btrfs: fix return value check of btrfs_join_transaction()
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/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 compress_type = BTRFS_COMPRESS_NONE;
126
127         if (compressed_size && compressed_pages) {
128                 compress_type = root->fs_info->compress_type;
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 (compress_type != BTRFS_COMPRESS_NONE) {
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                                                   compress_type);
180         } else {
181                 page = find_get_page(inode->i_mapping,
182                                      start >> PAGE_CACHE_SHIFT);
183                 btrfs_set_file_extent_compression(leaf, ei, 0);
184                 kaddr = kmap_atomic(page, KM_USER0);
185                 offset = start & (PAGE_CACHE_SIZE - 1);
186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187                 kunmap_atomic(kaddr, KM_USER0);
188                 page_cache_release(page);
189         }
190         btrfs_mark_buffer_dirty(leaf);
191         btrfs_free_path(path);
192
193         /*
194          * we're an inline extent, so nobody can
195          * extend the file past i_size without locking
196          * a page we already have locked.
197          *
198          * We must do any isize and inode updates
199          * before we unlock the pages.  Otherwise we
200          * could end up racing with unlink.
201          */
202         BTRFS_I(inode)->disk_i_size = inode->i_size;
203         btrfs_update_inode(trans, root, inode);
204
205         return 0;
206 fail:
207         btrfs_free_path(path);
208         return err;
209 }
210
211
212 /*
213  * conditionally insert an inline extent into the file.  This
214  * does the checks required to make sure the data is small enough
215  * to fit as an inline extent.
216  */
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218                                  struct btrfs_root *root,
219                                  struct inode *inode, u64 start, u64 end,
220                                  size_t compressed_size,
221                                  struct page **compressed_pages)
222 {
223         u64 isize = i_size_read(inode);
224         u64 actual_end = min(end + 1, isize);
225         u64 inline_len = actual_end - start;
226         u64 aligned_end = (end + root->sectorsize - 1) &
227                         ~((u64)root->sectorsize - 1);
228         u64 hint_byte;
229         u64 data_len = inline_len;
230         int ret;
231
232         if (compressed_size)
233                 data_len = compressed_size;
234
235         if (start > 0 ||
236             actual_end >= PAGE_CACHE_SIZE ||
237             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238             (!compressed_size &&
239             (actual_end & (root->sectorsize - 1)) == 0) ||
240             end + 1 < isize ||
241             data_len > root->fs_info->max_inline) {
242                 return 1;
243         }
244
245         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
246                                  &hint_byte, 1);
247         BUG_ON(ret);
248
249         if (isize > actual_end)
250                 inline_len = min_t(u64, isize, actual_end);
251         ret = insert_inline_extent(trans, root, inode, start,
252                                    inline_len, compressed_size,
253                                    compressed_pages);
254         BUG_ON(ret);
255         btrfs_delalloc_release_metadata(inode, end + 1 - start);
256         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
257         return 0;
258 }
259
260 struct async_extent {
261         u64 start;
262         u64 ram_size;
263         u64 compressed_size;
264         struct page **pages;
265         unsigned long nr_pages;
266         int compress_type;
267         struct list_head list;
268 };
269
270 struct async_cow {
271         struct inode *inode;
272         struct btrfs_root *root;
273         struct page *locked_page;
274         u64 start;
275         u64 end;
276         struct list_head extents;
277         struct btrfs_work work;
278 };
279
280 static noinline int add_async_extent(struct async_cow *cow,
281                                      u64 start, u64 ram_size,
282                                      u64 compressed_size,
283                                      struct page **pages,
284                                      unsigned long nr_pages,
285                                      int compress_type)
286 {
287         struct async_extent *async_extent;
288
289         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
290         async_extent->start = start;
291         async_extent->ram_size = ram_size;
292         async_extent->compressed_size = compressed_size;
293         async_extent->pages = pages;
294         async_extent->nr_pages = nr_pages;
295         async_extent->compress_type = compress_type;
296         list_add_tail(&async_extent->list, &cow->extents);
297         return 0;
298 }
299
300 /*
301  * we create compressed extents in two phases.  The first
302  * phase compresses a range of pages that have already been
303  * locked (both pages and state bits are locked).
304  *
305  * This is done inside an ordered work queue, and the compression
306  * is spread across many cpus.  The actual IO submission is step
307  * two, and the ordered work queue takes care of making sure that
308  * happens in the same order things were put onto the queue by
309  * writepages and friends.
310  *
311  * If this code finds it can't get good compression, it puts an
312  * entry onto the work queue to write the uncompressed bytes.  This
313  * makes sure that both compressed inodes and uncompressed inodes
314  * are written in the same order that pdflush sent them down.
315  */
316 static noinline int compress_file_range(struct inode *inode,
317                                         struct page *locked_page,
318                                         u64 start, u64 end,
319                                         struct async_cow *async_cow,
320                                         int *num_added)
321 {
322         struct btrfs_root *root = BTRFS_I(inode)->root;
323         struct btrfs_trans_handle *trans;
324         u64 num_bytes;
325         u64 blocksize = root->sectorsize;
326         u64 actual_end;
327         u64 isize = i_size_read(inode);
328         int ret = 0;
329         struct page **pages = NULL;
330         unsigned long nr_pages;
331         unsigned long nr_pages_ret = 0;
332         unsigned long total_compressed = 0;
333         unsigned long total_in = 0;
334         unsigned long max_compressed = 128 * 1024;
335         unsigned long max_uncompressed = 128 * 1024;
336         int i;
337         int will_compress;
338         int compress_type = root->fs_info->compress_type;
339
340         actual_end = min_t(u64, isize, end + 1);
341 again:
342         will_compress = 0;
343         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
345
346         /*
347          * we don't want to send crud past the end of i_size through
348          * compression, that's just a waste of CPU time.  So, if the
349          * end of the file is before the start of our current
350          * requested range of bytes, we bail out to the uncompressed
351          * cleanup code that can deal with all of this.
352          *
353          * It isn't really the fastest way to fix things, but this is a
354          * very uncommon corner.
355          */
356         if (actual_end <= start)
357                 goto cleanup_and_bail_uncompressed;
358
359         total_compressed = actual_end - start;
360
361         /* we want to make sure that amount of ram required to uncompress
362          * an extent is reasonable, so we limit the total size in ram
363          * of a compressed extent to 128k.  This is a crucial number
364          * because it also controls how easily we can spread reads across
365          * cpus for decompression.
366          *
367          * We also want to make sure the amount of IO required to do
368          * a random read is reasonably small, so we limit the size of
369          * a compressed extent to 128k.
370          */
371         total_compressed = min(total_compressed, max_uncompressed);
372         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373         num_bytes = max(blocksize,  num_bytes);
374         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                 if (BTRFS_I(inode)->force_compress)
389                         compress_type = BTRFS_I(inode)->force_compress;
390
391                 ret = btrfs_compress_pages(compress_type,
392                                            inode->i_mapping, start,
393                                            total_compressed, pages,
394                                            nr_pages, &nr_pages_ret,
395                                            &total_in,
396                                            &total_compressed,
397                                            max_compressed);
398
399                 if (!ret) {
400                         unsigned long offset = total_compressed &
401                                 (PAGE_CACHE_SIZE - 1);
402                         struct page *page = pages[nr_pages_ret - 1];
403                         char *kaddr;
404
405                         /* zero the tail end of the last page, we might be
406                          * sending it down to disk
407                          */
408                         if (offset) {
409                                 kaddr = kmap_atomic(page, KM_USER0);
410                                 memset(kaddr + offset, 0,
411                                        PAGE_CACHE_SIZE - offset);
412                                 kunmap_atomic(kaddr, KM_USER0);
413                         }
414                         will_compress = 1;
415                 }
416         }
417         if (start == 0) {
418                 trans = btrfs_join_transaction(root, 1);
419                 BUG_ON(IS_ERR(trans));
420                 btrfs_set_trans_block_group(trans, inode);
421                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
422
423                 /* lets try to make an inline extent */
424                 if (ret || total_in < (actual_end - start)) {
425                         /* we didn't compress the entire range, try
426                          * to make an uncompressed inline extent.
427                          */
428                         ret = cow_file_range_inline(trans, root, inode,
429                                                     start, end, 0, NULL);
430                 } else {
431                         /* try making a compressed inline extent */
432                         ret = cow_file_range_inline(trans, root, inode,
433                                                     start, end,
434                                                     total_compressed, pages);
435                 }
436                 if (ret == 0) {
437                         /*
438                          * inline extent creation worked, we don't need
439                          * to create any more async work items.  Unlock
440                          * and free up our temp pages.
441                          */
442                         extent_clear_unlock_delalloc(inode,
443                              &BTRFS_I(inode)->io_tree,
444                              start, end, NULL,
445                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
446                              EXTENT_CLEAR_DELALLOC |
447                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
448
449                         btrfs_end_transaction(trans, root);
450                         goto free_pages_out;
451                 }
452                 btrfs_end_transaction(trans, root);
453         }
454
455         if (will_compress) {
456                 /*
457                  * we aren't doing an inline extent round the compressed size
458                  * up to a block size boundary so the allocator does sane
459                  * things
460                  */
461                 total_compressed = (total_compressed + blocksize - 1) &
462                         ~(blocksize - 1);
463
464                 /*
465                  * one last check to make sure the compression is really a
466                  * win, compare the page count read with the blocks on disk
467                  */
468                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
469                         ~(PAGE_CACHE_SIZE - 1);
470                 if (total_compressed >= total_in) {
471                         will_compress = 0;
472                 } else {
473                         num_bytes = total_in;
474                 }
475         }
476         if (!will_compress && pages) {
477                 /*
478                  * the compression code ran but failed to make things smaller,
479                  * free any pages it allocated and our page pointer array
480                  */
481                 for (i = 0; i < nr_pages_ret; i++) {
482                         WARN_ON(pages[i]->mapping);
483                         page_cache_release(pages[i]);
484                 }
485                 kfree(pages);
486                 pages = NULL;
487                 total_compressed = 0;
488                 nr_pages_ret = 0;
489
490                 /* flag the file so we don't compress in the future */
491                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
492                     !(BTRFS_I(inode)->force_compress)) {
493                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
494                 }
495         }
496         if (will_compress) {
497                 *num_added += 1;
498
499                 /* the async work queues will take care of doing actual
500                  * allocation on disk for these compressed pages,
501                  * and will submit them to the elevator.
502                  */
503                 add_async_extent(async_cow, start, num_bytes,
504                                  total_compressed, pages, nr_pages_ret,
505                                  compress_type);
506
507                 if (start + num_bytes < end) {
508                         start += num_bytes;
509                         pages = NULL;
510                         cond_resched();
511                         goto again;
512                 }
513         } else {
514 cleanup_and_bail_uncompressed:
515                 /*
516                  * No compression, but we still need to write the pages in
517                  * the file we've been given so far.  redirty the locked
518                  * page if it corresponds to our extent and set things up
519                  * for the async work queue to run cow_file_range to do
520                  * the normal delalloc dance
521                  */
522                 if (page_offset(locked_page) >= start &&
523                     page_offset(locked_page) <= end) {
524                         __set_page_dirty_nobuffers(locked_page);
525                         /* unlocked later on in the async handlers */
526                 }
527                 add_async_extent(async_cow, start, end - start + 1,
528                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
529                 *num_added += 1;
530         }
531
532 out:
533         return 0;
534
535 free_pages_out:
536         for (i = 0; i < nr_pages_ret; i++) {
537                 WARN_ON(pages[i]->mapping);
538                 page_cache_release(pages[i]);
539         }
540         kfree(pages);
541
542         goto out;
543 }
544
545 /*
546  * phase two of compressed writeback.  This is the ordered portion
547  * of the code, which only gets called in the order the work was
548  * queued.  We walk all the async extents created by compress_file_range
549  * and send them down to the disk.
550  */
551 static noinline int submit_compressed_extents(struct inode *inode,
552                                               struct async_cow *async_cow)
553 {
554         struct async_extent *async_extent;
555         u64 alloc_hint = 0;
556         struct btrfs_trans_handle *trans;
557         struct btrfs_key ins;
558         struct extent_map *em;
559         struct btrfs_root *root = BTRFS_I(inode)->root;
560         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
561         struct extent_io_tree *io_tree;
562         int ret = 0;
563
564         if (list_empty(&async_cow->extents))
565                 return 0;
566
567
568         while (!list_empty(&async_cow->extents)) {
569                 async_extent = list_entry(async_cow->extents.next,
570                                           struct async_extent, list);
571                 list_del(&async_extent->list);
572
573                 io_tree = &BTRFS_I(inode)->io_tree;
574
575 retry:
576                 /* did the compression code fall back to uncompressed IO? */
577                 if (!async_extent->pages) {
578                         int page_started = 0;
579                         unsigned long nr_written = 0;
580
581                         lock_extent(io_tree, async_extent->start,
582                                          async_extent->start +
583                                          async_extent->ram_size - 1, GFP_NOFS);
584
585                         /* allocate blocks */
586                         ret = cow_file_range(inode, async_cow->locked_page,
587                                              async_extent->start,
588                                              async_extent->start +
589                                              async_extent->ram_size - 1,
590                                              &page_started, &nr_written, 0);
591
592                         /*
593                          * if page_started, cow_file_range inserted an
594                          * inline extent and took care of all the unlocking
595                          * and IO for us.  Otherwise, we need to submit
596                          * all those pages down to the drive.
597                          */
598                         if (!page_started && !ret)
599                                 extent_write_locked_range(io_tree,
600                                                   inode, async_extent->start,
601                                                   async_extent->start +
602                                                   async_extent->ram_size - 1,
603                                                   btrfs_get_extent,
604                                                   WB_SYNC_ALL);
605                         kfree(async_extent);
606                         cond_resched();
607                         continue;
608                 }
609
610                 lock_extent(io_tree, async_extent->start,
611                             async_extent->start + async_extent->ram_size - 1,
612                             GFP_NOFS);
613
614                 trans = btrfs_join_transaction(root, 1);
615                 BUG_ON(IS_ERR(trans));
616                 ret = btrfs_reserve_extent(trans, root,
617                                            async_extent->compressed_size,
618                                            async_extent->compressed_size,
619                                            0, alloc_hint,
620                                            (u64)-1, &ins, 1);
621                 btrfs_end_transaction(trans, root);
622
623                 if (ret) {
624                         int i;
625                         for (i = 0; i < async_extent->nr_pages; i++) {
626                                 WARN_ON(async_extent->pages[i]->mapping);
627                                 page_cache_release(async_extent->pages[i]);
628                         }
629                         kfree(async_extent->pages);
630                         async_extent->nr_pages = 0;
631                         async_extent->pages = NULL;
632                         unlock_extent(io_tree, async_extent->start,
633                                       async_extent->start +
634                                       async_extent->ram_size - 1, GFP_NOFS);
635                         goto retry;
636                 }
637
638                 /*
639                  * here we're doing allocation and writeback of the
640                  * compressed pages
641                  */
642                 btrfs_drop_extent_cache(inode, async_extent->start,
643                                         async_extent->start +
644                                         async_extent->ram_size - 1, 0);
645
646                 em = alloc_extent_map(GFP_NOFS);
647                 em->start = async_extent->start;
648                 em->len = async_extent->ram_size;
649                 em->orig_start = em->start;
650
651                 em->block_start = ins.objectid;
652                 em->block_len = ins.offset;
653                 em->bdev = root->fs_info->fs_devices->latest_bdev;
654                 em->compress_type = async_extent->compress_type;
655                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
656                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
657
658                 while (1) {
659                         write_lock(&em_tree->lock);
660                         ret = add_extent_mapping(em_tree, em);
661                         write_unlock(&em_tree->lock);
662                         if (ret != -EEXIST) {
663                                 free_extent_map(em);
664                                 break;
665                         }
666                         btrfs_drop_extent_cache(inode, async_extent->start,
667                                                 async_extent->start +
668                                                 async_extent->ram_size - 1, 0);
669                 }
670
671                 ret = btrfs_add_ordered_extent_compress(inode,
672                                                 async_extent->start,
673                                                 ins.objectid,
674                                                 async_extent->ram_size,
675                                                 ins.offset,
676                                                 BTRFS_ORDERED_COMPRESSED,
677                                                 async_extent->compress_type);
678                 BUG_ON(ret);
679
680                 /*
681                  * clear dirty, set writeback and unlock the pages.
682                  */
683                 extent_clear_unlock_delalloc(inode,
684                                 &BTRFS_I(inode)->io_tree,
685                                 async_extent->start,
686                                 async_extent->start +
687                                 async_extent->ram_size - 1,
688                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
689                                 EXTENT_CLEAR_UNLOCK |
690                                 EXTENT_CLEAR_DELALLOC |
691                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
692
693                 ret = btrfs_submit_compressed_write(inode,
694                                     async_extent->start,
695                                     async_extent->ram_size,
696                                     ins.objectid,
697                                     ins.offset, async_extent->pages,
698                                     async_extent->nr_pages);
699
700                 BUG_ON(ret);
701                 alloc_hint = ins.objectid + ins.offset;
702                 kfree(async_extent);
703                 cond_resched();
704         }
705
706         return 0;
707 }
708
709 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
710                                       u64 num_bytes)
711 {
712         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
713         struct extent_map *em;
714         u64 alloc_hint = 0;
715
716         read_lock(&em_tree->lock);
717         em = search_extent_mapping(em_tree, start, num_bytes);
718         if (em) {
719                 /*
720                  * if block start isn't an actual block number then find the
721                  * first block in this inode and use that as a hint.  If that
722                  * block is also bogus then just don't worry about it.
723                  */
724                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
725                         free_extent_map(em);
726                         em = search_extent_mapping(em_tree, 0, 0);
727                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
728                                 alloc_hint = em->block_start;
729                         if (em)
730                                 free_extent_map(em);
731                 } else {
732                         alloc_hint = em->block_start;
733                         free_extent_map(em);
734                 }
735         }
736         read_unlock(&em_tree->lock);
737
738         return alloc_hint;
739 }
740
741 /*
742  * when extent_io.c finds a delayed allocation range in the file,
743  * the call backs end up in this code.  The basic idea is to
744  * allocate extents on disk for the range, and create ordered data structs
745  * in ram to track those extents.
746  *
747  * locked_page is the page that writepage had locked already.  We use
748  * it to make sure we don't do extra locks or unlocks.
749  *
750  * *page_started is set to one if we unlock locked_page and do everything
751  * required to start IO on it.  It may be clean and already done with
752  * IO when we return.
