]> bbs.cooldavid.org Git - net-next-2.6.git/blob - fs/btrfs/inode.c
Btrfs: allow unaligned DIO
[net-next-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.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 const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const 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 btrfs_trans_handle *trans,
93                                      struct inode *inode,  struct inode *dir)
94 {
95         int err;
96
97         err = btrfs_init_acl(trans, inode, dir);
98         if (!err)
99                 err = btrfs_xattr_security_init(trans, inode, dir);
100         return err;
101 }
102
103 /*
104  * this does all the hard work for inserting an inline extent into
105  * the btree.  The caller should have done a btrfs_drop_extents so that
106  * no overlapping inline items exist in the btree
107  */
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109                                 struct btrfs_root *root, struct inode *inode,
110                                 u64 start, size_t size, size_t compressed_size,
111                                 struct page **compressed_pages)
112 {
113         struct btrfs_key key;
114         struct btrfs_path *path;
115         struct extent_buffer *leaf;
116         struct page *page = NULL;
117         char *kaddr;
118         unsigned long ptr;
119         struct btrfs_file_extent_item *ei;
120         int err = 0;
121         int ret;
122         size_t cur_size = size;
123         size_t datasize;
124         unsigned long offset;
125         int use_compress = 0;
126
127         if (compressed_size && compressed_pages) {
128                 use_compress = 1;
129                 cur_size = compressed_size;
130         }
131
132         path = btrfs_alloc_path();
133         if (!path)
134                 return -ENOMEM;
135
136         path->leave_spinning = 1;
137         btrfs_set_trans_block_group(trans, inode);
138
139         key.objectid = inode->i_ino;
140         key.offset = start;
141         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142         datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144         inode_add_bytes(inode, size);
145         ret = btrfs_insert_empty_item(trans, root, path, &key,
146                                       datasize);
147         BUG_ON(ret);
148         if (ret) {
149                 err = ret;
150                 goto fail;
151         }
152         leaf = path->nodes[0];
153         ei = btrfs_item_ptr(leaf, path->slots[0],
154                             struct btrfs_file_extent_item);
155         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157         btrfs_set_file_extent_encryption(leaf, ei, 0);
158         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160         ptr = btrfs_file_extent_inline_start(ei);
161
162         if (use_compress) {
163                 struct page *cpage;
164                 int i = 0;
165                 while (compressed_size > 0) {
166                         cpage = compressed_pages[i];
167                         cur_size = min_t(unsigned long, compressed_size,
168                                        PAGE_CACHE_SIZE);
169
170                         kaddr = kmap_atomic(cpage, KM_USER0);
171                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
172                         kunmap_atomic(kaddr, KM_USER0);
173
174                         i++;
175                         ptr += cur_size;
176                         compressed_size -= cur_size;
177                 }
178                 btrfs_set_file_extent_compression(leaf, ei,
179                                                   BTRFS_COMPRESS_ZLIB);
180         } else {
181                 page = find_get_page(inode->i_mapping,
182                                      start >> PAGE_CACHE_SHIFT);
183                 btrfs_set_file_extent_compression(leaf, ei, 0);
184                 kaddr = kmap_atomic(page, KM_USER0);
185                 offset = start & (PAGE_CACHE_SIZE - 1);
186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187                 kunmap_atomic(kaddr, KM_USER0);
188                 page_cache_release(page);
189         }
190         btrfs_mark_buffer_dirty(leaf);
191         btrfs_free_path(path);
192
193         /*
194          * we're an inline extent, so nobody can
195          * extend the file past i_size without locking
196          * a page we already have locked.
197          *
198          * We must do any isize and inode updates
199          * before we unlock the pages.  Otherwise we
200          * could end up racing with unlink.
201          */
202         BTRFS_I(inode)->disk_i_size = inode->i_size;
203         btrfs_update_inode(trans, root, inode);
204
205         return 0;
206 fail:
207         btrfs_free_path(path);
208         return err;
209 }
210
211
212 /*
213  * conditionally insert an inline extent into the file.  This
214  * does the checks required to make sure the data is small enough
215  * to fit as an inline extent.
216  */
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218                                  struct btrfs_root *root,
219                                  struct inode *inode, u64 start, u64 end,
220                                  size_t compressed_size,
221                                  struct page **compressed_pages)
222 {
223         u64 isize = i_size_read(inode);
224         u64 actual_end = min(end + 1, isize);
225         u64 inline_len = actual_end - start;
226         u64 aligned_end = (end + root->sectorsize - 1) &
227                         ~((u64)root->sectorsize - 1);
228         u64 hint_byte;
229         u64 data_len = inline_len;
230         int ret;
231
232         if (compressed_size)
233                 data_len = compressed_size;
234
235         if (start > 0 ||
236             actual_end >= PAGE_CACHE_SIZE ||
237             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238             (!compressed_size &&
239             (actual_end & (root->sectorsize - 1)) == 0) ||
240             end + 1 < isize ||
241             data_len > root->fs_info->max_inline) {
242                 return 1;
243         }
244
245         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
246                                  &hint_byte, 1);
247         BUG_ON(ret);
248
249         if (isize > actual_end)
250                 inline_len = min_t(u64, isize, actual_end);
251         ret = insert_inline_extent(trans, root, inode, start,
252                                    inline_len, compressed_size,
253                                    compressed_pages);
254         BUG_ON(ret);
255         btrfs_delalloc_release_metadata(inode, end + 1 - start);
256         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
257         return 0;
258 }
259
260 struct async_extent {
261         u64 start;
262         u64 ram_size;
263         u64 compressed_size;
264         struct page **pages;
265         unsigned long nr_pages;
266         struct list_head list;
267 };
268
269 struct async_cow {
270         struct inode *inode;
271         struct btrfs_root *root;
272         struct page *locked_page;
273         u64 start;
274         u64 end;
275         struct list_head extents;
276         struct btrfs_work work;
277 };
278
279 static noinline int add_async_extent(struct async_cow *cow,
280                                      u64 start, u64 ram_size,
281                                      u64 compressed_size,
282                                      struct page **pages,
283                                      unsigned long nr_pages)
284 {
285         struct async_extent *async_extent;
286
287         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
288         async_extent->start = start;
289         async_extent->ram_size = ram_size;
290         async_extent->compressed_size = compressed_size;
291         async_extent->pages = pages;
292         async_extent->nr_pages = nr_pages;
293         list_add_tail(&async_extent->list, &cow->extents);
294         return 0;
295 }
296
297 /*
298  * we create compressed extents in two phases.  The first
299  * phase compresses a range of pages that have already been
300  * locked (both pages and state bits are locked).
301  *
302  * This is done inside an ordered work queue, and the compression
303  * is spread across many cpus.  The actual IO submission is step
304  * two, and the ordered work queue takes care of making sure that
305  * happens in the same order things were put onto the queue by
306  * writepages and friends.
307  *
308  * If this code finds it can't get good compression, it puts an
309  * entry onto the work queue to write the uncompressed bytes.  This
310  * makes sure that both compressed inodes and uncompressed inodes
311  * are written in the same order that pdflush sent them down.
312  */
313 static noinline int compress_file_range(struct inode *inode,
314                                         struct page *locked_page,
315                                         u64 start, u64 end,
316                                         struct async_cow *async_cow,
317                                         int *num_added)
318 {
319         struct btrfs_root *root = BTRFS_I(inode)->root;
320         struct btrfs_trans_handle *trans;
321         u64 num_bytes;
322         u64 orig_start;
323         u64 disk_num_bytes;
324         u64 blocksize = root->sectorsize;
325         u64 actual_end;
326         u64 isize = i_size_read(inode);
327         int ret = 0;
328         struct page **pages = NULL;
329         unsigned long nr_pages;
330         unsigned long nr_pages_ret = 0;
331         unsigned long total_compressed = 0;
332         unsigned long total_in = 0;
333         unsigned long max_compressed = 128 * 1024;
334         unsigned long max_uncompressed = 128 * 1024;
335         int i;
336         int will_compress;
337
338         orig_start = start;
339
340         actual_end = min_t(u64, isize, end + 1);
341 again:
342         will_compress = 0;
343         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
345
346         /*
347          * we don't want to send crud past the end of i_size through
348          * compression, that's just a waste of CPU time.  So, if the
349          * end of the file is before the start of our current
350          * requested range of bytes, we bail out to the uncompressed
351          * cleanup code that can deal with all of this.
352          *
353          * It isn't really the fastest way to fix things, but this is a
354          * very uncommon corner.
355          */
356         if (actual_end <= start)
357                 goto cleanup_and_bail_uncompressed;
358
359         total_compressed = actual_end - start;
360
361         /* we want to make sure that amount of ram required to uncompress
362          * an extent is reasonable, so we limit the total size in ram
363          * of a compressed extent to 128k.  This is a crucial number
364          * because it also controls how easily we can spread reads across
365          * cpus for decompression.
366          *
367          * We also want to make sure the amount of IO required to do
368          * a random read is reasonably small, so we limit the size of
369          * a compressed extent to 128k.
370          */
371         total_compressed = min(total_compressed, max_uncompressed);
372         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373         num_bytes = max(blocksize,  num_bytes);
374         disk_num_bytes = num_bytes;
375         total_in = 0;
376         ret = 0;
377
378         /*
379          * we do compression for mount -o compress and when the
380          * inode has not been flagged as nocompress.  This flag can
381          * change at any time if we discover bad compression ratios.
382          */
383         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384             (btrfs_test_opt(root, COMPRESS) ||
385              (BTRFS_I(inode)->force_compress))) {
386                 WARN_ON(pages);
387                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388
389                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
390                                                 total_compressed, pages,
391                                                 nr_pages, &nr_pages_ret,
392                                                 &total_in,
393                                                 &total_compressed,
394                                                 max_compressed);
395
396                 if (!ret) {
397                         unsigned long offset = total_compressed &
398                                 (PAGE_CACHE_SIZE - 1);
399                         struct page *page = pages[nr_pages_ret - 1];
400                         char *kaddr;
401
402                         /* zero the tail end of the last page, we might be
403                          * sending it down to disk
404                          */
405                         if (offset) {
406                                 kaddr = kmap_atomic(page, KM_USER0);
407                                 memset(kaddr + offset, 0,
408                                        PAGE_CACHE_SIZE - offset);
409                                 kunmap_atomic(kaddr, KM_USER0);
410                         }
411                         will_compress = 1;
412                 }
413         }
414         if (start == 0) {
415                 trans = btrfs_join_transaction(root, 1);
416                 BUG_ON(!trans);
417                 btrfs_set_trans_block_group(trans, inode);
418                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
419
420                 /* lets try to make an inline extent */
421                 if (ret || total_in < (actual_end - start)) {
422                         /* we didn't compress the entire range, try
423                          * to make an uncompressed inline extent.
424                          */
425                         ret = cow_file_range_inline(trans, root, inode,
426                                                     start, end, 0, NULL);
427                 } else {
428                         /* try making a compressed inline extent */
429                         ret = cow_file_range_inline(trans, root, inode,
430                                                     start, end,
431                                                     total_compressed, pages);
432                 }
433                 if (ret == 0) {
434                         /*
435                          * inline extent creation worked, we don't need
436                          * to create any more async work items.  Unlock
437                          * and free up our temp pages.
438                          */
439                         extent_clear_unlock_delalloc(inode,
440                              &BTRFS_I(inode)->io_tree,
441                              start, end, NULL,
442                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
443                              EXTENT_CLEAR_DELALLOC |
444                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
445
446                         btrfs_end_transaction(trans, root);
447                         goto free_pages_out;
448                 }
449                 btrfs_end_transaction(trans, root);
450         }
451
452         if (will_compress) {
453                 /*
454                  * we aren't doing an inline extent round the compressed size
455                  * up to a block size boundary so the allocator does sane
456                  * things
457                  */
458                 total_compressed = (total_compressed + blocksize - 1) &
459                         ~(blocksize - 1);
460
461                 /*
462                  * one last check to make sure the compression is really a
463                  * win, compare the page count read with the blocks on disk
464                  */
465                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
466                         ~(PAGE_CACHE_SIZE - 1);
467                 if (total_compressed >= total_in) {
468                         will_compress = 0;
469                 } else {
470                         disk_num_bytes = total_compressed;
471                         num_bytes = total_in;
472                 }
473         }
474         if (!will_compress && pages) {
475                 /*
476                  * the compression code ran but failed to make things smaller,
477                  * free any pages it allocated and our page pointer array
478                  */
479                 for (i = 0; i < nr_pages_ret; i++) {
480                         WARN_ON(pages[i]->mapping);
481                         page_cache_release(pages[i]);
482                 }
483                 kfree(pages);
484                 pages = NULL;
485                 total_compressed = 0;
486                 nr_pages_ret = 0;
487
488                 /* flag the file so we don't compress in the future */
489                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
490                     !(BTRFS_I(inode)->force_compress)) {
491                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
492                 }
493         }
494         if (will_compress) {
495                 *num_added += 1;
496
497                 /* the async work queues will take care of doing actual
498                  * allocation on disk for these compressed pages,
499                  * and will submit them to the elevator.
500                  */
501                 add_async_extent(async_cow, start, num_bytes,
502                                  total_compressed, pages, nr_pages_ret);
503
504                 if (start + num_bytes < end && start + num_bytes < actual_end) {
505                         start += num_bytes;
506                         pages = NULL;
507                         cond_resched();
508                         goto again;
509                 }
510         } else {
511 cleanup_and_bail_uncompressed:
512                 /*
513                  * No compression, but we still need to write the pages in
514                  * the file we've been given so far.  redirty the locked
515                  * page if it corresponds to our extent and set things up
516                  * for the async work queue to run cow_file_range to do
517                  * the normal delalloc dance
518                  */
519                 if (page_offset(locked_page) >= start &&
520                     page_offset(locked_page) <= end) {
521                         __set_page_dirty_nobuffers(locked_page);
522                         /* unlocked later on in the async handlers */
523                 }
524                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
525                 *num_added += 1;
526         }
527
528 out:
529         return 0;
530
531 free_pages_out:
532         for (i = 0; i < nr_pages_ret; i++) {
533                 WARN_ON(pages[i]->mapping);
534                 page_cache_release(pages[i]);
535         }
536         kfree(pages);
537
538         goto out;
539 }
540
541 /*
542  * phase two of compressed writeback.  This is the ordered portion
543  * of the code, which only gets called in the order the work was
544  * queued.  We walk all the async extents created by compress_file_range
545  * and send them down to the disk.
546  */
547 static noinline int submit_compressed_extents(struct inode *inode,
548                                               struct async_cow *async_cow)
549 {
550         struct async_extent *async_extent;
551         u64 alloc_hint = 0;
552         struct btrfs_trans_handle *trans;
553         struct btrfs_key ins;
554         struct extent_map *em;
555         struct btrfs_root *root = BTRFS_I(inode)->root;
556         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
557         struct extent_io_tree *io_tree;
558         int ret = 0;
559
560         if (list_empty(&async_cow->extents))
561                 return 0;
562
563
564         while (!list_empty(&async_cow->extents)) {
565                 async_extent = list_entry(async_cow->extents.next,
566                                           struct async_extent, list);
567                 list_del(&async_extent->list);
568
569                 io_tree = &BTRFS_I(inode)->io_tree;
570
571 retry:
572                 /* did the compression code fall back to uncompressed IO? */
573                 if (!async_extent->pages) {
574                         int page_started = 0;
575                         unsigned long nr_written = 0;
576
577                         lock_extent(io_tree, async_extent->start,
578                                          async_extent->start +
579                                          async_extent->ram_size - 1, GFP_NOFS);
580
581                         /* allocate blocks */
582                         ret = cow_file_range(inode, async_cow->locked_page,
583                                              async_extent->start,
584                                              async_extent->start +
585                                              async_extent->ram_size - 1,
586                                              &page_started, &nr_written, 0);
587
588                         /*
589                          * if page_started, cow_file_range inserted an
590                          * inline extent and took care of all the unlocking
591                          * and IO for us.  Otherwise, we need to submit
592                          * all those pages down to the drive.
593                          */
594                         if (!page_started && !ret)
595                                 extent_write_locked_range(io_tree,
596                                                   inode, async_extent->start,
597                                                   async_extent->start +
598                                                   async_extent->ram_size - 1,
599                                                   btrfs_get_extent,
600                                                   WB_SYNC_ALL);
601                         kfree(async_extent);
602                         cond_resched();
603                         continue;
604                 }
605
606                 lock_extent(io_tree, async_extent->start,
607                             async_extent->start + async_extent->ram_size - 1,
608                             GFP_NOFS);
609
610                 trans = btrfs_join_transaction(root, 1);
611                 ret = btrfs_reserve_extent(trans, root,
612                                            async_extent->compressed_size,
613                                            async_extent->compressed_size,
614                                            0, alloc_hint,
615                                            (u64)-1, &ins, 1);
616                 btrfs_end_transaction(trans, root);
617
618                 if (ret) {
619                         int i;
620                         for (i = 0; i < async_extent->nr_pages; i++) {
621                                 WARN_ON(async_extent->pages[i]->mapping);
622                                 page_cache_release(async_extent->pages[i]);
623                         }
624                         kfree(async_extent->pages);
625                         async_extent->nr_pages = 0;
626                         async_extent->pages = NULL;
627                         unlock_extent(io_tree, async_extent->start,
628                                       async_extent->start +
629                                       async_extent->ram_size - 1, GFP_NOFS);
630                         goto retry;
631                 }
632
633                 /*
634                  * here we're doing allocation and writeback of the
635                  * compressed pages
636                  */
637                 btrfs_drop_extent_cache(inode, async_extent->start,
638                                         async_extent->start +
639                                         async_extent->ram_size - 1, 0);
640
641                 em = alloc_extent_map(GFP_NOFS);
642                 em->start = async_extent->start;
643                 em->len = async_extent->ram_size;
644                 em->orig_start = em->start;
645
646                 em->block_start = ins.objectid;
647                 em->block_len = ins.offset;
648                 em->bdev = root->fs_info->fs_devices->latest_bdev;
649                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
650                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
651
652                 while (1) {
653                         write_lock(&em_tree->lock);
654                         ret = add_extent_mapping(em_tree, em);
655                         write_unlock(&em_tree->lock);
656                         if (ret != -EEXIST) {
657                                 free_extent_map(em);
658                                 break;
659                         }
660                         btrfs_drop_extent_cache(inode, async_extent->start,
661                                                 async_extent->start +
662                                                 async_extent->ram_size - 1, 0);
663                 }
664
665                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
666                                                ins.objectid,
667                                                async_extent->ram_size,
668                                                ins.offset,
669                                                BTRFS_ORDERED_COMPRESSED);
670                 BUG_ON(ret);
671
672                 /*
673                  * clear dirty, set writeback and unlock the pages.
674                  */
675                 extent_clear_unlock_delalloc(inode,
676                                 &BTRFS_I(inode)->io_tree,
677                                 async_extent->start,
678                                 async_extent->start +
679                                 async_extent->ram_size - 1,
680                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
681                                 EXTENT_CLEAR_UNLOCK |
682                                 EXTENT_CLEAR_DELALLOC |
683                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
684
685                 ret = btrfs_submit_compressed_write(inode,
686                                     async_extent->start,
687                                     async_extent->ram_size,
688                                     ins.objectid,
689                                     ins.offset, async_extent->pages,
690                                     async_extent->nr_pages);
691
692                 BUG_ON(ret);
693                 alloc_hint = ins.objectid + ins.offset;
694                 kfree(async_extent);
695                 cond_resched();
696         }
697
698         return 0;
699 }
700
701 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
702                                       u64 num_bytes)
703 {
704         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
705         struct extent_map *em;
706         u64 alloc_hint = 0;
707
708         read_lock(&em_tree->lock);
709         em = search_extent_mapping(em_tree, start, num_bytes);
710         if (em) {
711                 /*
712                  * if block start isn't an actual block number then find the
713                  * first block in this inode and use that as a hint.  If that
714                  * block is also bogus then just don't worry about it.
715                  */
716                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
717                         free_extent_map(em);
718                         em = search_extent_mapping(em_tree, 0, 0);
719                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
720                                 alloc_hint = em->block_start;
721                         if (em)
722                                 free_extent_map(em);
723                 } else {
724                         alloc_hint = em->block_start;
725                         free_extent_map(em);
726                 }
727         }
728         read_unlock(&em_tree->lock);
729
730         return alloc_hint;
731 }
732
733 /*
734  * when extent_io.c finds a delayed allocation range in the file,
735  * the call backs end up in this code.  The basic idea is to
736  * allocate extents on disk for the range, and create ordered data structs
737  * in ram to track those extents.
738  *
739  * locked_page is the page that writepage had locked already.  We use
740  * it to make sure we don't do extra locks or unlocks.
741  *
742  * *page_started is set to one if we unlock locked_page and do everything
743  * required to start IO on it.  It may be clean and already done with
744  * IO when we return.
745  */
746 static noinline int cow_file_range(struct inode *inode,
747                                    struct page *locked_page,
748                                    u64 start, u64 end, int *page_started,
749                                    unsigned long *nr_written,
750                                    int unlock)
751 {
752         struct btrfs_root *root = BTRFS_I(inode)->root;
753         struct btrfs_trans_handle *trans;
754         u64 alloc_hint = 0;
755         u64 num_bytes;
756         unsigned long ram_size;
757         u64 disk_num_bytes;
758         u64 cur_alloc_size;
759         u64 blocksize = root->sectorsize;
760         u64 actual_end;
761         u64 isize = i_size_read(inode);
762         struct btrfs_key ins;
763         struct extent_map *em;
764         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
765         int ret = 0;
766
767         trans = btrfs_join_transaction(root, 1);
768         BUG_ON(!trans);
769         btrfs_set_trans_block_group(trans, inode);
770         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771
772         actual_end = min_t(u64, isize, end + 1);
773
774         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
775         num_bytes = max(blocksize,  num_bytes);
776         disk_num_bytes = num_bytes;
777         ret = 0;
778
779         if (start == 0) {
780                 /* lets try to make an inline extent */
781                 ret = cow_file_range_inline(trans, root, inode,
782                                             start, end, 0, NULL);
783                 if (ret == 0) {
784                         extent_clear_unlock_delalloc(inode,
785                                      &BTRFS_I(inode)->io_tree,
786                                      start, end, NULL,
787                                      EXTENT_CLEAR_UNLOCK_PAGE |
788                                      EXTENT_CLEAR_UNLOCK |
789                                      EXTENT_CLEAR_DELALLOC |
790                                      EXTENT_CLEAR_DIRTY |
791                                      EXTENT_SET_WRITEBACK |
792                                      EXTENT_END_WRITEBACK);
793
794                         *nr_written = *nr_written +
795                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
796                         *page_started = 1;
797                         ret = 0;
798                         goto out;
799                 }
800         }
801
802         BUG_ON(disk_num_bytes >
803                btrfs_super_total_bytes(&root->fs_info->super_copy));
804
805         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
806         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
807
808         while (disk_num_bytes > 0) {
809                 unsigned long op;
810
811                 cur_alloc_size = disk_num_bytes;
812                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
813                                            root->sectorsize, 0, alloc_hint,
814                                            (u64)-1, &ins, 1);
815                 BUG_ON(ret);
816
817                 em = alloc_extent_map(GFP_NOFS);
818                 em->start = start;
819                 em->orig_start = em->start;
820                 ram_size = ins.offset;
821                 em->len = ins.offset;
822
823                 em->block_start = ins.objectid;
824                 em->block_len = ins.offset;
825                 em->bdev = root->fs_info->fs_devices->latest_bdev;
826                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
827
828                 while (1) {
829                         write_lock(&em_tree->lock);
830                         ret = add_extent_mapping(em_tree, em);
831                         write_unlock(&em_tree->lock);
832                         if (ret != -EEXIST) {
833                                 free_extent_map(em);
834                                 break;
835                         }
836                         btrfs_drop_extent_cache(inode, start,
837                                                 start + ram_size - 1, 0);
838                 }
839
840                 cur_alloc_size = ins.offset;
841                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
842                                                ram_size, cur_alloc_size, 0);
843                 BUG_ON(ret);
844
845                 if (root->root_key.objectid ==
846                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
847                         ret = btrfs_reloc_clone_csums(inode, start,
848                                                       cur_alloc_size);
849                         BUG_ON(ret);
850                 }
851
852                 if (disk_num_bytes < cur_alloc_size)
853                         break;
854
855                 /* we're not doing compressed IO, don't unlock the first
856                  * page (which the caller expects to stay locked), don't
857                  * clear any dirty bits and don't set any writeback bits
858                  *
859                  * Do set the Private2 bit so we know this page was properly
860                  * setup for writepage
861                  */
862                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
863                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
864                         EXTENT_SET_PRIVATE2;
865
866                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
867                                              start, start + ram_size - 1,
868                                              locked_page, op);
869                 disk_num_bytes -= cur_alloc_size;
870                 num_bytes -= cur_alloc_size;
871                 alloc_hint = ins.objectid + ins.offset;
872                 start += cur_alloc_size;
873         }
874 out:
875         ret = 0;
876         btrfs_end_transaction(trans, root);
877
878         return ret;
879 }
880
881 /*
882  * work queue call back to started compression on a file and pages
883  */
884 static noinline void async_cow_start(struct btrfs_work *work)
885 {
886         struct async_cow *async_cow;
887         int num_added = 0;
888         async_cow = container_of(work, struct async_cow, work);
889
890         compress_file_range(async_cow->inode, async_cow->locked_page,
891                             async_cow->start, async_cow->end, async_cow,
892                             &num_added);
893         if (num_added == 0)
894                 async_cow->inode = NULL;
895 }
896
897 /*
898  * work queue call back to submit previously compressed pages
899  */
900 static noinline void async_cow_submit(struct btrfs_work *work)
901 {
902         struct async_cow *async_cow;
903         struct btrfs_root *root;
904         unsigned long nr_pages;
905
906         async_cow = container_of(work, struct async_cow, work);
907
908         root = async_cow->root;
909         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
910                 PAGE_CACHE_SHIFT;
911
912         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
913
914         if (atomic_read(&root->fs_info->async_delalloc_pages) <
915             5 * 1042 * 1024 &&
916             waitqueue_active(&root->fs_info->async_submit_wait))
917                 wake_up(&root->fs_info->async_submit_wait);
918
919         if (async_cow->inode)
920                 submit_compressed_extents(async_cow->inode, async_cow);
921 }
922
923 static noinline void async_cow_free(struct btrfs_work *work)
924 {
925         struct async_cow *async_cow;
926         async_cow = container_of(work, struct async_cow, work);
927         kfree(async_cow);
928 }
929
930 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
931                                 u64 start, u64 end, int *page_started,
932                                 unsigned long *nr_written)
933 {
934         struct async_cow *async_cow;
935         struct btrfs_root *root = BTRFS_I(inode)->root;
936         unsigned long nr_pages;
937         u64 cur_end;
938         int limit = 10 * 1024 * 1042;
939
940         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
941                          1, 0, NULL, GFP_NOFS);
942         while (start < end) {
943                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
944                 async_cow->inode = inode;
945                 async_cow->root = root;
946                 async_cow->locked_page = locked_page;
947                 async_cow->start = start;
948
949                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
950                         cur_end = end;
951                 else
952                         cur_end = min(end, start + 512 * 1024 - 1);
953
954                 async_cow->end = cur_end;
955                 INIT_LIST_HEAD(&async_cow->extents);
956
957                 async_cow->work.func = async_cow_start;
958                 async_cow->work.ordered_func = async_cow_submit;
959                 async_cow->work.ordered_free = async_cow_free;
960                 async_cow->work.flags = 0;
961
962                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
963                         PAGE_CACHE_SHIFT;
964                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
965
966                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
967                                    &async_cow->work);
968
969                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
970                         wait_event(root->fs_info->async_submit_wait,
971                            (atomic_read(&root->fs_info->async_delalloc_pages) <
972                             limit));
973                 }
974
975                 while (atomic_read(&root->fs_info->async_submit_draining) &&
976                       atomic_read(&root->fs_info->async_delalloc_pages)) {
977                         wait_event(root->fs_info->async_submit_wait,
978                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
979                            0));
980                 }
981
982                 *nr_written += nr_pages;
983                 start = cur_end + 1;
984         }
985         *page_started = 1;
986         return 0;
987 }
988
989 static noinline int csum_exist_in_range(struct btrfs_root *root,
990                                         u64 bytenr, u64 num_bytes)
991 {
992         int ret;
993         struct btrfs_ordered_sum *sums;
994         LIST_HEAD(list);
995
996         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
997                                        bytenr + num_bytes - 1, &list);
998         if (ret == 0 && list_empty(&list))
999                 return 0;
1000
1001         while (!list_empty(&list)) {
1002                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1003                 list_del(&sums->list);
1004                 kfree(sums);
1005         }
1006         return 1;
1007 }
1008
1009 /*
1010  * when nowcow writeback call back.  This checks for snapshots or COW copies
1011  * of the extents that exist in the file, and COWs the file as required.
