Merge branch 'hwpoison' of git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux...
[sfrench/cifs-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/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 "compat.h"
40 #include "ctree.h"
41 #include "disk-io.h"
42 #include "transaction.h"
43 #include "btrfs_inode.h"
44 #include "ioctl.h"
45 #include "print-tree.h"
46 #include "volumes.h"
47 #include "ordered-data.h"
48 #include "xattr.h"
49 #include "tree-log.h"
50 #include "compression.h"
51 #include "locking.h"
52
53 struct btrfs_iget_args {
54         u64 ino;
55         struct btrfs_root *root;
56 };
57
58 static const struct inode_operations btrfs_dir_inode_operations;
59 static const struct inode_operations btrfs_symlink_inode_operations;
60 static const struct inode_operations btrfs_dir_ro_inode_operations;
61 static const struct inode_operations btrfs_special_inode_operations;
62 static const struct inode_operations btrfs_file_inode_operations;
63 static const struct address_space_operations btrfs_aops;
64 static const struct address_space_operations btrfs_symlink_aops;
65 static struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
67
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
72
73 #define S_SHIFT 12
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
76         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
77         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
78         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
79         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
80         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
81         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
82 };
83
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87                                    struct page *locked_page,
88                                    u64 start, u64 end, int *page_started,
89                                    unsigned long *nr_written, int unlock);
90
91 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
92 {
93         int err;
94
95         err = btrfs_init_acl(inode, dir);
96         if (!err)
97                 err = btrfs_xattr_security_init(inode, dir);
98         return err;
99 }
100
101 /*
102  * this does all the hard work for inserting an inline extent into
103  * the btree.  The caller should have done a btrfs_drop_extents so that
104  * no overlapping inline items exist in the btree
105  */
106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
107                                 struct btrfs_root *root, struct inode *inode,
108                                 u64 start, size_t size, size_t compressed_size,
109                                 struct page **compressed_pages)
110 {
111         struct btrfs_key key;
112         struct btrfs_path *path;
113         struct extent_buffer *leaf;
114         struct page *page = NULL;
115         char *kaddr;
116         unsigned long ptr;
117         struct btrfs_file_extent_item *ei;
118         int err = 0;
119         int ret;
120         size_t cur_size = size;
121         size_t datasize;
122         unsigned long offset;
123         int use_compress = 0;
124
125         if (compressed_size && compressed_pages) {
126                 use_compress = 1;
127                 cur_size = compressed_size;
128         }
129
130         path = btrfs_alloc_path();
131         if (!path)
132                 return -ENOMEM;
133
134         path->leave_spinning = 1;
135         btrfs_set_trans_block_group(trans, inode);
136
137         key.objectid = inode->i_ino;
138         key.offset = start;
139         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
140         datasize = btrfs_file_extent_calc_inline_size(cur_size);
141
142         inode_add_bytes(inode, size);
143         ret = btrfs_insert_empty_item(trans, root, path, &key,
144                                       datasize);
145         BUG_ON(ret);
146         if (ret) {
147                 err = ret;
148                 goto fail;
149         }
150         leaf = path->nodes[0];
151         ei = btrfs_item_ptr(leaf, path->slots[0],
152                             struct btrfs_file_extent_item);
153         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
154         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
155         btrfs_set_file_extent_encryption(leaf, ei, 0);
156         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
157         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
158         ptr = btrfs_file_extent_inline_start(ei);
159
160         if (use_compress) {
161                 struct page *cpage;
162                 int i = 0;
163                 while (compressed_size > 0) {
164                         cpage = compressed_pages[i];
165                         cur_size = min_t(unsigned long, compressed_size,
166                                        PAGE_CACHE_SIZE);
167
168                         kaddr = kmap_atomic(cpage, KM_USER0);
169                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
170                         kunmap_atomic(kaddr, KM_USER0);
171
172                         i++;
173                         ptr += cur_size;
174                         compressed_size -= cur_size;
175                 }
176                 btrfs_set_file_extent_compression(leaf, ei,
177                                                   BTRFS_COMPRESS_ZLIB);
178         } else {
179                 page = find_get_page(inode->i_mapping,
180                                      start >> PAGE_CACHE_SHIFT);
181                 btrfs_set_file_extent_compression(leaf, ei, 0);
182                 kaddr = kmap_atomic(page, KM_USER0);
183                 offset = start & (PAGE_CACHE_SIZE - 1);
184                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
185                 kunmap_atomic(kaddr, KM_USER0);
186                 page_cache_release(page);
187         }
188         btrfs_mark_buffer_dirty(leaf);
189         btrfs_free_path(path);
190
191         BTRFS_I(inode)->disk_i_size = inode->i_size;
192         btrfs_update_inode(trans, root, inode);
193         return 0;
194 fail:
195         btrfs_free_path(path);
196         return err;
197 }
198
199
200 /*
201  * conditionally insert an inline extent into the file.  This
202  * does the checks required to make sure the data is small enough
203  * to fit as an inline extent.
204  */
205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
206                                  struct btrfs_root *root,
207                                  struct inode *inode, u64 start, u64 end,
208                                  size_t compressed_size,
209                                  struct page **compressed_pages)
210 {
211         u64 isize = i_size_read(inode);
212         u64 actual_end = min(end + 1, isize);
213         u64 inline_len = actual_end - start;
214         u64 aligned_end = (end + root->sectorsize - 1) &
215                         ~((u64)root->sectorsize - 1);
216         u64 hint_byte;
217         u64 data_len = inline_len;
218         int ret;
219
220         if (compressed_size)
221                 data_len = compressed_size;
222
223         if (start > 0 ||
224             actual_end >= PAGE_CACHE_SIZE ||
225             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
226             (!compressed_size &&
227             (actual_end & (root->sectorsize - 1)) == 0) ||
228             end + 1 < isize ||
229             data_len > root->fs_info->max_inline) {
230                 return 1;
231         }
232
233         ret = btrfs_drop_extents(trans, root, inode, start,
234                                  aligned_end, aligned_end, start, &hint_byte);
235         BUG_ON(ret);
236
237         if (isize > actual_end)
238                 inline_len = min_t(u64, isize, actual_end);
239         ret = insert_inline_extent(trans, root, inode, start,
240                                    inline_len, compressed_size,
241                                    compressed_pages);
242         BUG_ON(ret);
243         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
244         return 0;
245 }
246
247 struct async_extent {
248         u64 start;
249         u64 ram_size;
250         u64 compressed_size;
251         struct page **pages;
252         unsigned long nr_pages;
253         struct list_head list;
254 };
255
256 struct async_cow {
257         struct inode *inode;
258         struct btrfs_root *root;
259         struct page *locked_page;
260         u64 start;
261         u64 end;
262         struct list_head extents;
263         struct btrfs_work work;
264 };
265
266 static noinline int add_async_extent(struct async_cow *cow,
267                                      u64 start, u64 ram_size,
268                                      u64 compressed_size,
269                                      struct page **pages,
270                                      unsigned long nr_pages)
271 {
272         struct async_extent *async_extent;
273
274         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
275         async_extent->start = start;
276         async_extent->ram_size = ram_size;
277         async_extent->compressed_size = compressed_size;
278         async_extent->pages = pages;
279         async_extent->nr_pages = nr_pages;
280         list_add_tail(&async_extent->list, &cow->extents);
281         return 0;
282 }
283
284 /*
285  * we create compressed extents in two phases.  The first
286  * phase compresses a range of pages that have already been
287  * locked (both pages and state bits are locked).
288  *
289  * This is done inside an ordered work queue, and the compression
290  * is spread across many cpus.  The actual IO submission is step
291  * two, and the ordered work queue takes care of making sure that
292  * happens in the same order things were put onto the queue by
293  * writepages and friends.
294  *
295  * If this code finds it can't get good compression, it puts an
296  * entry onto the work queue to write the uncompressed bytes.  This
297  * makes sure that both compressed inodes and uncompressed inodes
298  * are written in the same order that pdflush sent them down.
299  */
300 static noinline int compress_file_range(struct inode *inode,
301                                         struct page *locked_page,
302                                         u64 start, u64 end,
303                                         struct async_cow *async_cow,
304                                         int *num_added)
305 {
306         struct btrfs_root *root = BTRFS_I(inode)->root;
307         struct btrfs_trans_handle *trans;
308         u64 num_bytes;
309         u64 orig_start;
310         u64 disk_num_bytes;
311         u64 blocksize = root->sectorsize;
312         u64 actual_end;
313         u64 isize = i_size_read(inode);
314         int ret = 0;
315         struct page **pages = NULL;
316         unsigned long nr_pages;
317         unsigned long nr_pages_ret = 0;
318         unsigned long total_compressed = 0;
319         unsigned long total_in = 0;
320         unsigned long max_compressed = 128 * 1024;
321         unsigned long max_uncompressed = 128 * 1024;
322         int i;
323         int will_compress;
324
325         orig_start = start;
326
327         actual_end = min_t(u64, isize, end + 1);
328 again:
329         will_compress = 0;
330         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
331         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
332
333         /*
334          * we don't want to send crud past the end of i_size through
335          * compression, that's just a waste of CPU time.  So, if the
336          * end of the file is before the start of our current
337          * requested range of bytes, we bail out to the uncompressed
338          * cleanup code that can deal with all of this.
339          *
340          * It isn't really the fastest way to fix things, but this is a
341          * very uncommon corner.
342          */
343         if (actual_end <= start)
344                 goto cleanup_and_bail_uncompressed;
345
346         total_compressed = actual_end - start;
347
348         /* we want to make sure that amount of ram required to uncompress
349          * an extent is reasonable, so we limit the total size in ram
350          * of a compressed extent to 128k.  This is a crucial number
351          * because it also controls how easily we can spread reads across
352          * cpus for decompression.
353          *
354          * We also want to make sure the amount of IO required to do
355          * a random read is reasonably small, so we limit the size of
356          * a compressed extent to 128k.
357          */
358         total_compressed = min(total_compressed, max_uncompressed);
359         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
360         num_bytes = max(blocksize,  num_bytes);
361         disk_num_bytes = num_bytes;
362         total_in = 0;
363         ret = 0;
364
365         /*
366          * we do compression for mount -o compress and when the
367          * inode has not been flagged as nocompress.  This flag can
368          * change at any time if we discover bad compression ratios.
369          */
370         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
371             btrfs_test_opt(root, COMPRESS)) {
372                 WARN_ON(pages);
373                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
374
375                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
376                                                 total_compressed, pages,
377                                                 nr_pages, &nr_pages_ret,
378                                                 &total_in,
379                                                 &total_compressed,
380                                                 max_compressed);
381
382                 if (!ret) {
383                         unsigned long offset = total_compressed &
384                                 (PAGE_CACHE_SIZE - 1);
385                         struct page *page = pages[nr_pages_ret - 1];
386                         char *kaddr;
387
388                         /* zero the tail end of the last page, we might be
389                          * sending it down to disk
390                          */
391                         if (offset) {
392                                 kaddr = kmap_atomic(page, KM_USER0);
393                                 memset(kaddr + offset, 0,
394                                        PAGE_CACHE_SIZE - offset);
395                                 kunmap_atomic(kaddr, KM_USER0);
396                         }
397                         will_compress = 1;
398                 }
399         }
400         if (start == 0) {
401                 trans = btrfs_join_transaction(root, 1);
402                 BUG_ON(!trans);
403                 btrfs_set_trans_block_group(trans, inode);
404
405                 /* lets try to make an inline extent */
406                 if (ret || total_in < (actual_end - start)) {
407                         /* we didn't compress the entire range, try
408                          * to make an uncompressed inline extent.
409                          */
410                         ret = cow_file_range_inline(trans, root, inode,
411                                                     start, end, 0, NULL);
412                 } else {
413                         /* try making a compressed inline extent */
414                         ret = cow_file_range_inline(trans, root, inode,
415                                                     start, end,
416                                                     total_compressed, pages);
417                 }
418                 btrfs_end_transaction(trans, root);
419                 if (ret == 0) {
420                         /*
421                          * inline extent creation worked, we don't need
422                          * to create any more async work items.  Unlock
423                          * and free up our temp pages.
424                          */
425                         extent_clear_unlock_delalloc(inode,
426                                                      &BTRFS_I(inode)->io_tree,
427                                                      start, end, NULL, 1, 0,
428                                                      0, 1, 1, 1);
429                         ret = 0;
430                         goto free_pages_out;
431                 }
432         }
433
434         if (will_compress) {
435                 /*
436                  * we aren't doing an inline extent round the compressed size
437                  * up to a block size boundary so the allocator does sane
438                  * things
439                  */
440                 total_compressed = (total_compressed + blocksize - 1) &
441                         ~(blocksize - 1);
442
443                 /*
444                  * one last check to make sure the compression is really a
445                  * win, compare the page count read with the blocks on disk
446                  */
447                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
448                         ~(PAGE_CACHE_SIZE - 1);
449                 if (total_compressed >= total_in) {
450                         will_compress = 0;
451                 } else {
452                         disk_num_bytes = total_compressed;
453                         num_bytes = total_in;
454                 }
455         }
456         if (!will_compress && pages) {
457                 /*
458                  * the compression code ran but failed to make things smaller,
459                  * free any pages it allocated and our page pointer array
460                  */
461                 for (i = 0; i < nr_pages_ret; i++) {
462                         WARN_ON(pages[i]->mapping);
463                         page_cache_release(pages[i]);
464                 }
465                 kfree(pages);
466                 pages = NULL;
467                 total_compressed = 0;
468                 nr_pages_ret = 0;
469
470                 /* flag the file so we don't compress in the future */
471                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
472         }
473         if (will_compress) {
474                 *num_added += 1;
475
476                 /* the async work queues will take care of doing actual
477                  * allocation on disk for these compressed pages,
478                  * and will submit them to the elevator.
479                  */
480                 add_async_extent(async_cow, start, num_bytes,
481                                  total_compressed, pages, nr_pages_ret);
482
483                 if (start + num_bytes < end && start + num_bytes < actual_end) {
484                         start += num_bytes;
485                         pages = NULL;
486                         cond_resched();
487                         goto again;
488                 }
489         } else {
490 cleanup_and_bail_uncompressed:
491                 /*
492                  * No compression, but we still need to write the pages in
493                  * the file we've been given so far.  redirty the locked
494                  * page if it corresponds to our extent and set things up
495                  * for the async work queue to run cow_file_range to do
496                  * the normal delalloc dance
497                  */
498                 if (page_offset(locked_page) >= start &&
499                     page_offset(locked_page) <= end) {
500                         __set_page_dirty_nobuffers(locked_page);
501                         /* unlocked later on in the async handlers */
502                 }
503                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
504                 *num_added += 1;
505         }
506
507 out:
508         return 0;
509
510 free_pages_out:
511         for (i = 0; i < nr_pages_ret; i++) {
512                 WARN_ON(pages[i]->mapping);
513                 page_cache_release(pages[i]);
514         }
515         kfree(pages);
516
517         goto out;
518 }
519
520 /*
521  * phase two of compressed writeback.  This is the ordered portion
522  * of the code, which only gets called in the order the work was
523  * queued.  We walk all the async extents created by compress_file_range
524  * and send them down to the disk.
525  */
526 static noinline int submit_compressed_extents(struct inode *inode,
527                                               struct async_cow *async_cow)
528 {
529         struct async_extent *async_extent;
530         u64 alloc_hint = 0;
531         struct btrfs_trans_handle *trans;
532         struct btrfs_key ins;
533         struct extent_map *em;
534         struct btrfs_root *root = BTRFS_I(inode)->root;
535         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
536         struct extent_io_tree *io_tree;
537         int ret;
538
539         if (list_empty(&async_cow->extents))
540                 return 0;
541
542         trans = btrfs_join_transaction(root, 1);
543
544         while (!list_empty(&async_cow->extents)) {
545                 async_extent = list_entry(async_cow->extents.next,
546                                           struct async_extent, list);
547                 list_del(&async_extent->list);
548
549                 io_tree = &BTRFS_I(inode)->io_tree;
550
551                 /* did the compression code fall back to uncompressed IO? */
552                 if (!async_extent->pages) {
553                         int page_started = 0;
554                         unsigned long nr_written = 0;
555
556                         lock_extent(io_tree, async_extent->start,
557                                     async_extent->start +
558                                     async_extent->ram_size - 1, GFP_NOFS);
559
560                         /* allocate blocks */
561                         cow_file_range(inode, async_cow->locked_page,
562                                        async_extent->start,
563                                        async_extent->start +
564                                        async_extent->ram_size - 1,
565                                        &page_started, &nr_written, 0);
566
567                         /*
568                          * if page_started, cow_file_range inserted an
569                          * inline extent and took care of all the unlocking
570                          * and IO for us.  Otherwise, we need to submit
571                          * all those pages down to the drive.
572                          */
573                         if (!page_started)
574                                 extent_write_locked_range(io_tree,
575                                                   inode, async_extent->start,
576                                                   async_extent->start +
577                                                   async_extent->ram_size - 1,
578                                                   btrfs_get_extent,
579                                                   WB_SYNC_ALL);
580                         kfree(async_extent);
581                         cond_resched();
582                         continue;
583                 }
584
585                 lock_extent(io_tree, async_extent->start,
586                             async_extent->start + async_extent->ram_size - 1,
587                             GFP_NOFS);
588                 /*
589                  * here we're doing allocation and writeback of the
590                  * compressed pages
591                  */
592                 btrfs_drop_extent_cache(inode, async_extent->start,
593                                         async_extent->start +
594                                         async_extent->ram_size - 1, 0);
595
596                 ret = btrfs_reserve_extent(trans, root,
597                                            async_extent->compressed_size,
598                                            async_extent->compressed_size,
599                                            0, alloc_hint,
600                                            (u64)-1, &ins, 1);
601                 BUG_ON(ret);
602                 em = alloc_extent_map(GFP_NOFS);
603                 em->start = async_extent->start;
604                 em->len = async_extent->ram_size;
605                 em->orig_start = em->start;
606
607                 em->block_start = ins.objectid;
608                 em->block_len = ins.offset;
609                 em->bdev = root->fs_info->fs_devices->latest_bdev;
610                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
611                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
612
613                 while (1) {
614                         spin_lock(&em_tree->lock);
615                         ret = add_extent_mapping(em_tree, em);
616                         spin_unlock(&em_tree->lock);
617                         if (ret != -EEXIST) {
618                                 free_extent_map(em);
619                                 break;
620                         }
621                         btrfs_drop_extent_cache(inode, async_extent->start,
622                                                 async_extent->start +
623                                                 async_extent->ram_size - 1, 0);
624                 }
625
626                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
627                                                ins.objectid,
628                                                async_extent->ram_size,
629                                                ins.offset,
630                                                BTRFS_ORDERED_COMPRESSED);
631                 BUG_ON(ret);
632
633                 btrfs_end_transaction(trans, root);
634
635                 /*
636                  * clear dirty, set writeback and unlock the pages.
637                  */
638                 extent_clear_unlock_delalloc(inode,
639                                              &BTRFS_I(inode)->io_tree,
640                                              async_extent->start,
641                                              async_extent->start +
642                                              async_extent->ram_size - 1,
643                                              NULL, 1, 1, 0, 1, 1, 0);
644
645                 ret = btrfs_submit_compressed_write(inode,
646                                     async_extent->start,
647                                     async_extent->ram_size,
648                                     ins.objectid,
649                                     ins.offset, async_extent->pages,
650                                     async_extent->nr_pages);
651
652                 BUG_ON(ret);
653                 trans = btrfs_join_transaction(root, 1);
654                 alloc_hint = ins.objectid + ins.offset;
655                 kfree(async_extent);
656                 cond_resched();
657         }
658
659         btrfs_end_transaction(trans, root);
660         return 0;
661 }
662
663 /*
664  * when extent_io.c finds a delayed allocation range in the file,
665  * the call backs end up in this code.  The basic idea is to
666  * allocate extents on disk for the range, and create ordered data structs
667  * in ram to track those extents.
668  *
669  * locked_page is the page that writepage had locked already.  We use
670  * it to make sure we don't do extra locks or unlocks.
671  *
672  * *page_started is set to one if we unlock locked_page and do everything
673  * required to start IO on it.  It may be clean and already done with
674  * IO when we return.
