Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93                                      struct inode *inode,  struct inode *dir)
94 {
95         int err;
96
97         err = btrfs_init_acl(trans, inode, dir);
98         if (!err)
99                 err = btrfs_xattr_security_init(trans, inode, dir);
100         return err;
101 }
102
103 /*
104  * this does all the hard work for inserting an inline extent into
105  * the btree.  The caller should have done a btrfs_drop_extents so that
106  * no overlapping inline items exist in the btree
107  */
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109                                 struct btrfs_root *root, struct inode *inode,
110                                 u64 start, size_t size, size_t compressed_size,
111                                 struct page **compressed_pages)
112 {
113         struct btrfs_key key;
114         struct btrfs_path *path;
115         struct extent_buffer *leaf;
116         struct page *page = NULL;
117         char *kaddr;
118         unsigned long ptr;
119         struct btrfs_file_extent_item *ei;
120         int err = 0;
121         int ret;
122         size_t cur_size = size;
123         size_t datasize;
124         unsigned long offset;
125         int use_compress = 0;
126
127         if (compressed_size && compressed_pages) {
128                 use_compress = 1;
129                 cur_size = compressed_size;
130         }
131
132         path = btrfs_alloc_path();
133         if (!path)
134                 return -ENOMEM;
135
136         path->leave_spinning = 1;
137         btrfs_set_trans_block_group(trans, inode);
138
139         key.objectid = inode->i_ino;
140         key.offset = start;
141         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142         datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144         inode_add_bytes(inode, size);
145         ret = btrfs_insert_empty_item(trans, root, path, &key,
146                                       datasize);
147         BUG_ON(ret);
148         if (ret) {
149                 err = ret;
150                 goto fail;
151         }
152         leaf = path->nodes[0];
153         ei = btrfs_item_ptr(leaf, path->slots[0],
154                             struct btrfs_file_extent_item);
155         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157         btrfs_set_file_extent_encryption(leaf, ei, 0);
158         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160         ptr = btrfs_file_extent_inline_start(ei);
161
162         if (use_compress) {
163                 struct page *cpage;
164                 int i = 0;
165                 while (compressed_size > 0) {
166                         cpage = compressed_pages[i];
167                         cur_size = min_t(unsigned long, compressed_size,
168                                        PAGE_CACHE_SIZE);
169
170                         kaddr = kmap_atomic(cpage, KM_USER0);
171                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
172                         kunmap_atomic(kaddr, KM_USER0);
173
174                         i++;
175                         ptr += cur_size;
176                         compressed_size -= cur_size;
177                 }
178                 btrfs_set_file_extent_compression(leaf, ei,
179                                                   BTRFS_COMPRESS_ZLIB);
180         } else {
181                 page = find_get_page(inode->i_mapping,
182                                      start >> PAGE_CACHE_SHIFT);
183                 btrfs_set_file_extent_compression(leaf, ei, 0);
184                 kaddr = kmap_atomic(page, KM_USER0);
185                 offset = start & (PAGE_CACHE_SIZE - 1);
186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187                 kunmap_atomic(kaddr, KM_USER0);
188                 page_cache_release(page);
189         }
190         btrfs_mark_buffer_dirty(leaf);
191         btrfs_free_path(path);
192
193         /*
194          * we're an inline extent, so nobody can
195          * extend the file past i_size without locking
196          * a page we already have locked.
197          *
198          * We must do any isize and inode updates
199          * before we unlock the pages.  Otherwise we
200          * could end up racing with unlink.
201          */
202         BTRFS_I(inode)->disk_i_size = inode->i_size;
203         btrfs_update_inode(trans, root, inode);
204
205         return 0;
206 fail:
207         btrfs_free_path(path);
208         return err;
209 }
210
211
212 /*
213  * conditionally insert an inline extent into the file.  This
214  * does the checks required to make sure the data is small enough
215  * to fit as an inline extent.
216  */
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218                                  struct btrfs_root *root,
219                                  struct inode *inode, u64 start, u64 end,
220                                  size_t compressed_size,
221                                  struct page **compressed_pages)
222 {
223         u64 isize = i_size_read(inode);
224         u64 actual_end = min(end + 1, isize);
225         u64 inline_len = actual_end - start;
226         u64 aligned_end = (end + root->sectorsize - 1) &
227                         ~((u64)root->sectorsize - 1);
228         u64 hint_byte;
229         u64 data_len = inline_len;
230         int ret;
231
232         if (compressed_size)
233                 data_len = compressed_size;
234
235         if (start > 0 ||
236             actual_end >= PAGE_CACHE_SIZE ||
237             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238             (!compressed_size &&
239             (actual_end & (root->sectorsize - 1)) == 0) ||
240             end + 1 < isize ||
241             data_len > root->fs_info->max_inline) {
242                 return 1;
243         }
244
245         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
246                                  &hint_byte, 1);
247         BUG_ON(ret);
248
249         if (isize > actual_end)
250                 inline_len = min_t(u64, isize, actual_end);
251         ret = insert_inline_extent(trans, root, inode, start,
252                                    inline_len, compressed_size,
253                                    compressed_pages);
254         BUG_ON(ret);
255         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
256         return 0;
257 }
258
259 struct async_extent {
260         u64 start;
261         u64 ram_size;
262         u64 compressed_size;
263         struct page **pages;
264         unsigned long nr_pages;
265         struct list_head list;
266 };
267
268 struct async_cow {
269         struct inode *inode;
270         struct btrfs_root *root;
271         struct page *locked_page;
272         u64 start;
273         u64 end;
274         struct list_head extents;
275         struct btrfs_work work;
276 };
277
278 static noinline int add_async_extent(struct async_cow *cow,
279                                      u64 start, u64 ram_size,
280                                      u64 compressed_size,
281                                      struct page **pages,
282                                      unsigned long nr_pages)
283 {
284         struct async_extent *async_extent;
285
286         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
287         async_extent->start = start;
288         async_extent->ram_size = ram_size;
289         async_extent->compressed_size = compressed_size;
290         async_extent->pages = pages;
291         async_extent->nr_pages = nr_pages;
292         list_add_tail(&async_extent->list, &cow->extents);
293         return 0;
294 }
295
296 /*
297  * we create compressed extents in two phases.  The first
298  * phase compresses a range of pages that have already been
299  * locked (both pages and state bits are locked).
300  *
301  * This is done inside an ordered work queue, and the compression
302  * is spread across many cpus.  The actual IO submission is step
303  * two, and the ordered work queue takes care of making sure that
304  * happens in the same order things were put onto the queue by
305  * writepages and friends.
306  *
307  * If this code finds it can't get good compression, it puts an
308  * entry onto the work queue to write the uncompressed bytes.  This
309  * makes sure that both compressed inodes and uncompressed inodes
310  * are written in the same order that pdflush sent them down.
311  */
312 static noinline int compress_file_range(struct inode *inode,
313                                         struct page *locked_page,
314                                         u64 start, u64 end,
315                                         struct async_cow *async_cow,
316                                         int *num_added)
317 {
318         struct btrfs_root *root = BTRFS_I(inode)->root;
319         struct btrfs_trans_handle *trans;
320         u64 num_bytes;
321         u64 orig_start;
322         u64 disk_num_bytes;
323         u64 blocksize = root->sectorsize;
324         u64 actual_end;
325         u64 isize = i_size_read(inode);
326         int ret = 0;
327         struct page **pages = NULL;
328         unsigned long nr_pages;
329         unsigned long nr_pages_ret = 0;
330         unsigned long total_compressed = 0;
331         unsigned long total_in = 0;
332         unsigned long max_compressed = 128 * 1024;
333         unsigned long max_uncompressed = 128 * 1024;
334         int i;
335         int will_compress;
336
337         orig_start = start;
338
339         actual_end = min_t(u64, isize, end + 1);
340 again:
341         will_compress = 0;
342         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
343         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
344
345         /*
346          * we don't want to send crud past the end of i_size through
347          * compression, that's just a waste of CPU time.  So, if the
348          * end of the file is before the start of our current
349          * requested range of bytes, we bail out to the uncompressed
350          * cleanup code that can deal with all of this.
351          *
352          * It isn't really the fastest way to fix things, but this is a
353          * very uncommon corner.
354          */
355         if (actual_end <= start)
356                 goto cleanup_and_bail_uncompressed;
357
358         total_compressed = actual_end - start;
359
360         /* we want to make sure that amount of ram required to uncompress
361          * an extent is reasonable, so we limit the total size in ram
362          * of a compressed extent to 128k.  This is a crucial number
363          * because it also controls how easily we can spread reads across
364          * cpus for decompression.
365          *
366          * We also want to make sure the amount of IO required to do
367          * a random read is reasonably small, so we limit the size of
368          * a compressed extent to 128k.
369          */
370         total_compressed = min(total_compressed, max_uncompressed);
371         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
372         num_bytes = max(blocksize,  num_bytes);
373         disk_num_bytes = num_bytes;
374         total_in = 0;
375         ret = 0;
376
377         /*
378          * we do compression for mount -o compress and when the
379          * inode has not been flagged as nocompress.  This flag can
380          * change at any time if we discover bad compression ratios.
381          */
382         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
383             (btrfs_test_opt(root, COMPRESS) ||
384              (BTRFS_I(inode)->force_compress))) {
385                 WARN_ON(pages);
386                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
387
388                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
389                                                 total_compressed, pages,
390                                                 nr_pages, &nr_pages_ret,
391                                                 &total_in,
392                                                 &total_compressed,
393                                                 max_compressed);
394
395                 if (!ret) {
396                         unsigned long offset = total_compressed &
397                                 (PAGE_CACHE_SIZE - 1);
398                         struct page *page = pages[nr_pages_ret - 1];
399                         char *kaddr;
400
401                         /* zero the tail end of the last page, we might be
402                          * sending it down to disk
403                          */
404                         if (offset) {
405                                 kaddr = kmap_atomic(page, KM_USER0);
406                                 memset(kaddr + offset, 0,
407                                        PAGE_CACHE_SIZE - offset);
408                                 kunmap_atomic(kaddr, KM_USER0);
409                         }
410                         will_compress = 1;
411                 }
412         }
413         if (start == 0) {
414                 trans = btrfs_join_transaction(root, 1);
415                 BUG_ON(!trans);
416                 btrfs_set_trans_block_group(trans, inode);
417
418                 /* lets try to make an inline extent */
419                 if (ret || total_in < (actual_end - start)) {
420                         /* we didn't compress the entire range, try
421                          * to make an uncompressed inline extent.
422                          */
423                         ret = cow_file_range_inline(trans, root, inode,
424                                                     start, end, 0, NULL);
425                 } else {
426                         /* try making a compressed inline extent */
427                         ret = cow_file_range_inline(trans, root, inode,
428                                                     start, end,
429                                                     total_compressed, pages);
430                 }
431                 if (ret == 0) {
432                         /*
433                          * inline extent creation worked, we don't need
434                          * to create any more async work items.  Unlock
435                          * and free up our temp pages.
436                          */
437                         extent_clear_unlock_delalloc(inode,
438                              &BTRFS_I(inode)->io_tree,
439                              start, end, NULL,
440                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
441                              EXTENT_CLEAR_DELALLOC |
442                              EXTENT_CLEAR_ACCOUNTING |
443                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
444
445                         btrfs_end_transaction(trans, root);
446                         goto free_pages_out;
447                 }
448                 btrfs_end_transaction(trans, root);
449         }
450
451         if (will_compress) {
452                 /*
453                  * we aren't doing an inline extent round the compressed size
454                  * up to a block size boundary so the allocator does sane
455                  * things
456                  */
457                 total_compressed = (total_compressed + blocksize - 1) &
458                         ~(blocksize - 1);
459
460                 /*
461                  * one last check to make sure the compression is really a
462                  * win, compare the page count read with the blocks on disk
463                  */
464                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
465                         ~(PAGE_CACHE_SIZE - 1);
466                 if (total_compressed >= total_in) {
467                         will_compress = 0;
468                 } else {
469                         disk_num_bytes = total_compressed;
470                         num_bytes = total_in;
471                 }
472         }
473         if (!will_compress && pages) {
474                 /*
475                  * the compression code ran but failed to make things smaller,
476                  * free any pages it allocated and our page pointer array
477                  */
478                 for (i = 0; i < nr_pages_ret; i++) {
479                         WARN_ON(pages[i]->mapping);
480                         page_cache_release(pages[i]);
481                 }
482                 kfree(pages);
483                 pages = NULL;
484                 total_compressed = 0;
485                 nr_pages_ret = 0;
486
487                 /* flag the file so we don't compress in the future */
488                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
489                     !(BTRFS_I(inode)->force_compress)) {
490                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
491                 }
492         }
493         if (will_compress) {
494                 *num_added += 1;
495
496                 /* the async work queues will take care of doing actual
497                  * allocation on disk for these compressed pages,
498                  * and will submit them to the elevator.
499                  */
500                 add_async_extent(async_cow, start, num_bytes,
501                                  total_compressed, pages, nr_pages_ret);
502
503                 if (start + num_bytes < end && start + num_bytes < actual_end) {
504                         start += num_bytes;
505                         pages = NULL;
506                         cond_resched();
507                         goto again;
508                 }
509         } else {
510 cleanup_and_bail_uncompressed:
511                 /*
512                  * No compression, but we still need to write the pages in
513                  * the file we've been given so far.  redirty the locked
514                  * page if it corresponds to our extent and set things up
515                  * for the async work queue to run cow_file_range to do
516                  * the normal delalloc dance
517                  */
518                 if (page_offset(locked_page) >= start &&
519                     page_offset(locked_page) <= end) {
520                         __set_page_dirty_nobuffers(locked_page);
521                         /* unlocked later on in the async handlers */
522                 }
523                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
524                 *num_added += 1;
525         }
526
527 out:
528         return 0;
529
530 free_pages_out:
531         for (i = 0; i < nr_pages_ret; i++) {
532                 WARN_ON(pages[i]->mapping);
533                 page_cache_release(pages[i]);
534         }
535         kfree(pages);
536
537         goto out;
538 }
539
540 /*
541  * phase two of compressed writeback.  This is the ordered portion
542  * of the code, which only gets called in the order the work was
543  * queued.  We walk all the async extents created by compress_file_range
544  * and send them down to the disk.
545  */
546 static noinline int submit_compressed_extents(struct inode *inode,
547                                               struct async_cow *async_cow)
548 {
549         struct async_extent *async_extent;
550         u64 alloc_hint = 0;
551         struct btrfs_trans_handle *trans;
552         struct btrfs_key ins;
553         struct extent_map *em;
554         struct btrfs_root *root = BTRFS_I(inode)->root;
555         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
556         struct extent_io_tree *io_tree;
557         int ret = 0;
558
559         if (list_empty(&async_cow->extents))
560                 return 0;
561
562
563         while (!list_empty(&async_cow->extents)) {
564                 async_extent = list_entry(async_cow->extents.next,
565                                           struct async_extent, list);
566                 list_del(&async_extent->list);
567
568                 io_tree = &BTRFS_I(inode)->io_tree;
569
570 retry:
571                 /* did the compression code fall back to uncompressed IO? */
572                 if (!async_extent->pages) {
573                         int page_started = 0;
574                         unsigned long nr_written = 0;
575
576                         lock_extent(io_tree, async_extent->start,
577                                          async_extent->start +
578                                          async_extent->ram_size - 1, GFP_NOFS);
579
580                         /* allocate blocks */
581                         ret = cow_file_range(inode, async_cow->locked_page,
582                                              async_extent->start,
583                                              async_extent->start +
584                                              async_extent->ram_size - 1,
585                                              &page_started, &nr_written, 0);
586
587                         /*
588                          * if page_started, cow_file_range inserted an
589                          * inline extent and took care of all the unlocking
590                          * and IO for us.  Otherwise, we need to submit
591                          * all those pages down to the drive.
592                          */
593                         if (!page_started && !ret)
594                                 extent_write_locked_range(io_tree,
595                                                   inode, async_extent->start,
596                                                   async_extent->start +
597                                                   async_extent->ram_size - 1,
598                                                   btrfs_get_extent,
599                                                   WB_SYNC_ALL);
600                         kfree(async_extent);
601                         cond_resched();
602                         continue;
603                 }
604
605                 lock_extent(io_tree, async_extent->start,
606                             async_extent->start + async_extent->ram_size - 1,
607                             GFP_NOFS);
608
609                 trans = btrfs_join_transaction(root, 1);
610                 ret = btrfs_reserve_extent(trans, root,
611                                            async_extent->compressed_size,
612                                            async_extent->compressed_size,
613                                            0, alloc_hint,
614                                            (u64)-1, &ins, 1);
615                 btrfs_end_transaction(trans, root);
616
617                 if (ret) {
618                         int i;
619                         for (i = 0; i < async_extent->nr_pages; i++) {
620                                 WARN_ON(async_extent->pages[i]->mapping);
621                                 page_cache_release(async_extent->pages[i]);
622                         }
623                         kfree(async_extent->pages);
624                         async_extent->nr_pages = 0;
625                         async_extent->pages = NULL;
626                         unlock_extent(io_tree, async_extent->start,
627                                       async_extent->start +
628                                       async_extent->ram_size - 1, GFP_NOFS);
629                         goto retry;
630                 }
631
632                 /*
633                  * here we're doing allocation and writeback of the
634                  * compressed pages
635                  */
636                 btrfs_drop_extent_cache(inode, async_extent->start,
637                                         async_extent->start +
638                                         async_extent->ram_size - 1, 0);
639
640                 em = alloc_extent_map(GFP_NOFS);
641                 em->start = async_extent->start;
642                 em->len = async_extent->ram_size;
643                 em->orig_start = em->start;
644
645                 em->block_start = ins.objectid;
646                 em->block_len = ins.offset;
647                 em->bdev = root->fs_info->fs_devices->latest_bdev;
648                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
649                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
650
651                 while (1) {
652                         write_lock(&em_tree->lock);
653                         ret = add_extent_mapping(em_tree, em);
654                         write_unlock(&em_tree->lock);
655                         if (ret != -EEXIST) {
656                                 free_extent_map(em);
657                                 break;
658                         }
659                         btrfs_drop_extent_cache(inode, async_extent->start,
660                                                 async_extent->start +
661                                                 async_extent->ram_size - 1, 0);
662                 }
663
664                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
665                                                ins.objectid,
666                                                async_extent->ram_size,
667                                                ins.offset,
668                                                BTRFS_ORDERED_COMPRESSED);
669                 BUG_ON(ret);
670
671                 /*
672                  * clear dirty, set writeback and unlock the pages.
673                  */
674                 extent_clear_unlock_delalloc(inode,
675                                 &BTRFS_I(inode)->io_tree,
676                                 async_extent->start,
677                                 async_extent->start +
678                                 async_extent->ram_size - 1,
679                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
680                                 EXTENT_CLEAR_UNLOCK |
681                                 EXTENT_CLEAR_DELALLOC |
682                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
683
684                 ret = btrfs_submit_compressed_write(inode,
685                                     async_extent->start,
686                                     async_extent->ram_size,
687                                     ins.objectid,
688                                     ins.offset, async_extent->pages,
689                                     async_extent->nr_pages);
690
691                 BUG_ON(ret);
692                 alloc_hint = ins.objectid + ins.offset;
693                 kfree(async_extent);
694                 cond_resched();
695         }
696
697         return 0;
698 }
699
700 /*
701  * when extent_io.c finds a delayed allocation range in the file,
702  * the call backs end up in this code.  The basic idea is to
703  * allocate extents on disk for the range, and create ordered data structs
704  * in ram to track those extents.
705  *
706  * locked_page is the page that writepage had locked already.  We use
707  * it to make sure we don't do extra locks or unlocks.
708  *
709  * *page_started is set to one if we unlock locked_page and do everything
710  * required to start IO on it.  It may be clean and already done with
711  * IO when we return.
712  */
713 static noinline int cow_file_range(struct inode *inode,
714                                    struct page *locked_page,
715                                    u64 start, u64 end, int *page_started,
716                                    unsigned long *nr_written,
717                                    int unlock)
718 {
719         struct btrfs_root *root = BTRFS_I(inode)->root;
720         struct btrfs_trans_handle *trans;
721         u64 alloc_hint = 0;
722         u64 num_bytes;
723         unsigned long ram_size;
724         u64 disk_num_bytes;
725         u64 cur_alloc_size;
726         u64 blocksize = root->sectorsize;
727         u64 actual_end;
728         u64 isize = i_size_read(inode);
729         struct btrfs_key ins;
730         struct extent_map *em;
731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
732         int ret = 0;
733
734         trans = btrfs_join_transaction(root, 1);
735         BUG_ON(!trans);
736         btrfs_set_trans_block_group(trans, inode);
737
738         actual_end = min_t(u64, isize, end + 1);
739
740         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
741         num_bytes = max(blocksize,  num_bytes);
742         disk_num_bytes = num_bytes;
743         ret = 0;
744
745         if (start == 0) {
746                 /* lets try to make an inline extent */
747                 ret = cow_file_range_inline(trans, root, inode,
748                                             start, end, 0, NULL);
749                 if (ret == 0) {
750                         extent_clear_unlock_delalloc(inode,
751                                      &BTRFS_I(inode)->io_tree,
752                                      start, end, NULL,
753                                      EXTENT_CLEAR_UNLOCK_PAGE |
754                                      EXTENT_CLEAR_UNLOCK |
755                                      EXTENT_CLEAR_DELALLOC |
756                                      EXTENT_CLEAR_ACCOUNTING |
757                                      EXTENT_CLEAR_DIRTY |
758                                      EXTENT_SET_WRITEBACK |
759                                      EXTENT_END_WRITEBACK);
760
761                         *nr_written = *nr_written +
762                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
763                         *page_started = 1;
764                         ret = 0;
765                         goto out;
766                 }
767         }
768
769         BUG_ON(disk_num_bytes >
770                btrfs_super_total_bytes(&root->fs_info->super_copy));
771
772
773         read_lock(&BTRFS_I(inode)->extent_tree.lock);
774         em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
775                                    start, num_bytes);
776         if (em) {
777                 /*
778                  * if block start isn't an actual block number then find the
779                  * first block in this inode and use that as a hint.  If that
780                  * block is also bogus then just don't worry about it.