1012  *
1013  * If no cow copies or snapshots exist, we write directly to the existing
1014  * blocks on disk
1015  */
1016 static noinline int run_delalloc_nocow(struct inode *inode,
1017                                        struct page *locked_page,
1018                               u64 start, u64 end, int *page_started, int force,
1019                               unsigned long *nr_written)
1020 {
1021         struct btrfs_root *root = BTRFS_I(inode)->root;
1022         struct btrfs_trans_handle *trans;
1023         struct extent_buffer *leaf;
1024         struct btrfs_path *path;
1025         struct btrfs_file_extent_item *fi;
1026         struct btrfs_key found_key;
1027         u64 cow_start;
1028         u64 cur_offset;
1029         u64 extent_end;
1030         u64 extent_offset;
1031         u64 disk_bytenr;
1032         u64 num_bytes;
1033         int extent_type;
1034         int ret;
1035         int type;
1036         int nocow;
1037         int check_prev = 1;
1038
1039         path = btrfs_alloc_path();
1040         BUG_ON(!path);
1041         trans = btrfs_join_transaction(root, 1);
1042         BUG_ON(!trans);
1043
1044         cow_start = (u64)-1;
1045         cur_offset = start;
1046         while (1) {
1047                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1048                                                cur_offset, 0);
1049                 BUG_ON(ret < 0);
1050                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1051                         leaf = path->nodes[0];
1052                         btrfs_item_key_to_cpu(leaf, &found_key,
1053                                               path->slots[0] - 1);
1054                         if (found_key.objectid == inode->i_ino &&
1055                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1056                                 path->slots[0]--;
1057                 }
1058                 check_prev = 0;
1059 next_slot:
1060                 leaf = path->nodes[0];
1061                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1062                         ret = btrfs_next_leaf(root, path);
1063                         if (ret < 0)
1064                                 BUG_ON(1);
1065                         if (ret > 0)
1066                                 break;
1067                         leaf = path->nodes[0];
1068                 }
1069
1070                 nocow = 0;
1071                 disk_bytenr = 0;
1072                 num_bytes = 0;
1073                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1074
1075                 if (found_key.objectid > inode->i_ino ||
1076                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1077                     found_key.offset > end)
1078                         break;
1079
1080                 if (found_key.offset > cur_offset) {
1081                         extent_end = found_key.offset;
1082                         extent_type = 0;
1083                         goto out_check;
1084                 }
1085
1086                 fi = btrfs_item_ptr(leaf, path->slots[0],
1087                                     struct btrfs_file_extent_item);
1088                 extent_type = btrfs_file_extent_type(leaf, fi);
1089
1090                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1091                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1092                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1093                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1094                         extent_end = found_key.offset +
1095                                 btrfs_file_extent_num_bytes(leaf, fi);
1096                         if (extent_end <= start) {
1097                                 path->slots[0]++;
1098                                 goto next_slot;
1099                         }
1100                         if (disk_bytenr == 0)
1101                                 goto out_check;
1102                         if (btrfs_file_extent_compression(leaf, fi) ||
1103                             btrfs_file_extent_encryption(leaf, fi) ||
1104                             btrfs_file_extent_other_encoding(leaf, fi))
1105                                 goto out_check;
1106                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1107                                 goto out_check;
1108                         if (btrfs_extent_readonly(root, disk_bytenr))
1109                                 goto out_check;
1110                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1111                                                   found_key.offset -
1112                                                   extent_offset, disk_bytenr))
1113                                 goto out_check;
1114                         disk_bytenr += extent_offset;
1115                         disk_bytenr += cur_offset - found_key.offset;
1116                         num_bytes = min(end + 1, extent_end) - cur_offset;
1117                         /*
1118                          * force cow if csum exists in the range.
1119                          * this ensure that csum for a given extent are
1120                          * either valid or do not exist.
1121                          */
1122                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1123                                 goto out_check;
1124                         nocow = 1;
1125                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1126                         extent_end = found_key.offset +
1127                                 btrfs_file_extent_inline_len(leaf, fi);
1128                         extent_end = ALIGN(extent_end, root->sectorsize);
1129                 } else {
1130                         BUG_ON(1);
1131                 }
1132 out_check:
1133                 if (extent_end <= start) {
1134                         path->slots[0]++;
1135                         goto next_slot;
1136                 }
1137                 if (!nocow) {
1138                         if (cow_start == (u64)-1)
1139                                 cow_start = cur_offset;
1140                         cur_offset = extent_end;
1141                         if (cur_offset > end)
1142                                 break;
1143                         path->slots[0]++;
1144                         goto next_slot;
1145                 }
1146
1147                 btrfs_release_path(root, path);
1148                 if (cow_start != (u64)-1) {
1149                         ret = cow_file_range(inode, locked_page, cow_start,
1150                                         found_key.offset - 1, page_started,
1151                                         nr_written, 1);
1152                         BUG_ON(ret);
1153                         cow_start = (u64)-1;
1154                 }
1155
1156                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1157                         struct extent_map *em;
1158                         struct extent_map_tree *em_tree;
1159                         em_tree = &BTRFS_I(inode)->extent_tree;
1160                         em = alloc_extent_map(GFP_NOFS);
1161                         em->start = cur_offset;
1162                         em->orig_start = em->start;
1163                         em->len = num_bytes;
1164                         em->block_len = num_bytes;
1165                         em->block_start = disk_bytenr;
1166                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1167                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1168                         while (1) {
1169                                 write_lock(&em_tree->lock);
1170                                 ret = add_extent_mapping(em_tree, em);
1171                                 write_unlock(&em_tree->lock);
1172                                 if (ret != -EEXIST) {
1173                                         free_extent_map(em);
1174                                         break;
1175                                 }
1176                                 btrfs_drop_extent_cache(inode, em->start,
1177                                                 em->start + em->len - 1, 0);
1178                         }
1179                         type = BTRFS_ORDERED_PREALLOC;
1180                 } else {
1181                         type = BTRFS_ORDERED_NOCOW;
1182                 }
1183
1184                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1185                                                num_bytes, num_bytes, type);
1186                 BUG_ON(ret);
1187
1188                 if (root->root_key.objectid ==
1189                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1190                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1191                                                       num_bytes);
1192                         BUG_ON(ret);
1193                 }
1194
1195                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1196                                 cur_offset, cur_offset + num_bytes - 1,
1197                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1198                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1199                                 EXTENT_SET_PRIVATE2);
1200                 cur_offset = extent_end;
1201                 if (cur_offset > end)
1202                         break;
1203         }
1204         btrfs_release_path(root, path);
1205
1206         if (cur_offset <= end && cow_start == (u64)-1)
1207                 cow_start = cur_offset;
1208         if (cow_start != (u64)-1) {
1209                 ret = cow_file_range(inode, locked_page, cow_start, end,
1210                                      page_started, nr_written, 1);
1211                 BUG_ON(ret);
1212         }
1213
1214         ret = btrfs_end_transaction(trans, root);
1215         BUG_ON(ret);
1216         btrfs_free_path(path);
1217         return 0;
1218 }
1219
1220 /*
1221  * extent_io.c call back to do delayed allocation processing
1222  */
1223 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1224                               u64 start, u64 end, int *page_started,
1225                               unsigned long *nr_written)
1226 {
1227         int ret;
1228         struct btrfs_root *root = BTRFS_I(inode)->root;
1229
1230         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1231                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1232                                          page_started, 1, nr_written);
1233         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1234                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1235                                          page_started, 0, nr_written);
1236         else if (!btrfs_test_opt(root, COMPRESS) &&
1237                  !(BTRFS_I(inode)->force_compress))
1238                 ret = cow_file_range(inode, locked_page, start, end,
1239                                       page_started, nr_written, 1);
1240         else
1241                 ret = cow_file_range_async(inode, locked_page, start, end,
1242                                            page_started, nr_written);
1243         return ret;
1244 }
1245
1246 static int btrfs_split_extent_hook(struct inode *inode,
1247                                    struct extent_state *orig, u64 split)
1248 {
1249         /* not delalloc, ignore it */
1250         if (!(orig->state & EXTENT_DELALLOC))
1251                 return 0;
1252
1253         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1254         return 0;
1255 }
1256
1257 /*
1258  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1259  * extents so we can keep track of new extents that are just merged onto old
1260  * extents, such as when we are doing sequential writes, so we can properly
1261  * account for the metadata space we'll need.
1262  */
1263 static int btrfs_merge_extent_hook(struct inode *inode,
1264                                    struct extent_state *new,
1265                                    struct extent_state *other)
1266 {
1267         /* not delalloc, ignore it */
1268         if (!(other->state & EXTENT_DELALLOC))
1269                 return 0;
1270
1271         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1272         return 0;
1273 }
1274
1275 /*
1276  * extent_io.c set_bit_hook, used to track delayed allocation
1277  * bytes in this file, and to maintain the list of inodes that
1278  * have pending delalloc work to be done.
1279  */
1280 static int btrfs_set_bit_hook(struct inode *inode,
1281                               struct extent_state *state, int *bits)
1282 {
1283
1284         /*
1285          * set_bit and clear bit hooks normally require _irqsave/restore
1286          * but in this case, we are only testeing for the DELALLOC
1287          * bit, which is only set or cleared with irqs on
1288          */
1289         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1290                 struct btrfs_root *root = BTRFS_I(inode)->root;
1291                 u64 len = state->end + 1 - state->start;
1292
1293                 if (*bits & EXTENT_FIRST_DELALLOC)
1294                         *bits &= ~EXTENT_FIRST_DELALLOC;
1295                 else
1296                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1297
1298                 spin_lock(&root->fs_info->delalloc_lock);
1299                 BTRFS_I(inode)->delalloc_bytes += len;
1300                 root->fs_info->delalloc_bytes += len;
1301                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1302                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1303                                       &root->fs_info->delalloc_inodes);
1304                 }
1305                 spin_unlock(&root->fs_info->delalloc_lock);
1306         }
1307         return 0;
1308 }
1309
1310 /*
1311  * extent_io.c clear_bit_hook, see set_bit_hook for why
1312  */
1313 static int btrfs_clear_bit_hook(struct inode *inode,
1314                                 struct extent_state *state, int *bits)
1315 {
1316         /*
1317          * set_bit and clear bit hooks normally require _irqsave/restore
1318          * but in this case, we are only testeing for the DELALLOC
1319          * bit, which is only set or cleared with irqs on
1320          */
1321         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1322                 struct btrfs_root *root = BTRFS_I(inode)->root;
1323                 u64 len = state->end + 1 - state->start;
1324
1325                 if (*bits & EXTENT_FIRST_DELALLOC)
1326                         *bits &= ~EXTENT_FIRST_DELALLOC;
1327                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1328                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1329
1330                 if (*bits & EXTENT_DO_ACCOUNTING)
1331                         btrfs_delalloc_release_metadata(inode, len);
1332
1333                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1334                         btrfs_free_reserved_data_space(inode, len);
1335
1336                 spin_lock(&root->fs_info->delalloc_lock);
1337                 root->fs_info->delalloc_bytes -= len;
1338                 BTRFS_I(inode)->delalloc_bytes -= len;
1339
1340                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1341                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1342                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1343                 }
1344                 spin_unlock(&root->fs_info->delalloc_lock);
1345         }
1346         return 0;
1347 }
1348
1349 /*
1350  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1351  * we don't create bios that span stripes or chunks
1352  */
1353 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1354                          size_t size, struct bio *bio,
1355                          unsigned long bio_flags)
1356 {
1357         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1358         struct btrfs_mapping_tree *map_tree;
1359         u64 logical = (u64)bio->bi_sector << 9;
1360         u64 length = 0;
1361         u64 map_length;
1362         int ret;
1363
1364         if (bio_flags & EXTENT_BIO_COMPRESSED)
1365                 return 0;
1366
1367         length = bio->bi_size;
1368         map_tree = &root->fs_info->mapping_tree;
1369         map_length = length;
1370         ret = btrfs_map_block(map_tree, READ, logical,
1371                               &map_length, NULL, 0);
1372
1373         if (map_length < length + size)
1374                 return 1;
1375         return 0;
1376 }
1377
1378 /*
1379  * in order to insert checksums into the metadata in large chunks,
1380  * we wait until bio submission time.   All the pages in the bio are
1381  * checksummed and sums are attached onto the ordered extent record.
1382  *
1383  * At IO completion time the cums attached on the ordered extent record
1384  * are inserted into the btree
1385  */
1386 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1387                                     struct bio *bio, int mirror_num,
1388                                     unsigned long bio_flags,
1389                                     u64 bio_offset)
1390 {
1391         struct btrfs_root *root = BTRFS_I(inode)->root;
1392         int ret = 0;
1393
1394         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1395         BUG_ON(ret);
1396         return 0;
1397 }
1398
1399 /*
1400  * in order to insert checksums into the metadata in large chunks,
1401  * we wait until bio submission time.   All the pages in the bio are
1402  * checksummed and sums are attached onto the ordered extent record.
1403  *
1404  * At IO completion time the cums attached on the ordered extent record
1405  * are inserted into the btree
1406  */
1407 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1408                           int mirror_num, unsigned long bio_flags,
1409                           u64 bio_offset)
1410 {
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1413 }
1414
1415 /*
1416  * extent_io.c submission hook. This does the right thing for csum calculation
1417  * on write, or reading the csums from the tree before a read
1418  */
1419 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1420                           int mirror_num, unsigned long bio_flags,
1421                           u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425         int skip_sum;
1426
1427         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1428
1429         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1430         BUG_ON(ret);
1431
1432         if (!(rw & (1 << BIO_RW))) {
1433                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1434                         return btrfs_submit_compressed_read(inode, bio,
1435                                                     mirror_num, bio_flags);
1436                 } else if (!skip_sum)
1437                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1438                 goto mapit;
1439         } else if (!skip_sum) {
1440                 /* csum items have already been cloned */
1441                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1442                         goto mapit;
1443                 /* we're doing a write, do the async checksumming */
1444                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1445                                    inode, rw, bio, mirror_num,
1446                                    bio_flags, bio_offset,
1447                                    __btrfs_submit_bio_start,
1448                                    __btrfs_submit_bio_done);
1449         }
1450
1451 mapit:
1452         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1453 }
1454
1455 /*
1456  * given a list of ordered sums record them in the inode.  This happens
1457  * at IO completion time based on sums calculated at bio submission time.
1458  */
1459 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1460                              struct inode *inode, u64 file_offset,
1461                              struct list_head *list)
1462 {
1463         struct btrfs_ordered_sum *sum;
1464
1465         btrfs_set_trans_block_group(trans, inode);
1466
1467         list_for_each_entry(sum, list, list) {
1468                 btrfs_csum_file_blocks(trans,
1469                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1470         }
1471         return 0;
1472 }
1473
1474 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1475                               struct extent_state **cached_state)
1476 {
1477         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1478                 WARN_ON(1);
1479         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1480                                    cached_state, GFP_NOFS);
1481 }
1482
1483 /* see btrfs_writepage_start_hook for details on why this is required */
1484 struct btrfs_writepage_fixup {
1485         struct page *page;
1486         struct btrfs_work work;
1487 };
1488
1489 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1490 {
1491         struct btrfs_writepage_fixup *fixup;
1492         struct btrfs_ordered_extent *ordered;
1493         struct extent_state *cached_state = NULL;
1494         struct page *page;
1495         struct inode *inode;
1496         u64 page_start;
1497         u64 page_end;
1498
1499         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1500         page = fixup->page;
1501 again:
1502         lock_page(page);
1503         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1504                 ClearPageChecked(page);
1505                 goto out_page;
1506         }
1507
1508         inode = page->mapping->host;
1509         page_start = page_offset(page);
1510         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1511
1512         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1513                          &cached_state, GFP_NOFS);
1514
1515         /* already ordered? We're done */
1516         if (PagePrivate2(page))
1517                 goto out;
1518
1519         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1520         if (ordered) {
1521                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1522                                      page_end, &cached_state, GFP_NOFS);
1523                 unlock_page(page);
1524                 btrfs_start_ordered_extent(inode, ordered, 1);
1525                 goto again;
1526         }
1527
1528         BUG();
1529         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1530         ClearPageChecked(page);
1531 out:
1532         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1533                              &cached_state, GFP_NOFS);
1534 out_page:
1535         unlock_page(page);
1536         page_cache_release(page);
1537 }
1538
1539 /*
1540  * There are a few paths in the higher layers of the kernel that directly
1541  * set the page dirty bit without asking the filesystem if it is a
1542  * good idea.  This causes problems because we want to make sure COW
1543  * properly happens and the data=ordered rules are followed.
1544  *
1545  * In our case any range that doesn't have the ORDERED bit set
1546  * hasn't been properly setup for IO.  We kick off an async process
1547  * to fix it up.  The async helper will wait for ordered extents, set
1548  * the delalloc bit and make it safe to write the page.
1549  */
1550 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1551 {
1552         struct inode *inode = page->mapping->host;
1553         struct btrfs_writepage_fixup *fixup;
1554         struct btrfs_root *root = BTRFS_I(inode)->root;
1555
1556         /* this page is properly in the ordered list */
1557         if (TestClearPagePrivate2(page))
1558                 return 0;
1559
1560         if (PageChecked(page))
1561                 return -EAGAIN;
1562
1563         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1564         if (!fixup)
1565                 return -EAGAIN;
1566
1567         SetPageChecked(page);
1568         page_cache_get(page);
1569         fixup->work.func = btrfs_writepage_fixup_worker;
1570         fixup->page = page;
1571         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1572         return -EAGAIN;
1573 }
1574
1575 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1576                                        struct inode *inode, u64 file_pos,
1577                                        u64 disk_bytenr, u64 disk_num_bytes,
1578                                        u64 num_bytes, u64 ram_bytes,
1579                                        u8 compression, u8 encryption,
1580                                        u16 other_encoding, int extent_type)
1581 {
1582         struct btrfs_root *root = BTRFS_I(inode)->root;
1583         struct btrfs_file_extent_item *fi;
1584         struct btrfs_path *path;
1585         struct extent_buffer *leaf;
1586         struct btrfs_key ins;
1587         u64 hint;
1588         int ret;
1589
1590         path = btrfs_alloc_path();
1591         BUG_ON(!path);
1592
1593         path->leave_spinning = 1;
1594
1595         /*
1596          * we may be replacing one extent in the tree with another.
1597          * The new extent is pinned in the extent map, and we don't want
1598          * to drop it from the cache until it is completely in the btree.
1599          *
1600          * So, tell btrfs_drop_extents to leave this extent in the cache.
1601          * the caller is expected to unpin it and allow it to be merged
1602          * with the others.
1603          */
1604         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1605                                  &hint, 0);
1606         BUG_ON(ret);
1607
1608         ins.objectid = inode->i_ino;
1609         ins.offset = file_pos;
1610         ins.type = BTRFS_EXTENT_DATA_KEY;
1611         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1612         BUG_ON(ret);
1613         leaf = path->nodes[0];
1614         fi = btrfs_item_ptr(leaf, path->slots[0],
1615                             struct btrfs_file_extent_item);
1616         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1617         btrfs_set_file_extent_type(leaf, fi, extent_type);
1618         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1619         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1620         btrfs_set_file_extent_offset(leaf, fi, 0);
1621         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1622         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1623         btrfs_set_file_extent_compression(leaf, fi, compression);
1624         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1625         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1626
1627         btrfs_unlock_up_safe(path, 1);
1628         btrfs_set_lock_blocking(leaf);
1629
1630         btrfs_mark_buffer_dirty(leaf);
1631
1632         inode_add_bytes(inode, num_bytes);
1633
1634         ins.objectid = disk_bytenr;
1635         ins.offset = disk_num_bytes;
1636         ins.type = BTRFS_EXTENT_ITEM_KEY;
1637         ret = btrfs_alloc_reserved_file_extent(trans, root,
1638                                         root->root_key.objectid,
1639                                         inode->i_ino, file_pos, &ins);
1640         BUG_ON(ret);
1641         btrfs_free_path(path);
1642
1643         return 0;
1644 }
1645
1646 /*
1647  * helper function for btrfs_finish_ordered_io, this
1648  * just reads in some of the csum leaves to prime them into ram
1649  * before we start the transaction.  It limits the amount of btree
1650  * reads required while inside the transaction.
1651  */
1652 /* as ordered data IO finishes, this gets called so we can finish
1653  * an ordered extent if the range of bytes in the file it covers are
1654  * fully written.
1655  */
1656 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1657 {
1658         struct btrfs_root *root = BTRFS_I(inode)->root;
1659         struct btrfs_trans_handle *trans = NULL;
1660         struct btrfs_ordered_extent *ordered_extent = NULL;
1661         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1662         struct extent_state *cached_state = NULL;
1663         int compressed = 0;
1664         int ret;
1665
1666         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1667                                              end - start + 1);
1668         if (!ret)
1669                 return 0;
1670         BUG_ON(!ordered_extent);
1671
1672         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1673                 BUG_ON(!list_empty(&ordered_extent->list));
1674                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1675                 if (!ret) {
1676                         trans = btrfs_join_transaction(root, 1);
1677                         btrfs_set_trans_block_group(trans, inode);
1678                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1679                         ret = btrfs_update_inode(trans, root, inode);
1680                         BUG_ON(ret);
1681                 }
1682                 goto out;
1683         }
1684
1685         lock_extent_bits(io_tree, ordered_extent->file_offset,
1686                          ordered_extent->file_offset + ordered_extent->len - 1,
1687                          0, &cached_state, GFP_NOFS);
1688
1689         trans = btrfs_join_transaction(root, 1);
1690         btrfs_set_trans_block_group(trans, inode);
1691         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1692
1693         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1694                 compressed = 1;
1695         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1696                 BUG_ON(compressed);
1697                 ret = btrfs_mark_extent_written(trans, inode,
1698                                                 ordered_extent->file_offset,
1699                                                 ordered_extent->file_offset +
1700                                                 ordered_extent->len);
1701                 BUG_ON(ret);
1702         } else {
1703                 ret = insert_reserved_file_extent(trans, inode,
1704                                                 ordered_extent->file_offset,
1705                                                 ordered_extent->start,
1706                                                 ordered_extent->disk_len,
1707                                                 ordered_extent->len,
1708                                                 ordered_extent->len,
1709                                                 compressed, 0, 0,
1710                                                 BTRFS_FILE_EXTENT_REG);
1711                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1712                                    ordered_extent->file_offset,
1713                                    ordered_extent->len);
1714                 BUG_ON(ret);
1715         }
1716         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1717                              ordered_extent->file_offset +
1718                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1719
1720         add_pending_csums(trans, inode, ordered_extent->file_offset,
1721                           &ordered_extent->list);
1722
1723         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1724         ret = btrfs_update_inode(trans, root, inode);
1725         BUG_ON(ret);
1726 out:
1727         btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1728         if (trans)
1729                 btrfs_end_transaction(trans, root);
1730         /* once for us */
1731         btrfs_put_ordered_extent(ordered_extent);
1732         /* once for the tree */
1733         btrfs_put_ordered_extent(ordered_extent);
1734
1735         return 0;
1736 }
1737
1738 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1739                                 struct extent_state *state, int uptodate)
1740 {
1741         ClearPagePrivate2(page);
1742         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1743 }
1744
1745 /*
1746  * When IO fails, either with EIO or csum verification fails, we
1747  * try other mirrors that might have a good copy of the data.  This
1748  * io_failure_record is used to record state as we go through all the
1749  * mirrors.  If another mirror has good data, the page is set up to date
1750  * and things continue.  If a good mirror can't be found, the original
1751  * bio end_io callback is called to indicate things have failed.