675  */
676 static noinline int cow_file_range(struct inode *inode,
677                                    struct page *locked_page,
678                                    u64 start, u64 end, int *page_started,
679                                    unsigned long *nr_written,
680                                    int unlock)
681 {
682         struct btrfs_root *root = BTRFS_I(inode)->root;
683         struct btrfs_trans_handle *trans;
684         u64 alloc_hint = 0;
685         u64 num_bytes;
686         unsigned long ram_size;
687         u64 disk_num_bytes;
688         u64 cur_alloc_size;
689         u64 blocksize = root->sectorsize;
690         u64 actual_end;
691         u64 isize = i_size_read(inode);
692         struct btrfs_key ins;
693         struct extent_map *em;
694         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
695         int ret = 0;
696
697         trans = btrfs_join_transaction(root, 1);
698         BUG_ON(!trans);
699         btrfs_set_trans_block_group(trans, inode);
700
701         actual_end = min_t(u64, isize, end + 1);
702
703         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
704         num_bytes = max(blocksize,  num_bytes);
705         disk_num_bytes = num_bytes;
706         ret = 0;
707
708         if (start == 0) {
709                 /* lets try to make an inline extent */
710                 ret = cow_file_range_inline(trans, root, inode,
711                                             start, end, 0, NULL);
712                 if (ret == 0) {
713                         extent_clear_unlock_delalloc(inode,
714                                                      &BTRFS_I(inode)->io_tree,
715                                                      start, end, NULL, 1, 1,
716                                                      1, 1, 1, 1);
717                         *nr_written = *nr_written +
718                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
719                         *page_started = 1;
720                         ret = 0;
721                         goto out;
722                 }
723         }
724
725         BUG_ON(disk_num_bytes >
726                btrfs_super_total_bytes(&root->fs_info->super_copy));
727
728         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
729
730         while (disk_num_bytes > 0) {
731                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
732                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
733                                            root->sectorsize, 0, alloc_hint,
734                                            (u64)-1, &ins, 1);
735                 BUG_ON(ret);
736
737                 em = alloc_extent_map(GFP_NOFS);
738                 em->start = start;
739                 em->orig_start = em->start;
740
741                 ram_size = ins.offset;
742                 em->len = ins.offset;
743
744                 em->block_start = ins.objectid;
745                 em->block_len = ins.offset;
746                 em->bdev = root->fs_info->fs_devices->latest_bdev;
747                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
748
749                 while (1) {
750                         spin_lock(&em_tree->lock);
751                         ret = add_extent_mapping(em_tree, em);
752                         spin_unlock(&em_tree->lock);
753                         if (ret != -EEXIST) {
754                                 free_extent_map(em);
755                                 break;
756                         }
757                         btrfs_drop_extent_cache(inode, start,
758                                                 start + ram_size - 1, 0);
759                 }
760
761                 cur_alloc_size = ins.offset;
762                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
763                                                ram_size, cur_alloc_size, 0);
764                 BUG_ON(ret);
765
766                 if (root->root_key.objectid ==
767                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
768                         ret = btrfs_reloc_clone_csums(inode, start,
769                                                       cur_alloc_size);
770                         BUG_ON(ret);
771                 }
772
773                 if (disk_num_bytes < cur_alloc_size)
774                         break;
775
776                 /* we're not doing compressed IO, don't unlock the first
777                  * page (which the caller expects to stay locked), don't
778                  * clear any dirty bits and don't set any writeback bits
779                  */
780                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
781                                              start, start + ram_size - 1,
782                                              locked_page, unlock, 1,
783                                              1, 0, 0, 0);
784                 disk_num_bytes -= cur_alloc_size;
785                 num_bytes -= cur_alloc_size;
786                 alloc_hint = ins.objectid + ins.offset;
787                 start += cur_alloc_size;
788         }
789 out:
790         ret = 0;
791         btrfs_end_transaction(trans, root);
792
793         return ret;
794 }
795
796 /*
797  * work queue call back to started compression on a file and pages
798  */
799 static noinline void async_cow_start(struct btrfs_work *work)
800 {
801         struct async_cow *async_cow;
802         int num_added = 0;
803         async_cow = container_of(work, struct async_cow, work);
804
805         compress_file_range(async_cow->inode, async_cow->locked_page,
806                             async_cow->start, async_cow->end, async_cow,
807                             &num_added);
808         if (num_added == 0)
809                 async_cow->inode = NULL;
810 }
811
812 /*
813  * work queue call back to submit previously compressed pages
814  */
815 static noinline void async_cow_submit(struct btrfs_work *work)
816 {
817         struct async_cow *async_cow;
818         struct btrfs_root *root;
819         unsigned long nr_pages;
820
821         async_cow = container_of(work, struct async_cow, work);
822
823         root = async_cow->root;
824         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
825                 PAGE_CACHE_SHIFT;
826
827         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
828
829         if (atomic_read(&root->fs_info->async_delalloc_pages) <
830             5 * 1042 * 1024 &&
831             waitqueue_active(&root->fs_info->async_submit_wait))
832                 wake_up(&root->fs_info->async_submit_wait);
833
834         if (async_cow->inode)
835                 submit_compressed_extents(async_cow->inode, async_cow);
836 }
837
838 static noinline void async_cow_free(struct btrfs_work *work)
839 {
840         struct async_cow *async_cow;
841         async_cow = container_of(work, struct async_cow, work);
842         kfree(async_cow);
843 }
844
845 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
846                                 u64 start, u64 end, int *page_started,
847                                 unsigned long *nr_written)
848 {
849         struct async_cow *async_cow;
850         struct btrfs_root *root = BTRFS_I(inode)->root;
851         unsigned long nr_pages;
852         u64 cur_end;
853         int limit = 10 * 1024 * 1042;
854
855         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
856                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
857         while (start < end) {
858                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
859                 async_cow->inode = inode;
860                 async_cow->root = root;
861                 async_cow->locked_page = locked_page;
862                 async_cow->start = start;
863
864                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
865                         cur_end = end;
866                 else
867                         cur_end = min(end, start + 512 * 1024 - 1);
868
869                 async_cow->end = cur_end;
870                 INIT_LIST_HEAD(&async_cow->extents);
871
872                 async_cow->work.func = async_cow_start;
873                 async_cow->work.ordered_func = async_cow_submit;
874                 async_cow->work.ordered_free = async_cow_free;
875                 async_cow->work.flags = 0;
876
877                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
878                         PAGE_CACHE_SHIFT;
879                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
880
881                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
882                                    &async_cow->work);
883
884                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
885                         wait_event(root->fs_info->async_submit_wait,
886                            (atomic_read(&root->fs_info->async_delalloc_pages) <
887                             limit));
888                 }
889
890                 while (atomic_read(&root->fs_info->async_submit_draining) &&
891                       atomic_read(&root->fs_info->async_delalloc_pages)) {
892                         wait_event(root->fs_info->async_submit_wait,
893                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
894                            0));
895                 }
896
897                 *nr_written += nr_pages;
898                 start = cur_end + 1;
899         }
900         *page_started = 1;
901         return 0;
902 }
903
904 static noinline int csum_exist_in_range(struct btrfs_root *root,
905                                         u64 bytenr, u64 num_bytes)
906 {
907         int ret;
908         struct btrfs_ordered_sum *sums;
909         LIST_HEAD(list);
910
911         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
912                                        bytenr + num_bytes - 1, &list);
913         if (ret == 0 && list_empty(&list))
914                 return 0;
915
916         while (!list_empty(&list)) {
917                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
918                 list_del(&sums->list);
919                 kfree(sums);
920         }
921         return 1;
922 }
923
924 /*
925  * when nowcow writeback call back.  This checks for snapshots or COW copies
926  * of the extents that exist in the file, and COWs the file as required.
927  *
928  * If no cow copies or snapshots exist, we write directly to the existing
929  * blocks on disk
930  */
931 static noinline int run_delalloc_nocow(struct inode *inode,
932                                        struct page *locked_page,
933                               u64 start, u64 end, int *page_started, int force,
934                               unsigned long *nr_written)
935 {
936         struct btrfs_root *root = BTRFS_I(inode)->root;
937         struct btrfs_trans_handle *trans;
938         struct extent_buffer *leaf;
939         struct btrfs_path *path;
940         struct btrfs_file_extent_item *fi;
941         struct btrfs_key found_key;
942         u64 cow_start;
943         u64 cur_offset;
944         u64 extent_end;
945         u64 extent_offset;
946         u64 disk_bytenr;
947         u64 num_bytes;
948         int extent_type;
949         int ret;
950         int type;
951         int nocow;
952         int check_prev = 1;
953
954         path = btrfs_alloc_path();
955         BUG_ON(!path);
956         trans = btrfs_join_transaction(root, 1);
957         BUG_ON(!trans);
958
959         cow_start = (u64)-1;
960         cur_offset = start;
961         while (1) {
962                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
963                                                cur_offset, 0);
964                 BUG_ON(ret < 0);
965                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
966                         leaf = path->nodes[0];
967                         btrfs_item_key_to_cpu(leaf, &found_key,
968                                               path->slots[0] - 1);
969                         if (found_key.objectid == inode->i_ino &&
970                             found_key.type == BTRFS_EXTENT_DATA_KEY)
971                                 path->slots[0]--;
972                 }
973                 check_prev = 0;
974 next_slot:
975                 leaf = path->nodes[0];
976                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
977                         ret = btrfs_next_leaf(root, path);
978                         if (ret < 0)
979                                 BUG_ON(1);
980                         if (ret > 0)
981                                 break;
982                         leaf = path->nodes[0];
983                 }
984
985                 nocow = 0;
986                 disk_bytenr = 0;
987                 num_bytes = 0;
988                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
989
990                 if (found_key.objectid > inode->i_ino ||
991                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
992                     found_key.offset > end)
993                         break;
994
995                 if (found_key.offset > cur_offset) {
996                         extent_end = found_key.offset;
997                         goto out_check;
998                 }
999
1000                 fi = btrfs_item_ptr(leaf, path->slots[0],
1001                                     struct btrfs_file_extent_item);
1002                 extent_type = btrfs_file_extent_type(leaf, fi);
1003
1004                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1005                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1006                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1007                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1008                         extent_end = found_key.offset +
1009                                 btrfs_file_extent_num_bytes(leaf, fi);
1010                         if (extent_end <= start) {
1011                                 path->slots[0]++;
1012                                 goto next_slot;
1013                         }
1014                         if (disk_bytenr == 0)
1015                                 goto out_check;
1016                         if (btrfs_file_extent_compression(leaf, fi) ||
1017                             btrfs_file_extent_encryption(leaf, fi) ||
1018                             btrfs_file_extent_other_encoding(leaf, fi))
1019                                 goto out_check;
1020                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1021                                 goto out_check;
1022                         if (btrfs_extent_readonly(root, disk_bytenr))
1023                                 goto out_check;
1024                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1025                                                   found_key.offset -
1026                                                   extent_offset, disk_bytenr))
1027                                 goto out_check;
1028                         disk_bytenr += extent_offset;
1029                         disk_bytenr += cur_offset - found_key.offset;
1030                         num_bytes = min(end + 1, extent_end) - cur_offset;
1031                         /*
1032                          * force cow if csum exists in the range.
1033                          * this ensure that csum for a given extent are
1034                          * either valid or do not exist.
1035                          */
1036                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1037                                 goto out_check;
1038                         nocow = 1;
1039                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1040                         extent_end = found_key.offset +
1041                                 btrfs_file_extent_inline_len(leaf, fi);
1042                         extent_end = ALIGN(extent_end, root->sectorsize);
1043                 } else {
1044                         BUG_ON(1);
1045                 }
1046 out_check:
1047                 if (extent_end <= start) {
1048                         path->slots[0]++;
1049                         goto next_slot;
1050                 }
1051                 if (!nocow) {
1052                         if (cow_start == (u64)-1)
1053                                 cow_start = cur_offset;
1054                         cur_offset = extent_end;
1055                         if (cur_offset > end)
1056                                 break;
1057                         path->slots[0]++;
1058                         goto next_slot;
1059                 }
1060
1061                 btrfs_release_path(root, path);
1062                 if (cow_start != (u64)-1) {
1063                         ret = cow_file_range(inode, locked_page, cow_start,
1064                                         found_key.offset - 1, page_started,
1065                                         nr_written, 1);
1066                         BUG_ON(ret);
1067                         cow_start = (u64)-1;
1068                 }
1069
1070                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1071                         struct extent_map *em;
1072                         struct extent_map_tree *em_tree;
1073                         em_tree = &BTRFS_I(inode)->extent_tree;
1074                         em = alloc_extent_map(GFP_NOFS);
1075                         em->start = cur_offset;
1076                         em->orig_start = em->start;
1077                         em->len = num_bytes;
1078                         em->block_len = num_bytes;
1079                         em->block_start = disk_bytenr;
1080                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1081                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082                         while (1) {
1083                                 spin_lock(&em_tree->lock);
1084                                 ret = add_extent_mapping(em_tree, em);
1085                                 spin_unlock(&em_tree->lock);
1086                                 if (ret != -EEXIST) {
1087                                         free_extent_map(em);
1088                                         break;
1089                                 }
1090                                 btrfs_drop_extent_cache(inode, em->start,
1091                                                 em->start + em->len - 1, 0);
1092                         }
1093                         type = BTRFS_ORDERED_PREALLOC;
1094                 } else {
1095                         type = BTRFS_ORDERED_NOCOW;
1096                 }
1097
1098                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1099                                                num_bytes, num_bytes, type);
1100                 BUG_ON(ret);
1101
1102                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103                                         cur_offset, cur_offset + num_bytes - 1,
1104                                         locked_page, 1, 1, 1, 0, 0, 0);
1105                 cur_offset = extent_end;
1106                 if (cur_offset > end)
1107                         break;
1108         }
1109         btrfs_release_path(root, path);
1110
1111         if (cur_offset <= end && cow_start == (u64)-1)
1112                 cow_start = cur_offset;
1113         if (cow_start != (u64)-1) {
1114                 ret = cow_file_range(inode, locked_page, cow_start, end,
1115                                      page_started, nr_written, 1);
1116                 BUG_ON(ret);
1117         }
1118
1119         ret = btrfs_end_transaction(trans, root);
1120         BUG_ON(ret);
1121         btrfs_free_path(path);
1122         return 0;
1123 }
1124
1125 /*
1126  * extent_io.c call back to do delayed allocation processing
1127  */
1128 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1129                               u64 start, u64 end, int *page_started,
1130                               unsigned long *nr_written)
1131 {
1132         int ret;
1133         struct btrfs_root *root = BTRFS_I(inode)->root;
1134
1135         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1136                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1137                                          page_started, 1, nr_written);
1138         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1139                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1140                                          page_started, 0, nr_written);
1141         else if (!btrfs_test_opt(root, COMPRESS))
1142                 ret = cow_file_range(inode, locked_page, start, end,
1143                                       page_started, nr_written, 1);
1144         else
1145                 ret = cow_file_range_async(inode, locked_page, start, end,
1146                                            page_started, nr_written);
1147         return ret;
1148 }
1149
1150 /*
1151  * extent_io.c set_bit_hook, used to track delayed allocation
1152  * bytes in this file, and to maintain the list of inodes that
1153  * have pending delalloc work to be done.
1154  */
1155 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156                        unsigned long old, unsigned long bits)
1157 {
1158         /*
1159          * set_bit and clear bit hooks normally require _irqsave/restore
1160          * but in this case, we are only testeing for the DELALLOC
1161          * bit, which is only set or cleared with irqs on
1162          */
1163         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164                 struct btrfs_root *root = BTRFS_I(inode)->root;
1165                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166                 spin_lock(&root->fs_info->delalloc_lock);
1167                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1168                 root->fs_info->delalloc_bytes += end - start + 1;
1169                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171                                       &root->fs_info->delalloc_inodes);
1172                 }
1173                 spin_unlock(&root->fs_info->delalloc_lock);
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * extent_io.c clear_bit_hook, see set_bit_hook for why
1180  */
1181 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1182                          unsigned long old, unsigned long bits)
1183 {
1184         /*
1185          * set_bit and clear bit hooks normally require _irqsave/restore
1186          * but in this case, we are only testeing for the DELALLOC
1187          * bit, which is only set or cleared with irqs on
1188          */
1189         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190                 struct btrfs_root *root = BTRFS_I(inode)->root;
1191
1192                 spin_lock(&root->fs_info->delalloc_lock);
1193                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1194                         printk(KERN_INFO "btrfs warning: delalloc account "
1195                                "%llu %llu\n",
1196                                (unsigned long long)end - start + 1,
1197                                (unsigned long long)
1198                                root->fs_info->delalloc_bytes);
1199                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1200                         root->fs_info->delalloc_bytes = 0;
1201                         BTRFS_I(inode)->delalloc_bytes = 0;
1202                 } else {
1203                         btrfs_delalloc_free_space(root, inode,
1204                                                   end - start + 1);
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct btrfs_ordered_sum *sum;
1328
1329         btrfs_set_trans_block_group(trans, inode);
1330
1331         list_for_each_entry(sum, list, list) {
1332                 btrfs_csum_file_blocks(trans,
1333                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1334         }
1335         return 0;
1336 }
1337
1338 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339 {
1340         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1341                 WARN_ON(1);
1342         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343                                    GFP_NOFS);
1344 }
1345
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup {
1348         struct page *page;
1349         struct btrfs_work work;
1350 };
1351
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1353 {
1354         struct btrfs_writepage_fixup *fixup;
1355         struct btrfs_ordered_extent *ordered;
1356         struct page *page;
1357         struct inode *inode;
1358         u64 page_start;
1359         u64 page_end;
1360
1361         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362         page = fixup->page;
1363 again:
1364         lock_page(page);
1365         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366                 ClearPageChecked(page);
1367                 goto out_page;
1368         }
1369
1370         inode = page->mapping->host;
1371         page_start = page_offset(page);
1372         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375
1376         /* already ordered? We're done */
1377         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378                              EXTENT_ORDERED, 0)) {
1379                 goto out;
1380         }
1381
1382         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383         if (ordered) {
1384                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385                               page_end, GFP_NOFS);
1386                 unlock_page(page);
1387                 btrfs_start_ordered_extent(inode, ordered, 1);
1388                 goto again;
1389         }
1390
1391         btrfs_set_extent_delalloc(inode, page_start, page_end);
1392         ClearPageChecked(page);
1393 out:
1394         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395 out_page:
1396         unlock_page(page);
1397         page_cache_release(page);
1398 }
1399
1400 /*
1401  * There are a few paths in the higher layers of the kernel that directly
1402  * set the page dirty bit without asking the filesystem if it is a
1403  * good idea.  This causes problems because we want to make sure COW
1404  * properly happens and the data=ordered rules are followed.
1405  *
1406  * In our case any range that doesn't have the ORDERED bit set
1407  * hasn't been properly setup for IO.  We kick off an async process
1408  * to fix it up.  The async helper will wait for ordered extents, set
1409  * the delalloc bit and make it safe to write the page.