753  */
754 static noinline int cow_file_range(struct inode *inode,
755                                    struct page *locked_page,
756                                    u64 start, u64 end, int *page_started,
757                                    unsigned long *nr_written,
758                                    int unlock)
759 {
760         struct btrfs_root *root = BTRFS_I(inode)->root;
761         struct btrfs_trans_handle *trans;
762         u64 alloc_hint = 0;
763         u64 num_bytes;
764         unsigned long ram_size;
765         u64 disk_num_bytes;
766         u64 cur_alloc_size;
767         u64 blocksize = root->sectorsize;
768         struct btrfs_key ins;
769         struct extent_map *em;
770         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
771         int ret = 0;
772
773         BUG_ON(root == root->fs_info->tree_root);
774         trans = btrfs_join_transaction(root, 1);
775         BUG_ON(IS_ERR(trans));
776         btrfs_set_trans_block_group(trans, inode);
777         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
778
779         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
780         num_bytes = max(blocksize,  num_bytes);
781         disk_num_bytes = num_bytes;
782         ret = 0;
783
784         if (start == 0) {
785                 /* lets try to make an inline extent */
786                 ret = cow_file_range_inline(trans, root, inode,
787                                             start, end, 0, NULL);
788                 if (ret == 0) {
789                         extent_clear_unlock_delalloc(inode,
790                                      &BTRFS_I(inode)->io_tree,
791                                      start, end, NULL,
792                                      EXTENT_CLEAR_UNLOCK_PAGE |
793                                      EXTENT_CLEAR_UNLOCK |
794                                      EXTENT_CLEAR_DELALLOC |
795                                      EXTENT_CLEAR_DIRTY |
796                                      EXTENT_SET_WRITEBACK |
797                                      EXTENT_END_WRITEBACK);
798
799                         *nr_written = *nr_written +
800                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
801                         *page_started = 1;
802                         ret = 0;
803                         goto out;
804                 }
805         }
806
807         BUG_ON(disk_num_bytes >
808                btrfs_super_total_bytes(&root->fs_info->super_copy));
809
810         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
811         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
812
813         while (disk_num_bytes > 0) {
814                 unsigned long op;
815
816                 cur_alloc_size = disk_num_bytes;
817                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
818                                            root->sectorsize, 0, alloc_hint,
819                                            (u64)-1, &ins, 1);
820                 BUG_ON(ret);
821
822                 em = alloc_extent_map(GFP_NOFS);
823                 em->start = start;
824                 em->orig_start = em->start;
825                 ram_size = ins.offset;
826                 em->len = ins.offset;
827
828                 em->block_start = ins.objectid;
829                 em->block_len = ins.offset;
830                 em->bdev = root->fs_info->fs_devices->latest_bdev;
831                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
832
833                 while (1) {
834                         write_lock(&em_tree->lock);
835                         ret = add_extent_mapping(em_tree, em);
836                         write_unlock(&em_tree->lock);
837                         if (ret != -EEXIST) {
838                                 free_extent_map(em);
839                                 break;
840                         }
841                         btrfs_drop_extent_cache(inode, start,
842                                                 start + ram_size - 1, 0);
843                 }
844
845                 cur_alloc_size = ins.offset;
846                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
847                                                ram_size, cur_alloc_size, 0);
848                 BUG_ON(ret);
849
850                 if (root->root_key.objectid ==
851                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
852                         ret = btrfs_reloc_clone_csums(inode, start,
853                                                       cur_alloc_size);
854                         BUG_ON(ret);
855                 }
856
857                 if (disk_num_bytes < cur_alloc_size)
858                         break;
859
860                 /* we're not doing compressed IO, don't unlock the first
861                  * page (which the caller expects to stay locked), don't
862                  * clear any dirty bits and don't set any writeback bits
863                  *
864                  * Do set the Private2 bit so we know this page was properly
865                  * setup for writepage
866                  */
867                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
868                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
869                         EXTENT_SET_PRIVATE2;
870
871                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
872                                              start, start + ram_size - 1,
873                                              locked_page, op);
874                 disk_num_bytes -= cur_alloc_size;
875                 num_bytes -= cur_alloc_size;
876                 alloc_hint = ins.objectid + ins.offset;
877                 start += cur_alloc_size;
878         }
879 out:
880         ret = 0;
881         btrfs_end_transaction(trans, root);
882
883         return ret;
884 }
885
886 /*
887  * work queue call back to started compression on a file and pages
888  */
889 static noinline void async_cow_start(struct btrfs_work *work)
890 {
891         struct async_cow *async_cow;
892         int num_added = 0;
893         async_cow = container_of(work, struct async_cow, work);
894
895         compress_file_range(async_cow->inode, async_cow->locked_page,
896                             async_cow->start, async_cow->end, async_cow,
897                             &num_added);
898         if (num_added == 0)
899                 async_cow->inode = NULL;
900 }
901
902 /*
903  * work queue call back to submit previously compressed pages
904  */
905 static noinline void async_cow_submit(struct btrfs_work *work)
906 {
907         struct async_cow *async_cow;
908         struct btrfs_root *root;
909         unsigned long nr_pages;
910
911         async_cow = container_of(work, struct async_cow, work);
912
913         root = async_cow->root;
914         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
915                 PAGE_CACHE_SHIFT;
916
917         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
918
919         if (atomic_read(&root->fs_info->async_delalloc_pages) <
920             5 * 1042 * 1024 &&
921             waitqueue_active(&root->fs_info->async_submit_wait))
922                 wake_up(&root->fs_info->async_submit_wait);
923
924         if (async_cow->inode)
925                 submit_compressed_extents(async_cow->inode, async_cow);
926 }
927
928 static noinline void async_cow_free(struct btrfs_work *work)
929 {
930         struct async_cow *async_cow;
931         async_cow = container_of(work, struct async_cow, work);
932         kfree(async_cow);
933 }
934
935 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
936                                 u64 start, u64 end, int *page_started,
937                                 unsigned long *nr_written)
938 {
939         struct async_cow *async_cow;
940         struct btrfs_root *root = BTRFS_I(inode)->root;
941         unsigned long nr_pages;
942         u64 cur_end;
943         int limit = 10 * 1024 * 1042;
944
945         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
946                          1, 0, NULL, GFP_NOFS);
947         while (start < end) {
948                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
949                 async_cow->inode = inode;
950                 async_cow->root = root;
951                 async_cow->locked_page = locked_page;
952                 async_cow->start = start;
953
954                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
955                         cur_end = end;
956                 else
957                         cur_end = min(end, start + 512 * 1024 - 1);
958
959                 async_cow->end = cur_end;
960                 INIT_LIST_HEAD(&async_cow->extents);
961
962                 async_cow->work.func = async_cow_start;
963                 async_cow->work.ordered_func = async_cow_submit;
964                 async_cow->work.ordered_free = async_cow_free;
965                 async_cow->work.flags = 0;
966
967                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
968                         PAGE_CACHE_SHIFT;
969                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
970
971                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
972                                    &async_cow->work);
973
974                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
975                         wait_event(root->fs_info->async_submit_wait,
976                            (atomic_read(&root->fs_info->async_delalloc_pages) <
977                             limit));
978                 }
979
980                 while (atomic_read(&root->fs_info->async_submit_draining) &&
981                       atomic_read(&root->fs_info->async_delalloc_pages)) {
982                         wait_event(root->fs_info->async_submit_wait,
983                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
984                            0));
985                 }
986
987                 *nr_written += nr_pages;
988                 start = cur_end + 1;
989         }
990         *page_started = 1;
991         return 0;
992 }
993
994 static noinline int csum_exist_in_range(struct btrfs_root *root,
995                                         u64 bytenr, u64 num_bytes)
996 {
997         int ret;
998         struct btrfs_ordered_sum *sums;
999         LIST_HEAD(list);
1000
1001         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1002                                        bytenr + num_bytes - 1, &list);
1003         if (ret == 0 && list_empty(&list))
1004                 return 0;
1005
1006         while (!list_empty(&list)) {
1007                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1008                 list_del(&sums->list);
1009                 kfree(sums);
1010         }
1011         return 1;
1012 }
1013
1014 /*
1015  * when nowcow writeback call back.  This checks for snapshots or COW copies
1016  * of the extents that exist in the file, and COWs the file as required.
1017  *
1018  * If no cow copies or snapshots exist, we write directly to the existing
1019  * blocks on disk
1020  */
1021 static noinline int run_delalloc_nocow(struct inode *inode,
1022                                        struct page *locked_page,
1023                               u64 start, u64 end, int *page_started, int force,
1024                               unsigned long *nr_written)
1025 {
1026         struct btrfs_root *root = BTRFS_I(inode)->root;
1027         struct btrfs_trans_handle *trans;
1028         struct extent_buffer *leaf;
1029         struct btrfs_path *path;
1030         struct btrfs_file_extent_item *fi;
1031         struct btrfs_key found_key;
1032         u64 cow_start;
1033         u64 cur_offset;
1034         u64 extent_end;
1035         u64 extent_offset;
1036         u64 disk_bytenr;
1037         u64 num_bytes;
1038         int extent_type;
1039         int ret;
1040         int type;
1041         int nocow;
1042         int check_prev = 1;
1043         bool nolock = false;
1044
1045         path = btrfs_alloc_path();
1046         BUG_ON(!path);
1047         if (root == root->fs_info->tree_root) {
1048                 nolock = true;
1049                 trans = btrfs_join_transaction_nolock(root, 1);
1050         } else {
1051                 trans = btrfs_join_transaction(root, 1);
1052         }
1053         BUG_ON(IS_ERR(trans));
1054
1055         cow_start = (u64)-1;
1056         cur_offset = start;
1057         while (1) {
1058                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1059                                                cur_offset, 0);
1060                 BUG_ON(ret < 0);
1061                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1062                         leaf = path->nodes[0];
1063                         btrfs_item_key_to_cpu(leaf, &found_key,
1064                                               path->slots[0] - 1);
1065                         if (found_key.objectid == inode->i_ino &&
1066                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1067                                 path->slots[0]--;
1068                 }
1069                 check_prev = 0;
1070 next_slot:
1071                 leaf = path->nodes[0];
1072                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073                         ret = btrfs_next_leaf(root, path);
1074                         if (ret < 0)
1075                                 BUG_ON(1);
1076                         if (ret > 0)
1077                                 break;
1078                         leaf = path->nodes[0];
1079                 }
1080
1081                 nocow = 0;
1082                 disk_bytenr = 0;
1083                 num_bytes = 0;
1084                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1085
1086                 if (found_key.objectid > inode->i_ino ||
1087                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1088                     found_key.offset > end)
1089                         break;
1090
1091                 if (found_key.offset > cur_offset) {
1092                         extent_end = found_key.offset;
1093                         extent_type = 0;
1094                         goto out_check;
1095                 }
1096
1097                 fi = btrfs_item_ptr(leaf, path->slots[0],
1098                                     struct btrfs_file_extent_item);
1099                 extent_type = btrfs_file_extent_type(leaf, fi);
1100
1101                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1102                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1103                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1104                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1105                         extent_end = found_key.offset +
1106                                 btrfs_file_extent_num_bytes(leaf, fi);
1107                         if (extent_end <= start) {
1108                                 path->slots[0]++;
1109                                 goto next_slot;
1110                         }
1111                         if (disk_bytenr == 0)
1112                                 goto out_check;
1113                         if (btrfs_file_extent_compression(leaf, fi) ||
1114                             btrfs_file_extent_encryption(leaf, fi) ||
1115                             btrfs_file_extent_other_encoding(leaf, fi))
1116                                 goto out_check;
1117                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1118                                 goto out_check;
1119                         if (btrfs_extent_readonly(root, disk_bytenr))
1120                                 goto out_check;
1121                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1122                                                   found_key.offset -
1123                                                   extent_offset, disk_bytenr))
1124                                 goto out_check;
1125                         disk_bytenr += extent_offset;
1126                         disk_bytenr += cur_offset - found_key.offset;
1127                         num_bytes = min(end + 1, extent_end) - cur_offset;
1128                         /*
1129                          * force cow if csum exists in the range.
1130                          * this ensure that csum for a given extent are
1131                          * either valid or do not exist.
1132                          */
1133                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1134                                 goto out_check;
1135                         nocow = 1;
1136                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1137                         extent_end = found_key.offset +
1138                                 btrfs_file_extent_inline_len(leaf, fi);
1139                         extent_end = ALIGN(extent_end, root->sectorsize);
1140                 } else {
1141                         BUG_ON(1);
1142                 }
1143 out_check:
1144                 if (extent_end <= start) {
1145                         path->slots[0]++;
1146                         goto next_slot;
1147                 }
1148                 if (!nocow) {
1149                         if (cow_start == (u64)-1)
1150                                 cow_start = cur_offset;
1151                         cur_offset = extent_end;
1152                         if (cur_offset > end)
1153                                 break;
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157
1158                 btrfs_release_path(root, path);
1159                 if (cow_start != (u64)-1) {
1160                         ret = cow_file_range(inode, locked_page, cow_start,
1161                                         found_key.offset - 1, page_started,
1162                                         nr_written, 1);
1163                         BUG_ON(ret);
1164                         cow_start = (u64)-1;
1165                 }
1166
1167                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1168                         struct extent_map *em;
1169                         struct extent_map_tree *em_tree;
1170                         em_tree = &BTRFS_I(inode)->extent_tree;
1171                         em = alloc_extent_map(GFP_NOFS);
1172                         em->start = cur_offset;
1173                         em->orig_start = em->start;
1174                         em->len = num_bytes;
1175                         em->block_len = num_bytes;
1176                         em->block_start = disk_bytenr;
1177                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1178                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1179                         while (1) {
1180                                 write_lock(&em_tree->lock);
1181                                 ret = add_extent_mapping(em_tree, em);
1182                                 write_unlock(&em_tree->lock);
1183                                 if (ret != -EEXIST) {
1184                                         free_extent_map(em);
1185                                         break;
1186                                 }
1187                                 btrfs_drop_extent_cache(inode, em->start,
1188                                                 em->start + em->len - 1, 0);
1189                         }
1190                         type = BTRFS_ORDERED_PREALLOC;
1191                 } else {
1192                         type = BTRFS_ORDERED_NOCOW;
1193                 }
1194
1195                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1196                                                num_bytes, num_bytes, type);
1197                 BUG_ON(ret);
1198
1199                 if (root->root_key.objectid ==
1200                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1201                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1202                                                       num_bytes);
1203                         BUG_ON(ret);
1204                 }
1205
1206                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1207                                 cur_offset, cur_offset + num_bytes - 1,
1208                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1209                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1210                                 EXTENT_SET_PRIVATE2);
1211                 cur_offset = extent_end;
1212                 if (cur_offset > end)
1213                         break;
1214         }
1215         btrfs_release_path(root, path);
1216
1217         if (cur_offset <= end && cow_start == (u64)-1)
1218                 cow_start = cur_offset;
1219         if (cow_start != (u64)-1) {
1220                 ret = cow_file_range(inode, locked_page, cow_start, end,
1221                                      page_started, nr_written, 1);
1222                 BUG_ON(ret);
1223         }
1224
1225         if (nolock) {
1226                 ret = btrfs_end_transaction_nolock(trans, root);
1227                 BUG_ON(ret);
1228         } else {
1229                 ret = btrfs_end_transaction(trans, root);
1230                 BUG_ON(ret);
1231         }
1232         btrfs_free_path(path);
1233         return 0;
1234 }
1235
1236 /*
1237  * extent_io.c call back to do delayed allocation processing
1238  */
1239 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1240                               u64 start, u64 end, int *page_started,
1241                               unsigned long *nr_written)
1242 {
1243         int ret;
1244         struct btrfs_root *root = BTRFS_I(inode)->root;
1245
1246         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1247                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1248                                          page_started, 1, nr_written);
1249         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1250                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1251                                          page_started, 0, nr_written);
1252         else if (!btrfs_test_opt(root, COMPRESS) &&
1253                  !(BTRFS_I(inode)->force_compress))
1254                 ret = cow_file_range(inode, locked_page, start, end,
1255                                       page_started, nr_written, 1);
1256         else
1257                 ret = cow_file_range_async(inode, locked_page, start, end,
1258                                            page_started, nr_written);
1259         return ret;
1260 }
1261
1262 static int btrfs_split_extent_hook(struct inode *inode,
1263                                    struct extent_state *orig, u64 split)
1264 {
1265         /* not delalloc, ignore it */
1266         if (!(orig->state & EXTENT_DELALLOC))
1267                 return 0;
1268
1269         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1270         return 0;
1271 }
1272
1273 /*
1274  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1275  * extents so we can keep track of new extents that are just merged onto old
1276  * extents, such as when we are doing sequential writes, so we can properly
1277  * account for the metadata space we'll need.
1278  */
1279 static int btrfs_merge_extent_hook(struct inode *inode,
1280                                    struct extent_state *new,
1281                                    struct extent_state *other)
1282 {
1283         /* not delalloc, ignore it */
1284         if (!(other->state & EXTENT_DELALLOC))
1285                 return 0;
1286
1287         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1288         return 0;
1289 }
1290
1291 /*
1292  * extent_io.c set_bit_hook, used to track delayed allocation
1293  * bytes in this file, and to maintain the list of inodes that
1294  * have pending delalloc work to be done.
1295  */
1296 static int btrfs_set_bit_hook(struct inode *inode,
1297                               struct extent_state *state, int *bits)
1298 {
1299
1300         /*
1301          * set_bit and clear bit hooks normally require _irqsave/restore
1302          * but in this case, we are only testeing for the DELALLOC
1303          * bit, which is only set or cleared with irqs on
1304          */
1305         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1306                 struct btrfs_root *root = BTRFS_I(inode)->root;
1307                 u64 len = state->end + 1 - state->start;
1308                 int do_list = (root->root_key.objectid !=
1309                                BTRFS_ROOT_TREE_OBJECTID);
1310
1311                 if (*bits & EXTENT_FIRST_DELALLOC)
1312                         *bits &= ~EXTENT_FIRST_DELALLOC;
1313                 else
1314                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1315
1316                 spin_lock(&root->fs_info->delalloc_lock);
1317                 BTRFS_I(inode)->delalloc_bytes += len;
1318                 root->fs_info->delalloc_bytes += len;
1319                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1320                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1321                                       &root->fs_info->delalloc_inodes);
1322                 }
1323                 spin_unlock(&root->fs_info->delalloc_lock);
1324         }
1325         return 0;
1326 }
1327
1328 /*
1329  * extent_io.c clear_bit_hook, see set_bit_hook for why
1330  */
1331 static int btrfs_clear_bit_hook(struct inode *inode,
1332                                 struct extent_state *state, int *bits)
1333 {
1334         /*
1335          * set_bit and clear bit hooks normally require _irqsave/restore
1336          * but in this case, we are only testeing for the DELALLOC
1337          * bit, which is only set or cleared with irqs on
1338          */
1339         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1340                 struct btrfs_root *root = BTRFS_I(inode)->root;
1341                 u64 len = state->end + 1 - state->start;
1342                 int do_list = (root->root_key.objectid !=
1343                                BTRFS_ROOT_TREE_OBJECTID);
1344
1345                 if (*bits & EXTENT_FIRST_DELALLOC)
1346                         *bits &= ~EXTENT_FIRST_DELALLOC;
1347                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1348                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1349
1350                 if (*bits & EXTENT_DO_ACCOUNTING)
1351                         btrfs_delalloc_release_metadata(inode, len);
1352
1353                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1354                     && do_list)
1355                         btrfs_free_reserved_data_space(inode, len);
1356
1357                 spin_lock(&root->fs_info->delalloc_lock);
1358                 root->fs_info->delalloc_bytes -= len;
1359                 BTRFS_I(inode)->delalloc_bytes -= len;
1360
1361                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1362                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1363                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1364                 }
1365                 spin_unlock(&root->fs_info->delalloc_lock);
1366         }
1367         return 0;
1368 }
1369
1370 /*
1371  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1372  * we don't create bios that span stripes or chunks
1373  */
1374 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1375                          size_t size, struct bio *bio,
1376                          unsigned long bio_flags)
1377 {
1378         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1379         struct btrfs_mapping_tree *map_tree;
1380         u64 logical = (u64)bio->bi_sector << 9;
1381         u64 length = 0;
1382         u64 map_length;
1383         int ret;
1384
1385         if (bio_flags & EXTENT_BIO_COMPRESSED)
1386                 return 0;
1387
1388         length = bio->bi_size;
1389         map_tree = &root->fs_info->mapping_tree;
1390         map_length = length;
1391         ret = btrfs_map_block(map_tree, READ, logical,
1392                               &map_length, NULL, 0);
1393
1394         if (map_length < length + size)
1395                 return 1;
1396         return ret;
1397 }
1398
1399 /*
1400  * in order to insert checksums into the metadata in large chunks,
1401  * we wait until bio submission time.   All the pages in the bio are
1402  * checksummed and sums are attached onto the ordered extent record.
1403  *
1404  * At IO completion time the cums attached on the ordered extent record
1405  * are inserted into the btree
1406  */
1407 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1408                                     struct bio *bio, int mirror_num,
1409                                     unsigned long bio_flags,
1410                                     u64 bio_offset)
1411 {
1412         struct btrfs_root *root = BTRFS_I(inode)->root;
1413         int ret = 0;
1414
1415         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1416         BUG_ON(ret);
1417         return 0;
1418 }
1419
1420 /*
1421  * in order to insert checksums into the metadata in large chunks,
1422  * we wait until bio submission time.   All the pages in the bio are
1423  * checksummed and sums are attached onto the ordered extent record.