1752  */
1753 struct io_failure_record {
1754         struct page *page;
1755         u64 start;
1756         u64 len;
1757         u64 logical;
1758         unsigned long bio_flags;
1759         int last_mirror;
1760 };
1761
1762 static int btrfs_io_failed_hook(struct bio *failed_bio,
1763                          struct page *page, u64 start, u64 end,
1764                          struct extent_state *state)
1765 {
1766         struct io_failure_record *failrec = NULL;
1767         u64 private;
1768         struct extent_map *em;
1769         struct inode *inode = page->mapping->host;
1770         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1771         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1772         struct bio *bio;
1773         int num_copies;
1774         int ret;
1775         int rw;
1776         u64 logical;
1777
1778         ret = get_state_private(failure_tree, start, &private);
1779         if (ret) {
1780                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1781                 if (!failrec)
1782                         return -ENOMEM;
1783                 failrec->start = start;
1784                 failrec->len = end - start + 1;
1785                 failrec->last_mirror = 0;
1786                 failrec->bio_flags = 0;
1787
1788                 read_lock(&em_tree->lock);
1789                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1790                 if (em->start > start || em->start + em->len < start) {
1791                         free_extent_map(em);
1792                         em = NULL;
1793                 }
1794                 read_unlock(&em_tree->lock);
1795
1796                 if (!em || IS_ERR(em)) {
1797                         kfree(failrec);
1798                         return -EIO;
1799                 }
1800                 logical = start - em->start;
1801                 logical = em->block_start + logical;
1802                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1803                         logical = em->block_start;
1804                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1805                 }
1806                 failrec->logical = logical;
1807                 free_extent_map(em);
1808                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1809                                 EXTENT_DIRTY, GFP_NOFS);
1810                 set_state_private(failure_tree, start,
1811                                  (u64)(unsigned long)failrec);
1812         } else {
1813                 failrec = (struct io_failure_record *)(unsigned long)private;
1814         }
1815         num_copies = btrfs_num_copies(
1816                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1817                               failrec->logical, failrec->len);
1818         failrec->last_mirror++;
1819         if (!state) {
1820                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1821                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1822                                                     failrec->start,
1823                                                     EXTENT_LOCKED);
1824                 if (state && state->start != failrec->start)
1825                         state = NULL;
1826                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1827         }
1828         if (!state || failrec->last_mirror > num_copies) {
1829                 set_state_private(failure_tree, failrec->start, 0);
1830                 clear_extent_bits(failure_tree, failrec->start,
1831                                   failrec->start + failrec->len - 1,
1832                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1833                 kfree(failrec);
1834                 return -EIO;
1835         }
1836         bio = bio_alloc(GFP_NOFS, 1);
1837         bio->bi_private = state;
1838         bio->bi_end_io = failed_bio->bi_end_io;
1839         bio->bi_sector = failrec->logical >> 9;
1840         bio->bi_bdev = failed_bio->bi_bdev;
1841         bio->bi_size = 0;
1842
1843         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1844         if (failed_bio->bi_rw & (1 << BIO_RW))
1845                 rw = WRITE;
1846         else
1847                 rw = READ;
1848
1849         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1850                                                       failrec->last_mirror,
1851                                                       failrec->bio_flags, 0);
1852         return 0;
1853 }
1854
1855 /*
1856  * each time an IO finishes, we do a fast check in the IO failure tree
1857  * to see if we need to process or clean up an io_failure_record
1858  */
1859 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1860 {
1861         u64 private;
1862         u64 private_failure;
1863         struct io_failure_record *failure;
1864         int ret;
1865
1866         private = 0;
1867         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1868                              (u64)-1, 1, EXTENT_DIRTY)) {
1869                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1870                                         start, &private_failure);
1871                 if (ret == 0) {
1872                         failure = (struct io_failure_record *)(unsigned long)
1873                                    private_failure;
1874                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1875                                           failure->start, 0);
1876                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1877                                           failure->start,
1878                                           failure->start + failure->len - 1,
1879                                           EXTENT_DIRTY | EXTENT_LOCKED,
1880                                           GFP_NOFS);
1881                         kfree(failure);
1882                 }
1883         }
1884         return 0;
1885 }
1886
1887 /*
1888  * when reads are done, we need to check csums to verify the data is correct
1889  * if there's a match, we allow the bio to finish.  If not, we go through
1890  * the io_failure_record routines to find good copies
1891  */
1892 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1893                                struct extent_state *state)
1894 {
1895         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1896         struct inode *inode = page->mapping->host;
1897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1898         char *kaddr;
1899         u64 private = ~(u32)0;
1900         int ret;
1901         struct btrfs_root *root = BTRFS_I(inode)->root;
1902         u32 csum = ~(u32)0;
1903
1904         if (PageChecked(page)) {
1905                 ClearPageChecked(page);
1906                 goto good;
1907         }
1908
1909         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1910                 return 0;
1911
1912         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1913             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1914                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1915                                   GFP_NOFS);
1916                 return 0;
1917         }
1918
1919         if (state && state->start == start) {
1920                 private = state->private;
1921                 ret = 0;
1922         } else {
1923                 ret = get_state_private(io_tree, start, &private);
1924         }
1925         kaddr = kmap_atomic(page, KM_USER0);
1926         if (ret)
1927                 goto zeroit;
1928
1929         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1930         btrfs_csum_final(csum, (char *)&csum);
1931         if (csum != private)
1932                 goto zeroit;
1933
1934         kunmap_atomic(kaddr, KM_USER0);
1935 good:
1936         /* if the io failure tree for this inode is non-empty,
1937          * check to see if we've recovered from a failed IO
1938          */
1939         btrfs_clean_io_failures(inode, start);
1940         return 0;
1941
1942 zeroit:
1943         if (printk_ratelimit()) {
1944                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1945                        "private %llu\n", page->mapping->host->i_ino,
1946                        (unsigned long long)start, csum,
1947                        (unsigned long long)private);
1948         }
1949         memset(kaddr + offset, 1, end - start + 1);
1950         flush_dcache_page(page);
1951         kunmap_atomic(kaddr, KM_USER0);
1952         if (private == 0)
1953                 return 0;
1954         return -EIO;
1955 }
1956
1957 struct delayed_iput {
1958         struct list_head list;
1959         struct inode *inode;
1960 };
1961
1962 void btrfs_add_delayed_iput(struct inode *inode)
1963 {
1964         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1965         struct delayed_iput *delayed;
1966
1967         if (atomic_add_unless(&inode->i_count, -1, 1))
1968                 return;
1969
1970         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1971         delayed->inode = inode;
1972
1973         spin_lock(&fs_info->delayed_iput_lock);
1974         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1975         spin_unlock(&fs_info->delayed_iput_lock);
1976 }
1977
1978 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1979 {
1980         LIST_HEAD(list);
1981         struct btrfs_fs_info *fs_info = root->fs_info;
1982         struct delayed_iput *delayed;
1983         int empty;
1984
1985         spin_lock(&fs_info->delayed_iput_lock);
1986         empty = list_empty(&fs_info->delayed_iputs);
1987         spin_unlock(&fs_info->delayed_iput_lock);
1988         if (empty)
1989                 return;
1990
1991         down_read(&root->fs_info->cleanup_work_sem);
1992         spin_lock(&fs_info->delayed_iput_lock);
1993         list_splice_init(&fs_info->delayed_iputs, &list);
1994         spin_unlock(&fs_info->delayed_iput_lock);
1995
1996         while (!list_empty(&list)) {
1997                 delayed = list_entry(list.next, struct delayed_iput, list);
1998                 list_del(&delayed->list);
1999                 iput(delayed->inode);
2000                 kfree(delayed);
2001         }
2002         up_read(&root->fs_info->cleanup_work_sem);
2003 }
2004
2005 /*
2006  * calculate extra metadata reservation when snapshotting a subvolume
2007  * contains orphan files.
2008  */
2009 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2010                                 struct btrfs_pending_snapshot *pending,
2011                                 u64 *bytes_to_reserve)
2012 {
2013         struct btrfs_root *root;
2014         struct btrfs_block_rsv *block_rsv;
2015         u64 num_bytes;
2016         int index;
2017
2018         root = pending->root;
2019         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2020                 return;
2021
2022         block_rsv = root->orphan_block_rsv;
2023
2024         /* orphan block reservation for the snapshot */
2025         num_bytes = block_rsv->size;
2026
2027         /*
2028          * after the snapshot is created, COWing tree blocks may use more
2029          * space than it frees. So we should make sure there is enough
2030          * reserved space.
2031          */
2032         index = trans->transid & 0x1;
2033         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2034                 num_bytes += block_rsv->size -
2035                              (block_rsv->reserved + block_rsv->freed[index]);
2036         }
2037
2038         *bytes_to_reserve += num_bytes;
2039 }
2040
2041 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2042                                 struct btrfs_pending_snapshot *pending)
2043 {
2044         struct btrfs_root *root = pending->root;
2045         struct btrfs_root *snap = pending->snap;
2046         struct btrfs_block_rsv *block_rsv;
2047         u64 num_bytes;
2048         int index;
2049         int ret;
2050
2051         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2052                 return;
2053
2054         /* refill source subvolume's orphan block reservation */
2055         block_rsv = root->orphan_block_rsv;
2056         index = trans->transid & 0x1;
2057         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2058                 num_bytes = block_rsv->size -
2059                             (block_rsv->reserved + block_rsv->freed[index]);
2060                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2061                                               root->orphan_block_rsv,
2062                                               num_bytes);
2063                 BUG_ON(ret);
2064         }
2065
2066         /* setup orphan block reservation for the snapshot */
2067         block_rsv = btrfs_alloc_block_rsv(snap);
2068         BUG_ON(!block_rsv);
2069
2070         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2071         snap->orphan_block_rsv = block_rsv;
2072
2073         num_bytes = root->orphan_block_rsv->size;
2074         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2075                                       block_rsv, num_bytes);
2076         BUG_ON(ret);
2077
2078 #if 0
2079         /* insert orphan item for the snapshot */
2080         WARN_ON(!root->orphan_item_inserted);
2081         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2082                                        snap->root_key.objectid);
2083         BUG_ON(ret);
2084         snap->orphan_item_inserted = 1;
2085 #endif
2086 }
2087
2088 enum btrfs_orphan_cleanup_state {
2089         ORPHAN_CLEANUP_STARTED  = 1,
2090         ORPHAN_CLEANUP_DONE     = 2,
2091 };
2092
2093 /*
2094  * This is called in transaction commmit time. If there are no orphan
2095  * files in the subvolume, it removes orphan item and frees block_rsv
2096  * structure.
2097  */
2098 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2099                               struct btrfs_root *root)
2100 {
2101         int ret;
2102
2103         if (!list_empty(&root->orphan_list) ||
2104             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2105                 return;
2106
2107         if (root->orphan_item_inserted &&
2108             btrfs_root_refs(&root->root_item) > 0) {
2109                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2110                                             root->root_key.objectid);
2111                 BUG_ON(ret);
2112                 root->orphan_item_inserted = 0;
2113         }
2114
2115         if (root->orphan_block_rsv) {
2116                 WARN_ON(root->orphan_block_rsv->size > 0);
2117                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2118                 root->orphan_block_rsv = NULL;
2119         }
2120 }
2121
2122 /*
2123  * This creates an orphan entry for the given inode in case something goes
2124  * wrong in the middle of an unlink/truncate.
2125  *
2126  * NOTE: caller of this function should reserve 5 units of metadata for
2127  *       this function.
2128  */
2129 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2130 {
2131         struct btrfs_root *root = BTRFS_I(inode)->root;
2132         struct btrfs_block_rsv *block_rsv = NULL;
2133         int reserve = 0;
2134         int insert = 0;
2135         int ret;
2136
2137         if (!root->orphan_block_rsv) {
2138                 block_rsv = btrfs_alloc_block_rsv(root);
2139                 BUG_ON(!block_rsv);
2140         }
2141
2142         spin_lock(&root->orphan_lock);
2143         if (!root->orphan_block_rsv) {
2144                 root->orphan_block_rsv = block_rsv;
2145         } else if (block_rsv) {
2146                 btrfs_free_block_rsv(root, block_rsv);
2147                 block_rsv = NULL;
2148         }
2149
2150         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2151                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2152 #if 0
2153                 /*
2154                  * For proper ENOSPC handling, we should do orphan
2155                  * cleanup when mounting. But this introduces backward
2156                  * compatibility issue.
2157                  */
2158                 if (!xchg(&root->orphan_item_inserted, 1))
2159                         insert = 2;
2160                 else
2161                         insert = 1;
2162 #endif
2163                 insert = 1;
2164         } else {
2165                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2166         }
2167
2168         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2169                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2170                 reserve = 1;
2171         }
2172         spin_unlock(&root->orphan_lock);
2173
2174         if (block_rsv)
2175                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2176
2177         /* grab metadata reservation from transaction handle */
2178         if (reserve) {
2179                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2180                 BUG_ON(ret);
2181         }
2182
2183         /* insert an orphan item to track this unlinked/truncated file */
2184         if (insert >= 1) {
2185                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2186                 BUG_ON(ret);
2187         }
2188
2189         /* insert an orphan item to track subvolume contains orphan files */
2190         if (insert >= 2) {
2191                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2192                                                root->root_key.objectid);
2193                 BUG_ON(ret);
2194         }
2195         return 0;
2196 }
2197
2198 /*
2199  * We have done the truncate/delete so we can go ahead and remove the orphan
2200  * item for this particular inode.
2201  */
2202 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2203 {
2204         struct btrfs_root *root = BTRFS_I(inode)->root;
2205         int delete_item = 0;
2206         int release_rsv = 0;
2207         int ret = 0;
2208
2209         spin_lock(&root->orphan_lock);
2210         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2211                 list_del_init(&BTRFS_I(inode)->i_orphan);
2212                 delete_item = 1;
2213         }
2214
2215         if (BTRFS_I(inode)->orphan_meta_reserved) {
2216                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2217                 release_rsv = 1;
2218         }
2219         spin_unlock(&root->orphan_lock);
2220
2221         if (trans && delete_item) {
2222                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2223                 BUG_ON(ret);
2224         }
2225
2226         if (release_rsv)
2227                 btrfs_orphan_release_metadata(inode);
2228
2229         return 0;
2230 }
2231
2232 /*
2233  * this cleans up any orphans that may be left on the list from the last use
2234  * of this root.
2235  */
2236 void btrfs_orphan_cleanup(struct btrfs_root *root)
2237 {
2238         struct btrfs_path *path;
2239         struct extent_buffer *leaf;
2240         struct btrfs_item *item;
2241         struct btrfs_key key, found_key;
2242         struct btrfs_trans_handle *trans;
2243         struct inode *inode;
2244         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2245
2246         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2247                 return;
2248
2249         path = btrfs_alloc_path();
2250         BUG_ON(!path);
2251         path->reada = -1;
2252
2253         key.objectid = BTRFS_ORPHAN_OBJECTID;
2254         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2255         key.offset = (u64)-1;
2256
2257         while (1) {
2258                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2259                 if (ret < 0) {
2260                         printk(KERN_ERR "Error searching slot for orphan: %d"
2261                                "\n", ret);
2262                         break;
2263                 }
2264
2265                 /*
2266                  * if ret == 0 means we found what we were searching for, which
2267                  * is weird, but possible, so only screw with path if we didnt
2268                  * find the key and see if we have stuff that matches
2269                  */
2270                 if (ret > 0) {
2271                         if (path->slots[0] == 0)
2272                                 break;
2273                         path->slots[0]--;
2274                 }
2275
2276                 /* pull out the item */
2277                 leaf = path->nodes[0];
2278                 item = btrfs_item_nr(leaf, path->slots[0]);
2279                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2280
2281                 /* make sure the item matches what we want */
2282                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2283                         break;
2284                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2285                         break;
2286
2287                 /* release the path since we're done with it */
2288                 btrfs_release_path(root, path);
2289
2290                 /*
2291                  * this is where we are basically btrfs_lookup, without the
2292                  * crossing root thing.  we store the inode number in the
2293                  * offset of the orphan item.
2294                  */
2295                 found_key.objectid = found_key.offset;
2296                 found_key.type = BTRFS_INODE_ITEM_KEY;
2297                 found_key.offset = 0;
2298                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2299                 BUG_ON(IS_ERR(inode));
2300
2301                 /*
2302                  * add this inode to the orphan list so btrfs_orphan_del does
2303                  * the proper thing when we hit it
2304                  */
2305                 spin_lock(&root->orphan_lock);
2306                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2307                 spin_unlock(&root->orphan_lock);
2308
2309                 /*
2310                  * if this is a bad inode, means we actually succeeded in
2311                  * removing the inode, but not the orphan record, which means
2312                  * we need to manually delete the orphan since iput will just
2313                  * do a destroy_inode
2314                  */
2315                 if (is_bad_inode(inode)) {
2316                         trans = btrfs_start_transaction(root, 0);
2317                         btrfs_orphan_del(trans, inode);
2318                         btrfs_end_transaction(trans, root);
2319                         iput(inode);
2320                         continue;
2321                 }
2322
2323                 /* if we have links, this was a truncate, lets do that */
2324                 if (inode->i_nlink) {
2325                         nr_truncate++;
2326                         btrfs_truncate(inode);
2327                 } else {
2328                         nr_unlink++;
2329                 }
2330
2331                 /* this will do delete_inode and everything for us */
2332                 iput(inode);
2333         }
2334         btrfs_free_path(path);
2335
2336         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2337
2338         if (root->orphan_block_rsv)
2339                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2340                                         (u64)-1);
2341
2342         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2343                 trans = btrfs_join_transaction(root, 1);
2344                 btrfs_end_transaction(trans, root);
2345         }
2346
2347         if (nr_unlink)
2348                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2349         if (nr_truncate)
2350                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2351 }
2352
2353 /*
2354  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2355  * don't find any xattrs, we know there can't be any acls.
2356  *
2357  * slot is the slot the inode is in, objectid is the objectid of the inode
2358  */
2359 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2360                                           int slot, u64 objectid)
2361 {
2362         u32 nritems = btrfs_header_nritems(leaf);
2363         struct btrfs_key found_key;
2364         int scanned = 0;
2365
2366         slot++;
2367         while (slot < nritems) {
2368                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2369
2370                 /* we found a different objectid, there must not be acls */
2371                 if (found_key.objectid != objectid)
2372                         return 0;
2373
2374                 /* we found an xattr, assume we've got an acl */
2375                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2376                         return 1;
2377
2378                 /*
2379                  * we found a key greater than an xattr key, there can't
2380                  * be any acls later on
2381                  */
2382                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2383                         return 0;
2384
2385                 slot++;
2386                 scanned++;
2387
2388                 /*
2389                  * it goes inode, inode backrefs, xattrs, extents,
2390                  * so if there are a ton of hard links to an inode there can
2391                  * be a lot of backrefs.  Don't waste time searching too hard,
2392                  * this is just an optimization
2393                  */
2394                 if (scanned >= 8)
2395                         break;
2396         }
2397         /* we hit the end of the leaf before we found an xattr or
2398          * something larger than an xattr.  We have to assume the inode
2399          * has acls
2400          */
2401         return 1;
2402 }
2403
2404 /*
2405  * read an inode from the btree into the in-memory inode
2406  */
2407 static void btrfs_read_locked_inode(struct inode *inode)
2408 {
2409         struct btrfs_path *path;
2410         struct extent_buffer *leaf;
2411         struct btrfs_inode_item *inode_item;
2412         struct btrfs_timespec *tspec;
2413         struct btrfs_root *root = BTRFS_I(inode)->root;
2414         struct btrfs_key location;
2415         int maybe_acls;
2416         u64 alloc_group_block;
2417         u32 rdev;
2418         int ret;
2419
2420         path = btrfs_alloc_path();
2421         BUG_ON(!path);
2422         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2423
2424         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2425         if (ret)
2426                 goto make_bad;
2427
2428         leaf = path->nodes[0];
2429         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2430                                     struct btrfs_inode_item);
2431
2432         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2433         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2434         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2435         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2436         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2437
2438         tspec = btrfs_inode_atime(inode_item);
2439         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2440         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2441
2442         tspec = btrfs_inode_mtime(inode_item);
2443         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2444         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2445
2446         tspec = btrfs_inode_ctime(inode_item);
2447         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2448         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2449
2450         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2451         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2452         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2453         inode->i_generation = BTRFS_I(inode)->generation;
2454         inode->i_rdev = 0;
2455         rdev = btrfs_inode_rdev(leaf, inode_item);
2456
2457         BTRFS_I(inode)->index_cnt = (u64)-1;
2458         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2459
2460         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2461
2462         /*
2463          * try to precache a NULL acl entry for files that don't have
2464          * any xattrs or acls
2465          */
2466         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2467         if (!maybe_acls)
2468                 cache_no_acl(inode);
2469
2470         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2471                                                 alloc_group_block, 0);
2472         btrfs_free_path(path);
2473         inode_item = NULL;
2474
2475         switch (inode->i_mode & S_IFMT) {
2476         case S_IFREG:
2477                 inode->i_mapping->a_ops = &btrfs_aops;
2478                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2479                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2480                 inode->i_fop = &btrfs_file_operations;
2481                 inode->i_op = &btrfs_file_inode_operations;
2482                 break;
2483         case S_IFDIR:
2484                 inode->i_fop = &btrfs_dir_file_operations;
2485                 if (root == root->fs_info->tree_root)
2486                         inode->i_op = &btrfs_dir_ro_inode_operations;
2487                 else
2488                         inode->i_op = &btrfs_dir_inode_operations;
2489                 break;
2490         case S_IFLNK:
2491                 inode->i_op = &btrfs_symlink_inode_operations;
2492                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2493                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2494                 break;
2495         default:
2496                 inode->i_op = &btrfs_special_inode_operations;
2497                 init_special_inode(inode, inode->i_mode, rdev);
2498                 break;
2499         }
2500
2501         btrfs_update_iflags(inode);
2502         return;
2503
2504 make_bad:
2505         btrfs_free_path(path);
2506         make_bad_inode(inode);
2507 }
2508
2509 /*
2510  * given a leaf and an inode, copy the inode fields into the leaf
2511  */
2512 static void fill_inode_item(struct btrfs_trans_handle *trans,
2513                             struct extent_buffer *leaf,
2514                             struct btrfs_inode_item *item,
2515                             struct inode *inode)
2516 {
2517         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2518         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2519         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2520         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2521         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2522
2523         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2524                                inode->i_atime.tv_sec);
2525         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2526                                 inode->i_atime.tv_nsec);
2527
2528         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2529                                inode->i_mtime.tv_sec);
2530         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2531                                 inode->i_mtime.tv_nsec);
2532
2533         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2534                                inode->i_ctime.tv_sec);
2535         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2536                                 inode->i_ctime.tv_nsec);
2537
2538         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2539         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2540         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2541         btrfs_set_inode_transid(leaf, item, trans->transid);
2542         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2543         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2544         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2545 }
2546
2547 /*
2548  * copy everything in the in-memory inode into the btree.
2549  */
2550 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2551                                 struct btrfs_root *root, struct inode *inode)
2552 {
2553         struct btrfs_inode_item *inode_item;
2554         struct btrfs_path *path;
2555         struct extent_buffer *leaf;
2556         int ret;
2557
2558         path = btrfs_alloc_path();
2559         BUG_ON(!path);
2560         path->leave_spinning = 1;
2561         ret = btrfs_lookup_inode(trans, root, path,
2562                                  &BTRFS_I(inode)->location, 1);
2563         if (ret) {
2564                 if (ret > 0)
2565                         ret = -ENOENT;
2566                 goto failed;
2567         }
2568
2569         btrfs_unlock_up_safe(path, 1);
2570         leaf = path->nodes[0];
2571         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2572                                   struct btrfs_inode_item);
2573
2574         fill_inode_item(trans, leaf, inode_item, inode);
2575         btrfs_mark_buffer_dirty(leaf);
2576         btrfs_set_inode_last_trans(trans, inode);
2577         ret = 0;
2578 failed:
2579         btrfs_free_path(path);
2580         return ret;
2581 }
2582
2583
2584 /*
2585  * unlink helper that gets used here in inode.c and in the tree logging
2586  * recovery code.  It remove a link in a directory with a given name, and
2587  * also drops the back refs in the inode to the directory
2588  */
2589 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2590                        struct btrfs_root *root,
2591                        struct inode *dir, struct inode *inode,
2592                        const char *name, int name_len)
2593 {
2594         struct btrfs_path *path;
2595         int ret = 0;
2596         struct extent_buffer *leaf;
2597         struct btrfs_dir_item *di;
2598         struct btrfs_key key;
2599         u64 index;
2600
2601         path = btrfs_alloc_path();
2602         if (!path) {
2603                 ret = -ENOMEM;
2604                 goto err;
2605         }
2606
2607         path->leave_spinning = 1;
2608         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2609                                     name, name_len, -1);
2610         if (IS_ERR(di)) {
2611                 ret = PTR_ERR(di);
2612                 goto err;
2613         }
2614         if (!di) {
2615                 ret = -ENOENT;
2616                 goto err;
2617         }
2618         leaf = path->nodes[0];
2619         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2620         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2621         if (ret)
2622                 goto err;
2623         btrfs_release_path(root, path);
2624
2625         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2626                                   inode->i_ino,
2627                                   dir->i_ino, &index);
2628         if (ret) {
2629                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2630                        "inode %lu parent %lu\n", name_len, name,
2631                        inode->i_ino, dir->i_ino);
2632                 goto err;
2633         }
2634
2635         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2636                                          index, name, name_len, -1);
2637         if (IS_ERR(di)) {
2638                 ret = PTR_ERR(di);
2639                 goto err;
2640         }
2641         if (!di) {
2642                 ret = -ENOENT;
2643                 goto err;
2644         }
2645         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2646         btrfs_release_path(root, path);
2647
2648         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2649                                          inode, dir->i_ino);
2650         BUG_ON(ret != 0 && ret != -ENOENT);
2651
2652         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2653                                            dir, index);
2654         BUG_ON(ret);
2655 err:
2656         btrfs_free_path(path);
2657         if (ret)
2658                 goto out;
2659
2660         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2661         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2662         btrfs_update_inode(trans, root, dir);
2663         btrfs_drop_nlink(inode);
2664         ret = btrfs_update_inode(trans, root, inode);
2665 out:
2666         return ret;
2667 }
2668
2669 /* helper to check if there is any shared block in the path */
2670 static int check_path_shared(struct btrfs_root *root,
2671                              struct btrfs_path *path)
2672 {
2673         struct extent_buffer *eb;
2674         int level;
2675         int ret;
2676         u64 refs;
2677
2678         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2679                 if (!path->nodes[level])
2680                         break;
2681                 eb = path->nodes[level];
2682                 if (!btrfs_block_can_be_shared(root, eb))
2683                         continue;
2684                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2685                                                &refs, NULL);
2686                 if (refs > 1)
2687                         return 1;
2688         }
2689         return 0;
2690 }
2691
2692 /*
2693  * helper to start transaction for unlink and rmdir.
2694  *
2695  * unlink and rmdir are special in btrfs, they do not always free space.
2696  * so in enospc case, we should make sure they will free space before
2697  * allowing them to use the global metadata reservation.