1410  */
1411 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1412 {
1413         struct inode *inode = page->mapping->host;
1414         struct btrfs_writepage_fixup *fixup;
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret;
1417
1418         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419                              EXTENT_ORDERED, 0);
1420         if (ret)
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u64 locked_end,
1443                                        u8 compression, u8 encryption,
1444                                        u16 other_encoding, int extent_type)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         struct btrfs_file_extent_item *fi;
1448         struct btrfs_path *path;
1449         struct extent_buffer *leaf;
1450         struct btrfs_key ins;
1451         u64 hint;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         BUG_ON(!path);
1456
1457         path->leave_spinning = 1;
1458         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1459                                  file_pos + num_bytes, locked_end,
1460                                  file_pos, &hint);
1461         BUG_ON(ret);
1462
1463         ins.objectid = inode->i_ino;
1464         ins.offset = file_pos;
1465         ins.type = BTRFS_EXTENT_DATA_KEY;
1466         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1467         BUG_ON(ret);
1468         leaf = path->nodes[0];
1469         fi = btrfs_item_ptr(leaf, path->slots[0],
1470                             struct btrfs_file_extent_item);
1471         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1472         btrfs_set_file_extent_type(leaf, fi, extent_type);
1473         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1474         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1475         btrfs_set_file_extent_offset(leaf, fi, 0);
1476         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1477         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1478         btrfs_set_file_extent_compression(leaf, fi, compression);
1479         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1480         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1481
1482         btrfs_unlock_up_safe(path, 1);
1483         btrfs_set_lock_blocking(leaf);
1484
1485         btrfs_mark_buffer_dirty(leaf);
1486
1487         inode_add_bytes(inode, num_bytes);
1488         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1489
1490         ins.objectid = disk_bytenr;
1491         ins.offset = disk_num_bytes;
1492         ins.type = BTRFS_EXTENT_ITEM_KEY;
1493         ret = btrfs_alloc_reserved_file_extent(trans, root,
1494                                         root->root_key.objectid,
1495                                         inode->i_ino, file_pos, &ins);
1496         BUG_ON(ret);
1497         btrfs_free_path(path);
1498
1499         return 0;
1500 }
1501
1502 /*
1503  * helper function for btrfs_finish_ordered_io, this
1504  * just reads in some of the csum leaves to prime them into ram
1505  * before we start the transaction.  It limits the amount of btree
1506  * reads required while inside the transaction.
1507  */
1508 static noinline void reada_csum(struct btrfs_root *root,
1509                                 struct btrfs_path *path,
1510                                 struct btrfs_ordered_extent *ordered_extent)
1511 {
1512         struct btrfs_ordered_sum *sum;
1513         u64 bytenr;
1514
1515         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1516                          list);
1517         bytenr = sum->sums[0].bytenr;
1518
1519         /*
1520          * we don't care about the results, the point of this search is
1521          * just to get the btree leaves into ram
1522          */
1523         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1524 }
1525
1526 /* as ordered data IO finishes, this gets called so we can finish
1527  * an ordered extent if the range of bytes in the file it covers are
1528  * fully written.
1529  */
1530 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1531 {
1532         struct btrfs_root *root = BTRFS_I(inode)->root;
1533         struct btrfs_trans_handle *trans;
1534         struct btrfs_ordered_extent *ordered_extent = NULL;
1535         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1536         struct btrfs_path *path;
1537         int compressed = 0;
1538         int ret;
1539
1540         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1541         if (!ret)
1542                 return 0;
1543
1544         /*
1545          * before we join the transaction, try to do some of our IO.
1546          * This will limit the amount of IO that we have to do with
1547          * the transaction running.  We're unlikely to need to do any
1548          * IO if the file extents are new, the disk_i_size checks
1549          * covers the most common case.
1550          */
1551         if (start < BTRFS_I(inode)->disk_i_size) {
1552                 path = btrfs_alloc_path();
1553                 if (path) {
1554                         ret = btrfs_lookup_file_extent(NULL, root, path,
1555                                                        inode->i_ino,
1556                                                        start, 0);
1557                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1558                                                                      start);
1559                         if (!list_empty(&ordered_extent->list)) {
1560                                 btrfs_release_path(root, path);
1561                                 reada_csum(root, path, ordered_extent);
1562                         }
1563                         btrfs_free_path(path);
1564                 }
1565         }
1566
1567         trans = btrfs_join_transaction(root, 1);
1568
1569         if (!ordered_extent)
1570                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1571         BUG_ON(!ordered_extent);
1572         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1573                 goto nocow;
1574
1575         lock_extent(io_tree, ordered_extent->file_offset,
1576                     ordered_extent->file_offset + ordered_extent->len - 1,
1577                     GFP_NOFS);
1578
1579         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1580                 compressed = 1;
1581         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1582                 BUG_ON(compressed);
1583                 ret = btrfs_mark_extent_written(trans, root, inode,
1584                                                 ordered_extent->file_offset,
1585                                                 ordered_extent->file_offset +
1586                                                 ordered_extent->len);
1587                 BUG_ON(ret);
1588         } else {
1589                 ret = insert_reserved_file_extent(trans, inode,
1590                                                 ordered_extent->file_offset,
1591                                                 ordered_extent->start,
1592                                                 ordered_extent->disk_len,
1593                                                 ordered_extent->len,
1594                                                 ordered_extent->len,
1595                                                 ordered_extent->file_offset +
1596                                                 ordered_extent->len,
1597                                                 compressed, 0, 0,
1598                                                 BTRFS_FILE_EXTENT_REG);
1599                 BUG_ON(ret);
1600         }
1601         unlock_extent(io_tree, ordered_extent->file_offset,
1602                     ordered_extent->file_offset + ordered_extent->len - 1,
1603                     GFP_NOFS);
1604 nocow:
1605         add_pending_csums(trans, inode, ordered_extent->file_offset,
1606                           &ordered_extent->list);
1607
1608         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1609         btrfs_ordered_update_i_size(inode, ordered_extent);
1610         btrfs_update_inode(trans, root, inode);
1611         btrfs_remove_ordered_extent(inode, ordered_extent);
1612         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1613
1614         /* once for us */
1615         btrfs_put_ordered_extent(ordered_extent);
1616         /* once for the tree */
1617         btrfs_put_ordered_extent(ordered_extent);
1618
1619         btrfs_end_transaction(trans, root);
1620         return 0;
1621 }
1622
1623 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1624                                 struct extent_state *state, int uptodate)
1625 {
1626         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1627 }
1628
1629 /*
1630  * When IO fails, either with EIO or csum verification fails, we
1631  * try other mirrors that might have a good copy of the data.  This
1632  * io_failure_record is used to record state as we go through all the
1633  * mirrors.  If another mirror has good data, the page is set up to date
1634  * and things continue.  If a good mirror can't be found, the original
1635  * bio end_io callback is called to indicate things have failed.
1636  */
1637 struct io_failure_record {
1638         struct page *page;
1639         u64 start;
1640         u64 len;
1641         u64 logical;
1642         unsigned long bio_flags;
1643         int last_mirror;
1644 };
1645
1646 static int btrfs_io_failed_hook(struct bio *failed_bio,
1647                          struct page *page, u64 start, u64 end,
1648                          struct extent_state *state)
1649 {
1650         struct io_failure_record *failrec = NULL;
1651         u64 private;
1652         struct extent_map *em;
1653         struct inode *inode = page->mapping->host;
1654         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1655         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1656         struct bio *bio;
1657         int num_copies;
1658         int ret;
1659         int rw;
1660         u64 logical;
1661
1662         ret = get_state_private(failure_tree, start, &private);
1663         if (ret) {
1664                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1665                 if (!failrec)
1666                         return -ENOMEM;
1667                 failrec->start = start;
1668                 failrec->len = end - start + 1;
1669                 failrec->last_mirror = 0;
1670                 failrec->bio_flags = 0;
1671
1672                 spin_lock(&em_tree->lock);
1673                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1674                 if (em->start > start || em->start + em->len < start) {
1675                         free_extent_map(em);
1676                         em = NULL;
1677                 }
1678                 spin_unlock(&em_tree->lock);
1679
1680                 if (!em || IS_ERR(em)) {
1681                         kfree(failrec);
1682                         return -EIO;
1683                 }
1684                 logical = start - em->start;
1685                 logical = em->block_start + logical;
1686                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1687                         logical = em->block_start;
1688                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1689                 }
1690                 failrec->logical = logical;
1691                 free_extent_map(em);
1692                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1693                                 EXTENT_DIRTY, GFP_NOFS);
1694                 set_state_private(failure_tree, start,
1695                                  (u64)(unsigned long)failrec);
1696         } else {
1697                 failrec = (struct io_failure_record *)(unsigned long)private;
1698         }
1699         num_copies = btrfs_num_copies(
1700                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1701                               failrec->logical, failrec->len);
1702         failrec->last_mirror++;
1703         if (!state) {
1704                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1705                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1706                                                     failrec->start,
1707                                                     EXTENT_LOCKED);
1708                 if (state && state->start != failrec->start)
1709                         state = NULL;
1710                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1711         }
1712         if (!state || failrec->last_mirror > num_copies) {
1713                 set_state_private(failure_tree, failrec->start, 0);
1714                 clear_extent_bits(failure_tree, failrec->start,
1715                                   failrec->start + failrec->len - 1,
1716                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1717                 kfree(failrec);
1718                 return -EIO;
1719         }
1720         bio = bio_alloc(GFP_NOFS, 1);
1721         bio->bi_private = state;
1722         bio->bi_end_io = failed_bio->bi_end_io;
1723         bio->bi_sector = failrec->logical >> 9;
1724         bio->bi_bdev = failed_bio->bi_bdev;
1725         bio->bi_size = 0;
1726
1727         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1728         if (failed_bio->bi_rw & (1 << BIO_RW))
1729                 rw = WRITE;
1730         else
1731                 rw = READ;
1732
1733         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1734                                                       failrec->last_mirror,
1735                                                       failrec->bio_flags);
1736         return 0;
1737 }
1738
1739 /*
1740  * each time an IO finishes, we do a fast check in the IO failure tree
1741  * to see if we need to process or clean up an io_failure_record
1742  */
1743 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1744 {
1745         u64 private;
1746         u64 private_failure;
1747         struct io_failure_record *failure;
1748         int ret;
1749
1750         private = 0;
1751         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1752                              (u64)-1, 1, EXTENT_DIRTY)) {
1753                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1754                                         start, &private_failure);
1755                 if (ret == 0) {
1756                         failure = (struct io_failure_record *)(unsigned long)
1757                                    private_failure;
1758                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1759                                           failure->start, 0);
1760                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1761                                           failure->start,
1762                                           failure->start + failure->len - 1,
1763                                           EXTENT_DIRTY | EXTENT_LOCKED,
1764                                           GFP_NOFS);
1765                         kfree(failure);
1766                 }
1767         }
1768         return 0;
1769 }
1770
1771 /*
1772  * when reads are done, we need to check csums to verify the data is correct
1773  * if there's a match, we allow the bio to finish.  If not, we go through
1774  * the io_failure_record routines to find good copies
1775  */
1776 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1777                                struct extent_state *state)
1778 {
1779         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1780         struct inode *inode = page->mapping->host;
1781         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1782         char *kaddr;
1783         u64 private = ~(u32)0;
1784         int ret;
1785         struct btrfs_root *root = BTRFS_I(inode)->root;
1786         u32 csum = ~(u32)0;
1787
1788         if (PageChecked(page)) {
1789                 ClearPageChecked(page);
1790                 goto good;
1791         }
1792
1793         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1794                 return 0;
1795
1796         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1797             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1798                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1799                                   GFP_NOFS);
1800                 return 0;
1801         }
1802
1803         if (state && state->start == start) {
1804                 private = state->private;
1805                 ret = 0;
1806         } else {
1807                 ret = get_state_private(io_tree, start, &private);
1808         }
1809         kaddr = kmap_atomic(page, KM_USER0);
1810         if (ret)
1811                 goto zeroit;
1812
1813         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1814         btrfs_csum_final(csum, (char *)&csum);
1815         if (csum != private)
1816                 goto zeroit;
1817
1818         kunmap_atomic(kaddr, KM_USER0);
1819 good:
1820         /* if the io failure tree for this inode is non-empty,
1821          * check to see if we've recovered from a failed IO
1822          */
1823         btrfs_clean_io_failures(inode, start);
1824         return 0;
1825
1826 zeroit:
1827         if (printk_ratelimit()) {
1828                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1829                        "private %llu\n", page->mapping->host->i_ino,
1830                        (unsigned long long)start, csum,
1831                        (unsigned long long)private);
1832         }
1833         memset(kaddr + offset, 1, end - start + 1);
1834         flush_dcache_page(page);
1835         kunmap_atomic(kaddr, KM_USER0);
1836         if (private == 0)
1837                 return 0;
1838         return -EIO;
1839 }
1840
1841 /*
1842  * This creates an orphan entry for the given inode in case something goes
1843  * wrong in the middle of an unlink/truncate.
1844  */
1845 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1846 {
1847         struct btrfs_root *root = BTRFS_I(inode)->root;
1848         int ret = 0;
1849
1850         spin_lock(&root->list_lock);
1851
1852         /* already on the orphan list, we're good */
1853         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1854                 spin_unlock(&root->list_lock);
1855                 return 0;
1856         }
1857
1858         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1859
1860         spin_unlock(&root->list_lock);
1861
1862         /*
1863          * insert an orphan item to track this unlinked/truncated file
1864          */
1865         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1866
1867         return ret;
1868 }
1869
1870 /*
1871  * We have done the truncate/delete so we can go ahead and remove the orphan
1872  * item for this particular inode.
1873  */
1874 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1875 {
1876         struct btrfs_root *root = BTRFS_I(inode)->root;
1877         int ret = 0;
1878
1879         spin_lock(&root->list_lock);
1880
1881         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1882                 spin_unlock(&root->list_lock);
1883                 return 0;
1884         }
1885
1886         list_del_init(&BTRFS_I(inode)->i_orphan);
1887         if (!trans) {
1888                 spin_unlock(&root->list_lock);
1889                 return 0;
1890         }
1891
1892         spin_unlock(&root->list_lock);
1893
1894         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1895
1896         return ret;
1897 }
1898
1899 /*
1900  * this cleans up any orphans that may be left on the list from the last use
1901  * of this root.
1902  */
1903 void btrfs_orphan_cleanup(struct btrfs_root *root)
1904 {
1905         struct btrfs_path *path;
1906         struct extent_buffer *leaf;
1907         struct btrfs_item *item;
1908         struct btrfs_key key, found_key;
1909         struct btrfs_trans_handle *trans;
1910         struct inode *inode;
1911         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1912
1913         path = btrfs_alloc_path();
1914         if (!path)
1915                 return;
1916         path->reada = -1;
1917
1918         key.objectid = BTRFS_ORPHAN_OBJECTID;
1919         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1920         key.offset = (u64)-1;
1921
1922
1923         while (1) {
1924                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1925                 if (ret < 0) {
1926                         printk(KERN_ERR "Error searching slot for orphan: %d"
1927                                "\n", ret);
1928                         break;
1929                 }
1930
1931                 /*
1932                  * if ret == 0 means we found what we were searching for, which
1933                  * is weird, but possible, so only screw with path if we didnt
1934                  * find the key and see if we have stuff that matches
1935                  */
1936                 if (ret > 0) {
1937                         if (path->slots[0] == 0)
1938                                 break;
1939                         path->slots[0]--;
1940                 }
1941
1942                 /* pull out the item */
1943                 leaf = path->nodes[0];
1944                 item = btrfs_item_nr(leaf, path->slots[0]);
1945                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1946
1947                 /* make sure the item matches what we want */
1948                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1949                         break;
1950                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1951                         break;
1952
1953                 /* release the path since we're done with it */
1954                 btrfs_release_path(root, path);
1955
1956                 /*
1957                  * this is where we are basically btrfs_lookup, without the
1958                  * crossing root thing.  we store the inode number in the
1959                  * offset of the orphan item.
1960                  */
1961                 found_key.objectid = found_key.offset;
1962                 found_key.type = BTRFS_INODE_ITEM_KEY;
1963                 found_key.offset = 0;
1964                 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1965                 if (IS_ERR(inode))
1966                         break;
1967
1968                 /*
1969                  * add this inode to the orphan list so btrfs_orphan_del does
1970                  * the proper thing when we hit it
1971                  */
1972                 spin_lock(&root->list_lock);
1973                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1974                 spin_unlock(&root->list_lock);
1975
1976                 /*
1977                  * if this is a bad inode, means we actually succeeded in
1978                  * removing the inode, but not the orphan record, which means
1979                  * we need to manually delete the orphan since iput will just
1980                  * do a destroy_inode
1981                  */
1982                 if (is_bad_inode(inode)) {
1983                         trans = btrfs_start_transaction(root, 1);
1984                         btrfs_orphan_del(trans, inode);
1985                         btrfs_end_transaction(trans, root);
1986                         iput(inode);
1987                         continue;
1988                 }
1989
1990                 /* if we have links, this was a truncate, lets do that */
1991                 if (inode->i_nlink) {
1992                         nr_truncate++;
1993                         btrfs_truncate(inode);
1994                 } else {
1995                         nr_unlink++;
1996                 }
1997
1998                 /* this will do delete_inode and everything for us */
1999                 iput(inode);
2000         }
2001
2002         if (nr_unlink)
2003                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2004         if (nr_truncate)
2005                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2006
2007         btrfs_free_path(path);
2008 }
2009
2010 /*
2011  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2012  * don't find any xattrs, we know there can't be any acls.
2013  *
2014  * slot is the slot the inode is in, objectid is the objectid of the inode
2015  */
2016 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2017                                           int slot, u64 objectid)
2018 {
2019         u32 nritems = btrfs_header_nritems(leaf);
2020         struct btrfs_key found_key;
2021         int scanned = 0;
2022
2023         slot++;
2024         while (slot < nritems) {
2025                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2026
2027                 /* we found a different objectid, there must not be acls */
2028                 if (found_key.objectid != objectid)
2029                         return 0;
2030
2031                 /* we found an xattr, assume we've got an acl */
2032                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2033                         return 1;
2034
2035                 /*
2036                  * we found a key greater than an xattr key, there can't
2037                  * be any acls later on
2038                  */
2039                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2040                         return 0;
2041
2042                 slot++;
2043                 scanned++;
2044
2045                 /*
2046                  * it goes inode, inode backrefs, xattrs, extents,
2047                  * so if there are a ton of hard links to an inode there can
2048                  * be a lot of backrefs.  Don't waste time searching too hard,
2049                  * this is just an optimization
2050                  */
2051                 if (scanned >= 8)
2052                         break;
2053         }
2054         /* we hit the end of the leaf before we found an xattr or
2055          * something larger than an xattr.  We have to assume the inode
2056          * has acls
2057          */
2058         return 1;
2059 }
2060
2061 /*
2062  * read an inode from the btree into the in-memory inode
2063  */
2064 static void btrfs_read_locked_inode(struct inode *inode)
2065 {
2066         struct btrfs_path *path;
2067         struct extent_buffer *leaf;
2068         struct btrfs_inode_item *inode_item;
2069         struct btrfs_timespec *tspec;
2070         struct btrfs_root *root = BTRFS_I(inode)->root;
2071         struct btrfs_key location;
2072         int maybe_acls;
2073         u64 alloc_group_block;
2074         u32 rdev;
2075         int ret;
2076
2077         path = btrfs_alloc_path();
2078         BUG_ON(!path);
2079         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2080
2081         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2082         if (ret)
2083                 goto make_bad;
2084
2085         leaf = path->nodes[0];
2086         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2087                                     struct btrfs_inode_item);
2088
2089         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2090         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2091         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2092         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2093         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2094
2095         tspec = btrfs_inode_atime(inode_item);
2096         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2097         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2098
2099         tspec = btrfs_inode_mtime(inode_item);
2100         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2101         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2102
2103         tspec = btrfs_inode_ctime(inode_item);
2104         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2105         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2106
2107         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2108         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2109         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2110         inode->i_generation = BTRFS_I(inode)->generation;
2111         inode->i_rdev = 0;
2112         rdev = btrfs_inode_rdev(leaf, inode_item);
2113
2114         BTRFS_I(inode)->index_cnt = (u64)-1;
2115         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2116
2117         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2118
2119         /*
2120          * try to precache a NULL acl entry for files that don't have
2121          * any xattrs or acls
2122          */
2123         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2124         if (!maybe_acls)
2125                 cache_no_acl(inode);
2126
2127         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2128                                                 alloc_group_block, 0);
2129         btrfs_free_path(path);
2130         inode_item = NULL;
2131
2132         switch (inode->i_mode & S_IFMT) {
2133         case S_IFREG:
2134                 inode->i_mapping->a_ops = &btrfs_aops;
2135                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2136                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2137                 inode->i_fop = &btrfs_file_operations;
2138                 inode->i_op = &btrfs_file_inode_operations;
2139                 break;
2140         case S_IFDIR:
2141                 inode->i_fop = &btrfs_dir_file_operations;
2142                 if (root == root->fs_info->tree_root)
2143                         inode->i_op = &btrfs_dir_ro_inode_operations;
2144                 else
2145                         inode->i_op = &btrfs_dir_inode_operations;
2146                 break;
2147         case S_IFLNK:
2148                 inode->i_op = &btrfs_symlink_inode_operations;
2149                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2150                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2151                 break;
2152         default:
2153                 inode->i_op = &btrfs_special_inode_operations;
2154                 init_special_inode(inode, inode->i_mode, rdev);
2155                 break;
2156         }
2157
2158         btrfs_update_iflags(inode);
2159         return;
2160
2161 make_bad:
2162         btrfs_free_path(path);
2163         make_bad_inode(inode);
2164 }
2165
2166 /*
2167  * given a leaf and an inode, copy the inode fields into the leaf
2168  */
2169 static void fill_inode_item(struct btrfs_trans_handle *trans,
2170                             struct extent_buffer *leaf,
2171                             struct btrfs_inode_item *item,
2172                             struct inode *inode)
2173 {
2174         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2175         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2176         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2177         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2178         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2179
2180         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2181                                inode->i_atime.tv_sec);
2182         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2183                                 inode->i_atime.tv_nsec);
2184
2185         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2186                                inode->i_mtime.tv_sec);
2187         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2188                                 inode->i_mtime.tv_nsec);
2189
2190         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2191                                inode->i_ctime.tv_sec);
2192         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2193                                 inode->i_ctime.tv_nsec);
2194
2195         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2196         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2197         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2198         btrfs_set_inode_transid(leaf, item, trans->transid);
2199         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2200         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2201         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2202 }
2203
2204 /*
2205  * copy everything in the in-memory inode into the btree.
