Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static struct inode_operations btrfs_dir_inode_operations;
60 static struct inode_operations btrfs_symlink_inode_operations;
61 static struct inode_operations btrfs_dir_ro_inode_operations;
62 static struct inode_operations btrfs_special_inode_operations;
63 static struct inode_operations btrfs_file_inode_operations;
64 static struct address_space_operations btrfs_aops;
65 static struct address_space_operations btrfs_symlink_aops;
66 static struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
93 {
94         int err;
95
96         err = btrfs_init_acl(inode, dir);
97         if (!err)
98                 err = btrfs_xattr_security_init(inode, dir);
99         return err;
100 }
101
102 /*
103  * this does all the hard work for inserting an inline extent into
104  * the btree.  The caller should have done a btrfs_drop_extents so that
105  * no overlapping inline items exist in the btree
106  */
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108                                 struct btrfs_root *root, struct inode *inode,
109                                 u64 start, size_t size, size_t compressed_size,
110                                 struct page **compressed_pages)
111 {
112         struct btrfs_key key;
113         struct btrfs_path *path;
114         struct extent_buffer *leaf;
115         struct page *page = NULL;
116         char *kaddr;
117         unsigned long ptr;
118         struct btrfs_file_extent_item *ei;
119         int err = 0;
120         int ret;
121         size_t cur_size = size;
122         size_t datasize;
123         unsigned long offset;
124         int use_compress = 0;
125
126         if (compressed_size && compressed_pages) {
127                 use_compress = 1;
128                 cur_size = compressed_size;
129         }
130
131         path = btrfs_alloc_path();
132         if (!path)
133                 return -ENOMEM;
134
135         path->leave_spinning = 1;
136         btrfs_set_trans_block_group(trans, inode);
137
138         key.objectid = inode->i_ino;
139         key.offset = start;
140         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141         datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143         inode_add_bytes(inode, size);
144         ret = btrfs_insert_empty_item(trans, root, path, &key,
145                                       datasize);
146         BUG_ON(ret);
147         if (ret) {
148                 err = ret;
149                 goto fail;
150         }
151         leaf = path->nodes[0];
152         ei = btrfs_item_ptr(leaf, path->slots[0],
153                             struct btrfs_file_extent_item);
154         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156         btrfs_set_file_extent_encryption(leaf, ei, 0);
157         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159         ptr = btrfs_file_extent_inline_start(ei);
160
161         if (use_compress) {
162                 struct page *cpage;
163                 int i = 0;
164                 while (compressed_size > 0) {
165                         cpage = compressed_pages[i];
166                         cur_size = min_t(unsigned long, compressed_size,
167                                        PAGE_CACHE_SIZE);
168
169                         kaddr = kmap_atomic(cpage, KM_USER0);
170                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
171                         kunmap_atomic(kaddr, KM_USER0);
172
173                         i++;
174                         ptr += cur_size;
175                         compressed_size -= cur_size;
176                 }
177                 btrfs_set_file_extent_compression(leaf, ei,
178                                                   BTRFS_COMPRESS_ZLIB);
179         } else {
180                 page = find_get_page(inode->i_mapping,
181                                      start >> PAGE_CACHE_SHIFT);
182                 btrfs_set_file_extent_compression(leaf, ei, 0);
183                 kaddr = kmap_atomic(page, KM_USER0);
184                 offset = start & (PAGE_CACHE_SIZE - 1);
185                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186                 kunmap_atomic(kaddr, KM_USER0);
187                 page_cache_release(page);
188         }
189         btrfs_mark_buffer_dirty(leaf);
190         btrfs_free_path(path);
191
192         BTRFS_I(inode)->disk_i_size = inode->i_size;
193         btrfs_update_inode(trans, root, inode);
194         return 0;
195 fail:
196         btrfs_free_path(path);
197         return err;
198 }
199
200
201 /*
202  * conditionally insert an inline extent into the file.  This
203  * does the checks required to make sure the data is small enough
204  * to fit as an inline extent.
205  */
206 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
207                                  struct btrfs_root *root,
208                                  struct inode *inode, u64 start, u64 end,
209                                  size_t compressed_size,
210                                  struct page **compressed_pages)
211 {
212         u64 isize = i_size_read(inode);
213         u64 actual_end = min(end + 1, isize);
214         u64 inline_len = actual_end - start;
215         u64 aligned_end = (end + root->sectorsize - 1) &
216                         ~((u64)root->sectorsize - 1);
217         u64 hint_byte;
218         u64 data_len = inline_len;
219         int ret;
220
221         if (compressed_size)
222                 data_len = compressed_size;
223
224         if (start > 0 ||
225             actual_end >= PAGE_CACHE_SIZE ||
226             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227             (!compressed_size &&
228             (actual_end & (root->sectorsize - 1)) == 0) ||
229             end + 1 < isize ||
230             data_len > root->fs_info->max_inline) {
231                 return 1;
232         }
233
234         ret = btrfs_drop_extents(trans, root, inode, start,
235                                  aligned_end, aligned_end, start, &hint_byte);
236         BUG_ON(ret);
237
238         if (isize > actual_end)
239                 inline_len = min_t(u64, isize, actual_end);
240         ret = insert_inline_extent(trans, root, inode, start,
241                                    inline_len, compressed_size,
242                                    compressed_pages);
243         BUG_ON(ret);
244         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
245         return 0;
246 }
247
248 struct async_extent {
249         u64 start;
250         u64 ram_size;
251         u64 compressed_size;
252         struct page **pages;
253         unsigned long nr_pages;
254         struct list_head list;
255 };
256
257 struct async_cow {
258         struct inode *inode;
259         struct btrfs_root *root;
260         struct page *locked_page;
261         u64 start;
262         u64 end;
263         struct list_head extents;
264         struct btrfs_work work;
265 };
266
267 static noinline int add_async_extent(struct async_cow *cow,
268                                      u64 start, u64 ram_size,
269                                      u64 compressed_size,
270                                      struct page **pages,
271                                      unsigned long nr_pages)
272 {
273         struct async_extent *async_extent;
274
275         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276         async_extent->start = start;
277         async_extent->ram_size = ram_size;
278         async_extent->compressed_size = compressed_size;
279         async_extent->pages = pages;
280         async_extent->nr_pages = nr_pages;
281         list_add_tail(&async_extent->list, &cow->extents);
282         return 0;
283 }
284
285 /*
286  * we create compressed extents in two phases.  The first
287  * phase compresses a range of pages that have already been
288  * locked (both pages and state bits are locked).
289  *
290  * This is done inside an ordered work queue, and the compression
291  * is spread across many cpus.  The actual IO submission is step
292  * two, and the ordered work queue takes care of making sure that
293  * happens in the same order things were put onto the queue by
294  * writepages and friends.
295  *
296  * If this code finds it can't get good compression, it puts an
297  * entry onto the work queue to write the uncompressed bytes.  This
298  * makes sure that both compressed inodes and uncompressed inodes
299  * are written in the same order that pdflush sent them down.
300  */
301 static noinline int compress_file_range(struct inode *inode,
302                                         struct page *locked_page,
303                                         u64 start, u64 end,
304                                         struct async_cow *async_cow,
305                                         int *num_added)
306 {
307         struct btrfs_root *root = BTRFS_I(inode)->root;
308         struct btrfs_trans_handle *trans;
309         u64 num_bytes;
310         u64 orig_start;
311         u64 disk_num_bytes;
312         u64 blocksize = root->sectorsize;
313         u64 actual_end;
314         u64 isize = i_size_read(inode);
315         int ret = 0;
316         struct page **pages = NULL;
317         unsigned long nr_pages;
318         unsigned long nr_pages_ret = 0;
319         unsigned long total_compressed = 0;
320         unsigned long total_in = 0;
321         unsigned long max_compressed = 128 * 1024;
322         unsigned long max_uncompressed = 128 * 1024;
323         int i;
324         int will_compress;
325
326         orig_start = start;
327
328         actual_end = min_t(u64, isize, end + 1);
329 again:
330         will_compress = 0;
331         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
333
334         /*
335          * we don't want to send crud past the end of i_size through
336          * compression, that's just a waste of CPU time.  So, if the
337          * end of the file is before the start of our current
338          * requested range of bytes, we bail out to the uncompressed
339          * cleanup code that can deal with all of this.
340          *
341          * It isn't really the fastest way to fix things, but this is a
342          * very uncommon corner.
343          */
344         if (actual_end <= start)
345                 goto cleanup_and_bail_uncompressed;
346
347         total_compressed = actual_end - start;
348
349         /* we want to make sure that amount of ram required to uncompress
350          * an extent is reasonable, so we limit the total size in ram
351          * of a compressed extent to 128k.  This is a crucial number
352          * because it also controls how easily we can spread reads across
353          * cpus for decompression.
354          *
355          * We also want to make sure the amount of IO required to do
356          * a random read is reasonably small, so we limit the size of
357          * a compressed extent to 128k.
358          */
359         total_compressed = min(total_compressed, max_uncompressed);
360         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361         num_bytes = max(blocksize,  num_bytes);
362         disk_num_bytes = num_bytes;
363         total_in = 0;
364         ret = 0;
365
366         /*
367          * we do compression for mount -o compress and when the
368          * inode has not been flagged as nocompress.  This flag can
369          * change at any time if we discover bad compression ratios.
370          */
371         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
372             btrfs_test_opt(root, COMPRESS)) {
373                 WARN_ON(pages);
374                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
375
376                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377                                                 total_compressed, pages,
378                                                 nr_pages, &nr_pages_ret,
379                                                 &total_in,
380                                                 &total_compressed,
381                                                 max_compressed);
382
383                 if (!ret) {
384                         unsigned long offset = total_compressed &
385                                 (PAGE_CACHE_SIZE - 1);
386                         struct page *page = pages[nr_pages_ret - 1];
387                         char *kaddr;
388
389                         /* zero the tail end of the last page, we might be
390                          * sending it down to disk
391                          */
392                         if (offset) {
393                                 kaddr = kmap_atomic(page, KM_USER0);
394                                 memset(kaddr + offset, 0,
395                                        PAGE_CACHE_SIZE - offset);
396                                 kunmap_atomic(kaddr, KM_USER0);
397                         }
398                         will_compress = 1;
399                 }
400         }
401         if (start == 0) {
402                 trans = btrfs_join_transaction(root, 1);
403                 BUG_ON(!trans);
404                 btrfs_set_trans_block_group(trans, inode);
405
406                 /* lets try to make an inline extent */
407                 if (ret || total_in < (actual_end - start)) {
408                         /* we didn't compress the entire range, try
409                          * to make an uncompressed inline extent.
410                          */
411                         ret = cow_file_range_inline(trans, root, inode,
412                                                     start, end, 0, NULL);
413                 } else {
414                         /* try making a compressed inline extent */
415                         ret = cow_file_range_inline(trans, root, inode,
416                                                     start, end,
417                                                     total_compressed, pages);
418                 }
419                 btrfs_end_transaction(trans, root);
420                 if (ret == 0) {
421                         /*
422                          * inline extent creation worked, we don't need
423                          * to create any more async work items.  Unlock
424                          * and free up our temp pages.
425                          */
426                         extent_clear_unlock_delalloc(inode,
427                                                      &BTRFS_I(inode)->io_tree,
428                                                      start, end, NULL, 1, 0,
429                                                      0, 1, 1, 1);
430                         ret = 0;
431                         goto free_pages_out;
432                 }
433         }
434
435         if (will_compress) {
436                 /*
437                  * we aren't doing an inline extent round the compressed size
438                  * up to a block size boundary so the allocator does sane
439                  * things
440                  */
441                 total_compressed = (total_compressed + blocksize - 1) &
442                         ~(blocksize - 1);
443
444                 /*
445                  * one last check to make sure the compression is really a
446                  * win, compare the page count read with the blocks on disk
447                  */
448                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
449                         ~(PAGE_CACHE_SIZE - 1);
450                 if (total_compressed >= total_in) {
451                         will_compress = 0;
452                 } else {
453                         disk_num_bytes = total_compressed;
454                         num_bytes = total_in;
455                 }
456         }
457         if (!will_compress && pages) {
458                 /*
459                  * the compression code ran but failed to make things smaller,
460                  * free any pages it allocated and our page pointer array
461                  */
462                 for (i = 0; i < nr_pages_ret; i++) {
463                         WARN_ON(pages[i]->mapping);
464                         page_cache_release(pages[i]);
465                 }
466                 kfree(pages);
467                 pages = NULL;
468                 total_compressed = 0;
469                 nr_pages_ret = 0;
470
471                 /* flag the file so we don't compress in the future */
472                 btrfs_set_flag(inode, NOCOMPRESS);
473         }
474         if (will_compress) {
475                 *num_added += 1;
476
477                 /* the async work queues will take care of doing actual
478                  * allocation on disk for these compressed pages,
479                  * and will submit them to the elevator.
480                  */
481                 add_async_extent(async_cow, start, num_bytes,
482                                  total_compressed, pages, nr_pages_ret);
483
484                 if (start + num_bytes < end && start + num_bytes < actual_end) {
485                         start += num_bytes;
486                         pages = NULL;
487                         cond_resched();
488                         goto again;
489                 }
490         } else {
491 cleanup_and_bail_uncompressed:
492                 /*
493                  * No compression, but we still need to write the pages in
494                  * the file we've been given so far.  redirty the locked
495                  * page if it corresponds to our extent and set things up
496                  * for the async work queue to run cow_file_range to do
497                  * the normal delalloc dance
498                  */
499                 if (page_offset(locked_page) >= start &&
500                     page_offset(locked_page) <= end) {
501                         __set_page_dirty_nobuffers(locked_page);
502                         /* unlocked later on in the async handlers */
503                 }
504                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
505                 *num_added += 1;
506         }
507
508 out:
509         return 0;
510
511 free_pages_out:
512         for (i = 0; i < nr_pages_ret; i++) {
513                 WARN_ON(pages[i]->mapping);
514                 page_cache_release(pages[i]);
515         }
516         kfree(pages);
517
518         goto out;
519 }
520
521 /*
522  * phase two of compressed writeback.  This is the ordered portion
523  * of the code, which only gets called in the order the work was
524  * queued.  We walk all the async extents created by compress_file_range
525  * and send them down to the disk.
526  */
527 static noinline int submit_compressed_extents(struct inode *inode,
528                                               struct async_cow *async_cow)
529 {
530         struct async_extent *async_extent;
531         u64 alloc_hint = 0;
532         struct btrfs_trans_handle *trans;
533         struct btrfs_key ins;
534         struct extent_map *em;
535         struct btrfs_root *root = BTRFS_I(inode)->root;
536         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537         struct extent_io_tree *io_tree;
538         int ret;
539
540         if (list_empty(&async_cow->extents))
541                 return 0;
542
543         trans = btrfs_join_transaction(root, 1);
544
545         while (!list_empty(&async_cow->extents)) {
546                 async_extent = list_entry(async_cow->extents.next,
547                                           struct async_extent, list);
548                 list_del(&async_extent->list);
549
550                 io_tree = &BTRFS_I(inode)->io_tree;
551
552                 /* did the compression code fall back to uncompressed IO? */
553                 if (!async_extent->pages) {
554                         int page_started = 0;
555                         unsigned long nr_written = 0;
556
557                         lock_extent(io_tree, async_extent->start,
558                                     async_extent->start +
559                                     async_extent->ram_size - 1, GFP_NOFS);
560
561                         /* allocate blocks */
562                         cow_file_range(inode, async_cow->locked_page,
563                                        async_extent->start,
564                                        async_extent->start +
565                                        async_extent->ram_size - 1,
566                                        &page_started, &nr_written, 0);
567
568                         /*
569                          * if page_started, cow_file_range inserted an
570                          * inline extent and took care of all the unlocking
571                          * and IO for us.  Otherwise, we need to submit
572                          * all those pages down to the drive.
573                          */
574                         if (!page_started)
575                                 extent_write_locked_range(io_tree,
576                                                   inode, async_extent->start,
577                                                   async_extent->start +
578                                                   async_extent->ram_size - 1,
579                                                   btrfs_get_extent,
580                                                   WB_SYNC_ALL);
581                         kfree(async_extent);
582                         cond_resched();
583                         continue;
584                 }
585
586                 lock_extent(io_tree, async_extent->start,
587                             async_extent->start + async_extent->ram_size - 1,
588                             GFP_NOFS);
589                 /*
590                  * here we're doing allocation and writeback of the
591                  * compressed pages
592                  */
593                 btrfs_drop_extent_cache(inode, async_extent->start,
594                                         async_extent->start +
595                                         async_extent->ram_size - 1, 0);
596
597                 ret = btrfs_reserve_extent(trans, root,
598                                            async_extent->compressed_size,
599                                            async_extent->compressed_size,
600                                            0, alloc_hint,
601                                            (u64)-1, &ins, 1);
602                 BUG_ON(ret);
603                 em = alloc_extent_map(GFP_NOFS);
604                 em->start = async_extent->start;
605                 em->len = async_extent->ram_size;
606                 em->orig_start = em->start;
607
608                 em->block_start = ins.objectid;
609                 em->block_len = ins.offset;
610                 em->bdev = root->fs_info->fs_devices->latest_bdev;
611                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
612                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
613
614                 while (1) {
615                         spin_lock(&em_tree->lock);
616                         ret = add_extent_mapping(em_tree, em);
617                         spin_unlock(&em_tree->lock);
618                         if (ret != -EEXIST) {
619                                 free_extent_map(em);
620                                 break;
621                         }
622                         btrfs_drop_extent_cache(inode, async_extent->start,
623                                                 async_extent->start +
624                                                 async_extent->ram_size - 1, 0);
625                 }
626
627                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
628                                                ins.objectid,
629                                                async_extent->ram_size,
630                                                ins.offset,
631                                                BTRFS_ORDERED_COMPRESSED);
632                 BUG_ON(ret);
633
634                 btrfs_end_transaction(trans, root);
635
636                 /*
637                  * clear dirty, set writeback and unlock the pages.
638                  */
639                 extent_clear_unlock_delalloc(inode,
640                                              &BTRFS_I(inode)->io_tree,
641                                              async_extent->start,
642                                              async_extent->start +
643                                              async_extent->ram_size - 1,
644                                              NULL, 1, 1, 0, 1, 1, 0);
645
646                 ret = btrfs_submit_compressed_write(inode,
647                                     async_extent->start,
648                                     async_extent->ram_size,
649                                     ins.objectid,
650                                     ins.offset, async_extent->pages,
651                                     async_extent->nr_pages);
652
653                 BUG_ON(ret);
654                 trans = btrfs_join_transaction(root, 1);
655                 alloc_hint = ins.objectid + ins.offset;
656                 kfree(async_extent);
657                 cond_resched();
658         }
659
660         btrfs_end_transaction(trans, root);
661         return 0;
662 }
663
664 /*
665  * when extent_io.c finds a delayed allocation range in the file,
666  * the call backs end up in this code.  The basic idea is to
667  * allocate extents on disk for the range, and create ordered data structs
668  * in ram to track those extents.