781                  */
782                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
783                         free_extent_map(em);
784                         em = search_extent_mapping(em_tree, 0, 0);
785                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
786                                 alloc_hint = em->block_start;
787                         if (em)
788                                 free_extent_map(em);
789                 } else {
790                         alloc_hint = em->block_start;
791                         free_extent_map(em);
792                 }
793         }
794         read_unlock(&BTRFS_I(inode)->extent_tree.lock);
795         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
796
797         while (disk_num_bytes > 0) {
798                 unsigned long op;
799
800                 cur_alloc_size = disk_num_bytes;
801                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
802                                            root->sectorsize, 0, alloc_hint,
803                                            (u64)-1, &ins, 1);
804                 BUG_ON(ret);
805
806                 em = alloc_extent_map(GFP_NOFS);
807                 em->start = start;
808                 em->orig_start = em->start;
809                 ram_size = ins.offset;
810                 em->len = ins.offset;
811
812                 em->block_start = ins.objectid;
813                 em->block_len = ins.offset;
814                 em->bdev = root->fs_info->fs_devices->latest_bdev;
815                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
816
817                 while (1) {
818                         write_lock(&em_tree->lock);
819                         ret = add_extent_mapping(em_tree, em);
820                         write_unlock(&em_tree->lock);
821                         if (ret != -EEXIST) {
822                                 free_extent_map(em);
823                                 break;
824                         }
825                         btrfs_drop_extent_cache(inode, start,
826                                                 start + ram_size - 1, 0);
827                 }
828
829                 cur_alloc_size = ins.offset;
830                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
831                                                ram_size, cur_alloc_size, 0);
832                 BUG_ON(ret);
833
834                 if (root->root_key.objectid ==
835                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
836                         ret = btrfs_reloc_clone_csums(inode, start,
837                                                       cur_alloc_size);
838                         BUG_ON(ret);
839                 }
840
841                 if (disk_num_bytes < cur_alloc_size)
842                         break;
843
844                 /* we're not doing compressed IO, don't unlock the first
845                  * page (which the caller expects to stay locked), don't
846                  * clear any dirty bits and don't set any writeback bits
847                  *
848                  * Do set the Private2 bit so we know this page was properly
849                  * setup for writepage
850                  */
851                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
852                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
853                         EXTENT_SET_PRIVATE2;
854
855                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
856                                              start, start + ram_size - 1,
857                                              locked_page, op);
858                 disk_num_bytes -= cur_alloc_size;
859                 num_bytes -= cur_alloc_size;
860                 alloc_hint = ins.objectid + ins.offset;
861                 start += cur_alloc_size;
862         }
863 out:
864         ret = 0;
865         btrfs_end_transaction(trans, root);
866
867         return ret;
868 }
869
870 /*
871  * work queue call back to started compression on a file and pages
872  */
873 static noinline void async_cow_start(struct btrfs_work *work)
874 {
875         struct async_cow *async_cow;
876         int num_added = 0;
877         async_cow = container_of(work, struct async_cow, work);
878
879         compress_file_range(async_cow->inode, async_cow->locked_page,
880                             async_cow->start, async_cow->end, async_cow,
881                             &num_added);
882         if (num_added == 0)
883                 async_cow->inode = NULL;
884 }
885
886 /*
887  * work queue call back to submit previously compressed pages
888  */
889 static noinline void async_cow_submit(struct btrfs_work *work)
890 {
891         struct async_cow *async_cow;
892         struct btrfs_root *root;
893         unsigned long nr_pages;
894
895         async_cow = container_of(work, struct async_cow, work);
896
897         root = async_cow->root;
898         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
899                 PAGE_CACHE_SHIFT;
900
901         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
902
903         if (atomic_read(&root->fs_info->async_delalloc_pages) <
904             5 * 1042 * 1024 &&
905             waitqueue_active(&root->fs_info->async_submit_wait))
906                 wake_up(&root->fs_info->async_submit_wait);
907
908         if (async_cow->inode)
909                 submit_compressed_extents(async_cow->inode, async_cow);
910 }
911
912 static noinline void async_cow_free(struct btrfs_work *work)
913 {
914         struct async_cow *async_cow;
915         async_cow = container_of(work, struct async_cow, work);
916         kfree(async_cow);
917 }
918
919 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
920                                 u64 start, u64 end, int *page_started,
921                                 unsigned long *nr_written)
922 {
923         struct async_cow *async_cow;
924         struct btrfs_root *root = BTRFS_I(inode)->root;
925         unsigned long nr_pages;
926         u64 cur_end;
927         int limit = 10 * 1024 * 1042;
928
929         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
930                          1, 0, NULL, GFP_NOFS);
931         while (start < end) {
932                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
933                 async_cow->inode = inode;
934                 async_cow->root = root;
935                 async_cow->locked_page = locked_page;
936                 async_cow->start = start;
937
938                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
939                         cur_end = end;
940                 else
941                         cur_end = min(end, start + 512 * 1024 - 1);
942
943                 async_cow->end = cur_end;
944                 INIT_LIST_HEAD(&async_cow->extents);
945
946                 async_cow->work.func = async_cow_start;
947                 async_cow->work.ordered_func = async_cow_submit;
948                 async_cow->work.ordered_free = async_cow_free;
949                 async_cow->work.flags = 0;
950
951                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
952                         PAGE_CACHE_SHIFT;
953                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
954
955                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
956                                    &async_cow->work);
957
958                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
959                         wait_event(root->fs_info->async_submit_wait,
960                            (atomic_read(&root->fs_info->async_delalloc_pages) <
961                             limit));
962                 }
963
964                 while (atomic_read(&root->fs_info->async_submit_draining) &&
965                       atomic_read(&root->fs_info->async_delalloc_pages)) {
966                         wait_event(root->fs_info->async_submit_wait,
967                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
968                            0));
969                 }
970
971                 *nr_written += nr_pages;
972                 start = cur_end + 1;
973         }
974         *page_started = 1;
975         return 0;
976 }
977
978 static noinline int csum_exist_in_range(struct btrfs_root *root,
979                                         u64 bytenr, u64 num_bytes)
980 {
981         int ret;
982         struct btrfs_ordered_sum *sums;
983         LIST_HEAD(list);
984
985         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
986                                        bytenr + num_bytes - 1, &list);
987         if (ret == 0 && list_empty(&list))
988                 return 0;
989
990         while (!list_empty(&list)) {
991                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
992                 list_del(&sums->list);
993                 kfree(sums);
994         }
995         return 1;
996 }
997
998 /*
999  * when nowcow writeback call back.  This checks for snapshots or COW copies
1000  * of the extents that exist in the file, and COWs the file as required.
1001  *
1002  * If no cow copies or snapshots exist, we write directly to the existing
1003  * blocks on disk
1004  */
1005 static noinline int run_delalloc_nocow(struct inode *inode,
1006                                        struct page *locked_page,
1007                               u64 start, u64 end, int *page_started, int force,
1008                               unsigned long *nr_written)
1009 {
1010         struct btrfs_root *root = BTRFS_I(inode)->root;
1011         struct btrfs_trans_handle *trans;
1012         struct extent_buffer *leaf;
1013         struct btrfs_path *path;
1014         struct btrfs_file_extent_item *fi;
1015         struct btrfs_key found_key;
1016         u64 cow_start;
1017         u64 cur_offset;
1018         u64 extent_end;
1019         u64 extent_offset;
1020         u64 disk_bytenr;
1021         u64 num_bytes;
1022         int extent_type;
1023         int ret;
1024         int type;
1025         int nocow;
1026         int check_prev = 1;
1027
1028         path = btrfs_alloc_path();
1029         BUG_ON(!path);
1030         trans = btrfs_join_transaction(root, 1);
1031         BUG_ON(!trans);
1032
1033         cow_start = (u64)-1;
1034         cur_offset = start;
1035         while (1) {
1036                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1037                                                cur_offset, 0);
1038                 BUG_ON(ret < 0);
1039                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1040                         leaf = path->nodes[0];
1041                         btrfs_item_key_to_cpu(leaf, &found_key,
1042                                               path->slots[0] - 1);
1043                         if (found_key.objectid == inode->i_ino &&
1044                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1045                                 path->slots[0]--;
1046                 }
1047                 check_prev = 0;
1048 next_slot:
1049                 leaf = path->nodes[0];
1050                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1051                         ret = btrfs_next_leaf(root, path);
1052                         if (ret < 0)
1053                                 BUG_ON(1);
1054                         if (ret > 0)
1055                                 break;
1056                         leaf = path->nodes[0];
1057                 }
1058
1059                 nocow = 0;
1060                 disk_bytenr = 0;
1061                 num_bytes = 0;
1062                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1063
1064                 if (found_key.objectid > inode->i_ino ||
1065                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1066                     found_key.offset > end)
1067                         break;
1068
1069                 if (found_key.offset > cur_offset) {
1070                         extent_end = found_key.offset;
1071                         extent_type = 0;
1072                         goto out_check;
1073                 }
1074
1075                 fi = btrfs_item_ptr(leaf, path->slots[0],
1076                                     struct btrfs_file_extent_item);
1077                 extent_type = btrfs_file_extent_type(leaf, fi);
1078
1079                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1080                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1081                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1082                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1083                         extent_end = found_key.offset +
1084                                 btrfs_file_extent_num_bytes(leaf, fi);
1085                         if (extent_end <= start) {
1086                                 path->slots[0]++;
1087                                 goto next_slot;
1088                         }
1089                         if (disk_bytenr == 0)
1090                                 goto out_check;
1091                         if (btrfs_file_extent_compression(leaf, fi) ||
1092                             btrfs_file_extent_encryption(leaf, fi) ||
1093                             btrfs_file_extent_other_encoding(leaf, fi))
1094                                 goto out_check;
1095                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1096                                 goto out_check;
1097                         if (btrfs_extent_readonly(root, disk_bytenr))
1098                                 goto out_check;
1099                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1100                                                   found_key.offset -
1101                                                   extent_offset, disk_bytenr))
1102                                 goto out_check;
1103                         disk_bytenr += extent_offset;
1104                         disk_bytenr += cur_offset - found_key.offset;
1105                         num_bytes = min(end + 1, extent_end) - cur_offset;
1106                         /*
1107                          * force cow if csum exists in the range.
1108                          * this ensure that csum for a given extent are
1109                          * either valid or do not exist.
1110                          */
1111                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1112                                 goto out_check;
1113                         nocow = 1;
1114                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1115                         extent_end = found_key.offset +
1116                                 btrfs_file_extent_inline_len(leaf, fi);
1117                         extent_end = ALIGN(extent_end, root->sectorsize);
1118                 } else {
1119                         BUG_ON(1);
1120                 }
1121 out_check:
1122                 if (extent_end <= start) {
1123                         path->slots[0]++;
1124                         goto next_slot;
1125                 }
1126                 if (!nocow) {
1127                         if (cow_start == (u64)-1)
1128                                 cow_start = cur_offset;
1129                         cur_offset = extent_end;
1130                         if (cur_offset > end)
1131                                 break;
1132                         path->slots[0]++;
1133                         goto next_slot;
1134                 }
1135
1136                 btrfs_release_path(root, path);
1137                 if (cow_start != (u64)-1) {
1138                         ret = cow_file_range(inode, locked_page, cow_start,
1139                                         found_key.offset - 1, page_started,
1140                                         nr_written, 1);
1141                         BUG_ON(ret);
1142                         cow_start = (u64)-1;
1143                 }
1144
1145                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1146                         struct extent_map *em;
1147                         struct extent_map_tree *em_tree;
1148                         em_tree = &BTRFS_I(inode)->extent_tree;
1149                         em = alloc_extent_map(GFP_NOFS);
1150                         em->start = cur_offset;
1151                         em->orig_start = em->start;
1152                         em->len = num_bytes;
1153                         em->block_len = num_bytes;
1154                         em->block_start = disk_bytenr;
1155                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1156                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1157                         while (1) {
1158                                 write_lock(&em_tree->lock);
1159                                 ret = add_extent_mapping(em_tree, em);
1160                                 write_unlock(&em_tree->lock);
1161                                 if (ret != -EEXIST) {
1162                                         free_extent_map(em);
1163                                         break;
1164                                 }
1165                                 btrfs_drop_extent_cache(inode, em->start,
1166                                                 em->start + em->len - 1, 0);
1167                         }
1168                         type = BTRFS_ORDERED_PREALLOC;
1169                 } else {
1170                         type = BTRFS_ORDERED_NOCOW;
1171                 }
1172
1173                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1174                                                num_bytes, num_bytes, type);
1175                 BUG_ON(ret);
1176
1177                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1178                                 cur_offset, cur_offset + num_bytes - 1,
1179                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1180                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1181                                 EXTENT_SET_PRIVATE2);
1182                 cur_offset = extent_end;
1183                 if (cur_offset > end)
1184                         break;
1185         }
1186         btrfs_release_path(root, path);
1187
1188         if (cur_offset <= end && cow_start == (u64)-1)
1189                 cow_start = cur_offset;
1190         if (cow_start != (u64)-1) {
1191                 ret = cow_file_range(inode, locked_page, cow_start, end,
1192                                      page_started, nr_written, 1);
1193                 BUG_ON(ret);
1194         }
1195
1196         ret = btrfs_end_transaction(trans, root);
1197         BUG_ON(ret);
1198         btrfs_free_path(path);
1199         return 0;
1200 }
1201
1202 /*
1203  * extent_io.c call back to do delayed allocation processing
1204  */
1205 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1206                               u64 start, u64 end, int *page_started,
1207                               unsigned long *nr_written)
1208 {
1209         int ret;
1210         struct btrfs_root *root = BTRFS_I(inode)->root;
1211
1212         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1213                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1214                                          page_started, 1, nr_written);
1215         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1216                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1217                                          page_started, 0, nr_written);
1218         else if (!btrfs_test_opt(root, COMPRESS) &&
1219                  !(BTRFS_I(inode)->force_compress))
1220                 ret = cow_file_range(inode, locked_page, start, end,
1221                                       page_started, nr_written, 1);
1222         else
1223                 ret = cow_file_range_async(inode, locked_page, start, end,
1224                                            page_started, nr_written);
1225         return ret;
1226 }
1227
1228 static int btrfs_split_extent_hook(struct inode *inode,
1229                                     struct extent_state *orig, u64 split)
1230 {
1231         if (!(orig->state & EXTENT_DELALLOC))
1232                 return 0;
1233
1234         spin_lock(&BTRFS_I(inode)->accounting_lock);
1235         BTRFS_I(inode)->outstanding_extents++;
1236         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1237
1238         return 0;
1239 }
1240
1241 /*
1242  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1243  * extents so we can keep track of new extents that are just merged onto old
1244  * extents, such as when we are doing sequential writes, so we can properly
1245  * account for the metadata space we'll need.
1246  */
1247 static int btrfs_merge_extent_hook(struct inode *inode,
1248                                    struct extent_state *new,
1249                                    struct extent_state *other)
1250 {
1251         /* not delalloc, ignore it */
1252         if (!(other->state & EXTENT_DELALLOC))
1253                 return 0;
1254
1255         spin_lock(&BTRFS_I(inode)->accounting_lock);
1256         BTRFS_I(inode)->outstanding_extents--;
1257         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1258
1259         return 0;
1260 }
1261
1262 /*
1263  * extent_io.c set_bit_hook, used to track delayed allocation
1264  * bytes in this file, and to maintain the list of inodes that
1265  * have pending delalloc work to be done.
1266  */
1267 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1268                        unsigned long old, unsigned long bits)
1269 {
1270
1271         /*
1272          * set_bit and clear bit hooks normally require _irqsave/restore
1273          * but in this case, we are only testeing for the DELALLOC
1274          * bit, which is only set or cleared with irqs on
1275          */
1276         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1277                 struct btrfs_root *root = BTRFS_I(inode)->root;
1278
1279                 spin_lock(&BTRFS_I(inode)->accounting_lock);
1280                 BTRFS_I(inode)->outstanding_extents++;
1281                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1282                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1283
1284                 spin_lock(&root->fs_info->delalloc_lock);
1285                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1286                 root->fs_info->delalloc_bytes += end - start + 1;
1287                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1288                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1289                                       &root->fs_info->delalloc_inodes);
1290                 }
1291                 spin_unlock(&root->fs_info->delalloc_lock);
1292         }
1293         return 0;
1294 }
1295
1296 /*
1297  * extent_io.c clear_bit_hook, see set_bit_hook for why
1298  */
1299 static int btrfs_clear_bit_hook(struct inode *inode,
1300                                 struct extent_state *state, unsigned long bits)
1301 {
1302         /*
1303          * set_bit and clear bit hooks normally require _irqsave/restore
1304          * but in this case, we are only testeing for the DELALLOC
1305          * bit, which is only set or cleared with irqs on
1306          */
1307         if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1308                 struct btrfs_root *root = BTRFS_I(inode)->root;
1309
1310                 if (bits & EXTENT_DO_ACCOUNTING) {
1311                         spin_lock(&BTRFS_I(inode)->accounting_lock);
1312                         WARN_ON(!BTRFS_I(inode)->outstanding_extents);
1313                         BTRFS_I(inode)->outstanding_extents--;
1314                         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1315                         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1316                 }
1317
1318                 spin_lock(&root->fs_info->delalloc_lock);
1319                 if (state->end - state->start + 1 >
1320                     root->fs_info->delalloc_bytes) {
1321                         printk(KERN_INFO "btrfs warning: delalloc account "
1322                                "%llu %llu\n",
1323                                (unsigned long long)
1324                                state->end - state->start + 1,
1325                                (unsigned long long)
1326                                root->fs_info->delalloc_bytes);
1327                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1328                         root->fs_info->delalloc_bytes = 0;
1329                         BTRFS_I(inode)->delalloc_bytes = 0;
1330                 } else {
1331                         btrfs_delalloc_free_space(root, inode,
1332                                                   state->end -
1333                                                   state->start + 1);
1334                         root->fs_info->delalloc_bytes -= state->end -
1335                                 state->start + 1;
1336                         BTRFS_I(inode)->delalloc_bytes -= state->end -
1337                                 state->start + 1;
1338                 }
1339                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1340                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1341                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1342                 }
1343                 spin_unlock(&root->fs_info->delalloc_lock);
1344         }
1345         return 0;
1346 }
1347
1348 /*
1349  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1350  * we don't create bios that span stripes or chunks
1351  */
1352 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1353                          size_t size, struct bio *bio,
1354                          unsigned long bio_flags)
1355 {
1356         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1357         struct btrfs_mapping_tree *map_tree;
1358         u64 logical = (u64)bio->bi_sector << 9;
1359         u64 length = 0;
1360         u64 map_length;
1361         int ret;
1362
1363         if (bio_flags & EXTENT_BIO_COMPRESSED)
1364                 return 0;
1365
1366         length = bio->bi_size;
1367         map_tree = &root->fs_info->mapping_tree;
1368         map_length = length;
1369         ret = btrfs_map_block(map_tree, READ, logical,
1370                               &map_length, NULL, 0);
1371
1372         if (map_length < length + size)
1373                 return 1;
1374         return 0;
1375 }
1376
1377 /*
1378  * in order to insert checksums into the metadata in large chunks,
1379  * we wait until bio submission time.   All the pages in the bio are
1380  * checksummed and sums are attached onto the ordered extent record.
1381  *
1382  * At IO completion time the cums attached on the ordered extent record
1383  * are inserted into the btree
1384  */
1385 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1386                                     struct bio *bio, int mirror_num,
1387                                     unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(inode)->root;
1390         int ret = 0;
1391
1392         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1393         BUG_ON(ret);
1394         return 0;
1395 }
1396
1397 /*
1398  * in order to insert checksums into the metadata in large chunks,
1399  * we wait until bio submission time.   All the pages in the bio are
1400  * checksummed and sums are attached onto the ordered extent record.
1401  *
1402  * At IO completion time the cums attached on the ordered extent record
1403  * are inserted into the btree
1404  */
1405 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1406                           int mirror_num, unsigned long bio_flags)
1407 {
1408         struct btrfs_root *root = BTRFS_I(inode)->root;
1409         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1410 }
1411
1412 /*
1413  * extent_io.c submission hook. This does the right thing for csum calculation
1414  * on write, or reading the csums from the tree before a read
1415  */
1416 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1417                           int mirror_num, unsigned long bio_flags)
1418 {
1419         struct btrfs_root *root = BTRFS_I(inode)->root;
1420         int ret = 0;
1421         int skip_sum;
1422
1423         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1424
1425         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1426         BUG_ON(ret);
1427
1428         if (!(rw & (1 << BIO_RW))) {
1429                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1430                         return btrfs_submit_compressed_read(inode, bio,
1431                                                     mirror_num, bio_flags);
1432                 } else if (!skip_sum)
1433                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1434                 goto mapit;
1435         } else if (!skip_sum) {
1436                 /* csum items have already been cloned */
1437                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1438                         goto mapit;
1439                 /* we're doing a write, do the async checksumming */
1440                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1441                                    inode, rw, bio, mirror_num,
1442                                    bio_flags, __btrfs_submit_bio_start,
1443                                    __btrfs_submit_bio_done);
1444         }
1445
1446 mapit:
1447         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1448 }
1449
1450 /*
1451  * given a list of ordered sums record them in the inode.  This happens
1452  * at IO completion time based on sums calculated at bio submission time.