1424  *
1425  * At IO completion time the cums attached on the ordered extent record
1426  * are inserted into the btree
1427  */
1428 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1429                           int mirror_num, unsigned long bio_flags,
1430                           u64 bio_offset)
1431 {
1432         struct btrfs_root *root = BTRFS_I(inode)->root;
1433         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1434 }
1435
1436 /*
1437  * extent_io.c submission hook. This does the right thing for csum calculation
1438  * on write, or reading the csums from the tree before a read
1439  */
1440 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1441                           int mirror_num, unsigned long bio_flags,
1442                           u64 bio_offset)
1443 {
1444         struct btrfs_root *root = BTRFS_I(inode)->root;
1445         int ret = 0;
1446         int skip_sum;
1447
1448         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1449
1450         if (root == root->fs_info->tree_root)
1451                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1452         else
1453                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1454         BUG_ON(ret);
1455
1456         if (!(rw & REQ_WRITE)) {
1457                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1458                         return btrfs_submit_compressed_read(inode, bio,
1459                                                     mirror_num, bio_flags);
1460                 } else if (!skip_sum)
1461                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1462                 goto mapit;
1463         } else if (!skip_sum) {
1464                 /* csum items have already been cloned */
1465                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1466                         goto mapit;
1467                 /* we're doing a write, do the async checksumming */
1468                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1469                                    inode, rw, bio, mirror_num,
1470                                    bio_flags, bio_offset,
1471                                    __btrfs_submit_bio_start,
1472                                    __btrfs_submit_bio_done);
1473         }
1474
1475 mapit:
1476         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1477 }
1478
1479 /*
1480  * given a list of ordered sums record them in the inode.  This happens
1481  * at IO completion time based on sums calculated at bio submission time.
1482  */
1483 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1484                              struct inode *inode, u64 file_offset,
1485                              struct list_head *list)
1486 {
1487         struct btrfs_ordered_sum *sum;
1488
1489         btrfs_set_trans_block_group(trans, inode);
1490
1491         list_for_each_entry(sum, list, list) {
1492                 btrfs_csum_file_blocks(trans,
1493                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1494         }
1495         return 0;
1496 }
1497
1498 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1499                               struct extent_state **cached_state)
1500 {
1501         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1502                 WARN_ON(1);
1503         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1504                                    cached_state, GFP_NOFS);
1505 }
1506
1507 /* see btrfs_writepage_start_hook for details on why this is required */
1508 struct btrfs_writepage_fixup {
1509         struct page *page;
1510         struct btrfs_work work;
1511 };
1512
1513 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1514 {
1515         struct btrfs_writepage_fixup *fixup;
1516         struct btrfs_ordered_extent *ordered;
1517         struct extent_state *cached_state = NULL;
1518         struct page *page;
1519         struct inode *inode;
1520         u64 page_start;
1521         u64 page_end;
1522
1523         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1524         page = fixup->page;
1525 again:
1526         lock_page(page);
1527         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1528                 ClearPageChecked(page);
1529                 goto out_page;
1530         }
1531
1532         inode = page->mapping->host;
1533         page_start = page_offset(page);
1534         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1535
1536         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1537                          &cached_state, GFP_NOFS);
1538
1539         /* already ordered? We're done */
1540         if (PagePrivate2(page))
1541                 goto out;
1542
1543         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1544         if (ordered) {
1545                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1546                                      page_end, &cached_state, GFP_NOFS);
1547                 unlock_page(page);
1548                 btrfs_start_ordered_extent(inode, ordered, 1);
1549                 goto again;
1550         }
1551
1552         BUG();
1553         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1554         ClearPageChecked(page);
1555 out:
1556         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1557                              &cached_state, GFP_NOFS);
1558 out_page:
1559         unlock_page(page);
1560         page_cache_release(page);
1561         kfree(fixup);
1562 }
1563
1564 /*
1565  * There are a few paths in the higher layers of the kernel that directly
1566  * set the page dirty bit without asking the filesystem if it is a
1567  * good idea.  This causes problems because we want to make sure COW
1568  * properly happens and the data=ordered rules are followed.
1569  *
1570  * In our case any range that doesn't have the ORDERED bit set
1571  * hasn't been properly setup for IO.  We kick off an async process
1572  * to fix it up.  The async helper will wait for ordered extents, set
1573  * the delalloc bit and make it safe to write the page.
1574  */
1575 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1576 {
1577         struct inode *inode = page->mapping->host;
1578         struct btrfs_writepage_fixup *fixup;
1579         struct btrfs_root *root = BTRFS_I(inode)->root;
1580
1581         /* this page is properly in the ordered list */
1582         if (TestClearPagePrivate2(page))
1583                 return 0;
1584
1585         if (PageChecked(page))
1586                 return -EAGAIN;
1587
1588         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1589         if (!fixup)
1590                 return -EAGAIN;
1591
1592         SetPageChecked(page);
1593         page_cache_get(page);
1594         fixup->work.func = btrfs_writepage_fixup_worker;
1595         fixup->page = page;
1596         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1597         return -EAGAIN;
1598 }
1599
1600 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1601                                        struct inode *inode, u64 file_pos,
1602                                        u64 disk_bytenr, u64 disk_num_bytes,
1603                                        u64 num_bytes, u64 ram_bytes,
1604                                        u8 compression, u8 encryption,
1605                                        u16 other_encoding, int extent_type)
1606 {
1607         struct btrfs_root *root = BTRFS_I(inode)->root;
1608         struct btrfs_file_extent_item *fi;
1609         struct btrfs_path *path;
1610         struct extent_buffer *leaf;
1611         struct btrfs_key ins;
1612         u64 hint;
1613         int ret;
1614
1615         path = btrfs_alloc_path();
1616         BUG_ON(!path);
1617
1618         path->leave_spinning = 1;
1619
1620         /*
1621          * we may be replacing one extent in the tree with another.
1622          * The new extent is pinned in the extent map, and we don't want
1623          * to drop it from the cache until it is completely in the btree.
1624          *
1625          * So, tell btrfs_drop_extents to leave this extent in the cache.
1626          * the caller is expected to unpin it and allow it to be merged
1627          * with the others.
1628          */
1629         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1630                                  &hint, 0);
1631         BUG_ON(ret);
1632
1633         ins.objectid = inode->i_ino;
1634         ins.offset = file_pos;
1635         ins.type = BTRFS_EXTENT_DATA_KEY;
1636         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1637         BUG_ON(ret);
1638         leaf = path->nodes[0];
1639         fi = btrfs_item_ptr(leaf, path->slots[0],
1640                             struct btrfs_file_extent_item);
1641         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1642         btrfs_set_file_extent_type(leaf, fi, extent_type);
1643         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1644         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1645         btrfs_set_file_extent_offset(leaf, fi, 0);
1646         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1647         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1648         btrfs_set_file_extent_compression(leaf, fi, compression);
1649         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1650         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1651
1652         btrfs_unlock_up_safe(path, 1);
1653         btrfs_set_lock_blocking(leaf);
1654
1655         btrfs_mark_buffer_dirty(leaf);
1656
1657         inode_add_bytes(inode, num_bytes);
1658
1659         ins.objectid = disk_bytenr;
1660         ins.offset = disk_num_bytes;
1661         ins.type = BTRFS_EXTENT_ITEM_KEY;
1662         ret = btrfs_alloc_reserved_file_extent(trans, root,
1663                                         root->root_key.objectid,
1664                                         inode->i_ino, file_pos, &ins);
1665         BUG_ON(ret);
1666         btrfs_free_path(path);
1667
1668         return 0;
1669 }
1670
1671 /*
1672  * helper function for btrfs_finish_ordered_io, this
1673  * just reads in some of the csum leaves to prime them into ram
1674  * before we start the transaction.  It limits the amount of btree
1675  * reads required while inside the transaction.
1676  */
1677 /* as ordered data IO finishes, this gets called so we can finish
1678  * an ordered extent if the range of bytes in the file it covers are
1679  * fully written.
1680  */
1681 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1682 {
1683         struct btrfs_root *root = BTRFS_I(inode)->root;
1684         struct btrfs_trans_handle *trans = NULL;
1685         struct btrfs_ordered_extent *ordered_extent = NULL;
1686         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1687         struct extent_state *cached_state = NULL;
1688         int compress_type = 0;
1689         int ret;
1690         bool nolock = false;
1691
1692         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1693                                              end - start + 1);
1694         if (!ret)
1695                 return 0;
1696         BUG_ON(!ordered_extent);
1697
1698         nolock = (root == root->fs_info->tree_root);
1699
1700         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1701                 BUG_ON(!list_empty(&ordered_extent->list));
1702                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1703                 if (!ret) {
1704                         if (nolock)
1705                                 trans = btrfs_join_transaction_nolock(root, 1);
1706                         else
1707                                 trans = btrfs_join_transaction(root, 1);
1708                         BUG_ON(IS_ERR(trans));
1709                         btrfs_set_trans_block_group(trans, inode);
1710                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1711                         ret = btrfs_update_inode(trans, root, inode);
1712                         BUG_ON(ret);
1713                 }
1714                 goto out;
1715         }
1716
1717         lock_extent_bits(io_tree, ordered_extent->file_offset,
1718                          ordered_extent->file_offset + ordered_extent->len - 1,
1719                          0, &cached_state, GFP_NOFS);
1720
1721         if (nolock)
1722                 trans = btrfs_join_transaction_nolock(root, 1);
1723         else
1724                 trans = btrfs_join_transaction(root, 1);
1725         BUG_ON(IS_ERR(trans));
1726         btrfs_set_trans_block_group(trans, inode);
1727         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1728
1729         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1730                 compress_type = ordered_extent->compress_type;
1731         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1732                 BUG_ON(compress_type);
1733                 ret = btrfs_mark_extent_written(trans, inode,
1734                                                 ordered_extent->file_offset,
1735                                                 ordered_extent->file_offset +
1736                                                 ordered_extent->len);
1737                 BUG_ON(ret);
1738         } else {
1739                 BUG_ON(root == root->fs_info->tree_root);
1740                 ret = insert_reserved_file_extent(trans, inode,
1741                                                 ordered_extent->file_offset,
1742                                                 ordered_extent->start,
1743                                                 ordered_extent->disk_len,
1744                                                 ordered_extent->len,
1745                                                 ordered_extent->len,
1746                                                 compress_type, 0, 0,
1747                                                 BTRFS_FILE_EXTENT_REG);
1748                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1749                                    ordered_extent->file_offset,
1750                                    ordered_extent->len);
1751                 BUG_ON(ret);
1752         }
1753         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1754                              ordered_extent->file_offset +
1755                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1756
1757         add_pending_csums(trans, inode, ordered_extent->file_offset,
1758                           &ordered_extent->list);
1759
1760         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1761         ret = btrfs_update_inode(trans, root, inode);
1762         BUG_ON(ret);
1763 out:
1764         if (nolock) {
1765                 if (trans)
1766                         btrfs_end_transaction_nolock(trans, root);
1767         } else {
1768                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1769                 if (trans)
1770                         btrfs_end_transaction(trans, root);
1771         }
1772
1773         /* once for us */
1774         btrfs_put_ordered_extent(ordered_extent);
1775         /* once for the tree */
1776         btrfs_put_ordered_extent(ordered_extent);
1777
1778         return 0;
1779 }
1780
1781 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1782                                 struct extent_state *state, int uptodate)
1783 {
1784         ClearPagePrivate2(page);
1785         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1786 }
1787
1788 /*
1789  * When IO fails, either with EIO or csum verification fails, we
1790  * try other mirrors that might have a good copy of the data.  This
1791  * io_failure_record is used to record state as we go through all the
1792  * mirrors.  If another mirror has good data, the page is set up to date
1793  * and things continue.  If a good mirror can't be found, the original
1794  * bio end_io callback is called to indicate things have failed.
1795  */
1796 struct io_failure_record {
1797         struct page *page;
1798         u64 start;
1799         u64 len;
1800         u64 logical;
1801         unsigned long bio_flags;
1802         int last_mirror;
1803 };
1804
1805 static int btrfs_io_failed_hook(struct bio *failed_bio,
1806                          struct page *page, u64 start, u64 end,
1807                          struct extent_state *state)
1808 {
1809         struct io_failure_record *failrec = NULL;
1810         u64 private;
1811         struct extent_map *em;
1812         struct inode *inode = page->mapping->host;
1813         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1814         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1815         struct bio *bio;
1816         int num_copies;
1817         int ret;
1818         int rw;
1819         u64 logical;
1820
1821         ret = get_state_private(failure_tree, start, &private);
1822         if (ret) {
1823                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1824                 if (!failrec)
1825                         return -ENOMEM;
1826                 failrec->start = start;
1827                 failrec->len = end - start + 1;
1828                 failrec->last_mirror = 0;
1829                 failrec->bio_flags = 0;
1830
1831                 read_lock(&em_tree->lock);
1832                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1833                 if (em->start > start || em->start + em->len < start) {
1834                         free_extent_map(em);
1835                         em = NULL;
1836                 }
1837                 read_unlock(&em_tree->lock);
1838
1839                 if (!em || IS_ERR(em)) {
1840                         kfree(failrec);
1841                         return -EIO;
1842                 }
1843                 logical = start - em->start;
1844                 logical = em->block_start + logical;
1845                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1846                         logical = em->block_start;
1847                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1848                         extent_set_compress_type(&failrec->bio_flags,
1849                                                  em->compress_type);
1850                 }
1851                 failrec->logical = logical;
1852                 free_extent_map(em);
1853                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1854                                 EXTENT_DIRTY, GFP_NOFS);
1855                 set_state_private(failure_tree, start,
1856                                  (u64)(unsigned long)failrec);
1857         } else {
1858                 failrec = (struct io_failure_record *)(unsigned long)private;
1859         }
1860         num_copies = btrfs_num_copies(
1861                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1862                               failrec->logical, failrec->len);
1863         failrec->last_mirror++;
1864         if (!state) {
1865                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1866                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1867                                                     failrec->start,
1868                                                     EXTENT_LOCKED);
1869                 if (state && state->start != failrec->start)
1870                         state = NULL;
1871                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1872         }
1873         if (!state || failrec->last_mirror > num_copies) {
1874                 set_state_private(failure_tree, failrec->start, 0);
1875                 clear_extent_bits(failure_tree, failrec->start,
1876                                   failrec->start + failrec->len - 1,
1877                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1878                 kfree(failrec);
1879                 return -EIO;
1880         }
1881         bio = bio_alloc(GFP_NOFS, 1);
1882         bio->bi_private = state;
1883         bio->bi_end_io = failed_bio->bi_end_io;
1884         bio->bi_sector = failrec->logical >> 9;
1885         bio->bi_bdev = failed_bio->bi_bdev;
1886         bio->bi_size = 0;
1887
1888         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1889         if (failed_bio->bi_rw & REQ_WRITE)
1890                 rw = WRITE;
1891         else
1892                 rw = READ;
1893
1894         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1895                                                       failrec->last_mirror,
1896                                                       failrec->bio_flags, 0);
1897         return 0;
1898 }
1899
1900 /*
1901  * each time an IO finishes, we do a fast check in the IO failure tree
1902  * to see if we need to process or clean up an io_failure_record
1903  */
1904 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1905 {
1906         u64 private;
1907         u64 private_failure;
1908         struct io_failure_record *failure;
1909         int ret;
1910
1911         private = 0;
1912         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1913                              (u64)-1, 1, EXTENT_DIRTY)) {
1914                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1915                                         start, &private_failure);
1916                 if (ret == 0) {
1917                         failure = (struct io_failure_record *)(unsigned long)
1918                                    private_failure;
1919                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1920                                           failure->start, 0);
1921                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1922                                           failure->start,
1923                                           failure->start + failure->len - 1,
1924                                           EXTENT_DIRTY | EXTENT_LOCKED,
1925                                           GFP_NOFS);
1926                         kfree(failure);
1927                 }
1928         }
1929         return 0;
1930 }
1931
1932 /*
1933  * when reads are done, we need to check csums to verify the data is correct
1934  * if there's a match, we allow the bio to finish.  If not, we go through
1935  * the io_failure_record routines to find good copies
1936  */
1937 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1938                                struct extent_state *state)
1939 {
1940         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1941         struct inode *inode = page->mapping->host;
1942         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1943         char *kaddr;
1944         u64 private = ~(u32)0;
1945         int ret;
1946         struct btrfs_root *root = BTRFS_I(inode)->root;
1947         u32 csum = ~(u32)0;
1948
1949         if (PageChecked(page)) {
1950                 ClearPageChecked(page);
1951                 goto good;
1952         }
1953
1954         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1955                 return 0;
1956
1957         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1958             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1959                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1960                                   GFP_NOFS);
1961                 return 0;
1962         }
1963
1964         if (state && state->start == start) {
1965                 private = state->private;
1966                 ret = 0;
1967         } else {
1968                 ret = get_state_private(io_tree, start, &private);
1969         }
1970         kaddr = kmap_atomic(page, KM_USER0);
1971         if (ret)
1972                 goto zeroit;
1973
1974         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1975         btrfs_csum_final(csum, (char *)&csum);
1976         if (csum != private)
1977                 goto zeroit;
1978
1979         kunmap_atomic(kaddr, KM_USER0);
1980 good:
1981         /* if the io failure tree for this inode is non-empty,
1982          * check to see if we've recovered from a failed IO
1983          */
1984         btrfs_clean_io_failures(inode, start);
1985         return 0;
1986
1987 zeroit:
1988         if (printk_ratelimit()) {
1989                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1990                        "private %llu\n", page->mapping->host->i_ino,
1991                        (unsigned long long)start, csum,
1992                        (unsigned long long)private);
1993         }
1994         memset(kaddr + offset, 1, end - start + 1);
1995         flush_dcache_page(page);
1996         kunmap_atomic(kaddr, KM_USER0);
1997         if (private == 0)
1998                 return 0;
1999         return -EIO;
2000 }
2001
2002 struct delayed_iput {
2003         struct list_head list;
2004         struct inode *inode;
2005 };
2006
2007 void btrfs_add_delayed_iput(struct inode *inode)
2008 {
2009         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2010         struct delayed_iput *delayed;
2011
2012         if (atomic_add_unless(&inode->i_count, -1, 1))
2013                 return;
2014
2015         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2016         delayed->inode = inode;
2017
2018         spin_lock(&fs_info->delayed_iput_lock);
2019         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2020         spin_unlock(&fs_info->delayed_iput_lock);
2021 }
2022
2023 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2024 {
2025         LIST_HEAD(list);
2026         struct btrfs_fs_info *fs_info = root->fs_info;
2027         struct delayed_iput *delayed;
2028         int empty;
2029
2030         spin_lock(&fs_info->delayed_iput_lock);
2031         empty = list_empty(&fs_info->delayed_iputs);
2032         spin_unlock(&fs_info->delayed_iput_lock);
2033         if (empty)
2034                 return;
2035
2036         down_read(&root->fs_info->cleanup_work_sem);
2037         spin_lock(&fs_info->delayed_iput_lock);
2038         list_splice_init(&fs_info->delayed_iputs, &list);
2039         spin_unlock(&fs_info->delayed_iput_lock);
2040
2041         while (!list_empty(&list)) {
2042                 delayed = list_entry(list.next, struct delayed_iput, list);
2043                 list_del(&delayed->list);
2044                 iput(delayed->inode);
2045                 kfree(delayed);
2046         }
2047         up_read(&root->fs_info->cleanup_work_sem);
2048 }
2049
2050 /*
2051  * calculate extra metadata reservation when snapshotting a subvolume
2052  * contains orphan files.