2206  */
2207 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2208                                 struct btrfs_root *root, struct inode *inode)
2209 {
2210         struct btrfs_inode_item *inode_item;
2211         struct btrfs_path *path;
2212         struct extent_buffer *leaf;
2213         int ret;
2214
2215         path = btrfs_alloc_path();
2216         BUG_ON(!path);
2217         path->leave_spinning = 1;
2218         ret = btrfs_lookup_inode(trans, root, path,
2219                                  &BTRFS_I(inode)->location, 1);
2220         if (ret) {
2221                 if (ret > 0)
2222                         ret = -ENOENT;
2223                 goto failed;
2224         }
2225
2226         btrfs_unlock_up_safe(path, 1);
2227         leaf = path->nodes[0];
2228         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2229                                   struct btrfs_inode_item);
2230
2231         fill_inode_item(trans, leaf, inode_item, inode);
2232         btrfs_mark_buffer_dirty(leaf);
2233         btrfs_set_inode_last_trans(trans, inode);
2234         ret = 0;
2235 failed:
2236         btrfs_free_path(path);
2237         return ret;
2238 }
2239
2240
2241 /*
2242  * unlink helper that gets used here in inode.c and in the tree logging
2243  * recovery code.  It remove a link in a directory with a given name, and
2244  * also drops the back refs in the inode to the directory
2245  */
2246 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2247                        struct btrfs_root *root,
2248                        struct inode *dir, struct inode *inode,
2249                        const char *name, int name_len)
2250 {
2251         struct btrfs_path *path;
2252         int ret = 0;
2253         struct extent_buffer *leaf;
2254         struct btrfs_dir_item *di;
2255         struct btrfs_key key;
2256         u64 index;
2257
2258         path = btrfs_alloc_path();
2259         if (!path) {
2260                 ret = -ENOMEM;
2261                 goto err;
2262         }
2263
2264         path->leave_spinning = 1;
2265         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2266                                     name, name_len, -1);
2267         if (IS_ERR(di)) {
2268                 ret = PTR_ERR(di);
2269                 goto err;
2270         }
2271         if (!di) {
2272                 ret = -ENOENT;
2273                 goto err;
2274         }
2275         leaf = path->nodes[0];
2276         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2277         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2278         if (ret)
2279                 goto err;
2280         btrfs_release_path(root, path);
2281
2282         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2283                                   inode->i_ino,
2284                                   dir->i_ino, &index);
2285         if (ret) {
2286                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2287                        "inode %lu parent %lu\n", name_len, name,
2288                        inode->i_ino, dir->i_ino);
2289                 goto err;
2290         }
2291
2292         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2293                                          index, name, name_len, -1);
2294         if (IS_ERR(di)) {
2295                 ret = PTR_ERR(di);
2296                 goto err;
2297         }
2298         if (!di) {
2299                 ret = -ENOENT;
2300                 goto err;
2301         }
2302         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2303         btrfs_release_path(root, path);
2304
2305         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2306                                          inode, dir->i_ino);
2307         BUG_ON(ret != 0 && ret != -ENOENT);
2308
2309         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2310                                            dir, index);
2311         BUG_ON(ret);
2312 err:
2313         btrfs_free_path(path);
2314         if (ret)
2315                 goto out;
2316
2317         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2318         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2319         btrfs_update_inode(trans, root, dir);
2320         btrfs_drop_nlink(inode);
2321         ret = btrfs_update_inode(trans, root, inode);
2322 out:
2323         return ret;
2324 }
2325
2326 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2327 {
2328         struct btrfs_root *root;
2329         struct btrfs_trans_handle *trans;
2330         struct inode *inode = dentry->d_inode;
2331         int ret;
2332         unsigned long nr = 0;
2333
2334         root = BTRFS_I(dir)->root;
2335
2336         trans = btrfs_start_transaction(root, 1);
2337
2338         btrfs_set_trans_block_group(trans, dir);
2339
2340         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2341
2342         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2343                                  dentry->d_name.name, dentry->d_name.len);
2344
2345         if (inode->i_nlink == 0)
2346                 ret = btrfs_orphan_add(trans, inode);
2347
2348         nr = trans->blocks_used;
2349
2350         btrfs_end_transaction_throttle(trans, root);
2351         btrfs_btree_balance_dirty(root, nr);
2352         return ret;
2353 }
2354
2355 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2356 {
2357         struct inode *inode = dentry->d_inode;
2358         int err = 0;
2359         int ret;
2360         struct btrfs_root *root = BTRFS_I(dir)->root;
2361         struct btrfs_trans_handle *trans;
2362         unsigned long nr = 0;
2363
2364         /*
2365          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2366          * the root of a subvolume or snapshot
2367          */
2368         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2369             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2370                 return -ENOTEMPTY;
2371         }
2372
2373         trans = btrfs_start_transaction(root, 1);
2374         btrfs_set_trans_block_group(trans, dir);
2375
2376         err = btrfs_orphan_add(trans, inode);
2377         if (err)
2378                 goto fail_trans;
2379
2380         /* now the directory is empty */
2381         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2382                                  dentry->d_name.name, dentry->d_name.len);
2383         if (!err)
2384                 btrfs_i_size_write(inode, 0);
2385
2386 fail_trans:
2387         nr = trans->blocks_used;
2388         ret = btrfs_end_transaction_throttle(trans, root);
2389         btrfs_btree_balance_dirty(root, nr);
2390
2391         if (ret && !err)
2392                 err = ret;
2393         return err;
2394 }
2395
2396 #if 0
2397 /*
2398  * when truncating bytes in a file, it is possible to avoid reading
2399  * the leaves that contain only checksum items.  This can be the
2400  * majority of the IO required to delete a large file, but it must
2401  * be done carefully.
2402  *
2403  * The keys in the level just above the leaves are checked to make sure
2404  * the lowest key in a given leaf is a csum key, and starts at an offset
2405  * after the new  size.
2406  *
2407  * Then the key for the next leaf is checked to make sure it also has
2408  * a checksum item for the same file.  If it does, we know our target leaf
2409  * contains only checksum items, and it can be safely freed without reading
2410  * it.
2411  *
2412  * This is just an optimization targeted at large files.  It may do
2413  * nothing.  It will return 0 unless things went badly.
2414  */
2415 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2416                                      struct btrfs_root *root,
2417                                      struct btrfs_path *path,
2418                                      struct inode *inode, u64 new_size)
2419 {
2420         struct btrfs_key key;
2421         int ret;
2422         int nritems;
2423         struct btrfs_key found_key;
2424         struct btrfs_key other_key;
2425         struct btrfs_leaf_ref *ref;
2426         u64 leaf_gen;
2427         u64 leaf_start;
2428
2429         path->lowest_level = 1;
2430         key.objectid = inode->i_ino;
2431         key.type = BTRFS_CSUM_ITEM_KEY;
2432         key.offset = new_size;
2433 again:
2434         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2435         if (ret < 0)
2436                 goto out;
2437
2438         if (path->nodes[1] == NULL) {
2439                 ret = 0;
2440                 goto out;
2441         }
2442         ret = 0;
2443         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2444         nritems = btrfs_header_nritems(path->nodes[1]);
2445
2446         if (!nritems)
2447                 goto out;
2448
2449         if (path->slots[1] >= nritems)
2450                 goto next_node;
2451
2452         /* did we find a key greater than anything we want to delete? */
2453         if (found_key.objectid > inode->i_ino ||
2454            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2455                 goto out;
2456
2457         /* we check the next key in the node to make sure the leave contains
2458          * only checksum items.  This comparison doesn't work if our
2459          * leaf is the last one in the node
2460          */
2461         if (path->slots[1] + 1 >= nritems) {
2462 next_node:
2463                 /* search forward from the last key in the node, this
2464                  * will bring us into the next node in the tree
2465                  */
2466                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2467
2468                 /* unlikely, but we inc below, so check to be safe */
2469                 if (found_key.offset == (u64)-1)
2470                         goto out;
2471
2472                 /* search_forward needs a path with locks held, do the
2473                  * search again for the original key.  It is possible
2474                  * this will race with a balance and return a path that
2475                  * we could modify, but this drop is just an optimization
2476                  * and is allowed to miss some leaves.
2477                  */
2478                 btrfs_release_path(root, path);
2479                 found_key.offset++;
2480
2481                 /* setup a max key for search_forward */
2482                 other_key.offset = (u64)-1;
2483                 other_key.type = key.type;
2484                 other_key.objectid = key.objectid;
2485
2486                 path->keep_locks = 1;
2487                 ret = btrfs_search_forward(root, &found_key, &other_key,
2488                                            path, 0, 0);
2489                 path->keep_locks = 0;
2490                 if (ret || found_key.objectid != key.objectid ||
2491                     found_key.type != key.type) {
2492                         ret = 0;
2493                         goto out;
2494                 }
2495
2496                 key.offset = found_key.offset;
2497                 btrfs_release_path(root, path);
2498                 cond_resched();
2499                 goto again;
2500         }
2501
2502         /* we know there's one more slot after us in the tree,
2503          * read that key so we can verify it is also a checksum item
2504          */
2505         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2506
2507         if (found_key.objectid < inode->i_ino)
2508                 goto next_key;
2509
2510         if (found_key.type != key.type || found_key.offset < new_size)
2511                 goto next_key;
2512
2513         /*
2514          * if the key for the next leaf isn't a csum key from this objectid,
2515          * we can't be sure there aren't good items inside this leaf.
2516          * Bail out
2517          */
2518         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2519                 goto out;
2520
2521         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2522         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2523         /*
2524          * it is safe to delete this leaf, it contains only
2525          * csum items from this inode at an offset >= new_size
2526          */
2527         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2528         BUG_ON(ret);
2529
2530         if (root->ref_cows && leaf_gen < trans->transid) {
2531                 ref = btrfs_alloc_leaf_ref(root, 0);
2532                 if (ref) {
2533                         ref->root_gen = root->root_key.offset;
2534                         ref->bytenr = leaf_start;
2535                         ref->owner = 0;
2536                         ref->generation = leaf_gen;
2537                         ref->nritems = 0;
2538
2539                         btrfs_sort_leaf_ref(ref);
2540
2541                         ret = btrfs_add_leaf_ref(root, ref, 0);
2542                         WARN_ON(ret);
2543                         btrfs_free_leaf_ref(root, ref);
2544                 } else {
2545                         WARN_ON(1);
2546                 }
2547         }
2548 next_key:
2549         btrfs_release_path(root, path);
2550
2551         if (other_key.objectid == inode->i_ino &&
2552             other_key.type == key.type && other_key.offset > key.offset) {
2553                 key.offset = other_key.offset;
2554                 cond_resched();
2555                 goto again;
2556         }
2557         ret = 0;
2558 out:
2559         /* fixup any changes we've made to the path */
2560         path->lowest_level = 0;
2561         path->keep_locks = 0;
2562         btrfs_release_path(root, path);
2563         return ret;
2564 }
2565
2566 #endif
2567
2568 /*
2569  * this can truncate away extent items, csum items and directory items.
2570  * It starts at a high offset and removes keys until it can't find
2571  * any higher than new_size
2572  *
2573  * csum items that cross the new i_size are truncated to the new size
2574  * as well.
2575  *
2576  * min_type is the minimum key type to truncate down to.  If set to 0, this
2577  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2578  */
2579 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2580                                         struct btrfs_root *root,
2581                                         struct inode *inode,
2582                                         u64 new_size, u32 min_type)
2583 {
2584         int ret;
2585         struct btrfs_path *path;
2586         struct btrfs_key key;
2587         struct btrfs_key found_key;
2588         u32 found_type = (u8)-1;
2589         struct extent_buffer *leaf;
2590         struct btrfs_file_extent_item *fi;
2591         u64 extent_start = 0;
2592         u64 extent_num_bytes = 0;
2593         u64 extent_offset = 0;
2594         u64 item_end = 0;
2595         int found_extent;
2596         int del_item;
2597         int pending_del_nr = 0;
2598         int pending_del_slot = 0;
2599         int extent_type = -1;
2600         int encoding;
2601         u64 mask = root->sectorsize - 1;
2602
2603         if (root->ref_cows)
2604                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2605         path = btrfs_alloc_path();
2606         BUG_ON(!path);
2607         path->reada = -1;
2608
2609         /* FIXME, add redo link to tree so we don't leak on crash */
2610         key.objectid = inode->i_ino;
2611         key.offset = (u64)-1;
2612         key.type = (u8)-1;
2613
2614 search_again:
2615         path->leave_spinning = 1;
2616         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2617         if (ret < 0)
2618                 goto error;
2619
2620         if (ret > 0) {
2621                 /* there are no items in the tree for us to truncate, we're
2622                  * done
2623                  */
2624                 if (path->slots[0] == 0) {
2625                         ret = 0;
2626                         goto error;
2627                 }
2628                 path->slots[0]--;
2629         }
2630
2631         while (1) {
2632                 fi = NULL;
2633                 leaf = path->nodes[0];
2634                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2635                 found_type = btrfs_key_type(&found_key);
2636                 encoding = 0;
2637
2638                 if (found_key.objectid != inode->i_ino)
2639                         break;
2640
2641                 if (found_type < min_type)
2642                         break;
2643
2644                 item_end = found_key.offset;
2645                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2646                         fi = btrfs_item_ptr(leaf, path->slots[0],
2647                                             struct btrfs_file_extent_item);
2648                         extent_type = btrfs_file_extent_type(leaf, fi);
2649                         encoding = btrfs_file_extent_compression(leaf, fi);
2650                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2651                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2652
2653                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2654                                 item_end +=
2655                                     btrfs_file_extent_num_bytes(leaf, fi);
2656                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2657                                 item_end += btrfs_file_extent_inline_len(leaf,
2658                                                                          fi);
2659                         }
2660                         item_end--;
2661                 }
2662                 if (item_end < new_size) {
2663                         if (found_type == BTRFS_DIR_ITEM_KEY)
2664                                 found_type = BTRFS_INODE_ITEM_KEY;
2665                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2666                                 found_type = BTRFS_EXTENT_DATA_KEY;
2667                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2668                                 found_type = BTRFS_XATTR_ITEM_KEY;
2669                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2670                                 found_type = BTRFS_INODE_REF_KEY;
2671                         else if (found_type)
2672                                 found_type--;
2673                         else
2674                                 break;
2675                         btrfs_set_key_type(&key, found_type);
2676                         goto next;
2677                 }
2678                 if (found_key.offset >= new_size)
2679                         del_item = 1;
2680                 else
2681                         del_item = 0;
2682                 found_extent = 0;
2683
2684                 /* FIXME, shrink the extent if the ref count is only 1 */
2685                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2686                         goto delete;
2687
2688                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2689                         u64 num_dec;
2690                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2691                         if (!del_item && !encoding) {
2692                                 u64 orig_num_bytes =
2693                                         btrfs_file_extent_num_bytes(leaf, fi);
2694                                 extent_num_bytes = new_size -
2695                                         found_key.offset + root->sectorsize - 1;
2696                                 extent_num_bytes = extent_num_bytes &
2697                                         ~((u64)root->sectorsize - 1);
2698                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2699                                                          extent_num_bytes);
2700                                 num_dec = (orig_num_bytes -
2701                                            extent_num_bytes);
2702                                 if (root->ref_cows && extent_start != 0)
2703                                         inode_sub_bytes(inode, num_dec);
2704                                 btrfs_mark_buffer_dirty(leaf);
2705                         } else {
2706                                 extent_num_bytes =
2707                                         btrfs_file_extent_disk_num_bytes(leaf,
2708                                                                          fi);
2709                                 extent_offset = found_key.offset -
2710                                         btrfs_file_extent_offset(leaf, fi);
2711
2712                                 /* FIXME blocksize != 4096 */
2713                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2714                                 if (extent_start != 0) {
2715                                         found_extent = 1;
2716                                         if (root->ref_cows)
2717                                                 inode_sub_bytes(inode, num_dec);
2718                                 }
2719                         }
2720                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2721                         /*
2722                          * we can't truncate inline items that have had
2723                          * special encodings
2724                          */
2725                         if (!del_item &&
2726                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2727                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2728                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2729                                 u32 size = new_size - found_key.offset;
2730
2731                                 if (root->ref_cows) {
2732                                         inode_sub_bytes(inode, item_end + 1 -
2733                                                         new_size);
2734                                 }
2735                                 size =
2736                                     btrfs_file_extent_calc_inline_size(size);
2737                                 ret = btrfs_truncate_item(trans, root, path,
2738                                                           size, 1);
2739                                 BUG_ON(ret);
2740                         } else if (root->ref_cows) {
2741                                 inode_sub_bytes(inode, item_end + 1 -
2742                                                 found_key.offset);
2743                         }
2744                 }
2745 delete:
2746                 if (del_item) {
2747                         if (!pending_del_nr) {
2748                                 /* no pending yet, add ourselves */
2749                                 pending_del_slot = path->slots[0];
2750                                 pending_del_nr = 1;
2751                         } else if (pending_del_nr &&
2752                                    path->slots[0] + 1 == pending_del_slot) {
2753                                 /* hop on the pending chunk */
2754                                 pending_del_nr++;
2755                                 pending_del_slot = path->slots[0];
2756                         } else {
2757                                 BUG();
2758                         }
2759                 } else {
2760                         break;
2761                 }
2762                 if (found_extent && root->ref_cows) {
2763                         btrfs_set_path_blocking(path);
2764                         ret = btrfs_free_extent(trans, root, extent_start,
2765                                                 extent_num_bytes, 0,
2766                                                 btrfs_header_owner(leaf),
2767                                                 inode->i_ino, extent_offset);
2768                         BUG_ON(ret);
2769                 }
2770 next:
2771                 if (path->slots[0] == 0) {
2772                         if (pending_del_nr)
2773                                 goto del_pending;
2774                         btrfs_release_path(root, path);
2775                         if (found_type == BTRFS_INODE_ITEM_KEY)
2776                                 break;
2777                         goto search_again;
2778                 }
2779
2780                 path->slots[0]--;
2781                 if (pending_del_nr &&
2782                     path->slots[0] + 1 != pending_del_slot) {
2783                         struct btrfs_key debug;
2784 del_pending:
2785                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2786                                               pending_del_slot);
2787                         ret = btrfs_del_items(trans, root, path,
2788                                               pending_del_slot,
2789                                               pending_del_nr);
2790                         BUG_ON(ret);
2791                         pending_del_nr = 0;
2792                         btrfs_release_path(root, path);
2793                         if (found_type == BTRFS_INODE_ITEM_KEY)
2794                                 break;
2795                         goto search_again;
2796                 }
2797         }
2798         ret = 0;
2799 error:
2800         if (pending_del_nr) {
2801                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2802                                       pending_del_nr);
2803         }
2804         btrfs_free_path(path);
2805         return ret;
2806 }
2807
2808 /*
2809  * taken from block_truncate_page, but does cow as it zeros out
2810  * any bytes left in the last page in the file.