669  *
670  * locked_page is the page that writepage had locked already.  We use
671  * it to make sure we don't do extra locks or unlocks.
672  *
673  * *page_started is set to one if we unlock locked_page and do everything
674  * required to start IO on it.  It may be clean and already done with
675  * IO when we return.
676  */
677 static noinline int cow_file_range(struct inode *inode,
678                                    struct page *locked_page,
679                                    u64 start, u64 end, int *page_started,
680                                    unsigned long *nr_written,
681                                    int unlock)
682 {
683         struct btrfs_root *root = BTRFS_I(inode)->root;
684         struct btrfs_trans_handle *trans;
685         u64 alloc_hint = 0;
686         u64 num_bytes;
687         unsigned long ram_size;
688         u64 disk_num_bytes;
689         u64 cur_alloc_size;
690         u64 blocksize = root->sectorsize;
691         u64 actual_end;
692         u64 isize = i_size_read(inode);
693         struct btrfs_key ins;
694         struct extent_map *em;
695         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
696         int ret = 0;
697
698         trans = btrfs_join_transaction(root, 1);
699         BUG_ON(!trans);
700         btrfs_set_trans_block_group(trans, inode);
701
702         actual_end = min_t(u64, isize, end + 1);
703
704         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
705         num_bytes = max(blocksize,  num_bytes);
706         disk_num_bytes = num_bytes;
707         ret = 0;
708
709         if (start == 0) {
710                 /* lets try to make an inline extent */
711                 ret = cow_file_range_inline(trans, root, inode,
712                                             start, end, 0, NULL);
713                 if (ret == 0) {
714                         extent_clear_unlock_delalloc(inode,
715                                                      &BTRFS_I(inode)->io_tree,
716                                                      start, end, NULL, 1, 1,
717                                                      1, 1, 1, 1);
718                         *nr_written = *nr_written +
719                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
720                         *page_started = 1;
721                         ret = 0;
722                         goto out;
723                 }
724         }
725
726         BUG_ON(disk_num_bytes >
727                btrfs_super_total_bytes(&root->fs_info->super_copy));
728
729         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
730
731         while (disk_num_bytes > 0) {
732                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
733                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
734                                            root->sectorsize, 0, alloc_hint,
735                                            (u64)-1, &ins, 1);
736                 BUG_ON(ret);
737
738                 em = alloc_extent_map(GFP_NOFS);
739                 em->start = start;
740                 em->orig_start = em->start;
741
742                 ram_size = ins.offset;
743                 em->len = ins.offset;
744
745                 em->block_start = ins.objectid;
746                 em->block_len = ins.offset;
747                 em->bdev = root->fs_info->fs_devices->latest_bdev;
748                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
749
750                 while (1) {
751                         spin_lock(&em_tree->lock);
752                         ret = add_extent_mapping(em_tree, em);
753                         spin_unlock(&em_tree->lock);
754                         if (ret != -EEXIST) {
755                                 free_extent_map(em);
756                                 break;
757                         }
758                         btrfs_drop_extent_cache(inode, start,
759                                                 start + ram_size - 1, 0);
760                 }
761
762                 cur_alloc_size = ins.offset;
763                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
764                                                ram_size, cur_alloc_size, 0);
765                 BUG_ON(ret);
766
767                 if (root->root_key.objectid ==
768                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
769                         ret = btrfs_reloc_clone_csums(inode, start,
770                                                       cur_alloc_size);
771                         BUG_ON(ret);
772                 }
773
774                 if (disk_num_bytes < cur_alloc_size)
775                         break;
776
777                 /* we're not doing compressed IO, don't unlock the first
778                  * page (which the caller expects to stay locked), don't
779                  * clear any dirty bits and don't set any writeback bits
780                  */
781                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
782                                              start, start + ram_size - 1,
783                                              locked_page, unlock, 1,
784                                              1, 0, 0, 0);
785                 disk_num_bytes -= cur_alloc_size;
786                 num_bytes -= cur_alloc_size;
787                 alloc_hint = ins.objectid + ins.offset;
788                 start += cur_alloc_size;
789         }
790 out:
791         ret = 0;
792         btrfs_end_transaction(trans, root);
793
794         return ret;
795 }
796
797 /*
798  * work queue call back to started compression on a file and pages
799  */
800 static noinline void async_cow_start(struct btrfs_work *work)
801 {
802         struct async_cow *async_cow;
803         int num_added = 0;
804         async_cow = container_of(work, struct async_cow, work);
805
806         compress_file_range(async_cow->inode, async_cow->locked_page,
807                             async_cow->start, async_cow->end, async_cow,
808                             &num_added);
809         if (num_added == 0)
810                 async_cow->inode = NULL;
811 }
812
813 /*
814  * work queue call back to submit previously compressed pages
815  */
816 static noinline void async_cow_submit(struct btrfs_work *work)
817 {
818         struct async_cow *async_cow;
819         struct btrfs_root *root;
820         unsigned long nr_pages;
821
822         async_cow = container_of(work, struct async_cow, work);
823
824         root = async_cow->root;
825         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
826                 PAGE_CACHE_SHIFT;
827
828         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
829
830         if (atomic_read(&root->fs_info->async_delalloc_pages) <
831             5 * 1042 * 1024 &&
832             waitqueue_active(&root->fs_info->async_submit_wait))
833                 wake_up(&root->fs_info->async_submit_wait);
834
835         if (async_cow->inode)
836                 submit_compressed_extents(async_cow->inode, async_cow);
837 }
838
839 static noinline void async_cow_free(struct btrfs_work *work)
840 {
841         struct async_cow *async_cow;
842         async_cow = container_of(work, struct async_cow, work);
843         kfree(async_cow);
844 }
845
846 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
847                                 u64 start, u64 end, int *page_started,
848                                 unsigned long *nr_written)
849 {
850         struct async_cow *async_cow;
851         struct btrfs_root *root = BTRFS_I(inode)->root;
852         unsigned long nr_pages;
853         u64 cur_end;
854         int limit = 10 * 1024 * 1042;
855
856         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
857                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
858         while (start < end) {
859                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
860                 async_cow->inode = inode;
861                 async_cow->root = root;
862                 async_cow->locked_page = locked_page;
863                 async_cow->start = start;
864
865                 if (btrfs_test_flag(inode, NOCOMPRESS))
866                         cur_end = end;
867                 else
868                         cur_end = min(end, start + 512 * 1024 - 1);
869
870                 async_cow->end = cur_end;
871                 INIT_LIST_HEAD(&async_cow->extents);
872
873                 async_cow->work.func = async_cow_start;
874                 async_cow->work.ordered_func = async_cow_submit;
875                 async_cow->work.ordered_free = async_cow_free;
876                 async_cow->work.flags = 0;
877
878                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
879                         PAGE_CACHE_SHIFT;
880                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
881
882                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
883                                    &async_cow->work);
884
885                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
886                         wait_event(root->fs_info->async_submit_wait,
887                            (atomic_read(&root->fs_info->async_delalloc_pages) <
888                             limit));
889                 }
890
891                 while (atomic_read(&root->fs_info->async_submit_draining) &&
892                       atomic_read(&root->fs_info->async_delalloc_pages)) {
893                         wait_event(root->fs_info->async_submit_wait,
894                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
895                            0));
896                 }
897
898                 *nr_written += nr_pages;
899                 start = cur_end + 1;
900         }
901         *page_started = 1;
902         return 0;
903 }
904
905 static noinline int csum_exist_in_range(struct btrfs_root *root,
906                                         u64 bytenr, u64 num_bytes)
907 {
908         int ret;
909         struct btrfs_ordered_sum *sums;
910         LIST_HEAD(list);
911
912         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
913                                        bytenr + num_bytes - 1, &list);
914         if (ret == 0 && list_empty(&list))
915                 return 0;
916
917         while (!list_empty(&list)) {
918                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
919                 list_del(&sums->list);
920                 kfree(sums);
921         }
922         return 1;
923 }
924
925 /*
926  * when nowcow writeback call back.  This checks for snapshots or COW copies
927  * of the extents that exist in the file, and COWs the file as required.
928  *
929  * If no cow copies or snapshots exist, we write directly to the existing
930  * blocks on disk
931  */
932 static noinline int run_delalloc_nocow(struct inode *inode,
933                                        struct page *locked_page,
934                               u64 start, u64 end, int *page_started, int force,
935                               unsigned long *nr_written)
936 {
937         struct btrfs_root *root = BTRFS_I(inode)->root;
938         struct btrfs_trans_handle *trans;
939         struct extent_buffer *leaf;
940         struct btrfs_path *path;
941         struct btrfs_file_extent_item *fi;
942         struct btrfs_key found_key;
943         u64 cow_start;
944         u64 cur_offset;
945         u64 extent_end;
946         u64 extent_offset;
947         u64 disk_bytenr;
948         u64 num_bytes;
949         int extent_type;
950         int ret;
951         int type;
952         int nocow;
953         int check_prev = 1;
954
955         path = btrfs_alloc_path();
956         BUG_ON(!path);
957         trans = btrfs_join_transaction(root, 1);
958         BUG_ON(!trans);
959
960         cow_start = (u64)-1;
961         cur_offset = start;
962         while (1) {
963                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
964                                                cur_offset, 0);
965                 BUG_ON(ret < 0);
966                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
967                         leaf = path->nodes[0];
968                         btrfs_item_key_to_cpu(leaf, &found_key,
969                                               path->slots[0] - 1);
970                         if (found_key.objectid == inode->i_ino &&
971                             found_key.type == BTRFS_EXTENT_DATA_KEY)
972                                 path->slots[0]--;
973                 }
974                 check_prev = 0;
975 next_slot:
976                 leaf = path->nodes[0];
977                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
978                         ret = btrfs_next_leaf(root, path);
979                         if (ret < 0)
980                                 BUG_ON(1);
981                         if (ret > 0)
982                                 break;
983                         leaf = path->nodes[0];
984                 }
985
986                 nocow = 0;
987                 disk_bytenr = 0;
988                 num_bytes = 0;
989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
990
991                 if (found_key.objectid > inode->i_ino ||
992                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
993                     found_key.offset > end)
994                         break;
995
996                 if (found_key.offset > cur_offset) {
997                         extent_end = found_key.offset;
998                         goto out_check;
999                 }
1000
1001                 fi = btrfs_item_ptr(leaf, path->slots[0],
1002                                     struct btrfs_file_extent_item);
1003                 extent_type = btrfs_file_extent_type(leaf, fi);
1004
1005                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1006                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1007                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1008                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1009                         extent_end = found_key.offset +
1010                                 btrfs_file_extent_num_bytes(leaf, fi);
1011                         if (extent_end <= start) {
1012                                 path->slots[0]++;
1013                                 goto next_slot;
1014                         }
1015                         if (disk_bytenr == 0)
1016                                 goto out_check;
1017                         if (btrfs_file_extent_compression(leaf, fi) ||
1018                             btrfs_file_extent_encryption(leaf, fi) ||
1019                             btrfs_file_extent_other_encoding(leaf, fi))
1020                                 goto out_check;
1021                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022                                 goto out_check;
1023                         if (btrfs_extent_readonly(root, disk_bytenr))
1024                                 goto out_check;
1025                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026                                                   found_key.offset -
1027                                                   extent_offset, disk_bytenr))
1028                                 goto out_check;
1029                         disk_bytenr += extent_offset;
1030                         disk_bytenr += cur_offset - found_key.offset;
1031                         num_bytes = min(end + 1, extent_end) - cur_offset;
1032                         /*
1033                          * force cow if csum exists in the range.
1034                          * this ensure that csum for a given extent are
1035                          * either valid or do not exist.
1036                          */
1037                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1038                                 goto out_check;
1039                         nocow = 1;
1040                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1041                         extent_end = found_key.offset +
1042                                 btrfs_file_extent_inline_len(leaf, fi);
1043                         extent_end = ALIGN(extent_end, root->sectorsize);
1044                 } else {
1045                         BUG_ON(1);
1046                 }
1047 out_check:
1048                 if (extent_end <= start) {
1049                         path->slots[0]++;
1050                         goto next_slot;
1051                 }
1052                 if (!nocow) {
1053                         if (cow_start == (u64)-1)
1054                                 cow_start = cur_offset;
1055                         cur_offset = extent_end;
1056                         if (cur_offset > end)
1057                                 break;
1058                         path->slots[0]++;
1059                         goto next_slot;
1060                 }
1061
1062                 btrfs_release_path(root, path);
1063                 if (cow_start != (u64)-1) {
1064                         ret = cow_file_range(inode, locked_page, cow_start,
1065                                         found_key.offset - 1, page_started,
1066                                         nr_written, 1);
1067                         BUG_ON(ret);
1068                         cow_start = (u64)-1;
1069                 }
1070
1071                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1072                         struct extent_map *em;
1073                         struct extent_map_tree *em_tree;
1074                         em_tree = &BTRFS_I(inode)->extent_tree;
1075                         em = alloc_extent_map(GFP_NOFS);
1076                         em->start = cur_offset;
1077                         em->orig_start = em->start;
1078                         em->len = num_bytes;
1079                         em->block_len = num_bytes;
1080                         em->block_start = disk_bytenr;
1081                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1082                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1083                         while (1) {
1084                                 spin_lock(&em_tree->lock);
1085                                 ret = add_extent_mapping(em_tree, em);
1086                                 spin_unlock(&em_tree->lock);
1087                                 if (ret != -EEXIST) {
1088                                         free_extent_map(em);
1089                                         break;
1090                                 }
1091                                 btrfs_drop_extent_cache(inode, em->start,
1092                                                 em->start + em->len - 1, 0);
1093                         }
1094                         type = BTRFS_ORDERED_PREALLOC;
1095                 } else {
1096                         type = BTRFS_ORDERED_NOCOW;
1097                 }
1098
1099                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1100                                                num_bytes, num_bytes, type);
1101                 BUG_ON(ret);
1102
1103                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1104                                         cur_offset, cur_offset + num_bytes - 1,
1105                                         locked_page, 1, 1, 1, 0, 0, 0);
1106                 cur_offset = extent_end;
1107                 if (cur_offset > end)
1108                         break;
1109         }
1110         btrfs_release_path(root, path);
1111
1112         if (cur_offset <= end && cow_start == (u64)-1)
1113                 cow_start = cur_offset;
1114         if (cow_start != (u64)-1) {
1115                 ret = cow_file_range(inode, locked_page, cow_start, end,
1116                                      page_started, nr_written, 1);
1117                 BUG_ON(ret);
1118         }
1119
1120         ret = btrfs_end_transaction(trans, root);
1121         BUG_ON(ret);
1122         btrfs_free_path(path);
1123         return 0;
1124 }
1125
1126 /*
1127  * extent_io.c call back to do delayed allocation processing
1128  */
1129 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1130                               u64 start, u64 end, int *page_started,
1131                               unsigned long *nr_written)
1132 {
1133         int ret;
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135
1136         if (btrfs_test_flag(inode, NODATACOW))
1137                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1138                                          page_started, 1, nr_written);
1139         else if (btrfs_test_flag(inode, PREALLOC))
1140                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1141                                          page_started, 0, nr_written);
1142         else if (!btrfs_test_opt(root, COMPRESS))
1143                 ret = cow_file_range(inode, locked_page, start, end,
1144                                       page_started, nr_written, 1);
1145         else
1146                 ret = cow_file_range_async(inode, locked_page, start, end,
1147                                            page_started, nr_written);
1148         return ret;
1149 }
1150
1151 /*
1152  * extent_io.c set_bit_hook, used to track delayed allocation
1153  * bytes in this file, and to maintain the list of inodes that
1154  * have pending delalloc work to be done.
1155  */
1156 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1157                        unsigned long old, unsigned long bits)
1158 {
1159         /*
1160          * set_bit and clear bit hooks normally require _irqsave/restore
1161          * but in this case, we are only testeing for the DELALLOC
1162          * bit, which is only set or cleared with irqs on
1163          */
1164         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165                 struct btrfs_root *root = BTRFS_I(inode)->root;
1166                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1167                 spin_lock(&root->fs_info->delalloc_lock);
1168                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1169                 root->fs_info->delalloc_bytes += end - start + 1;
1170                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1171                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1172                                       &root->fs_info->delalloc_inodes);
1173                 }
1174                 spin_unlock(&root->fs_info->delalloc_lock);
1175         }
1176         return 0;
1177 }
1178
1179 /*
1180  * extent_io.c clear_bit_hook, see set_bit_hook for why
1181  */
1182 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1183                          unsigned long old, unsigned long bits)
1184 {
1185         /*
1186          * set_bit and clear bit hooks normally require _irqsave/restore
1187          * but in this case, we are only testeing for the DELALLOC
1188          * bit, which is only set or cleared with irqs on
1189          */
1190         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1191                 struct btrfs_root *root = BTRFS_I(inode)->root;
1192
1193                 spin_lock(&root->fs_info->delalloc_lock);
1194                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1195                         printk(KERN_INFO "btrfs warning: delalloc account "
1196                                "%llu %llu\n",
1197                                (unsigned long long)end - start + 1,
1198                                (unsigned long long)
1199                                root->fs_info->delalloc_bytes);
1200                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1201                         root->fs_info->delalloc_bytes = 0;
1202                         BTRFS_I(inode)->delalloc_bytes = 0;
1203                 } else {
1204                         btrfs_delalloc_free_space(root, inode,
1205                                                   end - start + 1);
1206                         root->fs_info->delalloc_bytes -= end - start + 1;
1207                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1208                 }
1209                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1210                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1211                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1212                 }
1213                 spin_unlock(&root->fs_info->delalloc_lock);
1214         }
1215         return 0;
1216 }
1217
1218 /*
1219  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1220  * we don't create bios that span stripes or chunks
1221  */
1222 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1223                          size_t size, struct bio *bio,
1224                          unsigned long bio_flags)
1225 {
1226         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1227         struct btrfs_mapping_tree *map_tree;
1228         u64 logical = (u64)bio->bi_sector << 9;
1229         u64 length = 0;
1230         u64 map_length;
1231         int ret;
1232
1233         if (bio_flags & EXTENT_BIO_COMPRESSED)
1234                 return 0;
1235
1236         length = bio->bi_size;
1237         map_tree = &root->fs_info->mapping_tree;
1238         map_length = length;
1239         ret = btrfs_map_block(map_tree, READ, logical,
1240                               &map_length, NULL, 0);
1241
1242         if (map_length < length + size)
1243                 return 1;
1244         return 0;
1245 }
1246
1247 /*
1248  * in order to insert checksums into the metadata in large chunks,
1249  * we wait until bio submission time.   All the pages in the bio are
1250  * checksummed and sums are attached onto the ordered extent record.