2053  */
2054 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2055                                 struct btrfs_pending_snapshot *pending,
2056                                 u64 *bytes_to_reserve)
2057 {
2058         struct btrfs_root *root;
2059         struct btrfs_block_rsv *block_rsv;
2060         u64 num_bytes;
2061         int index;
2062
2063         root = pending->root;
2064         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2065                 return;
2066
2067         block_rsv = root->orphan_block_rsv;
2068
2069         /* orphan block reservation for the snapshot */
2070         num_bytes = block_rsv->size;
2071
2072         /*
2073          * after the snapshot is created, COWing tree blocks may use more
2074          * space than it frees. So we should make sure there is enough
2075          * reserved space.
2076          */
2077         index = trans->transid & 0x1;
2078         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2079                 num_bytes += block_rsv->size -
2080                              (block_rsv->reserved + block_rsv->freed[index]);
2081         }
2082
2083         *bytes_to_reserve += num_bytes;
2084 }
2085
2086 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2087                                 struct btrfs_pending_snapshot *pending)
2088 {
2089         struct btrfs_root *root = pending->root;
2090         struct btrfs_root *snap = pending->snap;
2091         struct btrfs_block_rsv *block_rsv;
2092         u64 num_bytes;
2093         int index;
2094         int ret;
2095
2096         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2097                 return;
2098
2099         /* refill source subvolume's orphan block reservation */
2100         block_rsv = root->orphan_block_rsv;
2101         index = trans->transid & 0x1;
2102         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2103                 num_bytes = block_rsv->size -
2104                             (block_rsv->reserved + block_rsv->freed[index]);
2105                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2106                                               root->orphan_block_rsv,
2107                                               num_bytes);
2108                 BUG_ON(ret);
2109         }
2110
2111         /* setup orphan block reservation for the snapshot */
2112         block_rsv = btrfs_alloc_block_rsv(snap);
2113         BUG_ON(!block_rsv);
2114
2115         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2116         snap->orphan_block_rsv = block_rsv;
2117
2118         num_bytes = root->orphan_block_rsv->size;
2119         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2120                                       block_rsv, num_bytes);
2121         BUG_ON(ret);
2122
2123 #if 0
2124         /* insert orphan item for the snapshot */
2125         WARN_ON(!root->orphan_item_inserted);
2126         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2127                                        snap->root_key.objectid);
2128         BUG_ON(ret);
2129         snap->orphan_item_inserted = 1;
2130 #endif
2131 }
2132
2133 enum btrfs_orphan_cleanup_state {
2134         ORPHAN_CLEANUP_STARTED  = 1,
2135         ORPHAN_CLEANUP_DONE     = 2,
2136 };
2137
2138 /*
2139  * This is called in transaction commmit time. If there are no orphan
2140  * files in the subvolume, it removes orphan item and frees block_rsv
2141  * structure.
2142  */
2143 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2144                               struct btrfs_root *root)
2145 {
2146         int ret;
2147
2148         if (!list_empty(&root->orphan_list) ||
2149             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2150                 return;
2151
2152         if (root->orphan_item_inserted &&
2153             btrfs_root_refs(&root->root_item) > 0) {
2154                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2155                                             root->root_key.objectid);
2156                 BUG_ON(ret);
2157                 root->orphan_item_inserted = 0;
2158         }
2159
2160         if (root->orphan_block_rsv) {
2161                 WARN_ON(root->orphan_block_rsv->size > 0);
2162                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2163                 root->orphan_block_rsv = NULL;
2164         }
2165 }
2166
2167 /*
2168  * This creates an orphan entry for the given inode in case something goes
2169  * wrong in the middle of an unlink/truncate.
2170  *
2171  * NOTE: caller of this function should reserve 5 units of metadata for
2172  *       this function.
2173  */
2174 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2175 {
2176         struct btrfs_root *root = BTRFS_I(inode)->root;
2177         struct btrfs_block_rsv *block_rsv = NULL;
2178         int reserve = 0;
2179         int insert = 0;
2180         int ret;
2181
2182         if (!root->orphan_block_rsv) {
2183                 block_rsv = btrfs_alloc_block_rsv(root);
2184                 BUG_ON(!block_rsv);
2185         }
2186
2187         spin_lock(&root->orphan_lock);
2188         if (!root->orphan_block_rsv) {
2189                 root->orphan_block_rsv = block_rsv;
2190         } else if (block_rsv) {
2191                 btrfs_free_block_rsv(root, block_rsv);
2192                 block_rsv = NULL;
2193         }
2194
2195         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2196                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2197 #if 0
2198                 /*
2199                  * For proper ENOSPC handling, we should do orphan
2200                  * cleanup when mounting. But this introduces backward
2201                  * compatibility issue.
2202                  */
2203                 if (!xchg(&root->orphan_item_inserted, 1))
2204                         insert = 2;
2205                 else
2206                         insert = 1;
2207 #endif
2208                 insert = 1;
2209         } else {
2210                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2211         }
2212
2213         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2214                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2215                 reserve = 1;
2216         }
2217         spin_unlock(&root->orphan_lock);
2218
2219         if (block_rsv)
2220                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2221
2222         /* grab metadata reservation from transaction handle */
2223         if (reserve) {
2224                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2225                 BUG_ON(ret);
2226         }
2227
2228         /* insert an orphan item to track this unlinked/truncated file */
2229         if (insert >= 1) {
2230                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2231                 BUG_ON(ret);
2232         }
2233
2234         /* insert an orphan item to track subvolume contains orphan files */
2235         if (insert >= 2) {
2236                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2237                                                root->root_key.objectid);
2238                 BUG_ON(ret);
2239         }
2240         return 0;
2241 }
2242
2243 /*
2244  * We have done the truncate/delete so we can go ahead and remove the orphan
2245  * item for this particular inode.
2246  */
2247 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2248 {
2249         struct btrfs_root *root = BTRFS_I(inode)->root;
2250         int delete_item = 0;
2251         int release_rsv = 0;
2252         int ret = 0;
2253
2254         spin_lock(&root->orphan_lock);
2255         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2256                 list_del_init(&BTRFS_I(inode)->i_orphan);
2257                 delete_item = 1;
2258         }
2259
2260         if (BTRFS_I(inode)->orphan_meta_reserved) {
2261                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2262                 release_rsv = 1;
2263         }
2264         spin_unlock(&root->orphan_lock);
2265
2266         if (trans && delete_item) {
2267                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2268                 BUG_ON(ret);
2269         }
2270
2271         if (release_rsv)
2272                 btrfs_orphan_release_metadata(inode);
2273
2274         return 0;
2275 }
2276
2277 /*
2278  * this cleans up any orphans that may be left on the list from the last use
2279  * of this root.
2280  */
2281 void btrfs_orphan_cleanup(struct btrfs_root *root)
2282 {
2283         struct btrfs_path *path;
2284         struct extent_buffer *leaf;
2285         struct btrfs_key key, found_key;
2286         struct btrfs_trans_handle *trans;
2287         struct inode *inode;
2288         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2289
2290         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2291                 return;
2292
2293         path = btrfs_alloc_path();
2294         BUG_ON(!path);
2295         path->reada = -1;
2296
2297         key.objectid = BTRFS_ORPHAN_OBJECTID;
2298         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2299         key.offset = (u64)-1;
2300
2301         while (1) {
2302                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2303                 if (ret < 0) {
2304                         printk(KERN_ERR "Error searching slot for orphan: %d"
2305                                "\n", ret);
2306                         break;
2307                 }
2308
2309                 /*
2310                  * if ret == 0 means we found what we were searching for, which
2311                  * is weird, but possible, so only screw with path if we didnt
2312                  * find the key and see if we have stuff that matches
2313                  */
2314                 if (ret > 0) {
2315                         if (path->slots[0] == 0)
2316                                 break;
2317                         path->slots[0]--;
2318                 }
2319
2320                 /* pull out the item */
2321                 leaf = path->nodes[0];
2322                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2323
2324                 /* make sure the item matches what we want */
2325                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2326                         break;
2327                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2328                         break;
2329
2330                 /* release the path since we're done with it */
2331                 btrfs_release_path(root, path);
2332
2333                 /*
2334                  * this is where we are basically btrfs_lookup, without the
2335                  * crossing root thing.  we store the inode number in the
2336                  * offset of the orphan item.
2337                  */
2338                 found_key.objectid = found_key.offset;
2339                 found_key.type = BTRFS_INODE_ITEM_KEY;
2340                 found_key.offset = 0;
2341                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2342                 BUG_ON(IS_ERR(inode));
2343
2344                 /*
2345                  * add this inode to the orphan list so btrfs_orphan_del does
2346                  * the proper thing when we hit it
2347                  */
2348                 spin_lock(&root->orphan_lock);
2349                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2350                 spin_unlock(&root->orphan_lock);
2351
2352                 /*
2353                  * if this is a bad inode, means we actually succeeded in
2354                  * removing the inode, but not the orphan record, which means
2355                  * we need to manually delete the orphan since iput will just
2356                  * do a destroy_inode
2357                  */
2358                 if (is_bad_inode(inode)) {
2359                         trans = btrfs_start_transaction(root, 0);
2360                         btrfs_orphan_del(trans, inode);
2361                         btrfs_end_transaction(trans, root);
2362                         iput(inode);
2363                         continue;
2364                 }
2365
2366                 /* if we have links, this was a truncate, lets do that */
2367                 if (inode->i_nlink) {
2368                         nr_truncate++;
2369                         btrfs_truncate(inode);
2370                 } else {
2371                         nr_unlink++;
2372                 }
2373
2374                 /* this will do delete_inode and everything for us */
2375                 iput(inode);
2376         }
2377         btrfs_free_path(path);
2378
2379         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2380
2381         if (root->orphan_block_rsv)
2382                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2383                                         (u64)-1);
2384
2385         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2386                 trans = btrfs_join_transaction(root, 1);
2387                 BUG_ON(IS_ERR(trans));
2388                 btrfs_end_transaction(trans, root);
2389         }
2390
2391         if (nr_unlink)
2392                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2393         if (nr_truncate)
2394                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2395 }
2396
2397 /*
2398  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2399  * don't find any xattrs, we know there can't be any acls.
2400  *
2401  * slot is the slot the inode is in, objectid is the objectid of the inode
2402  */
2403 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2404                                           int slot, u64 objectid)
2405 {
2406         u32 nritems = btrfs_header_nritems(leaf);
2407         struct btrfs_key found_key;
2408         int scanned = 0;
2409
2410         slot++;
2411         while (slot < nritems) {
2412                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2413
2414                 /* we found a different objectid, there must not be acls */
2415                 if (found_key.objectid != objectid)
2416                         return 0;
2417
2418                 /* we found an xattr, assume we've got an acl */
2419                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2420                         return 1;
2421
2422                 /*
2423                  * we found a key greater than an xattr key, there can't
2424                  * be any acls later on
2425                  */
2426                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2427                         return 0;
2428
2429                 slot++;
2430                 scanned++;
2431
2432                 /*
2433                  * it goes inode, inode backrefs, xattrs, extents,
2434                  * so if there are a ton of hard links to an inode there can
2435                  * be a lot of backrefs.  Don't waste time searching too hard,
2436                  * this is just an optimization
2437                  */
2438                 if (scanned >= 8)
2439                         break;
2440         }
2441         /* we hit the end of the leaf before we found an xattr or
2442          * something larger than an xattr.  We have to assume the inode
2443          * has acls
2444          */
2445         return 1;
2446 }
2447
2448 /*
2449  * read an inode from the btree into the in-memory inode
2450  */
2451 static void btrfs_read_locked_inode(struct inode *inode)
2452 {
2453         struct btrfs_path *path;
2454         struct extent_buffer *leaf;
2455         struct btrfs_inode_item *inode_item;
2456         struct btrfs_timespec *tspec;
2457         struct btrfs_root *root = BTRFS_I(inode)->root;
2458         struct btrfs_key location;
2459         int maybe_acls;
2460         u64 alloc_group_block;
2461         u32 rdev;
2462         int ret;
2463
2464         path = btrfs_alloc_path();
2465         BUG_ON(!path);
2466         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2467
2468         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2469         if (ret)
2470                 goto make_bad;
2471
2472         leaf = path->nodes[0];
2473         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2474                                     struct btrfs_inode_item);
2475
2476         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2477         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2478         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2479         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2480         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2481
2482         tspec = btrfs_inode_atime(inode_item);
2483         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2484         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2485
2486         tspec = btrfs_inode_mtime(inode_item);
2487         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2488         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2489
2490         tspec = btrfs_inode_ctime(inode_item);
2491         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2492         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2493
2494         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2495         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2496         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2497         inode->i_generation = BTRFS_I(inode)->generation;
2498         inode->i_rdev = 0;
2499         rdev = btrfs_inode_rdev(leaf, inode_item);
2500
2501         BTRFS_I(inode)->index_cnt = (u64)-1;
2502         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2503
2504         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2505
2506         /*
2507          * try to precache a NULL acl entry for files that don't have
2508          * any xattrs or acls
2509          */
2510         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2511         if (!maybe_acls)
2512                 cache_no_acl(inode);
2513
2514         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2515                                                 alloc_group_block, 0);
2516         btrfs_free_path(path);
2517         inode_item = NULL;
2518
2519         switch (inode->i_mode & S_IFMT) {
2520         case S_IFREG:
2521                 inode->i_mapping->a_ops = &btrfs_aops;
2522                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2523                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2524                 inode->i_fop = &btrfs_file_operations;
2525                 inode->i_op = &btrfs_file_inode_operations;
2526                 break;
2527         case S_IFDIR:
2528                 inode->i_fop = &btrfs_dir_file_operations;
2529                 if (root == root->fs_info->tree_root)
2530                         inode->i_op = &btrfs_dir_ro_inode_operations;
2531                 else
2532                         inode->i_op = &btrfs_dir_inode_operations;
2533                 break;
2534         case S_IFLNK:
2535                 inode->i_op = &btrfs_symlink_inode_operations;
2536                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2537                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2538                 break;
2539         default:
2540                 inode->i_op = &btrfs_special_inode_operations;
2541                 init_special_inode(inode, inode->i_mode, rdev);
2542                 break;
2543         }
2544
2545         btrfs_update_iflags(inode);
2546         return;
2547
2548 make_bad:
2549         btrfs_free_path(path);
2550         make_bad_inode(inode);
2551 }
2552
2553 /*
2554  * given a leaf and an inode, copy the inode fields into the leaf
2555  */
2556 static void fill_inode_item(struct btrfs_trans_handle *trans,
2557                             struct extent_buffer *leaf,
2558                             struct btrfs_inode_item *item,
2559                             struct inode *inode)
2560 {
2561         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2562         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2563         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2564         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2565         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2566
2567         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2568                                inode->i_atime.tv_sec);
2569         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2570                                 inode->i_atime.tv_nsec);
2571
2572         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2573                                inode->i_mtime.tv_sec);
2574         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2575                                 inode->i_mtime.tv_nsec);
2576
2577         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2578                                inode->i_ctime.tv_sec);
2579         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2580                                 inode->i_ctime.tv_nsec);
2581
2582         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2583         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2584         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2585         btrfs_set_inode_transid(leaf, item, trans->transid);
2586         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2587         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2588         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2589 }
2590
2591 /*
2592  * copy everything in the in-memory inode into the btree.
2593  */
2594 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2595                                 struct btrfs_root *root, struct inode *inode)
2596 {
2597         struct btrfs_inode_item *inode_item;
2598         struct btrfs_path *path;
2599         struct extent_buffer *leaf;
2600         int ret;
2601
2602         path = btrfs_alloc_path();
2603         BUG_ON(!path);
2604         path->leave_spinning = 1;
2605         ret = btrfs_lookup_inode(trans, root, path,
2606                                  &BTRFS_I(inode)->location, 1);
2607         if (ret) {
2608                 if (ret > 0)
2609                         ret = -ENOENT;
2610                 goto failed;
2611         }
2612
2613         btrfs_unlock_up_safe(path, 1);
2614         leaf = path->nodes[0];
2615         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2616                                   struct btrfs_inode_item);
2617
2618         fill_inode_item(trans, leaf, inode_item, inode);
2619         btrfs_mark_buffer_dirty(leaf);
2620         btrfs_set_inode_last_trans(trans, inode);
2621         ret = 0;
2622 failed:
2623         btrfs_free_path(path);
2624         return ret;
2625 }
2626
2627
2628 /*
2629  * unlink helper that gets used here in inode.c and in the tree logging
2630  * recovery code.  It remove a link in a directory with a given name, and
2631  * also drops the back refs in the inode to the directory
2632  */
2633 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2634                        struct btrfs_root *root,
2635                        struct inode *dir, struct inode *inode,
2636                        const char *name, int name_len)
2637 {
2638         struct btrfs_path *path;
2639         int ret = 0;
2640         struct extent_buffer *leaf;
2641         struct btrfs_dir_item *di;
2642         struct btrfs_key key;
2643         u64 index;
2644
2645         path = btrfs_alloc_path();
2646         if (!path) {
2647                 ret = -ENOMEM;
2648                 goto err;
2649         }
2650
2651         path->leave_spinning = 1;
2652         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2653                                     name, name_len, -1);
2654         if (IS_ERR(di)) {
2655                 ret = PTR_ERR(di);
2656                 goto err;
2657         }
2658         if (!di) {
2659                 ret = -ENOENT;
2660                 goto err;
2661         }
2662         leaf = path->nodes[0];
2663         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2664         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2665         if (ret)
2666                 goto err;
2667         btrfs_release_path(root, path);
2668
2669         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2670                                   inode->i_ino,
2671                                   dir->i_ino, &index);
2672         if (ret) {
2673                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2674                        "inode %lu parent %lu\n", name_len, name,
2675                        inode->i_ino, dir->i_ino);
2676                 goto err;
2677         }
2678
2679         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2680                                          index, name, name_len, -1);
2681         if (IS_ERR(di)) {
2682                 ret = PTR_ERR(di);
2683                 goto err;
2684         }
2685         if (!di) {
2686                 ret = -ENOENT;
2687                 goto err;
2688         }
2689         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2690         btrfs_release_path(root, path);
2691
2692         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2693                                          inode, dir->i_ino);
2694         BUG_ON(ret != 0 && ret != -ENOENT);
2695
2696         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2697                                            dir, index);
2698         if (ret == -ENOENT)
2699                 ret = 0;
2700 err:
2701         btrfs_free_path(path);
2702         if (ret)
2703                 goto out;
2704
2705         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2706         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2707         btrfs_update_inode(trans, root, dir);
2708         btrfs_drop_nlink(inode);
2709         ret = btrfs_update_inode(trans, root, inode);
2710 out:
2711         return ret;
2712 }
2713
2714 /* helper to check if there is any shared block in the path */
2715 static int check_path_shared(struct btrfs_root *root,
2716                              struct btrfs_path *path)
2717 {
2718         struct extent_buffer *eb;
2719         int level;
2720         u64 refs = 1;
2721         int uninitialized_var(ret);
2722
2723         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2724                 if (!path->nodes[level])
2725                         break;
2726                 eb = path->nodes[level];
2727                 if (!btrfs_block_can_be_shared(root, eb))
2728                         continue;
2729                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2730                                                &refs, NULL);
2731                 if (refs > 1)
2732                         return 1;
2733         }
2734         return ret; /* XXX callers? */
2735 }
2736
2737 /*
2738  * helper to start transaction for unlink and rmdir.
2739  *
2740  * unlink and rmdir are special in btrfs, they do not always free space.
2741  * so in enospc case, we should make sure they will free space before
2742  * allowing them to use the global metadata reservation.