1453  */
1454 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1455                              struct inode *inode, u64 file_offset,
1456                              struct list_head *list)
1457 {
1458         struct btrfs_ordered_sum *sum;
1459
1460         btrfs_set_trans_block_group(trans, inode);
1461
1462         list_for_each_entry(sum, list, list) {
1463                 btrfs_csum_file_blocks(trans,
1464                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1465         }
1466         return 0;
1467 }
1468
1469 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1470                               struct extent_state **cached_state)
1471 {
1472         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1473                 WARN_ON(1);
1474         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1475                                    cached_state, GFP_NOFS);
1476 }
1477
1478 /* see btrfs_writepage_start_hook for details on why this is required */
1479 struct btrfs_writepage_fixup {
1480         struct page *page;
1481         struct btrfs_work work;
1482 };
1483
1484 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1485 {
1486         struct btrfs_writepage_fixup *fixup;
1487         struct btrfs_ordered_extent *ordered;
1488         struct extent_state *cached_state = NULL;
1489         struct page *page;
1490         struct inode *inode;
1491         u64 page_start;
1492         u64 page_end;
1493
1494         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1495         page = fixup->page;
1496 again:
1497         lock_page(page);
1498         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1499                 ClearPageChecked(page);
1500                 goto out_page;
1501         }
1502
1503         inode = page->mapping->host;
1504         page_start = page_offset(page);
1505         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1506
1507         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1508                          &cached_state, GFP_NOFS);
1509
1510         /* already ordered? We're done */
1511         if (PagePrivate2(page))
1512                 goto out;
1513
1514         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1515         if (ordered) {
1516                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1517                                      page_end, &cached_state, GFP_NOFS);
1518                 unlock_page(page);
1519                 btrfs_start_ordered_extent(inode, ordered, 1);
1520                 goto again;
1521         }
1522
1523         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1524         ClearPageChecked(page);
1525 out:
1526         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1527                              &cached_state, GFP_NOFS);
1528 out_page:
1529         unlock_page(page);
1530         page_cache_release(page);
1531 }
1532
1533 /*
1534  * There are a few paths in the higher layers of the kernel that directly
1535  * set the page dirty bit without asking the filesystem if it is a
1536  * good idea.  This causes problems because we want to make sure COW
1537  * properly happens and the data=ordered rules are followed.
1538  *
1539  * In our case any range that doesn't have the ORDERED bit set
1540  * hasn't been properly setup for IO.  We kick off an async process
1541  * to fix it up.  The async helper will wait for ordered extents, set
1542  * the delalloc bit and make it safe to write the page.
1543  */
1544 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1545 {
1546         struct inode *inode = page->mapping->host;
1547         struct btrfs_writepage_fixup *fixup;
1548         struct btrfs_root *root = BTRFS_I(inode)->root;
1549
1550         /* this page is properly in the ordered list */
1551         if (TestClearPagePrivate2(page))
1552                 return 0;
1553
1554         if (PageChecked(page))
1555                 return -EAGAIN;
1556
1557         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1558         if (!fixup)
1559                 return -EAGAIN;
1560
1561         SetPageChecked(page);
1562         page_cache_get(page);
1563         fixup->work.func = btrfs_writepage_fixup_worker;
1564         fixup->page = page;
1565         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1566         return -EAGAIN;
1567 }
1568
1569 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1570                                        struct inode *inode, u64 file_pos,
1571                                        u64 disk_bytenr, u64 disk_num_bytes,
1572                                        u64 num_bytes, u64 ram_bytes,
1573                                        u8 compression, u8 encryption,
1574                                        u16 other_encoding, int extent_type)
1575 {
1576         struct btrfs_root *root = BTRFS_I(inode)->root;
1577         struct btrfs_file_extent_item *fi;
1578         struct btrfs_path *path;
1579         struct extent_buffer *leaf;
1580         struct btrfs_key ins;
1581         u64 hint;
1582         int ret;
1583
1584         path = btrfs_alloc_path();
1585         BUG_ON(!path);
1586
1587         path->leave_spinning = 1;
1588
1589         /*
1590          * we may be replacing one extent in the tree with another.
1591          * The new extent is pinned in the extent map, and we don't want
1592          * to drop it from the cache until it is completely in the btree.
1593          *
1594          * So, tell btrfs_drop_extents to leave this extent in the cache.
1595          * the caller is expected to unpin it and allow it to be merged
1596          * with the others.
1597          */
1598         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1599                                  &hint, 0);
1600         BUG_ON(ret);
1601
1602         ins.objectid = inode->i_ino;
1603         ins.offset = file_pos;
1604         ins.type = BTRFS_EXTENT_DATA_KEY;
1605         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1606         BUG_ON(ret);
1607         leaf = path->nodes[0];
1608         fi = btrfs_item_ptr(leaf, path->slots[0],
1609                             struct btrfs_file_extent_item);
1610         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1611         btrfs_set_file_extent_type(leaf, fi, extent_type);
1612         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1613         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1614         btrfs_set_file_extent_offset(leaf, fi, 0);
1615         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1616         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1617         btrfs_set_file_extent_compression(leaf, fi, compression);
1618         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1619         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1620
1621         btrfs_unlock_up_safe(path, 1);
1622         btrfs_set_lock_blocking(leaf);
1623
1624         btrfs_mark_buffer_dirty(leaf);
1625
1626         inode_add_bytes(inode, num_bytes);
1627
1628         ins.objectid = disk_bytenr;
1629         ins.offset = disk_num_bytes;
1630         ins.type = BTRFS_EXTENT_ITEM_KEY;
1631         ret = btrfs_alloc_reserved_file_extent(trans, root,
1632                                         root->root_key.objectid,
1633                                         inode->i_ino, file_pos, &ins);
1634         BUG_ON(ret);
1635         btrfs_free_path(path);
1636
1637         return 0;
1638 }
1639
1640 /*
1641  * helper function for btrfs_finish_ordered_io, this
1642  * just reads in some of the csum leaves to prime them into ram
1643  * before we start the transaction.  It limits the amount of btree
1644  * reads required while inside the transaction.
1645  */
1646 /* as ordered data IO finishes, this gets called so we can finish
1647  * an ordered extent if the range of bytes in the file it covers are
1648  * fully written.
1649  */
1650 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1651 {
1652         struct btrfs_root *root = BTRFS_I(inode)->root;
1653         struct btrfs_trans_handle *trans;
1654         struct btrfs_ordered_extent *ordered_extent = NULL;
1655         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1656         struct extent_state *cached_state = NULL;
1657         int compressed = 0;
1658         int ret;
1659
1660         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1661                                              end - start + 1);
1662         if (!ret)
1663                 return 0;
1664         BUG_ON(!ordered_extent);
1665
1666         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1667                 BUG_ON(!list_empty(&ordered_extent->list));
1668                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1669                 if (!ret) {
1670                         trans = btrfs_join_transaction(root, 1);
1671                         ret = btrfs_update_inode(trans, root, inode);
1672                         BUG_ON(ret);
1673                         btrfs_end_transaction(trans, root);
1674                 }
1675                 goto out;
1676         }
1677
1678         lock_extent_bits(io_tree, ordered_extent->file_offset,
1679                          ordered_extent->file_offset + ordered_extent->len - 1,
1680                          0, &cached_state, GFP_NOFS);
1681
1682         trans = btrfs_join_transaction(root, 1);
1683
1684         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1685                 compressed = 1;
1686         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1687                 BUG_ON(compressed);
1688                 ret = btrfs_mark_extent_written(trans, inode,
1689                                                 ordered_extent->file_offset,
1690                                                 ordered_extent->file_offset +
1691                                                 ordered_extent->len);
1692                 BUG_ON(ret);
1693         } else {
1694                 ret = insert_reserved_file_extent(trans, inode,
1695                                                 ordered_extent->file_offset,
1696                                                 ordered_extent->start,
1697                                                 ordered_extent->disk_len,
1698                                                 ordered_extent->len,
1699                                                 ordered_extent->len,
1700                                                 compressed, 0, 0,
1701                                                 BTRFS_FILE_EXTENT_REG);
1702                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1703                                    ordered_extent->file_offset,
1704                                    ordered_extent->len);
1705                 BUG_ON(ret);
1706         }
1707         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1708                              ordered_extent->file_offset +
1709                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1710
1711         add_pending_csums(trans, inode, ordered_extent->file_offset,
1712                           &ordered_extent->list);
1713
1714         /* this also removes the ordered extent from the tree */
1715         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716         ret = btrfs_update_inode(trans, root, inode);
1717         BUG_ON(ret);
1718         btrfs_end_transaction(trans, root);
1719 out:
1720         /* once for us */
1721         btrfs_put_ordered_extent(ordered_extent);
1722         /* once for the tree */
1723         btrfs_put_ordered_extent(ordered_extent);
1724
1725         return 0;
1726 }
1727
1728 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1729                                 struct extent_state *state, int uptodate)
1730 {
1731         ClearPagePrivate2(page);
1732         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1733 }
1734
1735 /*
1736  * When IO fails, either with EIO or csum verification fails, we
1737  * try other mirrors that might have a good copy of the data.  This
1738  * io_failure_record is used to record state as we go through all the
1739  * mirrors.  If another mirror has good data, the page is set up to date
1740  * and things continue.  If a good mirror can't be found, the original
1741  * bio end_io callback is called to indicate things have failed.
1742  */
1743 struct io_failure_record {
1744         struct page *page;
1745         u64 start;
1746         u64 len;
1747         u64 logical;
1748         unsigned long bio_flags;
1749         int last_mirror;
1750 };
1751
1752 static int btrfs_io_failed_hook(struct bio *failed_bio,
1753                          struct page *page, u64 start, u64 end,
1754                          struct extent_state *state)
1755 {
1756         struct io_failure_record *failrec = NULL;
1757         u64 private;
1758         struct extent_map *em;
1759         struct inode *inode = page->mapping->host;
1760         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1761         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1762         struct bio *bio;
1763         int num_copies;
1764         int ret;
1765         int rw;
1766         u64 logical;
1767
1768         ret = get_state_private(failure_tree, start, &private);
1769         if (ret) {
1770                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1771                 if (!failrec)
1772                         return -ENOMEM;
1773                 failrec->start = start;
1774                 failrec->len = end - start + 1;
1775                 failrec->last_mirror = 0;
1776                 failrec->bio_flags = 0;
1777
1778                 read_lock(&em_tree->lock);
1779                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1780                 if (em->start > start || em->start + em->len < start) {
1781                         free_extent_map(em);
1782                         em = NULL;
1783                 }
1784                 read_unlock(&em_tree->lock);
1785
1786                 if (!em || IS_ERR(em)) {
1787                         kfree(failrec);
1788                         return -EIO;
1789                 }
1790                 logical = start - em->start;
1791                 logical = em->block_start + logical;
1792                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1793                         logical = em->block_start;
1794                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1795                 }
1796                 failrec->logical = logical;
1797                 free_extent_map(em);
1798                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1799                                 EXTENT_DIRTY, GFP_NOFS);
1800                 set_state_private(failure_tree, start,
1801                                  (u64)(unsigned long)failrec);
1802         } else {
1803                 failrec = (struct io_failure_record *)(unsigned long)private;
1804         }
1805         num_copies = btrfs_num_copies(
1806                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1807                               failrec->logical, failrec->len);
1808         failrec->last_mirror++;
1809         if (!state) {
1810                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1811                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1812                                                     failrec->start,
1813                                                     EXTENT_LOCKED);
1814                 if (state && state->start != failrec->start)
1815                         state = NULL;
1816                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1817         }
1818         if (!state || failrec->last_mirror > num_copies) {
1819                 set_state_private(failure_tree, failrec->start, 0);
1820                 clear_extent_bits(failure_tree, failrec->start,
1821                                   failrec->start + failrec->len - 1,
1822                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1823                 kfree(failrec);
1824                 return -EIO;
1825         }
1826         bio = bio_alloc(GFP_NOFS, 1);
1827         bio->bi_private = state;
1828         bio->bi_end_io = failed_bio->bi_end_io;
1829         bio->bi_sector = failrec->logical >> 9;
1830         bio->bi_bdev = failed_bio->bi_bdev;
1831         bio->bi_size = 0;
1832
1833         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1834         if (failed_bio->bi_rw & (1 << BIO_RW))
1835                 rw = WRITE;
1836         else
1837                 rw = READ;
1838
1839         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1840                                                       failrec->last_mirror,
1841                                                       failrec->bio_flags);
1842         return 0;
1843 }
1844
1845 /*
1846  * each time an IO finishes, we do a fast check in the IO failure tree
1847  * to see if we need to process or clean up an io_failure_record
1848  */
1849 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1850 {
1851         u64 private;
1852         u64 private_failure;
1853         struct io_failure_record *failure;
1854         int ret;
1855
1856         private = 0;
1857         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1858                              (u64)-1, 1, EXTENT_DIRTY)) {
1859                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1860                                         start, &private_failure);
1861                 if (ret == 0) {
1862                         failure = (struct io_failure_record *)(unsigned long)
1863                                    private_failure;
1864                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1865                                           failure->start, 0);
1866                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1867                                           failure->start,
1868                                           failure->start + failure->len - 1,
1869                                           EXTENT_DIRTY | EXTENT_LOCKED,
1870                                           GFP_NOFS);
1871                         kfree(failure);
1872                 }
1873         }
1874         return 0;
1875 }
1876
1877 /*
1878  * when reads are done, we need to check csums to verify the data is correct
1879  * if there's a match, we allow the bio to finish.  If not, we go through
1880  * the io_failure_record routines to find good copies
1881  */
1882 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1883                                struct extent_state *state)
1884 {
1885         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1886         struct inode *inode = page->mapping->host;
1887         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1888         char *kaddr;
1889         u64 private = ~(u32)0;
1890         int ret;
1891         struct btrfs_root *root = BTRFS_I(inode)->root;
1892         u32 csum = ~(u32)0;
1893
1894         if (PageChecked(page)) {
1895                 ClearPageChecked(page);
1896                 goto good;
1897         }
1898
1899         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1900                 return 0;
1901
1902         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1903             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1904                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1905                                   GFP_NOFS);
1906                 return 0;
1907         }
1908
1909         if (state && state->start == start) {
1910                 private = state->private;
1911                 ret = 0;
1912         } else {
1913                 ret = get_state_private(io_tree, start, &private);
1914         }
1915         kaddr = kmap_atomic(page, KM_USER0);
1916         if (ret)
1917                 goto zeroit;
1918
1919         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1920         btrfs_csum_final(csum, (char *)&csum);
1921         if (csum != private)
1922                 goto zeroit;
1923
1924         kunmap_atomic(kaddr, KM_USER0);
1925 good:
1926         /* if the io failure tree for this inode is non-empty,
1927          * check to see if we've recovered from a failed IO
1928          */
1929         btrfs_clean_io_failures(inode, start);
1930         return 0;
1931
1932 zeroit:
1933         if (printk_ratelimit()) {
1934                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1935                        "private %llu\n", page->mapping->host->i_ino,
1936                        (unsigned long long)start, csum,
1937                        (unsigned long long)private);
1938         }
1939         memset(kaddr + offset, 1, end - start + 1);
1940         flush_dcache_page(page);
1941         kunmap_atomic(kaddr, KM_USER0);
1942         if (private == 0)
1943                 return 0;
1944         return -EIO;
1945 }
1946
1947 struct delayed_iput {
1948         struct list_head list;
1949         struct inode *inode;
1950 };
1951
1952 void btrfs_add_delayed_iput(struct inode *inode)
1953 {
1954         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1955         struct delayed_iput *delayed;
1956
1957         if (atomic_add_unless(&inode->i_count, -1, 1))
1958                 return;
1959
1960         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1961         delayed->inode = inode;
1962
1963         spin_lock(&fs_info->delayed_iput_lock);
1964         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1965         spin_unlock(&fs_info->delayed_iput_lock);
1966 }
1967
1968 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1969 {
1970         LIST_HEAD(list);
1971         struct btrfs_fs_info *fs_info = root->fs_info;
1972         struct delayed_iput *delayed;
1973         int empty;
1974
1975         spin_lock(&fs_info->delayed_iput_lock);
1976         empty = list_empty(&fs_info->delayed_iputs);
1977         spin_unlock(&fs_info->delayed_iput_lock);
1978         if (empty)
1979                 return;
1980
1981         down_read(&root->fs_info->cleanup_work_sem);
1982         spin_lock(&fs_info->delayed_iput_lock);
1983         list_splice_init(&fs_info->delayed_iputs, &list);
1984         spin_unlock(&fs_info->delayed_iput_lock);
1985
1986         while (!list_empty(&list)) {
1987                 delayed = list_entry(list.next, struct delayed_iput, list);
1988                 list_del(&delayed->list);
1989                 iput(delayed->inode);
1990                 kfree(delayed);
1991         }
1992         up_read(&root->fs_info->cleanup_work_sem);
1993 }
1994
1995 /*
1996  * This creates an orphan entry for the given inode in case something goes
1997  * wrong in the middle of an unlink/truncate.
1998  */
1999 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2000 {
2001         struct btrfs_root *root = BTRFS_I(inode)->root;
2002         int ret = 0;
2003
2004         spin_lock(&root->list_lock);
2005
2006         /* already on the orphan list, we're good */
2007         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2008                 spin_unlock(&root->list_lock);
2009                 return 0;
2010         }
2011
2012         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2013
2014         spin_unlock(&root->list_lock);
2015
2016         /*
2017          * insert an orphan item to track this unlinked/truncated file
2018          */
2019         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2020
2021         return ret;
2022 }
2023
2024 /*
2025  * We have done the truncate/delete so we can go ahead and remove the orphan
2026  * item for this particular inode.
2027  */
2028 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2029 {
2030         struct btrfs_root *root = BTRFS_I(inode)->root;
2031         int ret = 0;
2032
2033         spin_lock(&root->list_lock);
2034
2035         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2036                 spin_unlock(&root->list_lock);
2037                 return 0;
2038         }
2039
2040         list_del_init(&BTRFS_I(inode)->i_orphan);
2041         if (!trans) {
2042                 spin_unlock(&root->list_lock);
2043                 return 0;
2044         }
2045
2046         spin_unlock(&root->list_lock);
2047
2048         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2049
2050         return ret;
2051 }
2052
2053 /*
2054  * this cleans up any orphans that may be left on the list from the last use
2055  * of this root.
2056  */
2057 void btrfs_orphan_cleanup(struct btrfs_root *root)
2058 {
2059         struct btrfs_path *path;
2060         struct extent_buffer *leaf;
2061         struct btrfs_item *item;
2062         struct btrfs_key key, found_key;
2063         struct btrfs_trans_handle *trans;
2064         struct inode *inode;
2065         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2066
2067         if (!xchg(&root->clean_orphans, 0))
2068                 return;
2069
2070         path = btrfs_alloc_path();
2071         BUG_ON(!path);
2072         path->reada = -1;
2073
2074         key.objectid = BTRFS_ORPHAN_OBJECTID;
2075         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2076         key.offset = (u64)-1;
2077
2078         while (1) {
2079                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2080                 if (ret < 0) {
2081                         printk(KERN_ERR "Error searching slot for orphan: %d"
2082                                "\n", ret);
2083                         break;
2084                 }
2085
2086                 /*
2087                  * if ret == 0 means we found what we were searching for, which
2088                  * is weird, but possible, so only screw with path if we didnt
2089                  * find the key and see if we have stuff that matches
2090                  */
2091                 if (ret > 0) {
2092                         if (path->slots[0] == 0)
2093                                 break;
2094                         path->slots[0]--;
2095                 }
2096
2097                 /* pull out the item */
2098                 leaf = path->nodes[0];
2099                 item = btrfs_item_nr(leaf, path->slots[0]);
2100                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2101
2102                 /* make sure the item matches what we want */
2103                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2104                         break;
2105                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2106                         break;
2107
2108                 /* release the path since we're done with it */
2109                 btrfs_release_path(root, path);
2110
2111                 /*
2112                  * this is where we are basically btrfs_lookup, without the
2113                  * crossing root thing.  we store the inode number in the
2114                  * offset of the orphan item.
2115                  */
2116                 found_key.objectid = found_key.offset;
2117                 found_key.type = BTRFS_INODE_ITEM_KEY;
2118                 found_key.offset = 0;
2119                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2120                 if (IS_ERR(inode))
2121                         break;
2122
2123                 /*
2124                  * add this inode to the orphan list so btrfs_orphan_del does
2125                  * the proper thing when we hit it
2126                  */
2127                 spin_lock(&root->list_lock);
2128                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2129                 spin_unlock(&root->list_lock);
2130
2131                 /*
2132                  * if this is a bad inode, means we actually succeeded in
2133                  * removing the inode, but not the orphan record, which means
2134                  * we need to manually delete the orphan since iput will just
2135                  * do a destroy_inode
2136                  */
2137                 if (is_bad_inode(inode)) {
2138                         trans = btrfs_start_transaction(root, 1);
2139                         btrfs_orphan_del(trans, inode);
2140                         btrfs_end_transaction(trans, root);
2141                         iput(inode);
2142                         continue;
2143                 }
2144
2145                 /* if we have links, this was a truncate, lets do that */
2146                 if (inode->i_nlink) {
2147                         nr_truncate++;
2148                         btrfs_truncate(inode);
2149                 } else {
2150                         nr_unlink++;
2151                 }
2152
2153                 /* this will do delete_inode and everything for us */
2154                 iput(inode);
2155         }
2156
2157         if (nr_unlink)
2158                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2159         if (nr_truncate)
2160                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2161
2162         btrfs_free_path(path);
2163 }
2164
2165 /*
2166  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2167  * don't find any xattrs, we know there can't be any acls.