1251  *
1252  * At IO completion time the cums attached on the ordered extent record
1253  * are inserted into the btree
1254  */
1255 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1256                                     struct bio *bio, int mirror_num,
1257                                     unsigned long bio_flags)
1258 {
1259         struct btrfs_root *root = BTRFS_I(inode)->root;
1260         int ret = 0;
1261
1262         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1263         BUG_ON(ret);
1264         return 0;
1265 }
1266
1267 /*
1268  * in order to insert checksums into the metadata in large chunks,
1269  * we wait until bio submission time.   All the pages in the bio are
1270  * checksummed and sums are attached onto the ordered extent record.
1271  *
1272  * At IO completion time the cums attached on the ordered extent record
1273  * are inserted into the btree
1274  */
1275 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1276                           int mirror_num, unsigned long bio_flags)
1277 {
1278         struct btrfs_root *root = BTRFS_I(inode)->root;
1279         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1280 }
1281
1282 /*
1283  * extent_io.c submission hook. This does the right thing for csum calculation
1284  * on write, or reading the csums from the tree before a read
1285  */
1286 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1287                           int mirror_num, unsigned long bio_flags)
1288 {
1289         struct btrfs_root *root = BTRFS_I(inode)->root;
1290         int ret = 0;
1291         int skip_sum;
1292
1293         skip_sum = btrfs_test_flag(inode, NODATASUM);
1294
1295         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1296         BUG_ON(ret);
1297
1298         if (!(rw & (1 << BIO_RW))) {
1299                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1300                         return btrfs_submit_compressed_read(inode, bio,
1301                                                     mirror_num, bio_flags);
1302                 } else if (!skip_sum)
1303                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1304                 goto mapit;
1305         } else if (!skip_sum) {
1306                 /* csum items have already been cloned */
1307                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1308                         goto mapit;
1309                 /* we're doing a write, do the async checksumming */
1310                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1311                                    inode, rw, bio, mirror_num,
1312                                    bio_flags, __btrfs_submit_bio_start,
1313                                    __btrfs_submit_bio_done);
1314         }
1315
1316 mapit:
1317         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1318 }
1319
1320 /*
1321  * given a list of ordered sums record them in the inode.  This happens
1322  * at IO completion time based on sums calculated at bio submission time.
1323  */
1324 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1325                              struct inode *inode, u64 file_offset,
1326                              struct list_head *list)
1327 {
1328         struct btrfs_ordered_sum *sum;
1329
1330         btrfs_set_trans_block_group(trans, inode);
1331
1332         list_for_each_entry(sum, list, list) {
1333                 btrfs_csum_file_blocks(trans,
1334                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1335         }
1336         return 0;
1337 }
1338
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340 {
1341         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1342                 WARN_ON(1);
1343         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344                                    GFP_NOFS);
1345 }
1346
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1349         struct page *page;
1350         struct btrfs_work work;
1351 };
1352
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1354 {
1355         struct btrfs_writepage_fixup *fixup;
1356         struct btrfs_ordered_extent *ordered;
1357         struct page *page;
1358         struct inode *inode;
1359         u64 page_start;
1360         u64 page_end;
1361
1362         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363         page = fixup->page;
1364 again:
1365         lock_page(page);
1366         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367                 ClearPageChecked(page);
1368                 goto out_page;
1369         }
1370
1371         inode = page->mapping->host;
1372         page_start = page_offset(page);
1373         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1376
1377         /* already ordered? We're done */
1378         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379                              EXTENT_ORDERED, 0)) {
1380                 goto out;
1381         }
1382
1383         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384         if (ordered) {
1385                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386                               page_end, GFP_NOFS);
1387                 unlock_page(page);
1388                 btrfs_start_ordered_extent(inode, ordered, 1);
1389                 goto again;
1390         }
1391
1392         btrfs_set_extent_delalloc(inode, page_start, page_end);
1393         ClearPageChecked(page);
1394 out:
1395         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396 out_page:
1397         unlock_page(page);
1398         page_cache_release(page);
1399 }
1400
1401 /*
1402  * There are a few paths in the higher layers of the kernel that directly
1403  * set the page dirty bit without asking the filesystem if it is a
1404  * good idea.  This causes problems because we want to make sure COW
1405  * properly happens and the data=ordered rules are followed.
1406  *
1407  * In our case any range that doesn't have the ORDERED bit set
1408  * hasn't been properly setup for IO.  We kick off an async process
1409  * to fix it up.  The async helper will wait for ordered extents, set
1410  * the delalloc bit and make it safe to write the page.
1411  */
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 {
1414         struct inode *inode = page->mapping->host;
1415         struct btrfs_writepage_fixup *fixup;
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret;
1418
1419         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420                              EXTENT_ORDERED, 0);
1421         if (ret)
1422                 return 0;
1423
1424         if (PageChecked(page))
1425                 return -EAGAIN;
1426
1427         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428         if (!fixup)
1429                 return -EAGAIN;
1430
1431         SetPageChecked(page);
1432         page_cache_get(page);
1433         fixup->work.func = btrfs_writepage_fixup_worker;
1434         fixup->page = page;
1435         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436         return -EAGAIN;
1437 }
1438
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440                                        struct inode *inode, u64 file_pos,
1441                                        u64 disk_bytenr, u64 disk_num_bytes,
1442                                        u64 num_bytes, u64 ram_bytes,
1443                                        u64 locked_end,
1444                                        u8 compression, u8 encryption,
1445                                        u16 other_encoding, int extent_type)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         struct btrfs_file_extent_item *fi;
1449         struct btrfs_path *path;
1450         struct extent_buffer *leaf;
1451         struct btrfs_key ins;
1452         u64 hint;
1453         int ret;
1454
1455         path = btrfs_alloc_path();
1456         BUG_ON(!path);
1457
1458         path->leave_spinning = 1;
1459         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1460                                  file_pos + num_bytes, locked_end,
1461                                  file_pos, &hint);
1462         BUG_ON(ret);
1463
1464         ins.objectid = inode->i_ino;
1465         ins.offset = file_pos;
1466         ins.type = BTRFS_EXTENT_DATA_KEY;
1467         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1468         BUG_ON(ret);
1469         leaf = path->nodes[0];
1470         fi = btrfs_item_ptr(leaf, path->slots[0],
1471                             struct btrfs_file_extent_item);
1472         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1473         btrfs_set_file_extent_type(leaf, fi, extent_type);
1474         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1475         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1476         btrfs_set_file_extent_offset(leaf, fi, 0);
1477         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1478         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1479         btrfs_set_file_extent_compression(leaf, fi, compression);
1480         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1481         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1482
1483         btrfs_unlock_up_safe(path, 1);
1484         btrfs_set_lock_blocking(leaf);
1485
1486         btrfs_mark_buffer_dirty(leaf);
1487
1488         inode_add_bytes(inode, num_bytes);
1489         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1490
1491         ins.objectid = disk_bytenr;
1492         ins.offset = disk_num_bytes;
1493         ins.type = BTRFS_EXTENT_ITEM_KEY;
1494         ret = btrfs_alloc_reserved_file_extent(trans, root,
1495                                         root->root_key.objectid,
1496                                         inode->i_ino, file_pos, &ins);
1497         BUG_ON(ret);
1498         btrfs_free_path(path);
1499
1500         return 0;
1501 }
1502
1503 /*
1504  * helper function for btrfs_finish_ordered_io, this
1505  * just reads in some of the csum leaves to prime them into ram
1506  * before we start the transaction.  It limits the amount of btree
1507  * reads required while inside the transaction.
1508  */
1509 static noinline void reada_csum(struct btrfs_root *root,
1510                                 struct btrfs_path *path,
1511                                 struct btrfs_ordered_extent *ordered_extent)
1512 {
1513         struct btrfs_ordered_sum *sum;
1514         u64 bytenr;
1515
1516         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1517                          list);
1518         bytenr = sum->sums[0].bytenr;
1519
1520         /*
1521          * we don't care about the results, the point of this search is
1522          * just to get the btree leaves into ram
1523          */
1524         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1525 }
1526
1527 /* as ordered data IO finishes, this gets called so we can finish
1528  * an ordered extent if the range of bytes in the file it covers are
1529  * fully written.
1530  */
1531 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1532 {
1533         struct btrfs_root *root = BTRFS_I(inode)->root;
1534         struct btrfs_trans_handle *trans;
1535         struct btrfs_ordered_extent *ordered_extent = NULL;
1536         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1537         struct btrfs_path *path;
1538         int compressed = 0;
1539         int ret;
1540
1541         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1542         if (!ret)
1543                 return 0;
1544
1545         /*
1546          * before we join the transaction, try to do some of our IO.
1547          * This will limit the amount of IO that we have to do with
1548          * the transaction running.  We're unlikely to need to do any
1549          * IO if the file extents are new, the disk_i_size checks
1550          * covers the most common case.
1551          */
1552         if (start < BTRFS_I(inode)->disk_i_size) {
1553                 path = btrfs_alloc_path();
1554                 if (path) {
1555                         ret = btrfs_lookup_file_extent(NULL, root, path,
1556                                                        inode->i_ino,
1557                                                        start, 0);
1558                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1559                                                                      start);
1560                         if (!list_empty(&ordered_extent->list)) {
1561                                 btrfs_release_path(root, path);
1562                                 reada_csum(root, path, ordered_extent);
1563                         }
1564                         btrfs_free_path(path);
1565                 }
1566         }
1567
1568         trans = btrfs_join_transaction(root, 1);
1569
1570         if (!ordered_extent)
1571                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1572         BUG_ON(!ordered_extent);
1573         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1574                 goto nocow;
1575
1576         lock_extent(io_tree, ordered_extent->file_offset,
1577                     ordered_extent->file_offset + ordered_extent->len - 1,
1578                     GFP_NOFS);
1579
1580         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1581                 compressed = 1;
1582         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1583                 BUG_ON(compressed);
1584                 ret = btrfs_mark_extent_written(trans, root, inode,
1585                                                 ordered_extent->file_offset,
1586                                                 ordered_extent->file_offset +
1587                                                 ordered_extent->len);
1588                 BUG_ON(ret);
1589         } else {
1590                 ret = insert_reserved_file_extent(trans, inode,
1591                                                 ordered_extent->file_offset,
1592                                                 ordered_extent->start,
1593                                                 ordered_extent->disk_len,
1594                                                 ordered_extent->len,
1595                                                 ordered_extent->len,
1596                                                 ordered_extent->file_offset +
1597                                                 ordered_extent->len,
1598                                                 compressed, 0, 0,
1599                                                 BTRFS_FILE_EXTENT_REG);
1600                 BUG_ON(ret);
1601         }
1602         unlock_extent(io_tree, ordered_extent->file_offset,
1603                     ordered_extent->file_offset + ordered_extent->len - 1,
1604                     GFP_NOFS);
1605 nocow:
1606         add_pending_csums(trans, inode, ordered_extent->file_offset,
1607                           &ordered_extent->list);
1608
1609         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1610         btrfs_ordered_update_i_size(inode, ordered_extent);
1611         btrfs_update_inode(trans, root, inode);
1612         btrfs_remove_ordered_extent(inode, ordered_extent);
1613         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1614
1615         /* once for us */
1616         btrfs_put_ordered_extent(ordered_extent);
1617         /* once for the tree */
1618         btrfs_put_ordered_extent(ordered_extent);
1619
1620         btrfs_end_transaction(trans, root);
1621         return 0;
1622 }
1623
1624 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1625                                 struct extent_state *state, int uptodate)
1626 {
1627         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1628 }
1629
1630 /*
1631  * When IO fails, either with EIO or csum verification fails, we
1632  * try other mirrors that might have a good copy of the data.  This
1633  * io_failure_record is used to record state as we go through all the
1634  * mirrors.  If another mirror has good data, the page is set up to date
1635  * and things continue.  If a good mirror can't be found, the original
1636  * bio end_io callback is called to indicate things have failed.
1637  */
1638 struct io_failure_record {
1639         struct page *page;
1640         u64 start;
1641         u64 len;
1642         u64 logical;
1643         unsigned long bio_flags;
1644         int last_mirror;
1645 };
1646
1647 static int btrfs_io_failed_hook(struct bio *failed_bio,
1648                          struct page *page, u64 start, u64 end,
1649                          struct extent_state *state)
1650 {
1651         struct io_failure_record *failrec = NULL;
1652         u64 private;
1653         struct extent_map *em;
1654         struct inode *inode = page->mapping->host;
1655         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1656         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1657         struct bio *bio;
1658         int num_copies;
1659         int ret;
1660         int rw;
1661         u64 logical;
1662
1663         ret = get_state_private(failure_tree, start, &private);
1664         if (ret) {
1665                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1666                 if (!failrec)
1667                         return -ENOMEM;
1668                 failrec->start = start;
1669                 failrec->len = end - start + 1;
1670                 failrec->last_mirror = 0;
1671                 failrec->bio_flags = 0;
1672
1673                 spin_lock(&em_tree->lock);
1674                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1675                 if (em->start > start || em->start + em->len < start) {
1676                         free_extent_map(em);
1677                         em = NULL;
1678                 }
1679                 spin_unlock(&em_tree->lock);
1680
1681                 if (!em || IS_ERR(em)) {
1682                         kfree(failrec);
1683                         return -EIO;
1684                 }
1685                 logical = start - em->start;
1686                 logical = em->block_start + logical;
1687                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1688                         logical = em->block_start;
1689                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1690                 }
1691                 failrec->logical = logical;
1692                 free_extent_map(em);
1693                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1694                                 EXTENT_DIRTY, GFP_NOFS);
1695                 set_state_private(failure_tree, start,
1696                                  (u64)(unsigned long)failrec);
1697         } else {
1698                 failrec = (struct io_failure_record *)(unsigned long)private;
1699         }
1700         num_copies = btrfs_num_copies(
1701                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1702                               failrec->logical, failrec->len);
1703         failrec->last_mirror++;
1704         if (!state) {
1705                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1706                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1707                                                     failrec->start,
1708                                                     EXTENT_LOCKED);
1709                 if (state && state->start != failrec->start)
1710                         state = NULL;
1711                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1712         }
1713         if (!state || failrec->last_mirror > num_copies) {
1714                 set_state_private(failure_tree, failrec->start, 0);
1715                 clear_extent_bits(failure_tree, failrec->start,
1716                                   failrec->start + failrec->len - 1,
1717                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1718                 kfree(failrec);
1719                 return -EIO;
1720         }
1721         bio = bio_alloc(GFP_NOFS, 1);
1722         bio->bi_private = state;
1723         bio->bi_end_io = failed_bio->bi_end_io;
1724         bio->bi_sector = failrec->logical >> 9;
1725         bio->bi_bdev = failed_bio->bi_bdev;
1726         bio->bi_size = 0;
1727
1728         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1729         if (failed_bio->bi_rw & (1 << BIO_RW))
1730                 rw = WRITE;
1731         else
1732                 rw = READ;
1733
1734         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1735                                                       failrec->last_mirror,
1736                                                       failrec->bio_flags);
1737         return 0;
1738 }
1739
1740 /*
1741  * each time an IO finishes, we do a fast check in the IO failure tree
1742  * to see if we need to process or clean up an io_failure_record
1743  */
1744 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1745 {
1746         u64 private;
1747         u64 private_failure;
1748         struct io_failure_record *failure;
1749         int ret;
1750
1751         private = 0;
1752         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1753                              (u64)-1, 1, EXTENT_DIRTY)) {
1754                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1755                                         start, &private_failure);
1756                 if (ret == 0) {
1757                         failure = (struct io_failure_record *)(unsigned long)
1758                                    private_failure;
1759                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1760                                           failure->start, 0);
1761                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1762                                           failure->start,
1763                                           failure->start + failure->len - 1,
1764                                           EXTENT_DIRTY | EXTENT_LOCKED,
1765                                           GFP_NOFS);
1766                         kfree(failure);
1767                 }
1768         }
1769         return 0;
1770 }
1771
1772 /*
1773  * when reads are done, we need to check csums to verify the data is correct
1774  * if there's a match, we allow the bio to finish.  If not, we go through
1775  * the io_failure_record routines to find good copies
1776  */
1777 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1778                                struct extent_state *state)
1779 {
1780         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1781         struct inode *inode = page->mapping->host;
1782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1783         char *kaddr;
1784         u64 private = ~(u32)0;
1785         int ret;
1786         struct btrfs_root *root = BTRFS_I(inode)->root;
1787         u32 csum = ~(u32)0;
1788
1789         if (PageChecked(page)) {
1790                 ClearPageChecked(page);
1791                 goto good;
1792         }
1793         if (btrfs_test_flag(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                 BTRFS_I(inode)->i_acl = NULL;
2126                 BTRFS_I(inode)->i_default_acl = NULL;
2127         }
2128
2129         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2130                                                 alloc_group_block, 0);
2131         btrfs_free_path(path);
2132         inode_item = NULL;
2133
2134         switch (inode->i_mode & S_IFMT) {
2135         case S_IFREG:
2136                 inode->i_mapping->a_ops = &btrfs_aops;
2137                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2138                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2139                 inode->i_fop = &btrfs_file_operations;
2140                 inode->i_op = &btrfs_file_inode_operations;
2141                 break;
2142         case S_IFDIR:
2143                 inode->i_fop = &btrfs_dir_file_operations;
2144                 if (root == root->fs_info->tree_root)
2145                         inode->i_op = &btrfs_dir_ro_inode_operations;
2146                 else
2147                         inode->i_op = &btrfs_dir_inode_operations;
2148                 break;
2149         case S_IFLNK:
2150                 inode->i_op = &btrfs_symlink_inode_operations;
2151                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2152                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2153                 break;
2154         default:
2155                 inode->i_op = &btrfs_special_inode_operations;
2156                 init_special_inode(inode, inode->i_mode, rdev);
2157                 break;
2158         }
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         dir->i_sb->s_dirt = 1;
2323 out:
2324         return ret;
2325 }
2326
2327 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2328 {
2329         struct btrfs_root *root;
2330         struct btrfs_trans_handle *trans;
2331         struct inode *inode = dentry->d_inode;
2332         int ret;
2333         unsigned long nr = 0;
2334
2335         root = BTRFS_I(dir)->root;
2336
2337         trans = btrfs_start_transaction(root, 1);
2338
2339         btrfs_set_trans_block_group(trans, dir);
2340
2341         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2342
2343         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2344                                  dentry->d_name.name, dentry->d_name.len);
2345
2346         if (inode->i_nlink == 0)
2347                 ret = btrfs_orphan_add(trans, inode);
2348
2349         nr = trans->blocks_used;
2350
2351         btrfs_end_transaction_throttle(trans, root);
2352         btrfs_btree_balance_dirty(root, nr);
2353         return ret;
2354 }
2355
2356 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2357 {
2358         struct inode *inode = dentry->d_inode;
2359         int err = 0;
2360         int ret;
2361         struct btrfs_root *root = BTRFS_I(dir)->root;
2362         struct btrfs_trans_handle *trans;
2363         unsigned long nr = 0;
2364
2365         /*
2366          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2367          * the root of a subvolume or snapshot
2368          */
2369         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2370             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2371                 return -ENOTEMPTY;
2372         }
2373
2374         trans = btrfs_start_transaction(root, 1);
2375         btrfs_set_trans_block_group(trans, dir);
2376
2377         err = btrfs_orphan_add(trans, inode);
2378         if (err)
2379                 goto fail_trans;
2380
2381         /* now the directory is empty */
2382         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2383                                  dentry->d_name.name, dentry->d_name.len);
2384         if (!err)
2385                 btrfs_i_size_write(inode, 0);
2386
2387 fail_trans:
2388         nr = trans->blocks_used;
2389         ret = btrfs_end_transaction_throttle(trans, root);
2390         btrfs_btree_balance_dirty(root, nr);
2391
2392         if (ret && !err)
2393                 err = ret;
2394         return err;
2395 }
2396
2397 #if 0
2398 /*
2399  * when truncating bytes in a file, it is possible to avoid reading
2400  * the leaves that contain only checksum items.  This can be the
2401  * majority of the IO required to delete a large file, but it must
2402  * be done carefully.