2743  */
2744 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2745                                                        struct dentry *dentry)
2746 {
2747         struct btrfs_trans_handle *trans;
2748         struct btrfs_root *root = BTRFS_I(dir)->root;
2749         struct btrfs_path *path;
2750         struct btrfs_inode_ref *ref;
2751         struct btrfs_dir_item *di;
2752         struct inode *inode = dentry->d_inode;
2753         u64 index;
2754         int check_link = 1;
2755         int err = -ENOSPC;
2756         int ret;
2757
2758         trans = btrfs_start_transaction(root, 10);
2759         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2760                 return trans;
2761
2762         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2763                 return ERR_PTR(-ENOSPC);
2764
2765         /* check if there is someone else holds reference */
2766         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2767                 return ERR_PTR(-ENOSPC);
2768
2769         if (atomic_read(&inode->i_count) > 2)
2770                 return ERR_PTR(-ENOSPC);
2771
2772         if (xchg(&root->fs_info->enospc_unlink, 1))
2773                 return ERR_PTR(-ENOSPC);
2774
2775         path = btrfs_alloc_path();
2776         if (!path) {
2777                 root->fs_info->enospc_unlink = 0;
2778                 return ERR_PTR(-ENOMEM);
2779         }
2780
2781         trans = btrfs_start_transaction(root, 0);
2782         if (IS_ERR(trans)) {
2783                 btrfs_free_path(path);
2784                 root->fs_info->enospc_unlink = 0;
2785                 return trans;
2786         }
2787
2788         path->skip_locking = 1;
2789         path->search_commit_root = 1;
2790
2791         ret = btrfs_lookup_inode(trans, root, path,
2792                                 &BTRFS_I(dir)->location, 0);
2793         if (ret < 0) {
2794                 err = ret;
2795                 goto out;
2796         }
2797         if (ret == 0) {
2798                 if (check_path_shared(root, path))
2799                         goto out;
2800         } else {
2801                 check_link = 0;
2802         }
2803         btrfs_release_path(root, path);
2804
2805         ret = btrfs_lookup_inode(trans, root, path,
2806                                 &BTRFS_I(inode)->location, 0);
2807         if (ret < 0) {
2808                 err = ret;
2809                 goto out;
2810         }
2811         if (ret == 0) {
2812                 if (check_path_shared(root, path))
2813                         goto out;
2814         } else {
2815                 check_link = 0;
2816         }
2817         btrfs_release_path(root, path);
2818
2819         if (ret == 0 && S_ISREG(inode->i_mode)) {
2820                 ret = btrfs_lookup_file_extent(trans, root, path,
2821                                                inode->i_ino, (u64)-1, 0);
2822                 if (ret < 0) {
2823                         err = ret;
2824                         goto out;
2825                 }
2826                 BUG_ON(ret == 0);
2827                 if (check_path_shared(root, path))
2828                         goto out;
2829                 btrfs_release_path(root, path);
2830         }
2831
2832         if (!check_link) {
2833                 err = 0;
2834                 goto out;
2835         }
2836
2837         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2838                                 dentry->d_name.name, dentry->d_name.len, 0);
2839         if (IS_ERR(di)) {
2840                 err = PTR_ERR(di);
2841                 goto out;
2842         }
2843         if (di) {
2844                 if (check_path_shared(root, path))
2845                         goto out;
2846         } else {
2847                 err = 0;
2848                 goto out;
2849         }
2850         btrfs_release_path(root, path);
2851
2852         ref = btrfs_lookup_inode_ref(trans, root, path,
2853                                 dentry->d_name.name, dentry->d_name.len,
2854                                 inode->i_ino, dir->i_ino, 0);
2855         if (IS_ERR(ref)) {
2856                 err = PTR_ERR(ref);
2857                 goto out;
2858         }
2859         BUG_ON(!ref);
2860         if (check_path_shared(root, path))
2861                 goto out;
2862         index = btrfs_inode_ref_index(path->nodes[0], ref);
2863         btrfs_release_path(root, path);
2864
2865         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2866                                 dentry->d_name.name, dentry->d_name.len, 0);
2867         if (IS_ERR(di)) {
2868                 err = PTR_ERR(di);
2869                 goto out;
2870         }
2871         BUG_ON(ret == -ENOENT);
2872         if (check_path_shared(root, path))
2873                 goto out;
2874
2875         err = 0;
2876 out:
2877         btrfs_free_path(path);
2878         if (err) {
2879                 btrfs_end_transaction(trans, root);
2880                 root->fs_info->enospc_unlink = 0;
2881                 return ERR_PTR(err);
2882         }
2883
2884         trans->block_rsv = &root->fs_info->global_block_rsv;
2885         return trans;
2886 }
2887
2888 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2889                                struct btrfs_root *root)
2890 {
2891         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2892                 BUG_ON(!root->fs_info->enospc_unlink);
2893                 root->fs_info->enospc_unlink = 0;
2894         }
2895         btrfs_end_transaction_throttle(trans, root);
2896 }
2897
2898 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2899 {
2900         struct btrfs_root *root = BTRFS_I(dir)->root;
2901         struct btrfs_trans_handle *trans;
2902         struct inode *inode = dentry->d_inode;
2903         int ret;
2904         unsigned long nr = 0;
2905
2906         trans = __unlink_start_trans(dir, dentry);
2907         if (IS_ERR(trans))
2908                 return PTR_ERR(trans);
2909
2910         btrfs_set_trans_block_group(trans, dir);
2911
2912         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2913
2914         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2915                                  dentry->d_name.name, dentry->d_name.len);
2916         BUG_ON(ret);
2917
2918         if (inode->i_nlink == 0) {
2919                 ret = btrfs_orphan_add(trans, inode);
2920                 BUG_ON(ret);
2921         }
2922
2923         nr = trans->blocks_used;
2924         __unlink_end_trans(trans, root);
2925         btrfs_btree_balance_dirty(root, nr);
2926         return ret;
2927 }
2928
2929 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2930                         struct btrfs_root *root,
2931                         struct inode *dir, u64 objectid,
2932                         const char *name, int name_len)
2933 {
2934         struct btrfs_path *path;
2935         struct extent_buffer *leaf;
2936         struct btrfs_dir_item *di;
2937         struct btrfs_key key;
2938         u64 index;
2939         int ret;
2940
2941         path = btrfs_alloc_path();
2942         if (!path)
2943                 return -ENOMEM;
2944
2945         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2946                                    name, name_len, -1);
2947         BUG_ON(!di || IS_ERR(di));
2948
2949         leaf = path->nodes[0];
2950         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2951         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2952         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2953         BUG_ON(ret);
2954         btrfs_release_path(root, path);
2955
2956         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2957                                  objectid, root->root_key.objectid,
2958                                  dir->i_ino, &index, name, name_len);
2959         if (ret < 0) {
2960                 BUG_ON(ret != -ENOENT);
2961                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2962                                                  name, name_len);
2963                 BUG_ON(!di || IS_ERR(di));
2964
2965                 leaf = path->nodes[0];
2966                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2967                 btrfs_release_path(root, path);
2968                 index = key.offset;
2969         }
2970
2971         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2972                                          index, name, name_len, -1);
2973         BUG_ON(!di || IS_ERR(di));
2974
2975         leaf = path->nodes[0];
2976         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2977         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2978         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2979         BUG_ON(ret);
2980         btrfs_release_path(root, path);
2981
2982         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2983         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2984         ret = btrfs_update_inode(trans, root, dir);
2985         BUG_ON(ret);
2986
2987         btrfs_free_path(path);
2988         return 0;
2989 }
2990
2991 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2992 {
2993         struct inode *inode = dentry->d_inode;
2994         int err = 0;
2995         struct btrfs_root *root = BTRFS_I(dir)->root;
2996         struct btrfs_trans_handle *trans;
2997         unsigned long nr = 0;
2998
2999         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3000             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3001                 return -ENOTEMPTY;
3002
3003         trans = __unlink_start_trans(dir, dentry);
3004         if (IS_ERR(trans))
3005                 return PTR_ERR(trans);
3006
3007         btrfs_set_trans_block_group(trans, dir);
3008
3009         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3010                 err = btrfs_unlink_subvol(trans, root, dir,
3011                                           BTRFS_I(inode)->location.objectid,
3012                                           dentry->d_name.name,
3013                                           dentry->d_name.len);
3014                 goto out;
3015         }
3016
3017         err = btrfs_orphan_add(trans, inode);
3018         if (err)
3019                 goto out;
3020
3021         /* now the directory is empty */
3022         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3023                                  dentry->d_name.name, dentry->d_name.len);
3024         if (!err)
3025                 btrfs_i_size_write(inode, 0);
3026 out:
3027         nr = trans->blocks_used;
3028         __unlink_end_trans(trans, root);
3029         btrfs_btree_balance_dirty(root, nr);
3030
3031         return err;
3032 }
3033
3034 #if 0
3035 /*
3036  * when truncating bytes in a file, it is possible to avoid reading
3037  * the leaves that contain only checksum items.  This can be the
3038  * majority of the IO required to delete a large file, but it must
3039  * be done carefully.
3040  *
3041  * The keys in the level just above the leaves are checked to make sure
3042  * the lowest key in a given leaf is a csum key, and starts at an offset
3043  * after the new  size.
3044  *
3045  * Then the key for the next leaf is checked to make sure it also has
3046  * a checksum item for the same file.  If it does, we know our target leaf
3047  * contains only checksum items, and it can be safely freed without reading
3048  * it.
3049  *
3050  * This is just an optimization targeted at large files.  It may do
3051  * nothing.  It will return 0 unless things went badly.
3052  */
3053 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3054                                      struct btrfs_root *root,
3055                                      struct btrfs_path *path,
3056                                      struct inode *inode, u64 new_size)
3057 {
3058         struct btrfs_key key;
3059         int ret;
3060         int nritems;
3061         struct btrfs_key found_key;
3062         struct btrfs_key other_key;
3063         struct btrfs_leaf_ref *ref;
3064         u64 leaf_gen;
3065         u64 leaf_start;
3066
3067         path->lowest_level = 1;
3068         key.objectid = inode->i_ino;
3069         key.type = BTRFS_CSUM_ITEM_KEY;
3070         key.offset = new_size;
3071 again:
3072         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3073         if (ret < 0)
3074                 goto out;
3075
3076         if (path->nodes[1] == NULL) {
3077                 ret = 0;
3078                 goto out;
3079         }
3080         ret = 0;
3081         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3082         nritems = btrfs_header_nritems(path->nodes[1]);
3083
3084         if (!nritems)
3085                 goto out;
3086
3087         if (path->slots[1] >= nritems)
3088                 goto next_node;
3089
3090         /* did we find a key greater than anything we want to delete? */
3091         if (found_key.objectid > inode->i_ino ||
3092            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3093                 goto out;
3094
3095         /* we check the next key in the node to make sure the leave contains
3096          * only checksum items.  This comparison doesn't work if our
3097          * leaf is the last one in the node
3098          */
3099         if (path->slots[1] + 1 >= nritems) {
3100 next_node:
3101                 /* search forward from the last key in the node, this
3102                  * will bring us into the next node in the tree
3103                  */
3104                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3105
3106                 /* unlikely, but we inc below, so check to be safe */
3107                 if (found_key.offset == (u64)-1)
3108                         goto out;
3109
3110                 /* search_forward needs a path with locks held, do the
3111                  * search again for the original key.  It is possible
3112                  * this will race with a balance and return a path that
3113                  * we could modify, but this drop is just an optimization
3114                  * and is allowed to miss some leaves.
3115                  */
3116                 btrfs_release_path(root, path);
3117                 found_key.offset++;
3118
3119                 /* setup a max key for search_forward */
3120                 other_key.offset = (u64)-1;
3121                 other_key.type = key.type;
3122                 other_key.objectid = key.objectid;
3123
3124                 path->keep_locks = 1;
3125                 ret = btrfs_search_forward(root, &found_key, &other_key,
3126                                            path, 0, 0);
3127                 path->keep_locks = 0;
3128                 if (ret || found_key.objectid != key.objectid ||
3129                     found_key.type != key.type) {
3130                         ret = 0;
3131                         goto out;
3132                 }
3133
3134                 key.offset = found_key.offset;
3135                 btrfs_release_path(root, path);
3136                 cond_resched();
3137                 goto again;
3138         }
3139
3140         /* we know there's one more slot after us in the tree,
3141          * read that key so we can verify it is also a checksum item
3142          */
3143         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3144
3145         if (found_key.objectid < inode->i_ino)
3146                 goto next_key;
3147
3148         if (found_key.type != key.type || found_key.offset < new_size)
3149                 goto next_key;
3150
3151         /*
3152          * if the key for the next leaf isn't a csum key from this objectid,
3153          * we can't be sure there aren't good items inside this leaf.
3154          * Bail out
3155          */
3156         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3157                 goto out;
3158
3159         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3160         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3161         /*
3162          * it is safe to delete this leaf, it contains only
3163          * csum items from this inode at an offset >= new_size
3164          */
3165         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3166         BUG_ON(ret);
3167
3168         if (root->ref_cows && leaf_gen < trans->transid) {
3169                 ref = btrfs_alloc_leaf_ref(root, 0);
3170                 if (ref) {
3171                         ref->root_gen = root->root_key.offset;
3172                         ref->bytenr = leaf_start;
3173                         ref->owner = 0;
3174                         ref->generation = leaf_gen;
3175                         ref->nritems = 0;
3176
3177                         btrfs_sort_leaf_ref(ref);
3178
3179                         ret = btrfs_add_leaf_ref(root, ref, 0);
3180                         WARN_ON(ret);
3181                         btrfs_free_leaf_ref(root, ref);
3182                 } else {
3183                         WARN_ON(1);
3184                 }
3185         }
3186 next_key:
3187         btrfs_release_path(root, path);
3188
3189         if (other_key.objectid == inode->i_ino &&
3190             other_key.type == key.type && other_key.offset > key.offset) {
3191                 key.offset = other_key.offset;
3192                 cond_resched();
3193                 goto again;
3194         }
3195         ret = 0;
3196 out:
3197         /* fixup any changes we've made to the path */
3198         path->lowest_level = 0;
3199         path->keep_locks = 0;
3200         btrfs_release_path(root, path);
3201         return ret;
3202 }
3203
3204 #endif
3205
3206 /*
3207  * this can truncate away extent items, csum items and directory items.
3208  * It starts at a high offset and removes keys until it can't find
3209  * any higher than new_size
3210  *
3211  * csum items that cross the new i_size are truncated to the new size
3212  * as well.
3213  *
3214  * min_type is the minimum key type to truncate down to.  If set to 0, this
3215  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3216  */
3217 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3218                                struct btrfs_root *root,
3219                                struct inode *inode,
3220                                u64 new_size, u32 min_type)
3221 {
3222         struct btrfs_path *path;
3223         struct extent_buffer *leaf;
3224         struct btrfs_file_extent_item *fi;
3225         struct btrfs_key key;
3226         struct btrfs_key found_key;
3227         u64 extent_start = 0;
3228         u64 extent_num_bytes = 0;
3229         u64 extent_offset = 0;
3230         u64 item_end = 0;
3231         u64 mask = root->sectorsize - 1;
3232         u32 found_type = (u8)-1;
3233         int found_extent;
3234         int del_item;
3235         int pending_del_nr = 0;
3236         int pending_del_slot = 0;
3237         int extent_type = -1;
3238         int encoding;
3239         int ret;
3240         int err = 0;
3241
3242         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3243
3244         if (root->ref_cows || root == root->fs_info->tree_root)
3245                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3246
3247         path = btrfs_alloc_path();
3248         BUG_ON(!path);
3249         path->reada = -1;
3250
3251         key.objectid = inode->i_ino;
3252         key.offset = (u64)-1;
3253         key.type = (u8)-1;
3254
3255 search_again:
3256         path->leave_spinning = 1;
3257         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3258         if (ret < 0) {
3259                 err = ret;
3260                 goto out;
3261         }
3262
3263         if (ret > 0) {
3264                 /* there are no items in the tree for us to truncate, we're
3265                  * done
3266                  */
3267                 if (path->slots[0] == 0)
3268                         goto out;
3269                 path->slots[0]--;
3270         }
3271
3272         while (1) {
3273                 fi = NULL;
3274                 leaf = path->nodes[0];
3275                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3276                 found_type = btrfs_key_type(&found_key);
3277                 encoding = 0;
3278
3279                 if (found_key.objectid != inode->i_ino)
3280                         break;
3281
3282                 if (found_type < min_type)
3283                         break;
3284
3285                 item_end = found_key.offset;
3286                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3287                         fi = btrfs_item_ptr(leaf, path->slots[0],
3288                                             struct btrfs_file_extent_item);
3289                         extent_type = btrfs_file_extent_type(leaf, fi);
3290                         encoding = btrfs_file_extent_compression(leaf, fi);
3291                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3292                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3293
3294                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3295                                 item_end +=
3296                                     btrfs_file_extent_num_bytes(leaf, fi);
3297                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3298                                 item_end += btrfs_file_extent_inline_len(leaf,
3299                                                                          fi);
3300                         }
3301                         item_end--;
3302                 }
3303                 if (found_type > min_type) {
3304                         del_item = 1;
3305                 } else {
3306                         if (item_end < new_size)
3307                                 break;
3308                         if (found_key.offset >= new_size)
3309                                 del_item = 1;
3310                         else
3311                                 del_item = 0;
3312                 }
3313                 found_extent = 0;
3314                 /* FIXME, shrink the extent if the ref count is only 1 */
3315                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3316                         goto delete;
3317
3318                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3319                         u64 num_dec;
3320                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3321                         if (!del_item && !encoding) {
3322                                 u64 orig_num_bytes =
3323                                         btrfs_file_extent_num_bytes(leaf, fi);
3324                                 extent_num_bytes = new_size -
3325                                         found_key.offset + root->sectorsize - 1;
3326                                 extent_num_bytes = extent_num_bytes &
3327                                         ~((u64)root->sectorsize - 1);
3328                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3329                                                          extent_num_bytes);
3330                                 num_dec = (orig_num_bytes -
3331                                            extent_num_bytes);
3332                                 if (root->ref_cows && extent_start != 0)
3333                                         inode_sub_bytes(inode, num_dec);
3334                                 btrfs_mark_buffer_dirty(leaf);
3335                         } else {
3336                                 extent_num_bytes =
3337                                         btrfs_file_extent_disk_num_bytes(leaf,
3338                                                                          fi);
3339                                 extent_offset = found_key.offset -
3340                                         btrfs_file_extent_offset(leaf, fi);
3341
3342                                 /* FIXME blocksize != 4096 */
3343                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3344                                 if (extent_start != 0) {
3345                                         found_extent = 1;
3346                                         if (root->ref_cows)
3347                                                 inode_sub_bytes(inode, num_dec);
3348                                 }
3349                         }
3350                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3351                         /*
3352                          * we can't truncate inline items that have had
3353                          * special encodings
3354                          */
3355                         if (!del_item &&
3356                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3357                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3358                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3359                                 u32 size = new_size - found_key.offset;
3360
3361                                 if (root->ref_cows) {
3362                                         inode_sub_bytes(inode, item_end + 1 -
3363                                                         new_size);
3364                                 }
3365                                 size =
3366                                     btrfs_file_extent_calc_inline_size(size);
3367                                 ret = btrfs_truncate_item(trans, root, path,
3368                                                           size, 1);
3369                                 BUG_ON(ret);
3370                         } else if (root->ref_cows) {
3371                                 inode_sub_bytes(inode, item_end + 1 -
3372                                                 found_key.offset);
3373                         }
3374                 }
3375 delete:
3376                 if (del_item) {
3377                         if (!pending_del_nr) {
3378                                 /* no pending yet, add ourselves */
3379                                 pending_del_slot = path->slots[0];
3380                                 pending_del_nr = 1;
3381                         } else if (pending_del_nr &&
3382                                    path->slots[0] + 1 == pending_del_slot) {
3383                                 /* hop on the pending chunk */
3384                                 pending_del_nr++;
3385                                 pending_del_slot = path->slots[0];
3386                         } else {
3387                                 BUG();
3388                         }
3389                 } else {
3390                         break;
3391                 }
3392                 if (found_extent && (root->ref_cows ||
3393                                      root == root->fs_info->tree_root)) {
3394                         btrfs_set_path_blocking(path);
3395                         ret = btrfs_free_extent(trans, root, extent_start,
3396                                                 extent_num_bytes, 0,
3397                                                 btrfs_header_owner(leaf),
3398                                                 inode->i_ino, extent_offset);
3399                         BUG_ON(ret);
3400                 }
3401
3402                 if (found_type == BTRFS_INODE_ITEM_KEY)
3403                         break;
3404
3405                 if (path->slots[0] == 0 ||
3406                     path->slots[0] != pending_del_slot) {
3407                         if (root->ref_cows) {
3408                                 err = -EAGAIN;
3409                                 goto out;
3410                         }
3411                         if (pending_del_nr) {
3412                                 ret = btrfs_del_items(trans, root, path,
3413                                                 pending_del_slot,
3414                                                 pending_del_nr);
3415                                 BUG_ON(ret);
3416                                 pending_del_nr = 0;
3417                         }
3418                         btrfs_release_path(root, path);
3419                         goto search_again;
3420                 } else {
3421                         path->slots[0]--;
3422                 }
3423         }
3424 out:
3425         if (pending_del_nr) {
3426                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3427                                       pending_del_nr);
3428                 BUG_ON(ret);
3429         }
3430         btrfs_free_path(path);
3431         return err;
3432 }
3433
3434 /*
3435  * taken from block_truncate_page, but does cow as it zeros out
3436  * any bytes left in the last page in the file.