2168  *
2169  * slot is the slot the inode is in, objectid is the objectid of the inode
2170  */
2171 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2172                                           int slot, u64 objectid)
2173 {
2174         u32 nritems = btrfs_header_nritems(leaf);
2175         struct btrfs_key found_key;
2176         int scanned = 0;
2177
2178         slot++;
2179         while (slot < nritems) {
2180                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2181
2182                 /* we found a different objectid, there must not be acls */
2183                 if (found_key.objectid != objectid)
2184                         return 0;
2185
2186                 /* we found an xattr, assume we've got an acl */
2187                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2188                         return 1;
2189
2190                 /*
2191                  * we found a key greater than an xattr key, there can't
2192                  * be any acls later on
2193                  */
2194                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2195                         return 0;
2196
2197                 slot++;
2198                 scanned++;
2199
2200                 /*
2201                  * it goes inode, inode backrefs, xattrs, extents,
2202                  * so if there are a ton of hard links to an inode there can
2203                  * be a lot of backrefs.  Don't waste time searching too hard,
2204                  * this is just an optimization
2205                  */
2206                 if (scanned >= 8)
2207                         break;
2208         }
2209         /* we hit the end of the leaf before we found an xattr or
2210          * something larger than an xattr.  We have to assume the inode
2211          * has acls
2212          */
2213         return 1;
2214 }
2215
2216 /*
2217  * read an inode from the btree into the in-memory inode
2218  */
2219 static void btrfs_read_locked_inode(struct inode *inode)
2220 {
2221         struct btrfs_path *path;
2222         struct extent_buffer *leaf;
2223         struct btrfs_inode_item *inode_item;
2224         struct btrfs_timespec *tspec;
2225         struct btrfs_root *root = BTRFS_I(inode)->root;
2226         struct btrfs_key location;
2227         int maybe_acls;
2228         u64 alloc_group_block;
2229         u32 rdev;
2230         int ret;
2231
2232         path = btrfs_alloc_path();
2233         BUG_ON(!path);
2234         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2235
2236         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2237         if (ret)
2238                 goto make_bad;
2239
2240         leaf = path->nodes[0];
2241         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2242                                     struct btrfs_inode_item);
2243
2244         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2245         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2246         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2247         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2248         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2249
2250         tspec = btrfs_inode_atime(inode_item);
2251         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2252         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2253
2254         tspec = btrfs_inode_mtime(inode_item);
2255         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2256         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2257
2258         tspec = btrfs_inode_ctime(inode_item);
2259         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2260         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2261
2262         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2263         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2264         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2265         inode->i_generation = BTRFS_I(inode)->generation;
2266         inode->i_rdev = 0;
2267         rdev = btrfs_inode_rdev(leaf, inode_item);
2268
2269         BTRFS_I(inode)->index_cnt = (u64)-1;
2270         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2271
2272         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2273
2274         /*
2275          * try to precache a NULL acl entry for files that don't have
2276          * any xattrs or acls
2277          */
2278         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2279         if (!maybe_acls)
2280                 cache_no_acl(inode);
2281
2282         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2283                                                 alloc_group_block, 0);
2284         btrfs_free_path(path);
2285         inode_item = NULL;
2286
2287         switch (inode->i_mode & S_IFMT) {
2288         case S_IFREG:
2289                 inode->i_mapping->a_ops = &btrfs_aops;
2290                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2291                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2292                 inode->i_fop = &btrfs_file_operations;
2293                 inode->i_op = &btrfs_file_inode_operations;
2294                 break;
2295         case S_IFDIR:
2296                 inode->i_fop = &btrfs_dir_file_operations;
2297                 if (root == root->fs_info->tree_root)
2298                         inode->i_op = &btrfs_dir_ro_inode_operations;
2299                 else
2300                         inode->i_op = &btrfs_dir_inode_operations;
2301                 break;
2302         case S_IFLNK:
2303                 inode->i_op = &btrfs_symlink_inode_operations;
2304                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2305                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2306                 break;
2307         default:
2308                 inode->i_op = &btrfs_special_inode_operations;
2309                 init_special_inode(inode, inode->i_mode, rdev);
2310                 break;
2311         }
2312
2313         btrfs_update_iflags(inode);
2314         return;
2315
2316 make_bad:
2317         btrfs_free_path(path);
2318         make_bad_inode(inode);
2319 }
2320
2321 /*
2322  * given a leaf and an inode, copy the inode fields into the leaf
2323  */
2324 static void fill_inode_item(struct btrfs_trans_handle *trans,
2325                             struct extent_buffer *leaf,
2326                             struct btrfs_inode_item *item,
2327                             struct inode *inode)
2328 {
2329         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2330         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2331         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2332         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2333         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2334
2335         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2336                                inode->i_atime.tv_sec);
2337         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2338                                 inode->i_atime.tv_nsec);
2339
2340         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2341                                inode->i_mtime.tv_sec);
2342         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2343                                 inode->i_mtime.tv_nsec);
2344
2345         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2346                                inode->i_ctime.tv_sec);
2347         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2348                                 inode->i_ctime.tv_nsec);
2349
2350         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2351         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2352         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2353         btrfs_set_inode_transid(leaf, item, trans->transid);
2354         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2355         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2356         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2357 }
2358
2359 /*
2360  * copy everything in the in-memory inode into the btree.
2361  */
2362 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2363                                 struct btrfs_root *root, struct inode *inode)
2364 {
2365         struct btrfs_inode_item *inode_item;
2366         struct btrfs_path *path;
2367         struct extent_buffer *leaf;
2368         int ret;
2369
2370         path = btrfs_alloc_path();
2371         BUG_ON(!path);
2372         path->leave_spinning = 1;
2373         ret = btrfs_lookup_inode(trans, root, path,
2374                                  &BTRFS_I(inode)->location, 1);
2375         if (ret) {
2376                 if (ret > 0)
2377                         ret = -ENOENT;
2378                 goto failed;
2379         }
2380
2381         btrfs_unlock_up_safe(path, 1);
2382         leaf = path->nodes[0];
2383         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2384                                   struct btrfs_inode_item);
2385
2386         fill_inode_item(trans, leaf, inode_item, inode);
2387         btrfs_mark_buffer_dirty(leaf);
2388         btrfs_set_inode_last_trans(trans, inode);
2389         ret = 0;
2390 failed:
2391         btrfs_free_path(path);
2392         return ret;
2393 }
2394
2395
2396 /*
2397  * unlink helper that gets used here in inode.c and in the tree logging
2398  * recovery code.  It remove a link in a directory with a given name, and
2399  * also drops the back refs in the inode to the directory
2400  */
2401 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2402                        struct btrfs_root *root,
2403                        struct inode *dir, struct inode *inode,
2404                        const char *name, int name_len)
2405 {
2406         struct btrfs_path *path;
2407         int ret = 0;
2408         struct extent_buffer *leaf;
2409         struct btrfs_dir_item *di;
2410         struct btrfs_key key;
2411         u64 index;
2412
2413         path = btrfs_alloc_path();
2414         if (!path) {
2415                 ret = -ENOMEM;
2416                 goto err;
2417         }
2418
2419         path->leave_spinning = 1;
2420         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2421                                     name, name_len, -1);
2422         if (IS_ERR(di)) {
2423                 ret = PTR_ERR(di);
2424                 goto err;
2425         }
2426         if (!di) {
2427                 ret = -ENOENT;
2428                 goto err;
2429         }
2430         leaf = path->nodes[0];
2431         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2432         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2433         if (ret)
2434                 goto err;
2435         btrfs_release_path(root, path);
2436
2437         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2438                                   inode->i_ino,
2439                                   dir->i_ino, &index);
2440         if (ret) {
2441                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2442                        "inode %lu parent %lu\n", name_len, name,
2443                        inode->i_ino, dir->i_ino);
2444                 goto err;
2445         }
2446
2447         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2448                                          index, name, name_len, -1);
2449         if (IS_ERR(di)) {
2450                 ret = PTR_ERR(di);
2451                 goto err;
2452         }
2453         if (!di) {
2454                 ret = -ENOENT;
2455                 goto err;
2456         }
2457         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2458         btrfs_release_path(root, path);
2459
2460         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2461                                          inode, dir->i_ino);
2462         BUG_ON(ret != 0 && ret != -ENOENT);
2463
2464         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2465                                            dir, index);
2466         BUG_ON(ret);
2467 err:
2468         btrfs_free_path(path);
2469         if (ret)
2470                 goto out;
2471
2472         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2473         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2474         btrfs_update_inode(trans, root, dir);
2475         btrfs_drop_nlink(inode);
2476         ret = btrfs_update_inode(trans, root, inode);
2477 out:
2478         return ret;
2479 }
2480
2481 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2482 {
2483         struct btrfs_root *root;
2484         struct btrfs_trans_handle *trans;
2485         struct inode *inode = dentry->d_inode;
2486         int ret;
2487         unsigned long nr = 0;
2488
2489         root = BTRFS_I(dir)->root;
2490
2491         /*
2492          * 5 items for unlink inode
2493          * 1 for orphan
2494          */
2495         ret = btrfs_reserve_metadata_space(root, 6);
2496         if (ret)
2497                 return ret;
2498
2499         trans = btrfs_start_transaction(root, 1);
2500         if (IS_ERR(trans)) {
2501                 btrfs_unreserve_metadata_space(root, 6);
2502                 return PTR_ERR(trans);
2503         }
2504
2505         btrfs_set_trans_block_group(trans, dir);
2506
2507         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2508
2509         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2510                                  dentry->d_name.name, dentry->d_name.len);
2511
2512         if (inode->i_nlink == 0)
2513                 ret = btrfs_orphan_add(trans, inode);
2514
2515         nr = trans->blocks_used;
2516
2517         btrfs_end_transaction_throttle(trans, root);
2518         btrfs_unreserve_metadata_space(root, 6);
2519         btrfs_btree_balance_dirty(root, nr);
2520         return ret;
2521 }
2522
2523 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2524                         struct btrfs_root *root,
2525                         struct inode *dir, u64 objectid,
2526                         const char *name, int name_len)
2527 {
2528         struct btrfs_path *path;
2529         struct extent_buffer *leaf;
2530         struct btrfs_dir_item *di;
2531         struct btrfs_key key;
2532         u64 index;
2533         int ret;
2534
2535         path = btrfs_alloc_path();
2536         if (!path)
2537                 return -ENOMEM;
2538
2539         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2540                                    name, name_len, -1);
2541         BUG_ON(!di || IS_ERR(di));
2542
2543         leaf = path->nodes[0];
2544         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2545         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2546         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2547         BUG_ON(ret);
2548         btrfs_release_path(root, path);
2549
2550         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2551                                  objectid, root->root_key.objectid,
2552                                  dir->i_ino, &index, name, name_len);
2553         if (ret < 0) {
2554                 BUG_ON(ret != -ENOENT);
2555                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2556                                                  name, name_len);
2557                 BUG_ON(!di || IS_ERR(di));
2558
2559                 leaf = path->nodes[0];
2560                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2561                 btrfs_release_path(root, path);
2562                 index = key.offset;
2563         }
2564
2565         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2566                                          index, name, name_len, -1);
2567         BUG_ON(!di || IS_ERR(di));
2568
2569         leaf = path->nodes[0];
2570         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2571         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2572         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2573         BUG_ON(ret);
2574         btrfs_release_path(root, path);
2575
2576         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2577         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2578         ret = btrfs_update_inode(trans, root, dir);
2579         BUG_ON(ret);
2580         dir->i_sb->s_dirt = 1;
2581
2582         btrfs_free_path(path);
2583         return 0;
2584 }
2585
2586 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2587 {
2588         struct inode *inode = dentry->d_inode;
2589         int err = 0;
2590         int ret;
2591         struct btrfs_root *root = BTRFS_I(dir)->root;
2592         struct btrfs_trans_handle *trans;
2593         unsigned long nr = 0;
2594
2595         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2596             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2597                 return -ENOTEMPTY;
2598
2599         ret = btrfs_reserve_metadata_space(root, 5);
2600         if (ret)
2601                 return ret;
2602
2603         trans = btrfs_start_transaction(root, 1);
2604         if (IS_ERR(trans)) {
2605                 btrfs_unreserve_metadata_space(root, 5);
2606                 return PTR_ERR(trans);
2607         }
2608
2609         btrfs_set_trans_block_group(trans, dir);
2610
2611         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2612                 err = btrfs_unlink_subvol(trans, root, dir,
2613                                           BTRFS_I(inode)->location.objectid,
2614                                           dentry->d_name.name,
2615                                           dentry->d_name.len);
2616                 goto out;
2617         }
2618
2619         err = btrfs_orphan_add(trans, inode);
2620         if (err)
2621                 goto out;
2622
2623         /* now the directory is empty */
2624         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2625                                  dentry->d_name.name, dentry->d_name.len);
2626         if (!err)
2627                 btrfs_i_size_write(inode, 0);
2628 out:
2629         nr = trans->blocks_used;
2630         ret = btrfs_end_transaction_throttle(trans, root);
2631         btrfs_unreserve_metadata_space(root, 5);
2632         btrfs_btree_balance_dirty(root, nr);
2633
2634         if (ret && !err)
2635                 err = ret;
2636         return err;
2637 }
2638
2639 #if 0
2640 /*
2641  * when truncating bytes in a file, it is possible to avoid reading
2642  * the leaves that contain only checksum items.  This can be the
2643  * majority of the IO required to delete a large file, but it must
2644  * be done carefully.
2645  *
2646  * The keys in the level just above the leaves are checked to make sure
2647  * the lowest key in a given leaf is a csum key, and starts at an offset
2648  * after the new  size.
2649  *
2650  * Then the key for the next leaf is checked to make sure it also has
2651  * a checksum item for the same file.  If it does, we know our target leaf
2652  * contains only checksum items, and it can be safely freed without reading
2653  * it.
2654  *
2655  * This is just an optimization targeted at large files.  It may do
2656  * nothing.  It will return 0 unless things went badly.
2657  */
2658 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2659                                      struct btrfs_root *root,
2660                                      struct btrfs_path *path,
2661                                      struct inode *inode, u64 new_size)
2662 {
2663         struct btrfs_key key;
2664         int ret;
2665         int nritems;
2666         struct btrfs_key found_key;
2667         struct btrfs_key other_key;
2668         struct btrfs_leaf_ref *ref;
2669         u64 leaf_gen;
2670         u64 leaf_start;
2671
2672         path->lowest_level = 1;
2673         key.objectid = inode->i_ino;
2674         key.type = BTRFS_CSUM_ITEM_KEY;
2675         key.offset = new_size;
2676 again:
2677         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2678         if (ret < 0)
2679                 goto out;
2680
2681         if (path->nodes[1] == NULL) {
2682                 ret = 0;
2683                 goto out;
2684         }
2685         ret = 0;
2686         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2687         nritems = btrfs_header_nritems(path->nodes[1]);
2688
2689         if (!nritems)
2690                 goto out;
2691
2692         if (path->slots[1] >= nritems)
2693                 goto next_node;
2694
2695         /* did we find a key greater than anything we want to delete? */
2696         if (found_key.objectid > inode->i_ino ||
2697            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2698                 goto out;
2699
2700         /* we check the next key in the node to make sure the leave contains
2701          * only checksum items.  This comparison doesn't work if our
2702          * leaf is the last one in the node
2703          */
2704         if (path->slots[1] + 1 >= nritems) {
2705 next_node:
2706                 /* search forward from the last key in the node, this
2707                  * will bring us into the next node in the tree
2708                  */
2709                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2710
2711                 /* unlikely, but we inc below, so check to be safe */
2712                 if (found_key.offset == (u64)-1)
2713                         goto out;
2714
2715                 /* search_forward needs a path with locks held, do the
2716                  * search again for the original key.  It is possible
2717                  * this will race with a balance and return a path that
2718                  * we could modify, but this drop is just an optimization
2719                  * and is allowed to miss some leaves.
2720                  */
2721                 btrfs_release_path(root, path);
2722                 found_key.offset++;
2723
2724                 /* setup a max key for search_forward */
2725                 other_key.offset = (u64)-1;
2726                 other_key.type = key.type;
2727                 other_key.objectid = key.objectid;
2728
2729                 path->keep_locks = 1;
2730                 ret = btrfs_search_forward(root, &found_key, &other_key,
2731                                            path, 0, 0);
2732                 path->keep_locks = 0;
2733                 if (ret || found_key.objectid != key.objectid ||
2734                     found_key.type != key.type) {
2735                         ret = 0;
2736                         goto out;
2737                 }
2738
2739                 key.offset = found_key.offset;
2740                 btrfs_release_path(root, path);
2741                 cond_resched();
2742                 goto again;
2743         }
2744
2745         /* we know there's one more slot after us in the tree,
2746          * read that key so we can verify it is also a checksum item
2747          */
2748         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2749
2750         if (found_key.objectid < inode->i_ino)
2751                 goto next_key;
2752
2753         if (found_key.type != key.type || found_key.offset < new_size)
2754                 goto next_key;
2755
2756         /*
2757          * if the key for the next leaf isn't a csum key from this objectid,
2758          * we can't be sure there aren't good items inside this leaf.
2759          * Bail out
2760          */
2761         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2762                 goto out;
2763
2764         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2765         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2766         /*
2767          * it is safe to delete this leaf, it contains only
2768          * csum items from this inode at an offset >= new_size
2769          */
2770         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2771         BUG_ON(ret);
2772
2773         if (root->ref_cows && leaf_gen < trans->transid) {
2774                 ref = btrfs_alloc_leaf_ref(root, 0);
2775                 if (ref) {
2776                         ref->root_gen = root->root_key.offset;
2777                         ref->bytenr = leaf_start;
2778                         ref->owner = 0;
2779                         ref->generation = leaf_gen;
2780                         ref->nritems = 0;
2781
2782                         btrfs_sort_leaf_ref(ref);
2783
2784                         ret = btrfs_add_leaf_ref(root, ref, 0);
2785                         WARN_ON(ret);
2786                         btrfs_free_leaf_ref(root, ref);
2787                 } else {
2788                         WARN_ON(1);
2789                 }
2790         }
2791 next_key:
2792         btrfs_release_path(root, path);
2793
2794         if (other_key.objectid == inode->i_ino &&
2795             other_key.type == key.type && other_key.offset > key.offset) {
2796                 key.offset = other_key.offset;
2797                 cond_resched();
2798                 goto again;
2799         }
2800         ret = 0;
2801 out:
2802         /* fixup any changes we've made to the path */
2803         path->lowest_level = 0;
2804         path->keep_locks = 0;
2805         btrfs_release_path(root, path);
2806         return ret;
2807 }
2808
2809 #endif
2810
2811 /*
2812  * this can truncate away extent items, csum items and directory items.
2813  * It starts at a high offset and removes keys until it can't find
2814  * any higher than new_size
2815  *
2816  * csum items that cross the new i_size are truncated to the new size
2817  * as well.