2403  *
2404  * The keys in the level just above the leaves are checked to make sure
2405  * the lowest key in a given leaf is a csum key, and starts at an offset
2406  * after the new  size.
2407  *
2408  * Then the key for the next leaf is checked to make sure it also has
2409  * a checksum item for the same file.  If it does, we know our target leaf
2410  * contains only checksum items, and it can be safely freed without reading
2411  * it.
2412  *
2413  * This is just an optimization targeted at large files.  It may do
2414  * nothing.  It will return 0 unless things went badly.
2415  */
2416 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2417                                      struct btrfs_root *root,
2418                                      struct btrfs_path *path,
2419                                      struct inode *inode, u64 new_size)
2420 {
2421         struct btrfs_key key;
2422         int ret;
2423         int nritems;
2424         struct btrfs_key found_key;
2425         struct btrfs_key other_key;
2426         struct btrfs_leaf_ref *ref;
2427         u64 leaf_gen;
2428         u64 leaf_start;
2429
2430         path->lowest_level = 1;
2431         key.objectid = inode->i_ino;
2432         key.type = BTRFS_CSUM_ITEM_KEY;
2433         key.offset = new_size;
2434 again:
2435         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2436         if (ret < 0)
2437                 goto out;
2438
2439         if (path->nodes[1] == NULL) {
2440                 ret = 0;
2441                 goto out;
2442         }
2443         ret = 0;
2444         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2445         nritems = btrfs_header_nritems(path->nodes[1]);
2446
2447         if (!nritems)
2448                 goto out;
2449
2450         if (path->slots[1] >= nritems)
2451                 goto next_node;
2452
2453         /* did we find a key greater than anything we want to delete? */
2454         if (found_key.objectid > inode->i_ino ||
2455            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2456                 goto out;
2457
2458         /* we check the next key in the node to make sure the leave contains
2459          * only checksum items.  This comparison doesn't work if our
2460          * leaf is the last one in the node
2461          */
2462         if (path->slots[1] + 1 >= nritems) {
2463 next_node:
2464                 /* search forward from the last key in the node, this
2465                  * will bring us into the next node in the tree
2466                  */
2467                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2468
2469                 /* unlikely, but we inc below, so check to be safe */
2470                 if (found_key.offset == (u64)-1)
2471                         goto out;
2472
2473                 /* search_forward needs a path with locks held, do the
2474                  * search again for the original key.  It is possible
2475                  * this will race with a balance and return a path that
2476                  * we could modify, but this drop is just an optimization
2477                  * and is allowed to miss some leaves.
2478                  */
2479                 btrfs_release_path(root, path);
2480                 found_key.offset++;
2481
2482                 /* setup a max key for search_forward */
2483                 other_key.offset = (u64)-1;
2484                 other_key.type = key.type;
2485                 other_key.objectid = key.objectid;
2486
2487                 path->keep_locks = 1;
2488                 ret = btrfs_search_forward(root, &found_key, &other_key,
2489                                            path, 0, 0);
2490                 path->keep_locks = 0;
2491                 if (ret || found_key.objectid != key.objectid ||
2492                     found_key.type != key.type) {
2493                         ret = 0;
2494                         goto out;
2495                 }
2496
2497                 key.offset = found_key.offset;
2498                 btrfs_release_path(root, path);
2499                 cond_resched();
2500                 goto again;
2501         }
2502
2503         /* we know there's one more slot after us in the tree,
2504          * read that key so we can verify it is also a checksum item
2505          */
2506         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2507
2508         if (found_key.objectid < inode->i_ino)
2509                 goto next_key;
2510
2511         if (found_key.type != key.type || found_key.offset < new_size)
2512                 goto next_key;
2513
2514         /*
2515          * if the key for the next leaf isn't a csum key from this objectid,
2516          * we can't be sure there aren't good items inside this leaf.
2517          * Bail out
2518          */
2519         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2520                 goto out;
2521
2522         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2523         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2524         /*
2525          * it is safe to delete this leaf, it contains only
2526          * csum items from this inode at an offset >= new_size
2527          */
2528         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2529         BUG_ON(ret);
2530
2531         if (root->ref_cows && leaf_gen < trans->transid) {
2532                 ref = btrfs_alloc_leaf_ref(root, 0);
2533                 if (ref) {
2534                         ref->root_gen = root->root_key.offset;
2535                         ref->bytenr = leaf_start;
2536                         ref->owner = 0;
2537                         ref->generation = leaf_gen;
2538                         ref->nritems = 0;
2539
2540                         btrfs_sort_leaf_ref(ref);
2541
2542                         ret = btrfs_add_leaf_ref(root, ref, 0);
2543                         WARN_ON(ret);
2544                         btrfs_free_leaf_ref(root, ref);
2545                 } else {
2546                         WARN_ON(1);
2547                 }
2548         }
2549 next_key:
2550         btrfs_release_path(root, path);
2551
2552         if (other_key.objectid == inode->i_ino &&
2553             other_key.type == key.type && other_key.offset > key.offset) {
2554                 key.offset = other_key.offset;
2555                 cond_resched();
2556                 goto again;
2557         }
2558         ret = 0;
2559 out:
2560         /* fixup any changes we've made to the path */
2561         path->lowest_level = 0;
2562         path->keep_locks = 0;
2563         btrfs_release_path(root, path);
2564         return ret;
2565 }
2566
2567 #endif
2568
2569 /*
2570  * this can truncate away extent items, csum items and directory items.
2571  * It starts at a high offset and removes keys until it can't find
2572  * any higher than new_size
2573  *
2574  * csum items that cross the new i_size are truncated to the new size
2575  * as well.
2576  *
2577  * min_type is the minimum key type to truncate down to.  If set to 0, this
2578  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2579  */
2580 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2581                                         struct btrfs_root *root,
2582                                         struct inode *inode,
2583                                         u64 new_size, u32 min_type)
2584 {
2585         int ret;
2586         struct btrfs_path *path;
2587         struct btrfs_key key;
2588         struct btrfs_key found_key;
2589         u32 found_type = (u8)-1;
2590         struct extent_buffer *leaf;
2591         struct btrfs_file_extent_item *fi;
2592         u64 extent_start = 0;
2593         u64 extent_num_bytes = 0;
2594         u64 extent_offset = 0;
2595         u64 item_end = 0;
2596         int found_extent;
2597         int del_item;
2598         int pending_del_nr = 0;
2599         int pending_del_slot = 0;
2600         int extent_type = -1;
2601         int encoding;
2602         u64 mask = root->sectorsize - 1;
2603
2604         if (root->ref_cows)
2605                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2606         path = btrfs_alloc_path();
2607         path->reada = -1;
2608         BUG_ON(!path);
2609
2610         /* FIXME, add redo link to tree so we don't leak on crash */
2611         key.objectid = inode->i_ino;
2612         key.offset = (u64)-1;
2613         key.type = (u8)-1;
2614
2615 search_again:
2616         path->leave_spinning = 1;
2617         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2618         if (ret < 0)
2619                 goto error;
2620
2621         if (ret > 0) {
2622                 /* there are no items in the tree for us to truncate, we're
2623                  * done
2624                  */
2625                 if (path->slots[0] == 0) {
2626                         ret = 0;
2627                         goto error;
2628                 }
2629                 path->slots[0]--;
2630         }
2631
2632         while (1) {
2633                 fi = NULL;
2634                 leaf = path->nodes[0];
2635                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2636                 found_type = btrfs_key_type(&found_key);
2637                 encoding = 0;
2638
2639                 if (found_key.objectid != inode->i_ino)
2640                         break;
2641
2642                 if (found_type < min_type)
2643                         break;
2644
2645                 item_end = found_key.offset;
2646                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2647                         fi = btrfs_item_ptr(leaf, path->slots[0],
2648                                             struct btrfs_file_extent_item);
2649                         extent_type = btrfs_file_extent_type(leaf, fi);
2650                         encoding = btrfs_file_extent_compression(leaf, fi);
2651                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2652                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2653
2654                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2655                                 item_end +=
2656                                     btrfs_file_extent_num_bytes(leaf, fi);
2657                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2658                                 item_end += btrfs_file_extent_inline_len(leaf,
2659                                                                          fi);
2660                         }
2661                         item_end--;
2662                 }
2663                 if (item_end < new_size) {
2664                         if (found_type == BTRFS_DIR_ITEM_KEY)
2665                                 found_type = BTRFS_INODE_ITEM_KEY;
2666                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2667                                 found_type = BTRFS_EXTENT_DATA_KEY;
2668                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2669                                 found_type = BTRFS_XATTR_ITEM_KEY;
2670                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2671                                 found_type = BTRFS_INODE_REF_KEY;
2672                         else if (found_type)
2673                                 found_type--;
2674                         else
2675                                 break;
2676                         btrfs_set_key_type(&key, found_type);
2677                         goto next;
2678                 }
2679                 if (found_key.offset >= new_size)
2680                         del_item = 1;
2681                 else
2682                         del_item = 0;
2683                 found_extent = 0;
2684
2685                 /* FIXME, shrink the extent if the ref count is only 1 */
2686                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2687                         goto delete;
2688
2689                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2690                         u64 num_dec;
2691                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2692                         if (!del_item && !encoding) {
2693                                 u64 orig_num_bytes =
2694                                         btrfs_file_extent_num_bytes(leaf, fi);
2695                                 extent_num_bytes = new_size -
2696                                         found_key.offset + root->sectorsize - 1;
2697                                 extent_num_bytes = extent_num_bytes &
2698                                         ~((u64)root->sectorsize - 1);
2699                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2700                                                          extent_num_bytes);
2701                                 num_dec = (orig_num_bytes -
2702                                            extent_num_bytes);
2703                                 if (root->ref_cows && extent_start != 0)
2704                                         inode_sub_bytes(inode, num_dec);
2705                                 btrfs_mark_buffer_dirty(leaf);
2706                         } else {
2707                                 extent_num_bytes =
2708                                         btrfs_file_extent_disk_num_bytes(leaf,
2709                                                                          fi);
2710                                 extent_offset = found_key.offset -
2711                                         btrfs_file_extent_offset(leaf, fi);
2712
2713                                 /* FIXME blocksize != 4096 */
2714                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2715                                 if (extent_start != 0) {
2716                                         found_extent = 1;
2717                                         if (root->ref_cows)
2718                                                 inode_sub_bytes(inode, num_dec);
2719                                 }
2720                         }
2721                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2722                         /*
2723                          * we can't truncate inline items that have had
2724                          * special encodings
2725                          */
2726                         if (!del_item &&
2727                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2728                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2729                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2730                                 u32 size = new_size - found_key.offset;
2731
2732                                 if (root->ref_cows) {
2733                                         inode_sub_bytes(inode, item_end + 1 -
2734                                                         new_size);
2735                                 }
2736                                 size =
2737                                     btrfs_file_extent_calc_inline_size(size);
2738                                 ret = btrfs_truncate_item(trans, root, path,
2739                                                           size, 1);
2740                                 BUG_ON(ret);
2741                         } else if (root->ref_cows) {
2742                                 inode_sub_bytes(inode, item_end + 1 -
2743                                                 found_key.offset);
2744                         }
2745                 }
2746 delete:
2747                 if (del_item) {
2748                         if (!pending_del_nr) {
2749                                 /* no pending yet, add ourselves */
2750                                 pending_del_slot = path->slots[0];
2751                                 pending_del_nr = 1;
2752                         } else if (pending_del_nr &&
2753                                    path->slots[0] + 1 == pending_del_slot) {
2754                                 /* hop on the pending chunk */
2755                                 pending_del_nr++;
2756                                 pending_del_slot = path->slots[0];
2757                         } else {
2758                                 BUG();
2759                         }
2760                 } else {
2761                         break;
2762                 }
2763                 if (found_extent && root->ref_cows) {
2764                         btrfs_set_path_blocking(path);
2765                         ret = btrfs_free_extent(trans, root, extent_start,
2766                                                 extent_num_bytes, 0,
2767                                                 btrfs_header_owner(leaf),
2768                                                 inode->i_ino, extent_offset);
2769                         BUG_ON(ret);
2770                 }
2771 next:
2772                 if (path->slots[0] == 0) {
2773                         if (pending_del_nr)
2774                                 goto del_pending;
2775                         btrfs_release_path(root, path);
2776                         if (found_type == BTRFS_INODE_ITEM_KEY)
2777                                 break;
2778                         goto search_again;
2779                 }
2780
2781                 path->slots[0]--;
2782                 if (pending_del_nr &&
2783                     path->slots[0] + 1 != pending_del_slot) {
2784                         struct btrfs_key debug;
2785 del_pending:
2786                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2787                                               pending_del_slot);
2788                         ret = btrfs_del_items(trans, root, path,
2789                                               pending_del_slot,
2790                                               pending_del_nr);
2791                         BUG_ON(ret);
2792                         pending_del_nr = 0;
2793                         btrfs_release_path(root, path);
2794                         if (found_type == BTRFS_INODE_ITEM_KEY)
2795                                 break;
2796                         goto search_again;
2797                 }
2798         }
2799         ret = 0;
2800 error:
2801         if (pending_del_nr) {
2802                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2803                                       pending_del_nr);
2804         }
2805         btrfs_free_path(path);
2806         inode->i_sb->s_dirt = 1;
2807         return ret;
2808 }
2809
2810 /*
2811  * taken from block_truncate_page, but does cow as it zeros out
2812  * any bytes left in the last page in the file.
2813  */
2814 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2815 {
2816         struct inode *inode = mapping->host;
2817         struct btrfs_root *root = BTRFS_I(inode)->root;
2818         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2819         struct btrfs_ordered_extent *ordered;
2820         char *kaddr;
2821         u32 blocksize = root->sectorsize;
2822         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2823         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2824         struct page *page;
2825         int ret = 0;
2826         u64 page_start;
2827         u64 page_end;
2828
2829         if ((offset & (blocksize - 1)) == 0)
2830                 goto out;
2831
2832         ret = -ENOMEM;
2833 again:
2834         page = grab_cache_page(mapping, index);
2835         if (!page)
2836                 goto out;
2837
2838         page_start = page_offset(page);
2839         page_end = page_start + PAGE_CACHE_SIZE - 1;
2840
2841         if (!PageUptodate(page)) {
2842                 ret = btrfs_readpage(NULL, page);
2843                 lock_page(page);
2844                 if (page->mapping != mapping) {
2845                         unlock_page(page);
2846                         page_cache_release(page);
2847                         goto again;
2848                 }
2849                 if (!PageUptodate(page)) {
2850                         ret = -EIO;
2851                         goto out_unlock;
2852                 }
2853         }
2854         wait_on_page_writeback(page);
2855
2856         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2857         set_page_extent_mapped(page);
2858
2859         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2860         if (ordered) {
2861                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2862                 unlock_page(page);
2863                 page_cache_release(page);
2864                 btrfs_start_ordered_extent(inode, ordered, 1);
2865                 btrfs_put_ordered_extent(ordered);
2866                 goto again;
2867         }
2868
2869         btrfs_set_extent_delalloc(inode, page_start, page_end);
2870         ret = 0;
2871         if (offset != PAGE_CACHE_SIZE) {
2872                 kaddr = kmap(page);
2873                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2874                 flush_dcache_page(page);
2875                 kunmap(page);
2876         }
2877         ClearPageChecked(page);
2878         set_page_dirty(page);
2879         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2880
2881 out_unlock:
2882         unlock_page(page);
2883         page_cache_release(page);
2884 out:
2885         return ret;
2886 }
2887
2888 int btrfs_cont_expand(struct inode *inode, loff_t size)
2889 {
2890         struct btrfs_trans_handle *trans;
2891         struct btrfs_root *root = BTRFS_I(inode)->root;
2892         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2893         struct extent_map *em;
2894         u64 mask = root->sectorsize - 1;
2895         u64 hole_start = (inode->i_size + mask) & ~mask;
2896         u64 block_end = (size + mask) & ~mask;
2897         u64 last_byte;
2898         u64 cur_offset;
2899         u64 hole_size;
2900         int err;
2901
2902         if (size <= hole_start)
2903                 return 0;
2904
2905         err = btrfs_check_metadata_free_space(root);
2906         if (err)
2907                 return err;
2908
2909         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2910
2911         while (1) {
2912                 struct btrfs_ordered_extent *ordered;
2913                 btrfs_wait_ordered_range(inode, hole_start,
2914                                          block_end - hole_start);
2915                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2916                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2917                 if (!ordered)
2918                         break;
2919                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2920                 btrfs_put_ordered_extent(ordered);
2921         }
2922
2923         trans = btrfs_start_transaction(root, 1);
2924         btrfs_set_trans_block_group(trans, inode);
2925
2926         cur_offset = hole_start;
2927         while (1) {
2928                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2929                                 block_end - cur_offset, 0);
2930                 BUG_ON(IS_ERR(em) || !em);
2931                 last_byte = min(extent_map_end(em), block_end);
2932                 last_byte = (last_byte + mask) & ~mask;
2933                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2934                         u64 hint_byte = 0;
2935                         hole_size = last_byte - cur_offset;
2936                         err = btrfs_drop_extents(trans, root, inode,
2937                                                  cur_offset,
2938                                                  cur_offset + hole_size,
2939                                                  block_end,
2940                                                  cur_offset, &hint_byte);
2941                         if (err)
2942                                 break;
2943                         err = btrfs_insert_file_extent(trans, root,
2944                                         inode->i_ino, cur_offset, 0,
2945                                         0, hole_size, 0, hole_size,
2946                                         0, 0, 0);
2947                         btrfs_drop_extent_cache(inode, hole_start,
2948                                         last_byte - 1, 0);
2949                 }
2950                 free_extent_map(em);
2951                 cur_offset = last_byte;
2952                 if (err || cur_offset >= block_end)
2953                         break;
2954         }
2955
2956         btrfs_end_transaction(trans, root);
2957         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2958         return err;
2959 }
2960
2961 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2962 {
2963         struct inode *inode = dentry->d_inode;
2964         int err;
2965
2966         err = inode_change_ok(inode, attr);
2967         if (err)
2968                 return err;
2969
2970         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2971                 if (attr->ia_size > inode->i_size) {
2972                         err = btrfs_cont_expand(inode, attr->ia_size);
2973                         if (err)
2974                                 return err;
2975                 } else if (inode->i_size > 0 &&
2976                            attr->ia_size == 0) {
2977
2978                         /* we're truncating a file that used to have good
2979                          * data down to zero.  Make sure it gets into
2980                          * the ordered flush list so that any new writes
2981                          * get down to disk quickly.