3437  */
3438 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3439 {
3440         struct inode *inode = mapping->host;
3441         struct btrfs_root *root = BTRFS_I(inode)->root;
3442         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3443         struct btrfs_ordered_extent *ordered;
3444         struct extent_state *cached_state = NULL;
3445         char *kaddr;
3446         u32 blocksize = root->sectorsize;
3447         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3448         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3449         struct page *page;
3450         int ret = 0;
3451         u64 page_start;
3452         u64 page_end;
3453
3454         if ((offset & (blocksize - 1)) == 0)
3455                 goto out;
3456         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3457         if (ret)
3458                 goto out;
3459
3460         ret = -ENOMEM;
3461 again:
3462         page = grab_cache_page(mapping, index);
3463         if (!page) {
3464                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3465                 goto out;
3466         }
3467
3468         page_start = page_offset(page);
3469         page_end = page_start + PAGE_CACHE_SIZE - 1;
3470
3471         if (!PageUptodate(page)) {
3472                 ret = btrfs_readpage(NULL, page);
3473                 lock_page(page);
3474                 if (page->mapping != mapping) {
3475                         unlock_page(page);
3476                         page_cache_release(page);
3477                         goto again;
3478                 }
3479                 if (!PageUptodate(page)) {
3480                         ret = -EIO;
3481                         goto out_unlock;
3482                 }
3483         }
3484         wait_on_page_writeback(page);
3485
3486         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3487                          GFP_NOFS);
3488         set_page_extent_mapped(page);
3489
3490         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3491         if (ordered) {
3492                 unlock_extent_cached(io_tree, page_start, page_end,
3493                                      &cached_state, GFP_NOFS);
3494                 unlock_page(page);
3495                 page_cache_release(page);
3496                 btrfs_start_ordered_extent(inode, ordered, 1);
3497                 btrfs_put_ordered_extent(ordered);
3498                 goto again;
3499         }
3500
3501         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3502                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3503                           0, 0, &cached_state, GFP_NOFS);
3504
3505         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3506                                         &cached_state);
3507         if (ret) {
3508                 unlock_extent_cached(io_tree, page_start, page_end,
3509                                      &cached_state, GFP_NOFS);
3510                 goto out_unlock;
3511         }
3512
3513         ret = 0;
3514         if (offset != PAGE_CACHE_SIZE) {
3515                 kaddr = kmap(page);
3516                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3517                 flush_dcache_page(page);
3518                 kunmap(page);
3519         }
3520         ClearPageChecked(page);
3521         set_page_dirty(page);
3522         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3523                              GFP_NOFS);
3524
3525 out_unlock:
3526         if (ret)
3527                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3528         unlock_page(page);
3529         page_cache_release(page);
3530 out:
3531         return ret;
3532 }
3533
3534 int btrfs_cont_expand(struct inode *inode, loff_t size)
3535 {
3536         struct btrfs_trans_handle *trans;
3537         struct btrfs_root *root = BTRFS_I(inode)->root;
3538         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3539         struct extent_map *em = NULL;
3540         struct extent_state *cached_state = NULL;
3541         u64 mask = root->sectorsize - 1;
3542         u64 hole_start = (inode->i_size + mask) & ~mask;
3543         u64 block_end = (size + mask) & ~mask;
3544         u64 last_byte;
3545         u64 cur_offset;
3546         u64 hole_size;
3547         int err = 0;
3548
3549         if (size <= hole_start)
3550                 return 0;
3551
3552         while (1) {
3553                 struct btrfs_ordered_extent *ordered;
3554                 btrfs_wait_ordered_range(inode, hole_start,
3555                                          block_end - hole_start);
3556                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3557                                  &cached_state, GFP_NOFS);
3558                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3559                 if (!ordered)
3560                         break;
3561                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3562                                      &cached_state, GFP_NOFS);
3563                 btrfs_put_ordered_extent(ordered);
3564         }
3565
3566         cur_offset = hole_start;
3567         while (1) {
3568                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3569                                 block_end - cur_offset, 0);
3570                 BUG_ON(IS_ERR(em) || !em);
3571                 last_byte = min(extent_map_end(em), block_end);
3572                 last_byte = (last_byte + mask) & ~mask;
3573                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3574                         u64 hint_byte = 0;
3575                         hole_size = last_byte - cur_offset;
3576
3577                         trans = btrfs_start_transaction(root, 2);
3578                         if (IS_ERR(trans)) {
3579                                 err = PTR_ERR(trans);
3580                                 break;
3581                         }
3582                         btrfs_set_trans_block_group(trans, inode);
3583
3584                         err = btrfs_drop_extents(trans, inode, cur_offset,
3585                                                  cur_offset + hole_size,
3586                                                  &hint_byte, 1);
3587                         BUG_ON(err);
3588
3589                         err = btrfs_insert_file_extent(trans, root,
3590                                         inode->i_ino, cur_offset, 0,
3591                                         0, hole_size, 0, hole_size,
3592                                         0, 0, 0);
3593                         BUG_ON(err);
3594
3595                         btrfs_drop_extent_cache(inode, hole_start,
3596                                         last_byte - 1, 0);
3597
3598                         btrfs_end_transaction(trans, root);
3599                 }
3600                 free_extent_map(em);
3601                 em = NULL;
3602                 cur_offset = last_byte;
3603                 if (cur_offset >= block_end)
3604                         break;
3605         }
3606
3607         free_extent_map(em);
3608         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3609                              GFP_NOFS);
3610         return err;
3611 }
3612
3613 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3614 {
3615         struct btrfs_root *root = BTRFS_I(inode)->root;
3616         struct btrfs_trans_handle *trans;
3617         unsigned long nr;
3618         int ret;
3619
3620         if (attr->ia_size == inode->i_size)
3621                 return 0;
3622
3623         if (attr->ia_size > inode->i_size) {
3624                 unsigned long limit;
3625                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3626                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3627                         return -EFBIG;
3628                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3629                         send_sig(SIGXFSZ, current, 0);
3630                         return -EFBIG;
3631                 }
3632         }
3633
3634         trans = btrfs_start_transaction(root, 5);
3635         if (IS_ERR(trans))
3636                 return PTR_ERR(trans);
3637
3638         btrfs_set_trans_block_group(trans, inode);
3639
3640         ret = btrfs_orphan_add(trans, inode);
3641         BUG_ON(ret);
3642
3643         nr = trans->blocks_used;
3644         btrfs_end_transaction(trans, root);
3645         btrfs_btree_balance_dirty(root, nr);
3646
3647         if (attr->ia_size > inode->i_size) {
3648                 ret = btrfs_cont_expand(inode, attr->ia_size);
3649                 if (ret) {
3650                         btrfs_truncate(inode);
3651                         return ret;
3652                 }
3653
3654                 i_size_write(inode, attr->ia_size);
3655                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3656
3657                 trans = btrfs_start_transaction(root, 0);
3658                 BUG_ON(IS_ERR(trans));
3659                 btrfs_set_trans_block_group(trans, inode);
3660                 trans->block_rsv = root->orphan_block_rsv;
3661                 BUG_ON(!trans->block_rsv);
3662
3663                 ret = btrfs_update_inode(trans, root, inode);
3664                 BUG_ON(ret);
3665                 if (inode->i_nlink > 0) {
3666                         ret = btrfs_orphan_del(trans, inode);
3667                         BUG_ON(ret);
3668                 }
3669                 nr = trans->blocks_used;
3670                 btrfs_end_transaction(trans, root);
3671                 btrfs_btree_balance_dirty(root, nr);
3672                 return 0;
3673         }
3674
3675         /*
3676          * We're truncating a file that used to have good data down to
3677          * zero. Make sure it gets into the ordered flush list so that
3678          * any new writes get down to disk quickly.
3679          */
3680         if (attr->ia_size == 0)
3681                 BTRFS_I(inode)->ordered_data_close = 1;
3682
3683         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3684         ret = vmtruncate(inode, attr->ia_size);
3685         BUG_ON(ret);
3686
3687         return 0;
3688 }
3689
3690 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3691 {
3692         struct inode *inode = dentry->d_inode;
3693         struct btrfs_root *root = BTRFS_I(inode)->root;
3694         int err;
3695
3696         if (btrfs_root_readonly(root))
3697                 return -EROFS;
3698
3699         err = inode_change_ok(inode, attr);
3700         if (err)
3701                 return err;
3702
3703         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3704                 err = btrfs_setattr_size(inode, attr);
3705                 if (err)
3706                         return err;
3707         }
3708
3709         if (attr->ia_valid) {
3710                 setattr_copy(inode, attr);
3711                 mark_inode_dirty(inode);
3712
3713                 if (attr->ia_valid & ATTR_MODE)
3714                         err = btrfs_acl_chmod(inode);
3715         }
3716
3717         return err;
3718 }
3719
3720 void btrfs_evict_inode(struct inode *inode)
3721 {
3722         struct btrfs_trans_handle *trans;
3723         struct btrfs_root *root = BTRFS_I(inode)->root;
3724         unsigned long nr;
3725         int ret;
3726
3727         truncate_inode_pages(&inode->i_data, 0);
3728         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3729                                root == root->fs_info->tree_root))
3730                 goto no_delete;
3731
3732         if (is_bad_inode(inode)) {
3733                 btrfs_orphan_del(NULL, inode);
3734                 goto no_delete;
3735         }
3736         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3737         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3738
3739         if (root->fs_info->log_root_recovering) {
3740                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3741                 goto no_delete;
3742         }
3743
3744         if (inode->i_nlink > 0) {
3745                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3746                 goto no_delete;
3747         }
3748
3749         btrfs_i_size_write(inode, 0);
3750
3751         while (1) {
3752                 trans = btrfs_start_transaction(root, 0);
3753                 BUG_ON(IS_ERR(trans));
3754                 btrfs_set_trans_block_group(trans, inode);
3755                 trans->block_rsv = root->orphan_block_rsv;
3756
3757                 ret = btrfs_block_rsv_check(trans, root,
3758                                             root->orphan_block_rsv, 0, 5);
3759                 if (ret) {
3760                         BUG_ON(ret != -EAGAIN);
3761                         ret = btrfs_commit_transaction(trans, root);
3762                         BUG_ON(ret);
3763                         continue;
3764                 }
3765
3766                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3767                 if (ret != -EAGAIN)
3768                         break;
3769
3770                 nr = trans->blocks_used;
3771                 btrfs_end_transaction(trans, root);
3772                 trans = NULL;
3773                 btrfs_btree_balance_dirty(root, nr);
3774
3775         }
3776
3777         if (ret == 0) {
3778                 ret = btrfs_orphan_del(trans, inode);
3779                 BUG_ON(ret);
3780         }
3781
3782         nr = trans->blocks_used;
3783         btrfs_end_transaction(trans, root);
3784         btrfs_btree_balance_dirty(root, nr);
3785 no_delete:
3786         end_writeback(inode);
3787         return;
3788 }
3789
3790 /*
3791  * this returns the key found in the dir entry in the location pointer.
3792  * If no dir entries were found, location->objectid is 0.
3793  */
3794 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3795                                struct btrfs_key *location)
3796 {
3797         const char *name = dentry->d_name.name;
3798         int namelen = dentry->d_name.len;
3799         struct btrfs_dir_item *di;
3800         struct btrfs_path *path;
3801         struct btrfs_root *root = BTRFS_I(dir)->root;
3802         int ret = 0;
3803
3804         path = btrfs_alloc_path();
3805         BUG_ON(!path);
3806
3807         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3808                                     namelen, 0);
3809         if (IS_ERR(di))
3810                 ret = PTR_ERR(di);
3811
3812         if (!di || IS_ERR(di))
3813                 goto out_err;
3814
3815         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3816 out:
3817         btrfs_free_path(path);
3818         return ret;
3819 out_err:
3820         location->objectid = 0;
3821         goto out;
3822 }
3823
3824 /*
3825  * when we hit a tree root in a directory, the btrfs part of the inode
3826  * needs to be changed to reflect the root directory of the tree root.  This
3827  * is kind of like crossing a mount point.
3828  */
3829 static int fixup_tree_root_location(struct btrfs_root *root,
3830                                     struct inode *dir,
3831                                     struct dentry *dentry,
3832                                     struct btrfs_key *location,
3833                                     struct btrfs_root **sub_root)
3834 {
3835         struct btrfs_path *path;
3836         struct btrfs_root *new_root;
3837         struct btrfs_root_ref *ref;
3838         struct extent_buffer *leaf;
3839         int ret;
3840         int err = 0;
3841
3842         path = btrfs_alloc_path();
3843         if (!path) {
3844                 err = -ENOMEM;
3845                 goto out;
3846         }
3847
3848         err = -ENOENT;
3849         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3850                                   BTRFS_I(dir)->root->root_key.objectid,
3851                                   location->objectid);
3852         if (ret) {
3853                 if (ret < 0)
3854                         err = ret;
3855                 goto out;
3856         }
3857
3858         leaf = path->nodes[0];
3859         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3860         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3861             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3862                 goto out;
3863
3864         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3865                                    (unsigned long)(ref + 1),
3866                                    dentry->d_name.len);
3867         if (ret)
3868                 goto out;
3869
3870         btrfs_release_path(root->fs_info->tree_root, path);
3871
3872         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3873         if (IS_ERR(new_root)) {
3874                 err = PTR_ERR(new_root);
3875                 goto out;
3876         }
3877
3878         if (btrfs_root_refs(&new_root->root_item) == 0) {
3879                 err = -ENOENT;
3880                 goto out;
3881         }
3882
3883         *sub_root = new_root;
3884         location->objectid = btrfs_root_dirid(&new_root->root_item);
3885         location->type = BTRFS_INODE_ITEM_KEY;
3886         location->offset = 0;
3887         err = 0;
3888 out:
3889         btrfs_free_path(path);
3890         return err;
3891 }
3892
3893 static void inode_tree_add(struct inode *inode)
3894 {
3895         struct btrfs_root *root = BTRFS_I(inode)->root;
3896         struct btrfs_inode *entry;
3897         struct rb_node **p;
3898         struct rb_node *parent;
3899 again:
3900         p = &root->inode_tree.rb_node;
3901         parent = NULL;
3902
3903         if (hlist_unhashed(&inode->i_hash))
3904                 return;
3905
3906         spin_lock(&root->inode_lock);
3907         while (*p) {
3908                 parent = *p;
3909                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3910
3911                 if (inode->i_ino < entry->vfs_inode.i_ino)
3912                         p = &parent->rb_left;
3913                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3914                         p = &parent->rb_right;
3915                 else {
3916                         WARN_ON(!(entry->vfs_inode.i_state &
3917                                   (I_WILL_FREE | I_FREEING)));
3918                         rb_erase(parent, &root->inode_tree);
3919                         RB_CLEAR_NODE(parent);
3920                         spin_unlock(&root->inode_lock);
3921                         goto again;
3922                 }
3923         }
3924         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3925         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3926         spin_unlock(&root->inode_lock);
3927 }
3928
3929 static void inode_tree_del(struct inode *inode)
3930 {
3931         struct btrfs_root *root = BTRFS_I(inode)->root;
3932         int empty = 0;
3933
3934         spin_lock(&root->inode_lock);
3935         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3936                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3937                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3938                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3939         }
3940         spin_unlock(&root->inode_lock);
3941
3942         /*
3943          * Free space cache has inodes in the tree root, but the tree root has a
3944          * root_refs of 0, so this could end up dropping the tree root as a
3945          * snapshot, so we need the extra !root->fs_info->tree_root check to
3946          * make sure we don't drop it.
3947          */
3948         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3949             root != root->fs_info->tree_root) {
3950                 synchronize_srcu(&root->fs_info->subvol_srcu);
3951                 spin_lock(&root->inode_lock);
3952                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3953                 spin_unlock(&root->inode_lock);
3954                 if (empty)
3955                         btrfs_add_dead_root(root);
3956         }
3957 }
3958
3959 int btrfs_invalidate_inodes(struct btrfs_root *root)
3960 {
3961         struct rb_node *node;
3962         struct rb_node *prev;
3963         struct btrfs_inode *entry;
3964         struct inode *inode;
3965         u64 objectid = 0;
3966
3967         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3968
3969         spin_lock(&root->inode_lock);
3970 again:
3971         node = root->inode_tree.rb_node;
3972         prev = NULL;
3973         while (node) {
3974                 prev = node;
3975                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3976
3977                 if (objectid < entry->vfs_inode.i_ino)
3978                         node = node->rb_left;
3979                 else if (objectid > entry->vfs_inode.i_ino)
3980                         node = node->rb_right;
3981                 else
3982                         break;
3983         }
3984         if (!node) {
3985                 while (prev) {
3986                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3987                         if (objectid <= entry->vfs_inode.i_ino) {
3988                                 node = prev;
3989                                 break;
3990                         }
3991                         prev = rb_next(prev);
3992                 }
3993         }
3994         while (node) {
3995                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3996                 objectid = entry->vfs_inode.i_ino + 1;
3997                 inode = igrab(&entry->vfs_inode);
3998                 if (inode) {
3999                         spin_unlock(&root->inode_lock);
4000                         if (atomic_read(&inode->i_count) > 1)
4001                                 d_prune_aliases(inode);
4002                         /*
4003                          * btrfs_drop_inode will have it removed from
4004                          * the inode cache when its usage count
4005                          * hits zero.
4006                          */
4007                         iput(inode);
4008                         cond_resched();
4009                         spin_lock(&root->inode_lock);
4010                         goto again;
4011                 }
4012
4013                 if (cond_resched_lock(&root->inode_lock))
4014                         goto again;
4015
4016                 node = rb_next(node);
4017         }
4018         spin_unlock(&root->inode_lock);
4019         return 0;
4020 }
4021
4022 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4023 {
4024         struct btrfs_iget_args *args = p;
4025         inode->i_ino = args->ino;
4026         BTRFS_I(inode)->root = args->root;
4027         btrfs_set_inode_space_info(args->root, inode);
4028         return 0;
4029 }
4030
4031 static int btrfs_find_actor(struct inode *inode, void *opaque)
4032 {
4033         struct btrfs_iget_args *args = opaque;
4034         return args->ino == inode->i_ino &&
4035                 args->root == BTRFS_I(inode)->root;
4036 }
4037
4038 static struct inode *btrfs_iget_locked(struct super_block *s,
4039                                        u64 objectid,
4040                                        struct btrfs_root *root)
4041 {
4042         struct inode *inode;
4043         struct btrfs_iget_args args;
4044         args.ino = objectid;
4045         args.root = root;
4046
4047         inode = iget5_locked(s, objectid, btrfs_find_actor,
4048                              btrfs_init_locked_inode,
4049                              (void *)&args);
4050         return inode;
4051 }
4052
4053 /* Get an inode object given its location and corresponding root.