2818  *
2819  * min_type is the minimum key type to truncate down to.  If set to 0, this
2820  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2821  */
2822 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2823                                struct btrfs_root *root,
2824                                struct inode *inode,
2825                                u64 new_size, u32 min_type)
2826 {
2827         struct btrfs_path *path;
2828         struct extent_buffer *leaf;
2829         struct btrfs_file_extent_item *fi;
2830         struct btrfs_key key;
2831         struct btrfs_key found_key;
2832         u64 extent_start = 0;
2833         u64 extent_num_bytes = 0;
2834         u64 extent_offset = 0;
2835         u64 item_end = 0;
2836         u64 mask = root->sectorsize - 1;
2837         u32 found_type = (u8)-1;
2838         int found_extent;
2839         int del_item;
2840         int pending_del_nr = 0;
2841         int pending_del_slot = 0;
2842         int extent_type = -1;
2843         int encoding;
2844         int ret;
2845         int err = 0;
2846
2847         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2848
2849         if (root->ref_cows)
2850                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2851
2852         path = btrfs_alloc_path();
2853         BUG_ON(!path);
2854         path->reada = -1;
2855
2856         key.objectid = inode->i_ino;
2857         key.offset = (u64)-1;
2858         key.type = (u8)-1;
2859
2860 search_again:
2861         path->leave_spinning = 1;
2862         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2863         if (ret < 0) {
2864                 err = ret;
2865                 goto out;
2866         }
2867
2868         if (ret > 0) {
2869                 /* there are no items in the tree for us to truncate, we're
2870                  * done
2871                  */
2872                 if (path->slots[0] == 0)
2873                         goto out;
2874                 path->slots[0]--;
2875         }
2876
2877         while (1) {
2878                 fi = NULL;
2879                 leaf = path->nodes[0];
2880                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2881                 found_type = btrfs_key_type(&found_key);
2882                 encoding = 0;
2883
2884                 if (found_key.objectid != inode->i_ino)
2885                         break;
2886
2887                 if (found_type < min_type)
2888                         break;
2889
2890                 item_end = found_key.offset;
2891                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2892                         fi = btrfs_item_ptr(leaf, path->slots[0],
2893                                             struct btrfs_file_extent_item);
2894                         extent_type = btrfs_file_extent_type(leaf, fi);
2895                         encoding = btrfs_file_extent_compression(leaf, fi);
2896                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2897                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2898
2899                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2900                                 item_end +=
2901                                     btrfs_file_extent_num_bytes(leaf, fi);
2902                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2903                                 item_end += btrfs_file_extent_inline_len(leaf,
2904                                                                          fi);
2905                         }
2906                         item_end--;
2907                 }
2908                 if (found_type > min_type) {
2909                         del_item = 1;
2910                 } else {
2911                         if (item_end < new_size)
2912                                 break;
2913                         if (found_key.offset >= new_size)
2914                                 del_item = 1;
2915                         else
2916                                 del_item = 0;
2917                 }
2918                 found_extent = 0;
2919                 /* FIXME, shrink the extent if the ref count is only 1 */
2920                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2921                         goto delete;
2922
2923                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2924                         u64 num_dec;
2925                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2926                         if (!del_item && !encoding) {
2927                                 u64 orig_num_bytes =
2928                                         btrfs_file_extent_num_bytes(leaf, fi);
2929                                 extent_num_bytes = new_size -
2930                                         found_key.offset + root->sectorsize - 1;
2931                                 extent_num_bytes = extent_num_bytes &
2932                                         ~((u64)root->sectorsize - 1);
2933                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2934                                                          extent_num_bytes);
2935                                 num_dec = (orig_num_bytes -
2936                                            extent_num_bytes);
2937                                 if (root->ref_cows && extent_start != 0)
2938                                         inode_sub_bytes(inode, num_dec);
2939                                 btrfs_mark_buffer_dirty(leaf);
2940                         } else {
2941                                 extent_num_bytes =
2942                                         btrfs_file_extent_disk_num_bytes(leaf,
2943                                                                          fi);
2944                                 extent_offset = found_key.offset -
2945                                         btrfs_file_extent_offset(leaf, fi);
2946
2947                                 /* FIXME blocksize != 4096 */
2948                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2949                                 if (extent_start != 0) {
2950                                         found_extent = 1;
2951                                         if (root->ref_cows)
2952                                                 inode_sub_bytes(inode, num_dec);
2953                                 }
2954                         }
2955                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2956                         /*
2957                          * we can't truncate inline items that have had
2958                          * special encodings
2959                          */
2960                         if (!del_item &&
2961                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2962                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2963                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2964                                 u32 size = new_size - found_key.offset;
2965
2966                                 if (root->ref_cows) {
2967                                         inode_sub_bytes(inode, item_end + 1 -
2968                                                         new_size);
2969                                 }
2970                                 size =
2971                                     btrfs_file_extent_calc_inline_size(size);
2972                                 ret = btrfs_truncate_item(trans, root, path,
2973                                                           size, 1);
2974                                 BUG_ON(ret);
2975                         } else if (root->ref_cows) {
2976                                 inode_sub_bytes(inode, item_end + 1 -
2977                                                 found_key.offset);
2978                         }
2979                 }
2980 delete:
2981                 if (del_item) {
2982                         if (!pending_del_nr) {
2983                                 /* no pending yet, add ourselves */
2984                                 pending_del_slot = path->slots[0];
2985                                 pending_del_nr = 1;
2986                         } else if (pending_del_nr &&
2987                                    path->slots[0] + 1 == pending_del_slot) {
2988                                 /* hop on the pending chunk */
2989                                 pending_del_nr++;
2990                                 pending_del_slot = path->slots[0];
2991                         } else {
2992                                 BUG();
2993                         }
2994                 } else {
2995                         break;
2996                 }
2997                 if (found_extent && root->ref_cows) {
2998                         btrfs_set_path_blocking(path);
2999                         ret = btrfs_free_extent(trans, root, extent_start,
3000                                                 extent_num_bytes, 0,
3001                                                 btrfs_header_owner(leaf),
3002                                                 inode->i_ino, extent_offset);
3003                         BUG_ON(ret);
3004                 }
3005
3006                 if (found_type == BTRFS_INODE_ITEM_KEY)
3007                         break;
3008
3009                 if (path->slots[0] == 0 ||
3010                     path->slots[0] != pending_del_slot) {
3011                         if (root->ref_cows) {
3012                                 err = -EAGAIN;
3013                                 goto out;
3014                         }
3015                         if (pending_del_nr) {
3016                                 ret = btrfs_del_items(trans, root, path,
3017                                                 pending_del_slot,
3018                                                 pending_del_nr);
3019                                 BUG_ON(ret);
3020                                 pending_del_nr = 0;
3021                         }
3022                         btrfs_release_path(root, path);
3023                         goto search_again;
3024                 } else {
3025                         path->slots[0]--;
3026                 }
3027         }
3028 out:
3029         if (pending_del_nr) {
3030                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3031                                       pending_del_nr);
3032         }
3033         btrfs_free_path(path);
3034         return err;
3035 }
3036
3037 /*
3038  * taken from block_truncate_page, but does cow as it zeros out
3039  * any bytes left in the last page in the file.
3040  */
3041 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3042 {
3043         struct inode *inode = mapping->host;
3044         struct btrfs_root *root = BTRFS_I(inode)->root;
3045         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3046         struct btrfs_ordered_extent *ordered;
3047         struct extent_state *cached_state = NULL;
3048         char *kaddr;
3049         u32 blocksize = root->sectorsize;
3050         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3051         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3052         struct page *page;
3053         int ret = 0;
3054         u64 page_start;
3055         u64 page_end;
3056
3057         if ((offset & (blocksize - 1)) == 0)
3058                 goto out;
3059         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3060         if (ret)
3061                 goto out;
3062
3063         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3064         if (ret)
3065                 goto out;
3066
3067         ret = -ENOMEM;
3068 again:
3069         page = grab_cache_page(mapping, index);
3070         if (!page) {
3071                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3072                 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3073                 goto out;
3074         }
3075
3076         page_start = page_offset(page);
3077         page_end = page_start + PAGE_CACHE_SIZE - 1;
3078
3079         if (!PageUptodate(page)) {
3080                 ret = btrfs_readpage(NULL, page);
3081                 lock_page(page);
3082                 if (page->mapping != mapping) {
3083                         unlock_page(page);
3084                         page_cache_release(page);
3085                         goto again;
3086                 }
3087                 if (!PageUptodate(page)) {
3088                         ret = -EIO;
3089                         goto out_unlock;
3090                 }
3091         }
3092         wait_on_page_writeback(page);
3093
3094         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3095                          GFP_NOFS);
3096         set_page_extent_mapped(page);
3097
3098         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3099         if (ordered) {
3100                 unlock_extent_cached(io_tree, page_start, page_end,
3101                                      &cached_state, GFP_NOFS);
3102                 unlock_page(page);
3103                 page_cache_release(page);
3104                 btrfs_start_ordered_extent(inode, ordered, 1);
3105                 btrfs_put_ordered_extent(ordered);
3106                 goto again;
3107         }
3108
3109         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3110                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3111                           0, 0, &cached_state, GFP_NOFS);
3112
3113         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3114                                         &cached_state);
3115         if (ret) {
3116                 unlock_extent_cached(io_tree, page_start, page_end,
3117                                      &cached_state, GFP_NOFS);
3118                 goto out_unlock;
3119         }
3120
3121         ret = 0;
3122         if (offset != PAGE_CACHE_SIZE) {
3123                 kaddr = kmap(page);
3124                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3125                 flush_dcache_page(page);
3126                 kunmap(page);
3127         }
3128         ClearPageChecked(page);
3129         set_page_dirty(page);
3130         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3131                              GFP_NOFS);
3132
3133 out_unlock:
3134         if (ret)
3135                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3136         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3137         unlock_page(page);
3138         page_cache_release(page);
3139 out:
3140         return ret;
3141 }
3142
3143 int btrfs_cont_expand(struct inode *inode, loff_t size)
3144 {
3145         struct btrfs_trans_handle *trans;
3146         struct btrfs_root *root = BTRFS_I(inode)->root;
3147         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3148         struct extent_map *em;
3149         struct extent_state *cached_state = NULL;
3150         u64 mask = root->sectorsize - 1;
3151         u64 hole_start = (inode->i_size + mask) & ~mask;
3152         u64 block_end = (size + mask) & ~mask;
3153         u64 last_byte;
3154         u64 cur_offset;
3155         u64 hole_size;
3156         int err = 0;
3157
3158         if (size <= hole_start)
3159                 return 0;
3160
3161         while (1) {
3162                 struct btrfs_ordered_extent *ordered;
3163                 btrfs_wait_ordered_range(inode, hole_start,
3164                                          block_end - hole_start);
3165                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3166                                  &cached_state, GFP_NOFS);
3167                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3168                 if (!ordered)
3169                         break;
3170                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3171                                      &cached_state, GFP_NOFS);
3172                 btrfs_put_ordered_extent(ordered);
3173         }
3174
3175         cur_offset = hole_start;
3176         while (1) {
3177                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3178                                 block_end - cur_offset, 0);
3179                 BUG_ON(IS_ERR(em) || !em);
3180                 last_byte = min(extent_map_end(em), block_end);
3181                 last_byte = (last_byte + mask) & ~mask;
3182                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3183                         u64 hint_byte = 0;
3184                         hole_size = last_byte - cur_offset;
3185
3186                         err = btrfs_reserve_metadata_space(root, 2);
3187                         if (err)
3188                                 break;
3189
3190                         trans = btrfs_start_transaction(root, 1);
3191                         btrfs_set_trans_block_group(trans, inode);
3192
3193                         err = btrfs_drop_extents(trans, inode, cur_offset,
3194                                                  cur_offset + hole_size,
3195                                                  &hint_byte, 1);
3196                         BUG_ON(err);
3197
3198                         err = btrfs_insert_file_extent(trans, root,
3199                                         inode->i_ino, cur_offset, 0,
3200                                         0, hole_size, 0, hole_size,
3201                                         0, 0, 0);
3202                         BUG_ON(err);
3203
3204                         btrfs_drop_extent_cache(inode, hole_start,
3205                                         last_byte - 1, 0);
3206
3207                         btrfs_end_transaction(trans, root);
3208                         btrfs_unreserve_metadata_space(root, 2);
3209                 }
3210                 free_extent_map(em);
3211                 cur_offset = last_byte;
3212                 if (cur_offset >= block_end)
3213                         break;
3214         }
3215
3216         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3217                              GFP_NOFS);
3218         return err;
3219 }
3220
3221 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3222 {
3223         struct btrfs_root *root = BTRFS_I(inode)->root;
3224         struct btrfs_trans_handle *trans;
3225         unsigned long nr;
3226         int ret;
3227
3228         if (attr->ia_size == inode->i_size)
3229                 return 0;
3230
3231         if (attr->ia_size > inode->i_size) {
3232                 unsigned long limit;
3233                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3234                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3235                         return -EFBIG;
3236                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3237                         send_sig(SIGXFSZ, current, 0);
3238                         return -EFBIG;
3239                 }
3240         }
3241
3242         ret = btrfs_reserve_metadata_space(root, 1);
3243         if (ret)
3244                 return ret;
3245
3246         trans = btrfs_start_transaction(root, 1);
3247         btrfs_set_trans_block_group(trans, inode);
3248
3249         ret = btrfs_orphan_add(trans, inode);
3250         BUG_ON(ret);
3251
3252         nr = trans->blocks_used;
3253         btrfs_end_transaction(trans, root);
3254         btrfs_unreserve_metadata_space(root, 1);
3255         btrfs_btree_balance_dirty(root, nr);
3256
3257         if (attr->ia_size > inode->i_size) {
3258                 ret = btrfs_cont_expand(inode, attr->ia_size);
3259                 if (ret) {
3260                         btrfs_truncate(inode);
3261                         return ret;
3262                 }
3263
3264                 i_size_write(inode, attr->ia_size);
3265                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3266
3267                 trans = btrfs_start_transaction(root, 1);
3268                 btrfs_set_trans_block_group(trans, inode);
3269
3270                 ret = btrfs_update_inode(trans, root, inode);
3271                 BUG_ON(ret);
3272                 if (inode->i_nlink > 0) {
3273                         ret = btrfs_orphan_del(trans, inode);
3274                         BUG_ON(ret);
3275                 }
3276                 nr = trans->blocks_used;
3277                 btrfs_end_transaction(trans, root);
3278                 btrfs_btree_balance_dirty(root, nr);
3279                 return 0;
3280         }
3281
3282         /*
3283          * We're truncating a file that used to have good data down to
3284          * zero. Make sure it gets into the ordered flush list so that
3285          * any new writes get down to disk quickly.
3286          */
3287         if (attr->ia_size == 0)
3288                 BTRFS_I(inode)->ordered_data_close = 1;
3289
3290         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3291         ret = vmtruncate(inode, attr->ia_size);
3292         BUG_ON(ret);
3293
3294         return 0;
3295 }
3296
3297 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3298 {
3299         struct inode *inode = dentry->d_inode;
3300         int err;
3301
3302         err = inode_change_ok(inode, attr);
3303         if (err)
3304                 return err;
3305
3306         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3307                 err = btrfs_setattr_size(inode, attr);
3308                 if (err)
3309                         return err;
3310         }
3311         attr->ia_valid &= ~ATTR_SIZE;
3312
3313         if (attr->ia_valid)
3314                 err = inode_setattr(inode, attr);
3315
3316         if (!err && ((attr->ia_valid & ATTR_MODE)))
3317                 err = btrfs_acl_chmod(inode);
3318         return err;
3319 }
3320
3321 void btrfs_delete_inode(struct inode *inode)
3322 {
3323         struct btrfs_trans_handle *trans;
3324         struct btrfs_root *root = BTRFS_I(inode)->root;
3325         unsigned long nr;
3326         int ret;
3327
3328         truncate_inode_pages(&inode->i_data, 0);
3329         if (is_bad_inode(inode)) {
3330                 btrfs_orphan_del(NULL, inode);
3331                 goto no_delete;
3332         }
3333         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3334
3335         if (root->fs_info->log_root_recovering) {
3336                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3337                 goto no_delete;
3338         }
3339
3340         if (inode->i_nlink > 0) {
3341                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3342                 goto no_delete;
3343         }
3344
3345         btrfs_i_size_write(inode, 0);
3346
3347         while (1) {
3348                 trans = btrfs_start_transaction(root, 1);
3349                 btrfs_set_trans_block_group(trans, inode);
3350                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3351
3352                 if (ret != -EAGAIN)
3353                         break;
3354
3355                 nr = trans->blocks_used;
3356                 btrfs_end_transaction(trans, root);
3357                 trans = NULL;
3358                 btrfs_btree_balance_dirty(root, nr);
3359         }
3360
3361         if (ret == 0) {
3362                 ret = btrfs_orphan_del(trans, inode);
3363                 BUG_ON(ret);
3364         }
3365
3366         nr = trans->blocks_used;
3367         btrfs_end_transaction(trans, root);
3368         btrfs_btree_balance_dirty(root, nr);
3369 no_delete:
3370         clear_inode(inode);
3371         return;
3372 }
3373
3374 /*
3375  * this returns the key found in the dir entry in the location pointer.
3376  * If no dir entries were found, location->objectid is 0.
3377  */
3378 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3379                                struct btrfs_key *location)
3380 {
3381         const char *name = dentry->d_name.name;
3382         int namelen = dentry->d_name.len;
3383         struct btrfs_dir_item *di;
3384         struct btrfs_path *path;
3385         struct btrfs_root *root = BTRFS_I(dir)->root;
3386         int ret = 0;
3387
3388         path = btrfs_alloc_path();
3389         BUG_ON(!path);
3390
3391         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3392                                     namelen, 0);
3393         if (IS_ERR(di))
3394                 ret = PTR_ERR(di);
3395
3396         if (!di || IS_ERR(di))
3397                 goto out_err;
3398
3399         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3400 out:
3401         btrfs_free_path(path);
3402         return ret;
3403 out_err:
3404         location->objectid = 0;
3405         goto out;
3406 }
3407
3408 /*
3409  * when we hit a tree root in a directory, the btrfs part of the inode
3410  * needs to be changed to reflect the root directory of the tree root.  This
3411  * is kind of like crossing a mount point.
3412  */
3413 static int fixup_tree_root_location(struct btrfs_root *root,
3414                                     struct inode *dir,
3415                                     struct dentry *dentry,
3416                                     struct btrfs_key *location,
3417                                     struct btrfs_root **sub_root)
3418 {
3419         struct btrfs_path *path;
3420         struct btrfs_root *new_root;
3421         struct btrfs_root_ref *ref;
3422         struct extent_buffer *leaf;
3423         int ret;
3424         int err = 0;
3425
3426         path = btrfs_alloc_path();
3427         if (!path) {
3428                 err = -ENOMEM;
3429                 goto out;
3430         }
3431
3432         err = -ENOENT;
3433         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3434                                   BTRFS_I(dir)->root->root_key.objectid,
3435                                   location->objectid);
3436         if (ret) {
3437                 if (ret < 0)
3438                         err = ret;
3439                 goto out;
3440         }
3441
3442         leaf = path->nodes[0];
3443         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3444         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3445             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3446                 goto out;
3447
3448         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3449                                    (unsigned long)(ref + 1),
3450                                    dentry->d_name.len);
3451         if (ret)
3452                 goto out;
3453
3454         btrfs_release_path(root->fs_info->tree_root, path);
3455
3456         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3457         if (IS_ERR(new_root)) {
3458                 err = PTR_ERR(new_root);
3459                 goto out;
3460         }
3461
3462         if (btrfs_root_refs(&new_root->root_item) == 0) {
3463                 err = -ENOENT;
3464                 goto out;
3465         }
3466
3467         *sub_root = new_root;
3468         location->objectid = btrfs_root_dirid(&new_root->root_item);
3469         location->type = BTRFS_INODE_ITEM_KEY;
3470         location->offset = 0;
3471         err = 0;
3472 out:
3473         btrfs_free_path(path);
3474         return err;
3475 }
3476
3477 static void inode_tree_add(struct inode *inode)
3478 {
3479         struct btrfs_root *root = BTRFS_I(inode)->root;
3480         struct btrfs_inode *entry;
3481         struct rb_node **p;
3482         struct rb_node *parent;
3483 again:
3484         p = &root->inode_tree.rb_node;
3485         parent = NULL;
3486
3487         if (hlist_unhashed(&inode->i_hash))
3488                 return;
3489
3490         spin_lock(&root->inode_lock);
3491         while (*p) {
3492                 parent = *p;
3493                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3494
3495                 if (inode->i_ino < entry->vfs_inode.i_ino)
3496                         p = &parent->rb_left;
3497                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3498                         p = &parent->rb_right;
3499                 else {
3500                         WARN_ON(!(entry->vfs_inode.i_state &
3501                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3502                         rb_erase(parent, &root->inode_tree);
3503                         RB_CLEAR_NODE(parent);
3504                         spin_unlock(&root->inode_lock);
3505                         goto again;
3506                 }
3507         }
3508         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3509         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3510         spin_unlock(&root->inode_lock);
3511 }
3512
3513 static void inode_tree_del(struct inode *inode)
3514 {
3515         struct btrfs_root *root = BTRFS_I(inode)->root;
3516         int empty = 0;
3517
3518         spin_lock(&root->inode_lock);
3519         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3520                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3521                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3522                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3523         }
3524         spin_unlock(&root->inode_lock);
3525
3526         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3527                 synchronize_srcu(&root->fs_info->subvol_srcu);
3528                 spin_lock(&root->inode_lock);
3529                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3530                 spin_unlock(&root->inode_lock);
3531                 if (empty)
3532                         btrfs_add_dead_root(root);
3533         }
3534 }
3535
3536 int btrfs_invalidate_inodes(struct btrfs_root *root)
3537 {
3538         struct rb_node *node;
3539         struct rb_node *prev;
3540         struct btrfs_inode *entry;
3541         struct inode *inode;
3542         u64 objectid = 0;
3543
3544         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3545
3546         spin_lock(&root->inode_lock);
3547 again:
3548         node = root->inode_tree.rb_node;
3549         prev = NULL;
3550         while (node) {
3551                 prev = node;
3552                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3553
3554                 if (objectid < entry->vfs_inode.i_ino)
3555                         node = node->rb_left;
3556                 else if (objectid > entry->vfs_inode.i_ino)
3557                         node = node->rb_right;
3558                 else
3559                         break;
3560         }
3561         if (!node) {
3562                 while (prev) {
3563                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3564                         if (objectid <= entry->vfs_inode.i_ino) {
3565                                 node = prev;
3566                                 break;
3567                         }
3568                         prev = rb_next(prev);
3569                 }
3570         }
3571         while (node) {
3572                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3573                 objectid = entry->vfs_inode.i_ino + 1;
3574                 inode = igrab(&entry->vfs_inode);
3575                 if (inode) {
3576                         spin_unlock(&root->inode_lock);
3577                         if (atomic_read(&inode->i_count) > 1)
3578                                 d_prune_aliases(inode);
3579                         /*
3580                          * btrfs_drop_inode will remove it from
3581                          * the inode cache when its usage count
3582                          * hits zero.
3583                          */
3584                         iput(inode);
3585                         cond_resched();
3586                         spin_lock(&root->inode_lock);
3587                         goto again;
3588                 }
3589
3590                 if (cond_resched_lock(&root->inode_lock))
3591                         goto again;
3592
3593                 node = rb_next(node);
3594         }
3595         spin_unlock(&root->inode_lock);
3596         return 0;
3597 }
3598
3599 static noinline void init_btrfs_i(struct inode *inode)
3600 {
3601         struct btrfs_inode *bi = BTRFS_I(inode);
3602
3603         bi->generation = 0;
3604         bi->sequence = 0;
3605         bi->last_trans = 0;
3606         bi->last_sub_trans = 0;
3607         bi->logged_trans = 0;
3608         bi->delalloc_bytes = 0;
3609         bi->reserved_bytes = 0;
3610         bi->disk_i_size = 0;
3611         bi->flags = 0;
3612         bi->index_cnt = (u64)-1;
3613         bi->last_unlink_trans = 0;
3614         bi->ordered_data_close = 0;
3615         bi->force_compress = 0;
3616         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3617         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3618                              inode->i_mapping, GFP_NOFS);
3619         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3620                              inode->i_mapping, GFP_NOFS);
3621         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3622         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3623         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3624         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3625         mutex_init(&BTRFS_I(inode)->log_mutex);
3626 }
3627
3628 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3629 {
3630         struct btrfs_iget_args *args = p;
3631         inode->i_ino = args->ino;
3632         init_btrfs_i(inode);
3633         BTRFS_I(inode)->root = args->root;
3634         btrfs_set_inode_space_info(args->root, inode);
3635         return 0;
3636 }
3637
3638 static int btrfs_find_actor(struct inode *inode, void *opaque)
3639 {
3640         struct btrfs_iget_args *args = opaque;
3641         return args->ino == inode->i_ino &&
3642                 args->root == BTRFS_I(inode)->root;
3643 }
3644
3645 static struct inode *btrfs_iget_locked(struct super_block *s,
3646                                        u64 objectid,
3647                                        struct btrfs_root *root)
3648 {
3649         struct inode *inode;
3650         struct btrfs_iget_args args;
3651         args.ino = objectid;
3652         args.root = root;
3653
3654         inode = iget5_locked(s, objectid, btrfs_find_actor,
3655                              btrfs_init_locked_inode,
3656                              (void *)&args);
3657         return inode;
3658 }
3659
3660 /* Get an inode object given its location and corresponding root.