2982                          */
2983                         BTRFS_I(inode)->ordered_data_close = 1;
2984                 }
2985         }
2986
2987         err = inode_setattr(inode, attr);
2988
2989         if (!err && ((attr->ia_valid & ATTR_MODE)))
2990                 err = btrfs_acl_chmod(inode);
2991         return err;
2992 }
2993
2994 void btrfs_delete_inode(struct inode *inode)
2995 {
2996         struct btrfs_trans_handle *trans;
2997         struct btrfs_root *root = BTRFS_I(inode)->root;
2998         unsigned long nr;
2999         int ret;
3000
3001         truncate_inode_pages(&inode->i_data, 0);
3002         if (is_bad_inode(inode)) {
3003                 btrfs_orphan_del(NULL, inode);
3004                 goto no_delete;
3005         }
3006         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3007
3008         btrfs_i_size_write(inode, 0);
3009         trans = btrfs_join_transaction(root, 1);
3010
3011         btrfs_set_trans_block_group(trans, inode);
3012         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3013         if (ret) {
3014                 btrfs_orphan_del(NULL, inode);
3015                 goto no_delete_lock;
3016         }
3017
3018         btrfs_orphan_del(trans, inode);
3019
3020         nr = trans->blocks_used;
3021         clear_inode(inode);
3022
3023         btrfs_end_transaction(trans, root);
3024         btrfs_btree_balance_dirty(root, nr);
3025         return;
3026
3027 no_delete_lock:
3028         nr = trans->blocks_used;
3029         btrfs_end_transaction(trans, root);
3030         btrfs_btree_balance_dirty(root, nr);
3031 no_delete:
3032         clear_inode(inode);
3033 }
3034
3035 /*
3036  * this returns the key found in the dir entry in the location pointer.
3037  * If no dir entries were found, location->objectid is 0.
3038  */
3039 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3040                                struct btrfs_key *location)
3041 {
3042         const char *name = dentry->d_name.name;
3043         int namelen = dentry->d_name.len;
3044         struct btrfs_dir_item *di;
3045         struct btrfs_path *path;
3046         struct btrfs_root *root = BTRFS_I(dir)->root;
3047         int ret = 0;
3048
3049         path = btrfs_alloc_path();
3050         BUG_ON(!path);
3051
3052         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3053                                     namelen, 0);
3054         if (IS_ERR(di))
3055                 ret = PTR_ERR(di);
3056
3057         if (!di || IS_ERR(di))
3058                 goto out_err;
3059
3060         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3061 out:
3062         btrfs_free_path(path);
3063         return ret;
3064 out_err:
3065         location->objectid = 0;
3066         goto out;
3067 }
3068
3069 /*
3070  * when we hit a tree root in a directory, the btrfs part of the inode
3071  * needs to be changed to reflect the root directory of the tree root.  This
3072  * is kind of like crossing a mount point.
3073  */
3074 static int fixup_tree_root_location(struct btrfs_root *root,
3075                              struct btrfs_key *location,
3076                              struct btrfs_root **sub_root,
3077                              struct dentry *dentry)
3078 {
3079         struct btrfs_root_item *ri;
3080
3081         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3082                 return 0;
3083         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3084                 return 0;
3085
3086         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3087                                         dentry->d_name.name,
3088                                         dentry->d_name.len);
3089         if (IS_ERR(*sub_root))
3090                 return PTR_ERR(*sub_root);
3091
3092         ri = &(*sub_root)->root_item;
3093         location->objectid = btrfs_root_dirid(ri);
3094         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3095         location->offset = 0;
3096
3097         return 0;
3098 }
3099
3100 static void inode_tree_add(struct inode *inode)
3101 {
3102         struct btrfs_root *root = BTRFS_I(inode)->root;
3103         struct btrfs_inode *entry;
3104         struct rb_node **p = &root->inode_tree.rb_node;
3105         struct rb_node *parent = NULL;
3106
3107         spin_lock(&root->inode_lock);
3108         while (*p) {
3109                 parent = *p;
3110                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3111
3112                 if (inode->i_ino < entry->vfs_inode.i_ino)
3113                         p = &(*p)->rb_left;
3114                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3115                         p = &(*p)->rb_right;
3116                 else {
3117                         WARN_ON(!(entry->vfs_inode.i_state &
3118                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3119                         break;
3120                 }
3121         }
3122         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3123         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3124         spin_unlock(&root->inode_lock);
3125 }
3126
3127 static void inode_tree_del(struct inode *inode)
3128 {
3129         struct btrfs_root *root = BTRFS_I(inode)->root;
3130
3131         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3132                 spin_lock(&root->inode_lock);
3133                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3134                 spin_unlock(&root->inode_lock);
3135                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3136         }
3137 }
3138
3139 static noinline void init_btrfs_i(struct inode *inode)
3140 {
3141         struct btrfs_inode *bi = BTRFS_I(inode);
3142
3143         bi->i_acl = BTRFS_ACL_NOT_CACHED;
3144         bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
3145
3146         bi->generation = 0;
3147         bi->sequence = 0;
3148         bi->last_trans = 0;
3149         bi->logged_trans = 0;
3150         bi->delalloc_bytes = 0;
3151         bi->reserved_bytes = 0;
3152         bi->disk_i_size = 0;
3153         bi->flags = 0;
3154         bi->index_cnt = (u64)-1;
3155         bi->last_unlink_trans = 0;
3156         bi->ordered_data_close = 0;
3157         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3158         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3159                              inode->i_mapping, GFP_NOFS);
3160         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3161                              inode->i_mapping, GFP_NOFS);
3162         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3163         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3164         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3165         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3166         mutex_init(&BTRFS_I(inode)->extent_mutex);
3167         mutex_init(&BTRFS_I(inode)->log_mutex);
3168 }
3169
3170 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3171 {
3172         struct btrfs_iget_args *args = p;
3173         inode->i_ino = args->ino;
3174         init_btrfs_i(inode);
3175         BTRFS_I(inode)->root = args->root;
3176         btrfs_set_inode_space_info(args->root, inode);
3177         return 0;
3178 }
3179
3180 static int btrfs_find_actor(struct inode *inode, void *opaque)
3181 {
3182         struct btrfs_iget_args *args = opaque;
3183         return args->ino == inode->i_ino &&
3184                 args->root == BTRFS_I(inode)->root;
3185 }
3186
3187 static struct inode *btrfs_iget_locked(struct super_block *s,
3188                                        u64 objectid,
3189                                        struct btrfs_root *root)
3190 {
3191         struct inode *inode;
3192         struct btrfs_iget_args args;
3193         args.ino = objectid;
3194         args.root = root;
3195
3196         inode = iget5_locked(s, objectid, btrfs_find_actor,
3197                              btrfs_init_locked_inode,
3198                              (void *)&args);
3199         return inode;
3200 }
3201
3202 /* Get an inode object given its location and corresponding root.
3203  * Returns in *is_new if the inode was read from disk
3204  */
3205 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3206                          struct btrfs_root *root)
3207 {
3208         struct inode *inode;
3209
3210         inode = btrfs_iget_locked(s, location->objectid, root);
3211         if (!inode)
3212                 return ERR_PTR(-ENOMEM);
3213
3214         if (inode->i_state & I_NEW) {
3215                 BTRFS_I(inode)->root = root;
3216                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3217                 btrfs_read_locked_inode(inode);
3218
3219                 inode_tree_add(inode);
3220                 unlock_new_inode(inode);
3221         }
3222
3223         return inode;
3224 }
3225
3226 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3227 {
3228         struct inode *inode;
3229         struct btrfs_inode *bi = BTRFS_I(dir);
3230         struct btrfs_root *root = bi->root;
3231         struct btrfs_root *sub_root = root;
3232         struct btrfs_key location;
3233         int ret;
3234
3235         if (dentry->d_name.len > BTRFS_NAME_LEN)
3236                 return ERR_PTR(-ENAMETOOLONG);
3237
3238         ret = btrfs_inode_by_name(dir, dentry, &location);
3239
3240         if (ret < 0)
3241                 return ERR_PTR(ret);
3242
3243         inode = NULL;
3244         if (location.objectid) {
3245                 ret = fixup_tree_root_location(root, &location, &sub_root,
3246                                                 dentry);
3247                 if (ret < 0)
3248                         return ERR_PTR(ret);
3249                 if (ret > 0)
3250                         return ERR_PTR(-ENOENT);
3251                 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3252                 if (IS_ERR(inode))
3253                         return ERR_CAST(inode);
3254         }
3255         return inode;
3256 }
3257
3258 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3259                                    struct nameidata *nd)
3260 {
3261         struct inode *inode;
3262
3263         if (dentry->d_name.len > BTRFS_NAME_LEN)
3264                 return ERR_PTR(-ENAMETOOLONG);
3265
3266         inode = btrfs_lookup_dentry(dir, dentry);
3267         if (IS_ERR(inode))
3268                 return ERR_CAST(inode);
3269
3270         return d_splice_alias(inode, dentry);
3271 }
3272
3273 static unsigned char btrfs_filetype_table[] = {
3274         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3275 };
3276
3277 static int btrfs_real_readdir(struct file *filp, void *dirent,
3278                               filldir_t filldir)
3279 {
3280         struct inode *inode = filp->f_dentry->d_inode;
3281         struct btrfs_root *root = BTRFS_I(inode)->root;
3282         struct btrfs_item *item;
3283         struct btrfs_dir_item *di;
3284         struct btrfs_key key;
3285         struct btrfs_key found_key;
3286         struct btrfs_path *path;
3287         int ret;
3288         u32 nritems;
3289         struct extent_buffer *leaf;
3290         int slot;
3291         int advance;
3292         unsigned char d_type;
3293         int over = 0;
3294         u32 di_cur;
3295         u32 di_total;
3296         u32 di_len;
3297         int key_type = BTRFS_DIR_INDEX_KEY;
3298         char tmp_name[32];
3299         char *name_ptr;
3300         int name_len;
3301
3302         /* FIXME, use a real flag for deciding about the key type */
3303         if (root->fs_info->tree_root == root)
3304                 key_type = BTRFS_DIR_ITEM_KEY;
3305
3306         /* special case for "." */
3307         if (filp->f_pos == 0) {
3308                 over = filldir(dirent, ".", 1,
3309                                1, inode->i_ino,
3310                                DT_DIR);
3311                 if (over)
3312                         return 0;
3313                 filp->f_pos = 1;
3314         }
3315         /* special case for .., just use the back ref */
3316         if (filp->f_pos == 1) {
3317                 u64 pino = parent_ino(filp->f_path.dentry);
3318                 over = filldir(dirent, "..", 2,
3319                                2, pino, DT_DIR);
3320                 if (over)
3321                         return 0;
3322                 filp->f_pos = 2;
3323         }
3324         path = btrfs_alloc_path();
3325         path->reada = 2;
3326
3327         btrfs_set_key_type(&key, key_type);
3328         key.offset = filp->f_pos;
3329         key.objectid = inode->i_ino;
3330
3331         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3332         if (ret < 0)
3333                 goto err;
3334         advance = 0;
3335
3336         while (1) {
3337                 leaf = path->nodes[0];
3338                 nritems = btrfs_header_nritems(leaf);
3339                 slot = path->slots[0];
3340                 if (advance || slot >= nritems) {
3341                         if (slot >= nritems - 1) {
3342                                 ret = btrfs_next_leaf(root, path);
3343                                 if (ret)
3344                                         break;
3345                                 leaf = path->nodes[0];
3346                                 nritems = btrfs_header_nritems(leaf);
3347                                 slot = path->slots[0];
3348                         } else {
3349                                 slot++;
3350                                 path->slots[0]++;
3351                         }
3352                 }
3353
3354                 advance = 1;
3355                 item = btrfs_item_nr(leaf, slot);
3356                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3357
3358                 if (found_key.objectid != key.objectid)
3359                         break;
3360                 if (btrfs_key_type(&found_key) != key_type)
3361                         break;
3362                 if (found_key.offset < filp->f_pos)
3363                         continue;
3364
3365                 filp->f_pos = found_key.offset;
3366
3367                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3368                 di_cur = 0;
3369                 di_total = btrfs_item_size(leaf, item);
3370
3371                 while (di_cur < di_total) {
3372                         struct btrfs_key location;
3373
3374                         name_len = btrfs_dir_name_len(leaf, di);
3375                         if (name_len <= sizeof(tmp_name)) {
3376                                 name_ptr = tmp_name;
3377                         } else {
3378                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3379                                 if (!name_ptr) {
3380                                         ret = -ENOMEM;
3381                                         goto err;
3382                                 }
3383                         }
3384                         read_extent_buffer(leaf, name_ptr,
3385                                            (unsigned long)(di + 1), name_len);
3386
3387                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3388                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3389
3390                         /* is this a reference to our own snapshot? If so
3391                          * skip it
3392                          */
3393                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3394                             location.objectid == root->root_key.objectid) {
3395                                 over = 0;
3396                                 goto skip;
3397                         }
3398                         over = filldir(dirent, name_ptr, name_len,
3399                                        found_key.offset, location.objectid,
3400                                        d_type);
3401
3402 skip:
3403                         if (name_ptr != tmp_name)
3404                                 kfree(name_ptr);
3405
3406                         if (over)
3407                                 goto nopos;
3408                         di_len = btrfs_dir_name_len(leaf, di) +
3409                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3410                         di_cur += di_len;
3411                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3412                 }
3413         }
3414
3415         /* Reached end of directory/root. Bump pos past the last item. */
3416         if (key_type == BTRFS_DIR_INDEX_KEY)
3417                 filp->f_pos = INT_LIMIT(off_t);
3418         else
3419                 filp->f_pos++;
3420 nopos:
3421         ret = 0;
3422 err:
3423         btrfs_free_path(path);
3424         return ret;
3425 }
3426
3427 int btrfs_write_inode(struct inode *inode, int wait)
3428 {
3429         struct btrfs_root *root = BTRFS_I(inode)->root;
3430         struct btrfs_trans_handle *trans;
3431         int ret = 0;
3432
3433         if (root->fs_info->btree_inode == inode)
3434                 return 0;
3435
3436         if (wait) {
3437                 trans = btrfs_join_transaction(root, 1);
3438                 btrfs_set_trans_block_group(trans, inode);
3439                 ret = btrfs_commit_transaction(trans, root);
3440         }
3441         return ret;
3442 }
3443
3444 /*
3445  * This is somewhat expensive, updating the tree every time the
3446  * inode changes.  But, it is most likely to find the inode in cache.
3447  * FIXME, needs more benchmarking...there are no reasons other than performance
3448  * to keep or drop this code.