2698  */
2699 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2700                                                        struct dentry *dentry)
2701 {
2702         struct btrfs_trans_handle *trans;
2703         struct btrfs_root *root = BTRFS_I(dir)->root;
2704         struct btrfs_path *path;
2705         struct btrfs_inode_ref *ref;
2706         struct btrfs_dir_item *di;
2707         struct inode *inode = dentry->d_inode;
2708         u64 index;
2709         int check_link = 1;
2710         int err = -ENOSPC;
2711         int ret;
2712
2713         trans = btrfs_start_transaction(root, 10);
2714         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2715                 return trans;
2716
2717         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2718                 return ERR_PTR(-ENOSPC);
2719
2720         /* check if there is someone else holds reference */
2721         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2722                 return ERR_PTR(-ENOSPC);
2723
2724         if (atomic_read(&inode->i_count) > 2)
2725                 return ERR_PTR(-ENOSPC);
2726
2727         if (xchg(&root->fs_info->enospc_unlink, 1))
2728                 return ERR_PTR(-ENOSPC);
2729
2730         path = btrfs_alloc_path();
2731         if (!path) {
2732                 root->fs_info->enospc_unlink = 0;
2733                 return ERR_PTR(-ENOMEM);
2734         }
2735
2736         trans = btrfs_start_transaction(root, 0);
2737         if (IS_ERR(trans)) {
2738                 btrfs_free_path(path);
2739                 root->fs_info->enospc_unlink = 0;
2740                 return trans;
2741         }
2742
2743         path->skip_locking = 1;
2744         path->search_commit_root = 1;
2745
2746         ret = btrfs_lookup_inode(trans, root, path,
2747                                 &BTRFS_I(dir)->location, 0);
2748         if (ret < 0) {
2749                 err = ret;
2750                 goto out;
2751         }
2752         if (ret == 0) {
2753                 if (check_path_shared(root, path))
2754                         goto out;
2755         } else {
2756                 check_link = 0;
2757         }
2758         btrfs_release_path(root, path);
2759
2760         ret = btrfs_lookup_inode(trans, root, path,
2761                                 &BTRFS_I(inode)->location, 0);
2762         if (ret < 0) {
2763                 err = ret;
2764                 goto out;
2765         }
2766         if (ret == 0) {
2767                 if (check_path_shared(root, path))
2768                         goto out;
2769         } else {
2770                 check_link = 0;
2771         }
2772         btrfs_release_path(root, path);
2773
2774         if (ret == 0 && S_ISREG(inode->i_mode)) {
2775                 ret = btrfs_lookup_file_extent(trans, root, path,
2776                                                inode->i_ino, (u64)-1, 0);
2777                 if (ret < 0) {
2778                         err = ret;
2779                         goto out;
2780                 }
2781                 BUG_ON(ret == 0);
2782                 if (check_path_shared(root, path))
2783                         goto out;
2784                 btrfs_release_path(root, path);
2785         }
2786
2787         if (!check_link) {
2788                 err = 0;
2789                 goto out;
2790         }
2791
2792         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2793                                 dentry->d_name.name, dentry->d_name.len, 0);
2794         if (IS_ERR(di)) {
2795                 err = PTR_ERR(di);
2796                 goto out;
2797         }
2798         if (di) {
2799                 if (check_path_shared(root, path))
2800                         goto out;
2801         } else {
2802                 err = 0;
2803                 goto out;
2804         }
2805         btrfs_release_path(root, path);
2806
2807         ref = btrfs_lookup_inode_ref(trans, root, path,
2808                                 dentry->d_name.name, dentry->d_name.len,
2809                                 inode->i_ino, dir->i_ino, 0);
2810         if (IS_ERR(ref)) {
2811                 err = PTR_ERR(ref);
2812                 goto out;
2813         }
2814         BUG_ON(!ref);
2815         if (check_path_shared(root, path))
2816                 goto out;
2817         index = btrfs_inode_ref_index(path->nodes[0], ref);
2818         btrfs_release_path(root, path);
2819
2820         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2821                                 dentry->d_name.name, dentry->d_name.len, 0);
2822         if (IS_ERR(di)) {
2823                 err = PTR_ERR(di);
2824                 goto out;
2825         }
2826         BUG_ON(ret == -ENOENT);
2827         if (check_path_shared(root, path))
2828                 goto out;
2829
2830         err = 0;
2831 out:
2832         btrfs_free_path(path);
2833         if (err) {
2834                 btrfs_end_transaction(trans, root);
2835                 root->fs_info->enospc_unlink = 0;
2836                 return ERR_PTR(err);
2837         }
2838
2839         trans->block_rsv = &root->fs_info->global_block_rsv;
2840         return trans;
2841 }
2842
2843 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2844                                struct btrfs_root *root)
2845 {
2846         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2847                 BUG_ON(!root->fs_info->enospc_unlink);
2848                 root->fs_info->enospc_unlink = 0;
2849         }
2850         btrfs_end_transaction_throttle(trans, root);
2851 }
2852
2853 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2854 {
2855         struct btrfs_root *root = BTRFS_I(dir)->root;
2856         struct btrfs_trans_handle *trans;
2857         struct inode *inode = dentry->d_inode;
2858         int ret;
2859         unsigned long nr = 0;
2860
2861         trans = __unlink_start_trans(dir, dentry);
2862         if (IS_ERR(trans))
2863                 return PTR_ERR(trans);
2864
2865         btrfs_set_trans_block_group(trans, dir);
2866
2867         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2868
2869         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2870                                  dentry->d_name.name, dentry->d_name.len);
2871         BUG_ON(ret);
2872
2873         if (inode->i_nlink == 0) {
2874                 ret = btrfs_orphan_add(trans, inode);
2875                 BUG_ON(ret);
2876         }
2877
2878         nr = trans->blocks_used;
2879         __unlink_end_trans(trans, root);
2880         btrfs_btree_balance_dirty(root, nr);
2881         return ret;
2882 }
2883
2884 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2885                         struct btrfs_root *root,
2886                         struct inode *dir, u64 objectid,
2887                         const char *name, int name_len)
2888 {
2889         struct btrfs_path *path;
2890         struct extent_buffer *leaf;
2891         struct btrfs_dir_item *di;
2892         struct btrfs_key key;
2893         u64 index;
2894         int ret;
2895
2896         path = btrfs_alloc_path();
2897         if (!path)
2898                 return -ENOMEM;
2899
2900         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2901                                    name, name_len, -1);
2902         BUG_ON(!di || IS_ERR(di));
2903
2904         leaf = path->nodes[0];
2905         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2906         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2907         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2908         BUG_ON(ret);
2909         btrfs_release_path(root, path);
2910
2911         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2912                                  objectid, root->root_key.objectid,
2913                                  dir->i_ino, &index, name, name_len);
2914         if (ret < 0) {
2915                 BUG_ON(ret != -ENOENT);
2916                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2917                                                  name, name_len);
2918                 BUG_ON(!di || IS_ERR(di));
2919
2920                 leaf = path->nodes[0];
2921                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2922                 btrfs_release_path(root, path);
2923                 index = key.offset;
2924         }
2925
2926         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2927                                          index, name, name_len, -1);
2928         BUG_ON(!di || IS_ERR(di));
2929
2930         leaf = path->nodes[0];
2931         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2932         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2933         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2934         BUG_ON(ret);
2935         btrfs_release_path(root, path);
2936
2937         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2938         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2939         ret = btrfs_update_inode(trans, root, dir);
2940         BUG_ON(ret);
2941         dir->i_sb->s_dirt = 1;
2942
2943         btrfs_free_path(path);
2944         return 0;
2945 }
2946
2947 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2948 {
2949         struct inode *inode = dentry->d_inode;
2950         int err = 0;
2951         struct btrfs_root *root = BTRFS_I(dir)->root;
2952         struct btrfs_trans_handle *trans;
2953         unsigned long nr = 0;
2954
2955         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2956             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2957                 return -ENOTEMPTY;
2958
2959         trans = __unlink_start_trans(dir, dentry);
2960         if (IS_ERR(trans))
2961                 return PTR_ERR(trans);
2962
2963         btrfs_set_trans_block_group(trans, dir);
2964
2965         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2966                 err = btrfs_unlink_subvol(trans, root, dir,
2967                                           BTRFS_I(inode)->location.objectid,
2968                                           dentry->d_name.name,
2969                                           dentry->d_name.len);
2970                 goto out;
2971         }
2972
2973         err = btrfs_orphan_add(trans, inode);
2974         if (err)
2975                 goto out;
2976
2977         /* now the directory is empty */
2978         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2979                                  dentry->d_name.name, dentry->d_name.len);
2980         if (!err)
2981                 btrfs_i_size_write(inode, 0);
2982 out:
2983         nr = trans->blocks_used;
2984         __unlink_end_trans(trans, root);
2985         btrfs_btree_balance_dirty(root, nr);
2986
2987         return err;
2988 }
2989
2990 #if 0
2991 /*
2992  * when truncating bytes in a file, it is possible to avoid reading
2993  * the leaves that contain only checksum items.  This can be the
2994  * majority of the IO required to delete a large file, but it must
2995  * be done carefully.
2996  *
2997  * The keys in the level just above the leaves are checked to make sure
2998  * the lowest key in a given leaf is a csum key, and starts at an offset
2999  * after the new  size.
3000  *
3001  * Then the key for the next leaf is checked to make sure it also has
3002  * a checksum item for the same file.  If it does, we know our target leaf
3003  * contains only checksum items, and it can be safely freed without reading
3004  * it.
3005  *
3006  * This is just an optimization targeted at large files.  It may do
3007  * nothing.  It will return 0 unless things went badly.
3008  */
3009 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3010                                      struct btrfs_root *root,
3011                                      struct btrfs_path *path,
3012                                      struct inode *inode, u64 new_size)
3013 {
3014         struct btrfs_key key;
3015         int ret;
3016         int nritems;
3017         struct btrfs_key found_key;
3018         struct btrfs_key other_key;
3019         struct btrfs_leaf_ref *ref;
3020         u64 leaf_gen;
3021         u64 leaf_start;
3022
3023         path->lowest_level = 1;
3024         key.objectid = inode->i_ino;
3025         key.type = BTRFS_CSUM_ITEM_KEY;
3026         key.offset = new_size;
3027 again:
3028         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3029         if (ret < 0)
3030                 goto out;
3031
3032         if (path->nodes[1] == NULL) {
3033                 ret = 0;
3034                 goto out;
3035         }
3036         ret = 0;
3037         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3038         nritems = btrfs_header_nritems(path->nodes[1]);
3039
3040         if (!nritems)
3041                 goto out;
3042
3043         if (path->slots[1] >= nritems)
3044                 goto next_node;
3045
3046         /* did we find a key greater than anything we want to delete? */
3047         if (found_key.objectid > inode->i_ino ||
3048            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3049                 goto out;
3050
3051         /* we check the next key in the node to make sure the leave contains
3052          * only checksum items.  This comparison doesn't work if our
3053          * leaf is the last one in the node
3054          */
3055         if (path->slots[1] + 1 >= nritems) {
3056 next_node:
3057                 /* search forward from the last key in the node, this
3058                  * will bring us into the next node in the tree
3059                  */
3060                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3061
3062                 /* unlikely, but we inc below, so check to be safe */
3063                 if (found_key.offset == (u64)-1)
3064                         goto out;
3065
3066                 /* search_forward needs a path with locks held, do the
3067                  * search again for the original key.  It is possible
3068                  * this will race with a balance and return a path that
3069                  * we could modify, but this drop is just an optimization
3070                  * and is allowed to miss some leaves.
3071                  */
3072                 btrfs_release_path(root, path);
3073                 found_key.offset++;
3074
3075                 /* setup a max key for search_forward */
3076                 other_key.offset = (u64)-1;
3077                 other_key.type = key.type;
3078                 other_key.objectid = key.objectid;
3079
3080                 path->keep_locks = 1;
3081                 ret = btrfs_search_forward(root, &found_key, &other_key,
3082                                            path, 0, 0);
3083                 path->keep_locks = 0;
3084                 if (ret || found_key.objectid != key.objectid ||
3085                     found_key.type != key.type) {
3086                         ret = 0;
3087                         goto out;
3088                 }
3089
3090                 key.offset = found_key.offset;
3091                 btrfs_release_path(root, path);
3092                 cond_resched();
3093                 goto again;
3094         }
3095
3096         /* we know there's one more slot after us in the tree,
3097          * read that key so we can verify it is also a checksum item
3098          */
3099         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3100
3101         if (found_key.objectid < inode->i_ino)
3102                 goto next_key;
3103
3104         if (found_key.type != key.type || found_key.offset < new_size)
3105                 goto next_key;
3106
3107         /*
3108          * if the key for the next leaf isn't a csum key from this objectid,
3109          * we can't be sure there aren't good items inside this leaf.
3110          * Bail out
3111          */
3112         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3113                 goto out;
3114
3115         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3116         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3117         /*
3118          * it is safe to delete this leaf, it contains only
3119          * csum items from this inode at an offset >= new_size
3120          */
3121         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3122         BUG_ON(ret);
3123
3124         if (root->ref_cows && leaf_gen < trans->transid) {
3125                 ref = btrfs_alloc_leaf_ref(root, 0);
3126                 if (ref) {
3127                         ref->root_gen = root->root_key.offset;
3128                         ref->bytenr = leaf_start;
3129                         ref->owner = 0;
3130                         ref->generation = leaf_gen;
3131                         ref->nritems = 0;
3132
3133                         btrfs_sort_leaf_ref(ref);
3134
3135                         ret = btrfs_add_leaf_ref(root, ref, 0);
3136                         WARN_ON(ret);
3137                         btrfs_free_leaf_ref(root, ref);
3138                 } else {
3139                         WARN_ON(1);
3140                 }
3141         }
3142 next_key:
3143         btrfs_release_path(root, path);
3144
3145         if (other_key.objectid == inode->i_ino &&
3146             other_key.type == key.type && other_key.offset > key.offset) {
3147                 key.offset = other_key.offset;
3148                 cond_resched();
3149                 goto again;
3150         }
3151         ret = 0;
3152 out:
3153         /* fixup any changes we've made to the path */
3154         path->lowest_level = 0;
3155         path->keep_locks = 0;
3156         btrfs_release_path(root, path);
3157         return ret;
3158 }
3159
3160 #endif
3161
3162 /*
3163  * this can truncate away extent items, csum items and directory items.
3164  * It starts at a high offset and removes keys until it can't find
3165  * any higher than new_size
3166  *
3167  * csum items that cross the new i_size are truncated to the new size
3168  * as well.
3169  *
3170  * min_type is the minimum key type to truncate down to.  If set to 0, this
3171  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3172  */
3173 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3174                                struct btrfs_root *root,
3175                                struct inode *inode,
3176                                u64 new_size, u32 min_type)
3177 {
3178         struct btrfs_path *path;
3179         struct extent_buffer *leaf;
3180         struct btrfs_file_extent_item *fi;
3181         struct btrfs_key key;
3182         struct btrfs_key found_key;
3183         u64 extent_start = 0;
3184         u64 extent_num_bytes = 0;
3185         u64 extent_offset = 0;
3186         u64 item_end = 0;
3187         u64 mask = root->sectorsize - 1;
3188         u32 found_type = (u8)-1;
3189         int found_extent;
3190         int del_item;
3191         int pending_del_nr = 0;
3192         int pending_del_slot = 0;
3193         int extent_type = -1;
3194         int encoding;
3195         int ret;
3196         int err = 0;
3197
3198         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3199
3200         if (root->ref_cows)
3201                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3202
3203         path = btrfs_alloc_path();
3204         BUG_ON(!path);
3205         path->reada = -1;
3206
3207         key.objectid = inode->i_ino;
3208         key.offset = (u64)-1;
3209         key.type = (u8)-1;
3210
3211 search_again:
3212         path->leave_spinning = 1;
3213         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3214         if (ret < 0) {
3215                 err = ret;
3216                 goto out;
3217         }
3218
3219         if (ret > 0) {
3220                 /* there are no items in the tree for us to truncate, we're
3221                  * done
3222                  */
3223                 if (path->slots[0] == 0)
3224                         goto out;
3225                 path->slots[0]--;
3226         }
3227
3228         while (1) {
3229                 fi = NULL;
3230                 leaf = path->nodes[0];
3231                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3232                 found_type = btrfs_key_type(&found_key);
3233                 encoding = 0;
3234
3235                 if (found_key.objectid != inode->i_ino)
3236                         break;
3237
3238                 if (found_type < min_type)
3239                         break;
3240
3241                 item_end = found_key.offset;
3242                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3243                         fi = btrfs_item_ptr(leaf, path->slots[0],
3244                                             struct btrfs_file_extent_item);
3245                         extent_type = btrfs_file_extent_type(leaf, fi);
3246                         encoding = btrfs_file_extent_compression(leaf, fi);
3247                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3248                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3249
3250                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3251                                 item_end +=
3252                                     btrfs_file_extent_num_bytes(leaf, fi);
3253                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3254                                 item_end += btrfs_file_extent_inline_len(leaf,
3255                                                                          fi);
3256                         }
3257                         item_end--;
3258                 }
3259                 if (found_type > min_type) {
3260                         del_item = 1;
3261                 } else {
3262                         if (item_end < new_size)
3263                                 break;
3264                         if (found_key.offset >= new_size)
3265                                 del_item = 1;
3266                         else
3267                                 del_item = 0;
3268                 }
3269                 found_extent = 0;
3270                 /* FIXME, shrink the extent if the ref count is only 1 */
3271                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3272                         goto delete;
3273
3274                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3275                         u64 num_dec;
3276                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3277                         if (!del_item && !encoding) {
3278                                 u64 orig_num_bytes =
3279                                         btrfs_file_extent_num_bytes(leaf, fi);
3280                                 extent_num_bytes = new_size -
3281                                         found_key.offset + root->sectorsize - 1;
3282                                 extent_num_bytes = extent_num_bytes &
3283                                         ~((u64)root->sectorsize - 1);
3284                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3285                                                          extent_num_bytes);
3286                                 num_dec = (orig_num_bytes -
3287                                            extent_num_bytes);
3288                                 if (root->ref_cows && extent_start != 0)
3289                                         inode_sub_bytes(inode, num_dec);
3290                                 btrfs_mark_buffer_dirty(leaf);
3291                         } else {
3292                                 extent_num_bytes =
3293                                         btrfs_file_extent_disk_num_bytes(leaf,
3294                                                                          fi);
3295                                 extent_offset = found_key.offset -
3296                                         btrfs_file_extent_offset(leaf, fi);
3297
3298                                 /* FIXME blocksize != 4096 */
3299                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3300                                 if (extent_start != 0) {
3301                                         found_extent = 1;
3302                                         if (root->ref_cows)
3303                                                 inode_sub_bytes(inode, num_dec);
3304                                 }
3305                         }
3306                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3307                         /*
3308                          * we can't truncate inline items that have had
3309                          * special encodings
3310                          */
3311                         if (!del_item &&
3312                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3313                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3314                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3315                                 u32 size = new_size - found_key.offset;
3316
3317                                 if (root->ref_cows) {
3318                                         inode_sub_bytes(inode, item_end + 1 -
3319                                                         new_size);
3320                                 }
3321                                 size =
3322                                     btrfs_file_extent_calc_inline_size(size);
3323                                 ret = btrfs_truncate_item(trans, root, path,
3324                                                           size, 1);
3325                                 BUG_ON(ret);
3326                         } else if (root->ref_cows) {
3327                                 inode_sub_bytes(inode, item_end + 1 -
3328                                                 found_key.offset);
3329                         }
3330                 }
3331 delete:
3332                 if (del_item) {
3333                         if (!pending_del_nr) {
3334                                 /* no pending yet, add ourselves */
3335                                 pending_del_slot = path->slots[0];
3336                                 pending_del_nr = 1;
3337                         } else if (pending_del_nr &&
3338                                    path->slots[0] + 1 == pending_del_slot) {
3339                                 /* hop on the pending chunk */
3340                                 pending_del_nr++;
3341                                 pending_del_slot = path->slots[0];
3342                         } else {
3343                                 BUG();
3344                         }
3345                 } else {
3346                         break;
3347                 }
3348                 if (found_extent && root->ref_cows) {
3349                         btrfs_set_path_blocking(path);
3350                         ret = btrfs_free_extent(trans, root, extent_start,
3351                                                 extent_num_bytes, 0,
3352                                                 btrfs_header_owner(leaf),
3353                                                 inode->i_ino, extent_offset);
3354                         BUG_ON(ret);
3355                 }
3356
3357                 if (found_type == BTRFS_INODE_ITEM_KEY)
3358                         break;
3359
3360                 if (path->slots[0] == 0 ||
3361                     path->slots[0] != pending_del_slot) {
3362                         if (root->ref_cows) {
3363                                 err = -EAGAIN;
3364                                 goto out;
3365                         }
3366                         if (pending_del_nr) {
3367                                 ret = btrfs_del_items(trans, root, path,
3368                                                 pending_del_slot,
3369                                                 pending_del_nr);
3370                                 BUG_ON(ret);
3371                                 pending_del_nr = 0;
3372                         }
3373                         btrfs_release_path(root, path);
3374                         goto search_again;
3375                 } else {
3376                         path->slots[0]--;
3377                 }
3378         }
3379 out:
3380         if (pending_del_nr) {
3381                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3382                                       pending_del_nr);
3383                 BUG_ON(ret);
3384         }
3385         btrfs_free_path(path);
3386         return err;
3387 }
3388
3389 /*
3390  * taken from block_truncate_page, but does cow as it zeros out
3391  * any bytes left in the last page in the file.
3392  */
3393 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3394 {
3395         struct inode *inode = mapping->host;
3396         struct btrfs_root *root = BTRFS_I(inode)->root;
3397         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3398         struct btrfs_ordered_extent *ordered;
3399         struct extent_state *cached_state = NULL;
3400         char *kaddr;
3401         u32 blocksize = root->sectorsize;
3402         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3403         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3404         struct page *page;
3405         int ret = 0;
3406         u64 page_start;
3407         u64 page_end;
3408
3409         if ((offset & (blocksize - 1)) == 0)
3410                 goto out;
3411         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3412         if (ret)
3413                 goto out;
3414
3415         ret = -ENOMEM;
3416 again:
3417         page = grab_cache_page(mapping, index);
3418         if (!page) {
3419                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3420                 goto out;
3421         }
3422
3423         page_start = page_offset(page);
3424         page_end = page_start + PAGE_CACHE_SIZE - 1;
3425
3426         if (!PageUptodate(page)) {
3427                 ret = btrfs_readpage(NULL, page);
3428                 lock_page(page);
3429                 if (page->mapping != mapping) {
3430                         unlock_page(page);
3431                         page_cache_release(page);
3432                         goto again;
3433                 }
3434                 if (!PageUptodate(page)) {
3435                         ret = -EIO;
3436                         goto out_unlock;
3437                 }
3438         }
3439         wait_on_page_writeback(page);
3440
3441         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3442                          GFP_NOFS);
3443         set_page_extent_mapped(page);
3444
3445         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3446         if (ordered) {
3447                 unlock_extent_cached(io_tree, page_start, page_end,
3448                                      &cached_state, GFP_NOFS);
3449                 unlock_page(page);
3450                 page_cache_release(page);
3451                 btrfs_start_ordered_extent(inode, ordered, 1);
3452                 btrfs_put_ordered_extent(ordered);
3453                 goto again;
3454         }
3455
3456         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3457                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3458                           0, 0, &cached_state, GFP_NOFS);
3459
3460         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3461                                         &cached_state);
3462         if (ret) {
3463                 unlock_extent_cached(io_tree, page_start, page_end,
3464                                      &cached_state, GFP_NOFS);
3465                 goto out_unlock;
3466         }
3467
3468         ret = 0;
3469         if (offset != PAGE_CACHE_SIZE) {
3470                 kaddr = kmap(page);
3471                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3472                 flush_dcache_page(page);
3473                 kunmap(page);
3474         }
3475         ClearPageChecked(page);
3476         set_page_dirty(page);
3477         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3478                              GFP_NOFS);
3479
3480 out_unlock:
3481         if (ret)
3482                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3483         unlock_page(page);
3484         page_cache_release(page);
3485 out:
3486         return ret;
3487 }
3488
3489 int btrfs_cont_expand(struct inode *inode, loff_t size)
3490 {
3491         struct btrfs_trans_handle *trans;
3492         struct btrfs_root *root = BTRFS_I(inode)->root;
3493         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494         struct extent_map *em = NULL;
3495         struct extent_state *cached_state = NULL;
3496         u64 mask = root->sectorsize - 1;
3497         u64 hole_start = (inode->i_size + mask) & ~mask;
3498         u64 block_end = (size + mask) & ~mask;
3499         u64 last_byte;
3500         u64 cur_offset;
3501         u64 hole_size;
3502         int err = 0;
3503
3504         if (size <= hole_start)
3505                 return 0;
3506
3507         while (1) {
3508                 struct btrfs_ordered_extent *ordered;
3509                 btrfs_wait_ordered_range(inode, hole_start,
3510                                          block_end - hole_start);
3511                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3512                                  &cached_state, GFP_NOFS);
3513                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3514                 if (!ordered)
3515                         break;
3516                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3517                                      &cached_state, GFP_NOFS);
3518                 btrfs_put_ordered_extent(ordered);
3519         }
3520
3521         cur_offset = hole_start;
3522         while (1) {
3523                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3524                                 block_end - cur_offset, 0);
3525                 BUG_ON(IS_ERR(em) || !em);
3526                 last_byte = min(extent_map_end(em), block_end);
3527                 last_byte = (last_byte + mask) & ~mask;
3528                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3529                         u64 hint_byte = 0;
3530                         hole_size = last_byte - cur_offset;
3531
3532                         trans = btrfs_start_transaction(root, 2);
3533                         if (IS_ERR(trans)) {
3534                                 err = PTR_ERR(trans);
3535                                 break;
3536                         }
3537                         btrfs_set_trans_block_group(trans, inode);
3538
3539                         err = btrfs_drop_extents(trans, inode, cur_offset,
3540                                                  cur_offset + hole_size,
3541                                                  &hint_byte, 1);
3542                         BUG_ON(err);
3543
3544                         err = btrfs_insert_file_extent(trans, root,
3545                                         inode->i_ino, cur_offset, 0,
3546                                         0, hole_size, 0, hole_size,
3547                                         0, 0, 0);
3548                         BUG_ON(err);
3549
3550                         btrfs_drop_extent_cache(inode, hole_start,
3551                                         last_byte - 1, 0);
3552
3553                         btrfs_end_transaction(trans, root);
3554                 }
3555                 free_extent_map(em);
3556                 em = NULL;
3557                 cur_offset = last_byte;
3558                 if (cur_offset >= block_end)
3559                         break;
3560         }
3561
3562         free_extent_map(em);
3563         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3564                              GFP_NOFS);
3565         return err;
3566 }
3567
3568 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3569 {
3570         struct btrfs_root *root = BTRFS_I(inode)->root;
3571         struct btrfs_trans_handle *trans;
3572         unsigned long nr;
3573         int ret;
3574
3575         if (attr->ia_size == inode->i_size)
3576                 return 0;
3577
3578         if (attr->ia_size > inode->i_size) {
3579                 unsigned long limit;
3580                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3581                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3582                         return -EFBIG;
3583                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3584                         send_sig(SIGXFSZ, current, 0);
3585                         return -EFBIG;
3586                 }
3587         }
3588
3589         trans = btrfs_start_transaction(root, 5);
3590         if (IS_ERR(trans))
3591                 return PTR_ERR(trans);
3592
3593         btrfs_set_trans_block_group(trans, inode);
3594
3595         ret = btrfs_orphan_add(trans, inode);
3596         BUG_ON(ret);
3597
3598         nr = trans->blocks_used;
3599         btrfs_end_transaction(trans, root);
3600         btrfs_btree_balance_dirty(root, nr);
3601
3602         if (attr->ia_size > inode->i_size) {
3603                 ret = btrfs_cont_expand(inode, attr->ia_size);
3604                 if (ret) {
3605                         btrfs_truncate(inode);
3606                         return ret;
3607                 }
3608
3609                 i_size_write(inode, attr->ia_size);
3610                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3611
3612                 trans = btrfs_start_transaction(root, 0);
3613                 BUG_ON(IS_ERR(trans));
3614                 btrfs_set_trans_block_group(trans, inode);
3615                 trans->block_rsv = root->orphan_block_rsv;
3616                 BUG_ON(!trans->block_rsv);
3617
3618                 ret = btrfs_update_inode(trans, root, inode);
3619                 BUG_ON(ret);
3620                 if (inode->i_nlink > 0) {
3621                         ret = btrfs_orphan_del(trans, inode);
3622                         BUG_ON(ret);
3623                 }
3624                 nr = trans->blocks_used;
3625                 btrfs_end_transaction(trans, root);
3626                 btrfs_btree_balance_dirty(root, nr);
3627                 return 0;
3628         }
3629
3630         /*
3631          * We're truncating a file that used to have good data down to
3632          * zero. Make sure it gets into the ordered flush list so that
3633          * any new writes get down to disk quickly.