3661  * Returns in *is_new if the inode was read from disk
3662  */
3663 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3664                          struct btrfs_root *root, int *new)
3665 {
3666         struct inode *inode;
3667
3668         inode = btrfs_iget_locked(s, location->objectid, root);
3669         if (!inode)
3670                 return ERR_PTR(-ENOMEM);
3671
3672         if (inode->i_state & I_NEW) {
3673                 BTRFS_I(inode)->root = root;
3674                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3675                 btrfs_read_locked_inode(inode);
3676
3677                 inode_tree_add(inode);
3678                 unlock_new_inode(inode);
3679                 if (new)
3680                         *new = 1;
3681         }
3682
3683         return inode;
3684 }
3685
3686 static struct inode *new_simple_dir(struct super_block *s,
3687                                     struct btrfs_key *key,
3688                                     struct btrfs_root *root)
3689 {
3690         struct inode *inode = new_inode(s);
3691
3692         if (!inode)
3693                 return ERR_PTR(-ENOMEM);
3694
3695         init_btrfs_i(inode);
3696
3697         BTRFS_I(inode)->root = root;
3698         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3699         BTRFS_I(inode)->dummy_inode = 1;
3700
3701         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3702         inode->i_op = &simple_dir_inode_operations;
3703         inode->i_fop = &simple_dir_operations;
3704         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3705         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3706
3707         return inode;
3708 }
3709
3710 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3711 {
3712         struct inode *inode;
3713         struct btrfs_root *root = BTRFS_I(dir)->root;
3714         struct btrfs_root *sub_root = root;
3715         struct btrfs_key location;
3716         int index;
3717         int ret;
3718
3719         dentry->d_op = &btrfs_dentry_operations;
3720
3721         if (dentry->d_name.len > BTRFS_NAME_LEN)
3722                 return ERR_PTR(-ENAMETOOLONG);
3723
3724         ret = btrfs_inode_by_name(dir, dentry, &location);
3725
3726         if (ret < 0)
3727                 return ERR_PTR(ret);
3728
3729         if (location.objectid == 0)
3730                 return NULL;
3731
3732         if (location.type == BTRFS_INODE_ITEM_KEY) {
3733                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3734                 return inode;
3735         }
3736
3737         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3738
3739         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3740         ret = fixup_tree_root_location(root, dir, dentry,
3741                                        &location, &sub_root);
3742         if (ret < 0) {
3743                 if (ret != -ENOENT)
3744                         inode = ERR_PTR(ret);
3745                 else
3746                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3747         } else {
3748                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3749         }
3750         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3751
3752         if (root != sub_root) {
3753                 down_read(&root->fs_info->cleanup_work_sem);
3754                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3755                         btrfs_orphan_cleanup(sub_root);
3756                 up_read(&root->fs_info->cleanup_work_sem);
3757         }
3758
3759         return inode;
3760 }
3761
3762 static int btrfs_dentry_delete(struct dentry *dentry)
3763 {
3764         struct btrfs_root *root;
3765
3766         if (!dentry->d_inode && !IS_ROOT(dentry))
3767                 dentry = dentry->d_parent;
3768
3769         if (dentry->d_inode) {
3770                 root = BTRFS_I(dentry->d_inode)->root;
3771                 if (btrfs_root_refs(&root->root_item) == 0)
3772                         return 1;
3773         }
3774         return 0;
3775 }
3776
3777 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3778                                    struct nameidata *nd)
3779 {
3780         struct inode *inode;
3781
3782         inode = btrfs_lookup_dentry(dir, dentry);
3783         if (IS_ERR(inode))
3784                 return ERR_CAST(inode);
3785
3786         return d_splice_alias(inode, dentry);
3787 }
3788
3789 static unsigned char btrfs_filetype_table[] = {
3790         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3791 };
3792
3793 static int btrfs_real_readdir(struct file *filp, void *dirent,
3794                               filldir_t filldir)
3795 {
3796         struct inode *inode = filp->f_dentry->d_inode;
3797         struct btrfs_root *root = BTRFS_I(inode)->root;
3798         struct btrfs_item *item;
3799         struct btrfs_dir_item *di;
3800         struct btrfs_key key;
3801         struct btrfs_key found_key;
3802         struct btrfs_path *path;
3803         int ret;
3804         u32 nritems;
3805         struct extent_buffer *leaf;
3806         int slot;
3807         int advance;
3808         unsigned char d_type;
3809         int over = 0;
3810         u32 di_cur;
3811         u32 di_total;
3812         u32 di_len;
3813         int key_type = BTRFS_DIR_INDEX_KEY;
3814         char tmp_name[32];
3815         char *name_ptr;
3816         int name_len;
3817
3818         /* FIXME, use a real flag for deciding about the key type */
3819         if (root->fs_info->tree_root == root)
3820                 key_type = BTRFS_DIR_ITEM_KEY;
3821
3822         /* special case for "." */
3823         if (filp->f_pos == 0) {
3824                 over = filldir(dirent, ".", 1,
3825                                1, inode->i_ino,
3826                                DT_DIR);
3827                 if (over)
3828                         return 0;
3829                 filp->f_pos = 1;
3830         }
3831         /* special case for .., just use the back ref */
3832         if (filp->f_pos == 1) {
3833                 u64 pino = parent_ino(filp->f_path.dentry);
3834                 over = filldir(dirent, "..", 2,
3835                                2, pino, DT_DIR);
3836                 if (over)
3837                         return 0;
3838                 filp->f_pos = 2;
3839         }
3840         path = btrfs_alloc_path();
3841         path->reada = 2;
3842
3843         btrfs_set_key_type(&key, key_type);
3844         key.offset = filp->f_pos;
3845         key.objectid = inode->i_ino;
3846
3847         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3848         if (ret < 0)
3849                 goto err;
3850         advance = 0;
3851
3852         while (1) {
3853                 leaf = path->nodes[0];
3854                 nritems = btrfs_header_nritems(leaf);
3855                 slot = path->slots[0];
3856                 if (advance || slot >= nritems) {
3857                         if (slot >= nritems - 1) {
3858                                 ret = btrfs_next_leaf(root, path);
3859                                 if (ret)
3860                                         break;
3861                                 leaf = path->nodes[0];
3862                                 nritems = btrfs_header_nritems(leaf);
3863                                 slot = path->slots[0];
3864                         } else {
3865                                 slot++;
3866                                 path->slots[0]++;
3867                         }
3868                 }
3869
3870                 advance = 1;
3871                 item = btrfs_item_nr(leaf, slot);
3872                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3873
3874                 if (found_key.objectid != key.objectid)
3875                         break;
3876                 if (btrfs_key_type(&found_key) != key_type)
3877                         break;
3878                 if (found_key.offset < filp->f_pos)
3879                         continue;
3880
3881                 filp->f_pos = found_key.offset;
3882
3883                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3884                 di_cur = 0;
3885                 di_total = btrfs_item_size(leaf, item);
3886
3887                 while (di_cur < di_total) {
3888                         struct btrfs_key location;
3889
3890                         name_len = btrfs_dir_name_len(leaf, di);
3891                         if (name_len <= sizeof(tmp_name)) {
3892                                 name_ptr = tmp_name;
3893                         } else {
3894                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3895                                 if (!name_ptr) {
3896                                         ret = -ENOMEM;
3897                                         goto err;
3898                                 }
3899                         }
3900                         read_extent_buffer(leaf, name_ptr,
3901                                            (unsigned long)(di + 1), name_len);
3902
3903                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3904                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3905
3906                         /* is this a reference to our own snapshot? If so
3907                          * skip it
3908                          */
3909                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3910                             location.objectid == root->root_key.objectid) {
3911                                 over = 0;
3912                                 goto skip;
3913                         }
3914                         over = filldir(dirent, name_ptr, name_len,
3915                                        found_key.offset, location.objectid,
3916                                        d_type);
3917
3918 skip:
3919                         if (name_ptr != tmp_name)
3920                                 kfree(name_ptr);
3921
3922                         if (over)
3923                                 goto nopos;
3924                         di_len = btrfs_dir_name_len(leaf, di) +
3925                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3926                         di_cur += di_len;
3927                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3928                 }
3929         }
3930
3931         /* Reached end of directory/root. Bump pos past the last item. */
3932         if (key_type == BTRFS_DIR_INDEX_KEY)
3933                 /*
3934                  * 32-bit glibc will use getdents64, but then strtol -
3935                  * so the last number we can serve is this.
3936                  */
3937                 filp->f_pos = 0x7fffffff;
3938         else
3939                 filp->f_pos++;
3940 nopos:
3941         ret = 0;
3942 err:
3943         btrfs_free_path(path);
3944         return ret;
3945 }
3946
3947 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
3948 {
3949         struct btrfs_root *root = BTRFS_I(inode)->root;
3950         struct btrfs_trans_handle *trans;
3951         int ret = 0;
3952
3953         if (root->fs_info->btree_inode == inode)
3954                 return 0;
3955
3956         if (wbc->sync_mode == WB_SYNC_ALL) {
3957                 trans = btrfs_join_transaction(root, 1);
3958                 btrfs_set_trans_block_group(trans, inode);
3959                 ret = btrfs_commit_transaction(trans, root);
3960         }
3961         return ret;
3962 }
3963
3964 /*
3965  * This is somewhat expensive, updating the tree every time the
3966  * inode changes.  But, it is most likely to find the inode in cache.
3967  * FIXME, needs more benchmarking...there are no reasons other than performance
3968  * to keep or drop this code.
3969  */
3970 void btrfs_dirty_inode(struct inode *inode)
3971 {
3972         struct btrfs_root *root = BTRFS_I(inode)->root;
3973         struct btrfs_trans_handle *trans;
3974
3975         trans = btrfs_join_transaction(root, 1);
3976         btrfs_set_trans_block_group(trans, inode);
3977         btrfs_update_inode(trans, root, inode);
3978         btrfs_end_transaction(trans, root);
3979 }
3980
3981 /*
3982  * find the highest existing sequence number in a directory
3983  * and then set the in-memory index_cnt variable to reflect
3984  * free sequence numbers
3985  */
3986 static int btrfs_set_inode_index_count(struct inode *inode)
3987 {
3988         struct btrfs_root *root = BTRFS_I(inode)->root;
3989         struct btrfs_key key, found_key;
3990         struct btrfs_path *path;
3991         struct extent_buffer *leaf;
3992         int ret;
3993
3994         key.objectid = inode->i_ino;
3995         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3996         key.offset = (u64)-1;
3997
3998         path = btrfs_alloc_path();
3999         if (!path)
4000                 return -ENOMEM;
4001
4002         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4003         if (ret < 0)
4004                 goto out;
4005         /* FIXME: we should be able to handle this */
4006         if (ret == 0)
4007                 goto out;
4008         ret = 0;
4009
4010         /*
4011          * MAGIC NUMBER EXPLANATION:
4012          * since we search a directory based on f_pos we have to start at 2
4013          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4014          * else has to start at 2
4015          */
4016         if (path->slots[0] == 0) {
4017                 BTRFS_I(inode)->index_cnt = 2;
4018                 goto out;
4019         }
4020
4021         path->slots[0]--;
4022
4023         leaf = path->nodes[0];
4024         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4025
4026         if (found_key.objectid != inode->i_ino ||
4027             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4028                 BTRFS_I(inode)->index_cnt = 2;
4029                 goto out;
4030         }
4031
4032         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4033 out:
4034         btrfs_free_path(path);
4035         return ret;
4036 }
4037
4038 /*
4039  * helper to find a free sequence number in a given directory.  This current
4040  * code is very simple, later versions will do smarter things in the btree
4041  */
4042 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4043 {
4044         int ret = 0;
4045
4046         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4047                 ret = btrfs_set_inode_index_count(dir);
4048                 if (ret)
4049                         return ret;
4050         }
4051
4052         *index = BTRFS_I(dir)->index_cnt;
4053         BTRFS_I(dir)->index_cnt++;
4054
4055         return ret;
4056 }
4057
4058 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4059                                      struct btrfs_root *root,
4060                                      struct inode *dir,
4061                                      const char *name, int name_len,
4062                                      u64 ref_objectid, u64 objectid,
4063                                      u64 alloc_hint, int mode, u64 *index)
4064 {
4065         struct inode *inode;
4066         struct btrfs_inode_item *inode_item;
4067         struct btrfs_key *location;
4068         struct btrfs_path *path;
4069         struct btrfs_inode_ref *ref;
4070         struct btrfs_key key[2];
4071         u32 sizes[2];
4072         unsigned long ptr;
4073         int ret;
4074         int owner;
4075
4076         path = btrfs_alloc_path();
4077         BUG_ON(!path);
4078
4079         inode = new_inode(root->fs_info->sb);
4080         if (!inode)
4081                 return ERR_PTR(-ENOMEM);
4082
4083         if (dir) {
4084                 ret = btrfs_set_inode_index(dir, index);
4085                 if (ret) {
4086                         iput(inode);
4087                         return ERR_PTR(ret);
4088                 }
4089         }
4090         /*
4091          * index_cnt is ignored for everything but a dir,
4092          * btrfs_get_inode_index_count has an explanation for the magic
4093          * number
4094          */
4095         init_btrfs_i(inode);
4096         BTRFS_I(inode)->index_cnt = 2;
4097         BTRFS_I(inode)->root = root;
4098         BTRFS_I(inode)->generation = trans->transid;
4099         btrfs_set_inode_space_info(root, inode);
4100
4101         if (mode & S_IFDIR)
4102                 owner = 0;
4103         else
4104                 owner = 1;
4105         BTRFS_I(inode)->block_group =
4106                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4107
4108         key[0].objectid = objectid;
4109         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4110         key[0].offset = 0;
4111
4112         key[1].objectid = objectid;
4113         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4114         key[1].offset = ref_objectid;
4115
4116         sizes[0] = sizeof(struct btrfs_inode_item);
4117         sizes[1] = name_len + sizeof(*ref);
4118
4119         path->leave_spinning = 1;
4120         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4121         if (ret != 0)
4122                 goto fail;
4123
4124         inode->i_uid = current_fsuid();
4125
4126         if (dir && (dir->i_mode & S_ISGID)) {
4127                 inode->i_gid = dir->i_gid;
4128                 if (S_ISDIR(mode))
4129                         mode |= S_ISGID;
4130         } else
4131                 inode->i_gid = current_fsgid();
4132
4133         inode->i_mode = mode;
4134         inode->i_ino = objectid;
4135         inode_set_bytes(inode, 0);
4136         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4137         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4138                                   struct btrfs_inode_item);
4139         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4140
4141         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4142                              struct btrfs_inode_ref);
4143         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4144         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4145         ptr = (unsigned long)(ref + 1);
4146         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4147
4148         btrfs_mark_buffer_dirty(path->nodes[0]);
4149         btrfs_free_path(path);
4150
4151         location = &BTRFS_I(inode)->location;
4152         location->objectid = objectid;
4153         location->offset = 0;
4154         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4155
4156         btrfs_inherit_iflags(inode, dir);
4157
4158         if ((mode & S_IFREG)) {
4159                 if (btrfs_test_opt(root, NODATASUM))
4160                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4161                 if (btrfs_test_opt(root, NODATACOW))
4162                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4163         }
4164
4165         insert_inode_hash(inode);
4166         inode_tree_add(inode);
4167         return inode;
4168 fail:
4169         if (dir)
4170                 BTRFS_I(dir)->index_cnt--;
4171         btrfs_free_path(path);
4172         iput(inode);
4173         return ERR_PTR(ret);
4174 }
4175
4176 static inline u8 btrfs_inode_type(struct inode *inode)
4177 {
4178         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4179 }
4180
4181 /*
4182  * utility function to add 'inode' into 'parent_inode' with
4183  * a give name and a given sequence number.
4184  * if 'add_backref' is true, also insert a backref from the
4185  * inode to the parent directory.