3449  */
3450 void btrfs_dirty_inode(struct inode *inode)
3451 {
3452         struct btrfs_root *root = BTRFS_I(inode)->root;
3453         struct btrfs_trans_handle *trans;
3454
3455         trans = btrfs_join_transaction(root, 1);
3456         btrfs_set_trans_block_group(trans, inode);
3457         btrfs_update_inode(trans, root, inode);
3458         btrfs_end_transaction(trans, root);
3459 }
3460
3461 /*
3462  * find the highest existing sequence number in a directory
3463  * and then set the in-memory index_cnt variable to reflect
3464  * free sequence numbers
3465  */
3466 static int btrfs_set_inode_index_count(struct inode *inode)
3467 {
3468         struct btrfs_root *root = BTRFS_I(inode)->root;
3469         struct btrfs_key key, found_key;
3470         struct btrfs_path *path;
3471         struct extent_buffer *leaf;
3472         int ret;
3473
3474         key.objectid = inode->i_ino;
3475         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3476         key.offset = (u64)-1;
3477
3478         path = btrfs_alloc_path();
3479         if (!path)
3480                 return -ENOMEM;
3481
3482         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483         if (ret < 0)
3484                 goto out;
3485         /* FIXME: we should be able to handle this */
3486         if (ret == 0)
3487                 goto out;
3488         ret = 0;
3489
3490         /*
3491          * MAGIC NUMBER EXPLANATION:
3492          * since we search a directory based on f_pos we have to start at 2
3493          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3494          * else has to start at 2
3495          */
3496         if (path->slots[0] == 0) {
3497                 BTRFS_I(inode)->index_cnt = 2;
3498                 goto out;
3499         }
3500
3501         path->slots[0]--;
3502
3503         leaf = path->nodes[0];
3504         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3505
3506         if (found_key.objectid != inode->i_ino ||
3507             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3508                 BTRFS_I(inode)->index_cnt = 2;
3509                 goto out;
3510         }
3511
3512         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3513 out:
3514         btrfs_free_path(path);
3515         return ret;
3516 }
3517
3518 /*
3519  * helper to find a free sequence number in a given directory.  This current
3520  * code is very simple, later versions will do smarter things in the btree
3521  */
3522 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3523 {
3524         int ret = 0;
3525
3526         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3527                 ret = btrfs_set_inode_index_count(dir);
3528                 if (ret)
3529                         return ret;
3530         }
3531
3532         *index = BTRFS_I(dir)->index_cnt;
3533         BTRFS_I(dir)->index_cnt++;
3534
3535         return ret;
3536 }
3537
3538 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3539                                      struct btrfs_root *root,
3540                                      struct inode *dir,
3541                                      const char *name, int name_len,
3542                                      u64 ref_objectid, u64 objectid,
3543                                      u64 alloc_hint, int mode, u64 *index)
3544 {
3545         struct inode *inode;
3546         struct btrfs_inode_item *inode_item;
3547         struct btrfs_key *location;
3548         struct btrfs_path *path;
3549         struct btrfs_inode_ref *ref;
3550         struct btrfs_key key[2];
3551         u32 sizes[2];
3552         unsigned long ptr;
3553         int ret;
3554         int owner;
3555
3556         path = btrfs_alloc_path();
3557         BUG_ON(!path);
3558
3559         inode = new_inode(root->fs_info->sb);
3560         if (!inode)
3561                 return ERR_PTR(-ENOMEM);
3562
3563         if (dir) {
3564                 ret = btrfs_set_inode_index(dir, index);
3565                 if (ret) {
3566                         iput(inode);
3567                         return ERR_PTR(ret);
3568                 }
3569         }
3570         /*
3571          * index_cnt is ignored for everything but a dir,
3572          * btrfs_get_inode_index_count has an explanation for the magic
3573          * number
3574          */
3575         init_btrfs_i(inode);
3576         BTRFS_I(inode)->index_cnt = 2;
3577         BTRFS_I(inode)->root = root;
3578         BTRFS_I(inode)->generation = trans->transid;
3579         btrfs_set_inode_space_info(root, inode);
3580
3581         if (mode & S_IFDIR)
3582                 owner = 0;
3583         else
3584                 owner = 1;
3585         BTRFS_I(inode)->block_group =
3586                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3587         if ((mode & S_IFREG)) {
3588                 if (btrfs_test_opt(root, NODATASUM))
3589                         btrfs_set_flag(inode, NODATASUM);
3590                 if (btrfs_test_opt(root, NODATACOW))
3591                         btrfs_set_flag(inode, NODATACOW);
3592         }
3593
3594         key[0].objectid = objectid;
3595         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3596         key[0].offset = 0;
3597
3598         key[1].objectid = objectid;
3599         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3600         key[1].offset = ref_objectid;
3601
3602         sizes[0] = sizeof(struct btrfs_inode_item);
3603         sizes[1] = name_len + sizeof(*ref);
3604
3605         path->leave_spinning = 1;
3606         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3607         if (ret != 0)
3608                 goto fail;
3609
3610         if (objectid > root->highest_inode)
3611                 root->highest_inode = objectid;
3612
3613         inode->i_uid = current_fsuid();
3614
3615         if (dir && (dir->i_mode & S_ISGID)) {
3616                 inode->i_gid = dir->i_gid;
3617                 if (S_ISDIR(mode))
3618                         mode |= S_ISGID;
3619         } else
3620                 inode->i_gid = current_fsgid();
3621
3622         inode->i_mode = mode;
3623         inode->i_ino = objectid;
3624         inode_set_bytes(inode, 0);
3625         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3626         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3627                                   struct btrfs_inode_item);
3628         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3629
3630         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3631                              struct btrfs_inode_ref);
3632         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3633         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3634         ptr = (unsigned long)(ref + 1);
3635         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3636
3637         btrfs_mark_buffer_dirty(path->nodes[0]);
3638         btrfs_free_path(path);
3639
3640         location = &BTRFS_I(inode)->location;
3641         location->objectid = objectid;
3642         location->offset = 0;
3643         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3644
3645         insert_inode_hash(inode);
3646         inode_tree_add(inode);
3647         return inode;
3648 fail:
3649         if (dir)
3650                 BTRFS_I(dir)->index_cnt--;
3651         btrfs_free_path(path);
3652         iput(inode);
3653         return ERR_PTR(ret);
3654 }
3655
3656 static inline u8 btrfs_inode_type(struct inode *inode)
3657 {
3658         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3659 }
3660
3661 /*
3662  * utility function to add 'inode' into 'parent_inode' with
3663  * a give name and a given sequence number.
3664  * if 'add_backref' is true, also insert a backref from the
3665  * inode to the parent directory.
3666  */
3667 int btrfs_add_link(struct btrfs_trans_handle *trans,
3668                    struct inode *parent_inode, struct inode *inode,
3669                    const char *name, int name_len, int add_backref, u64 index)
3670 {
3671         int ret;
3672         struct btrfs_key key;
3673         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3674
3675         key.objectid = inode->i_ino;
3676         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3677         key.offset = 0;
3678
3679         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3680                                     parent_inode->i_ino,
3681                                     &key, btrfs_inode_type(inode),
3682                                     index);
3683         if (ret == 0) {
3684                 if (add_backref) {
3685                         ret = btrfs_insert_inode_ref(trans, root,
3686                                                      name, name_len,
3687                                                      inode->i_ino,
3688                                                      parent_inode->i_ino,
3689                                                      index);
3690                 }
3691                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3692                                    name_len * 2);
3693                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3694                 ret = btrfs_update_inode(trans, root, parent_inode);
3695         }
3696         return ret;
3697 }
3698
3699 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3700                             struct dentry *dentry, struct inode *inode,
3701                             int backref, u64 index)
3702 {
3703         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3704                                  inode, dentry->d_name.name,
3705                                  dentry->d_name.len, backref, index);
3706         if (!err) {
3707                 d_instantiate(dentry, inode);
3708                 return 0;
3709         }
3710         if (err > 0)
3711                 err = -EEXIST;
3712         return err;
3713 }
3714
3715 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3716                         int mode, dev_t rdev)
3717 {
3718         struct btrfs_trans_handle *trans;
3719         struct btrfs_root *root = BTRFS_I(dir)->root;
3720         struct inode *inode = NULL;
3721         int err;
3722         int drop_inode = 0;
3723         u64 objectid;
3724         unsigned long nr = 0;
3725         u64 index = 0;
3726
3727         if (!new_valid_dev(rdev))
3728                 return -EINVAL;
3729
3730         err = btrfs_check_metadata_free_space(root);
3731         if (err)
3732                 goto fail;
3733
3734         trans = btrfs_start_transaction(root, 1);
3735         btrfs_set_trans_block_group(trans, dir);
3736
3737         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3738         if (err) {
3739                 err = -ENOSPC;
3740                 goto out_unlock;
3741         }
3742
3743         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3744                                 dentry->d_name.len,
3745                                 dentry->d_parent->d_inode->i_ino, objectid,
3746                                 BTRFS_I(dir)->block_group, mode, &index);
3747         err = PTR_ERR(inode);
3748         if (IS_ERR(inode))
3749                 goto out_unlock;
3750
3751         err = btrfs_init_inode_security(inode, dir);
3752         if (err) {
3753                 drop_inode = 1;
3754                 goto out_unlock;
3755         }
3756
3757         btrfs_set_trans_block_group(trans, inode);
3758         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3759         if (err)
3760                 drop_inode = 1;
3761         else {
3762                 inode->i_op = &btrfs_special_inode_operations;
3763                 init_special_inode(inode, inode->i_mode, rdev);
3764                 btrfs_update_inode(trans, root, inode);
3765         }
3766         dir->i_sb->s_dirt = 1;
3767         btrfs_update_inode_block_group(trans, inode);
3768         btrfs_update_inode_block_group(trans, dir);
3769 out_unlock:
3770         nr = trans->blocks_used;
3771         btrfs_end_transaction_throttle(trans, root);
3772 fail:
3773         if (drop_inode) {
3774                 inode_dec_link_count(inode);
3775                 iput(inode);
3776         }
3777         btrfs_btree_balance_dirty(root, nr);
3778         return err;
3779 }
3780
3781 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3782                         int mode, struct nameidata *nd)
3783 {
3784         struct btrfs_trans_handle *trans;
3785         struct btrfs_root *root = BTRFS_I(dir)->root;
3786         struct inode *inode = NULL;
3787         int err;
3788         int drop_inode = 0;
3789         unsigned long nr = 0;
3790         u64 objectid;
3791         u64 index = 0;
3792
3793         err = btrfs_check_metadata_free_space(root);
3794         if (err)
3795                 goto fail;
3796         trans = btrfs_start_transaction(root, 1);
3797         btrfs_set_trans_block_group(trans, dir);
3798
3799         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3800         if (err) {
3801                 err = -ENOSPC;
3802                 goto out_unlock;
3803         }
3804
3805         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3806                                 dentry->d_name.len,
3807                                 dentry->d_parent->d_inode->i_ino,
3808                                 objectid, BTRFS_I(dir)->block_group, mode,
3809                                 &index);
3810         err = PTR_ERR(inode);
3811         if (IS_ERR(inode))
3812                 goto out_unlock;
3813
3814         err = btrfs_init_inode_security(inode, dir);
3815         if (err) {
3816                 drop_inode = 1;
3817                 goto out_unlock;
3818         }
3819
3820         btrfs_set_trans_block_group(trans, inode);
3821         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3822         if (err)
3823                 drop_inode = 1;
3824         else {
3825                 inode->i_mapping->a_ops = &btrfs_aops;
3826                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3827                 inode->i_fop = &btrfs_file_operations;
3828                 inode->i_op = &btrfs_file_inode_operations;
3829                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3830         }
3831         dir->i_sb->s_dirt = 1;
3832         btrfs_update_inode_block_group(trans, inode);
3833         btrfs_update_inode_block_group(trans, dir);
3834 out_unlock:
3835         nr = trans->blocks_used;
3836         btrfs_end_transaction_throttle(trans, root);
3837 fail:
3838         if (drop_inode) {
3839                 inode_dec_link_count(inode);
3840                 iput(inode);
3841         }
3842         btrfs_btree_balance_dirty(root, nr);
3843         return err;
3844 }
3845
3846 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3847                       struct dentry *dentry)
3848 {
3849         struct btrfs_trans_handle *trans;
3850         struct btrfs_root *root = BTRFS_I(dir)->root;
3851         struct inode *inode = old_dentry->d_inode;
3852         u64 index;
3853         unsigned long nr = 0;
3854         int err;
3855         int drop_inode = 0;
3856
3857         if (inode->i_nlink == 0)
3858                 return -ENOENT;
3859
3860         btrfs_inc_nlink(inode);
3861         err = btrfs_check_metadata_free_space(root);
3862         if (err)
3863                 goto fail;
3864         err = btrfs_set_inode_index(dir, &index);
3865         if (err)
3866                 goto fail;
3867
3868         trans = btrfs_start_transaction(root, 1);
3869
3870         btrfs_set_trans_block_group(trans, dir);
3871         atomic_inc(&inode->i_count);
3872
3873         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3874
3875         if (err)
3876                 drop_inode = 1;
3877
3878         dir->i_sb->s_dirt = 1;
3879         btrfs_update_inode_block_group(trans, dir);
3880         err = btrfs_update_inode(trans, root, inode);
3881
3882         if (err)
3883                 drop_inode = 1;
3884
3885         nr = trans->blocks_used;
3886
3887         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3888         btrfs_end_transaction_throttle(trans, root);
3889 fail:
3890         if (drop_inode) {
3891                 inode_dec_link_count(inode);
3892                 iput(inode);
3893         }
3894         btrfs_btree_balance_dirty(root, nr);
3895         return err;
3896 }
3897
3898 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3899 {
3900         struct inode *inode = NULL;
3901         struct btrfs_trans_handle *trans;
3902         struct btrfs_root *root = BTRFS_I(dir)->root;
3903         int err = 0;
3904         int drop_on_err = 0;
3905         u64 objectid = 0;
3906         u64 index = 0;
3907         unsigned long nr = 1;
3908
3909         err = btrfs_check_metadata_free_space(root);
3910         if (err)
3911                 goto out_unlock;
3912
3913         trans = btrfs_start_transaction(root, 1);
3914         btrfs_set_trans_block_group(trans, dir);
3915
3916         if (IS_ERR(trans)) {
3917                 err = PTR_ERR(trans);
3918                 goto out_unlock;
3919         }
3920
3921         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3922         if (err) {
3923                 err = -ENOSPC;
3924                 goto out_unlock;
3925         }
3926
3927         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3928                                 dentry->d_name.len,
3929                                 dentry->d_parent->d_inode->i_ino, objectid,
3930                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3931                                 &index);
3932         if (IS_ERR(inode)) {
3933                 err = PTR_ERR(inode);
3934                 goto out_fail;
3935         }
3936
3937         drop_on_err = 1;
3938
3939         err = btrfs_init_inode_security(inode, dir);
3940         if (err)
3941                 goto out_fail;
3942
3943         inode->i_op = &btrfs_dir_inode_operations;
3944         inode->i_fop = &btrfs_dir_file_operations;
3945         btrfs_set_trans_block_group(trans, inode);
3946
3947         btrfs_i_size_write(inode, 0);
3948         err = btrfs_update_inode(trans, root, inode);
3949         if (err)
3950                 goto out_fail;
3951
3952         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3953                                  inode, dentry->d_name.name,
3954                                  dentry->d_name.len, 0, index);
3955         if (err)
3956                 goto out_fail;
3957
3958         d_instantiate(dentry, inode);
3959         drop_on_err = 0;
3960         dir->i_sb->s_dirt = 1;
3961         btrfs_update_inode_block_group(trans, inode);
3962         btrfs_update_inode_block_group(trans, dir);
3963
3964 out_fail:
3965         nr = trans->blocks_used;
3966         btrfs_end_transaction_throttle(trans, root);
3967
3968 out_unlock:
3969         if (drop_on_err)
3970                 iput(inode);
3971         btrfs_btree_balance_dirty(root, nr);
3972         return err;
3973 }
3974
3975 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3976  * and an extent that you want to insert, deal with overlap and insert
3977  * the new extent into the tree.
3978  */
3979 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3980                                 struct extent_map *existing,
3981                                 struct extent_map *em,
3982                                 u64 map_start, u64 map_len)
3983 {
3984         u64 start_diff;
3985
3986         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3987         start_diff = map_start - em->start;
3988         em->start = map_start;
3989         em->len = map_len;
3990         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3991             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3992                 em->block_start += start_diff;
3993                 em->block_len -= start_diff;
3994         }
3995         return add_extent_mapping(em_tree, em);
3996 }
3997
3998 static noinline int uncompress_inline(struct btrfs_path *path,
3999                                       struct inode *inode, struct page *page,
4000                                       size_t pg_offset, u64 extent_offset,
4001                                       struct btrfs_file_extent_item *item)
4002 {
4003         int ret;
4004         struct extent_buffer *leaf = path->nodes[0];
4005         char *tmp;
4006         size_t max_size;
4007         unsigned long inline_size;
4008         unsigned long ptr;
4009
4010         WARN_ON(pg_offset != 0);
4011         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4012         inline_size = btrfs_file_extent_inline_item_len(leaf,
4013                                         btrfs_item_nr(leaf, path->slots[0]));
4014         tmp = kmalloc(inline_size, GFP_NOFS);
4015         ptr = btrfs_file_extent_inline_start(item);
4016
4017         read_extent_buffer(leaf, tmp, ptr, inline_size);
4018
4019         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4020         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4021                                     inline_size, max_size);
4022         if (ret) {
4023                 char *kaddr = kmap_atomic(page, KM_USER0);
4024                 unsigned long copy_size = min_t(u64,
4025                                   PAGE_CACHE_SIZE - pg_offset,
4026                                   max_size - extent_offset);
4027                 memset(kaddr + pg_offset, 0, copy_size);
4028                 kunmap_atomic(kaddr, KM_USER0);
4029         }
4030         kfree(tmp);
4031         return 0;
4032 }
4033
4034 /*
4035  * a bit scary, this does extent mapping from logical file offset to the disk.
4036  * the ugly parts come from merging extents from the disk with the in-ram
4037  * representation.  This gets more complex because of the data=ordered code,
4038  * where the in-ram extents might be locked pending data=ordered completion.
4039  *
4040  * This also copies inline extents directly into the page.