3634          */
3635         if (attr->ia_size == 0)
3636                 BTRFS_I(inode)->ordered_data_close = 1;
3637
3638         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3639         ret = vmtruncate(inode, attr->ia_size);
3640         BUG_ON(ret);
3641
3642         return 0;
3643 }
3644
3645 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3646 {
3647         struct inode *inode = dentry->d_inode;
3648         int err;
3649
3650         err = inode_change_ok(inode, attr);
3651         if (err)
3652                 return err;
3653
3654         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3655                 err = btrfs_setattr_size(inode, attr);
3656                 if (err)
3657                         return err;
3658         }
3659         attr->ia_valid &= ~ATTR_SIZE;
3660
3661         if (attr->ia_valid)
3662                 err = inode_setattr(inode, attr);
3663
3664         if (!err && ((attr->ia_valid & ATTR_MODE)))
3665                 err = btrfs_acl_chmod(inode);
3666         return err;
3667 }
3668
3669 void btrfs_delete_inode(struct inode *inode)
3670 {
3671         struct btrfs_trans_handle *trans;
3672         struct btrfs_root *root = BTRFS_I(inode)->root;
3673         unsigned long nr;
3674         int ret;
3675
3676         truncate_inode_pages(&inode->i_data, 0);
3677         if (is_bad_inode(inode)) {
3678                 btrfs_orphan_del(NULL, inode);
3679                 goto no_delete;
3680         }
3681         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3682
3683         if (root->fs_info->log_root_recovering) {
3684                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3685                 goto no_delete;
3686         }
3687
3688         if (inode->i_nlink > 0) {
3689                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3690                 goto no_delete;
3691         }
3692
3693         btrfs_i_size_write(inode, 0);
3694
3695         while (1) {
3696                 trans = btrfs_start_transaction(root, 0);
3697                 BUG_ON(IS_ERR(trans));
3698                 btrfs_set_trans_block_group(trans, inode);
3699                 trans->block_rsv = root->orphan_block_rsv;
3700
3701                 ret = btrfs_block_rsv_check(trans, root,
3702                                             root->orphan_block_rsv, 0, 5);
3703                 if (ret) {
3704                         BUG_ON(ret != -EAGAIN);
3705                         ret = btrfs_commit_transaction(trans, root);
3706                         BUG_ON(ret);
3707                         continue;
3708                 }
3709
3710                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3711                 if (ret != -EAGAIN)
3712                         break;
3713
3714                 nr = trans->blocks_used;
3715                 btrfs_end_transaction(trans, root);
3716                 trans = NULL;
3717                 btrfs_btree_balance_dirty(root, nr);
3718
3719         }
3720
3721         if (ret == 0) {
3722                 ret = btrfs_orphan_del(trans, inode);
3723                 BUG_ON(ret);
3724         }
3725
3726         nr = trans->blocks_used;
3727         btrfs_end_transaction(trans, root);
3728         btrfs_btree_balance_dirty(root, nr);
3729 no_delete:
3730         clear_inode(inode);
3731         return;
3732 }
3733
3734 /*
3735  * this returns the key found in the dir entry in the location pointer.
3736  * If no dir entries were found, location->objectid is 0.
3737  */
3738 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3739                                struct btrfs_key *location)
3740 {
3741         const char *name = dentry->d_name.name;
3742         int namelen = dentry->d_name.len;
3743         struct btrfs_dir_item *di;
3744         struct btrfs_path *path;
3745         struct btrfs_root *root = BTRFS_I(dir)->root;
3746         int ret = 0;
3747
3748         path = btrfs_alloc_path();
3749         BUG_ON(!path);
3750
3751         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3752                                     namelen, 0);
3753         if (IS_ERR(di))
3754                 ret = PTR_ERR(di);
3755
3756         if (!di || IS_ERR(di))
3757                 goto out_err;
3758
3759         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3760 out:
3761         btrfs_free_path(path);
3762         return ret;
3763 out_err:
3764         location->objectid = 0;
3765         goto out;
3766 }
3767
3768 /*
3769  * when we hit a tree root in a directory, the btrfs part of the inode
3770  * needs to be changed to reflect the root directory of the tree root.  This
3771  * is kind of like crossing a mount point.
3772  */
3773 static int fixup_tree_root_location(struct btrfs_root *root,
3774                                     struct inode *dir,
3775                                     struct dentry *dentry,
3776                                     struct btrfs_key *location,
3777                                     struct btrfs_root **sub_root)
3778 {
3779         struct btrfs_path *path;
3780         struct btrfs_root *new_root;
3781         struct btrfs_root_ref *ref;
3782         struct extent_buffer *leaf;
3783         int ret;
3784         int err = 0;
3785
3786         path = btrfs_alloc_path();
3787         if (!path) {
3788                 err = -ENOMEM;
3789                 goto out;
3790         }
3791
3792         err = -ENOENT;
3793         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3794                                   BTRFS_I(dir)->root->root_key.objectid,
3795                                   location->objectid);
3796         if (ret) {
3797                 if (ret < 0)
3798                         err = ret;
3799                 goto out;
3800         }
3801
3802         leaf = path->nodes[0];
3803         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3804         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3805             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3806                 goto out;
3807
3808         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3809                                    (unsigned long)(ref + 1),
3810                                    dentry->d_name.len);
3811         if (ret)
3812                 goto out;
3813
3814         btrfs_release_path(root->fs_info->tree_root, path);
3815
3816         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3817         if (IS_ERR(new_root)) {
3818                 err = PTR_ERR(new_root);
3819                 goto out;
3820         }
3821
3822         if (btrfs_root_refs(&new_root->root_item) == 0) {
3823                 err = -ENOENT;
3824                 goto out;
3825         }
3826
3827         *sub_root = new_root;
3828         location->objectid = btrfs_root_dirid(&new_root->root_item);
3829         location->type = BTRFS_INODE_ITEM_KEY;
3830         location->offset = 0;
3831         err = 0;
3832 out:
3833         btrfs_free_path(path);
3834         return err;
3835 }
3836
3837 static void inode_tree_add(struct inode *inode)
3838 {
3839         struct btrfs_root *root = BTRFS_I(inode)->root;
3840         struct btrfs_inode *entry;
3841         struct rb_node **p;
3842         struct rb_node *parent;
3843 again:
3844         p = &root->inode_tree.rb_node;
3845         parent = NULL;
3846
3847         if (hlist_unhashed(&inode->i_hash))
3848                 return;
3849
3850         spin_lock(&root->inode_lock);
3851         while (*p) {
3852                 parent = *p;
3853                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3854
3855                 if (inode->i_ino < entry->vfs_inode.i_ino)
3856                         p = &parent->rb_left;
3857                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3858                         p = &parent->rb_right;
3859                 else {
3860                         WARN_ON(!(entry->vfs_inode.i_state &
3861                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3862                         rb_erase(parent, &root->inode_tree);
3863                         RB_CLEAR_NODE(parent);
3864                         spin_unlock(&root->inode_lock);
3865                         goto again;
3866                 }
3867         }
3868         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3869         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3870         spin_unlock(&root->inode_lock);
3871 }
3872
3873 static void inode_tree_del(struct inode *inode)
3874 {
3875         struct btrfs_root *root = BTRFS_I(inode)->root;
3876         int empty = 0;
3877
3878         spin_lock(&root->inode_lock);
3879         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3880                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3881                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3882                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3883         }
3884         spin_unlock(&root->inode_lock);
3885
3886         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3887                 synchronize_srcu(&root->fs_info->subvol_srcu);
3888                 spin_lock(&root->inode_lock);
3889                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3890                 spin_unlock(&root->inode_lock);
3891                 if (empty)
3892                         btrfs_add_dead_root(root);
3893         }
3894 }
3895
3896 int btrfs_invalidate_inodes(struct btrfs_root *root)
3897 {
3898         struct rb_node *node;
3899         struct rb_node *prev;
3900         struct btrfs_inode *entry;
3901         struct inode *inode;
3902         u64 objectid = 0;
3903
3904         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3905
3906         spin_lock(&root->inode_lock);
3907 again:
3908         node = root->inode_tree.rb_node;
3909         prev = NULL;
3910         while (node) {
3911                 prev = node;
3912                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3913
3914                 if (objectid < entry->vfs_inode.i_ino)
3915                         node = node->rb_left;
3916                 else if (objectid > entry->vfs_inode.i_ino)
3917                         node = node->rb_right;
3918                 else
3919                         break;
3920         }
3921         if (!node) {
3922                 while (prev) {
3923                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3924                         if (objectid <= entry->vfs_inode.i_ino) {
3925                                 node = prev;
3926                                 break;
3927                         }
3928                         prev = rb_next(prev);
3929                 }
3930         }
3931         while (node) {
3932                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3933                 objectid = entry->vfs_inode.i_ino + 1;
3934                 inode = igrab(&entry->vfs_inode);
3935                 if (inode) {
3936                         spin_unlock(&root->inode_lock);
3937                         if (atomic_read(&inode->i_count) > 1)
3938                                 d_prune_aliases(inode);
3939                         /*
3940                          * btrfs_drop_inode will remove it from
3941                          * the inode cache when its usage count
3942                          * hits zero.
3943                          */
3944                         iput(inode);
3945                         cond_resched();
3946                         spin_lock(&root->inode_lock);
3947                         goto again;
3948                 }
3949
3950                 if (cond_resched_lock(&root->inode_lock))
3951                         goto again;
3952
3953                 node = rb_next(node);
3954         }
3955         spin_unlock(&root->inode_lock);
3956         return 0;
3957 }
3958
3959 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3960 {
3961         struct btrfs_iget_args *args = p;
3962         inode->i_ino = args->ino;
3963         BTRFS_I(inode)->root = args->root;
3964         btrfs_set_inode_space_info(args->root, inode);
3965         return 0;
3966 }
3967
3968 static int btrfs_find_actor(struct inode *inode, void *opaque)
3969 {
3970         struct btrfs_iget_args *args = opaque;
3971         return args->ino == inode->i_ino &&
3972                 args->root == BTRFS_I(inode)->root;
3973 }
3974
3975 static struct inode *btrfs_iget_locked(struct super_block *s,
3976                                        u64 objectid,
3977                                        struct btrfs_root *root)
3978 {
3979         struct inode *inode;
3980         struct btrfs_iget_args args;
3981         args.ino = objectid;
3982         args.root = root;
3983
3984         inode = iget5_locked(s, objectid, btrfs_find_actor,
3985                              btrfs_init_locked_inode,
3986                              (void *)&args);
3987         return inode;
3988 }
3989
3990 /* Get an inode object given its location and corresponding root.
3991  * Returns in *is_new if the inode was read from disk
3992  */
3993 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3994                          struct btrfs_root *root, int *new)
3995 {
3996         struct inode *inode;
3997
3998         inode = btrfs_iget_locked(s, location->objectid, root);
3999         if (!inode)
4000                 return ERR_PTR(-ENOMEM);
4001
4002         if (inode->i_state & I_NEW) {
4003                 BTRFS_I(inode)->root = root;
4004                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4005                 btrfs_read_locked_inode(inode);
4006
4007                 inode_tree_add(inode);
4008                 unlock_new_inode(inode);
4009                 if (new)
4010                         *new = 1;
4011         }
4012
4013         return inode;
4014 }
4015
4016 static struct inode *new_simple_dir(struct super_block *s,
4017                                     struct btrfs_key *key,
4018                                     struct btrfs_root *root)
4019 {
4020         struct inode *inode = new_inode(s);
4021
4022         if (!inode)
4023                 return ERR_PTR(-ENOMEM);
4024
4025         BTRFS_I(inode)->root = root;
4026         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4027         BTRFS_I(inode)->dummy_inode = 1;
4028
4029         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4030         inode->i_op = &simple_dir_inode_operations;
4031         inode->i_fop = &simple_dir_operations;
4032         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4033         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4034
4035         return inode;
4036 }
4037
4038 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4039 {
4040         struct inode *inode;
4041         struct btrfs_root *root = BTRFS_I(dir)->root;
4042         struct btrfs_root *sub_root = root;
4043         struct btrfs_key location;
4044         int index;
4045         int ret;
4046
4047         dentry->d_op = &btrfs_dentry_operations;
4048
4049         if (dentry->d_name.len > BTRFS_NAME_LEN)
4050                 return ERR_PTR(-ENAMETOOLONG);
4051
4052         ret = btrfs_inode_by_name(dir, dentry, &location);
4053
4054         if (ret < 0)
4055                 return ERR_PTR(ret);
4056
4057         if (location.objectid == 0)
4058                 return NULL;
4059
4060         if (location.type == BTRFS_INODE_ITEM_KEY) {
4061                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4062                 return inode;
4063         }
4064
4065         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4066
4067         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4068         ret = fixup_tree_root_location(root, dir, dentry,
4069                                        &location, &sub_root);
4070         if (ret < 0) {
4071                 if (ret != -ENOENT)
4072                         inode = ERR_PTR(ret);
4073                 else
4074                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4075         } else {
4076                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4077         }
4078         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4079
4080         if (root != sub_root) {
4081                 down_read(&root->fs_info->cleanup_work_sem);
4082                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4083                         btrfs_orphan_cleanup(sub_root);
4084                 up_read(&root->fs_info->cleanup_work_sem);
4085         }
4086
4087         return inode;
4088 }
4089
4090 static int btrfs_dentry_delete(struct dentry *dentry)
4091 {
4092         struct btrfs_root *root;
4093
4094         if (!dentry->d_inode && !IS_ROOT(dentry))
4095                 dentry = dentry->d_parent;
4096
4097         if (dentry->d_inode) {
4098                 root = BTRFS_I(dentry->d_inode)->root;
4099                 if (btrfs_root_refs(&root->root_item) == 0)
4100                         return 1;
4101         }
4102         return 0;
4103 }
4104
4105 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4106                                    struct nameidata *nd)
4107 {
4108         struct inode *inode;
4109
4110         inode = btrfs_lookup_dentry(dir, dentry);
4111         if (IS_ERR(inode))
4112                 return ERR_CAST(inode);
4113
4114         return d_splice_alias(inode, dentry);
4115 }
4116
4117 static unsigned char btrfs_filetype_table[] = {
4118         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4119 };
4120
4121 static int btrfs_real_readdir(struct file *filp, void *dirent,
4122                               filldir_t filldir)
4123 {
4124         struct inode *inode = filp->f_dentry->d_inode;
4125         struct btrfs_root *root = BTRFS_I(inode)->root;
4126         struct btrfs_item *item;
4127         struct btrfs_dir_item *di;
4128         struct btrfs_key key;
4129         struct btrfs_key found_key;
4130         struct btrfs_path *path;
4131         int ret;
4132         u32 nritems;
4133         struct extent_buffer *leaf;
4134         int slot;
4135         int advance;
4136         unsigned char d_type;
4137         int over = 0;
4138         u32 di_cur;
4139         u32 di_total;
4140         u32 di_len;
4141         int key_type = BTRFS_DIR_INDEX_KEY;
4142         char tmp_name[32];
4143         char *name_ptr;
4144         int name_len;
4145
4146         /* FIXME, use a real flag for deciding about the key type */
4147         if (root->fs_info->tree_root == root)
4148                 key_type = BTRFS_DIR_ITEM_KEY;
4149
4150         /* special case for "." */
4151         if (filp->f_pos == 0) {
4152                 over = filldir(dirent, ".", 1,
4153                                1, inode->i_ino,
4154                                DT_DIR);
4155                 if (over)
4156                         return 0;
4157                 filp->f_pos = 1;
4158         }
4159         /* special case for .., just use the back ref */
4160         if (filp->f_pos == 1) {
4161                 u64 pino = parent_ino(filp->f_path.dentry);
4162                 over = filldir(dirent, "..", 2,
4163                                2, pino, DT_DIR);
4164                 if (over)
4165                         return 0;
4166                 filp->f_pos = 2;
4167         }
4168         path = btrfs_alloc_path();
4169         path->reada = 2;
4170
4171         btrfs_set_key_type(&key, key_type);
4172         key.offset = filp->f_pos;
4173         key.objectid = inode->i_ino;
4174
4175         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4176         if (ret < 0)
4177                 goto err;
4178         advance = 0;
4179
4180         while (1) {
4181                 leaf = path->nodes[0];
4182                 nritems = btrfs_header_nritems(leaf);
4183                 slot = path->slots[0];
4184                 if (advance || slot >= nritems) {
4185                         if (slot >= nritems - 1) {
4186                                 ret = btrfs_next_leaf(root, path);
4187                                 if (ret)
4188                                         break;
4189                                 leaf = path->nodes[0];
4190                                 nritems = btrfs_header_nritems(leaf);
4191                                 slot = path->slots[0];
4192                         } else {
4193                                 slot++;
4194                                 path->slots[0]++;
4195                         }
4196                 }
4197
4198                 advance = 1;
4199                 item = btrfs_item_nr(leaf, slot);
4200                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4201
4202                 if (found_key.objectid != key.objectid)
4203                         break;
4204                 if (btrfs_key_type(&found_key) != key_type)
4205                         break;
4206                 if (found_key.offset < filp->f_pos)
4207                         continue;
4208
4209                 filp->f_pos = found_key.offset;
4210
4211                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4212                 di_cur = 0;
4213                 di_total = btrfs_item_size(leaf, item);
4214
4215                 while (di_cur < di_total) {
4216                         struct btrfs_key location;
4217
4218                         name_len = btrfs_dir_name_len(leaf, di);
4219                         if (name_len <= sizeof(tmp_name)) {
4220                                 name_ptr = tmp_name;
4221                         } else {
4222                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4223                                 if (!name_ptr) {
4224                                         ret = -ENOMEM;
4225                                         goto err;
4226                                 }
4227                         }
4228                         read_extent_buffer(leaf, name_ptr,
4229                                            (unsigned long)(di + 1), name_len);
4230
4231                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4232                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4233
4234                         /* is this a reference to our own snapshot? If so
4235                          * skip it
4236                          */
4237                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4238                             location.objectid == root->root_key.objectid) {
4239                                 over = 0;
4240                                 goto skip;
4241                         }
4242                         over = filldir(dirent, name_ptr, name_len,
4243                                        found_key.offset, location.objectid,
4244                                        d_type);
4245
4246 skip:
4247                         if (name_ptr != tmp_name)
4248                                 kfree(name_ptr);
4249
4250                         if (over)
4251                                 goto nopos;
4252                         di_len = btrfs_dir_name_len(leaf, di) +
4253                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4254                         di_cur += di_len;
4255                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4256                 }
4257         }
4258
4259         /* Reached end of directory/root. Bump pos past the last item. */
4260         if (key_type == BTRFS_DIR_INDEX_KEY)
4261                 /*
4262                  * 32-bit glibc will use getdents64, but then strtol -
4263                  * so the last number we can serve is this.
4264                  */
4265                 filp->f_pos = 0x7fffffff;
4266         else
4267                 filp->f_pos++;
4268 nopos:
4269         ret = 0;
4270 err:
4271         btrfs_free_path(path);
4272         return ret;
4273 }
4274
4275 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4276 {
4277         struct btrfs_root *root = BTRFS_I(inode)->root;
4278         struct btrfs_trans_handle *trans;
4279         int ret = 0;
4280
4281         if (BTRFS_I(inode)->dummy_inode)
4282                 return 0;
4283
4284         if (wbc->sync_mode == WB_SYNC_ALL) {
4285                 trans = btrfs_join_transaction(root, 1);
4286                 btrfs_set_trans_block_group(trans, inode);
4287                 ret = btrfs_commit_transaction(trans, root);
4288         }
4289         return ret;
4290 }
4291
4292 /*
4293  * This is somewhat expensive, updating the tree every time the
4294  * inode changes.  But, it is most likely to find the inode in cache.
4295  * FIXME, needs more benchmarking...there are no reasons other than performance
4296  * to keep or drop this code.
4297  */
4298 void btrfs_dirty_inode(struct inode *inode)
4299 {
4300         struct btrfs_root *root = BTRFS_I(inode)->root;
4301         struct btrfs_trans_handle *trans;
4302         int ret;
4303
4304         if (BTRFS_I(inode)->dummy_inode)
4305                 return;
4306
4307         trans = btrfs_join_transaction(root, 1);
4308         btrfs_set_trans_block_group(trans, inode);
4309
4310         ret = btrfs_update_inode(trans, root, inode);
4311         if (ret && ret == -ENOSPC) {
4312                 /* whoops, lets try again with the full transaction */
4313                 btrfs_end_transaction(trans, root);
4314                 trans = btrfs_start_transaction(root, 1);
4315                 btrfs_set_trans_block_group(trans, inode);
4316
4317                 ret = btrfs_update_inode(trans, root, inode);
4318                 if (ret) {
4319                         printk(KERN_ERR"btrfs: fail to dirty inode %lu error %d\n",
4320                                 inode->i_ino, ret);
4321                 }
4322         }
4323         btrfs_end_transaction(trans, root);
4324 }
4325
4326 /*
4327  * find the highest existing sequence number in a directory
4328  * and then set the in-memory index_cnt variable to reflect
4329  * free sequence numbers
4330  */
4331 static int btrfs_set_inode_index_count(struct inode *inode)
4332 {
4333         struct btrfs_root *root = BTRFS_I(inode)->root;
4334         struct btrfs_key key, found_key;
4335         struct btrfs_path *path;
4336         struct extent_buffer *leaf;
4337         int ret;
4338
4339         key.objectid = inode->i_ino;
4340         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4341         key.offset = (u64)-1;
4342
4343         path = btrfs_alloc_path();
4344         if (!path)
4345                 return -ENOMEM;
4346
4347         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4348         if (ret < 0)
4349                 goto out;
4350         /* FIXME: we should be able to handle this */
4351         if (ret == 0)
4352                 goto out;
4353         ret = 0;
4354
4355         /*
4356          * MAGIC NUMBER EXPLANATION:
4357          * since we search a directory based on f_pos we have to start at 2
4358          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4359          * else has to start at 2
4360          */
4361         if (path->slots[0] == 0) {
4362                 BTRFS_I(inode)->index_cnt = 2;
4363                 goto out;
4364         }
4365
4366         path->slots[0]--;
4367
4368         leaf = path->nodes[0];
4369         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4370
4371         if (found_key.objectid != inode->i_ino ||
4372             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4373                 BTRFS_I(inode)->index_cnt = 2;
4374                 goto out;
4375         }
4376
4377         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4378 out:
4379         btrfs_free_path(path);
4380         return ret;
4381 }
4382
4383 /*
4384  * helper to find a free sequence number in a given directory.  This current
4385  * code is very simple, later versions will do smarter things in the btree
4386  */
4387 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4388 {
4389         int ret = 0;
4390
4391         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4392                 ret = btrfs_set_inode_index_count(dir);
4393                 if (ret)
4394                         return ret;
4395         }
4396
4397         *index = BTRFS_I(dir)->index_cnt;
4398         BTRFS_I(dir)->index_cnt++;
4399
4400         return ret;
4401 }
4402
4403 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4404                                      struct btrfs_root *root,
4405                                      struct inode *dir,
4406                                      const char *name, int name_len,
4407                                      u64 ref_objectid, u64 objectid,
4408                                      u64 alloc_hint, int mode, u64 *index)
4409 {
4410         struct inode *inode;
4411         struct btrfs_inode_item *inode_item;
4412         struct btrfs_key *location;
4413         struct btrfs_path *path;
4414         struct btrfs_inode_ref *ref;
4415         struct btrfs_key key[2];
4416         u32 sizes[2];
4417         unsigned long ptr;
4418         int ret;
4419         int owner;
4420
4421         path = btrfs_alloc_path();
4422         BUG_ON(!path);
4423
4424         inode = new_inode(root->fs_info->sb);
4425         if (!inode)
4426                 return ERR_PTR(-ENOMEM);
4427
4428         if (dir) {
4429                 ret = btrfs_set_inode_index(dir, index);
4430                 if (ret) {
4431                         iput(inode);
4432                         return ERR_PTR(ret);
4433                 }
4434         }
4435         /*
4436          * index_cnt is ignored for everything but a dir,
4437          * btrfs_get_inode_index_count has an explanation for the magic
4438          * number
4439          */
4440         BTRFS_I(inode)->index_cnt = 2;
4441         BTRFS_I(inode)->root = root;
4442         BTRFS_I(inode)->generation = trans->transid;
4443         btrfs_set_inode_space_info(root, inode);
4444
4445         if (mode & S_IFDIR)
4446                 owner = 0;
4447         else
4448                 owner = 1;
4449         BTRFS_I(inode)->block_group =
4450                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4451
4452         key[0].objectid = objectid;
4453         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4454         key[0].offset = 0;
4455
4456         key[1].objectid = objectid;
4457         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4458         key[1].offset = ref_objectid;
4459
4460         sizes[0] = sizeof(struct btrfs_inode_item);
4461         sizes[1] = name_len + sizeof(*ref);
4462
4463         path->leave_spinning = 1;
4464         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4465         if (ret != 0)
4466                 goto fail;
4467
4468         inode->i_uid = current_fsuid();
4469
4470         if (dir && (dir->i_mode & S_ISGID)) {
4471                 inode->i_gid = dir->i_gid;
4472                 if (S_ISDIR(mode))
4473                         mode |= S_ISGID;
4474         } else
4475                 inode->i_gid = current_fsgid();
4476
4477         inode->i_mode = mode;
4478         inode->i_ino = objectid;
4479         inode_set_bytes(inode, 0);
4480         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4481         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4482                                   struct btrfs_inode_item);
4483         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4484
4485         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4486                              struct btrfs_inode_ref);
4487         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4488         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4489         ptr = (unsigned long)(ref + 1);
4490         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4491
4492         btrfs_mark_buffer_dirty(path->nodes[0]);
4493         btrfs_free_path(path);
4494
4495         location = &BTRFS_I(inode)->location;
4496         location->objectid = objectid;
4497         location->offset = 0;
4498         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4499
4500         btrfs_inherit_iflags(inode, dir);
4501
4502         if ((mode & S_IFREG)) {
4503                 if (btrfs_test_opt(root, NODATASUM))
4504                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4505                 if (btrfs_test_opt(root, NODATACOW))
4506                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4507         }
4508
4509         insert_inode_hash(inode);
4510         inode_tree_add(inode);
4511         return inode;
4512 fail:
4513         if (dir)
4514                 BTRFS_I(dir)->index_cnt--;
4515         btrfs_free_path(path);
4516         iput(inode);
4517         return ERR_PTR(ret);
4518 }
4519
4520 static inline u8 btrfs_inode_type(struct inode *inode)
4521 {
4522         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4523 }
4524
4525 /*
4526  * utility function to add 'inode' into 'parent_inode' with
4527  * a give name and a given sequence number.
4528  * if 'add_backref' is true, also insert a backref from the
4529  * inode to the parent directory.