4186  */
4187 int btrfs_add_link(struct btrfs_trans_handle *trans,
4188                    struct inode *parent_inode, struct inode *inode,
4189                    const char *name, int name_len, int add_backref, u64 index)
4190 {
4191         int ret = 0;
4192         struct btrfs_key key;
4193         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4194
4195         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4196                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4197         } else {
4198                 key.objectid = inode->i_ino;
4199                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4200                 key.offset = 0;
4201         }
4202
4203         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4204                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4205                                          key.objectid, root->root_key.objectid,
4206                                          parent_inode->i_ino,
4207                                          index, name, name_len);
4208         } else if (add_backref) {
4209                 ret = btrfs_insert_inode_ref(trans, root,
4210                                              name, name_len, inode->i_ino,
4211                                              parent_inode->i_ino, index);
4212         }
4213
4214         if (ret == 0) {
4215                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4216                                             parent_inode->i_ino, &key,
4217                                             btrfs_inode_type(inode), index);
4218                 BUG_ON(ret);
4219
4220                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4221                                    name_len * 2);
4222                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4223                 ret = btrfs_update_inode(trans, root, parent_inode);
4224         }
4225         return ret;
4226 }
4227
4228 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4229                             struct dentry *dentry, struct inode *inode,
4230                             int backref, u64 index)
4231 {
4232         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4233                                  inode, dentry->d_name.name,
4234                                  dentry->d_name.len, backref, index);
4235         if (!err) {
4236                 d_instantiate(dentry, inode);
4237                 return 0;
4238         }
4239         if (err > 0)
4240                 err = -EEXIST;
4241         return err;
4242 }
4243
4244 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4245                         int mode, dev_t rdev)
4246 {
4247         struct btrfs_trans_handle *trans;
4248         struct btrfs_root *root = BTRFS_I(dir)->root;
4249         struct inode *inode = NULL;
4250         int err;
4251         int drop_inode = 0;
4252         u64 objectid;
4253         unsigned long nr = 0;
4254         u64 index = 0;
4255
4256         if (!new_valid_dev(rdev))
4257                 return -EINVAL;
4258
4259         /*
4260          * 2 for inode item and ref
4261          * 2 for dir items
4262          * 1 for xattr if selinux is on
4263          */
4264         err = btrfs_reserve_metadata_space(root, 5);
4265         if (err)
4266                 return err;
4267
4268         trans = btrfs_start_transaction(root, 1);
4269         if (!trans)
4270                 goto fail;
4271         btrfs_set_trans_block_group(trans, dir);
4272
4273         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4274         if (err) {
4275                 err = -ENOSPC;
4276                 goto out_unlock;
4277         }
4278
4279         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4280                                 dentry->d_name.len,
4281                                 dentry->d_parent->d_inode->i_ino, objectid,
4282                                 BTRFS_I(dir)->block_group, mode, &index);
4283         err = PTR_ERR(inode);
4284         if (IS_ERR(inode))
4285                 goto out_unlock;
4286
4287         err = btrfs_init_inode_security(trans, inode, dir);
4288         if (err) {
4289                 drop_inode = 1;
4290                 goto out_unlock;
4291         }
4292
4293         btrfs_set_trans_block_group(trans, inode);
4294         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4295         if (err)
4296                 drop_inode = 1;
4297         else {
4298                 inode->i_op = &btrfs_special_inode_operations;
4299                 init_special_inode(inode, inode->i_mode, rdev);
4300                 btrfs_update_inode(trans, root, inode);
4301         }
4302         btrfs_update_inode_block_group(trans, inode);
4303         btrfs_update_inode_block_group(trans, dir);
4304 out_unlock:
4305         nr = trans->blocks_used;
4306         btrfs_end_transaction_throttle(trans, root);
4307 fail:
4308         btrfs_unreserve_metadata_space(root, 5);
4309         if (drop_inode) {
4310                 inode_dec_link_count(inode);
4311                 iput(inode);
4312         }
4313         btrfs_btree_balance_dirty(root, nr);
4314         return err;
4315 }
4316
4317 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4318                         int mode, struct nameidata *nd)
4319 {
4320         struct btrfs_trans_handle *trans;
4321         struct btrfs_root *root = BTRFS_I(dir)->root;
4322         struct inode *inode = NULL;
4323         int err;
4324         int drop_inode = 0;
4325         unsigned long nr = 0;
4326         u64 objectid;
4327         u64 index = 0;
4328
4329         /*
4330          * 2 for inode item and ref
4331          * 2 for dir items
4332          * 1 for xattr if selinux is on
4333          */
4334         err = btrfs_reserve_metadata_space(root, 5);
4335         if (err)
4336                 return err;
4337
4338         trans = btrfs_start_transaction(root, 1);
4339         if (!trans)
4340                 goto fail;
4341         btrfs_set_trans_block_group(trans, dir);
4342
4343         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4344         if (err) {
4345                 err = -ENOSPC;
4346                 goto out_unlock;
4347         }
4348
4349         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4350                                 dentry->d_name.len,
4351                                 dentry->d_parent->d_inode->i_ino,
4352                                 objectid, BTRFS_I(dir)->block_group, mode,
4353                                 &index);
4354         err = PTR_ERR(inode);
4355         if (IS_ERR(inode))
4356                 goto out_unlock;
4357
4358         err = btrfs_init_inode_security(trans, inode, dir);
4359         if (err) {
4360                 drop_inode = 1;
4361                 goto out_unlock;
4362         }
4363
4364         btrfs_set_trans_block_group(trans, inode);
4365         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4366         if (err)
4367                 drop_inode = 1;
4368         else {
4369                 inode->i_mapping->a_ops = &btrfs_aops;
4370                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4371                 inode->i_fop = &btrfs_file_operations;
4372                 inode->i_op = &btrfs_file_inode_operations;
4373                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4374         }
4375         btrfs_update_inode_block_group(trans, inode);
4376         btrfs_update_inode_block_group(trans, dir);
4377 out_unlock:
4378         nr = trans->blocks_used;
4379         btrfs_end_transaction_throttle(trans, root);
4380 fail:
4381         btrfs_unreserve_metadata_space(root, 5);
4382         if (drop_inode) {
4383                 inode_dec_link_count(inode);
4384                 iput(inode);
4385         }
4386         btrfs_btree_balance_dirty(root, nr);
4387         return err;
4388 }
4389
4390 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4391                       struct dentry *dentry)
4392 {
4393         struct btrfs_trans_handle *trans;
4394         struct btrfs_root *root = BTRFS_I(dir)->root;
4395         struct inode *inode = old_dentry->d_inode;
4396         u64 index;
4397         unsigned long nr = 0;
4398         int err;
4399         int drop_inode = 0;
4400
4401         if (inode->i_nlink == 0)
4402                 return -ENOENT;
4403
4404         /* do not allow sys_link's with other subvols of the same device */
4405         if (root->objectid != BTRFS_I(inode)->root->objectid)
4406                 return -EPERM;
4407
4408         /*
4409          * 1 item for inode ref
4410          * 2 items for dir items
4411          */
4412         err = btrfs_reserve_metadata_space(root, 3);
4413         if (err)
4414                 return err;
4415
4416         btrfs_inc_nlink(inode);
4417
4418         err = btrfs_set_inode_index(dir, &index);
4419         if (err)
4420                 goto fail;
4421
4422         trans = btrfs_start_transaction(root, 1);
4423
4424         btrfs_set_trans_block_group(trans, dir);
4425         atomic_inc(&inode->i_count);
4426
4427         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4428
4429         if (err) {
4430                 drop_inode = 1;
4431         } else {
4432                 btrfs_update_inode_block_group(trans, dir);
4433                 err = btrfs_update_inode(trans, root, inode);
4434                 BUG_ON(err);
4435                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4436         }
4437
4438         nr = trans->blocks_used;
4439         btrfs_end_transaction_throttle(trans, root);
4440 fail:
4441         btrfs_unreserve_metadata_space(root, 3);
4442         if (drop_inode) {
4443                 inode_dec_link_count(inode);
4444                 iput(inode);
4445         }
4446         btrfs_btree_balance_dirty(root, nr);
4447         return err;
4448 }
4449
4450 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4451 {
4452         struct inode *inode = NULL;
4453         struct btrfs_trans_handle *trans;
4454         struct btrfs_root *root = BTRFS_I(dir)->root;
4455         int err = 0;
4456         int drop_on_err = 0;
4457         u64 objectid = 0;
4458         u64 index = 0;
4459         unsigned long nr = 1;
4460
4461         /*
4462          * 2 items for inode and ref
4463          * 2 items for dir items
4464          * 1 for xattr if selinux is on
4465          */
4466         err = btrfs_reserve_metadata_space(root, 5);
4467         if (err)
4468                 return err;
4469
4470         trans = btrfs_start_transaction(root, 1);
4471         if (!trans) {
4472                 err = -ENOMEM;
4473                 goto out_unlock;
4474         }
4475         btrfs_set_trans_block_group(trans, dir);
4476
4477         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4478         if (err) {
4479                 err = -ENOSPC;
4480                 goto out_fail;
4481         }
4482
4483         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4484                                 dentry->d_name.len,
4485                                 dentry->d_parent->d_inode->i_ino, objectid,
4486                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4487                                 &index);
4488         if (IS_ERR(inode)) {
4489                 err = PTR_ERR(inode);
4490                 goto out_fail;
4491         }
4492
4493         drop_on_err = 1;
4494
4495         err = btrfs_init_inode_security(trans, inode, dir);
4496         if (err)
4497                 goto out_fail;
4498
4499         inode->i_op = &btrfs_dir_inode_operations;
4500         inode->i_fop = &btrfs_dir_file_operations;
4501         btrfs_set_trans_block_group(trans, inode);
4502
4503         btrfs_i_size_write(inode, 0);
4504         err = btrfs_update_inode(trans, root, inode);
4505         if (err)
4506                 goto out_fail;
4507
4508         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4509                                  inode, dentry->d_name.name,
4510                                  dentry->d_name.len, 0, index);
4511         if (err)
4512                 goto out_fail;
4513
4514         d_instantiate(dentry, inode);
4515         drop_on_err = 0;
4516         btrfs_update_inode_block_group(trans, inode);
4517         btrfs_update_inode_block_group(trans, dir);
4518
4519 out_fail:
4520         nr = trans->blocks_used;
4521         btrfs_end_transaction_throttle(trans, root);
4522
4523 out_unlock:
4524         btrfs_unreserve_metadata_space(root, 5);
4525         if (drop_on_err)
4526                 iput(inode);
4527         btrfs_btree_balance_dirty(root, nr);
4528         return err;
4529 }
4530
4531 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4532  * and an extent that you want to insert, deal with overlap and insert
4533  * the new extent into the tree.
4534  */
4535 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4536                                 struct extent_map *existing,
4537                                 struct extent_map *em,
4538                                 u64 map_start, u64 map_len)
4539 {
4540         u64 start_diff;
4541
4542         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4543         start_diff = map_start - em->start;
4544         em->start = map_start;
4545         em->len = map_len;
4546         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4547             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4548                 em->block_start += start_diff;
4549                 em->block_len -= start_diff;
4550         }
4551         return add_extent_mapping(em_tree, em);
4552 }
4553
4554 static noinline int uncompress_inline(struct btrfs_path *path,
4555                                       struct inode *inode, struct page *page,
4556                                       size_t pg_offset, u64 extent_offset,
4557                                       struct btrfs_file_extent_item *item)
4558 {
4559         int ret;
4560         struct extent_buffer *leaf = path->nodes[0];
4561         char *tmp;
4562         size_t max_size;
4563         unsigned long inline_size;
4564         unsigned long ptr;
4565
4566         WARN_ON(pg_offset != 0);
4567         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4568         inline_size = btrfs_file_extent_inline_item_len(leaf,
4569                                         btrfs_item_nr(leaf, path->slots[0]));
4570         tmp = kmalloc(inline_size, GFP_NOFS);
4571         ptr = btrfs_file_extent_inline_start(item);
4572
4573         read_extent_buffer(leaf, tmp, ptr, inline_size);
4574
4575         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4576         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4577                                     inline_size, max_size);
4578         if (ret) {
4579                 char *kaddr = kmap_atomic(page, KM_USER0);
4580                 unsigned long copy_size = min_t(u64,
4581                                   PAGE_CACHE_SIZE - pg_offset,
4582                                   max_size - extent_offset);
4583                 memset(kaddr + pg_offset, 0, copy_size);
4584                 kunmap_atomic(kaddr, KM_USER0);
4585         }
4586         kfree(tmp);
4587         return 0;
4588 }
4589
4590 /*
4591  * a bit scary, this does extent mapping from logical file offset to the disk.
4592  * the ugly parts come from merging extents from the disk with the in-ram
4593  * representation.  This gets more complex because of the data=ordered code,
4594  * where the in-ram extents might be locked pending data=ordered completion.
4595  *
4596  * This also copies inline extents directly into the page.
4597  */
4598
4599 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4600                                     size_t pg_offset, u64 start, u64 len,
4601                                     int create)
4602 {
4603         int ret;
4604         int err = 0;
4605         u64 bytenr;
4606         u64 extent_start = 0;
4607         u64 extent_end = 0;
4608         u64 objectid = inode->i_ino;
4609         u32 found_type;
4610         struct btrfs_path *path = NULL;
4611         struct btrfs_root *root = BTRFS_I(inode)->root;
4612         struct btrfs_file_extent_item *item;
4613         struct extent_buffer *leaf;
4614         struct btrfs_key found_key;
4615         struct extent_map *em = NULL;
4616         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4617         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4618         struct btrfs_trans_handle *trans = NULL;
4619         int compressed;
4620
4621 again:
4622         read_lock(&em_tree->lock);
4623         em = lookup_extent_mapping(em_tree, start, len);
4624         if (em)
4625                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4626         read_unlock(&em_tree->lock);
4627
4628         if (em) {
4629                 if (em->start > start || em->start + em->len <= start)
4630                         free_extent_map(em);
4631                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4632                         free_extent_map(em);
4633                 else
4634                         goto out;
4635         }
4636         em = alloc_extent_map(GFP_NOFS);
4637         if (!em) {
4638                 err = -ENOMEM;
4639                 goto out;
4640         }
4641         em->bdev = root->fs_info->fs_devices->latest_bdev;
4642         em->start = EXTENT_MAP_HOLE;
4643         em->orig_start = EXTENT_MAP_HOLE;
4644         em->len = (u64)-1;
4645         em->block_len = (u64)-1;
4646
4647         if (!path) {
4648                 path = btrfs_alloc_path();
4649                 BUG_ON(!path);
4650         }
4651
4652         ret = btrfs_lookup_file_extent(trans, root, path,
4653                                        objectid, start, trans != NULL);
4654         if (ret < 0) {
4655                 err = ret;
4656                 goto out;
4657         }
4658
4659         if (ret != 0) {
4660                 if (path->slots[0] == 0)
4661                         goto not_found;
4662                 path->slots[0]--;
4663         }
4664
4665         leaf = path->nodes[0];
4666         item = btrfs_item_ptr(leaf, path->slots[0],
4667                               struct btrfs_file_extent_item);
4668         /* are we inside the extent that was found? */
4669         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4670         found_type = btrfs_key_type(&found_key);
4671         if (found_key.objectid != objectid ||
4672             found_type != BTRFS_EXTENT_DATA_KEY) {
4673                 goto not_found;
4674         }
4675
4676         found_type = btrfs_file_extent_type(leaf, item);
4677         extent_start = found_key.offset;
4678         compressed = btrfs_file_extent_compression(leaf, item);
4679         if (found_type == BTRFS_FILE_EXTENT_REG ||
4680             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4681                 extent_end = extent_start +
4682                        btrfs_file_extent_num_bytes(leaf, item);
4683         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4684                 size_t size;
4685                 size = btrfs_file_extent_inline_len(leaf, item);
4686                 extent_end = (extent_start + size + root->sectorsize - 1) &
4687                         ~((u64)root->sectorsize - 1);
4688         }
4689
4690         if (start >= extent_end) {
4691                 path->slots[0]++;
4692                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4693                         ret = btrfs_next_leaf(root, path);
4694                         if (ret < 0) {
4695                                 err = ret;
4696                                 goto out;
4697                         }
4698                         if (ret > 0)
4699                                 goto not_found;
4700                         leaf = path->nodes[0];
4701                 }
4702                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4703                 if (found_key.objectid != objectid ||
4704                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4705                         goto not_found;
4706                 if (start + len <= found_key.offset)
4707                         goto not_found;
4708                 em->start = start;
4709                 em->len = found_key.offset - start;
4710                 goto not_found_em;
4711         }
4712
4713         if (found_type == BTRFS_FILE_EXTENT_REG ||
4714             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4715                 em->start = extent_start;
4716                 em->len = extent_end - extent_start;
4717                 em->orig_start = extent_start -
4718                                  btrfs_file_extent_offset(leaf, item);
4719                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4720                 if (bytenr == 0) {
4721                         em->block_start = EXTENT_MAP_HOLE;
4722                         goto insert;
4723                 }
4724                 if (compressed) {
4725                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4726                         em->block_start = bytenr;
4727                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4728                                                                          item);
4729                 } else {
4730                         bytenr += btrfs_file_extent_offset(leaf, item);
4731                         em->block_start = bytenr;
4732                         em->block_len = em->len;
4733                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4734                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4735                 }
4736                 goto insert;
4737         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4738                 unsigned long ptr;
4739                 char *map;
4740                 size_t size;
4741                 size_t extent_offset;
4742                 size_t copy_size;
4743
4744                 em->block_start = EXTENT_MAP_INLINE;
4745                 if (!page || create) {
4746                         em->start = extent_start;
4747                         em->len = extent_end - extent_start;
4748                         goto out;
4749                 }
4750
4751                 size = btrfs_file_extent_inline_len(leaf, item);
4752                 extent_offset = page_offset(page) + pg_offset - extent_start;
4753                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4754                                 size - extent_offset);
4755                 em->start = extent_start + extent_offset;
4756                 em->len = (copy_size + root->sectorsize - 1) &
4757                         ~((u64)root->sectorsize - 1);
4758                 em->orig_start = EXTENT_MAP_INLINE;
4759                 if (compressed)
4760                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4761                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4762                 if (create == 0 && !PageUptodate(page)) {
4763                         if (btrfs_file_extent_compression(leaf, item) ==
4764                             BTRFS_COMPRESS_ZLIB) {
4765                                 ret = uncompress_inline(path, inode, page,
4766                                                         pg_offset,
4767                                                         extent_offset, item);
4768                                 BUG_ON(ret);
4769                         } else {
4770                                 map = kmap(page);
4771                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4772                                                    copy_size);
4773                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4774                                         memset(map + pg_offset + copy_size, 0,
4775                                                PAGE_CACHE_SIZE - pg_offset -
4776                                                copy_size);
4777                                 }
4778                                 kunmap(page);
4779                         }
4780                         flush_dcache_page(page);
4781                 } else if (create && PageUptodate(page)) {
4782                         if (!trans) {
4783                                 kunmap(page);
4784                                 free_extent_map(em);
4785                                 em = NULL;
4786                                 btrfs_release_path(root, path);
4787                                 trans = btrfs_join_transaction(root, 1);
4788                                 goto again;
4789                         }
4790                         map = kmap(page);
4791                         write_extent_buffer(leaf, map + pg_offset, ptr,
4792                                             copy_size);
4793                         kunmap(page);
4794                         btrfs_mark_buffer_dirty(leaf);
4795                 }
4796                 set_extent_uptodate(io_tree, em->start,
4797                                     extent_map_end(em) - 1, GFP_NOFS);
4798                 goto insert;
4799         } else {
4800                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4801                 WARN_ON(1);
4802         }
4803 not_found:
4804         em->start = start;
4805         em->len = len;
4806 not_found_em:
4807         em->block_start = EXTENT_MAP_HOLE;
4808         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4809 insert:
4810         btrfs_release_path(root, path);
4811         if (em->start > start || extent_map_end(em) <= start) {
4812                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4813                        "[%llu %llu]\n", (unsigned long long)em->start,
4814                        (unsigned long long)em->len,
4815                        (unsigned long long)start,
4816                        (unsigned long long)len);
4817                 err = -EIO;
4818                 goto out;
4819         }
4820
4821         err = 0;
4822         write_lock(&em_tree->lock);
4823         ret = add_extent_mapping(em_tree, em);
4824         /* it is possible that someone inserted the extent into the tree
4825          * while we had the lock dropped.  It is also possible that
4826          * an overlapping map exists in the tree
4827          */
4828         if (ret == -EEXIST) {
4829                 struct extent_map *existing;
4830
4831                 ret = 0;
4832
4833                 existing = lookup_extent_mapping(em_tree, start, len);
4834                 if (existing && (existing->start > start ||
4835                     existing->start + existing->len <= start)) {
4836                         free_extent_map(existing);
4837                         existing = NULL;
4838                 }
4839                 if (!existing) {
4840                         existing = lookup_extent_mapping(em_tree, em->start,
4841                                                          em->len);
4842                         if (existing) {
4843                                 err = merge_extent_mapping(em_tree, existing,
4844                                                            em, start,
4845                                                            root->sectorsize);
4846                                 free_extent_map(existing);
4847                                 if (err) {
4848                                         free_extent_map(em);
4849                                         em = NULL;
4850                                 }
4851                         } else {
4852                                 err = -EIO;
4853                                 free_extent_map(em);
4854                                 em = NULL;
4855                         }
4856                 } else {
4857                         free_extent_map(em);
4858                         em = existing;
4859                         err = 0;
4860                 }
4861         }
4862         write_unlock(&em_tree->lock);
4863 out:
4864         if (path)
4865                 btrfs_free_path(path);
4866         if (trans) {
4867                 ret = btrfs_end_transaction(trans, root);
4868                 if (!err)
4869                         err = ret;
4870         }
4871         if (err) {
4872                 free_extent_map(em);
4873                 return ERR_PTR(err);
4874         }
4875         return em;
4876 }
4877
4878 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4879                         const struct iovec *iov, loff_t offset,
4880                         unsigned long nr_segs)
4881 {
4882         return -EINVAL;
4883 }
4884
4885 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4886                 __u64 start, __u64 len)
4887 {
4888         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4889 }
4890
4891 int btrfs_readpage(struct file *file, struct page *page)
4892 {
4893         struct extent_io_tree *tree;
4894         tree = &BTRFS_I(page->mapping->host)->io_tree;
4895         return extent_read_full_page(tree, page, btrfs_get_extent);
4896 }
4897
4898 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4899 {
4900         struct extent_io_tree *tree;
4901
4902
4903         if (current->flags & PF_MEMALLOC) {
4904                 redirty_page_for_writepage(wbc, page);
4905                 unlock_page(page);
4906                 return 0;
4907         }
4908         tree = &BTRFS_I(page->mapping->host)->io_tree;
4909         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4910 }
4911
4912 int btrfs_writepages(struct address_space *mapping,
4913                      struct writeback_control *wbc)
4914 {
4915         struct extent_io_tree *tree;
4916
4917         tree = &BTRFS_I(mapping->host)->io_tree;
4918         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4919 }
4920
4921 static int
4922 btrfs_readpages(struct file *file, struct address_space *mapping,
4923                 struct list_head *pages, unsigned nr_pages)
4924 {
4925         struct extent_io_tree *tree;
4926         tree = &BTRFS_I(mapping->host)->io_tree;
4927         return extent_readpages(tree, mapping, pages, nr_pages,
4928                                 btrfs_get_extent);
4929 }
4930 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4931 {
4932         struct extent_io_tree *tree;
4933         struct extent_map_tree *map;
4934         int ret;
4935
4936         tree = &BTRFS_I(page->mapping->host)->io_tree;
4937         map = &BTRFS_I(page->mapping->host)->extent_tree;
4938         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4939         if (ret == 1) {
4940                 ClearPagePrivate(page);
4941                 set_page_private(page, 0);
4942                 page_cache_release(page);
4943         }
4944         return ret;
4945 }
4946
4947 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4948 {
4949         if (PageWriteback(page) || PageDirty(page))
4950                 return 0;
4951         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4952 }
4953
4954 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4955 {
4956         struct extent_io_tree *tree;
4957         struct btrfs_ordered_extent *ordered;
4958         struct extent_state *cached_state = NULL;
4959         u64 page_start = page_offset(page);
4960         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4961
4962
4963         /*
4964          * we have the page locked, so new writeback can't start,
4965          * and the dirty bit won't be cleared while we are here.
4966          *
4967          * Wait for IO on this page so that we can safely clear
4968          * the PagePrivate2 bit and do ordered accounting
4969          */
4970         wait_on_page_writeback(page);
4971
4972         tree = &BTRFS_I(page->mapping->host)->io_tree;
4973         if (offset) {
4974                 btrfs_releasepage(page, GFP_NOFS);
4975                 return;
4976         }
4977         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
4978                          GFP_NOFS);
4979         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4980                                            page_offset(page));
4981         if (ordered) {
4982                 /*
4983                  * IO on this page will never be started, so we need
4984                  * to account for any ordered extents now
4985                  */
4986                 clear_extent_bit(tree, page_start, page_end,
4987                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4988                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4989                                  &cached_state, GFP_NOFS);
4990                 /*
4991                  * whoever cleared the private bit is responsible
4992                  * for the finish_ordered_io
4993                  */
4994                 if (TestClearPagePrivate2(page)) {
4995                         btrfs_finish_ordered_io(page->mapping->host,
4996                                                 page_start, page_end);
4997                 }
4998                 btrfs_put_ordered_extent(ordered);
4999                 cached_state = NULL;
5000                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5001                                  GFP_NOFS);
5002         }
5003         clear_extent_bit(tree, page_start, page_end,
5004                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5005                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5006         __btrfs_releasepage(page, GFP_NOFS);
5007
5008         ClearPageChecked(page);
5009         if (PagePrivate(page)) {
5010                 ClearPagePrivate(page);
5011                 set_page_private(page, 0);
5012                 page_cache_release(page);
5013         }
5014 }
5015
5016 /*
5017  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5018  * called from a page fault handler when a page is first dirtied. Hence we must
5019  * be careful to check for EOF conditions here. We set the page up correctly
5020  * for a written page which means we get ENOSPC checking when writing into
5021  * holes and correct delalloc and unwritten extent mapping on filesystems that
5022  * support these features.
5023  *
5024  * We are not allowed to take the i_mutex here so we have to play games to
5025  * protect against truncate races as the page could now be beyond EOF.  Because
5026  * vmtruncate() writes the inode size before removing pages, once we have the
5027  * page lock we can determine safely if the page is beyond EOF. If it is not
5028  * beyond EOF, then the page is guaranteed safe against truncation until we
5029  * unlock the page.