2811  */
2812 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2813 {
2814         struct inode *inode = mapping->host;
2815         struct btrfs_root *root = BTRFS_I(inode)->root;
2816         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2817         struct btrfs_ordered_extent *ordered;
2818         char *kaddr;
2819         u32 blocksize = root->sectorsize;
2820         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2821         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2822         struct page *page;
2823         int ret = 0;
2824         u64 page_start;
2825         u64 page_end;
2826
2827         if ((offset & (blocksize - 1)) == 0)
2828                 goto out;
2829
2830         ret = -ENOMEM;
2831 again:
2832         page = grab_cache_page(mapping, index);
2833         if (!page)
2834                 goto out;
2835
2836         page_start = page_offset(page);
2837         page_end = page_start + PAGE_CACHE_SIZE - 1;
2838
2839         if (!PageUptodate(page)) {
2840                 ret = btrfs_readpage(NULL, page);
2841                 lock_page(page);
2842                 if (page->mapping != mapping) {
2843                         unlock_page(page);
2844                         page_cache_release(page);
2845                         goto again;
2846                 }
2847                 if (!PageUptodate(page)) {
2848                         ret = -EIO;
2849                         goto out_unlock;
2850                 }
2851         }
2852         wait_on_page_writeback(page);
2853
2854         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2855         set_page_extent_mapped(page);
2856
2857         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2858         if (ordered) {
2859                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2860                 unlock_page(page);
2861                 page_cache_release(page);
2862                 btrfs_start_ordered_extent(inode, ordered, 1);
2863                 btrfs_put_ordered_extent(ordered);
2864                 goto again;
2865         }
2866
2867         btrfs_set_extent_delalloc(inode, page_start, page_end);
2868         ret = 0;
2869         if (offset != PAGE_CACHE_SIZE) {
2870                 kaddr = kmap(page);
2871                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2872                 flush_dcache_page(page);
2873                 kunmap(page);
2874         }
2875         ClearPageChecked(page);
2876         set_page_dirty(page);
2877         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2878
2879 out_unlock:
2880         unlock_page(page);
2881         page_cache_release(page);
2882 out:
2883         return ret;
2884 }
2885
2886 int btrfs_cont_expand(struct inode *inode, loff_t size)
2887 {
2888         struct btrfs_trans_handle *trans;
2889         struct btrfs_root *root = BTRFS_I(inode)->root;
2890         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2891         struct extent_map *em;
2892         u64 mask = root->sectorsize - 1;
2893         u64 hole_start = (inode->i_size + mask) & ~mask;
2894         u64 block_end = (size + mask) & ~mask;
2895         u64 last_byte;
2896         u64 cur_offset;
2897         u64 hole_size;
2898         int err;
2899
2900         if (size <= hole_start)
2901                 return 0;
2902
2903         err = btrfs_check_metadata_free_space(root);
2904         if (err)
2905                 return err;
2906
2907         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2908
2909         while (1) {
2910                 struct btrfs_ordered_extent *ordered;
2911                 btrfs_wait_ordered_range(inode, hole_start,
2912                                          block_end - hole_start);
2913                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2914                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2915                 if (!ordered)
2916                         break;
2917                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2918                 btrfs_put_ordered_extent(ordered);
2919         }
2920
2921         trans = btrfs_start_transaction(root, 1);
2922         btrfs_set_trans_block_group(trans, inode);
2923
2924         cur_offset = hole_start;
2925         while (1) {
2926                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2927                                 block_end - cur_offset, 0);
2928                 BUG_ON(IS_ERR(em) || !em);
2929                 last_byte = min(extent_map_end(em), block_end);
2930                 last_byte = (last_byte + mask) & ~mask;
2931                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2932                         u64 hint_byte = 0;
2933                         hole_size = last_byte - cur_offset;
2934                         err = btrfs_drop_extents(trans, root, inode,
2935                                                  cur_offset,
2936                                                  cur_offset + hole_size,
2937                                                  block_end,
2938                                                  cur_offset, &hint_byte);
2939                         if (err)
2940                                 break;
2941                         err = btrfs_insert_file_extent(trans, root,
2942                                         inode->i_ino, cur_offset, 0,
2943                                         0, hole_size, 0, hole_size,
2944                                         0, 0, 0);
2945                         btrfs_drop_extent_cache(inode, hole_start,
2946                                         last_byte - 1, 0);
2947                 }
2948                 free_extent_map(em);
2949                 cur_offset = last_byte;
2950                 if (err || cur_offset >= block_end)
2951                         break;
2952         }
2953
2954         btrfs_end_transaction(trans, root);
2955         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2956         return err;
2957 }
2958
2959 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2960 {
2961         struct inode *inode = dentry->d_inode;
2962         int err;
2963
2964         err = inode_change_ok(inode, attr);
2965         if (err)
2966                 return err;
2967
2968         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2969                 if (attr->ia_size > inode->i_size) {
2970                         err = btrfs_cont_expand(inode, attr->ia_size);
2971                         if (err)
2972                                 return err;
2973                 } else if (inode->i_size > 0 &&
2974                            attr->ia_size == 0) {
2975
2976                         /* we're truncating a file that used to have good
2977                          * data down to zero.  Make sure it gets into
2978                          * the ordered flush list so that any new writes
2979                          * get down to disk quickly.
2980                          */
2981                         BTRFS_I(inode)->ordered_data_close = 1;
2982                 }
2983         }
2984
2985         err = inode_setattr(inode, attr);
2986
2987         if (!err && ((attr->ia_valid & ATTR_MODE)))
2988                 err = btrfs_acl_chmod(inode);
2989         return err;
2990 }
2991
2992 void btrfs_delete_inode(struct inode *inode)
2993 {
2994         struct btrfs_trans_handle *trans;
2995         struct btrfs_root *root = BTRFS_I(inode)->root;
2996         unsigned long nr;
2997         int ret;
2998
2999         truncate_inode_pages(&inode->i_data, 0);
3000         if (is_bad_inode(inode)) {
3001                 btrfs_orphan_del(NULL, inode);
3002                 goto no_delete;
3003         }
3004         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3005
3006         btrfs_i_size_write(inode, 0);
3007         trans = btrfs_join_transaction(root, 1);
3008
3009         btrfs_set_trans_block_group(trans, inode);
3010         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3011         if (ret) {
3012                 btrfs_orphan_del(NULL, inode);
3013                 goto no_delete_lock;
3014         }
3015
3016         btrfs_orphan_del(trans, inode);
3017
3018         nr = trans->blocks_used;
3019         clear_inode(inode);
3020
3021         btrfs_end_transaction(trans, root);
3022         btrfs_btree_balance_dirty(root, nr);
3023         return;
3024
3025 no_delete_lock:
3026         nr = trans->blocks_used;
3027         btrfs_end_transaction(trans, root);
3028         btrfs_btree_balance_dirty(root, nr);
3029 no_delete:
3030         clear_inode(inode);
3031 }
3032
3033 /*
3034  * this returns the key found in the dir entry in the location pointer.
3035  * If no dir entries were found, location->objectid is 0.
3036  */
3037 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3038                                struct btrfs_key *location)
3039 {
3040         const char *name = dentry->d_name.name;
3041         int namelen = dentry->d_name.len;
3042         struct btrfs_dir_item *di;
3043         struct btrfs_path *path;
3044         struct btrfs_root *root = BTRFS_I(dir)->root;
3045         int ret = 0;
3046
3047         path = btrfs_alloc_path();
3048         BUG_ON(!path);
3049
3050         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3051                                     namelen, 0);
3052         if (IS_ERR(di))
3053                 ret = PTR_ERR(di);
3054
3055         if (!di || IS_ERR(di))
3056                 goto out_err;
3057
3058         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3059 out:
3060         btrfs_free_path(path);
3061         return ret;
3062 out_err:
3063         location->objectid = 0;
3064         goto out;
3065 }
3066
3067 /*
3068  * when we hit a tree root in a directory, the btrfs part of the inode
3069  * needs to be changed to reflect the root directory of the tree root.  This
3070  * is kind of like crossing a mount point.
3071  */
3072 static int fixup_tree_root_location(struct btrfs_root *root,
3073                              struct btrfs_key *location,
3074                              struct btrfs_root **sub_root,
3075                              struct dentry *dentry)
3076 {
3077         struct btrfs_root_item *ri;
3078
3079         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3080                 return 0;
3081         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3082                 return 0;
3083
3084         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3085                                         dentry->d_name.name,
3086                                         dentry->d_name.len);
3087         if (IS_ERR(*sub_root))
3088                 return PTR_ERR(*sub_root);
3089
3090         ri = &(*sub_root)->root_item;
3091         location->objectid = btrfs_root_dirid(ri);
3092         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3093         location->offset = 0;
3094
3095         return 0;
3096 }
3097
3098 static void inode_tree_add(struct inode *inode)
3099 {
3100         struct btrfs_root *root = BTRFS_I(inode)->root;
3101         struct btrfs_inode *entry;
3102         struct rb_node **p;
3103         struct rb_node *parent;
3104
3105 again:
3106         p = &root->inode_tree.rb_node;
3107         parent = NULL;
3108
3109         spin_lock(&root->inode_lock);
3110         while (*p) {
3111                 parent = *p;
3112                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3113
3114                 if (inode->i_ino < entry->vfs_inode.i_ino)
3115                         p = &parent->rb_left;
3116                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3117                         p = &parent->rb_right;
3118                 else {
3119                         WARN_ON(!(entry->vfs_inode.i_state &
3120                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3121                         rb_erase(parent, &root->inode_tree);
3122                         RB_CLEAR_NODE(parent);
3123                         spin_unlock(&root->inode_lock);
3124                         goto again;
3125                 }
3126         }
3127         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3128         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3129         spin_unlock(&root->inode_lock);
3130 }
3131
3132 static void inode_tree_del(struct inode *inode)
3133 {
3134         struct btrfs_root *root = BTRFS_I(inode)->root;
3135
3136         spin_lock(&root->inode_lock);
3137         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3138                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3139                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3140         }
3141         spin_unlock(&root->inode_lock);
3142 }
3143
3144 static noinline void init_btrfs_i(struct inode *inode)
3145 {
3146         struct btrfs_inode *bi = BTRFS_I(inode);
3147
3148         bi->generation = 0;
3149         bi->sequence = 0;
3150         bi->last_trans = 0;
3151         bi->logged_trans = 0;
3152         bi->delalloc_bytes = 0;
3153         bi->reserved_bytes = 0;
3154         bi->disk_i_size = 0;
3155         bi->flags = 0;
3156         bi->index_cnt = (u64)-1;
3157         bi->last_unlink_trans = 0;
3158         bi->ordered_data_close = 0;
3159         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3160         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3161                              inode->i_mapping, GFP_NOFS);
3162         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3163                              inode->i_mapping, GFP_NOFS);
3164         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3165         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3166         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3167         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3168         mutex_init(&BTRFS_I(inode)->extent_mutex);
3169         mutex_init(&BTRFS_I(inode)->log_mutex);
3170 }
3171
3172 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3173 {
3174         struct btrfs_iget_args *args = p;
3175         inode->i_ino = args->ino;
3176         init_btrfs_i(inode);
3177         BTRFS_I(inode)->root = args->root;
3178         btrfs_set_inode_space_info(args->root, inode);
3179         return 0;
3180 }
3181
3182 static int btrfs_find_actor(struct inode *inode, void *opaque)
3183 {
3184         struct btrfs_iget_args *args = opaque;
3185         return args->ino == inode->i_ino &&
3186                 args->root == BTRFS_I(inode)->root;
3187 }
3188
3189 static struct inode *btrfs_iget_locked(struct super_block *s,
3190                                        u64 objectid,
3191                                        struct btrfs_root *root)
3192 {
3193         struct inode *inode;
3194         struct btrfs_iget_args args;
3195         args.ino = objectid;
3196         args.root = root;
3197
3198         inode = iget5_locked(s, objectid, btrfs_find_actor,
3199                              btrfs_init_locked_inode,
3200                              (void *)&args);
3201         return inode;
3202 }
3203
3204 /* Get an inode object given its location and corresponding root.
3205  * Returns in *is_new if the inode was read from disk
3206  */
3207 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3208                          struct btrfs_root *root)
3209 {
3210         struct inode *inode;
3211
3212         inode = btrfs_iget_locked(s, location->objectid, root);
3213         if (!inode)
3214                 return ERR_PTR(-ENOMEM);
3215
3216         if (inode->i_state & I_NEW) {
3217                 BTRFS_I(inode)->root = root;
3218                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3219                 btrfs_read_locked_inode(inode);
3220
3221                 inode_tree_add(inode);
3222                 unlock_new_inode(inode);
3223         }
3224
3225         return inode;
3226 }
3227
3228 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3229 {
3230         struct inode *inode;
3231         struct btrfs_inode *bi = BTRFS_I(dir);
3232         struct btrfs_root *root = bi->root;
3233         struct btrfs_root *sub_root = root;
3234         struct btrfs_key location;
3235         int ret;
3236
3237         if (dentry->d_name.len > BTRFS_NAME_LEN)
3238                 return ERR_PTR(-ENAMETOOLONG);
3239
3240         ret = btrfs_inode_by_name(dir, dentry, &location);
3241
3242         if (ret < 0)
3243                 return ERR_PTR(ret);
3244
3245         inode = NULL;
3246         if (location.objectid) {
3247                 ret = fixup_tree_root_location(root, &location, &sub_root,
3248                                                 dentry);
3249                 if (ret < 0)
3250                         return ERR_PTR(ret);
3251                 if (ret > 0)
3252                         return ERR_PTR(-ENOENT);
3253                 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3254                 if (IS_ERR(inode))
3255                         return ERR_CAST(inode);
3256         }
3257         return inode;
3258 }
3259
3260 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3261                                    struct nameidata *nd)
3262 {
3263         struct inode *inode;
3264
3265         if (dentry->d_name.len > BTRFS_NAME_LEN)
3266                 return ERR_PTR(-ENAMETOOLONG);
3267
3268         inode = btrfs_lookup_dentry(dir, dentry);
3269         if (IS_ERR(inode))
3270                 return ERR_CAST(inode);
3271
3272         return d_splice_alias(inode, dentry);
3273 }
3274
3275 static unsigned char btrfs_filetype_table[] = {
3276         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3277 };
3278
3279 static int btrfs_real_readdir(struct file *filp, void *dirent,
3280                               filldir_t filldir)
3281 {
3282         struct inode *inode = filp->f_dentry->d_inode;
3283         struct btrfs_root *root = BTRFS_I(inode)->root;
3284         struct btrfs_item *item;
3285         struct btrfs_dir_item *di;
3286         struct btrfs_key key;
3287         struct btrfs_key found_key;
3288         struct btrfs_path *path;
3289         int ret;
3290         u32 nritems;
3291         struct extent_buffer *leaf;
3292         int slot;
3293         int advance;
3294         unsigned char d_type;
3295         int over = 0;
3296         u32 di_cur;
3297         u32 di_total;
3298         u32 di_len;
3299         int key_type = BTRFS_DIR_INDEX_KEY;
3300         char tmp_name[32];
3301         char *name_ptr;
3302         int name_len;
3303
3304         /* FIXME, use a real flag for deciding about the key type */
3305         if (root->fs_info->tree_root == root)
3306                 key_type = BTRFS_DIR_ITEM_KEY;
3307
3308         /* special case for "." */
3309         if (filp->f_pos == 0) {
3310                 over = filldir(dirent, ".", 1,
3311                                1, inode->i_ino,
3312                                DT_DIR);
3313                 if (over)
3314                         return 0;
3315                 filp->f_pos = 1;
3316         }
3317         /* special case for .., just use the back ref */
3318         if (filp->f_pos == 1) {
3319                 u64 pino = parent_ino(filp->f_path.dentry);
3320                 over = filldir(dirent, "..", 2,
3321                                2, pino, DT_DIR);
3322                 if (over)
3323                         return 0;
3324                 filp->f_pos = 2;
3325         }
3326         path = btrfs_alloc_path();
3327         path->reada = 2;
3328
3329         btrfs_set_key_type(&key, key_type);
3330         key.offset = filp->f_pos;
3331         key.objectid = inode->i_ino;
3332
3333         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3334         if (ret < 0)
3335                 goto err;
3336         advance = 0;
3337
3338         while (1) {
3339                 leaf = path->nodes[0];
3340                 nritems = btrfs_header_nritems(leaf);
3341                 slot = path->slots[0];
3342                 if (advance || slot >= nritems) {
3343                         if (slot >= nritems - 1) {
3344                                 ret = btrfs_next_leaf(root, path);
3345                                 if (ret)
3346                                         break;
3347                                 leaf = path->nodes[0];
3348                                 nritems = btrfs_header_nritems(leaf);
3349                                 slot = path->slots[0];
3350                         } else {
3351                                 slot++;
3352                                 path->slots[0]++;
3353                         }
3354                 }
3355
3356                 advance = 1;
3357                 item = btrfs_item_nr(leaf, slot);
3358                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3359
3360                 if (found_key.objectid != key.objectid)
3361                         break;
3362                 if (btrfs_key_type(&found_key) != key_type)
3363                         break;
3364                 if (found_key.offset < filp->f_pos)
3365                         continue;
3366
3367                 filp->f_pos = found_key.offset;
3368
3369                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3370                 di_cur = 0;
3371                 di_total = btrfs_item_size(leaf, item);
3372
3373                 while (di_cur < di_total) {
3374                         struct btrfs_key location;
3375
3376                         name_len = btrfs_dir_name_len(leaf, di);
3377                         if (name_len <= sizeof(tmp_name)) {
3378                                 name_ptr = tmp_name;
3379                         } else {
3380                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3381                                 if (!name_ptr) {
3382                                         ret = -ENOMEM;
3383                                         goto err;
3384                                 }
3385                         }
3386                         read_extent_buffer(leaf, name_ptr,
3387                                            (unsigned long)(di + 1), name_len);
3388
3389                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3390                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3391
3392                         /* is this a reference to our own snapshot? If so
3393                          * skip it
3394                          */
3395                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3396                             location.objectid == root->root_key.objectid) {
3397                                 over = 0;
3398                                 goto skip;
3399                         }
3400                         over = filldir(dirent, name_ptr, name_len,
3401                                        found_key.offset, location.objectid,
3402                                        d_type);
3403
3404 skip:
3405                         if (name_ptr != tmp_name)
3406                                 kfree(name_ptr);
3407
3408                         if (over)
3409                                 goto nopos;
3410                         di_len = btrfs_dir_name_len(leaf, di) +
3411                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3412                         di_cur += di_len;
3413                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3414                 }
3415         }
3416
3417         /* Reached end of directory/root. Bump pos past the last item. */
3418         if (key_type == BTRFS_DIR_INDEX_KEY)
3419                 filp->f_pos = INT_LIMIT(off_t);
3420         else
3421                 filp->f_pos++;
3422 nopos:
3423         ret = 0;
3424 err:
3425         btrfs_free_path(path);
3426         return ret;
3427 }
3428
3429 int btrfs_write_inode(struct inode *inode, int wait)
3430 {
3431         struct btrfs_root *root = BTRFS_I(inode)->root;
3432         struct btrfs_trans_handle *trans;
3433         int ret = 0;
3434
3435         if (root->fs_info->btree_inode == inode)
3436                 return 0;
3437
3438         if (wait) {
3439                 trans = btrfs_join_transaction(root, 1);
3440                 btrfs_set_trans_block_group(trans, inode);
3441                 ret = btrfs_commit_transaction(trans, root);
3442         }
3443         return ret;
3444 }
3445
3446 /*
3447  * This is somewhat expensive, updating the tree every time the
3448  * inode changes.  But, it is most likely to find the inode in cache.
3449  * FIXME, needs more benchmarking...there are no reasons other than performance
3450  * to keep or drop this code.