4530  */
4531 int btrfs_add_link(struct btrfs_trans_handle *trans,
4532                    struct inode *parent_inode, struct inode *inode,
4533                    const char *name, int name_len, int add_backref, u64 index)
4534 {
4535         int ret = 0;
4536         struct btrfs_key key;
4537         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4538
4539         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4540                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4541         } else {
4542                 key.objectid = inode->i_ino;
4543                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4544                 key.offset = 0;
4545         }
4546
4547         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4548                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4549                                          key.objectid, root->root_key.objectid,
4550                                          parent_inode->i_ino,
4551                                          index, name, name_len);
4552         } else if (add_backref) {
4553                 ret = btrfs_insert_inode_ref(trans, root,
4554                                              name, name_len, inode->i_ino,
4555                                              parent_inode->i_ino, index);
4556         }
4557
4558         if (ret == 0) {
4559                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4560                                             parent_inode->i_ino, &key,
4561                                             btrfs_inode_type(inode), index);
4562                 BUG_ON(ret);
4563
4564                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4565                                    name_len * 2);
4566                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4567                 ret = btrfs_update_inode(trans, root, parent_inode);
4568         }
4569         return ret;
4570 }
4571
4572 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4573                             struct dentry *dentry, struct inode *inode,
4574                             int backref, u64 index)
4575 {
4576         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4577                                  inode, dentry->d_name.name,
4578                                  dentry->d_name.len, backref, index);
4579         if (!err) {
4580                 d_instantiate(dentry, inode);
4581                 return 0;
4582         }
4583         if (err > 0)
4584                 err = -EEXIST;
4585         return err;
4586 }
4587
4588 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4589                         int mode, dev_t rdev)
4590 {
4591         struct btrfs_trans_handle *trans;
4592         struct btrfs_root *root = BTRFS_I(dir)->root;
4593         struct inode *inode = NULL;
4594         int err;
4595         int drop_inode = 0;
4596         u64 objectid;
4597         unsigned long nr = 0;
4598         u64 index = 0;
4599
4600         if (!new_valid_dev(rdev))
4601                 return -EINVAL;
4602
4603         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4604         if (err)
4605                 return err;
4606
4607         /*
4608          * 2 for inode item and ref
4609          * 2 for dir items
4610          * 1 for xattr if selinux is on
4611          */
4612         trans = btrfs_start_transaction(root, 5);
4613         if (IS_ERR(trans))
4614                 return PTR_ERR(trans);
4615
4616         btrfs_set_trans_block_group(trans, dir);
4617
4618         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4619                                 dentry->d_name.len,
4620                                 dentry->d_parent->d_inode->i_ino, objectid,
4621                                 BTRFS_I(dir)->block_group, mode, &index);
4622         err = PTR_ERR(inode);
4623         if (IS_ERR(inode))
4624                 goto out_unlock;
4625
4626         err = btrfs_init_inode_security(trans, inode, dir);
4627         if (err) {
4628                 drop_inode = 1;
4629                 goto out_unlock;
4630         }
4631
4632         btrfs_set_trans_block_group(trans, inode);
4633         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4634         if (err)
4635                 drop_inode = 1;
4636         else {
4637                 inode->i_op = &btrfs_special_inode_operations;
4638                 init_special_inode(inode, inode->i_mode, rdev);
4639                 btrfs_update_inode(trans, root, inode);
4640         }
4641         btrfs_update_inode_block_group(trans, inode);
4642         btrfs_update_inode_block_group(trans, dir);
4643 out_unlock:
4644         nr = trans->blocks_used;
4645         btrfs_end_transaction_throttle(trans, root);
4646         btrfs_btree_balance_dirty(root, nr);
4647         if (drop_inode) {
4648                 inode_dec_link_count(inode);
4649                 iput(inode);
4650         }
4651         return err;
4652 }
4653
4654 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4655                         int mode, struct nameidata *nd)
4656 {
4657         struct btrfs_trans_handle *trans;
4658         struct btrfs_root *root = BTRFS_I(dir)->root;
4659         struct inode *inode = NULL;
4660         int drop_inode = 0;
4661         int err;
4662         unsigned long nr = 0;
4663         u64 objectid;
4664         u64 index = 0;
4665
4666         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4667         if (err)
4668                 return err;
4669         /*
4670          * 2 for inode item and ref
4671          * 2 for dir items
4672          * 1 for xattr if selinux is on
4673          */
4674         trans = btrfs_start_transaction(root, 5);
4675         if (IS_ERR(trans))
4676                 return PTR_ERR(trans);
4677
4678         btrfs_set_trans_block_group(trans, dir);
4679
4680         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4681                                 dentry->d_name.len,
4682                                 dentry->d_parent->d_inode->i_ino,
4683                                 objectid, BTRFS_I(dir)->block_group, mode,
4684                                 &index);
4685         err = PTR_ERR(inode);
4686         if (IS_ERR(inode))
4687                 goto out_unlock;
4688
4689         err = btrfs_init_inode_security(trans, inode, dir);
4690         if (err) {
4691                 drop_inode = 1;
4692                 goto out_unlock;
4693         }
4694
4695         btrfs_set_trans_block_group(trans, inode);
4696         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4697         if (err)
4698                 drop_inode = 1;
4699         else {
4700                 inode->i_mapping->a_ops = &btrfs_aops;
4701                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4702                 inode->i_fop = &btrfs_file_operations;
4703                 inode->i_op = &btrfs_file_inode_operations;
4704                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4705         }
4706         btrfs_update_inode_block_group(trans, inode);
4707         btrfs_update_inode_block_group(trans, dir);
4708 out_unlock:
4709         nr = trans->blocks_used;
4710         btrfs_end_transaction_throttle(trans, root);
4711         if (drop_inode) {
4712                 inode_dec_link_count(inode);
4713                 iput(inode);
4714         }
4715         btrfs_btree_balance_dirty(root, nr);
4716         return err;
4717 }
4718
4719 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4720                       struct dentry *dentry)
4721 {
4722         struct btrfs_trans_handle *trans;
4723         struct btrfs_root *root = BTRFS_I(dir)->root;
4724         struct inode *inode = old_dentry->d_inode;
4725         u64 index;
4726         unsigned long nr = 0;
4727         int err;
4728         int drop_inode = 0;
4729
4730         if (inode->i_nlink == 0)
4731                 return -ENOENT;
4732
4733         /* do not allow sys_link's with other subvols of the same device */
4734         if (root->objectid != BTRFS_I(inode)->root->objectid)
4735                 return -EPERM;
4736
4737         btrfs_inc_nlink(inode);
4738
4739         err = btrfs_set_inode_index(dir, &index);
4740         if (err)
4741                 goto fail;
4742
4743         /*
4744          * 1 item for inode ref
4745          * 2 items for dir items
4746          */
4747         trans = btrfs_start_transaction(root, 3);
4748         if (IS_ERR(trans)) {
4749                 err = PTR_ERR(trans);
4750                 goto fail;
4751         }
4752
4753         btrfs_set_trans_block_group(trans, dir);
4754         atomic_inc(&inode->i_count);
4755
4756         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4757
4758         if (err) {
4759                 drop_inode = 1;
4760         } else {
4761                 btrfs_update_inode_block_group(trans, dir);
4762                 err = btrfs_update_inode(trans, root, inode);
4763                 BUG_ON(err);
4764                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4765         }
4766
4767         nr = trans->blocks_used;
4768         btrfs_end_transaction_throttle(trans, root);
4769 fail:
4770         if (drop_inode) {
4771                 inode_dec_link_count(inode);
4772                 iput(inode);
4773         }
4774         btrfs_btree_balance_dirty(root, nr);
4775         return err;
4776 }
4777
4778 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4779 {
4780         struct inode *inode = NULL;
4781         struct btrfs_trans_handle *trans;
4782         struct btrfs_root *root = BTRFS_I(dir)->root;
4783         int err = 0;
4784         int drop_on_err = 0;
4785         u64 objectid = 0;
4786         u64 index = 0;
4787         unsigned long nr = 1;
4788
4789         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4790         if (err)
4791                 return err;
4792
4793         /*
4794          * 2 items for inode and ref
4795          * 2 items for dir items
4796          * 1 for xattr if selinux is on
4797          */
4798         trans = btrfs_start_transaction(root, 5);
4799         if (IS_ERR(trans))
4800                 return PTR_ERR(trans);
4801         btrfs_set_trans_block_group(trans, dir);
4802
4803         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4804                                 dentry->d_name.len,
4805                                 dentry->d_parent->d_inode->i_ino, objectid,
4806                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4807                                 &index);
4808         if (IS_ERR(inode)) {
4809                 err = PTR_ERR(inode);
4810                 goto out_fail;
4811         }
4812
4813         drop_on_err = 1;
4814
4815         err = btrfs_init_inode_security(trans, inode, dir);
4816         if (err)
4817                 goto out_fail;
4818
4819         inode->i_op = &btrfs_dir_inode_operations;
4820         inode->i_fop = &btrfs_dir_file_operations;
4821         btrfs_set_trans_block_group(trans, inode);
4822
4823         btrfs_i_size_write(inode, 0);
4824         err = btrfs_update_inode(trans, root, inode);
4825         if (err)
4826                 goto out_fail;
4827
4828         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4829                                  inode, dentry->d_name.name,
4830                                  dentry->d_name.len, 0, index);
4831         if (err)
4832                 goto out_fail;
4833
4834         d_instantiate(dentry, inode);
4835         drop_on_err = 0;
4836         btrfs_update_inode_block_group(trans, inode);
4837         btrfs_update_inode_block_group(trans, dir);
4838
4839 out_fail:
4840         nr = trans->blocks_used;
4841         btrfs_end_transaction_throttle(trans, root);
4842         if (drop_on_err)
4843                 iput(inode);
4844         btrfs_btree_balance_dirty(root, nr);
4845         return err;
4846 }
4847
4848 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4849  * and an extent that you want to insert, deal with overlap and insert
4850  * the new extent into the tree.
4851  */
4852 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4853                                 struct extent_map *existing,
4854                                 struct extent_map *em,
4855                                 u64 map_start, u64 map_len)
4856 {
4857         u64 start_diff;
4858
4859         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4860         start_diff = map_start - em->start;
4861         em->start = map_start;
4862         em->len = map_len;
4863         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4864             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4865                 em->block_start += start_diff;
4866                 em->block_len -= start_diff;
4867         }
4868         return add_extent_mapping(em_tree, em);
4869 }
4870
4871 static noinline int uncompress_inline(struct btrfs_path *path,
4872                                       struct inode *inode, struct page *page,
4873                                       size_t pg_offset, u64 extent_offset,
4874                                       struct btrfs_file_extent_item *item)
4875 {
4876         int ret;
4877         struct extent_buffer *leaf = path->nodes[0];
4878         char *tmp;
4879         size_t max_size;
4880         unsigned long inline_size;
4881         unsigned long ptr;
4882
4883         WARN_ON(pg_offset != 0);
4884         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4885         inline_size = btrfs_file_extent_inline_item_len(leaf,
4886                                         btrfs_item_nr(leaf, path->slots[0]));
4887         tmp = kmalloc(inline_size, GFP_NOFS);
4888         ptr = btrfs_file_extent_inline_start(item);
4889
4890         read_extent_buffer(leaf, tmp, ptr, inline_size);
4891
4892         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4893         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4894                                     inline_size, max_size);
4895         if (ret) {
4896                 char *kaddr = kmap_atomic(page, KM_USER0);
4897                 unsigned long copy_size = min_t(u64,
4898                                   PAGE_CACHE_SIZE - pg_offset,
4899                                   max_size - extent_offset);
4900                 memset(kaddr + pg_offset, 0, copy_size);
4901                 kunmap_atomic(kaddr, KM_USER0);
4902         }
4903         kfree(tmp);
4904         return 0;
4905 }
4906
4907 /*
4908  * a bit scary, this does extent mapping from logical file offset to the disk.
4909  * the ugly parts come from merging extents from the disk with the in-ram
4910  * representation.  This gets more complex because of the data=ordered code,
4911  * where the in-ram extents might be locked pending data=ordered completion.
4912  *
4913  * This also copies inline extents directly into the page.
4914  */
4915
4916 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4917                                     size_t pg_offset, u64 start, u64 len,
4918                                     int create)
4919 {
4920         int ret;
4921         int err = 0;
4922         u64 bytenr;
4923         u64 extent_start = 0;
4924         u64 extent_end = 0;
4925         u64 objectid = inode->i_ino;
4926         u32 found_type;
4927         struct btrfs_path *path = NULL;
4928         struct btrfs_root *root = BTRFS_I(inode)->root;
4929         struct btrfs_file_extent_item *item;
4930         struct extent_buffer *leaf;
4931         struct btrfs_key found_key;
4932         struct extent_map *em = NULL;
4933         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4934         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4935         struct btrfs_trans_handle *trans = NULL;
4936         int compressed;
4937
4938 again:
4939         read_lock(&em_tree->lock);
4940         em = lookup_extent_mapping(em_tree, start, len);
4941         if (em)
4942                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4943         read_unlock(&em_tree->lock);
4944
4945         if (em) {
4946                 if (em->start > start || em->start + em->len <= start)
4947                         free_extent_map(em);
4948                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4949                         free_extent_map(em);
4950                 else
4951                         goto out;
4952         }
4953         em = alloc_extent_map(GFP_NOFS);
4954         if (!em) {
4955                 err = -ENOMEM;
4956                 goto out;
4957         }
4958         em->bdev = root->fs_info->fs_devices->latest_bdev;
4959         em->start = EXTENT_MAP_HOLE;
4960         em->orig_start = EXTENT_MAP_HOLE;
4961         em->len = (u64)-1;
4962         em->block_len = (u64)-1;
4963
4964         if (!path) {
4965                 path = btrfs_alloc_path();
4966                 BUG_ON(!path);
4967         }
4968
4969         ret = btrfs_lookup_file_extent(trans, root, path,
4970                                        objectid, start, trans != NULL);
4971         if (ret < 0) {
4972                 err = ret;
4973                 goto out;
4974         }
4975
4976         if (ret != 0) {
4977                 if (path->slots[0] == 0)
4978                         goto not_found;
4979                 path->slots[0]--;
4980         }
4981
4982         leaf = path->nodes[0];
4983         item = btrfs_item_ptr(leaf, path->slots[0],
4984                               struct btrfs_file_extent_item);
4985         /* are we inside the extent that was found? */
4986         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4987         found_type = btrfs_key_type(&found_key);
4988         if (found_key.objectid != objectid ||
4989             found_type != BTRFS_EXTENT_DATA_KEY) {
4990                 goto not_found;
4991         }
4992
4993         found_type = btrfs_file_extent_type(leaf, item);
4994         extent_start = found_key.offset;
4995         compressed = btrfs_file_extent_compression(leaf, item);
4996         if (found_type == BTRFS_FILE_EXTENT_REG ||
4997             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4998                 extent_end = extent_start +
4999                        btrfs_file_extent_num_bytes(leaf, item);
5000         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5001                 size_t size;
5002                 size = btrfs_file_extent_inline_len(leaf, item);
5003                 extent_end = (extent_start + size + root->sectorsize - 1) &
5004                         ~((u64)root->sectorsize - 1);
5005         }
5006
5007         if (start >= extent_end) {
5008                 path->slots[0]++;
5009                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5010                         ret = btrfs_next_leaf(root, path);
5011                         if (ret < 0) {
5012                                 err = ret;
5013                                 goto out;
5014                         }
5015                         if (ret > 0)
5016                                 goto not_found;
5017                         leaf = path->nodes[0];
5018                 }
5019                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5020                 if (found_key.objectid != objectid ||
5021                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5022                         goto not_found;
5023                 if (start + len <= found_key.offset)
5024                         goto not_found;
5025                 em->start = start;
5026                 em->len = found_key.offset - start;
5027                 goto not_found_em;
5028         }
5029
5030         if (found_type == BTRFS_FILE_EXTENT_REG ||
5031             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5032                 em->start = extent_start;
5033                 em->len = extent_end - extent_start;
5034                 em->orig_start = extent_start -
5035                                  btrfs_file_extent_offset(leaf, item);
5036                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5037                 if (bytenr == 0) {
5038                         em->block_start = EXTENT_MAP_HOLE;
5039                         goto insert;
5040                 }
5041                 if (compressed) {
5042                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5043                         em->block_start = bytenr;
5044                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5045                                                                          item);
5046                 } else {
5047                         bytenr += btrfs_file_extent_offset(leaf, item);
5048                         em->block_start = bytenr;
5049                         em->block_len = em->len;
5050                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5051                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5052                 }
5053                 goto insert;
5054         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5055                 unsigned long ptr;
5056                 char *map;
5057                 size_t size;
5058                 size_t extent_offset;
5059                 size_t copy_size;
5060
5061                 em->block_start = EXTENT_MAP_INLINE;
5062                 if (!page || create) {
5063                         em->start = extent_start;
5064                         em->len = extent_end - extent_start;
5065                         goto out;
5066                 }
5067
5068                 size = btrfs_file_extent_inline_len(leaf, item);
5069                 extent_offset = page_offset(page) + pg_offset - extent_start;
5070                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5071                                 size - extent_offset);
5072                 em->start = extent_start + extent_offset;
5073                 em->len = (copy_size + root->sectorsize - 1) &
5074                         ~((u64)root->sectorsize - 1);
5075                 em->orig_start = EXTENT_MAP_INLINE;
5076                 if (compressed)
5077                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5078                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5079                 if (create == 0 && !PageUptodate(page)) {
5080                         if (btrfs_file_extent_compression(leaf, item) ==
5081                             BTRFS_COMPRESS_ZLIB) {
5082                                 ret = uncompress_inline(path, inode, page,
5083                                                         pg_offset,
5084                                                         extent_offset, item);
5085                                 BUG_ON(ret);
5086                         } else {
5087                                 map = kmap(page);
5088                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5089                                                    copy_size);
5090                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5091                                         memset(map + pg_offset + copy_size, 0,
5092                                                PAGE_CACHE_SIZE - pg_offset -
5093                                                copy_size);
5094                                 }
5095                                 kunmap(page);
5096                         }
5097                         flush_dcache_page(page);
5098                 } else if (create && PageUptodate(page)) {
5099                         WARN_ON(1);
5100                         if (!trans) {
5101                                 kunmap(page);
5102                                 free_extent_map(em);
5103                                 em = NULL;
5104                                 btrfs_release_path(root, path);
5105                                 trans = btrfs_join_transaction(root, 1);
5106                                 goto again;
5107                         }
5108                         map = kmap(page);
5109                         write_extent_buffer(leaf, map + pg_offset, ptr,
5110                                             copy_size);
5111                         kunmap(page);
5112                         btrfs_mark_buffer_dirty(leaf);
5113                 }
5114                 set_extent_uptodate(io_tree, em->start,
5115                                     extent_map_end(em) - 1, GFP_NOFS);
5116                 goto insert;
5117         } else {
5118                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5119                 WARN_ON(1);
5120         }
5121 not_found:
5122         em->start = start;
5123         em->len = len;
5124 not_found_em:
5125         em->block_start = EXTENT_MAP_HOLE;
5126         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5127 insert:
5128         btrfs_release_path(root, path);
5129         if (em->start > start || extent_map_end(em) <= start) {
5130                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5131                        "[%llu %llu]\n", (unsigned long long)em->start,
5132                        (unsigned long long)em->len,
5133                        (unsigned long long)start,
5134                        (unsigned long long)len);
5135                 err = -EIO;
5136                 goto out;
5137         }
5138
5139         err = 0;
5140         write_lock(&em_tree->lock);
5141         ret = add_extent_mapping(em_tree, em);
5142         /* it is possible that someone inserted the extent into the tree
5143          * while we had the lock dropped.  It is also possible that
5144          * an overlapping map exists in the tree
5145          */
5146         if (ret == -EEXIST) {
5147                 struct extent_map *existing;
5148
5149                 ret = 0;
5150
5151                 existing = lookup_extent_mapping(em_tree, start, len);
5152                 if (existing && (existing->start > start ||
5153                     existing->start + existing->len <= start)) {
5154                         free_extent_map(existing);
5155                         existing = NULL;
5156                 }
5157                 if (!existing) {
5158                         existing = lookup_extent_mapping(em_tree, em->start,
5159                                                          em->len);
5160                         if (existing) {
5161                                 err = merge_extent_mapping(em_tree, existing,
5162                                                            em, start,
5163                                                            root->sectorsize);
5164                                 free_extent_map(existing);
5165                                 if (err) {
5166                                         free_extent_map(em);
5167                                         em = NULL;
5168                                 }
5169                         } else {
5170                                 err = -EIO;
5171                                 free_extent_map(em);
5172                                 em = NULL;
5173                         }
5174                 } else {
5175                         free_extent_map(em);
5176                         em = existing;
5177                         err = 0;
5178                 }
5179         }
5180         write_unlock(&em_tree->lock);
5181 out:
5182         if (path)
5183                 btrfs_free_path(path);
5184         if (trans) {
5185                 ret = btrfs_end_transaction(trans, root);
5186                 if (!err)
5187                         err = ret;
5188         }
5189         if (err) {
5190                 free_extent_map(em);
5191                 return ERR_PTR(err);
5192         }
5193         return em;
5194 }
5195
5196 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5197                                                   u64 start, u64 len)
5198 {
5199         struct btrfs_root *root = BTRFS_I(inode)->root;
5200         struct btrfs_trans_handle *trans;
5201         struct extent_map *em;
5202         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5203         struct btrfs_key ins;
5204         u64 alloc_hint;
5205         int ret;
5206
5207         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5208
5209         trans = btrfs_join_transaction(root, 0);
5210         if (!trans)
5211                 return ERR_PTR(-ENOMEM);
5212
5213         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5214
5215         alloc_hint = get_extent_allocation_hint(inode, start, len);
5216         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5217                                    alloc_hint, (u64)-1, &ins, 1);
5218         if (ret) {
5219                 em = ERR_PTR(ret);
5220                 goto out;
5221         }
5222
5223         em = alloc_extent_map(GFP_NOFS);
5224         if (!em) {
5225                 em = ERR_PTR(-ENOMEM);
5226                 goto out;
5227         }
5228
5229         em->start = start;
5230         em->orig_start = em->start;
5231         em->len = ins.offset;
5232
5233         em->block_start = ins.objectid;
5234         em->block_len = ins.offset;
5235         em->bdev = root->fs_info->fs_devices->latest_bdev;
5236         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5237
5238         while (1) {
5239                 write_lock(&em_tree->lock);
5240                 ret = add_extent_mapping(em_tree, em);
5241                 write_unlock(&em_tree->lock);
5242                 if (ret != -EEXIST)
5243                         break;
5244                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5245         }
5246
5247         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5248                                            ins.offset, ins.offset, 0);
5249         if (ret) {
5250                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5251                 em = ERR_PTR(ret);
5252         }
5253 out:
5254         btrfs_end_transaction(trans, root);
5255         return em;
5256 }
5257
5258 /*
5259  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5260  * block must be cow'd
5261  */
5262 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5263                                       struct inode *inode, u64 offset, u64 len)
5264 {
5265         struct btrfs_path *path;
5266         int ret;
5267         struct extent_buffer *leaf;
5268         struct btrfs_root *root = BTRFS_I(inode)->root;
5269         struct btrfs_file_extent_item *fi;
5270         struct btrfs_key key;
5271         u64 disk_bytenr;
5272         u64 backref_offset;
5273         u64 extent_end;
5274         u64 num_bytes;
5275         int slot;
5276         int found_type;
5277
5278         path = btrfs_alloc_path();
5279         if (!path)
5280                 return -ENOMEM;
5281
5282         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5283                                        offset, 0);
5284         if (ret < 0)
5285                 goto out;
5286
5287         slot = path->slots[0];
5288         if (ret == 1) {
5289                 if (slot == 0) {
5290                         /* can't find the item, must cow */
5291                         ret = 0;
5292                         goto out;
5293                 }
5294                 slot--;
5295         }
5296         ret = 0;
5297         leaf = path->nodes[0];
5298         btrfs_item_key_to_cpu(leaf, &key, slot);
5299         if (key.objectid != inode->i_ino ||
5300             key.type != BTRFS_EXTENT_DATA_KEY) {
5301                 /* not our file or wrong item type, must cow */
5302                 goto out;
5303         }
5304
5305         if (key.offset > offset) {
5306                 /* Wrong offset, must cow */
5307                 goto out;
5308         }
5309
5310         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5311         found_type = btrfs_file_extent_type(leaf, fi);
5312         if (found_type != BTRFS_FILE_EXTENT_REG &&
5313             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5314                 /* not a regular extent, must cow */
5315                 goto out;
5316         }
5317         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5318         backref_offset = btrfs_file_extent_offset(leaf, fi);
5319
5320         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5321         if (extent_end < offset + len) {
5322                 /* extent doesn't include our full range, must cow */
5323                 goto out;
5324         }
5325
5326         if (btrfs_extent_readonly(root, disk_bytenr))
5327                 goto out;
5328
5329         /*
5330          * look for other files referencing this extent, if we
5331          * find any we must cow
5332          */
5333         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5334                                   key.offset - backref_offset, disk_bytenr))
5335                 goto out;
5336
5337         /*
5338          * adjust disk_bytenr and num_bytes to cover just the bytes
5339          * in this extent we are about to write.  If there
5340          * are any csums in that range we have to cow in order
5341          * to keep the csums correct
5342          */
5343         disk_bytenr += backref_offset;
5344         disk_bytenr += offset - key.offset;
5345         num_bytes = min(offset + len, extent_end) - offset;
5346         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5347                                 goto out;
5348         /*
5349          * all of the above have passed, it is safe to overwrite this extent
5350          * without cow
5351          */
5352         ret = 1;
5353 out:
5354         btrfs_free_path(path);
5355         return ret;
5356 }
5357
5358 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5359                                    struct buffer_head *bh_result, int create)
5360 {
5361         struct extent_map *em;
5362         struct btrfs_root *root = BTRFS_I(inode)->root;
5363         u64 start = iblock << inode->i_blkbits;
5364         u64 len = bh_result->b_size;
5365         struct btrfs_trans_handle *trans;
5366
5367         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5368         if (IS_ERR(em))
5369                 return PTR_ERR(em);
5370
5371         /*
5372          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5373          * io.  INLINE is special, and we could probably kludge it in here, but
5374          * it's still buffered so for safety lets just fall back to the generic
5375          * buffered path.
5376          *
5377          * For COMPRESSED we _have_ to read the entire extent in so we can
5378          * decompress it, so there will be buffering required no matter what we
5379          * do, so go ahead and fallback to buffered.
5380          *
5381          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5382          * to buffered IO.  Don't blame me, this is the price we pay for using
5383          * the generic code.
5384          */
5385         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5386             em->block_start == EXTENT_MAP_INLINE) {
5387                 free_extent_map(em);
5388                 return -ENOTBLK;
5389         }
5390
5391         /* Just a good old fashioned hole, return */
5392         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5393                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5394                 free_extent_map(em);
5395                 /* DIO will do one hole at a time, so just unlock a sector */
5396                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5397                               start + root->sectorsize - 1, GFP_NOFS);
5398                 return 0;
5399         }
5400
5401         /*
5402          * We don't allocate a new extent in the following cases
5403          *
5404          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5405          * existing extent.
