size;
2903         int err;
2904
2905         if (size <= hole_start)
2906                 return 0;
2907
2908         err = btrfs_check_metadata_free_space(root);
2909         if (err)
2910                 return err;
2911
2912         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2913
2914         while (1) {
2915                 struct btrfs_ordered_extent *ordered;
2916                 btrfs_wait_ordered_range(inode, hole_start,
2917                                          block_end - hole_start);
2918                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2919                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2920                 if (!ordered)
2921                         break;
2922                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2923                 btrfs_put_ordered_extent(ordered);
2924         }
2925
2926         trans = btrfs_start_transaction(root, 1);
2927         btrfs_set_trans_block_group(trans, inode);
2928
2929         cur_offset = hole_start;
2930         while (1) {
2931                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2932                                 block_end - cur_offset, 0);
2933                 BUG_ON(IS_ERR(em) || !em);
2934                 last_byte = min(extent_map_end(em), block_end);
2935                 last_byte = (last_byte + mask) & ~mask;
2936                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2937                         u64 hint_byte = 0;
2938                         hole_size = last_byte - cur_offset;
2939                         err = btrfs_drop_extents(trans, root, inode,
2940                                                  cur_offset,
2941                                                  cur_offset + hole_size,
2942                                                  block_end,
2943                                                  cur_offset, &hint_byte);
2944                         if (err)
2945                                 break;
2946                         err = btrfs_insert_file_extent(trans, root,
2947                                         inode->i_ino, cur_offset, 0,
2948                                         0, hole_size, 0, hole_size,
2949                                         0, 0, 0);
2950                         btrfs_drop_extent_cache(inode, hole_start,
2951                                         last_byte - 1, 0);
2952                 }
2953                 free_extent_map(em);
2954                 cur_offset = last_byte;
2955                 if (err || cur_offset >= block_end)
2956                         break;
2957         }
2958
2959         btrfs_end_transaction(trans, root);
2960         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2961         return err;
2962 }
2963
2964 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2965 {
2966         struct inode *inode = dentry->d_inode;
2967         int err;
2968
2969         err = inode_change_ok(inode, attr);
2970         if (err)
2971                 return err;
2972
2973         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2974                 if (attr->ia_size > inode->i_size) {
2975                         err = btrfs_cont_expand(inode, attr->ia_size);
2976                         if (err)
2977                                 return err;
2978                 } else if (inode->i_size > 0 &&
2979                            attr->ia_size == 0) {
2980
2981                         /* we're truncating a file that used to have good
2982                          * data down to zero.  Make sure it gets into
2983                          * the ordered flush list so that any new writes
2984                          * get down to disk quickly.
2985                          */
2986                         BTRFS_I(inode)->ordered_data_close = 1;
2987                 }
2988         }
2989
2990         err = inode_setattr(inode, attr);
2991
2992         if (!err && ((attr->ia_valid & ATTR_MODE)))
2993                 err = btrfs_acl_chmod(inode);
2994         return err;
2995 }
2996
2997 void btrfs_delete_inode(struct inode *inode)
2998 {
2999         struct btrfs_trans_handle *trans;
3000         struct btrfs_root *root = BTRFS_I(inode)->root;
3001         unsigned long nr;
3002         int ret;
3003
3004         truncate_inode_pages(&inode->i_data, 0);
3005         if (is_bad_inode(inode)) {
3006                 btrfs_orphan_del(NULL, inode);
3007                 goto no_delete;
3008         }
3009         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3010
3011         btrfs_i_size_write(inode, 0);
3012         trans = btrfs_join_transaction(root, 1);
3013
3014         btrfs_set_trans_block_group(trans, inode);
3015         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3016         if (ret) {
3017                 btrfs_orphan_del(NULL, inode);
3018                 goto no_delete_lock;
3019         }
3020
3021         btrfs_orphan_del(trans, inode);
3022
3023         nr = trans->blocks_used;
3024         clear_inode(inode);
3025
3026         btrfs_end_transaction(trans, root);
3027         btrfs_btree_balance_dirty(root, nr);
3028         return;
3029
3030 no_delete_lock:
3031         nr = trans->blocks_used;
3032         btrfs_end_transaction(trans, root);
3033         btrfs_btree_balance_dirty(root, nr);
3034 no_delete:
3035         clear_inode(inode);
3036 }
3037
3038 /*
3039  * this returns the key found in the dir entry in the location pointer.
3040  * If no dir entries were found, location->objectid is 0.
3041  */
3042 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3043                                struct btrfs_key *location)
3044 {
3045         const char *name = dentry->d_name.name;
3046         int namelen = dentry->d_name.len;
3047         struct btrfs_dir_item *di;
3048         struct btrfs_path *path;
3049         struct btrfs_root *root = BTRFS_I(dir)->root;
3050         int ret = 0;
3051
3052         path = btrfs_alloc_path();
3053         BUG_ON(!path);
3054