3451  */
3452 void btrfs_dirty_inode(struct inode *inode)
3453 {
3454         struct btrfs_root *root = BTRFS_I(inode)->root;
3455         struct btrfs_trans_handle *trans;
3456
3457         trans = btrfs_join_transaction(root, 1);
3458         btrfs_set_trans_block_group(trans, inode);
3459         btrfs_update_inode(trans, root, inode);
3460         btrfs_end_transaction(trans, root);
3461 }
3462
3463 /*
3464  * find the highest existing sequence number in a directory
3465  * and then set the in-memory index_cnt variable to reflect
3466  * free sequence numbers
3467  */
3468 static int btrfs_set_inode_index_count(struct inode *inode)
3469 {
3470         struct btrfs_root *root = BTRFS_I(inode)->root;
3471         struct btrfs_key key, found_key;
3472         struct btrfs_path *path;
3473         struct extent_buffer *leaf;
3474         int ret;
3475
3476         key.objectid = inode->i_ino;
3477         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3478         key.offset = (u64)-1;
3479
3480         path = btrfs_alloc_path();
3481         if (!path)
3482                 return -ENOMEM;
3483
3484         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3485         if (ret < 0)
3486                 goto out;
3487         /* FIXME: we should be able to handle this */
3488         if (ret == 0)
3489                 goto out;
3490         ret = 0;
3491
3492         /*
3493          * MAGIC NUMBER EXPLANATION:
3494          * since we search a directory based on f_pos we have to start at 2
3495          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3496          * else has to start at 2
3497          */
3498         if (path->slots[0] == 0) {
3499                 BTRFS_I(inode)->index_cnt = 2;
3500                 goto out;
3501         }
3502
3503         path->slots[0]--;
3504
3505         leaf = path->nodes[0];
3506         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3507
3508         if (found_key.objectid != inode->i_ino ||
3509             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3510                 BTRFS_I(inode)->index_cnt = 2;
3511                 goto out;
3512         }
3513
3514         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3515 out:
3516         btrfs_free_path(path);
3517         return ret;
3518 }
3519
3520 /*
3521  * helper to find a free sequence number in a given directory.  This current
3522  * code is very simple, later versions will do smarter things in the btree
3523  */
3524 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3525 {
3526         int ret = 0;
3527
3528         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3529                 ret = btrfs_set_inode_index_count(dir);
3530                 if (ret)
3531                         return ret;
3532         }
3533
3534         *index = BTRFS_I(dir)->index_cnt;
3535         BTRFS_I(dir)->index_cnt++;
3536
3537         return ret;
3538 }
3539
3540 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3541                                      struct btrfs_root *root,
3542                                      struct inode *dir,
3543                                      const char *name, int name_len,
3544                                      u64 ref_objectid, u64 objectid,
3545                                      u64 alloc_hint, int mode, u64 *index)
3546 {
3547         struct inode *inode;
3548         struct btrfs_inode_item *inode_item;
3549         struct btrfs_key *location;
3550         struct btrfs_path *path;
3551         struct btrfs_inode_ref *ref;
3552         struct btrfs_key key[2];
3553         u32 sizes[2];
3554         unsigned long ptr;
3555         int ret;
3556         int owner;
3557
3558         path = btrfs_alloc_path();
3559         BUG_ON(!path);
3560
3561         inode = new_inode(root->fs_info->sb);
3562         if (!inode)
3563                 return ERR_PTR(-ENOMEM);
3564
3565         if (dir) {
3566                 ret = btrfs_set_inode_index(dir, index);
3567                 if (ret) {
3568                         iput(inode);
3569                         return ERR_PTR(ret);
3570                 }
3571         }
3572         /*
3573          * index_cnt is ignored for everything but a dir,
3574          * btrfs_get_inode_index_count has an explanation for the magic
3575          * number
3576          */
3577         init_btrfs_i(inode);
3578         BTRFS_I(inode)->index_cnt = 2;
3579         BTRFS_I(inode)->root = root;
3580         BTRFS_I(inode)->generation = trans->transid;
3581         btrfs_set_inode_space_info(root, inode);
3582
3583         if (mode & S_IFDIR)
3584                 owner = 0;
3585         else
3586                 owner = 1;
3587         BTRFS_I(inode)->block_group =
3588                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3589
3590         key[0].objectid = objectid;
3591         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3592         key[0].offset = 0;
3593
3594         key[1].objectid = objectid;
3595         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3596         key[1].offset = ref_objectid;
3597
3598         sizes[0] = sizeof(struct btrfs_inode_item);
3599         sizes[1] = name_len + sizeof(*ref);
3600
3601         path->leave_spinning = 1;
3602         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3603         if (ret != 0)
3604                 goto fail;
3605
3606         if (objectid > root->highest_inode)
3607                 root->highest_inode = objectid;
3608
3609         inode->i_uid = current_fsuid();
3610
3611         if (dir && (dir->i_mode & S_ISGID)) {
3612                 inode->i_gid = dir->i_gid;
3613                 if (S_ISDIR(mode))
3614                         mode |= S_ISGID;
3615         } else
3616                 inode->i_gid = current_fsgid();
3617
3618         inode->i_mode = mode;
3619         inode->i_ino = objectid;
3620         inode_set_bytes(inode, 0);
3621         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3622         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3623                                   struct btrfs_inode_item);
3624         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3625
3626         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3627                              struct btrfs_inode_ref);
3628         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3629         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3630         ptr = (unsigned long)(ref + 1);
3631         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3632
3633         btrfs_mark_buffer_dirty(path->nodes[0]);
3634         btrfs_free_path(path);
3635
3636         location = &BTRFS_I(inode)->location;
3637         location->objectid = objectid;
3638         location->offset = 0;
3639         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3640
3641         btrfs_inherit_iflags(inode, dir);
3642
3643         if ((mode & S_IFREG)) {
3644                 if (btrfs_test_opt(root, NODATASUM))
3645                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
3646                 if (btrfs_test_opt(root, NODATACOW))
3647                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
3648         }
3649
3650         insert_inode_hash(inode);
3651         inode_tree_add(inode);
3652         return inode;
3653 fail:
3654         if (dir)
3655                 BTRFS_I(dir)->index_cnt--;
3656         btrfs_free_path(path);
3657         iput(inode);
3658         return ERR_PTR(ret);
3659 }
3660
3661 static inline u8 btrfs_inode_type(struct inode *inode)
3662 {
3663         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3664 }
3665
3666 /*
3667  * utility function to add 'inode' into 'parent_inode' with
3668  * a give name and a given sequence number.
3669  * if 'add_backref' is true, also insert a backref from the
3670  * inode to the parent directory.
3671  */
3672 int btrfs_add_link(struct btrfs_trans_handle *trans,
3673                    struct inode *parent_inode, struct inode *inode,
3674                    const char *name, int name_len, int add_backref, u64 index)
3675 {
3676         int ret;
3677         struct btrfs_key key;
3678         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3679
3680         key.objectid = inode->i_ino;
3681         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3682         key.offset = 0;
3683
3684         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3685                                     parent_inode->i_ino,
3686                                     &key, btrfs_inode_type(inode),
3687                                     index);
3688         if (ret == 0) {
3689                 if (add_backref) {
3690                         ret = btrfs_insert_inode_ref(trans, root,
3691                                                      name, name_len,
3692                                                      inode->i_ino,
3693                                                      parent_inode->i_ino,
3694                                                      index);
3695                 }
3696                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3697                                    name_len * 2);
3698                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3699                 ret = btrfs_update_inode(trans, root, parent_inode);
3700         }
3701         return ret;
3702 }
3703
3704 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3705                             struct dentry *dentry, struct inode *inode,
3706                             int backref, u64 index)
3707 {
3708         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3709                                  inode, dentry->d_name.name,
3710                                  dentry->d_name.len, backref, index);
3711         if (!err) {
3712                 d_instantiate(dentry, inode);
3713                 return 0;
3714         }
3715         if (err > 0)
3716                 err = -EEXIST;
3717         return err;
3718 }
3719
3720 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3721                         int mode, dev_t rdev)
3722 {
3723         struct btrfs_trans_handle *trans;
3724         struct btrfs_root *root = BTRFS_I(dir)->root;
3725         struct inode *inode = NULL;
3726         int err;
3727         int drop_inode = 0;
3728         u64 objectid;
3729         unsigned long nr = 0;
3730         u64 index = 0;
3731
3732         if (!new_valid_dev(rdev))
3733                 return -EINVAL;
3734
3735         err = btrfs_check_metadata_free_space(root);
3736         if (err)
3737                 goto fail;
3738
3739         trans = btrfs_start_transaction(root, 1);
3740         btrfs_set_trans_block_group(trans, dir);
3741
3742         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3743         if (err) {
3744                 err = -ENOSPC;
3745                 goto out_unlock;
3746         }
3747
3748         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3749                                 dentry->d_name.len,
3750                                 dentry->d_parent->d_inode->i_ino, objectid,
3751                                 BTRFS_I(dir)->block_group, mode, &index);
3752         err = PTR_ERR(inode);
3753         if (IS_ERR(inode))
3754                 goto out_unlock;
3755
3756         err = btrfs_init_inode_security(inode, dir);
3757         if (err) {
3758                 drop_inode = 1;
3759                 goto out_unlock;
3760         }
3761
3762         btrfs_set_trans_block_group(trans, inode);
3763         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3764         if (err)
3765                 drop_inode = 1;
3766         else {
3767                 inode->i_op = &btrfs_special_inode_operations;
3768                 init_special_inode(inode, inode->i_mode, rdev);
3769                 btrfs_update_inode(trans, root, inode);
3770         }
3771         btrfs_update_inode_block_group(trans, inode);
3772         btrfs_update_inode_block_group(trans, dir);
3773 out_unlock:
3774         nr = trans->blocks_used;
3775         btrfs_end_transaction_throttle(trans, root);
3776 fail:
3777         if (drop_inode) {
3778                 inode_dec_link_count(inode);
3779                 iput(inode);
3780         }
3781         btrfs_btree_balance_dirty(root, nr);
3782         return err;
3783 }
3784
3785 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3786                         int mode, struct nameidata *nd)
3787 {
3788         struct btrfs_trans_handle *trans;
3789         struct btrfs_root *root = BTRFS_I(dir)->root;
3790         struct inode *inode = NULL;
3791         int err;
3792         int drop_inode = 0;
3793         unsigned long nr = 0;
3794         u64 objectid;
3795         u64 index = 0;
3796
3797         err = btrfs_check_metadata_free_space(root);
3798         if (err)
3799                 goto fail;
3800         trans = btrfs_start_transaction(root, 1);
3801         btrfs_set_trans_block_group(trans, dir);
3802
3803         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3804         if (err) {
3805                 err = -ENOSPC;
3806                 goto out_unlock;
3807         }
3808
3809         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3810                                 dentry->d_name.len,
3811                                 dentry->d_parent->d_inode->i_ino,
3812                                 objectid, BTRFS_I(dir)->block_group, mode,
3813                                 &index);
3814         err = PTR_ERR(inode);
3815         if (IS_ERR(inode))
3816                 goto out_unlock;
3817
3818         err = btrfs_init_inode_security(inode, dir);
3819         if (err) {
3820                 drop_inode = 1;
3821                 goto out_unlock;
3822         }
3823
3824         btrfs_set_trans_block_group(trans, inode);
3825         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3826         if (err)
3827                 drop_inode = 1;
3828         else {
3829                 inode->i_mapping->a_ops = &btrfs_aops;
3830                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3831                 inode->i_fop = &btrfs_file_operations;
3832                 inode->i_op = &btrfs_file_inode_operations;
3833                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3834         }
3835         btrfs_update_inode_block_group(trans, inode);
3836         btrfs_update_inode_block_group(trans, dir);
3837 out_unlock:
3838         nr = trans->blocks_used;
3839         btrfs_end_transaction_throttle(trans, root);
3840 fail:
3841         if (drop_inode) {
3842                 inode_dec_link_count(inode);
3843                 iput(inode);
3844         }
3845         btrfs_btree_balance_dirty(root, nr);
3846         return err;
3847 }
3848
3849 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3850                       struct dentry *dentry)
3851 {
3852         struct btrfs_trans_handle *trans;
3853         struct btrfs_root *root = BTRFS_I(dir)->root;
3854         struct inode *inode = old_dentry->d_inode;
3855         u64 index;
3856         unsigned long nr = 0;
3857         int err;
3858         int drop_inode = 0;
3859
3860         if (inode->i_nlink == 0)
3861                 return -ENOENT;
3862
3863         btrfs_inc_nlink(inode);
3864         err = btrfs_check_metadata_free_space(root);
3865         if (err)
3866                 goto fail;
3867         err = btrfs_set_inode_index(dir, &index);
3868         if (err)
3869                 goto fail;
3870
3871         trans = btrfs_start_transaction(root, 1);
3872
3873         btrfs_set_trans_block_group(trans, dir);
3874         atomic_inc(&inode->i_count);
3875
3876         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3877
3878         if (err)
3879                 drop_inode = 1;
3880
3881         btrfs_update_inode_block_group(trans, dir);
3882         err = btrfs_update_inode(trans, root, inode);
3883
3884         if (err)
3885                 drop_inode = 1;
3886
3887         nr = trans->blocks_used;
3888
3889         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3890         btrfs_end_transaction_throttle(trans, root);
3891 fail:
3892         if (drop_inode) {
3893                 inode_dec_link_count(inode);
3894                 iput(inode);
3895         }
3896         btrfs_btree_balance_dirty(root, nr);
3897         return err;
3898 }
3899
3900 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3901 {
3902         struct inode *inode = NULL;
3903         struct btrfs_trans_handle *trans;
3904         struct btrfs_root *root = BTRFS_I(dir)->root;
3905         int err = 0;
3906         int drop_on_err = 0;
3907         u64 objectid = 0;
3908         u64 index = 0;
3909         unsigned long nr = 1;
3910
3911         err = btrfs_check_metadata_free_space(root);
3912         if (err)
3913                 goto out_unlock;
3914
3915         trans = btrfs_start_transaction(root, 1);
3916         btrfs_set_trans_block_group(trans, dir);
3917
3918         if (IS_ERR(trans)) {
3919                 err = PTR_ERR(trans);
3920                 goto out_unlock;
3921         }
3922
3923         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3924         if (err) {
3925                 err = -ENOSPC;
3926                 goto out_unlock;
3927         }
3928
3929         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3930                                 dentry->d_name.len,
3931                                 dentry->d_parent->d_inode->i_ino, objectid,
3932                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3933                                 &index);
3934         if (IS_ERR(inode)) {
3935                 err = PTR_ERR(inode);
3936                 goto out_fail;
3937         }
3938
3939         drop_on_err = 1;
3940
3941         err = btrfs_init_inode_security(inode, dir);
3942         if (err)
3943                 goto out_fail;
3944
3945         inode->i_op = &btrfs_dir_inode_operations;
3946         inode->i_fop = &btrfs_dir_file_operations;
3947         btrfs_set_trans_block_group(trans, inode);
3948
3949         btrfs_i_size_write(inode, 0);
3950         err = btrfs_update_inode(trans, root, inode);
3951         if (err)
3952                 goto out_fail;
3953
3954         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3955                                  inode, dentry->d_name.name,
3956                                  dentry->d_name.len, 0, index);
3957         if (err)
3958                 goto out_fail;
3959
3960         d_instantiate(dentry, inode);
3961         drop_on_err = 0;
3962         btrfs_update_inode_block_group(trans, inode);
3963         btrfs_update_inode_block_group(trans, dir);
3964
3965 out_fail:
3966         nr = trans->blocks_used;
3967         btrfs_end_transaction_throttle(trans, root);
3968
3969 out_unlock:
3970         if (drop_on_err)
3971                 iput(inode);
3972         btrfs_btree_balance_dirty(root, nr);
3973         return err;
3974 }
3975
3976 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3977  * and an extent that you want to insert, deal with overlap and insert
3978  * the new extent into the tree.
3979  */
3980 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3981                                 struct extent_map *existing,
3982                                 struct extent_map *em,
3983                                 u64 map_start, u64 map_len)
3984 {
3985         u64 start_diff;
3986
3987         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3988         start_diff = map_start - em->start;
3989         em->start = map_start;
3990         em->len = map_len;
3991         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3992             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3993                 em->block_start += start_diff;
3994                 em->block_len -= start_diff;
3995         }
3996         return add_extent_mapping(em_tree, em);
3997 }
3998
3999 static noinline int uncompress_inline(struct btrfs_path *path,
4000                                       struct inode *inode, struct page *page,
4001                                       size_t pg_offset, u64 extent_offset,
4002                                       struct btrfs_file_extent_item *item)
4003 {
4004         int ret;
4005         struct extent_buffer *leaf = path->nodes[0];
4006         char *tmp;
4007         size_t max_size;
4008         unsigned long inline_size;
4009         unsigned long ptr;
4010
4011         WARN_ON(pg_offset != 0);
4012         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4013         inline_size = btrfs_file_extent_inline_item_len(leaf,
4014                                         btrfs_item_nr(leaf, path->slots[0]));
4015         tmp = kmalloc(inline_size, GFP_NOFS);
4016         ptr = btrfs_file_extent_inline_start(item);
4017
4018         read_extent_buffer(leaf, tmp, ptr, inline_size);
4019
4020         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4021         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4022                                     inline_size, max_size);
4023         if (ret) {
4024                 char *kaddr = kmap_atomic(page, KM_USER0);
4025                 unsigned long copy_size = min_t(u64,
4026                                   PAGE_CACHE_SIZE - pg_offset,
4027                                   max_size - extent_offset);
4028                 memset(kaddr + pg_offset, 0, copy_size);
4029                 kunmap_atomic(kaddr, KM_USER0);
4030         }
4031         kfree(tmp);
4032         return 0;
4033 }
4034
4035 /*
4036  * a bit scary, this does extent mapping from logical file offset to the disk.
4037  * the ugly parts come from merging extents from the disk with the in-ram
4038  * representation.  This gets more complex because of the data=ordered code,
4039  * where the in-ram extents might be locked pending data=ordered completion.
4040  *
4041  * This also copies inline extents directly into the page.