5030  */
5031 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5032 {
5033         struct page *page = vmf->page;
5034         struct inode *inode = fdentry(vma->vm_file)->d_inode;
5035         struct btrfs_root *root = BTRFS_I(inode)->root;
5036         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5037         struct btrfs_ordered_extent *ordered;
5038         struct extent_state *cached_state = NULL;
5039         char *kaddr;
5040         unsigned long zero_start;
5041         loff_t size;
5042         int ret;
5043         u64 page_start;
5044         u64 page_end;
5045
5046         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5047         if (ret) {
5048                 if (ret == -ENOMEM)
5049                         ret = VM_FAULT_OOM;
5050                 else /* -ENOSPC, -EIO, etc */
5051                         ret = VM_FAULT_SIGBUS;
5052                 goto out;
5053         }
5054
5055         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5056         if (ret) {
5057                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5058                 ret = VM_FAULT_SIGBUS;
5059                 goto out;
5060         }
5061
5062         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5063 again:
5064         lock_page(page);
5065         size = i_size_read(inode);
5066         page_start = page_offset(page);
5067         page_end = page_start + PAGE_CACHE_SIZE - 1;
5068
5069         if ((page->mapping != inode->i_mapping) ||
5070             (page_start >= size)) {
5071                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5072                 /* page got truncated out from underneath us */
5073                 goto out_unlock;
5074         }
5075         wait_on_page_writeback(page);
5076
5077         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
5078                          GFP_NOFS);
5079         set_page_extent_mapped(page);
5080
5081         /*
5082          * we can't set the delalloc bits if there are pending ordered
5083          * extents.  Drop our locks and wait for them to finish
5084          */
5085         ordered = btrfs_lookup_ordered_extent(inode, page_start);
5086         if (ordered) {
5087                 unlock_extent_cached(io_tree, page_start, page_end,
5088                                      &cached_state, GFP_NOFS);
5089                 unlock_page(page);
5090                 btrfs_start_ordered_extent(inode, ordered, 1);
5091                 btrfs_put_ordered_extent(ordered);
5092                 goto again;
5093         }
5094
5095         /*
5096          * XXX - page_mkwrite gets called every time the page is dirtied, even
5097          * if it was already dirty, so for space accounting reasons we need to
5098          * clear any delalloc bits for the range we are fixing to save.  There
5099          * is probably a better way to do this, but for now keep consistent with
5100          * prepare_pages in the normal write path.
5101          */
5102         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5103                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5104                           0, 0, &cached_state, GFP_NOFS);
5105
5106         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
5107                                         &cached_state);
5108         if (ret) {
5109                 unlock_extent_cached(io_tree, page_start, page_end,
5110                                      &cached_state, GFP_NOFS);
5111                 ret = VM_FAULT_SIGBUS;
5112                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5113                 goto out_unlock;
5114         }
5115         ret = 0;
5116
5117         /* page is wholly or partially inside EOF */
5118         if (page_start + PAGE_CACHE_SIZE > size)
5119                 zero_start = size & ~PAGE_CACHE_MASK;
5120         else
5121                 zero_start = PAGE_CACHE_SIZE;
5122
5123         if (zero_start != PAGE_CACHE_SIZE) {
5124                 kaddr = kmap(page);
5125                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5126                 flush_dcache_page(page);
5127                 kunmap(page);
5128         }
5129         ClearPageChecked(page);
5130         set_page_dirty(page);
5131         SetPageUptodate(page);
5132
5133         BTRFS_I(inode)->last_trans = root->fs_info->generation;
5134         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5135
5136         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
5137
5138 out_unlock:
5139         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5140         if (!ret)
5141                 return VM_FAULT_LOCKED;
5142         unlock_page(page);
5143 out:
5144         return ret;
5145 }
5146
5147 static void btrfs_truncate(struct inode *inode)
5148 {
5149         struct btrfs_root *root = BTRFS_I(inode)->root;
5150         int ret;
5151         struct btrfs_trans_handle *trans;
5152         unsigned long nr;
5153         u64 mask = root->sectorsize - 1;
5154
5155         if (!S_ISREG(inode->i_mode)) {
5156                 WARN_ON(1);
5157                 return;
5158         }
5159
5160         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5161         if (ret)
5162                 return;
5163
5164         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5165         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5166
5167         trans = btrfs_start_transaction(root, 1);
5168         btrfs_set_trans_block_group(trans, inode);
5169
5170         /*
5171          * setattr is responsible for setting the ordered_data_close flag,
5172          * but that is only tested during the last file release.  That
5173          * could happen well after the next commit, leaving a great big
5174          * window where new writes may get lost if someone chooses to write
5175          * to this file after truncating to zero
5176          *
5177          * The inode doesn't have any dirty data here, and so if we commit
5178          * this is a noop.  If someone immediately starts writing to the inode
5179          * it is very likely we'll catch some of their writes in this
5180          * transaction, and the commit will find this file on the ordered
5181          * data list with good things to send down.
5182          *
5183          * This is a best effort solution, there is still a window where
5184          * using truncate to replace the contents of the file will
5185          * end up with a zero length file after a crash.
5186          */
5187         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5188                 btrfs_add_ordered_operation(trans, root, inode);
5189
5190         while (1) {
5191                 ret = btrfs_truncate_inode_items(trans, root, inode,
5192                                                  inode->i_size,
5193                                                  BTRFS_EXTENT_DATA_KEY);
5194                 if (ret != -EAGAIN)
5195                         break;
5196
5197                 ret = btrfs_update_inode(trans, root, inode);
5198                 BUG_ON(ret);
5199
5200                 nr = trans->blocks_used;
5201                 btrfs_end_transaction(trans, root);
5202                 btrfs_btree_balance_dirty(root, nr);
5203
5204                 trans = btrfs_start_transaction(root, 1);
5205                 btrfs_set_trans_block_group(trans, inode);
5206         }
5207
5208         if (ret == 0 && inode->i_nlink > 0) {
5209                 ret = btrfs_orphan_del(trans, inode);
5210                 BUG_ON(ret);
5211         }
5212
5213         ret = btrfs_update_inode(trans, root, inode);
5214         BUG_ON(ret);
5215
5216         nr = trans->blocks_used;
5217         ret = btrfs_end_transaction_throttle(trans, root);
5218         BUG_ON(ret);
5219         btrfs_btree_balance_dirty(root, nr);
5220 }
5221
5222 /*
5223  * create a new subvolume directory/inode (helper for the ioctl).
5224  */
5225 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5226                              struct btrfs_root *new_root,
5227                              u64 new_dirid, u64 alloc_hint)
5228 {
5229         struct inode *inode;
5230         int err;
5231         u64 index = 0;
5232
5233         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5234                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5235         if (IS_ERR(inode))
5236                 return PTR_ERR(inode);
5237         inode->i_op = &btrfs_dir_inode_operations;
5238         inode->i_fop = &btrfs_dir_file_operations;
5239
5240         inode->i_nlink = 1;
5241         btrfs_i_size_write(inode, 0);
5242
5243         err = btrfs_update_inode(trans, new_root, inode);
5244         BUG_ON(err);
5245
5246         iput(inode);
5247         return 0;
5248 }
5249
5250 /* helper function for file defrag and space balancing.  This
5251  * forces readahead on a given range of bytes in an inode
5252  */
5253 unsigned long btrfs_force_ra(struct address_space *mapping,
5254                               struct file_ra_state *ra, struct file *file,
5255                               pgoff_t offset, pgoff_t last_index)
5256 {
5257         pgoff_t req_size = last_index - offset + 1;
5258
5259         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5260         return offset + req_size;
5261 }
5262
5263 struct inode *btrfs_alloc_inode(struct super_block *sb)
5264 {
5265         struct btrfs_inode *ei;
5266
5267         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5268         if (!ei)
5269                 return NULL;
5270         ei->last_trans = 0;
5271         ei->last_sub_trans = 0;
5272         ei->logged_trans = 0;
5273         ei->outstanding_extents = 0;
5274         ei->reserved_extents = 0;
5275         ei->root = NULL;
5276         spin_lock_init(&ei->accounting_lock);
5277         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5278         INIT_LIST_HEAD(&ei->i_orphan);
5279         INIT_LIST_HEAD(&ei->ordered_operations);
5280         return &ei->vfs_inode;
5281 }
5282
5283 void btrfs_destroy_inode(struct inode *inode)
5284 {
5285         struct btrfs_ordered_extent *ordered;
5286         struct btrfs_root *root = BTRFS_I(inode)->root;
5287
5288         WARN_ON(!list_empty(&inode->i_dentry));
5289         WARN_ON(inode->i_data.nrpages);
5290
5291         /*
5292          * This can happen where we create an inode, but somebody else also
5293          * created the same inode and we need to destroy the one we already
5294          * created.
5295          */
5296         if (!root)
5297                 goto free;
5298
5299         /*
5300          * Make sure we're properly removed from the ordered operation
5301          * lists.
5302          */
5303         smp_mb();
5304         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5305                 spin_lock(&root->fs_info->ordered_extent_lock);
5306                 list_del_init(&BTRFS_I(inode)->ordered_operations);
5307                 spin_unlock(&root->fs_info->ordered_extent_lock);
5308         }
5309
5310         spin_lock(&root->list_lock);
5311         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5312                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5313                        inode->i_ino);
5314                 list_del_init(&BTRFS_I(inode)->i_orphan);
5315         }
5316         spin_unlock(&root->list_lock);
5317
5318         while (1) {
5319                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5320                 if (!ordered)
5321                         break;
5322                 else {
5323                         printk(KERN_ERR "btrfs found ordered "
5324                                "extent %llu %llu on inode cleanup\n",
5325                                (unsigned long long)ordered->file_offset,
5326                                (unsigned long long)ordered->len);
5327                         btrfs_remove_ordered_extent(inode, ordered);
5328                         btrfs_put_ordered_extent(ordered);
5329                         btrfs_put_ordered_extent(ordered);
5330                 }
5331         }
5332         inode_tree_del(inode);
5333         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5334 free:
5335         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5336 }
5337
5338 void btrfs_drop_inode(struct inode *inode)
5339 {
5340         struct btrfs_root *root = BTRFS_I(inode)->root;
5341         if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5342                 generic_delete_inode(inode);
5343         else
5344                 generic_drop_inode(inode);
5345 }
5346
5347 static void init_once(void *foo)
5348 {
5349         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5350
5351         inode_init_once(&ei->vfs_inode);
5352 }
5353
5354 void btrfs_destroy_cachep(void)
5355 {
5356         if (btrfs_inode_cachep)
5357                 kmem_cache_destroy(btrfs_inode_cachep);
5358         if (btrfs_trans_handle_cachep)
5359                 kmem_cache_destroy(btrfs_trans_handle_cachep);
5360         if (btrfs_transaction_cachep)
5361                 kmem_cache_destroy(btrfs_transaction_cachep);
5362         if (btrfs_path_cachep)
5363                 kmem_cache_destroy(btrfs_path_cachep);
5364 }
5365
5366 int btrfs_init_cachep(void)
5367 {
5368         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5369                         sizeof(struct btrfs_inode), 0,
5370                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5371         if (!btrfs_inode_cachep)
5372                 goto fail;
5373
5374         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5375                         sizeof(struct btrfs_trans_handle), 0,
5376                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5377         if (!btrfs_trans_handle_cachep)
5378                 goto fail;
5379
5380         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5381                         sizeof(struct btrfs_transaction), 0,
5382                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5383         if (!btrfs_transaction_cachep)
5384                 goto fail;
5385
5386         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5387                         sizeof(struct btrfs_path), 0,
5388                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5389         if (!btrfs_path_cachep)
5390                 goto fail;
5391
5392         return 0;
5393 fail:
5394         btrfs_destroy_cachep();
5395         return -ENOMEM;
5396 }
5397
5398 static int btrfs_getattr(struct vfsmount *mnt,
5399                          struct dentry *dentry, struct kstat *stat)
5400 {
5401         struct inode *inode = dentry->d_inode;
5402         generic_fillattr(inode, stat);
5403         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5404         stat->blksize = PAGE_CACHE_SIZE;
5405         stat->blocks = (inode_get_bytes(inode) +
5406                         BTRFS_I(inode)->delalloc_bytes) >> 9;
5407         return 0;
5408 }
5409
5410 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5411                            struct inode *new_dir, struct dentry *new_dentry)
5412 {
5413         struct btrfs_trans_handle *trans;
5414         struct btrfs_root *root = BTRFS_I(old_dir)->root;
5415         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5416         struct inode *new_inode = new_dentry->d_inode;
5417         struct inode *old_inode = old_dentry->d_inode;
5418         struct timespec ctime = CURRENT_TIME;
5419         u64 index = 0;
5420         u64 root_objectid;
5421         int ret;
5422
5423         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5424                 return -EPERM;
5425
5426         /* we only allow rename subvolume link between subvolumes */
5427         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5428                 return -EXDEV;
5429
5430         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5431             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5432                 return -ENOTEMPTY;
5433
5434         if (S_ISDIR(old_inode->i_mode) && new_inode &&
5435             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5436                 return -ENOTEMPTY;
5437
5438         /*
5439          * We want to reserve the absolute worst case amount of items.  So if
5440          * both inodes are subvols and we need to unlink them then that would
5441          * require 4 item modifications, but if they are both normal inodes it
5442          * would require 5 item modifications, so we'll assume their normal
5443          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5444          * should cover the worst case number of items we'll modify.
5445          */
5446         ret = btrfs_reserve_metadata_space(root, 11);
5447         if (ret)
5448                 return ret;
5449
5450         /*
5451          * we're using rename to replace one file with another.
5452          * and the replacement file is large.  Start IO on it now so
5453          * we don't add too much work to the end of the transaction
5454          */
5455         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5456             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5457                 filemap_flush(old_inode->i_mapping);
5458
5459         /* close the racy window with snapshot create/destroy ioctl */
5460         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5461                 down_read(&root->fs_info->subvol_sem);
5462
5463         trans = btrfs_start_transaction(root, 1);
5464         btrfs_set_trans_block_group(trans, new_dir);
5465
5466         if (dest != root)
5467                 btrfs_record_root_in_trans(trans, dest);
5468
5469         ret = btrfs_set_inode_index(new_dir, &index);
5470         if (ret)
5471                 goto out_fail;
5472
5473         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5474                 /* force full log commit if subvolume involved. */
5475                 root->fs_info->last_trans_log_full_commit = trans->transid;
5476         } else {
5477                 ret = btrfs_insert_inode_ref(trans, dest,
5478                                              new_dentry->d_name.name,
5479                                              new_dentry->d_name.len,
5480                                              old_inode->i_ino,
5481                                              new_dir->i_ino, index);
5482                 if (ret)
5483                         goto out_fail;
5484                 /*
5485                  * this is an ugly little race, but the rename is required
5486                  * to make sure that if we crash, the inode is either at the
5487                  * old name or the new one.  pinning the log transaction lets
5488                  * us make sure we don't allow a log commit to come in after
5489                  * we unlink the name but before we add the new name back in.
5490                  */
5491                 btrfs_pin_log_trans(root);
5492         }
5493         /*
5494          * make sure the inode gets flushed if it is replacing
5495          * something.
5496          */
5497         if (new_inode && new_inode->i_size &&
5498             old_inode && S_ISREG(old_inode->i_mode)) {
5499                 btrfs_add_ordered_operation(trans, root, old_inode);
5500         }
5501
5502         old_dir->i_ctime = old_dir->i_mtime = ctime;
5503         new_dir->i_ctime = new_dir->i_mtime = ctime;
5504         old_inode->i_ctime = ctime;
5505
5506         if (old_dentry->d_parent != new_dentry->d_parent)
5507                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5508
5509         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5510                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5511                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5512                                         old_dentry->d_name.name,
5513                                         old_dentry->d_name.len);
5514         } else {
5515                 btrfs_inc_nlink(old_dentry->d_inode);
5516                 ret = btrfs_unlink_inode(trans, root, old_dir,
5517                                          old_dentry->d_inode,
5518                                          old_dentry->d_name.name,
5519                                          old_dentry->d_name.len);
5520         }
5521         BUG_ON(ret);
5522
5523         if (new_inode) {
5524                 new_inode->i_ctime = CURRENT_TIME;
5525                 if (unlikely(new_inode->i_ino ==
5526                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5527                         root_objectid = BTRFS_I(new_inode)->location.objectid;
5528                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
5529                                                 root_objectid,
5530                                                 new_dentry->d_name.name,
5531                                                 new_dentry->d_name.len);
5532                         BUG_ON(new_inode->i_nlink == 0);
5533                 } else {
5534                         ret = btrfs_unlink_inode(trans, dest, new_dir,
5535                                                  new_dentry->d_inode,
5536                                                  new_dentry->d_name.name,
5537                                                  new_dentry->d_name.len);
5538                 }
5539                 BUG_ON(ret);
5540                 if (new_inode->i_nlink == 0) {
5541                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5542                         BUG_ON(ret);
5543                 }
5544         }
5545
5546         ret = btrfs_add_link(trans, new_dir, old_inode,
5547                              new_dentry->d_name.name,
5548                              new_dentry->d_name.len, 0, index);
5549         BUG_ON(ret);
5550
5551         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5552                 btrfs_log_new_name(trans, old_inode, old_dir,
5553                                    new_dentry->d_parent);
5554                 btrfs_end_log_trans(root);
5555         }
5556 out_fail:
5557         btrfs_end_transaction_throttle(trans, root);
5558
5559         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5560                 up_read(&root->fs_info->subvol_sem);
5561
5562         btrfs_unreserve_metadata_space(root, 11);
5563         return ret;
5564 }
5565
5566 /*
5567  * some fairly slow code that needs optimization. This walks the list
5568  * of all the inodes with pending delalloc and forces them to disk.
5569  */
5570 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5571 {
5572         struct list_head *head = &root->fs_info->delalloc_inodes;
5573         struct btrfs_inode *binode;
5574         struct inode *inode;
5575
5576         if (root->fs_info->sb->s_flags & MS_RDONLY)
5577                 return -EROFS;
5578
5579         spin_lock(&root->fs_info->delalloc_lock);
5580         while (!list_empty(head)) {
5581                 binode = list_entry(head->next, struct btrfs_inode,
5582                                     delalloc_inodes);
5583                 inode = igrab(&binode->vfs_inode);
5584                 if (!inode)
5585                         list_del_init(&binode->delalloc_inodes);
5586                 spin_unlock(&root->fs_info->delalloc_lock);
5587                 if (inode) {
5588                         filemap_flush(inode->i_mapping);
5589                         if (delay_iput)
5590                                 btrfs_add_delayed_iput(inode);
5591                         else
5592                                 iput(inode);
5593                 }
5594                 cond_resched();
5595                 spin_lock(&root->fs_info->delalloc_lock);
5596         }
5597         spin_unlock(&root->fs_info->delalloc_lock);
5598
5599         /* the filemap_flush will queue IO into the worker threads, but
5600          * we have to make sure the IO is actually started and that
5601          * ordered extents get created before we return
5602          */
5603         atomic_inc(&root->fs_info->async_submit_draining);
5604         while (atomic_read(&root->fs_info->nr_async_submits) ||
5605               atomic_read(&root->fs_info->async_delalloc_pages)) {
5606                 wait_event(root->fs_info->async_submit_wait,
5607                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5608                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5609         }
5610         atomic_dec(&root->fs_info->async_submit_draining);
5611         return 0;
5612 }
5613
5614 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5615                          const char *symname)
5616 {
5617         struct btrfs_trans_handle *trans;
5618         struct btrfs_root *root = BTRFS_I(dir)->root;
5619         struct btrfs_path *path;
5620         struct btrfs_key key;
5621         struct inode *inode = NULL;
5622         int err;
5623         int drop_inode = 0;
5624         u64 objectid;
5625         u64 index = 0 ;
5626         int name_len;
5627         int datasize;
5628         unsigned long ptr;
5629         struct btrfs_file_extent_item *ei;
5630         struct extent_buffer *leaf;
5631         unsigned long nr = 0;
5632
5633         name_len = strlen(symname) + 1;
5634         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5635                 return -ENAMETOOLONG;
5636
5637         /*
5638          * 2 items for inode item and ref
5639          * 2 items for dir items
5640          * 1 item for xattr if selinux is on
5641          */
5642         err = btrfs_reserve_metadata_space(root, 5);
5643         if (err)
5644                 return err;
5645
5646         trans = btrfs_start_transaction(root, 1);
5647         if (!trans)
5648                 goto out_fail;
5649         btrfs_set_trans_block_group(trans, dir);
5650
5651         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5652         if (err) {
5653                 err = -ENOSPC;
5654                 goto out_unlock;
5655         }
5656
5657         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5658                                 dentry->d_name.len,
5659                                 dentry->d_parent->d_inode->i_ino, objectid,
5660                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5661                                 &index);
5662         err = PTR_ERR(inode);
5663         if (IS_ERR(inode))
5664                 goto out_unlock;
5665
5666         err = btrfs_init_inode_security(trans, inode, dir);
5667         if (err) {
5668                 drop_inode = 1;
5669                 goto out_unlock;
5670         }
5671
5672         btrfs_set_trans_block_group(trans, inode);
5673         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5674         if (err)
5675                 drop_inode = 1;
5676         else {
5677                 inode->i_mapping->a_ops = &btrfs_aops;
5678                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5679                 inode->i_fop = &btrfs_file_operations;
5680                 inode->i_op = &btrfs_file_inode_operations;
5681                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5682         }
5683         btrfs_update_inode_block_group(trans, inode);
5684         btrfs_update_inode_block_group(trans, dir);
5685         if (drop_inode)
5686                 goto out_unlock;
5687
5688         path = btrfs_alloc_path();
5689         BUG_ON(!path);
5690         key.objectid = inode->i_ino;
5691         key.offset = 0;
5692         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);