5406          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5407          * just use the extent.
5408          *
5409          */
5410         if (!create) {
5411                 len = em->len - (start - em->start);
5412                 goto map;
5413         }
5414
5415         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5416             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5417              em->block_start != EXTENT_MAP_HOLE)) {
5418                 int type;
5419                 int ret;
5420                 u64 block_start;
5421
5422                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5423                         type = BTRFS_ORDERED_PREALLOC;
5424                 else
5425                         type = BTRFS_ORDERED_NOCOW;
5426                 len = min(len, em->len - (start - em->start));
5427                 block_start = em->block_start + (start - em->start);
5428
5429                 /*
5430                  * we're not going to log anything, but we do need
5431                  * to make sure the current transaction stays open
5432                  * while we look for nocow cross refs
5433                  */
5434                 trans = btrfs_join_transaction(root, 0);
5435                 if (!trans)
5436                         goto must_cow;
5437
5438                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5439                         ret = btrfs_add_ordered_extent_dio(inode, start,
5440                                            block_start, len, len, type);
5441                         btrfs_end_transaction(trans, root);
5442                         if (ret) {
5443                                 free_extent_map(em);
5444                                 return ret;
5445                         }
5446                         goto unlock;
5447                 }
5448                 btrfs_end_transaction(trans, root);
5449         }
5450 must_cow:
5451         /*
5452          * this will cow the extent, reset the len in case we changed
5453          * it above
5454          */
5455         len = bh_result->b_size;
5456         free_extent_map(em);
5457         em = btrfs_new_extent_direct(inode, start, len);
5458         if (IS_ERR(em))
5459                 return PTR_ERR(em);
5460         len = min(len, em->len - (start - em->start));
5461 unlock:
5462         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5463                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5464                           0, NULL, GFP_NOFS);
5465 map:
5466         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5467                 inode->i_blkbits;
5468         bh_result->b_size = len;
5469         bh_result->b_bdev = em->bdev;
5470         set_buffer_mapped(bh_result);
5471         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5472                 set_buffer_new(bh_result);
5473
5474         free_extent_map(em);
5475
5476         return 0;
5477 }
5478
5479 struct btrfs_dio_private {
5480         struct inode *inode;
5481         u64 logical_offset;
5482         u64 disk_bytenr;
5483         u64 bytes;
5484         u32 *csums;
5485         void *private;
5486 };
5487
5488 static void btrfs_endio_direct_read(struct bio *bio, int err)
5489 {
5490         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5491         struct bio_vec *bvec = bio->bi_io_vec;
5492         struct btrfs_dio_private *dip = bio->bi_private;
5493         struct inode *inode = dip->inode;
5494         struct btrfs_root *root = BTRFS_I(inode)->root;
5495         u64 start;
5496         u32 *private = dip->csums;
5497
5498         start = dip->logical_offset;
5499         do {
5500                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5501                         struct page *page = bvec->bv_page;
5502                         char *kaddr;
5503                         u32 csum = ~(u32)0;
5504                         unsigned long flags;
5505
5506                         local_irq_save(flags);
5507                         kaddr = kmap_atomic(page, KM_IRQ0);
5508                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5509                                                csum, bvec->bv_len);
5510                         btrfs_csum_final(csum, (char *)&csum);
5511                         kunmap_atomic(kaddr, KM_IRQ0);
5512                         local_irq_restore(flags);
5513
5514                         flush_dcache_page(bvec->bv_page);
5515                         if (csum != *private) {
5516                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5517                                       " %llu csum %u private %u\n",
5518                                       inode->i_ino, (unsigned long long)start,
5519                                       csum, *private);
5520                                 err = -EIO;
5521                         }
5522                 }
5523
5524                 start += bvec->bv_len;
5525                 private++;
5526                 bvec++;
5527         } while (bvec <= bvec_end);
5528
5529         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5530                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5531         bio->bi_private = dip->private;
5532
5533         kfree(dip->csums);
5534         kfree(dip);
5535         dio_end_io(bio, err);
5536 }
5537
5538 static void btrfs_endio_direct_write(struct bio *bio, int err)
5539 {
5540         struct btrfs_dio_private *dip = bio->bi_private;
5541         struct inode *inode = dip->inode;
5542         struct btrfs_root *root = BTRFS_I(inode)->root;
5543         struct btrfs_trans_handle *trans;
5544         struct btrfs_ordered_extent *ordered = NULL;
5545         struct extent_state *cached_state = NULL;
5546         int ret;
5547
5548         if (err)
5549                 goto out_done;
5550
5551         ret = btrfs_dec_test_ordered_pending(inode, &ordered,
5552                                              dip->logical_offset, dip->bytes);
5553         if (!ret)
5554                 goto out_done;
5555
5556         BUG_ON(!ordered);
5557
5558         trans = btrfs_join_transaction(root, 1);
5559         if (!trans) {
5560                 err = -ENOMEM;
5561                 goto out;
5562         }
5563         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5564
5565         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5566                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5567                 if (!ret)
5568                         ret = btrfs_update_inode(trans, root, inode);
5569                 err = ret;
5570                 goto out;
5571         }
5572
5573         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5574                          ordered->file_offset + ordered->len - 1, 0,
5575                          &cached_state, GFP_NOFS);
5576
5577         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5578                 ret = btrfs_mark_extent_written(trans, inode,
5579                                                 ordered->file_offset,
5580                                                 ordered->file_offset +
5581                                                 ordered->len);
5582                 if (ret) {
5583                         err = ret;
5584                         goto out_unlock;
5585                 }
5586         } else {
5587                 ret = insert_reserved_file_extent(trans, inode,
5588                                                   ordered->file_offset,
5589                                                   ordered->start,
5590                                                   ordered->disk_len,
5591                                                   ordered->len,
5592                                                   ordered->len,
5593                                                   0, 0, 0,
5594                                                   BTRFS_FILE_EXTENT_REG);
5595                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5596                                    ordered->file_offset, ordered->len);
5597                 if (ret) {
5598                         err = ret;
5599                         WARN_ON(1);
5600                         goto out_unlock;
5601                 }
5602         }
5603
5604         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5605         btrfs_ordered_update_i_size(inode, 0, ordered);
5606         btrfs_update_inode(trans, root, inode);
5607 out_unlock:
5608         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5609                              ordered->file_offset + ordered->len - 1,
5610                              &cached_state, GFP_NOFS);
5611 out:
5612         btrfs_delalloc_release_metadata(inode, ordered->len);
5613         btrfs_end_transaction(trans, root);
5614         btrfs_put_ordered_extent(ordered);
5615         btrfs_put_ordered_extent(ordered);
5616 out_done:
5617         bio->bi_private = dip->private;
5618
5619         kfree(dip->csums);
5620         kfree(dip);
5621         dio_end_io(bio, err);
5622 }
5623
5624 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5625                                     struct bio *bio, int mirror_num,
5626                                     unsigned long bio_flags, u64 offset)
5627 {
5628         int ret;
5629         struct btrfs_root *root = BTRFS_I(inode)->root;
5630         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5631         BUG_ON(ret);
5632         return 0;
5633 }
5634
5635 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5636                                 loff_t file_offset)
5637 {
5638         struct btrfs_root *root = BTRFS_I(inode)->root;
5639         struct btrfs_dio_private *dip;
5640         struct bio_vec *bvec = bio->bi_io_vec;
5641         u64 start;
5642         int skip_sum;
5643         int write = rw & (1 << BIO_RW);
5644         int ret = 0;
5645
5646         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5647
5648         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5649         if (!dip) {
5650                 ret = -ENOMEM;
5651                 goto free_ordered;
5652         }
5653         dip->csums = NULL;
5654
5655         if (!skip_sum) {
5656                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5657                 if (!dip->csums) {
5658                         ret = -ENOMEM;
5659                         goto free_ordered;
5660                 }
5661         }
5662
5663         dip->private = bio->bi_private;
5664         dip->inode = inode;
5665         dip->logical_offset = file_offset;
5666
5667         start = dip->logical_offset;
5668         dip->bytes = 0;
5669         do {
5670                 dip->bytes += bvec->bv_len;
5671                 bvec++;
5672         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5673
5674         dip->disk_bytenr = (u64)bio->bi_sector << 9;
5675         bio->bi_private = dip;
5676
5677         if (write)
5678                 bio->bi_end_io = btrfs_endio_direct_write;
5679         else
5680                 bio->bi_end_io = btrfs_endio_direct_read;
5681
5682         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5683         if (ret)
5684                 goto out_err;
5685
5686         if (write && !skip_sum) {
5687                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
5688                                    inode, rw, bio, 0, 0,
5689                                    dip->logical_offset,
5690                                    __btrfs_submit_bio_start_direct_io,
5691                                    __btrfs_submit_bio_done);
5692                 if (ret)
5693                         goto out_err;
5694                 return;
5695         } else if (!skip_sum)
5696                 btrfs_lookup_bio_sums_dio(root, inode, bio,
5697                                           dip->logical_offset, dip->csums);
5698
5699         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5700         if (ret)
5701                 goto out_err;
5702         return;
5703 out_err:
5704         kfree(dip->csums);
5705         kfree(dip);
5706 free_ordered:
5707         /*
5708          * If this is a write, we need to clean up the reserved space and kill
5709          * the ordered extent.
5710          */
5711         if (write) {
5712                 struct btrfs_ordered_extent *ordered;
5713                 ordered = btrfs_lookup_ordered_extent(inode,
5714                                                       dip->logical_offset);
5715                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
5716                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
5717                         btrfs_free_reserved_extent(root, ordered->start,
5718                                                    ordered->disk_len);
5719                 btrfs_put_ordered_extent(ordered);
5720                 btrfs_put_ordered_extent(ordered);
5721         }
5722         bio_endio(bio, ret);
5723 }
5724
5725 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
5726                         const struct iovec *iov, loff_t offset,
5727                         unsigned long nr_segs)
5728 {
5729         int seg;
5730         size_t size;
5731         unsigned long addr;
5732         unsigned blocksize_mask = root->sectorsize - 1;
5733         ssize_t retval = -EINVAL;
5734         loff_t end = offset;
5735
5736         if (offset & blocksize_mask)
5737                 goto out;
5738
5739         /* Check the memory alignment.  Blocks cannot straddle pages */
5740         for (seg = 0; seg < nr_segs; seg++) {
5741                 addr = (unsigned long)iov[seg].iov_base;
5742                 size = iov[seg].iov_len;
5743                 end += size;
5744                 if ((addr & blocksize_mask) || (size & blocksize_mask)) 
5745                         goto out;
5746         }
5747         retval = 0;
5748 out:
5749         return retval;
5750 }
5751 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
5752                         const struct iovec *iov, loff_t offset,
5753                         unsigned long nr_segs)
5754 {
5755         struct file *file = iocb->ki_filp;
5756         struct inode *inode = file->f_mapping->host;
5757         struct btrfs_ordered_extent *ordered;
5758         struct extent_state *cached_state = NULL;
5759         u64 lockstart, lockend;
5760         ssize_t ret;
5761         int writing = rw & WRITE;
5762         int write_bits = 0;
5763         size_t count = iov_length(iov, nr_segs);
5764
5765         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
5766                             offset, nr_segs)) {
5767                 return 0;
5768         }
5769
5770         lockstart = offset;
5771         lockend = offset + count - 1;
5772
5773         if (writing) {
5774                 ret = btrfs_delalloc_reserve_space(inode, count);
5775                 if (ret)
5776                         goto out;
5777         }
5778
5779         while (1) {
5780                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5781                                  0, &cached_state, GFP_NOFS);
5782                 /*
5783                  * We're concerned with the entire range that we're going to be
5784                  * doing DIO to, so we need to make sure theres no ordered
5785                  * extents in this range.
5786                  */
5787                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5788                                                      lockend - lockstart + 1);
5789                 if (!ordered)
5790                         break;
5791                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5792                                      &cached_state, GFP_NOFS);
5793                 btrfs_start_ordered_extent(inode, ordered, 1);
5794                 btrfs_put_ordered_extent(ordered);
5795                 cond_resched();
5796         }
5797
5798         /*
5799          * we don't use btrfs_set_extent_delalloc because we don't want
5800          * the dirty or uptodate bits
5801          */
5802         if (writing) {
5803                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
5804                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5805                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
5806                                      GFP_NOFS);
5807                 if (ret) {
5808                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5809                                          lockend, EXTENT_LOCKED | write_bits,
5810                                          1, 0, &cached_state, GFP_NOFS);
5811                         goto out;
5812                 }
5813         }
5814
5815         free_extent_state(cached_state);
5816         cached_state = NULL;
5817
5818         ret = __blockdev_direct_IO(rw, iocb, inode,
5819                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
5820                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
5821                    btrfs_submit_direct, 0);
5822
5823         if (ret < 0 && ret != -EIOCBQUEUED) {
5824                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
5825                               offset + iov_length(iov, nr_segs) - 1,
5826                               EXTENT_LOCKED | write_bits, 1, 0,
5827                               &cached_state, GFP_NOFS);
5828         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
5829                 /*
5830                  * We're falling back to buffered, unlock the section we didn't
5831                  * do IO on.
5832                  */
5833                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
5834                               offset + iov_length(iov, nr_segs) - 1,
5835                               EXTENT_LOCKED | write_bits, 1, 0,
5836                               &cached_state, GFP_NOFS);
5837         }
5838 out:
5839         free_extent_state(cached_state);
5840         return ret;
5841 }
5842
5843 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5844                 __u64 start, __u64 len)
5845 {
5846         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
5847 }
5848
5849 int btrfs_readpage(struct file *file, struct page *page)
5850 {
5851         struct extent_io_tree *tree;
5852         tree = &BTRFS_I(page->mapping->host)->io_tree;
5853         return extent_read_full_page(tree, page, btrfs_get_extent);
5854 }
5855
5856 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
5857 {
5858         struct extent_io_tree *tree;
5859
5860
5861         if (current->flags & PF_MEMALLOC) {
5862                 redirty_page_for_writepage(wbc, page);
5863                 unlock_page(page);
5864                 return 0;
5865         }
5866         tree = &BTRFS_I(page->mapping->host)->io_tree;
5867         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
5868 }
5869
5870 int btrfs_writepages(struct address_space *mapping,
5871                      struct writeback_control *wbc)
5872 {
5873         struct extent_io_tree *tree;
5874
5875         tree = &BTRFS_I(mapping->host)->io_tree;
5876         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
5877 }
5878
5879 static int
5880 btrfs_readpages(struct file *file, struct address_space *mapping,
5881                 struct list_head *pages, unsigned nr_pages)
5882 {
5883         struct extent_io_tree *tree;
5884         tree = &BTRFS_I(mapping->host)->io_tree;
5885         return extent_readpages(tree, mapping, pages, nr_pages,
5886                                 btrfs_get_extent);
5887 }
5888 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5889 {
5890         struct extent_io_tree *tree;
5891         struct extent_map_tree *map;
5892         int ret;
5893
5894         tree = &BTRFS_I(page->mapping->host)->io_tree;
5895         map = &BTRFS_I(page->mapping->host)->extent_tree;
5896         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
5897         if (ret == 1) {
5898                 ClearPagePrivate(page);
5899                 set_page_private(page, 0);
5900                 page_cache_release(page);
5901         }
5902         return ret;
5903 }
5904
5905 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5906 {
5907         if (PageWriteback(page) || PageDirty(page))
5908                 return 0;
5909         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5910 }
5911
5912 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5913 {
5914         struct extent_io_tree *tree;
5915         struct btrfs_ordered_extent *ordered;
5916         struct extent_state *cached_state = NULL;
5917         u64 page_start = page_offset(page);
5918         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5919
5920
5921         /*
5922          * we have the page locked, so new writeback can't start,
5923          * and the dirty bit won't be cleared while we are here.
5924          *
5925          * Wait for IO on this page so that we can safely clear
5926          * the PagePrivate2 bit and do ordered accounting
5927          */
5928         wait_on_page_writeback(page);
5929
5930         tree = &BTRFS_I(page->mapping->host)->io_tree;
5931         if (offset) {
5932                 btrfs_releasepage(page, GFP_NOFS);
5933                 return;
5934         }
5935         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5936                          GFP_NOFS);
5937         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5938                                            page_offset(page));
5939         if (ordered) {
5940                 /*
5941                  * IO on this page will never be started, so we need
5942                  * to account for any ordered extents now
5943                  */
5944                 clear_extent_bit(tree, page_start, page_end,
5945                                  EXTENT_DIRTY | EXTENT_DELALLOC |
5946                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5947                                  &cached_state, GFP_NOFS);
5948                 /*
5949                  * whoever cleared the private bit is responsible
5950                  * for the finish_ordered_io
5951                  */
5952                 if (TestClearPagePrivate2(page)) {
5953                         btrfs_finish_ordered_io(page->mapping->host,
5954                                                 page_start, page_end);
5955                 }
5956                 btrfs_put_ordered_extent(ordered);
5957                 cached_state = NULL;
5958                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5959                                  GFP_NOFS);
5960         }
5961         clear_extent_bit(tree, page_start, page_end,
5962                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5963                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5964         __btrfs_releasepage(page, GFP_NOFS);
5965
5966         ClearPageChecked(page);
5967         if (PagePrivate(page)) {
5968                 ClearPagePrivate(page);
5969                 set_page_private(page, 0);
5970                 page_cache_release(page);
5971         }
5972 }
5973
5974 /*
5975  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5976  * called from a page fault handler when a page is first dirtied. Hence we must
5977  * be careful to check for EOF conditions here. We set the page up correctly
5978  * for a written page which means we get ENOSPC checking when writing into
5979  * holes and correct delalloc and unwritten extent mapping on filesystems that
5980  * support these features.
5981  *
5982  * We are not allowed to take the i_mutex here so we have to play games to
5983  * protect against truncate races as the page could now be beyond EOF.  Because
5984  * vmtruncate() writes the inode size before removing pages, once we have the
5985  * page lock we can determine safely if the page is beyond EOF. If it is not
5986  * beyond EOF, then the page is guaranteed safe against truncation until we
5987  * unlock the page.
5988  */
5989 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5990 {
5991         struct page *page = vmf->page;
5992         struct inode *inode = fdentry(vma->vm_file)->d_inode;
5993         struct btrfs_root *root = BTRFS_I(inode)->root;
5994         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5995         struct btrfs_ordered_extent *ordered;
5996         struct extent_state *cached_state = NULL;
5997         char *kaddr;
5998         unsigned long zero_start;
5999         loff_t size;
6000         int ret;
6001         u64 page_start;
6002         u64 page_end;
6003
6004         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6005         if (ret) {
6006                 if (ret == -ENOMEM)
6007                         ret = VM_FAULT_OOM;
6008                 else /* -ENOSPC, -EIO, etc */
6009                         ret = VM_FAULT_SIGBUS;
6010                 goto out;
6011         }
6012
6013         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6014 again:
6015         lock_page(page);
6016         size = i_size_read(inode);
6017         page_start = page_offset(page);
6018         page_end = page_start + PAGE_CACHE_SIZE - 1;
6019
6020         if ((page->mapping != inode->i_mapping) ||
6021             (page_start >= size)) {
6022                 /* page got truncated out from underneath us */
6023                 goto out_unlock;
6024         }
6025         wait_on_page_writeback(page);
6026
6027         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6028                          GFP_NOFS);
6029         set_page_extent_mapped(page);
6030
6031         /*
6032          * we can't set the delalloc bits if there are pending ordered
6033          * extents.  Drop our locks and wait for them to finish
6034          */
6035         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6036         if (ordered) {
6037                 unlock_extent_cached(io_tree, page_start, page_end,
6038                                      &cached_state, GFP_NOFS);
6039                 unlock_page(page);
6040                 btrfs_start_ordered_extent(inode, ordered, 1);
6041                 btrfs_put_ordered_extent(ordered);
6042                 goto again;
6043         }
6044
6045         /*
6046          * XXX - page_mkwrite gets called every time the page is dirtied, even
6047          * if it was already dirty, so for space accounting reasons we need to
6048          * clear any delalloc bits for the range we are fixing to save.  There
6049          * is probably a better way to do this, but for now keep consistent with
6050          * prepare_pages in the normal write path.
6051          */
6052         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6053                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6054                           0, 0, &cached_state, GFP_NOFS);
6055
6056         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6057                                         &cached_state);
6058         if (ret) {
6059                 unlock_extent_cached(io_tree, page_start, page_end,
6060                                      &cached_state, GFP_NOFS);
6061                 ret = VM_FAULT_SIGBUS;
6062                 goto out_unlock;
6063         }
6064         ret = 0;
6065
6066         /* page is wholly or partially inside EOF */
6067         if (page_start + PAGE_CACHE_SIZE > size)
6068                 zero_start = size & ~PAGE_CACHE_MASK;
6069         else
6070                 zero_start = PAGE_CACHE_SIZE;
6071
6072         if (zero_start != PAGE_CACHE_SIZE) {
6073                 kaddr = kmap(page);
6074                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6075                 flush_dcache_page(page);
6076                 kunmap(page);
6077         }
6078         ClearPageChecked(page);
6079         set_page_dirty(page);
6080         SetPageUptodate(page);
6081
6082         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6083         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6084
6085         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6086
6087 out_unlock:
6088         if (!ret)
6089                 return VM_FAULT_LOCKED;
6090         unlock_page(page);
6091         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6092 out:
6093         return ret;
6094 }
6095
6096 static void btrfs_truncate(struct inode *inode)
6097 {
6098         struct btrfs_root *root = BTRFS_I(inode)->root;
6099         int ret;
6100         struct btrfs_trans_handle *trans;
6101         unsigned long nr;
6102         u64 mask = root->sectorsize - 1;
6103
6104         if (!S_ISREG(inode->i_mode)) {
6105                 WARN_ON(1);
6106                 return;
6107         }
6108
6109         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6110         if (ret)
6111                 return;
6112
6113         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6114         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6115
6116         trans = btrfs_start_transaction(root, 0);
6117         BUG_ON(IS_ERR(trans));
6118         btrfs_set_trans_block_group(trans, inode);
6119         trans->block_rsv = root->orphan_block_rsv;
6120
6121         /*
6122          * setattr is responsible for setting the ordered_data_close flag,
6123          * but that is only tested during the last file release.  That
6124          * could happen well after the next commit, leaving a great big
6125          * window where new writes may get lost if someone chooses to write
6126          * to this file after truncating to zero
6127          *
6128          * The inode doesn't have any dirty data here, and so if we commit
6129          * this is a noop.  If someone immediately starts writing to the inode
6130          * it is very likely we'll catch some of their writes in this
6131          * transaction, and the commit will find this file on the ordered
6132          * data list with good things to send down.
6133          *
6134          * This is a best effort solution, there is still a window where
6135          * using truncate to replace the contents of the file will
6136          * end up with a zero length file after a crash.
6137          */
6138         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6139                 btrfs_add_ordered_operation(trans, root, inode);
6140
6141         while (1) {
6142                 if (!trans) {
6143                         trans = btrfs_start_transaction(root, 0);
6144                         BUG_ON(IS_ERR(trans));
6145                         btrfs_set_trans_block_group(trans, inode);
6146                         trans->block_rsv = root->orphan_block_rsv;
6147                 }
6148
6149                 ret = btrfs_block_rsv_check(trans, root,
6150                                             root->orphan_block_rsv, 0, 5);
6151                 if (ret) {
6152                         BUG_ON(ret != -EAGAIN);
6153                         ret = btrfs_commit_transaction(trans, root);
6154                         BUG_ON(ret);
6155                         trans = NULL;
6156                         continue;
6157                 }
6158
6159                 ret = btrfs_truncate_inode_items(trans, root, inode,
6160                                                  inode->i_size,
6161                                                  BTRFS_EXTENT_DATA_KEY);
6162                 if (ret != -EAGAIN)
6163                         break;
6164
6165                 ret = btrfs_update_inode(trans, root, inode);
6166                 BUG_ON(ret);
6167
6168                 nr = trans->blocks_used;
6169                 btrfs_end_transaction(trans, root);
6170                 trans = NULL;
6171                 btrfs_btree_balance_dirty(root, nr);
6172         }
6173
6174         if (ret == 0 && inode->i_nlink > 0) {
6175                 ret = btrfs_orphan_del(trans, inode);
6176                 BUG_ON(ret);
6177         }
6178
6179         ret = btrfs_update_inode(trans, root, inode);
6180         BUG_ON(ret);
6181
6182         nr = trans->blocks_used;
6183         ret = btrfs_end_transaction_throttle(trans, root);
6184         BUG_ON(ret);
6185         btrfs_btree_balance_dirty(root, nr);
6186 }
6187
6188 /*
6189  * create a new subvolume directory/inode (helper for the ioctl).
6190  */
6191 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6192                              struct btrfs_root *new_root,
6193                              u64 new_dirid, u64 alloc_hint)
6194 {
6195         struct inode *inode;
6196         int err;
6197         u64 index = 0;
6198
6199         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6200                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6201         if (IS_ERR(inode))
6202                 return PTR_ERR(inode);
6203         inode->i_op = &btrfs_dir_inode_operations;
6204         inode->i_fop = &btrfs_dir_file_operations;
6205
6206         inode->i_nlink = 1;
6207         btrfs_i_size_write(inode, 0);
6208
6209         err = btrfs_update_inode(trans, new_root, inode);
6210         BUG_ON(err);
6211
6212         iput(inode);
6213         return 0;
6214 }
6215
6216 /* helper function for file defrag and space balancing.  This
6217  * forces readahead on a given range of bytes in an inode
6218  */
6219 unsigned long btrfs_force_ra(struct address_space *mapping,
6220                               struct file_ra_state *ra, struct file *file,
6221                               pgoff_t offset, pgoff_t last_index)
6222 {
6223         pgoff_t req_size = last_index - offset + 1;
6224
6225         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6226         return offset + req_size;
6227 }
6228
6229 struct inode *btrfs_alloc_inode(struct super_block *sb)
6230 {
6231         struct btrfs_inode *ei;
6232         struct inode *inode;
6233
6234         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6235         if (!ei)
6236                 return NULL;
6237
6238         ei->root = NULL;
6239         ei->space_info = NULL;
6240         ei->generation = 0;
6241         ei->sequence = 0;
6242         ei->last_trans = 0;
6243         ei->last_sub_trans = 0;
6244         ei->logged_trans = 0;
6245         ei->delalloc_bytes = 0;
6246         ei->reserved_bytes = 0;
6247         ei->disk_i_size = 0;
6248         ei->flags = 0;
6249         ei->index_cnt = (u64)-1;
6250         ei->last_unlink_trans = 0;
6251
6252         spin_lock_init(&ei->accounting_lock);
6253         atomic_set(&ei->outstanding_extents, 0);
6254         ei->reserved_extents = 0;
6255
6256         ei->ordered_data_close = 0;
6257         ei->orphan_meta_reserved = 0;
6258         ei->dummy_inode = 0;
6259         ei->force_compress = 0;
6260
6261         inode = &ei->vfs_inode;
6262         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6263         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6264         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6265         mutex_init(&ei->log_mutex);
6266         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6267         INIT_LIST_HEAD(&ei->i_orphan);
6268         INIT_LIST_HEAD(&ei->delalloc_inodes);
6269         INIT_LIST_HEAD(&ei->ordered_operations);
6270         RB_CLEAR_NODE(&ei->rb_node);
6271
6272         return inode;
6273 }
6274
6275 void btrfs_destroy_inode(struct inode *inode)
6276 {
6277         struct btrfs_ordered_extent *ordered;
6278         struct btrfs_root *root = BTRFS_I(inode)->root;
6279
6280         WARN_ON(!list_empty(&inode->i_dentry));
6281         WARN_ON(inode->i_data.nrpages);
6282         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6283         WARN_ON(BTRFS_I(inode)->reserved_extents);
6284
6285         /*
6286          * This can happen where we create an inode, but somebody else also
6287          * created the same inode and we need to destroy the one we already
6288          * created.