4042  */
4043
4044 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4045                                     size_t pg_offset, u64 start, u64 len,
4046                                     int create)
4047 {
4048         int ret;
4049         int err = 0;
4050         u64 bytenr;
4051         u64 extent_start = 0;
4052         u64 extent_end = 0;
4053         u64 objectid = inode->i_ino;
4054         u32 found_type;
4055         struct btrfs_path *path = NULL;
4056         struct btrfs_root *root = BTRFS_I(inode)->root;
4057         struct btrfs_file_extent_item *item;
4058         struct extent_buffer *leaf;
4059         struct btrfs_key found_key;
4060         struct extent_map *em = NULL;
4061         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4062         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4063         struct btrfs_trans_handle *trans = NULL;
4064         int compressed;
4065
4066 again:
4067         spin_lock(&em_tree->lock);
4068         em = lookup_extent_mapping(em_tree, start, len);
4069         if (em)
4070                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4071         spin_unlock(&em_tree->lock);
4072
4073         if (em) {
4074                 if (em->start > start || em->start + em->len <= start)
4075                         free_extent_map(em);
4076                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4077                         free_extent_map(em);
4078                 else
4079                         goto out;
4080         }
4081         em = alloc_extent_map(GFP_NOFS);
4082         if (!em) {
4083                 err = -ENOMEM;
4084                 goto out;
4085         }
4086         em->bdev = root->fs_info->fs_devices->latest_bdev;
4087         em->start = EXTENT_MAP_HOLE;
4088         em->orig_start = EXTENT_MAP_HOLE;
4089         em->len = (u64)-1;
4090         em->block_len = (u64)-1;
4091
4092         if (!path) {
4093                 path = btrfs_alloc_path();
4094                 BUG_ON(!path);
4095         }
4096
4097         ret = btrfs_lookup_file_extent(trans, root, path,
4098                                        objectid, start, trans != NULL);
4099         if (ret < 0) {
4100                 err = ret;
4101                 goto out;
4102         }
4103
4104         if (ret != 0) {
4105                 if (path->slots[0] == 0)
4106                         goto not_found;
4107                 path->slots[0]--;
4108         }
4109
4110         leaf = path->nodes[0];
4111         item = btrfs_item_ptr(leaf, path->slots[0],
4112                               struct btrfs_file_extent_item);
4113         /* are we inside the extent that was found? */
4114         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4115         found_type = btrfs_key_type(&found_key);
4116         if (found_key.objectid != objectid ||
4117             found_type != BTRFS_EXTENT_DATA_KEY) {
4118                 goto not_found;
4119         }
4120
4121         found_type = btrfs_file_extent_type(leaf, item);
4122         extent_start = found_key.offset;
4123         compressed = btrfs_file_extent_compression(leaf, item);
4124         if (found_type == BTRFS_FILE_EXTENT_REG ||
4125             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4126                 extent_end = extent_start +
4127                        btrfs_file_extent_num_bytes(leaf, item);
4128         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4129                 size_t size;
4130                 size = btrfs_file_extent_inline_len(leaf, item);
4131                 extent_end = (extent_start + size + root->sectorsize - 1) &
4132                         ~((u64)root->sectorsize - 1);
4133         }
4134
4135         if (start >= extent_end) {
4136                 path->slots[0]++;
4137                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4138                         ret = btrfs_next_leaf(root, path);
4139                         if (ret < 0) {
4140                                 err = ret;
4141                                 goto out;
4142                         }
4143                         if (ret > 0)
4144                                 goto not_found;
4145                         leaf = path->nodes[0];
4146                 }
4147                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4148                 if (found_key.objectid != objectid ||
4149                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4150                         goto not_found;
4151                 if (start + len <= found_key.offset)
4152                         goto not_found;
4153                 em->start = start;
4154                 em->len = found_key.offset - start;
4155                 goto not_found_em;
4156         }
4157
4158         if (found_type == BTRFS_FILE_EXTENT_REG ||
4159             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4160                 em->start = extent_start;
4161                 em->len = extent_end - extent_start;
4162                 em->orig_start = extent_start -
4163                                  btrfs_file_extent_offset(leaf, item);
4164                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4165                 if (bytenr == 0) {
4166                         em->block_start = EXTENT_MAP_HOLE;
4167                         goto insert;
4168                 }
4169                 if (compressed) {
4170                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4171                         em->block_start = bytenr;
4172                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4173                                                                          item);
4174                 } else {
4175                         bytenr += btrfs_file_extent_offset(leaf, item);
4176                         em->block_start = bytenr;
4177                         em->block_len = em->len;
4178                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4179                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4180                 }
4181                 goto insert;
4182         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4183                 unsigned long ptr;
4184                 char *map;
4185                 size_t size;
4186                 size_t extent_offset;
4187                 size_t copy_size;
4188
4189                 em->block_start = EXTENT_MAP_INLINE;
4190                 if (!page || create) {
4191                         em->start = extent_start;
4192                         em->len = extent_end - extent_start;
4193                         goto out;
4194                 }
4195
4196                 size = btrfs_file_extent_inline_len(leaf, item);
4197                 extent_offset = page_offset(page) + pg_offset - extent_start;
4198                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4199                                 size - extent_offset);
4200                 em->start = extent_start + extent_offset;
4201                 em->len = (copy_size + root->sectorsize - 1) &
4202                         ~((u64)root->sectorsize - 1);
4203                 em->orig_start = EXTENT_MAP_INLINE;
4204                 if (compressed)
4205                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4206                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4207                 if (create == 0 && !PageUptodate(page)) {
4208                         if (btrfs_file_extent_compression(leaf, item) ==
4209                             BTRFS_COMPRESS_ZLIB) {
4210                                 ret = uncompress_inline(path, inode, page,
4211                                                         pg_offset,
4212                                                         extent_offset, item);
4213                                 BUG_ON(ret);
4214                         } else {
4215                                 map = kmap(page);
4216                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4217                                                    copy_size);
4218                                 kunmap(page);
4219                         }
4220                         flush_dcache_page(page);
4221                 } else if (create && PageUptodate(page)) {
4222                         if (!trans) {
4223                                 kunmap(page);
4224                                 free_extent_map(em);
4225                                 em = NULL;
4226                                 btrfs_release_path(root, path);
4227                                 trans = btrfs_join_transaction(root, 1);
4228                                 goto again;
4229                         }
4230                         map = kmap(page);
4231                         write_extent_buffer(leaf, map + pg_offset, ptr,
4232                                             copy_size);
4233                         kunmap(page);
4234                         btrfs_mark_buffer_dirty(leaf);
4235                 }
4236                 set_extent_uptodate(io_tree, em->start,
4237                                     extent_map_end(em) - 1, GFP_NOFS);
4238                 goto insert;
4239         } else {
4240                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4241                 WARN_ON(1);
4242         }
4243 not_found:
4244         em->start = start;
4245         em->len = len;
4246 not_found_em:
4247         em->block_start = EXTENT_MAP_HOLE;
4248         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4249 insert:
4250         btrfs_release_path(root, path);
4251         if (em->start > start || extent_map_end(em) <= start) {
4252                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4253                        "[%llu %llu]\n", (unsigned long long)em->start,
4254                        (unsigned long long)em->len,
4255                        (unsigned long long)start,
4256                        (unsigned long long)len);
4257                 err = -EIO;
4258                 goto out;
4259         }
4260
4261         err = 0;
4262         spin_lock(&em_tree->lock);
4263         ret = add_extent_mapping(em_tree, em);
4264         /* it is possible that someone inserted the extent into the tree
4265          * while we had the lock dropped.  It is also possible that
4266          * an overlapping map exists in the tree
4267          */
4268         if (ret == -EEXIST) {
4269                 struct extent_map *existing;
4270
4271                 ret = 0;
4272
4273                 existing = lookup_extent_mapping(em_tree, start, len);
4274                 if (existing && (existing->start > start ||
4275                     existing->start + existing->len <= start)) {
4276                         free_extent_map(existing);
4277                         existing = NULL;
4278                 }
4279                 if (!existing) {
4280                         existing = lookup_extent_mapping(em_tree, em->start,
4281                                                          em->len);
4282                         if (existing) {
4283                                 err = merge_extent_mapping(em_tree, existing,
4284                                                            em, start,
4285                                                            root->sectorsize);
4286                                 free_extent_map(existing);
4287                                 if (err) {
4288                                         free_extent_map(em);
4289                                         em = NULL;
4290                                 }
4291                         } else {
4292                                 err = -EIO;
4293                                 free_extent_map(em);
4294                                 em = NULL;
4295                         }
4296                 } else {
4297                         free_extent_map(em);
4298                         em = existing;
4299                         err = 0;
4300                 }
4301         }
4302         spin_unlock(&em_tree->lock);
4303 out:
4304         if (path)
4305                 btrfs_free_path(path);
4306         if (trans) {
4307                 ret = btrfs_end_transaction(trans, root);
4308                 if (!err)
4309                         err = ret;
4310         }
4311         if (err) {
4312                 free_extent_map(em);
4313                 return ERR_PTR(err);
4314         }
4315         return em;
4316 }
4317
4318 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4319                         const struct iovec *iov, loff_t offset,
4320                         unsigned long nr_segs)
4321 {
4322         return -EINVAL;
4323 }
4324
4325 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4326                 __u64 start, __u64 len)
4327 {
4328         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4329 }
4330
4331 int btrfs_readpage(struct file *file, struct page *page)
4332 {
4333         struct extent_io_tree *tree;
4334         tree = &BTRFS_I(page->mapping->host)->io_tree;
4335         return extent_read_full_page(tree, page, btrfs_get_extent);
4336 }
4337
4338 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4339 {
4340         struct extent_io_tree *tree;
4341
4342
4343         if (current->flags & PF_MEMALLOC) {
4344                 redirty_page_for_writepage(wbc, page);
4345                 unlock_page(page);
4346                 return 0;
4347         }
4348         tree = &BTRFS_I(page->mapping->host)->io_tree;
4349         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4350 }
4351
4352 int btrfs_writepages(struct address_space *mapping,
4353                      struct writeback_control *wbc)
4354 {
4355         struct extent_io_tree *tree;
4356
4357         tree = &BTRFS_I(mapping->host)->io_tree;
4358         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4359 }
4360
4361 static int
4362 btrfs_readpages(struct file *file, struct address_space *mapping,
4363                 struct list_head *pages, unsigned nr_pages)
4364 {
4365         struct extent_io_tree *tree;
4366         tree = &BTRFS_I(mapping->host)->io_tree;
4367         return extent_readpages(tree, mapping, pages, nr_pages,
4368                                 btrfs_get_extent);
4369 }
4370 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4371 {
4372         struct extent_io_tree *tree;
4373         struct extent_map_tree *map;
4374         int ret;
4375
4376         tree = &BTRFS_I(page->mapping->host)->io_tree;
4377         map = &BTRFS_I(page->mapping->host)->extent_tree;
4378         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4379         if (ret == 1) {
4380                 ClearPagePrivate(page);
4381                 set_page_private(page, 0);
4382                 page_cache_release(page);
4383         }
4384         return ret;
4385 }
4386
4387 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4388 {
4389         if (PageWriteback(page) || PageDirty(page))
4390                 return 0;
4391         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4392 }
4393
4394 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4395 {
4396         struct extent_io_tree *tree;
4397         struct btrfs_ordered_extent *ordered;
4398         u64 page_start = page_offset(page);
4399         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4400
4401         wait_on_page_writeback(page);
4402         tree = &BTRFS_I(page->mapping->host)->io_tree;
4403         if (offset) {
4404                 btrfs_releasepage(page, GFP_NOFS);
4405                 return;
4406         }
4407
4408         lock_extent(tree, page_start, page_end, GFP_NOFS);
4409         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4410                                            page_offset(page));
4411         if (ordered) {
4412                 /*
4413                  * IO on this page will never be started, so we need
4414                  * to account for any ordered extents now
4415                  */
4416                 clear_extent_bit(tree, page_start, page_end,
4417                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4418                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4419                 btrfs_finish_ordered_io(page->mapping->host,
4420                                         page_start, page_end);
4421                 btrfs_put_ordered_extent(ordered);
4422                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4423         }
4424         clear_extent_bit(tree, page_start, page_end,
4425                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4426                  EXTENT_ORDERED,
4427                  1, 1, GFP_NOFS);
4428         __btrfs_releasepage(page, GFP_NOFS);
4429
4430         ClearPageChecked(page);
4431         if (PagePrivate(page)) {
4432                 ClearPagePrivate(page);
4433                 set_page_private(page, 0);
4434                 page_cache_release(page);
4435         }
4436 }
4437
4438 /*
4439  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4440  * called from a page fault handler when a page is first dirtied. Hence we must
4441  * be careful to check for EOF conditions here. We set the page up correctly
4442  * for a written page which means we get ENOSPC checking when writing into
4443  * holes and correct delalloc and unwritten extent mapping on filesystems that
4444  * support these features.
4445  *
4446  * We are not allowed to take the i_mutex here so we have to play games to
4447  * protect against truncate races as the page could now be beyond EOF.  Because
4448  * vmtruncate() writes the inode size before removing pages, once we have the
4449  * page lock we can determine safely if the page is beyond EOF. If it is not
4450  * beyond EOF, then the page is guaranteed safe against truncation until we
4451  * unlock the page.
4452  */
4453 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4454 {
4455         struct page *page = vmf->page;
4456         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4457         struct btrfs_root *root = BTRFS_I(inode)->root;
4458         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4459         struct btrfs_ordered_extent *ordered;
4460         char *kaddr;
4461         unsigned long zero_start;
4462         loff_t size;
4463         int ret;
4464         u64 page_start;
4465         u64 page_end;
4466
4467         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4468         if (ret) {
4469                 if (ret == -ENOMEM)
4470                         ret = VM_FAULT_OOM;
4471                 else /* -ENOSPC, -EIO, etc */
4472                         ret = VM_FAULT_SIGBUS;
4473                 goto out;
4474         }
4475
4476         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4477 again:
4478         lock_page(page);
4479         size = i_size_read(inode);
4480         page_start = page_offset(page);
4481         page_end = page_start + PAGE_CACHE_SIZE - 1;
4482
4483         if ((page->mapping != inode->i_mapping) ||
4484             (page_start >= size)) {
4485                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4486                 /* page got truncated out from underneath us */
4487                 goto out_unlock;
4488         }
4489         wait_on_page_writeback(page);
4490
4491         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4492         set_page_extent_mapped(page);
4493
4494         /*
4495          * we can't set the delalloc bits if there are pending ordered
4496          * extents.  Drop our locks and wait for them to finish
4497          */
4498         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4499         if (ordered) {
4500                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4501                 unlock_page(page);
4502                 btrfs_start_ordered_extent(inode, ordered, 1);
4503                 btrfs_put_ordered_extent(ordered);
4504                 goto again;
4505         }
4506
4507         btrfs_set_extent_delalloc(inode, page_start, page_end);
4508         ret = 0;
4509
4510         /* page is wholly or partially inside EOF */
4511         if (page_start + PAGE_CACHE_SIZE > size)
4512                 zero_start = size & ~PAGE_CACHE_MASK;
4513         else
4514                 zero_start = PAGE_CACHE_SIZE;
4515
4516         if (zero_start != PAGE_CACHE_SIZE) {
4517                 kaddr = kmap(page);
4518                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4519                 flush_dcache_page(page);
4520                 kunmap(page);
4521         }
4522         ClearPageChecked(page);
4523         set_page_dirty(page);
4524
4525         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4526         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4527
4528 out_unlock:
4529         unlock_page(page);
4530 out:
4531         return ret;
4532 }
4533
4534 static void btrfs_truncate(struct inode *inode)
4535 {
4536         struct btrfs_root *root = BTRFS_I(inode)->root;
4537         int ret;
4538         struct btrfs_trans_handle *trans;
4539         unsigned long nr;
4540         u64 mask = root->sectorsize - 1;
4541
4542         if (!S_ISREG(inode->i_mode))
4543                 return;
4544         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4545                 return;
4546
4547         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4548         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4549
4550         trans = btrfs_start_transaction(root, 1);
4551
4552         /*
4553          * setattr is responsible for setting the ordered_data_close flag,
4554          * but that is only tested during the last file release.  That
4555          * could happen well after the next commit, leaving a great big
4556          * window where new writes may get lost if someone chooses to write
4557          * to this file after truncating to zero
4558          *
4559          * The inode doesn't have any dirty data here, and so if we commit
4560          * this is a noop.  If someone immediately starts writing to the inode
4561          * it is very likely we'll catch some of their writes in this
4562          * transaction, and the commit will find this file on the ordered
4563          * data list with good things to send down.
4564          *
4565          * This is a best effort solution, there is still a window where
4566          * using truncate to replace the contents of the file will
4567          * end up with a zero length file after a crash.
4568          */
4569         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4570                 btrfs_add_ordered_operation(trans, root, inode);
4571
4572         btrfs_set_trans_block_group(trans, inode);
4573         btrfs_i_size_write(inode, inode->i_size);
4574
4575         ret = btrfs_orphan_add(trans, inode);
4576         if (ret)
4577                 goto out;
4578         /* FIXME, add redo link to tree so we don't leak on crash */
4579         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4580                                       BTRFS_EXTENT_DATA_KEY);
4581         btrfs_update_inode(trans, root, inode);
4582
4583         ret = btrfs_orphan_del(trans, inode);
4584         BUG_ON(ret);
4585
4586 out:
4587         nr = trans->blocks_used;
4588         ret = btrfs_end_transaction_throttle(trans, root);
4589         BUG_ON(ret);
4590         btrfs_btree_balance_dirty(root, nr);
4591 }
4592
4593 /*
4594  * create a new subvolume directory/inode (helper for the ioctl).
4595  */
4596 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4597                              struct btrfs_root *new_root, struct dentry *dentry,
4598                              u64 new_dirid, u64 alloc_hint)
4599 {
4600         struct inode *inode;
4601         int error;
4602         u64 index = 0;
4603
4604         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4605                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4606         if (IS_ERR(inode))
4607                 return PTR_ERR(inode);
4608         inode->i_op = &btrfs_dir_inode_operations;
4609         inode->i_fop = &btrfs_dir_file_operations;
4610
4611         inode->i_nlink = 1;
4612         btrfs_i_size_write(inode, 0);
4613
4614         error = btrfs_update_inode(trans, new_root, inode);
4615         if (error)
4616                 return error;
4617
4618         d_instantiate(dentry, inode);
4619         return 0;
4620 }
4621
4622 /* helper function for file defrag and space balancing.  This
4623  * forces readahead on a given range of bytes in an inode
4624  */
4625 unsigned long btrfs_force_ra(struct address_space *mapping,
4626                               struct file_ra_state *ra, struct file *file,
4627                               pgoff_t offset, pgoff_t last_index)
4628 {
4629         pgoff_t req_size = last_index - offset + 1;
4630
4631         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4632         return offset + req_size;
4633 }
4634
4635 struct inode *btrfs_alloc_inode(struct super_block *sb)
4636 {
4637         struct btrfs_inode *ei;
4638
4639         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4640         if (!ei)
4641                 return NULL;
4642         ei->last_trans = 0;
4643         ei->logged_trans = 0;
4644         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4645         INIT_LIST_HEAD(&ei->i_orphan);
4646         INIT_LIST_HEAD(&ei->ordered_operations);
4647         return &ei->vfs_inode;
4648 }
4649
4650 void btrfs_destroy_inode(struct inode *inode)
4651 {
4652         struct btrfs_ordered_extent *ordered;
4653         struct btrfs_root *root = BTRFS_I(inode)->root;
4654
4655         WARN_ON(!list_empty(&inode->i_dentry));
4656         WARN_ON(inode->i_data.nrpages);
4657
4658         /*
4659          * Make sure we're properly removed from the ordered operation
4660          * lists.
4661          */
4662         smp_mb();
4663         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4664                 spin_lock(&root->fs_info->ordered_extent_lock);
4665                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4666                 spin_unlock(&root->fs_info->ordered_extent_lock);
4667         }
4668
4669         spin_lock(&root->list_lock);
4670         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4671                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4672                        " list\n", inode->i_ino);
4673                 dump_stack();
4674         }
4675         spin_unlock(&root->list_lock);
4676
4677         while (1) {
4678                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4679                 if (!ordered)
4680                         break;
4681                 else {
4682                         printk(KERN_ERR "btrfs found ordered "
4683                                "extent %llu %llu on inode cleanup\n",
4684                                (unsigned long long)ordered->file_offset,
4685                                (unsigned long long)ordered->len);
4686                         btrfs_remove_ordered_extent(inode, ordered);
4687                         btrfs_put_ordered_extent(ordered);
4688                         btrfs_put_ordered_extent(ordered);
4689                 }
4690         }
4691         inode_tree_del(inode);
4692         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4693         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4694 }
4695
4696 static void init_once(void *foo)
4697 {
4698         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4699
4700         inode_init_once(&ei->vfs_inode);
4701 }
4702
4703 void btrfs_destroy_cachep(void)
4704 {
4705         if (btrfs_inode_cachep)
4706                 kmem_cache_destroy(btrfs_inode_cachep);
4707         if (btrfs_trans_handle_cachep)
4708                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4709         if (btrfs_transaction_cachep)
4710                 kmem_cache_destroy(btrfs_transaction_cachep);
4711         if (btrfs_path_cachep)
4712                 kmem_cache_destroy(btrfs_path_cachep);
4713 }
4714
4715 int btrfs_init_cachep(void)
4716 {
4717         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4718                         sizeof(struct btrfs_inode), 0,
4719                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4720         if (!btrfs_inode_cachep)
4721                 goto fail;
4722
4723         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4724                         sizeof(struct btrfs_trans_handle), 0,
4725                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4726         if (!btrfs_trans_handle_cachep)
4727                 goto fail;
4728
4729         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4730                         sizeof(struct btrfs_transaction), 0,
4731                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4732         if (!btrfs_transaction_cachep)
4733                 goto fail;
4734
4735         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4736                         sizeof(struct btrfs_path), 0,
4737                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4738         if (!btrfs_path_cachep)
4739                 goto fail;
4740
4741         return 0;
4742 fail:
4743         btrfs_destroy_cachep();
4744         return -ENOMEM;
4745 }
4746
4747 static int btrfs_getattr(struct vfsmount *mnt,
4748                          struct dentry *dentry, struct kstat *stat)
4749 {
4750         struct inode *inode = dentry->d_inode;
4751         generic_fillattr(inode, stat);
4752         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4753         stat->blksize = PAGE_CACHE_SIZE;
4754         stat->blocks = (inode_get_bytes(inode) +
4755                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4756         return 0;
4757 }
4758
4759 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4760                            struct inode *new_dir, struct dentry *new_dentry)
4761 {
4762         struct btrfs_trans_handle *trans;
4763         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4764         struct inode *new_inode = new_dentry->d_inode;
4765         struct inode *old_inode = old_dentry->d_inode;
4766         struct timespec ctime = CURRENT_TIME;
4767         u64 index = 0;
4768         int ret;
4769
4770         /* we're not allowed to rename between subvolumes */
4771         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4772             BTRFS_I(new_dir)->root->root_key.objectid)
4773                 return -EXDEV;
4774
4775         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4776             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4777                 return -ENOTEMPTY;
4778         }
4779
4780         /* to rename a snapshot or subvolume, we need to juggle the
4781          * backrefs.  This isn't coded yet
4782          */
4783         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4784                 return -EXDEV;
4785
4786         ret = btrfs_check_metadata_free_space(root);
4787         if (ret)
4788                 goto out_unlock;
4789
4790         /*
4791          * we're using rename to replace one file with another.