4041  */
4042
4043 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4044                                     size_t pg_offset, u64 start, u64 len,
4045                                     int create)
4046 {
4047         int ret;
4048         int err = 0;
4049         u64 bytenr;
4050         u64 extent_start = 0;
4051         u64 extent_end = 0;
4052         u64 objectid = inode->i_ino;
4053         u32 found_type;
4054         struct btrfs_path *path = NULL;
4055         struct btrfs_root *root = BTRFS_I(inode)->root;
4056         struct btrfs_file_extent_item *item;
4057         struct extent_buffer *leaf;
4058         struct btrfs_key found_key;
4059         struct extent_map *em = NULL;
4060         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4061         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4062         struct btrfs_trans_handle *trans = NULL;
4063         int compressed;
4064
4065 again:
4066         spin_lock(&em_tree->lock);
4067         em = lookup_extent_mapping(em_tree, start, len);
4068         if (em)
4069                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4070         spin_unlock(&em_tree->lock);
4071
4072         if (em) {
4073                 if (em->start > start || em->start + em->len <= start)
4074                         free_extent_map(em);
4075                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4076                         free_extent_map(em);
4077                 else
4078                         goto out;
4079         }
4080         em = alloc_extent_map(GFP_NOFS);
4081         if (!em) {
4082                 err = -ENOMEM;
4083                 goto out;
4084         }
4085         em->bdev = root->fs_info->fs_devices->latest_bdev;
4086         em->start = EXTENT_MAP_HOLE;
4087         em->orig_start = EXTENT_MAP_HOLE;
4088         em->len = (u64)-1;
4089         em->block_len = (u64)-1;
4090
4091         if (!path) {
4092                 path = btrfs_alloc_path();
4093                 BUG_ON(!path);
4094         }
4095
4096         ret = btrfs_lookup_file_extent(trans, root, path,
4097                                        objectid, start, trans != NULL);
4098         if (ret < 0) {
4099                 err = ret;
4100                 goto out;
4101         }
4102
4103         if (ret != 0) {
4104                 if (path->slots[0] == 0)
4105                         goto not_found;
4106                 path->slots[0]--;
4107         }
4108
4109         leaf = path->nodes[0];
4110         item = btrfs_item_ptr(leaf, path->slots[0],
4111                               struct btrfs_file_extent_item);
4112         /* are we inside the extent that was found? */
4113         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4114         found_type = btrfs_key_type(&found_key);
4115         if (found_key.objectid != objectid ||
4116             found_type != BTRFS_EXTENT_DATA_KEY) {
4117                 goto not_found;
4118         }
4119
4120         found_type = btrfs_file_extent_type(leaf, item);
4121         extent_start = found_key.offset;
4122         compressed = btrfs_file_extent_compression(leaf, item);
4123         if (found_type == BTRFS_FILE_EXTENT_REG ||
4124             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4125                 extent_end = extent_start +
4126                        btrfs_file_extent_num_bytes(leaf, item);
4127         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4128                 size_t size;
4129                 size = btrfs_file_extent_inline_len(leaf, item);
4130                 extent_end = (extent_start + size + root->sectorsize - 1) &
4131                         ~((u64)root->sectorsize - 1);
4132         }
4133
4134         if (start >= extent_end) {
4135                 path->slots[0]++;
4136                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4137                         ret = btrfs_next_leaf(root, path);
4138                         if (ret < 0) {
4139                                 err = ret;
4140                                 goto out;
4141                         }
4142                         if (ret > 0)
4143                                 goto not_found;
4144                         leaf = path->nodes[0];
4145                 }
4146                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4147                 if (found_key.objectid != objectid ||
4148                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4149                         goto not_found;
4150                 if (start + len <= found_key.offset)
4151                         goto not_found;
4152                 em->start = start;
4153                 em->len = found_key.offset - start;
4154                 goto not_found_em;
4155         }
4156
4157         if (found_type == BTRFS_FILE_EXTENT_REG ||
4158             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4159                 em->start = extent_start;
4160                 em->len = extent_end - extent_start;
4161                 em->orig_start = extent_start -
4162                                  btrfs_file_extent_offset(leaf, item);
4163                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4164                 if (bytenr == 0) {
4165                         em->block_start = EXTENT_MAP_HOLE;
4166                         goto insert;
4167                 }
4168                 if (compressed) {
4169                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4170                         em->block_start = bytenr;
4171                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4172                                                                          item);
4173                 } else {
4174                         bytenr += btrfs_file_extent_offset(leaf, item);
4175                         em->block_start = bytenr;
4176                         em->block_len = em->len;
4177                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4178                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4179                 }
4180                 goto insert;
4181         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4182                 unsigned long ptr;
4183                 char *map;
4184                 size_t size;
4185                 size_t extent_offset;
4186                 size_t copy_size;
4187
4188                 em->block_start = EXTENT_MAP_INLINE;
4189                 if (!page || create) {
4190                         em->start = extent_start;
4191                         em->len = extent_end - extent_start;
4192                         goto out;
4193                 }
4194
4195                 size = btrfs_file_extent_inline_len(leaf, item);
4196                 extent_offset = page_offset(page) + pg_offset - extent_start;
4197                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4198                                 size - extent_offset);
4199                 em->start = extent_start + extent_offset;
4200                 em->len = (copy_size + root->sectorsize - 1) &
4201                         ~((u64)root->sectorsize - 1);
4202                 em->orig_start = EXTENT_MAP_INLINE;
4203                 if (compressed)
4204                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4205                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4206                 if (create == 0 && !PageUptodate(page)) {
4207                         if (btrfs_file_extent_compression(leaf, item) ==
4208                             BTRFS_COMPRESS_ZLIB) {
4209                                 ret = uncompress_inline(path, inode, page,
4210                                                         pg_offset,
4211                                                         extent_offset, item);
4212                                 BUG_ON(ret);
4213                         } else {
4214                                 map = kmap(page);
4215                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4216                                                    copy_size);
4217                                 kunmap(page);
4218                         }
4219                         flush_dcache_page(page);
4220                 } else if (create && PageUptodate(page)) {
4221                         if (!trans) {
4222                                 kunmap(page);
4223                                 free_extent_map(em);
4224                                 em = NULL;
4225                                 btrfs_release_path(root, path);
4226                                 trans = btrfs_join_transaction(root, 1);
4227                                 goto again;
4228                         }
4229                         map = kmap(page);
4230                         write_extent_buffer(leaf, map + pg_offset, ptr,
4231                                             copy_size);
4232                         kunmap(page);
4233                         btrfs_mark_buffer_dirty(leaf);
4234                 }
4235                 set_extent_uptodate(io_tree, em->start,
4236                                     extent_map_end(em) - 1, GFP_NOFS);
4237                 goto insert;
4238         } else {
4239                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4240                 WARN_ON(1);
4241         }
4242 not_found:
4243         em->start = start;
4244         em->len = len;
4245 not_found_em:
4246         em->block_start = EXTENT_MAP_HOLE;
4247         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4248 insert:
4249         btrfs_release_path(root, path);
4250         if (em->start > start || extent_map_end(em) <= start) {
4251                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4252                        "[%llu %llu]\n", (unsigned long long)em->start,
4253                        (unsigned long long)em->len,
4254                        (unsigned long long)start,
4255                        (unsigned long long)len);
4256                 err = -EIO;
4257                 goto out;
4258         }
4259
4260         err = 0;
4261         spin_lock(&em_tree->lock);
4262         ret = add_extent_mapping(em_tree, em);
4263         /* it is possible that someone inserted the extent into the tree
4264          * while we had the lock dropped.  It is also possible that
4265          * an overlapping map exists in the tree
4266          */
4267         if (ret == -EEXIST) {
4268                 struct extent_map *existing;
4269
4270                 ret = 0;
4271
4272                 existing = lookup_extent_mapping(em_tree, start, len);
4273                 if (existing && (existing->start > start ||
4274                     existing->start + existing->len <= start)) {
4275                         free_extent_map(existing);
4276                         existing = NULL;
4277                 }
4278                 if (!existing) {
4279                         existing = lookup_extent_mapping(em_tree, em->start,
4280                                                          em->len);
4281                         if (existing) {
4282                                 err = merge_extent_mapping(em_tree, existing,
4283                                                            em, start,
4284                                                            root->sectorsize);
4285                                 free_extent_map(existing);
4286                                 if (err) {
4287                                         free_extent_map(em);
4288                                         em = NULL;
4289                                 }
4290                         } else {
4291                                 err = -EIO;
4292                                 free_extent_map(em);
4293                                 em = NULL;
4294                         }
4295                 } else {
4296                         free_extent_map(em);
4297                         em = existing;
4298                         err = 0;
4299                 }
4300         }
4301         spin_unlock(&em_tree->lock);
4302 out:
4303         if (path)
4304                 btrfs_free_path(path);
4305         if (trans) {
4306                 ret = btrfs_end_transaction(trans, root);
4307                 if (!err)
4308                         err = ret;
4309         }
4310         if (err) {
4311                 free_extent_map(em);
4312                 return ERR_PTR(err);
4313         }
4314         return em;
4315 }
4316
4317 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4318                         const struct iovec *iov, loff_t offset,
4319                         unsigned long nr_segs)
4320 {
4321         return -EINVAL;
4322 }
4323
4324 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4325                 __u64 start, __u64 len)
4326 {
4327         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4328 }
4329
4330 int btrfs_readpage(struct file *file, struct page *page)
4331 {
4332         struct extent_io_tree *tree;
4333         tree = &BTRFS_I(page->mapping->host)->io_tree;
4334         return extent_read_full_page(tree, page, btrfs_get_extent);
4335 }
4336
4337 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4338 {
4339         struct extent_io_tree *tree;
4340
4341
4342         if (current->flags & PF_MEMALLOC) {
4343                 redirty_page_for_writepage(wbc, page);
4344                 unlock_page(page);
4345                 return 0;
4346         }
4347         tree = &BTRFS_I(page->mapping->host)->io_tree;
4348         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4349 }
4350
4351 int btrfs_writepages(struct address_space *mapping,
4352                      struct writeback_control *wbc)
4353 {
4354         struct extent_io_tree *tree;
4355
4356         tree = &BTRFS_I(mapping->host)->io_tree;
4357         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4358 }
4359
4360 static int
4361 btrfs_readpages(struct file *file, struct address_space *mapping,
4362                 struct list_head *pages, unsigned nr_pages)
4363 {
4364         struct extent_io_tree *tree;
4365         tree = &BTRFS_I(mapping->host)->io_tree;
4366         return extent_readpages(tree, mapping, pages, nr_pages,
4367                                 btrfs_get_extent);
4368 }
4369 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4370 {
4371         struct extent_io_tree *tree;
4372         struct extent_map_tree *map;
4373         int ret;
4374
4375         tree = &BTRFS_I(page->mapping->host)->io_tree;
4376         map = &BTRFS_I(page->mapping->host)->extent_tree;
4377         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4378         if (ret == 1) {
4379                 ClearPagePrivate(page);
4380                 set_page_private(page, 0);
4381                 page_cache_release(page);
4382         }
4383         return ret;
4384 }
4385
4386 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4387 {
4388         if (PageWriteback(page) || PageDirty(page))
4389                 return 0;
4390         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4391 }
4392
4393 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4394 {
4395         struct extent_io_tree *tree;
4396         struct btrfs_ordered_extent *ordered;
4397         u64 page_start = page_offset(page);
4398         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4399
4400         wait_on_page_writeback(page);
4401         tree = &BTRFS_I(page->mapping->host)->io_tree;
4402         if (offset) {
4403                 btrfs_releasepage(page, GFP_NOFS);
4404                 return;
4405         }
4406
4407         lock_extent(tree, page_start, page_end, GFP_NOFS);
4408         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4409                                            page_offset(page));
4410         if (ordered) {
4411                 /*
4412                  * IO on this page will never be started, so we need
4413                  * to account for any ordered extents now
4414                  */
4415                 clear_extent_bit(tree, page_start, page_end,
4416                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4417                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4418                 btrfs_finish_ordered_io(page->mapping->host,
4419                                         page_start, page_end);
4420                 btrfs_put_ordered_extent(ordered);
4421                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4422         }
4423         clear_extent_bit(tree, page_start, page_end,
4424                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4425                  EXTENT_ORDERED,
4426                  1, 1, GFP_NOFS);
4427         __btrfs_releasepage(page, GFP_NOFS);
4428
4429         ClearPageChecked(page);
4430         if (PagePrivate(page)) {
4431                 ClearPagePrivate(page);
4432                 set_page_private(page, 0);
4433                 page_cache_release(page);
4434         }
4435 }
4436
4437 /*
4438  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4439  * called from a page fault handler when a page is first dirtied. Hence we must
4440  * be careful to check for EOF conditions here. We set the page up correctly
4441  * for a written page which means we get ENOSPC checking when writing into
4442  * holes and correct delalloc and unwritten extent mapping on filesystems that
4443  * support these features.
4444  *
4445  * We are not allowed to take the i_mutex here so we have to play games to
4446  * protect against truncate races as the page could now be beyond EOF.  Because
4447  * vmtruncate() writes the inode size before removing pages, once we have the
4448  * page lock we can determine safely if the page is beyond EOF. If it is not
4449  * beyond EOF, then the page is guaranteed safe against truncation until we
4450  * unlock the page.
4451  */
4452 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4453 {
4454         struct page *page = vmf->page;
4455         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4456         struct btrfs_root *root = BTRFS_I(inode)->root;
4457         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4458         struct btrfs_ordered_extent *ordered;
4459         char *kaddr;
4460         unsigned long zero_start;
4461         loff_t size;
4462         int ret;
4463         u64 page_start;
4464         u64 page_end;
4465
4466         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4467         if (ret) {
4468                 if (ret == -ENOMEM)
4469                         ret = VM_FAULT_OOM;
4470                 else /* -ENOSPC, -EIO, etc */
4471                         ret = VM_FAULT_SIGBUS;
4472                 goto out;
4473         }
4474
4475         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4476 again:
4477         lock_page(page);
4478         size = i_size_read(inode);
4479         page_start = page_offset(page);
4480         page_end = page_start + PAGE_CACHE_SIZE - 1;
4481
4482         if ((page->mapping != inode->i_mapping) ||
4483             (page_start >= size)) {
4484                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4485                 /* page got truncated out from underneath us */
4486                 goto out_unlock;
4487         }
4488         wait_on_page_writeback(page);
4489
4490         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4491         set_page_extent_mapped(page);
4492
4493         /*
4494          * we can't set the delalloc bits if there are pending ordered
4495          * extents.  Drop our locks and wait for them to finish
4496          */
4497         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4498         if (ordered) {
4499                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4500                 unlock_page(page);
4501                 btrfs_start_ordered_extent(inode, ordered, 1);
4502                 btrfs_put_ordered_extent(ordered);
4503                 goto again;
4504         }
4505
4506         btrfs_set_extent_delalloc(inode, page_start, page_end);
4507         ret = 0;
4508
4509         /* page is wholly or partially inside EOF */
4510         if (page_start + PAGE_CACHE_SIZE > size)
4511                 zero_start = size & ~PAGE_CACHE_MASK;
4512         else
4513                 zero_start = PAGE_CACHE_SIZE;
4514
4515         if (zero_start != PAGE_CACHE_SIZE) {
4516                 kaddr = kmap(page);
4517                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4518                 flush_dcache_page(page);
4519                 kunmap(page);
4520         }
4521         ClearPageChecked(page);
4522         set_page_dirty(page);
4523
4524         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4525         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4526
4527 out_unlock:
4528         unlock_page(page);
4529 out:
4530         return ret;
4531 }
4532
4533 static void btrfs_truncate(struct inode *inode)
4534 {
4535         struct btrfs_root *root = BTRFS_I(inode)->root;
4536         int ret;
4537         struct btrfs_trans_handle *trans;
4538         unsigned long nr;
4539         u64 mask = root->sectorsize - 1;
4540
4541         if (!S_ISREG(inode->i_mode))
4542                 return;
4543         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4544                 return;
4545
4546         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4547         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4548
4549         trans = btrfs_start_transaction(root, 1);
4550
4551         /*
4552          * setattr is responsible for setting the ordered_data_close flag,
4553          * but that is only tested during the last file release.  That
4554          * could happen well after the next commit, leaving a great big
4555          * window where new writes may get lost if someone chooses to write
4556          * to this file after truncating to zero
4557          *
4558          * The inode doesn't have any dirty data here, and so if we commit
4559          * this is a noop.  If someone immediately starts writing to the inode
4560          * it is very likely we'll catch some of their writes in this
4561          * transaction, and the commit will find this file on the ordered
4562          * data list with good things to send down.
4563          *
4564          * This is a best effort solution, there is still a window where
4565          * using truncate to replace the contents of the file will
4566          * end up with a zero length file after a crash.
4567          */
4568         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4569                 btrfs_add_ordered_operation(trans, root, inode);
4570
4571         btrfs_set_trans_block_group(trans, inode);
4572         btrfs_i_size_write(inode, inode->i_size);
4573
4574         ret = btrfs_orphan_add(trans, inode);
4575         if (ret)
4576                 goto out;
4577         /* FIXME, add redo link to tree so we don't leak on crash */
4578         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4579                                       BTRFS_EXTENT_DATA_KEY);
4580         btrfs_update_inode(trans, root, inode);
4581
4582         ret = btrfs_orphan_del(trans, inode);
4583         BUG_ON(ret);
4584
4585 out:
4586         nr = trans->blocks_used;
4587         ret = btrfs_end_transaction_throttle(trans, root);
4588         BUG_ON(ret);
4589         btrfs_btree_balance_dirty(root, nr);
4590 }
4591
4592 /*
4593  * create a new subvolume directory/inode (helper for the ioctl).
4594  */
4595 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4596                              struct btrfs_root *new_root, struct dentry *dentry,
4597                              u64 new_dirid, u64 alloc_hint)
4598 {
4599         struct inode *inode;
4600         int error;
4601         u64 index = 0;
4602
4603         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4604                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4605         if (IS_ERR(inode))
4606                 return PTR_ERR(inode);
4607         inode->i_op = &btrfs_dir_inode_operations;
4608         inode->i_fop = &btrfs_dir_file_operations;
4609
4610         inode->i_nlink = 1;
4611         btrfs_i_size_write(inode, 0);
4612
4613         error = btrfs_update_inode(trans, new_root, inode);
4614         if (error)
4615                 return error;
4616
4617         d_instantiate(dentry, inode);
4618         return 0;
4619 }
4620
4621 /* helper function for file defrag and space balancing.  This
4622  * forces readahead on a given range of bytes in an inode
4623  */
4624 unsigned long btrfs_force_ra(struct address_space *mapping,
4625                               struct file_ra_state *ra, struct file *file,
4626                               pgoff_t offset, pgoff_t last_index)
4627 {
4628         pgoff_t req_size = last_index - offset + 1;
4629
4630         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4631         return offset + req_size;
4632 }
4633
4634 struct inode *btrfs_alloc_inode(struct super_block *sb)
4635 {
4636         struct btrfs_inode *ei;
4637
4638         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4639         if (!ei)
4640                 return NULL;
4641         ei->last_trans = 0;
4642         ei->logged_trans = 0;
4643         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4644         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4645         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4646         INIT_LIST_HEAD(&ei->i_orphan);
4647         INIT_LIST_HEAD(&ei->ordered_operations);
4648         return &ei->vfs_inode;
4649 }
4650
4651 void btrfs_destroy_inode(struct inode *inode)
4652 {
4653         struct btrfs_ordered_extent *ordered;
4654         struct btrfs_root *root = BTRFS_I(inode)->root;
4655
4656         WARN_ON(!list_empty(&inode->i_dentry));
4657         WARN_ON(inode->i_data.nrpages);
4658
4659         if (BTRFS_I(inode)->i_acl &&
4660             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4661                 posix_acl_release(BTRFS_I(inode)->i_acl);
4662         if (BTRFS_I(inode)->i_default_acl &&
4663             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4664                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4665
4666         /*
4667          * Make sure we're properly removed from the ordered operation
4668          * lists.