4054  * Returns in *is_new if the inode was read from disk
4055  */
4056 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4057                          struct btrfs_root *root, int *new)
4058 {
4059         struct inode *inode;
4060
4061         inode = btrfs_iget_locked(s, location->objectid, root);
4062         if (!inode)
4063                 return ERR_PTR(-ENOMEM);
4064
4065         if (inode->i_state & I_NEW) {
4066                 BTRFS_I(inode)->root = root;
4067                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4068                 btrfs_read_locked_inode(inode);
4069
4070                 inode_tree_add(inode);
4071                 unlock_new_inode(inode);
4072                 if (new)
4073                         *new = 1;
4074         }
4075
4076         return inode;
4077 }
4078
4079 static struct inode *new_simple_dir(struct super_block *s,
4080                                     struct btrfs_key *key,
4081                                     struct btrfs_root *root)
4082 {
4083         struct inode *inode = new_inode(s);
4084
4085         if (!inode)
4086                 return ERR_PTR(-ENOMEM);
4087
4088         BTRFS_I(inode)->root = root;
4089         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4090         BTRFS_I(inode)->dummy_inode = 1;
4091
4092         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4093         inode->i_op = &simple_dir_inode_operations;
4094         inode->i_fop = &simple_dir_operations;
4095         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4096         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4097
4098         return inode;
4099 }
4100
4101 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4102 {
4103         struct inode *inode;
4104         struct btrfs_root *root = BTRFS_I(dir)->root;
4105         struct btrfs_root *sub_root = root;
4106         struct btrfs_key location;
4107         int index;
4108         int ret;
4109
4110         dentry->d_op = &btrfs_dentry_operations;
4111
4112         if (dentry->d_name.len > BTRFS_NAME_LEN)
4113                 return ERR_PTR(-ENAMETOOLONG);
4114
4115         ret = btrfs_inode_by_name(dir, dentry, &location);
4116
4117         if (ret < 0)
4118                 return ERR_PTR(ret);
4119
4120         if (location.objectid == 0)
4121                 return NULL;
4122
4123         if (location.type == BTRFS_INODE_ITEM_KEY) {
4124                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4125                 return inode;
4126         }
4127
4128         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4129
4130         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4131         ret = fixup_tree_root_location(root, dir, dentry,
4132                                        &location, &sub_root);
4133         if (ret < 0) {
4134                 if (ret != -ENOENT)
4135                         inode = ERR_PTR(ret);
4136                 else
4137                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4138         } else {
4139                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4140         }
4141         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4142
4143         if (!IS_ERR(inode) && root != sub_root) {
4144                 down_read(&root->fs_info->cleanup_work_sem);
4145                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4146                         btrfs_orphan_cleanup(sub_root);
4147                 up_read(&root->fs_info->cleanup_work_sem);
4148         }
4149
4150         return inode;
4151 }
4152
4153 static int btrfs_dentry_delete(struct dentry *dentry)
4154 {
4155         struct btrfs_root *root;
4156
4157         if (!dentry->d_inode && !IS_ROOT(dentry))
4158                 dentry = dentry->d_parent;
4159
4160         if (dentry->d_inode) {
4161                 root = BTRFS_I(dentry->d_inode)->root;
4162                 if (btrfs_root_refs(&root->root_item) == 0)
4163                         return 1;
4164         }
4165         return 0;
4166 }
4167
4168 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4169                                    struct nameidata *nd)
4170 {
4171         struct inode *inode;
4172
4173         inode = btrfs_lookup_dentry(dir, dentry);
4174         if (IS_ERR(inode))
4175                 return ERR_CAST(inode);
4176
4177         return d_splice_alias(inode, dentry);
4178 }
4179
4180 static unsigned char btrfs_filetype_table[] = {
4181         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4182 };
4183
4184 static int btrfs_real_readdir(struct file *filp, void *dirent,
4185                               filldir_t filldir)
4186 {
4187         struct inode *inode = filp->f_dentry->d_inode;
4188         struct btrfs_root *root = BTRFS_I(inode)->root;
4189         struct btrfs_item *item;
4190         struct btrfs_dir_item *di;
4191         struct btrfs_key key;
4192         struct btrfs_key found_key;
4193         struct btrfs_path *path;
4194         int ret;
4195         u32 nritems;
4196         struct extent_buffer *leaf;
4197         int slot;
4198         int advance;
4199         unsigned char d_type;
4200         int over = 0;
4201         u32 di_cur;
4202         u32 di_total;
4203         u32 di_len;
4204         int key_type = BTRFS_DIR_INDEX_KEY;
4205         char tmp_name[32];
4206         char *name_ptr;
4207         int name_len;
4208
4209         /* FIXME, use a real flag for deciding about the key type */
4210         if (root->fs_info->tree_root == root)
4211                 key_type = BTRFS_DIR_ITEM_KEY;
4212
4213         /* special case for "." */
4214         if (filp->f_pos == 0) {
4215                 over = filldir(dirent, ".", 1,
4216                                1, inode->i_ino,
4217                                DT_DIR);
4218                 if (over)
4219                         return 0;
4220                 filp->f_pos = 1;
4221         }
4222         /* special case for .., just use the back ref */
4223         if (filp->f_pos == 1) {
4224                 u64 pino = parent_ino(filp->f_path.dentry);
4225                 over = filldir(dirent, "..", 2,
4226                                2, pino, DT_DIR);
4227                 if (over)
4228                         return 0;
4229                 filp->f_pos = 2;
4230         }
4231         path = btrfs_alloc_path();
4232         path->reada = 2;
4233
4234         btrfs_set_key_type(&key, key_type);
4235         key.offset = filp->f_pos;
4236         key.objectid = inode->i_ino;
4237
4238         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4239         if (ret < 0)
4240                 goto err;
4241         advance = 0;
4242
4243         while (1) {
4244                 leaf = path->nodes[0];
4245                 nritems = btrfs_header_nritems(leaf);
4246                 slot = path->slots[0];
4247                 if (advance || slot >= nritems) {
4248                         if (slot >= nritems - 1) {
4249                                 ret = btrfs_next_leaf(root, path);
4250                                 if (ret)
4251                                         break;
4252                                 leaf = path->nodes[0];
4253                                 nritems = btrfs_header_nritems(leaf);
4254                                 slot = path->slots[0];
4255                         } else {
4256                                 slot++;
4257                                 path->slots[0]++;
4258                         }
4259                 }
4260
4261                 advance = 1;
4262                 item = btrfs_item_nr(leaf, slot);
4263                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4264
4265                 if (found_key.objectid != key.objectid)
4266                         break;
4267                 if (btrfs_key_type(&found_key) != key_type)
4268                         break;
4269                 if (found_key.offset < filp->f_pos)
4270                         continue;
4271
4272                 filp->f_pos = found_key.offset;
4273
4274                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4275                 di_cur = 0;
4276                 di_total = btrfs_item_size(leaf, item);
4277
4278                 while (di_cur < di_total) {
4279                         struct btrfs_key location;
4280
4281                         name_len = btrfs_dir_name_len(leaf, di);
4282                         if (name_len <= sizeof(tmp_name)) {
4283                                 name_ptr = tmp_name;
4284                         } else {
4285                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4286                                 if (!name_ptr) {
4287                                         ret = -ENOMEM;
4288                                         goto err;
4289                                 }
4290                         }
4291                         read_extent_buffer(leaf, name_ptr,
4292                                            (unsigned long)(di + 1), name_len);
4293
4294                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4295                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4296
4297                         /* is this a reference to our own snapshot? If so
4298                          * skip it
4299                          */
4300                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4301                             location.objectid == root->root_key.objectid) {
4302                                 over = 0;
4303                                 goto skip;
4304                         }
4305                         over = filldir(dirent, name_ptr, name_len,
4306                                        found_key.offset, location.objectid,
4307                                        d_type);
4308
4309 skip:
4310                         if (name_ptr != tmp_name)
4311                                 kfree(name_ptr);
4312
4313                         if (over)
4314                                 goto nopos;
4315                         di_len = btrfs_dir_name_len(leaf, di) +
4316                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4317                         di_cur += di_len;
4318                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4319                 }
4320         }
4321
4322         /* Reached end of directory/root. Bump pos past the last item. */
4323         if (key_type == BTRFS_DIR_INDEX_KEY)
4324                 /*
4325                  * 32-bit glibc will use getdents64, but then strtol -
4326                  * so the last number we can serve is this.
4327                  */
4328                 filp->f_pos = 0x7fffffff;
4329         else
4330                 filp->f_pos++;
4331 nopos:
4332         ret = 0;
4333 err:
4334         btrfs_free_path(path);
4335         return ret;
4336 }
4337
4338 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4339 {
4340         struct btrfs_root *root = BTRFS_I(inode)->root;
4341         struct btrfs_trans_handle *trans;
4342         int ret = 0;
4343         bool nolock = false;
4344
4345         if (BTRFS_I(inode)->dummy_inode)
4346                 return 0;
4347
4348         smp_mb();
4349         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4350
4351         if (wbc->sync_mode == WB_SYNC_ALL) {
4352                 if (nolock)
4353                         trans = btrfs_join_transaction_nolock(root, 1);
4354                 else
4355                         trans = btrfs_join_transaction(root, 1);
4356                 if (IS_ERR(trans))
4357                         return PTR_ERR(trans);
4358                 btrfs_set_trans_block_group(trans, inode);
4359                 if (nolock)
4360                         ret = btrfs_end_transaction_nolock(trans, root);
4361                 else
4362                         ret = btrfs_commit_transaction(trans, root);
4363         }
4364         return ret;
4365 }
4366
4367 /*
4368  * This is somewhat expensive, updating the tree every time the
4369  * inode changes.  But, it is most likely to find the inode in cache.
4370  * FIXME, needs more benchmarking...there are no reasons other than performance
4371  * to keep or drop this code.
4372  */
4373 void btrfs_dirty_inode(struct inode *inode)
4374 {
4375         struct btrfs_root *root = BTRFS_I(inode)->root;
4376         struct btrfs_trans_handle *trans;
4377         int ret;
4378
4379         if (BTRFS_I(inode)->dummy_inode)
4380                 return;
4381
4382         trans = btrfs_join_transaction(root, 1);
4383         BUG_ON(IS_ERR(trans));
4384         btrfs_set_trans_block_group(trans, inode);
4385
4386         ret = btrfs_update_inode(trans, root, inode);
4387         if (ret && ret == -ENOSPC) {
4388                 /* whoops, lets try again with the full transaction */
4389                 btrfs_end_transaction(trans, root);
4390                 trans = btrfs_start_transaction(root, 1);
4391                 if (IS_ERR(trans)) {
4392                         if (printk_ratelimit()) {
4393                                 printk(KERN_ERR "btrfs: fail to "
4394                                        "dirty  inode %lu error %ld\n",
4395                                        inode->i_ino, PTR_ERR(trans));
4396                         }
4397                         return;
4398                 }
4399                 btrfs_set_trans_block_group(trans, inode);
4400
4401                 ret = btrfs_update_inode(trans, root, inode);
4402                 if (ret) {
4403                         if (printk_ratelimit()) {
4404                                 printk(KERN_ERR "btrfs: fail to "
4405                                        "dirty  inode %lu error %d\n",
4406                                        inode->i_ino, ret);
4407                         }
4408                 }
4409         }
4410         btrfs_end_transaction(trans, root);
4411 }
4412
4413 /*
4414  * find the highest existing sequence number in a directory
4415  * and then set the in-memory index_cnt variable to reflect
4416  * free sequence numbers
4417  */
4418 static int btrfs_set_inode_index_count(struct inode *inode)
4419 {
4420         struct btrfs_root *root = BTRFS_I(inode)->root;
4421         struct btrfs_key key, found_key;
4422         struct btrfs_path *path;
4423         struct extent_buffer *leaf;
4424         int ret;
4425
4426         key.objectid = inode->i_ino;
4427         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4428         key.offset = (u64)-1;
4429
4430         path = btrfs_alloc_path();
4431         if (!path)
4432                 return -ENOMEM;
4433
4434         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4435         if (ret < 0)
4436                 goto out;
4437         /* FIXME: we should be able to handle this */
4438         if (ret == 0)
4439                 goto out;
4440         ret = 0;
4441
4442         /*
4443          * MAGIC NUMBER EXPLANATION:
4444          * since we search a directory based on f_pos we have to start at 2
4445          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4446          * else has to start at 2
4447          */
4448         if (path->slots[0] == 0) {
4449                 BTRFS_I(inode)->index_cnt = 2;
4450                 goto out;
4451         }
4452
4453         path->slots[0]--;
4454
4455         leaf = path->nodes[0];
4456         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4457
4458         if (found_key.objectid != inode->i_ino ||
4459             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4460                 BTRFS_I(inode)->index_cnt = 2;
4461                 goto out;
4462         }
4463
4464         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4465 out:
4466         btrfs_free_path(path);
4467         return ret;
4468 }
4469
4470 /*
4471  * helper to find a free sequence number in a given directory.  This current
4472  * code is very simple, later versions will do smarter things in the btree
4473  */
4474 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4475 {
4476         int ret = 0;
4477
4478         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4479                 ret = btrfs_set_inode_index_count(dir);
4480                 if (ret)
4481                         return ret;
4482         }
4483
4484         *index = BTRFS_I(dir)->index_cnt;
4485         BTRFS_I(dir)->index_cnt++;
4486
4487         return ret;
4488 }
4489
4490 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4491                                      struct btrfs_root *root,
4492                                      struct inode *dir,
4493                                      const char *name, int name_len,
4494                                      u64 ref_objectid, u64 objectid,
4495                                      u64 alloc_hint, int mode, u64 *index)
4496 {
4497         struct inode *inode;
4498         struct btrfs_inode_item *inode_item;
4499         struct btrfs_key *location;
4500         struct btrfs_path *path;
4501         struct btrfs_inode_ref *ref;
4502         struct btrfs_key key[2];
4503         u32 sizes[2];
4504         unsigned long ptr;
4505         int ret;
4506         int owner;
4507
4508         path = btrfs_alloc_path();
4509         BUG_ON(!path);
4510
4511         inode = new_inode(root->fs_info->sb);
4512         if (!inode)
4513                 return ERR_PTR(-ENOMEM);
4514
4515         if (dir) {
4516                 ret = btrfs_set_inode_index(dir, index);
4517                 if (ret) {
4518                         iput(inode);
4519                         return ERR_PTR(ret);
4520                 }
4521         }
4522         /*
4523          * index_cnt is ignored for everything but a dir,
4524          * btrfs_get_inode_index_count has an explanation for the magic
4525          * number
4526          */
4527         BTRFS_I(inode)->index_cnt = 2;
4528         BTRFS_I(inode)->root = root;
4529         BTRFS_I(inode)->generation = trans->transid;
4530         inode->i_generation = BTRFS_I(inode)->generation;
4531         btrfs_set_inode_space_info(root, inode);
4532
4533         if (mode & S_IFDIR)
4534                 owner = 0;
4535         else
4536                 owner = 1;
4537         BTRFS_I(inode)->block_group =
4538                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4539
4540         key[0].objectid = objectid;
4541         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4542         key[0].offset = 0;
4543
4544         key[1].objectid = objectid;
4545         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4546         key[1].offset = ref_objectid;
4547
4548         sizes[0] = sizeof(struct btrfs_inode_item);
4549         sizes[1] = name_len + sizeof(*ref);
4550
4551         path->leave_spinning = 1;
4552         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4553         if (ret != 0)
4554                 goto fail;
4555
4556         inode_init_owner(inode, dir, mode);
4557         inode->i_ino = objectid;
4558         inode_set_bytes(inode, 0);
4559         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4560         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4561                                   struct btrfs_inode_item);
4562         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4563
4564         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4565                              struct btrfs_inode_ref);
4566         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4567         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4568         ptr = (unsigned long)(ref + 1);
4569         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4570
4571         btrfs_mark_buffer_dirty(path->nodes[0]);
4572         btrfs_free_path(path);
4573
4574         location = &BTRFS_I(inode)->location;
4575         location->objectid = objectid;
4576         location->offset = 0;
4577         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4578
4579         btrfs_inherit_iflags(inode, dir);
4580
4581         if ((mode & S_IFREG)) {
4582                 if (btrfs_test_opt(root, NODATASUM))
4583                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4584                 if (btrfs_test_opt(root, NODATACOW))
4585                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4586         }
4587
4588         insert_inode_hash(inode);
4589         inode_tree_add(inode);
4590         return inode;
4591 fail:
4592         if (dir)
4593                 BTRFS_I(dir)->index_cnt--;
4594         btrfs_free_path(path);
4595         iput(inode);
4596         return ERR_PTR(ret);
4597 }
4598
4599 static inline u8 btrfs_inode_type(struct inode *inode)
4600 {
4601         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4602 }
4603
4604 /*
4605  * utility function to add 'inode' into 'parent_inode' with
4606  * a give name and a given sequence number.
4607  * if 'add_backref' is true, also insert a backref from the
4608  * inode to the parent directory.