4792          * and the replacement file is large.  Start IO on it now so
4793          * we don't add too much work to the end of the transaction
4794          */
4795         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
4796             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4797                 filemap_flush(old_inode->i_mapping);
4798
4799         trans = btrfs_start_transaction(root, 1);
4800
4801         /*
4802          * make sure the inode gets flushed if it is replacing
4803          * something.
4804          */
4805         if (new_inode && new_inode->i_size &&
4806             old_inode && S_ISREG(old_inode->i_mode)) {
4807                 btrfs_add_ordered_operation(trans, root, old_inode);
4808         }
4809
4810         /*
4811          * this is an ugly little race, but the rename is required to make
4812          * sure that if we crash, the inode is either at the old name
4813          * or the new one.  pinning the log transaction lets us make sure
4814          * we don't allow a log commit to come in after we unlink the
4815          * name but before we add the new name back in.
4816          */
4817         btrfs_pin_log_trans(root);
4818
4819         btrfs_set_trans_block_group(trans, new_dir);
4820
4821         btrfs_inc_nlink(old_dentry->d_inode);
4822         old_dir->i_ctime = old_dir->i_mtime = ctime;
4823         new_dir->i_ctime = new_dir->i_mtime = ctime;
4824         old_inode->i_ctime = ctime;
4825
4826         if (old_dentry->d_parent != new_dentry->d_parent)
4827                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4828
4829         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4830                                  old_dentry->d_name.name,
4831                                  old_dentry->d_name.len);
4832         if (ret)
4833                 goto out_fail;
4834
4835         if (new_inode) {
4836                 new_inode->i_ctime = CURRENT_TIME;
4837                 ret = btrfs_unlink_inode(trans, root, new_dir,
4838                                          new_dentry->d_inode,
4839                                          new_dentry->d_name.name,
4840                                          new_dentry->d_name.len);
4841                 if (ret)
4842                         goto out_fail;
4843                 if (new_inode->i_nlink == 0) {
4844                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4845                         if (ret)
4846                                 goto out_fail;
4847                 }
4848
4849         }
4850         ret = btrfs_set_inode_index(new_dir, &index);
4851         if (ret)
4852                 goto out_fail;
4853
4854         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4855                              old_inode, new_dentry->d_name.name,
4856                              new_dentry->d_name.len, 1, index);
4857         if (ret)
4858                 goto out_fail;
4859
4860         btrfs_log_new_name(trans, old_inode, old_dir,
4861                                        new_dentry->d_parent);
4862 out_fail:
4863
4864         /* this btrfs_end_log_trans just allows the current
4865          * log-sub transaction to complete
4866          */
4867         btrfs_end_log_trans(root);
4868         btrfs_end_transaction_throttle(trans, root);
4869 out_unlock:
4870         return ret;
4871 }
4872
4873 /*
4874  * some fairly slow code that needs optimization. This walks the list
4875  * of all the inodes with pending delalloc and forces them to disk.
4876  */
4877 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4878 {
4879         struct list_head *head = &root->fs_info->delalloc_inodes;
4880         struct btrfs_inode *binode;
4881         struct inode *inode;
4882
4883         if (root->fs_info->sb->s_flags & MS_RDONLY)
4884                 return -EROFS;
4885
4886         spin_lock(&root->fs_info->delalloc_lock);
4887         while (!list_empty(head)) {
4888                 binode = list_entry(head->next, struct btrfs_inode,
4889                                     delalloc_inodes);
4890                 inode = igrab(&binode->vfs_inode);
4891                 if (!inode)
4892                         list_del_init(&binode->delalloc_inodes);
4893                 spin_unlock(&root->fs_info->delalloc_lock);
4894                 if (inode) {
4895                         filemap_flush(inode->i_mapping);
4896                         iput(inode);
4897                 }
4898                 cond_resched();
4899                 spin_lock(&root->fs_info->delalloc_lock);
4900         }
4901         spin_unlock(&root->fs_info->delalloc_lock);
4902
4903         /* the filemap_flush will queue IO into the worker threads, but
4904          * we have to make sure the IO is actually started and that
4905          * ordered extents get created before we return
4906          */
4907         atomic_inc(&root->fs_info->async_submit_draining);
4908         while (atomic_read(&root->fs_info->nr_async_submits) ||
4909               atomic_read(&root->fs_info->async_delalloc_pages)) {
4910                 wait_event(root->fs_info->async_submit_wait,
4911                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4912                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4913         }
4914         atomic_dec(&root->fs_info->async_submit_draining);
4915         return 0;
4916 }
4917
4918 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4919                          const char *symname)
4920 {
4921         struct btrfs_trans_handle *trans;
4922         struct btrfs_root *root = BTRFS_I(dir)->root;
4923         struct btrfs_path *path;
4924         struct btrfs_key key;
4925         struct inode *inode = NULL;
4926         int err;
4927         int drop_inode = 0;
4928         u64 objectid;
4929         u64 index = 0 ;
4930         int name_len;
4931         int datasize;
4932         unsigned long ptr;
4933         struct btrfs_file_extent_item *ei;
4934         struct extent_buffer *leaf;
4935         unsigned long nr = 0;
4936
4937         name_len = strlen(symname) + 1;
4938         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4939                 return -ENAMETOOLONG;
4940
4941         err = btrfs_check_metadata_free_space(root);
4942         if (err)
4943                 goto out_fail;
4944
4945         trans = btrfs_start_transaction(root, 1);
4946         btrfs_set_trans_block_group(trans, dir);
4947
4948         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4949         if (err) {
4950                 err = -ENOSPC;
4951                 goto out_unlock;
4952         }
4953
4954         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4955                                 dentry->d_name.len,
4956                                 dentry->d_parent->d_inode->i_ino, objectid,
4957                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4958                                 &index);
4959         err = PTR_ERR(inode);
4960         if (IS_ERR(inode))
4961                 goto out_unlock;
4962
4963         err = btrfs_init_inode_security(inode, dir);
4964         if (err) {
4965                 drop_inode = 1;
4966                 goto out_unlock;
4967         }
4968
4969         btrfs_set_trans_block_group(trans, inode);
4970         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4971         if (err)
4972                 drop_inode = 1;
4973         else {
4974                 inode->i_mapping->a_ops = &btrfs_aops;
4975                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4976                 inode->i_fop = &btrfs_file_operations;
4977                 inode->i_op = &btrfs_file_inode_operations;
4978                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4979         }
4980         btrfs_update_inode_block_group(trans, inode);
4981         btrfs_update_inode_block_group(trans, dir);
4982         if (drop_inode)
4983                 goto out_unlock;
4984
4985         path = btrfs_alloc_path();
4986         BUG_ON(!path);
4987         key.objectid = inode->i_ino;
4988         key.offset = 0;
4989         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4990         datasize = btrfs_file_extent_calc_inline_size(name_len);
4991         err = btrfs_insert_empty_item(trans, root, path, &key,
4992                                       datasize);
4993         if (err) {
4994                 drop_inode = 1;
4995                 goto out_unlock;
4996         }
4997         leaf = path->nodes[0];
4998         ei = btrfs_item_ptr(leaf, path->slots[0],
4999                             struct btrfs_file_extent_item);
5000         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5001         btrfs_set_file_extent_type(leaf, ei,
5002                                    BTRFS_FILE_EXTENT_INLINE);
5003         btrfs_set_file_extent_encryption(leaf, ei, 0);
5004         btrfs_set_file_extent_compression(leaf, ei, 0);
5005         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5006         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5007
5008         ptr = btrfs_file_extent_inline_start(ei);
5009         write_extent_buffer(leaf, symname, ptr, name_len);
5010         btrfs_mark_buffer_dirty(leaf);
5011         btrfs_free_path(path);
5012
5013         inode->i_op = &btrfs_symlink_inode_operations;
5014         inode->i_mapping->a_ops = &btrfs_symlink_aops;
5015         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5016         inode_set_bytes(inode, name_len);
5017         btrfs_i_size_write(inode, name_len - 1);
5018         err = btrfs_update_inode(trans, root, inode);
5019         if (err)
5020                 drop_inode = 1;
5021
5022 out_unlock:
5023         nr = trans->blocks_used;
5024         btrfs_end_transaction_throttle(trans, root);
5025 out_fail:
5026         if (drop_inode) {
5027                 inode_dec_link_count(inode);
5028                 iput(inode);
5029         }
5030         btrfs_btree_balance_dirty(root, nr);
5031         return err;
5032 }
5033
5034 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5035                                struct inode *inode, u64 start, u64 end,
5036                                u64 locked_end, u64 alloc_hint, int mode)
5037 {
5038         struct btrfs_root *root = BTRFS_I(inode)->root;
5039         struct btrfs_key ins;
5040         u64 alloc_size;
5041         u64 cur_offset = start;
5042         u64 num_bytes = end - start;
5043         int ret = 0;
5044
5045         while (num_bytes > 0) {
5046                 alloc_size = min(num_bytes, root->fs_info->max_extent);
5047                 ret = btrfs_reserve_extent(trans, root, alloc_size,
5048                                            root->sectorsize, 0, alloc_hint,
5049                                            (u64)-1, &ins, 1);
5050                 if (ret) {
5051                         WARN_ON(1);
5052                         goto out;
5053                 }
5054                 ret = insert_reserved_file_extent(trans, inode,
5055                                                   cur_offset, ins.objectid,
5056                                                   ins.offset, ins.offset,
5057                                                   ins.offset, locked_end,
5058                                                   0, 0, 0,
5059                                                   BTRFS_FILE_EXTENT_PREALLOC);
5060                 BUG_ON(ret);
5061                 num_bytes -= ins.offset;
5062                 cur_offset += ins.offset;
5063                 alloc_hint = ins.objectid + ins.offset;
5064         }
5065 out:
5066         if (cur_offset > start) {
5067                 inode->i_ctime = CURRENT_TIME;
5068                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5069                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5070                     cur_offset > i_size_read(inode))
5071                         btrfs_i_size_write(inode, cur_offset);
5072                 ret = btrfs_update_inode(trans, root, inode);
5073                 BUG_ON(ret);
5074         }
5075
5076         return ret;
5077 }
5078
5079 static long btrfs_fallocate(struct inode *inode, int mode,
5080                             loff_t offset, loff_t len)
5081 {
5082         u64 cur_offset;
5083         u64 last_byte;
5084         u64 alloc_start;
5085         u64 alloc_end;
5086         u64 alloc_hint = 0;
5087         u64 locked_end;
5088         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5089         struct extent_map *em;
5090         struct btrfs_trans_handle *trans;
5091         struct btrfs_root *root;
5092         int ret;
5093
5094         alloc_start = offset & ~mask;
5095         alloc_end =  (offset + len + mask) & ~mask;
5096
5097         /*
5098          * wait for ordered IO before we have any locks.  We'll loop again
5099          * below with the locks held.
5100          */
5101         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5102
5103         mutex_lock(&inode->i_mutex);
5104         if (alloc_start > inode->i_size) {
5105                 ret = btrfs_cont_expand(inode, alloc_start);
5106                 if (ret)
5107                         goto out;
5108         }
5109
5110         root = BTRFS_I(inode)->root;
5111
5112         ret = btrfs_check_data_free_space(root, inode,
5113                                           alloc_end - alloc_start);
5114         if (ret)
5115                 goto out;
5116
5117         locked_end = alloc_end - 1;
5118         while (1) {
5119                 struct btrfs_ordered_extent *ordered;
5120
5121                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5122                 if (!trans) {
5123                         ret = -EIO;
5124                         goto out_free;
5125                 }
5126
5127                 /* the extent lock is ordered inside the running
5128                  * transaction
5129                  */
5130                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5131                             GFP_NOFS);
5132                 ordered = btrfs_lookup_first_ordered_extent(inode,
5133                                                             alloc_end - 1);
5134                 if (ordered &&
5135                     ordered->file_offset + ordered->len > alloc_start &&
5136                     ordered->file_offset < alloc_end) {
5137                         btrfs_put_ordered_extent(ordered);
5138                         unlock_extent(&BTRFS_I(inode)->io_tree,
5139                                       alloc_start, locked_end, GFP_NOFS);
5140                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5141
5142                         /*
5143                          * we can't wait on the range with the transaction
5144                          * running or with the extent lock held
5145                          */
5146                         btrfs_wait_ordered_range(inode, alloc_start,
5147                                                  alloc_end - alloc_start);
5148                 } else {
5149                         if (ordered)
5150                                 btrfs_put_ordered_extent(ordered);
5151                         break;
5152                 }
5153         }
5154
5155         cur_offset = alloc_start;
5156         while (1) {
5157                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5158                                       alloc_end - cur_offset, 0);
5159                 BUG_ON(IS_ERR(em) || !em);
5160                 last_byte = min(extent_map_end(em), alloc_end);
5161                 last_byte = (last_byte + mask) & ~mask;
5162                 if (em->block_start == EXTENT_MAP_HOLE) {
5163                         ret = prealloc_file_range(trans, inode, cur_offset,
5164                                         last_byte, locked_end + 1,
5165                                         alloc_hint, mode);
5166                         if (ret < 0) {
5167                                 free_extent_map(em);
5168                                 break;
5169                         }
5170                 }
5171                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5172                         alloc_hint = em->block_start;
5173                 free_extent_map(em);
5174
5175                 cur_offset = last_byte;
5176                 if (cur_offset >= alloc_end) {
5177                         ret = 0;
5178                         break;
5179                 }
5180         }
5181         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5182                       GFP_NOFS);
5183
5184         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5185 out_free:
5186         btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5187 out:
5188         mutex_unlock(&inode->i_mutex);
5189         return ret;
5190 }
5191
5192 static int btrfs_set_page_dirty(struct page *page)
5193 {
5194         return __set_page_dirty_nobuffers(page);
5195 }
5196
5197 static int btrfs_permission(struct inode *inode, int mask)
5198 {
5199         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5200                 return -EACCES;
5201         return generic_permission(inode, mask, btrfs_check_acl);
5202 }
5203
5204 static const struct inode_operations btrfs_dir_inode_operations = {
5205         .getattr        = btrfs_getattr,
5206         .lookup         = btrfs_lookup,
5207         .create         = btrfs_create,
5208         .unlink         = btrfs_unlink,
5209         .link           = btrfs_link,
5210         .mkdir          = btrfs_mkdir,
5211         .rmdir          = btrfs_rmdir,
5212         .rename         = btrfs_rename,
5213         .symlink        = btrfs_symlink,
5214         .setattr        = btrfs_setattr,
5215         .mknod          = btrfs_mknod,
5216         .setxattr       = btrfs_setxattr,
5217         .getxattr       = btrfs_getxattr,
5218         .listxattr      = btrfs_listxattr,
5219         .removexattr    = btrfs_removexattr,
5220         .permission     = btrfs_permission,
5221 };
5222 static const struct inode_operations btrfs_dir_ro_inode_operations = {
5223         .lookup         = btrfs_lookup,
5224         .permission     = btrfs_permission,
5225 };
5226 static struct file_operations btrfs_dir_file_operations = {
5227         .llseek         = generic_file_llseek,
5228         .read           = generic_read_dir,
5229         .readdir        = btrfs_real_readdir,
5230         .unlocked_ioctl = btrfs_ioctl,
5231 #ifdef CONFIG_COMPAT
5232         .compat_ioctl   = btrfs_ioctl,
5233 #endif
5234         .release        = btrfs_release_file,
5235         .fsync          = btrfs_sync_file,
5236 };
5237
5238 static struct extent_io_ops btrfs_extent_io_ops = {
5239         .fill_delalloc = run_delalloc_range,
5240         .submit_bio_hook = btrfs_submit_bio_hook,
5241         .merge_bio_hook = btrfs_merge_bio_hook,
5242         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5243         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5244         .writepage_start_hook = btrfs_writepage_start_hook,
5245         .readpage_io_failed_hook = btrfs_io_failed_hook,
5246         .set_bit_hook = btrfs_set_bit_hook,
5247         .clear_bit_hook = btrfs_clear_bit_hook,
5248 };
5249
5250 /*
5251  * btrfs doesn't support the bmap operation because swapfiles
5252  * use bmap to make a mapping of extents in the file.  They assume
5253  * these extents won't change over the life of the file and they
5254  * use the bmap result to do IO directly to the drive.
5255  *
5256  * the btrfs bmap call would return logical addresses that aren't
5257  * suitable for IO and they also will change frequently as COW
5258  * operations happen.  So, swapfile + btrfs == corruption.
5259  *
5260  * For now we're avoiding this by dropping bmap.
5261  */
5262 static const struct address_space_operations btrfs_aops = {
5263         .readpage       = btrfs_readpage,
5264         .writepage      = btrfs_writepage,
5265         .writepages     = btrfs_writepages,
5266         .readpages      = btrfs_readpages,
5267         .sync_page      = block_sync_page,
5268         .direct_IO      = btrfs_direct_IO,
5269         .invalidatepage = btrfs_invalidatepage,
5270         .releasepage    = btrfs_releasepage,
5271         .set_page_dirty = btrfs_set_page_dirty,
5272         .error_remove_page = generic_error_remove_page,
5273 };
5274
5275 static const struct address_space_operations btrfs_symlink_aops = {
5276         .readpage       = btrfs_readpage,
5277         .writepage      = btrfs_writepage,
5278         .invalidatepage = btrfs_invalidatepage,
5279         .releasepage    = btrfs_releasepage,
5280 };
5281
5282 static const struct inode_operations btrfs_file_inode_operations = {
5283         .truncate       = btrfs_truncate,
5284         .getattr        = btrfs_getattr,
5285         .setattr        = btrfs_setattr,
5286         .setxattr       = btrfs_setxattr,
5287         .getxattr       = btrfs_getxattr,
5288         .listxattr      = btrfs_listxattr,
5289         .removexattr    = btrfs_removexattr,
5290         .permission     = btrfs_permission,
5291         .fallocate      = btrfs_fallocate,
5292         .fiemap         = btrfs_fiemap,
5293 };
5294 static const struct inode_operations btrfs_special_inode_operations = {
5295         .getattr        = btrfs_getattr,
5296         .setattr        = btrfs_setattr,
5297         .permission     = btrfs_permission,
5298         .setxattr       = btrfs_setxattr,
5299         .getxattr       = btrfs_getxattr,
5300         .listxattr      = btrfs_listxattr,
5301         .removexattr    = btrfs_removexattr,
5302 };
5303 static const struct inode_operations btrfs_symlink_inode_operations = {
5304         .readlink       = generic_readlink,
5305         .follow_link    = page_follow_link_light,
5306         .put_link       = page_put_link,
5307         .permission     = btrfs_permission,
5308         .setxattr       = btrfs_setxattr,
5309         .getxattr       = btrfs_getxattr,
5310         .listxattr      = btrfs_listxattr,
5311         .removexattr    = btrfs_removexattr,
5312 };