4669          */
4670         smp_mb();
4671         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4672                 spin_lock(&root->fs_info->ordered_extent_lock);
4673                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4674                 spin_unlock(&root->fs_info->ordered_extent_lock);
4675         }
4676
4677         spin_lock(&root->list_lock);
4678         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4679                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4680                        " list\n", inode->i_ino);
4681                 dump_stack();
4682         }
4683         spin_unlock(&root->list_lock);
4684
4685         while (1) {
4686                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4687                 if (!ordered)
4688                         break;
4689                 else {
4690                         printk(KERN_ERR "btrfs found ordered "
4691                                "extent %llu %llu on inode cleanup\n",
4692                                (unsigned long long)ordered->file_offset,
4693                                (unsigned long long)ordered->len);
4694                         btrfs_remove_ordered_extent(inode, ordered);
4695                         btrfs_put_ordered_extent(ordered);
4696                         btrfs_put_ordered_extent(ordered);
4697                 }
4698         }
4699         inode_tree_del(inode);
4700         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4701         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4702 }
4703
4704 static void init_once(void *foo)
4705 {
4706         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4707
4708         inode_init_once(&ei->vfs_inode);
4709 }
4710
4711 void btrfs_destroy_cachep(void)
4712 {
4713         if (btrfs_inode_cachep)
4714                 kmem_cache_destroy(btrfs_inode_cachep);
4715         if (btrfs_trans_handle_cachep)
4716                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4717         if (btrfs_transaction_cachep)
4718                 kmem_cache_destroy(btrfs_transaction_cachep);
4719         if (btrfs_path_cachep)
4720                 kmem_cache_destroy(btrfs_path_cachep);
4721 }
4722
4723 int btrfs_init_cachep(void)
4724 {
4725         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4726                         sizeof(struct btrfs_inode), 0,
4727                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4728         if (!btrfs_inode_cachep)
4729                 goto fail;
4730
4731         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4732                         sizeof(struct btrfs_trans_handle), 0,
4733                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4734         if (!btrfs_trans_handle_cachep)
4735                 goto fail;
4736
4737         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4738                         sizeof(struct btrfs_transaction), 0,
4739                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4740         if (!btrfs_transaction_cachep)
4741                 goto fail;
4742
4743         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4744                         sizeof(struct btrfs_path), 0,
4745                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4746         if (!btrfs_path_cachep)
4747                 goto fail;
4748
4749         return 0;
4750 fail:
4751         btrfs_destroy_cachep();
4752         return -ENOMEM;
4753 }
4754
4755 static int btrfs_getattr(struct vfsmount *mnt,
4756                          struct dentry *dentry, struct kstat *stat)
4757 {
4758         struct inode *inode = dentry->d_inode;
4759         generic_fillattr(inode, stat);
4760         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4761         stat->blksize = PAGE_CACHE_SIZE;
4762         stat->blocks = (inode_get_bytes(inode) +
4763                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4764         return 0;
4765 }
4766
4767 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4768                            struct inode *new_dir, struct dentry *new_dentry)
4769 {
4770         struct btrfs_trans_handle *trans;
4771         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4772         struct inode *new_inode = new_dentry->d_inode;
4773         struct inode *old_inode = old_dentry->d_inode;
4774         struct timespec ctime = CURRENT_TIME;
4775         u64 index = 0;
4776         int ret;
4777
4778         /* we're not allowed to rename between subvolumes */
4779         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4780             BTRFS_I(new_dir)->root->root_key.objectid)
4781                 return -EXDEV;
4782
4783         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4784             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4785                 return -ENOTEMPTY;
4786         }
4787
4788         /* to rename a snapshot or subvolume, we need to juggle the
4789          * backrefs.  This isn't coded yet
4790          */
4791         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4792                 return -EXDEV;
4793
4794         ret = btrfs_check_metadata_free_space(root);
4795         if (ret)
4796                 goto out_unlock;
4797
4798         /*
4799          * we're using rename to replace one file with another.
4800          * and the replacement file is large.  Start IO on it now so
4801          * we don't add too much work to the end of the transaction
4802          */
4803         if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4804             new_inode->i_size &&
4805             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4806                 filemap_flush(old_inode->i_mapping);
4807
4808         trans = btrfs_start_transaction(root, 1);
4809
4810         /*
4811          * make sure the inode gets flushed if it is replacing
4812          * something.
4813          */
4814         if (new_inode && new_inode->i_size &&
4815             old_inode && S_ISREG(old_inode->i_mode)) {
4816                 btrfs_add_ordered_operation(trans, root, old_inode);
4817         }
4818
4819         /*
4820          * this is an ugly little race, but the rename is required to make
4821          * sure that if we crash, the inode is either at the old name
4822          * or the new one.  pinning the log transaction lets us make sure
4823          * we don't allow a log commit to come in after we unlink the
4824          * name but before we add the new name back in.
4825          */
4826         btrfs_pin_log_trans(root);
4827
4828         btrfs_set_trans_block_group(trans, new_dir);
4829
4830         btrfs_inc_nlink(old_dentry->d_inode);
4831         old_dir->i_ctime = old_dir->i_mtime = ctime;
4832         new_dir->i_ctime = new_dir->i_mtime = ctime;
4833         old_inode->i_ctime = ctime;
4834
4835         if (old_dentry->d_parent != new_dentry->d_parent)
4836                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4837
4838         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4839                                  old_dentry->d_name.name,
4840                                  old_dentry->d_name.len);
4841         if (ret)
4842                 goto out_fail;
4843
4844         if (new_inode) {
4845                 new_inode->i_ctime = CURRENT_TIME;
4846                 ret = btrfs_unlink_inode(trans, root, new_dir,
4847                                          new_dentry->d_inode,
4848                                          new_dentry->d_name.name,
4849                                          new_dentry->d_name.len);
4850                 if (ret)
4851                         goto out_fail;
4852                 if (new_inode->i_nlink == 0) {
4853                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4854                         if (ret)
4855                                 goto out_fail;
4856                 }
4857
4858         }
4859         ret = btrfs_set_inode_index(new_dir, &index);
4860         if (ret)
4861                 goto out_fail;
4862
4863         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4864                              old_inode, new_dentry->d_name.name,
4865                              new_dentry->d_name.len, 1, index);
4866         if (ret)
4867                 goto out_fail;
4868
4869         btrfs_log_new_name(trans, old_inode, old_dir,
4870                                        new_dentry->d_parent);
4871 out_fail:
4872
4873         /* this btrfs_end_log_trans just allows the current
4874          * log-sub transaction to complete
4875          */
4876         btrfs_end_log_trans(root);
4877         btrfs_end_transaction_throttle(trans, root);
4878 out_unlock:
4879         return ret;
4880 }
4881
4882 /*
4883  * some fairly slow code that needs optimization. This walks the list
4884  * of all the inodes with pending delalloc and forces them to disk.
4885  */
4886 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4887 {
4888         struct list_head *head = &root->fs_info->delalloc_inodes;
4889         struct btrfs_inode *binode;
4890         struct inode *inode;
4891
4892         if (root->fs_info->sb->s_flags & MS_RDONLY)
4893                 return -EROFS;
4894
4895         spin_lock(&root->fs_info->delalloc_lock);
4896         while (!list_empty(head)) {
4897                 binode = list_entry(head->next, struct btrfs_inode,
4898                                     delalloc_inodes);
4899                 inode = igrab(&binode->vfs_inode);
4900                 if (!inode)
4901                         list_del_init(&binode->delalloc_inodes);
4902                 spin_unlock(&root->fs_info->delalloc_lock);
4903                 if (inode) {
4904                         filemap_flush(inode->i_mapping);
4905                         iput(inode);
4906                 }
4907                 cond_resched();
4908                 spin_lock(&root->fs_info->delalloc_lock);
4909         }
4910         spin_unlock(&root->fs_info->delalloc_lock);
4911
4912         /* the filemap_flush will queue IO into the worker threads, but
4913          * we have to make sure the IO is actually started and that
4914          * ordered extents get created before we return
4915          */
4916         atomic_inc(&root->fs_info->async_submit_draining);
4917         while (atomic_read(&root->fs_info->nr_async_submits) ||
4918               atomic_read(&root->fs_info->async_delalloc_pages)) {
4919                 wait_event(root->fs_info->async_submit_wait,
4920                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4921                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4922         }
4923         atomic_dec(&root->fs_info->async_submit_draining);
4924         return 0;
4925 }
4926
4927 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4928                          const char *symname)
4929 {
4930         struct btrfs_trans_handle *trans;
4931         struct btrfs_root *root = BTRFS_I(dir)->root;
4932         struct btrfs_path *path;
4933         struct btrfs_key key;
4934         struct inode *inode = NULL;
4935         int err;
4936         int drop_inode = 0;
4937         u64 objectid;
4938         u64 index = 0 ;
4939         int name_len;
4940         int datasize;
4941         unsigned long ptr;
4942         struct btrfs_file_extent_item *ei;
4943         struct extent_buffer *leaf;
4944         unsigned long nr = 0;
4945
4946         name_len = strlen(symname) + 1;
4947         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4948                 return -ENAMETOOLONG;
4949
4950         err = btrfs_check_metadata_free_space(root);
4951         if (err)
4952                 goto out_fail;
4953
4954         trans = btrfs_start_transaction(root, 1);
4955         btrfs_set_trans_block_group(trans, dir);
4956
4957         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4958         if (err) {
4959                 err = -ENOSPC;
4960                 goto out_unlock;
4961         }
4962
4963         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4964                                 dentry->d_name.len,
4965                                 dentry->d_parent->d_inode->i_ino, objectid,
4966                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4967                                 &index);
4968         err = PTR_ERR(inode);
4969         if (IS_ERR(inode))
4970                 goto out_unlock;
4971
4972         err = btrfs_init_inode_security(inode, dir);
4973         if (err) {
4974                 drop_inode = 1;
4975                 goto out_unlock;
4976         }
4977
4978         btrfs_set_trans_block_group(trans, inode);
4979         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4980         if (err)
4981                 drop_inode = 1;
4982         else {
4983                 inode->i_mapping->a_ops = &btrfs_aops;
4984                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4985                 inode->i_fop = &btrfs_file_operations;
4986                 inode->i_op = &btrfs_file_inode_operations;
4987                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4988         }
4989         dir->i_sb->s_dirt = 1;
4990         btrfs_update_inode_block_group(trans, inode);
4991         btrfs_update_inode_block_group(trans, dir);
4992         if (drop_inode)
4993                 goto out_unlock;
4994
4995         path = btrfs_alloc_path();
4996         BUG_ON(!path);
4997         key.objectid = inode->i_ino;
4998         key.offset = 0;
4999         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5000         datasize = btrfs_file_extent_calc_inline_size(name_len);
5001         err = btrfs_insert_empty_item(trans, root, path, &key,
5002                                       datasize);
5003         if (err) {
5004                 drop_inode = 1;
5005                 goto out_unlock;
5006         }
5007         leaf = path->nodes[0];
5008         ei = btrfs_item_ptr(leaf, path->slots[0],
5009                             struct btrfs_file_extent_item);
5010         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5011         btrfs_set_file_extent_type(leaf, ei,
5012                                    BTRFS_FILE_EXTENT_INLINE);
5013         btrfs_set_file_extent_encryption(leaf, ei, 0);
5014         btrfs_set_file_extent_compression(leaf, ei, 0);
5015         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5016         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5017
5018         ptr = btrfs_file_extent_inline_start(ei);
5019         write_extent_buffer(leaf, symname, ptr, name_len);
5020         btrfs_mark_buffer_dirty(leaf);
5021         btrfs_free_path(path);
5022
5023         inode->i_op = &btrfs_symlink_inode_operations;
5024         inode->i_mapping->a_ops = &btrfs_symlink_aops;
5025         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5026         inode_set_bytes(inode, name_len);
5027         btrfs_i_size_write(inode, name_len - 1);
5028         err = btrfs_update_inode(trans, root, inode);
5029         if (err)
5030                 drop_inode = 1;
5031
5032 out_unlock:
5033         nr = trans->blocks_used;
5034         btrfs_end_transaction_throttle(trans, root);
5035 out_fail:
5036         if (drop_inode) {
5037                 inode_dec_link_count(inode);
5038                 iput(inode);
5039         }
5040         btrfs_btree_balance_dirty(root, nr);
5041         return err;
5042 }
5043
5044 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5045                                struct inode *inode, u64 start, u64 end,
5046                                u64 locked_end, u64 alloc_hint, int mode)
5047 {
5048         struct btrfs_root *root = BTRFS_I(inode)->root;
5049         struct btrfs_key ins;
5050         u64 alloc_size;
5051         u64 cur_offset = start;
5052         u64 num_bytes = end - start;
5053         int ret = 0;
5054
5055         while (num_bytes > 0) {
5056                 alloc_size = min(num_bytes, root->fs_info->max_extent);
5057                 ret = btrfs_reserve_extent(trans, root, alloc_size,
5058                                            root->sectorsize, 0, alloc_hint,
5059                                            (u64)-1, &ins, 1);
5060                 if (ret) {
5061                         WARN_ON(1);
5062                         goto out;
5063                 }
5064                 ret = insert_reserved_file_extent(trans, inode,
5065                                                   cur_offset, ins.objectid,
5066                                                   ins.offset, ins.offset,
5067                                                   ins.offset, locked_end,
5068                                                   0, 0, 0,
5069                                                   BTRFS_FILE_EXTENT_PREALLOC);
5070                 BUG_ON(ret);
5071                 num_bytes -= ins.offset;
5072                 cur_offset += ins.offset;
5073                 alloc_hint = ins.objectid + ins.offset;
5074         }
5075 out:
5076         if (cur_offset > start) {
5077                 inode->i_ctime = CURRENT_TIME;
5078                 btrfs_set_flag(inode, PREALLOC);
5079                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5080                     cur_offset > i_size_read(inode))
5081                         btrfs_i_size_write(inode, cur_offset);
5082                 ret = btrfs_update_inode(trans, root, inode);
5083                 BUG_ON(ret);
5084         }
5085
5086         return ret;
5087 }
5088
5089 static long btrfs_fallocate(struct inode *inode, int mode,
5090                             loff_t offset, loff_t len)
5091 {
5092         u64 cur_offset;
5093         u64 last_byte;
5094         u64 alloc_start;
5095         u64 alloc_end;
5096         u64 alloc_hint = 0;
5097         u64 locked_end;
5098         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5099         struct extent_map *em;
5100         struct btrfs_trans_handle *trans;
5101         int ret;
5102
5103         alloc_start = offset & ~mask;
5104         alloc_end =  (offset + len + mask) & ~mask;
5105
5106         /*
5107          * wait for ordered IO before we have any locks.  We'll loop again
5108          * below with the locks held.
5109          */
5110         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5111
5112         mutex_lock(&inode->i_mutex);
5113         if (alloc_start > inode->i_size) {
5114                 ret = btrfs_cont_expand(inode, alloc_start);
5115                 if (ret)
5116                         goto out;
5117         }
5118
5119         locked_end = alloc_end - 1;
5120         while (1) {
5121                 struct btrfs_ordered_extent *ordered;
5122
5123                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5124                 if (!trans) {
5125                         ret = -EIO;
5126                         goto out;
5127                 }
5128
5129                 /* the extent lock is ordered inside the running
5130                  * transaction
5131                  */
5132                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5133                             GFP_NOFS);
5134                 ordered = btrfs_lookup_first_ordered_extent(inode,
5135                                                             alloc_end - 1);
5136                 if (ordered &&
5137                     ordered->file_offset + ordered->len > alloc_start &&
5138                     ordered->file_offset < alloc_end) {
5139                         btrfs_put_ordered_extent(ordered);
5140                         unlock_extent(&BTRFS_I(inode)->io_tree,
5141                                       alloc_start, locked_end, GFP_NOFS);
5142                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5143
5144                         /*
5145                          * we can't wait on the range with the transaction
5146                          * running or with the extent lock held
5147                          */
5148                         btrfs_wait_ordered_range(inode, alloc_start,
5149                                                  alloc_end - alloc_start);
5150                 } else {
5151                         if (ordered)
5152                                 btrfs_put_ordered_extent(ordered);
5153                         break;
5154                 }
5155         }
5156
5157         cur_offset = alloc_start;
5158         while (1) {
5159                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5160                                       alloc_end - cur_offset, 0);
5161                 BUG_ON(IS_ERR(em) || !em);
5162                 last_byte = min(extent_map_end(em), alloc_end);
5163                 last_byte = (last_byte + mask) & ~mask;
5164                 if (em->block_start == EXTENT_MAP_HOLE) {
5165                         ret = prealloc_file_range(trans, inode, cur_offset,
5166                                         last_byte, locked_end + 1,
5167                                         alloc_hint, mode);
5168                         if (ret < 0) {
5169                                 free_extent_map(em);
5170                                 break;
5171                         }
5172                 }
5173                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5174                         alloc_hint = em->block_start;
5175                 free_extent_map(em);
5176
5177                 cur_offset = last_byte;
5178                 if (cur_offset >= alloc_end) {
5179                         ret = 0;
5180                         break;
5181                 }
5182         }
5183         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5184                       GFP_NOFS);
5185
5186         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
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_test_flag(inode, READONLY) && (mask & MAY_WRITE))
5200                 return -EACCES;
5201         return generic_permission(inode, mask, btrfs_check_acl);
5202 }
5203
5204 static 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 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 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 };
5273
5274 static struct address_space_operations btrfs_symlink_aops = {
5275         .readpage       = btrfs_readpage,
5276         .writepage      = btrfs_writepage,
5277         .invalidatepage = btrfs_invalidatepage,
5278         .releasepage    = btrfs_releasepage,
5279 };
5280
5281 static struct inode_operations btrfs_file_inode_operations = {
5282         .truncate       = btrfs_truncate,
5283         .getattr        = btrfs_getattr,
5284         .setattr        = btrfs_setattr,
5285         .setxattr       = btrfs_setxattr,
5286         .getxattr       = btrfs_getxattr,
5287         .listxattr      = btrfs_listxattr,
5288         .removexattr    = btrfs_removexattr,
5289         .permission     = btrfs_permission,
5290         .fallocate      = btrfs_fallocate,
5291         .fiemap         = btrfs_fiemap,
5292 };
5293 static struct inode_operations btrfs_special_inode_operations = {
5294         .getattr        = btrfs_getattr,
5295         .setattr        = btrfs_setattr,
5296         .permission     = btrfs_permission,
5297         .setxattr       = btrfs_setxattr,
5298         .getxattr       = btrfs_getxattr,
5299         .listxattr      = btrfs_listxattr,
5300         .removexattr    = btrfs_removexattr,
5301 };
5302 static struct inode_operations btrfs_symlink_inode_operations = {
5303         .readlink       = generic_readlink,
5304         .follow_link    = page_follow_link_light,
5305         .put_link       = page_put_link,
5306         .permission     = btrfs_permission,
5307         .setxattr       = btrfs_setxattr,
5308         .getxattr       = btrfs_getxattr,
5309         .listxattr      = btrfs_listxattr,
5310         .removexattr    = btrfs_removexattr,
5311 };