6289          */
6290         if (!root)
6291                 goto free;
6292
6293         /*
6294          * Make sure we're properly removed from the ordered operation
6295          * lists.
6296          */
6297         smp_mb();
6298         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6299                 spin_lock(&root->fs_info->ordered_extent_lock);
6300                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6301                 spin_unlock(&root->fs_info->ordered_extent_lock);
6302         }
6303
6304         spin_lock(&root->orphan_lock);
6305         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6306                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6307                        inode->i_ino);
6308                 list_del_init(&BTRFS_I(inode)->i_orphan);
6309         }
6310         spin_unlock(&root->orphan_lock);
6311
6312         while (1) {
6313                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6314                 if (!ordered)
6315                         break;
6316                 else {
6317                         printk(KERN_ERR "btrfs found ordered "
6318                                "extent %llu %llu on inode cleanup\n",
6319                                (unsigned long long)ordered->file_offset,
6320                                (unsigned long long)ordered->len);
6321                         btrfs_remove_ordered_extent(inode, ordered);
6322                         btrfs_put_ordered_extent(ordered);
6323                         btrfs_put_ordered_extent(ordered);
6324                 }
6325         }
6326         inode_tree_del(inode);
6327         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6328 free:
6329         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6330 }
6331
6332 void btrfs_drop_inode(struct inode *inode)
6333 {
6334         struct btrfs_root *root = BTRFS_I(inode)->root;
6335         if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
6336                 generic_delete_inode(inode);
6337         else
6338                 generic_drop_inode(inode);
6339 }
6340
6341 static void init_once(void *foo)
6342 {
6343         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6344
6345         inode_init_once(&ei->vfs_inode);
6346 }
6347
6348 void btrfs_destroy_cachep(void)
6349 {
6350         if (btrfs_inode_cachep)
6351                 kmem_cache_destroy(btrfs_inode_cachep);
6352         if (btrfs_trans_handle_cachep)
6353                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6354         if (btrfs_transaction_cachep)
6355                 kmem_cache_destroy(btrfs_transaction_cachep);
6356         if (btrfs_path_cachep)
6357                 kmem_cache_destroy(btrfs_path_cachep);
6358 }
6359
6360 int btrfs_init_cachep(void)
6361 {
6362         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6363                         sizeof(struct btrfs_inode), 0,
6364                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6365         if (!btrfs_inode_cachep)
6366                 goto fail;
6367
6368         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6369                         sizeof(struct btrfs_trans_handle), 0,
6370                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6371         if (!btrfs_trans_handle_cachep)
6372                 goto fail;
6373
6374         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6375                         sizeof(struct btrfs_transaction), 0,
6376                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6377         if (!btrfs_transaction_cachep)
6378                 goto fail;
6379
6380         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6381                         sizeof(struct btrfs_path), 0,
6382                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6383         if (!btrfs_path_cachep)
6384                 goto fail;
6385
6386         return 0;
6387 fail:
6388         btrfs_destroy_cachep();
6389         return -ENOMEM;
6390 }
6391
6392 static int btrfs_getattr(struct vfsmount *mnt,
6393                          struct dentry *dentry, struct kstat *stat)
6394 {
6395         struct inode *inode = dentry->d_inode;
6396         generic_fillattr(inode, stat);
6397         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6398         stat->blksize = PAGE_CACHE_SIZE;
6399         stat->blocks = (inode_get_bytes(inode) +
6400                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6401         return 0;
6402 }
6403
6404 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6405                            struct inode *new_dir, struct dentry *new_dentry)
6406 {
6407         struct btrfs_trans_handle *trans;
6408         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6409         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6410         struct inode *new_inode = new_dentry->d_inode;
6411         struct inode *old_inode = old_dentry->d_inode;
6412         struct timespec ctime = CURRENT_TIME;
6413         u64 index = 0;
6414         u64 root_objectid;
6415         int ret;
6416
6417         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6418                 return -EPERM;
6419
6420         /* we only allow rename subvolume link between subvolumes */
6421         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6422                 return -EXDEV;
6423
6424         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6425             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6426                 return -ENOTEMPTY;
6427
6428         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6429             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6430                 return -ENOTEMPTY;
6431         /*
6432          * we're using rename to replace one file with another.
6433          * and the replacement file is large.  Start IO on it now so
6434          * we don't add too much work to the end of the transaction
6435          */
6436         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6437             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6438                 filemap_flush(old_inode->i_mapping);
6439
6440         /* close the racy window with snapshot create/destroy ioctl */
6441         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6442                 down_read(&root->fs_info->subvol_sem);
6443         /*
6444          * We want to reserve the absolute worst case amount of items.  So if
6445          * both inodes are subvols and we need to unlink them then that would
6446          * require 4 item modifications, but if they are both normal inodes it
6447          * would require 5 item modifications, so we'll assume their normal
6448          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6449          * should cover the worst case number of items we'll modify.
6450          */
6451         trans = btrfs_start_transaction(root, 20);
6452         if (IS_ERR(trans))
6453                 return PTR_ERR(trans);
6454
6455         btrfs_set_trans_block_group(trans, new_dir);
6456
6457         if (dest != root)
6458                 btrfs_record_root_in_trans(trans, dest);
6459
6460         ret = btrfs_set_inode_index(new_dir, &index);
6461         if (ret)
6462                 goto out_fail;
6463
6464         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6465                 /* force full log commit if subvolume involved. */
6466                 root->fs_info->last_trans_log_full_commit = trans->transid;
6467         } else {
6468                 ret = btrfs_insert_inode_ref(trans, dest,
6469                                              new_dentry->d_name.name,
6470                                              new_dentry->d_name.len,
6471                                              old_inode->i_ino,
6472                                              new_dir->i_ino, index);
6473                 if (ret)
6474                         goto out_fail;
6475                 /*
6476                  * this is an ugly little race, but the rename is required
6477                  * to make sure that if we crash, the inode is either at the
6478                  * old name or the new one.  pinning the log transaction lets
6479                  * us make sure we don't allow a log commit to come in after
6480                  * we unlink the name but before we add the new name back in.
6481                  */
6482                 btrfs_pin_log_trans(root);
6483         }
6484         /*
6485          * make sure the inode gets flushed if it is replacing
6486          * something.
6487          */
6488         if (new_inode && new_inode->i_size &&
6489             old_inode && S_ISREG(old_inode->i_mode)) {
6490                 btrfs_add_ordered_operation(trans, root, old_inode);
6491         }
6492
6493         old_dir->i_ctime = old_dir->i_mtime = ctime;
6494         new_dir->i_ctime = new_dir->i_mtime = ctime;
6495         old_inode->i_ctime = ctime;
6496
6497         if (old_dentry->d_parent != new_dentry->d_parent)
6498                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6499
6500         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6501                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6502                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6503                                         old_dentry->d_name.name,
6504                                         old_dentry->d_name.len);
6505         } else {
6506                 btrfs_inc_nlink(old_dentry->d_inode);
6507                 ret = btrfs_unlink_inode(trans, root, old_dir,
6508                                          old_dentry->d_inode,
6509                                          old_dentry->d_name.name,
6510                                          old_dentry->d_name.len);
6511         }
6512         BUG_ON(ret);
6513
6514         if (new_inode) {
6515                 new_inode->i_ctime = CURRENT_TIME;
6516                 if (unlikely(new_inode->i_ino ==
6517                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6518                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6519                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6520                                                 root_objectid,
6521                                                 new_dentry->d_name.name,
6522                                                 new_dentry->d_name.len);
6523                         BUG_ON(new_inode->i_nlink == 0);
6524                 } else {
6525                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6526                                                  new_dentry->d_inode,
6527                                                  new_dentry->d_name.name,
6528                                                  new_dentry->d_name.len);
6529                 }
6530                 BUG_ON(ret);
6531                 if (new_inode->i_nlink == 0) {
6532                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6533                         BUG_ON(ret);
6534                 }
6535         }
6536
6537         ret = btrfs_add_link(trans, new_dir, old_inode,
6538                              new_dentry->d_name.name,
6539                              new_dentry->d_name.len, 0, index);
6540         BUG_ON(ret);
6541
6542         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6543                 btrfs_log_new_name(trans, old_inode, old_dir,
6544                                    new_dentry->d_parent);
6545                 btrfs_end_log_trans(root);
6546         }
6547 out_fail:
6548         btrfs_end_transaction_throttle(trans, root);
6549
6550         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6551                 up_read(&root->fs_info->subvol_sem);
6552
6553         return ret;
6554 }
6555
6556 /*
6557  * some fairly slow code that needs optimization. This walks the list
6558  * of all the inodes with pending delalloc and forces them to disk.
6559  */
6560 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6561 {
6562         struct list_head *head = &root->fs_info->delalloc_inodes;
6563         struct btrfs_inode *binode;
6564         struct inode *inode;
6565
6566         if (root->fs_info->sb->s_flags & MS_RDONLY)
6567                 return -EROFS;
6568
6569         spin_lock(&root->fs_info->delalloc_lock);
6570         while (!list_empty(head)) {
6571                 binode = list_entry(head->next, struct btrfs_inode,
6572                                     delalloc_inodes);
6573                 inode = igrab(&binode->vfs_inode);
6574                 if (!inode)
6575                         list_del_init(&binode->delalloc_inodes);
6576                 spin_unlock(&root->fs_info->delalloc_lock);
6577                 if (inode) {
6578                         filemap_flush(inode->i_mapping);
6579                         if (delay_iput)
6580                                 btrfs_add_delayed_iput(inode);
6581                         else
6582                                 iput(inode);
6583                 }
6584                 cond_resched();
6585                 spin_lock(&root->fs_info->delalloc_lock);
6586         }
6587         spin_unlock(&root->fs_info->delalloc_lock);
6588
6589         /* the filemap_flush will queue IO into the worker threads, but
6590          * we have to make sure the IO is actually started and that
6591          * ordered extents get created before we return
6592          */
6593         atomic_inc(&root->fs_info->async_submit_draining);
6594         while (atomic_read(&root->fs_info->nr_async_submits) ||
6595               atomic_read(&root->fs_info->async_delalloc_pages)) {
6596                 wait_event(root->fs_info->async_submit_wait,
6597                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
6598                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
6599         }
6600         atomic_dec(&root->fs_info->async_submit_draining);
6601         return 0;
6602 }
6603
6604 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput)
6605 {
6606         struct btrfs_inode *binode;
6607         struct inode *inode = NULL;
6608
6609         spin_lock(&root->fs_info->delalloc_lock);
6610         while (!list_empty(&root->fs_info->delalloc_inodes)) {
6611                 binode = list_entry(root->fs_info->delalloc_inodes.next,
6612                                     struct btrfs_inode, delalloc_inodes);
6613                 inode = igrab(&binode->vfs_inode);
6614                 if (inode) {
6615                         list_move_tail(&binode->delalloc_inodes,
6616                                        &root->fs_info->delalloc_inodes);
6617                         break;
6618                 }
6619
6620                 list_del_init(&binode->delalloc_inodes);
6621                 cond_resched_lock(&root->fs_info->delalloc_lock);
6622         }
6623         spin_unlock(&root->fs_info->delalloc_lock);
6624
6625         if (inode) {
6626                 write_inode_now(inode, 0);
6627                 if (delay_iput)
6628                         btrfs_add_delayed_iput(inode);
6629                 else
6630                         iput(inode);
6631                 return 1;
6632         }
6633         return 0;
6634 }
6635
6636 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
6637                          const char *symname)
6638 {
6639         struct btrfs_trans_handle *trans;
6640         struct btrfs_root *root = BTRFS_I(dir)->root;
6641         struct btrfs_path *path;
6642         struct btrfs_key key;
6643         struct inode *inode = NULL;
6644         int err;
6645         int drop_inode = 0;
6646         u64 objectid;
6647         u64 index = 0 ;
6648         int name_len;
6649         int datasize;
6650         unsigned long ptr;
6651         struct btrfs_file_extent_item *ei;
6652         struct extent_buffer *leaf;
6653         unsigned long nr = 0;
6654
6655         name_len = strlen(symname) + 1;
6656         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
6657                 return -ENAMETOOLONG;
6658
6659         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
6660         if (err)
6661                 return err;
6662         /*
6663          * 2 items for inode item and ref
6664          * 2 items for dir items
6665          * 1 item for xattr if selinux is on
6666          */
6667         trans = btrfs_start_transaction(root, 5);
6668         if (IS_ERR(trans))
6669                 return PTR_ERR(trans);
6670
6671         btrfs_set_trans_block_group(trans, dir);
6672
6673         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6674                                 dentry->d_name.len,
6675                                 dentry->d_parent->d_inode->i_ino, objectid,
6676                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
6677                                 &index);
6678         err = PTR_ERR(inode);
6679         if (IS_ERR(inode))
6680                 goto out_unlock;
6681
6682         err = btrfs_init_inode_security(trans, inode, dir);
6683         if (err) {
6684                 drop_inode = 1;
6685                 goto out_unlock;
6686         }
6687
6688         btrfs_set_trans_block_group(trans, inode);
6689         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
6690         if (err)
6691                 drop_inode = 1;
6692         else {
6693                 inode->i_mapping->a_ops = &btrfs_aops;
6694                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6695                 inode->i_fop = &btrfs_file_operations;
6696                 inode->i_op = &btrfs_file_inode_operations;
6697                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6698         }
6699         btrfs_update_inode_block_group(trans, inode);
6700         btrfs_update_inode_block_group(trans, dir);
6701         if (drop_inode)
6702                 goto out_unlock;
6703
6704         path = btrfs_alloc_path();
6705         BUG_ON(!path);
6706         key.objectid = inode->i_ino;
6707         key.offset = 0;
6708         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
6709         datasize = btrfs_file_extent_calc_inline_size(name_len);
6710         err = btrfs_insert_empty_item(trans, root, path, &key,
6711                                       datasize);
6712         if (err) {
6713                 drop_inode = 1;
6714                 goto out_unlock;
6715         }
6716         leaf = path->nodes[0];
6717         ei = btrfs_item_ptr(leaf, path->slots[0],
6718                             struct btrfs_file_extent_item);
6719         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
6720         btrfs_set_file_extent_type(leaf, ei,
6721                                    BTRFS_FILE_EXTENT_INLINE);
6722         btrfs_set_file_extent_encryption(leaf, ei, 0);
6723         btrfs_set_file_extent_compression(leaf, ei, 0);
6724         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
6725         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
6726
6727         ptr = btrfs_file_extent_inline_start(ei);
6728         write_extent_buffer(leaf, symname, ptr, name_len);
6729         btrfs_mark_buffer_dirty(leaf);
6730         btrfs_free_path(path);
6731
6732         inode->i_op = &btrfs_symlink_inode_operations;
6733         inode->i_mapping->a_ops = &btrfs_symlink_aops;
6734         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6735         inode_set_bytes(inode, name_len);
6736         btrfs_i_size_write(inode, name_len - 1);
6737         err = btrfs_update_inode(trans, root, inode);
6738         if (err)
6739                 drop_inode = 1;
6740
6741 out_unlock:
6742         nr = trans->blocks_used;
6743         btrfs_end_transaction_throttle(trans, root);
6744         if (drop_inode) {
6745                 inode_dec_link_count(inode);
6746                 iput(inode);
6747         }
6748         btrfs_btree_balance_dirty(root, nr);
6749         return err;
6750 }
6751
6752 int btrfs_prealloc_file_range(struct inode *inode, int mode,
6753                               u64 start, u64 num_bytes, u64 min_size,
6754                               loff_t actual_len, u64 *alloc_hint)
6755 {
6756         struct btrfs_trans_handle *trans;
6757         struct btrfs_root *root = BTRFS_I(inode)->root;
6758         struct btrfs_key ins;
6759         u64 cur_offset = start;
6760         int ret = 0;
6761
6762         while (num_bytes > 0) {
6763                 trans = btrfs_start_transaction(root, 3);
6764                 if (IS_ERR(trans)) {
6765                         ret = PTR_ERR(trans);
6766                         break;
6767                 }
6768
6769                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
6770                                            0, *alloc_hint, (u64)-1, &ins, 1);
6771                 if (ret) {
6772                         btrfs_end_transaction(trans, root);
6773                         break;
6774                 }
6775
6776                 ret = insert_reserved_file_extent(trans, inode,
6777                                                   cur_offset, ins.objectid,
6778                                                   ins.offset, ins.offset,
6779                                                   ins.offset, 0, 0, 0,
6780                                                   BTRFS_FILE_EXTENT_PREALLOC);
6781                 BUG_ON(ret);
6782                 btrfs_drop_extent_cache(inode, cur_offset,
6783                                         cur_offset + ins.offset -1, 0);
6784
6785                 num_bytes -= ins.offset;
6786                 cur_offset += ins.offset;
6787                 *alloc_hint = ins.objectid + ins.offset;
6788
6789                 inode->i_ctime = CURRENT_TIME;
6790                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
6791                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
6792                     (actual_len > inode->i_size) &&
6793                     (cur_offset > inode->i_size)) {
6794                         if (cur_offset > actual_len)
6795                                 i_size_write(inode, actual_len);
6796                         else
6797                                 i_size_write(inode, cur_offset);
6798                         i_size_write(inode, cur_offset);
6799                         btrfs_ordered_update_i_size(inode, cur_offset, NULL);
6800                 }
6801
6802                 ret = btrfs_update_inode(trans, root, inode);
6803                 BUG_ON(ret);
6804
6805                 btrfs_end_transaction(trans, root);
6806         }
6807         return ret;
6808 }
6809
6810 static long btrfs_fallocate(struct inode *inode, int mode,
6811                             loff_t offset, loff_t len)
6812 {
6813         struct extent_state *cached_state = NULL;
6814         u64 cur_offset;
6815         u64 last_byte;
6816         u64 alloc_start;
6817         u64 alloc_end;
6818         u64 alloc_hint = 0;
6819         u64 locked_end;
6820         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
6821         struct extent_map *em;
6822         int ret;
6823
6824         alloc_start = offset & ~mask;
6825         alloc_end =  (offset + len + mask) & ~mask;
6826
6827         /*
6828          * wait for ordered IO before we have any locks.  We'll loop again
6829          * below with the locks held.
6830          */
6831         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
6832
6833         mutex_lock(&inode->i_mutex);
6834         if (alloc_start > inode->i_size) {
6835                 ret = btrfs_cont_expand(inode, alloc_start);
6836                 if (ret)
6837                         goto out;
6838         }
6839
6840         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
6841         if (ret)
6842                 goto out;
6843
6844         locked_end = alloc_end - 1;
6845         while (1) {
6846                 struct btrfs_ordered_extent *ordered;
6847
6848                 /* the extent lock is ordered inside the running
6849                  * transaction
6850                  */
6851                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
6852                                  locked_end, 0, &cached_state, GFP_NOFS);
6853                 ordered = btrfs_lookup_first_ordered_extent(inode,
6854                                                             alloc_end - 1);
6855                 if (ordered &&
6856                     ordered->file_offset + ordered->len > alloc_start &&
6857                     ordered->file_offset < alloc_end) {
6858                         btrfs_put_ordered_extent(ordered);
6859                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
6860                                              alloc_start, locked_end,
6861                                              &cached_state, GFP_NOFS);
6862                         /*
6863                          * we can't wait on the range with the transaction
6864                          * running or with the extent lock held
6865                          */
6866                         btrfs_wait_ordered_range(inode, alloc_start,
6867                                                  alloc_end - alloc_start);
6868                 } else {
6869                         if (ordered)
6870                                 btrfs_put_ordered_extent(ordered);
6871                         break;
6872                 }
6873         }
6874
6875         cur_offset = alloc_start;
6876         while (1) {
6877                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
6878                                       alloc_end - cur_offset, 0);
6879                 BUG_ON(IS_ERR(em) || !em);
6880                 last_byte = min(extent_map_end(em), alloc_end);
6881                 last_byte = (last_byte + mask) & ~mask;
6882                 if (em->block_start == EXTENT_MAP_HOLE ||
6883                     (cur_offset >= inode->i_size &&
6884                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6885                         ret = btrfs_prealloc_file_range(inode, 0, cur_offset,
6886                                                         last_byte - cur_offset,
6887                                                         1 << inode->i_blkbits,
6888                                                         offset + len,
6889                                                         &alloc_hint);
6890                         if (ret < 0) {
6891                                 free_extent_map(em);
6892                                 break;
6893                         }
6894                 }
6895                 free_extent_map(em);
6896
6897                 cur_offset = last_byte;
6898                 if (cur_offset >= alloc_end) {
6899                         ret = 0;
6900                         break;
6901                 }
6902         }
6903         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
6904                              &cached_state, GFP_NOFS);
6905
6906         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
6907 out:
6908         mutex_unlock(&inode->i_mutex);
6909         return ret;
6910 }
6911
6912 static int btrfs_set_page_dirty(struct page *page)
6913 {
6914         return __set_page_dirty_nobuffers(page);
6915 }
6916
6917 static int btrfs_permission(struct inode *inode, int mask)
6918 {
6919         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
6920                 return -EACCES;
6921         return generic_permission(inode, mask, btrfs_check_acl);
6922 }
6923
6924 static const struct inode_operations btrfs_dir_inode_operations = {
6925         .getattr        = btrfs_getattr,
6926         .lookup         = btrfs_lookup,
6927         .create         = btrfs_create,
6928         .unlink         = btrfs_unlink,
6929         .link           = btrfs_link,
6930         .mkdir          = btrfs_mkdir,
6931         .rmdir          = btrfs_rmdir,
6932         .rename         = btrfs_rename,
6933         .symlink        = btrfs_symlink,
6934         .setattr        = btrfs_setattr,
6935         .mknod          = btrfs_mknod,
6936         .setxattr       = btrfs_setxattr,
6937         .getxattr       = btrfs_getxattr,
6938         .listxattr      = btrfs_listxattr,
6939         .removexattr    = btrfs_removexattr,
6940         .permission     = btrfs_permission,
6941 };
6942 static const struct inode_operations btrfs_dir_ro_inode_operations = {
6943         .lookup         = btrfs_lookup,
6944         .permission     = btrfs_permission,
6945 };
6946
6947 static const struct file_operations btrfs_dir_file_operations = {
6948         .llseek         = generic_file_llseek,
6949         .read           = generic_read_dir,
6950         .readdir        = btrfs_real_readdir,
6951         .unlocked_ioctl = btrfs_ioctl,
6952 #ifdef CONFIG_COMPAT
6953         .compat_ioctl   = btrfs_ioctl,
6954 #endif
6955         .release        = btrfs_release_file,
6956         .fsync          = btrfs_sync_file,
6957 };
6958
6959 static struct extent_io_ops btrfs_extent_io_ops = {
6960         .fill_delalloc = run_delalloc_range,
6961         .submit_bio_hook = btrfs_submit_bio_hook,
6962         .merge_bio_hook = btrfs_merge_bio_hook,
6963         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6964         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6965         .writepage_start_hook = btrfs_writepage_start_hook,
6966         .readpage_io_failed_hook = btrfs_io_failed_hook,
6967         .set_bit_hook = btrfs_set_bit_hook,
6968         .clear_bit_hook = btrfs_clear_bit_hook,
6969         .merge_extent_hook = btrfs_merge_extent_hook,
6970         .split_extent_hook = btrfs_split_extent_hook,
6971 };
6972
6973 /*
6974  * btrfs doesn't support the bmap operation because swapfiles
6975  * use bmap to make a mapping of extents in the file.  They assume
6976  * these extents won't change over the life of the file and they
6977  * use the bmap result to do IO directly to the drive.
6978  *
6979  * the btrfs bmap call would return logical addresses that aren't
6980  * suitable for IO and they also will change frequently as COW
6981  * operations happen.  So, swapfile + btrfs == corruption.
6982  *
6983  * For now we're avoiding this by dropping bmap.
6984  */
6985 static const struct address_space_operations btrfs_aops = {
6986         .readpage       = btrfs_readpage,
6987         .writepage      = btrfs_writepage,
6988         .writepages     = btrfs_writepages,
6989         .readpages      = btrfs_readpages,
6990         .sync_page      = block_sync_page,
6991         .direct_IO      = btrfs_direct_IO,
6992         .invalidatepage = btrfs_invalidatepage,
6993         .releasepage    = btrfs_releasepage,
6994         .set_page_dirty = btrfs_set_page_dirty,
6995         .error_remove_page = generic_error_remove_page,
6996 };
6997
6998 static const struct address_space_operations btrfs_symlink_aops = {
6999         .readpage       = btrfs_readpage,
7000         .writepage      = btrfs_writepage,
7001         .invalidatepage = btrfs_invalidatepage,
7002         .releasepage    = btrfs_releasepage,
7003 };
7004
7005 static const struct inode_operations btrfs_file_inode_operations = {
7006         .truncate       = btrfs_truncate,
7007         .getattr        = btrfs_getattr,
7008         .setattr        = btrfs_setattr,
7009         .setxattr       = btrfs_setxattr,
7010         .getxattr       = btrfs_getxattr,
7011         .listxattr      = btrfs_listxattr,
7012         .removexattr    = btrfs_removexattr,
7013         .permission     = btrfs_permission,
7014         .fallocate      = btrfs_fallocate,
7015         .fiemap         = btrfs_fiemap,
7016 };
7017 static const struct inode_operations btrfs_special_inode_operations = {
7018         .getattr        = btrfs_getattr,
7019         .setattr        = btrfs_setattr,
7020         .permission     = btrfs_permission,
7021         .setxattr       = btrfs_setxattr,
7022         .getxattr       = btrfs_getxattr,
7023         .listxattr      = btrfs_listxattr,
7024         .removexattr    = btrfs_removexattr,
7025 };
7026 static const struct inode_operations btrfs_symlink_inode_operations = {
7027         .readlink       = generic_readlink,
7028         .follow_link    = page_follow_link_light,
7029         .put_link       = page_put_link,
7030         .permission     = btrfs_permission,
7031         .setxattr       = btrfs_setxattr,
7032         .getxattr       = btrfs_getxattr,
7033         .listxattr      = btrfs_listxattr,
7034         .removexattr    = btrfs_removexattr,
7035 };
7036
7037 const struct dentry_operations btrfs_dentry_operations = {
7038         .d_delete       = btrfs_dentry_delete,
7039 };