4609  */
4610 int btrfs_add_link(struct btrfs_trans_handle *trans,
4611                    struct inode *parent_inode, struct inode *inode,
4612                    const char *name, int name_len, int add_backref, u64 index)
4613 {
4614         int ret = 0;
4615         struct btrfs_key key;
4616         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4617
4618         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4619                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4620         } else {
4621                 key.objectid = inode->i_ino;
4622                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4623                 key.offset = 0;
4624         }
4625
4626         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4627                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4628                                          key.objectid, root->root_key.objectid,
4629                                          parent_inode->i_ino,
4630                                          index, name, name_len);
4631         } else if (add_backref) {
4632                 ret = btrfs_insert_inode_ref(trans, root,
4633                                              name, name_len, inode->i_ino,
4634                                              parent_inode->i_ino, index);
4635         }
4636
4637         if (ret == 0) {
4638                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4639                                             parent_inode->i_ino, &key,
4640                                             btrfs_inode_type(inode), index);
4641                 BUG_ON(ret);
4642
4643                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4644                                    name_len * 2);
4645                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4646                 ret = btrfs_update_inode(trans, root, parent_inode);
4647         }
4648         return ret;
4649 }
4650
4651 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4652                             struct inode *dir, struct dentry *dentry,
4653                             struct inode *inode, int backref, u64 index)
4654 {
4655         int err = btrfs_add_link(trans, dir, inode,
4656                                  dentry->d_name.name, dentry->d_name.len,
4657                                  backref, index);
4658         if (!err) {
4659                 d_instantiate(dentry, inode);
4660                 return 0;
4661         }
4662         if (err > 0)
4663                 err = -EEXIST;
4664         return err;
4665 }
4666
4667 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4668                         int mode, dev_t rdev)
4669 {
4670         struct btrfs_trans_handle *trans;
4671         struct btrfs_root *root = BTRFS_I(dir)->root;
4672         struct inode *inode = NULL;
4673         int err;
4674         int drop_inode = 0;
4675         u64 objectid;
4676         unsigned long nr = 0;
4677         u64 index = 0;
4678
4679         if (!new_valid_dev(rdev))
4680                 return -EINVAL;
4681
4682         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4683         if (err)
4684                 return err;
4685
4686         /*
4687          * 2 for inode item and ref
4688          * 2 for dir items
4689          * 1 for xattr if selinux is on
4690          */
4691         trans = btrfs_start_transaction(root, 5);
4692         if (IS_ERR(trans))
4693                 return PTR_ERR(trans);
4694
4695         btrfs_set_trans_block_group(trans, dir);
4696
4697         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4698                                 dentry->d_name.len, dir->i_ino, objectid,
4699                                 BTRFS_I(dir)->block_group, mode, &index);
4700         err = PTR_ERR(inode);
4701         if (IS_ERR(inode))
4702                 goto out_unlock;
4703
4704         err = btrfs_init_inode_security(trans, inode, dir);
4705         if (err) {
4706                 drop_inode = 1;
4707                 goto out_unlock;
4708         }
4709
4710         btrfs_set_trans_block_group(trans, inode);
4711         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4712         if (err)
4713                 drop_inode = 1;
4714         else {
4715                 inode->i_op = &btrfs_special_inode_operations;
4716                 init_special_inode(inode, inode->i_mode, rdev);
4717                 btrfs_update_inode(trans, root, inode);
4718         }
4719         btrfs_update_inode_block_group(trans, inode);
4720         btrfs_update_inode_block_group(trans, dir);
4721 out_unlock:
4722         nr = trans->blocks_used;
4723         btrfs_end_transaction_throttle(trans, root);
4724         btrfs_btree_balance_dirty(root, nr);
4725         if (drop_inode) {
4726                 inode_dec_link_count(inode);
4727                 iput(inode);
4728         }
4729         return err;
4730 }
4731
4732 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4733                         int mode, struct nameidata *nd)
4734 {
4735         struct btrfs_trans_handle *trans;
4736         struct btrfs_root *root = BTRFS_I(dir)->root;
4737         struct inode *inode = NULL;
4738         int drop_inode = 0;
4739         int err;
4740         unsigned long nr = 0;
4741         u64 objectid;
4742         u64 index = 0;
4743
4744         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4745         if (err)
4746                 return err;
4747         /*
4748          * 2 for inode item and ref
4749          * 2 for dir items
4750          * 1 for xattr if selinux is on
4751          */
4752         trans = btrfs_start_transaction(root, 5);
4753         if (IS_ERR(trans))
4754                 return PTR_ERR(trans);
4755
4756         btrfs_set_trans_block_group(trans, dir);
4757
4758         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4759                                 dentry->d_name.len, dir->i_ino, objectid,
4760                                 BTRFS_I(dir)->block_group, mode, &index);
4761         err = PTR_ERR(inode);
4762         if (IS_ERR(inode))
4763                 goto out_unlock;
4764
4765         err = btrfs_init_inode_security(trans, inode, dir);
4766         if (err) {
4767                 drop_inode = 1;
4768                 goto out_unlock;
4769         }
4770
4771         btrfs_set_trans_block_group(trans, inode);
4772         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4773         if (err)
4774                 drop_inode = 1;
4775         else {
4776                 inode->i_mapping->a_ops = &btrfs_aops;
4777                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4778                 inode->i_fop = &btrfs_file_operations;
4779                 inode->i_op = &btrfs_file_inode_operations;
4780                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4781         }
4782         btrfs_update_inode_block_group(trans, inode);
4783         btrfs_update_inode_block_group(trans, dir);
4784 out_unlock:
4785         nr = trans->blocks_used;
4786         btrfs_end_transaction_throttle(trans, root);
4787         if (drop_inode) {
4788                 inode_dec_link_count(inode);
4789                 iput(inode);
4790         }
4791         btrfs_btree_balance_dirty(root, nr);
4792         return err;
4793 }
4794
4795 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4796                       struct dentry *dentry)
4797 {
4798         struct btrfs_trans_handle *trans;
4799         struct btrfs_root *root = BTRFS_I(dir)->root;
4800         struct inode *inode = old_dentry->d_inode;
4801         u64 index;
4802         unsigned long nr = 0;
4803         int err;
4804         int drop_inode = 0;
4805
4806         if (inode->i_nlink == 0)
4807                 return -ENOENT;
4808
4809         /* do not allow sys_link's with other subvols of the same device */
4810         if (root->objectid != BTRFS_I(inode)->root->objectid)
4811                 return -EPERM;
4812
4813         btrfs_inc_nlink(inode);
4814         inode->i_ctime = CURRENT_TIME;
4815
4816         err = btrfs_set_inode_index(dir, &index);
4817         if (err)
4818                 goto fail;
4819
4820         /*
4821          * 1 item for inode ref
4822          * 2 items for dir items
4823          */
4824         trans = btrfs_start_transaction(root, 3);
4825         if (IS_ERR(trans)) {
4826                 err = PTR_ERR(trans);
4827                 goto fail;
4828         }
4829
4830         btrfs_set_trans_block_group(trans, dir);
4831         atomic_inc(&inode->i_count);
4832
4833         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4834
4835         if (err) {
4836                 drop_inode = 1;
4837         } else {
4838                 struct dentry *parent = dget_parent(dentry);
4839                 btrfs_update_inode_block_group(trans, dir);
4840                 err = btrfs_update_inode(trans, root, inode);
4841                 BUG_ON(err);
4842                 btrfs_log_new_name(trans, inode, NULL, parent);
4843                 dput(parent);
4844         }
4845
4846         nr = trans->blocks_used;
4847         btrfs_end_transaction_throttle(trans, root);
4848 fail:
4849         if (drop_inode) {
4850                 inode_dec_link_count(inode);
4851                 iput(inode);
4852         }
4853         btrfs_btree_balance_dirty(root, nr);
4854         return err;
4855 }
4856
4857 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4858 {
4859         struct inode *inode = NULL;
4860         struct btrfs_trans_handle *trans;
4861         struct btrfs_root *root = BTRFS_I(dir)->root;
4862         int err = 0;
4863         int drop_on_err = 0;
4864         u64 objectid = 0;
4865         u64 index = 0;
4866         unsigned long nr = 1;
4867
4868         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4869         if (err)
4870                 return err;
4871
4872         /*
4873          * 2 items for inode and ref
4874          * 2 items for dir items
4875          * 1 for xattr if selinux is on
4876          */
4877         trans = btrfs_start_transaction(root, 5);
4878         if (IS_ERR(trans))
4879                 return PTR_ERR(trans);
4880         btrfs_set_trans_block_group(trans, dir);
4881
4882         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4883                                 dentry->d_name.len, dir->i_ino, objectid,
4884                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4885                                 &index);
4886         if (IS_ERR(inode)) {
4887                 err = PTR_ERR(inode);
4888                 goto out_fail;
4889         }
4890
4891         drop_on_err = 1;
4892
4893         err = btrfs_init_inode_security(trans, inode, dir);
4894         if (err)
4895                 goto out_fail;
4896
4897         inode->i_op = &btrfs_dir_inode_operations;
4898         inode->i_fop = &btrfs_dir_file_operations;
4899         btrfs_set_trans_block_group(trans, inode);
4900
4901         btrfs_i_size_write(inode, 0);
4902         err = btrfs_update_inode(trans, root, inode);
4903         if (err)
4904                 goto out_fail;
4905
4906         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4907                              dentry->d_name.len, 0, index);
4908         if (err)
4909                 goto out_fail;
4910
4911         d_instantiate(dentry, inode);
4912         drop_on_err = 0;
4913         btrfs_update_inode_block_group(trans, inode);
4914         btrfs_update_inode_block_group(trans, dir);
4915
4916 out_fail:
4917         nr = trans->blocks_used;
4918         btrfs_end_transaction_throttle(trans, root);
4919         if (drop_on_err)
4920                 iput(inode);
4921         btrfs_btree_balance_dirty(root, nr);
4922         return err;
4923 }
4924
4925 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4926  * and an extent that you want to insert, deal with overlap and insert
4927  * the new extent into the tree.
4928  */
4929 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4930                                 struct extent_map *existing,
4931                                 struct extent_map *em,
4932                                 u64 map_start, u64 map_len)
4933 {
4934         u64 start_diff;
4935
4936         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4937         start_diff = map_start - em->start;
4938         em->start = map_start;
4939         em->len = map_len;
4940         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4941             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4942                 em->block_start += start_diff;
4943                 em->block_len -= start_diff;
4944         }
4945         return add_extent_mapping(em_tree, em);
4946 }
4947
4948 static noinline int uncompress_inline(struct btrfs_path *path,
4949                                       struct inode *inode, struct page *page,
4950                                       size_t pg_offset, u64 extent_offset,
4951                                       struct btrfs_file_extent_item *item)
4952 {
4953         int ret;
4954         struct extent_buffer *leaf = path->nodes[0];
4955         char *tmp;
4956         size_t max_size;
4957         unsigned long inline_size;
4958         unsigned long ptr;
4959         int compress_type;
4960
4961         WARN_ON(pg_offset != 0);
4962         compress_type = btrfs_file_extent_compression(leaf, item);
4963         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4964         inline_size = btrfs_file_extent_inline_item_len(leaf,
4965                                         btrfs_item_nr(leaf, path->slots[0]));
4966         tmp = kmalloc(inline_size, GFP_NOFS);
4967         ptr = btrfs_file_extent_inline_start(item);
4968
4969         read_extent_buffer(leaf, tmp, ptr, inline_size);
4970
4971         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4972         ret = btrfs_decompress(compress_type, tmp, page,
4973                                extent_offset, inline_size, max_size);
4974         if (ret) {
4975                 char *kaddr = kmap_atomic(page, KM_USER0);
4976                 unsigned long copy_size = min_t(u64,
4977                                   PAGE_CACHE_SIZE - pg_offset,
4978                                   max_size - extent_offset);
4979                 memset(kaddr + pg_offset, 0, copy_size);
4980                 kunmap_atomic(kaddr, KM_USER0);
4981         }
4982         kfree(tmp);
4983         return 0;
4984 }
4985
4986 /*
4987  * a bit scary, this does extent mapping from logical file offset to the disk.
4988  * the ugly parts come from merging extents from the disk with the in-ram
4989  * representation.  This gets more complex because of the data=ordered code,
4990  * where the in-ram extents might be locked pending data=ordered completion.
4991  *
4992  * This also copies inline extents directly into the page.
4993  */
4994
4995 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4996                                     size_t pg_offset, u64 start, u64 len,
4997                                     int create)
4998 {
4999         int ret;
5000         int err = 0;
5001         u64 bytenr;
5002         u64 extent_start = 0;
5003         u64 extent_end = 0;
5004         u64 objectid = inode->i_ino;
5005         u32 found_type;
5006         struct btrfs_path *path = NULL;
5007         struct btrfs_root *root = BTRFS_I(inode)->root;
5008         struct btrfs_file_extent_item *item;
5009         struct extent_buffer *leaf;
5010         struct btrfs_key found_key;
5011         struct extent_map *em = NULL;
5012         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5013         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5014         struct btrfs_trans_handle *trans = NULL;
5015         int compress_type;
5016
5017 again:
5018         read_lock(&em_tree->lock);
5019         em = lookup_extent_mapping(em_tree, start, len);
5020         if (em)
5021                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5022         read_unlock(&em_tree->lock);
5023
5024         if (em) {
5025                 if (em->start > start || em->start + em->len <= start)
5026                         free_extent_map(em);
5027                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5028                         free_extent_map(em);
5029                 else
5030                         goto out;
5031         }
5032         em = alloc_extent_map(GFP_NOFS);
5033         if (!em) {
5034                 err = -ENOMEM;
5035                 goto out;
5036         }
5037         em->bdev = root->fs_info->fs_devices->latest_bdev;
5038         em->start = EXTENT_MAP_HOLE;
5039         em->orig_start = EXTENT_MAP_HOLE;
5040         em->len = (u64)-1;
5041         em->block_len = (u64)-1;
5042
5043         if (!path) {
5044                 path = btrfs_alloc_path();
5045                 BUG_ON(!path);
5046         }
5047
5048         ret = btrfs_lookup_file_extent(trans, root, path,
5049                                        objectid, start, trans != NULL);
5050         if (ret < 0) {
5051                 err = ret;
5052                 goto out;
5053         }
5054
5055         if (ret != 0) {
5056                 if (path->slots[0] == 0)
5057                         goto not_found;
5058                 path->slots[0]--;
5059         }
5060
5061         leaf = path->nodes[0];
5062         item = btrfs_item_ptr(leaf, path->slots[0],
5063                               struct btrfs_file_extent_item);
5064         /* are we inside the extent that was found? */
5065         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5066         found_type = btrfs_key_type(&found_key);
5067         if (found_key.objectid != objectid ||
5068             found_type != BTRFS_EXTENT_DATA_KEY) {
5069                 goto not_found;
5070         }
5071
5072         found_type = btrfs_file_extent_type(leaf, item);
5073         extent_start = found_key.offset;
5074         compress_type = btrfs_file_extent_compression(leaf, item);
5075         if (found_type == BTRFS_FILE_EXTENT_REG ||
5076             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5077                 extent_end = extent_start +
5078                        btrfs_file_extent_num_bytes(leaf, item);
5079         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5080                 size_t size;
5081                 size = btrfs_file_extent_inline_len(leaf, item);
5082                 extent_end = (extent_start + size + root->sectorsize - 1) &
5083                         ~((u64)root->sectorsize - 1);
5084         }
5085
5086         if (start >= extent_end) {
5087                 path->slots[0]++;
5088                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5089                         ret = btrfs_next_leaf(root, path);
5090                         if (ret < 0) {
5091                                 err = ret;
5092                                 goto out;
5093                         }
5094                         if (ret > 0)
5095                                 goto not_found;
5096                         leaf = path->nodes[0];
5097                 }
5098                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5099                 if (found_key.objectid != objectid ||
5100                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5101                         goto not_found;
5102                 if (start + len <= found_key.offset)
5103                         goto not_found;
5104                 em->start = start;
5105                 em->len = found_key.offset - start;
5106                 goto not_found_em;
5107         }
5108
5109         if (found_type == BTRFS_FILE_EXTENT_REG ||
5110             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5111                 em->start = extent_start;
5112                 em->len = extent_end - extent_start;
5113                 em->orig_start = extent_start -
5114                                  btrfs_file_extent_offset(leaf, item);
5115                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5116                 if (bytenr == 0) {
5117                         em->block_start = EXTENT_MAP_HOLE;
5118                         goto insert;
5119                 }
5120                 if (compress_type != BTRFS_COMPRESS_NONE) {
5121                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5122                         em->compress_type = compress_type;
5123                         em->block_start = bytenr;
5124                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5125                                                                          item);
5126                 } else {
5127                         bytenr += btrfs_file_extent_offset(leaf, item);
5128                         em->block_start = bytenr;
5129                         em->block_len = em->len;
5130                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5131                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5132                 }
5133                 goto insert;
5134         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5135                 unsigned long ptr;
5136                 char *map;
5137                 size_t size;
5138                 size_t extent_offset;
5139                 size_t copy_size;
5140
5141                 em->block_start = EXTENT_MAP_INLINE;
5142                 if (!page || create) {
5143                         em->start = extent_start;
5144                         em->len = extent_end - extent_start;
5145                         goto out;
5146                 }
5147
5148                 size = btrfs_file_extent_inline_len(leaf, item);
5149                 extent_offset = page_offset(page) + pg_offset - extent_start;
5150                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5151                                 size - extent_offset);
5152                 em->start = extent_start + extent_offset;
5153                 em->len = (copy_size + root->sectorsize - 1) &
5154                         ~((u64)root->sectorsize - 1);
5155                 em->orig_start = EXTENT_MAP_INLINE;
5156                 if (compress_type) {
5157                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5158                         em->compress_type = compress_type;
5159                 }
5160                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5161                 if (create == 0 && !PageUptodate(page)) {
5162                         if (btrfs_file_extent_compression(leaf, item) !=
5163                             BTRFS_COMPRESS_NONE) {
5164                                 ret = uncompress_inline(path, inode, page,
5165                                                         pg_offset,
5166                                                         extent_offset, item);
5167                                 BUG_ON(ret);
5168                         } else {
5169                                 map = kmap(page);
5170                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5171                                                    copy_size);
5172                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5173                                         memset(map + pg_offset + copy_size, 0,
5174                                                PAGE_CACHE_SIZE - pg_offset -
5175                                                copy_size);
5176                                 }
5177                                 kunmap(page);
5178                         }
5179                         flush_dcache_page(page);
5180                 } else if (create && PageUptodate(page)) {
5181                         WARN_ON(1);
5182                         if (!trans) {
5183                                 kunmap(page);
5184                                 free_extent_map(em);
5185                                 em = NULL;
5186                                 btrfs_release_path(root, path);
5187                                 trans = btrfs_join_transaction(root, 1);
5188                                 if (IS_ERR(trans))
5189                                         return ERR_CAST(trans);
5190                                 goto again;
5191                         }
5192                         map = kmap(page);
5193                         write_extent_buffer(leaf, map + pg_offset, ptr,
5194                                             copy_size);
5195                         kunmap(page);
5196                         btrfs_mark_buffer_dirty(leaf);
5197                 }
5198                 set_extent_uptodate(io_tree, em->start,
5199                                     extent_map_end(em) - 1, GFP_NOFS);
5200                 goto insert;
5201         } else {
5202                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5203                 WARN_ON(1);
5204         }
5205 not_found:
5206         em->start = start;
5207         em->len = len;
5208 not_found_em:
5209         em->block_start = EXTENT_MAP_HOLE;
5210         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5211 insert:
5212         btrfs_release_path(root, path);
5213         if (em->start > start || extent_map_end(em) <= start) {
5214                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5215                        "[%llu %llu]\n", (unsigned long long)em->start,
5216                        (unsigned long long)em->len,
5217                        (unsigned long long)start,
5218                        (unsigned long long)len);
5219                 err = -EIO;
5220                 goto out;
5221         }
5222
5223         err = 0;
5224         write_lock(&em_tree->lock);
5225         ret = add_extent_mapping(em_tree, em);
5226         /* it is possible that someone inserted the extent into the tree
5227          * while we had the lock dropped.  It is also possible that
5228          * an overlapping map exists in the tree
5229          */
5230         if (ret == -EEXIST) {
5231                 struct extent_map *existing;
5232
5233                 ret = 0;
5234
5235                 existing = lookup_extent_mapping(em_tree, start, len);
5236                 if (existing && (existing->start > start ||
5237                     existing->start + existing->len <= start)) {
5238                         free_extent_map(existing);
5239                         existing = NULL;
5240                 }
5241                 if (!existing) {
5242                         existing = lookup_extent_mapping(em_tree, em->start,
5243                                                          em->len);
5244                         if (existing) {
5245                                 err = merge_extent_mapping(em_tree, existing,
5246                                                            em, start,
5247                                                            root->sectorsize);
5248                                 free_extent_map(existing);
5249                                 if (err) {
5250                                         free_extent_map(em);
5251                                         em = NULL;
5252                                 }
5253                         } else {
5254                                 err = -EIO;
5255                                 free_extent_map(em);
5256                                 em = NULL;
5257                         }
5258                 } else {
5259                         free_extent_map(em);
5260                         em = existing;
5261                         err = 0;
5262                 }
5263         }
5264         write_unlock(&em_tree->lock);
5265 out:
5266         if (path)
5267                 btrfs_free_path(path);
5268         if (trans) {
5269                 ret = btrfs_end_transaction(trans, root);
5270                 if (!err)
5271                         err = ret;
5272         }
5273         if (err) {
5274                 free_extent_map(em);
5275                 return ERR_PTR(err);
5276         }
5277         return em;
5278 }
5279
5280 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5281                                                   u64 start, u64 len)
5282 {
5283         struct btrfs_root *root = BTRFS_I(inode)->root;
5284         struct btrfs_trans_handle *trans;
5285         struct extent_map *em;
5286         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5287         struct btrfs_key ins;
5288         u64 alloc_hint;
5289         int ret;
5290
5291         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5292
5293         trans = btrfs_join_transaction(root, 0);
5294         if (IS_ERR(trans))
5295                 return ERR_CAST(trans);
5296
5297         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5298
5299         alloc_hint = get_extent_allocation_hint(inode, start, len);
5300         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5301                                    alloc_hint, (u64)-1, &ins, 1);
5302         if (ret) {
5303                 em = ERR_PTR(ret);
5304                 goto out;
5305         }
5306
5307         em = alloc_extent_map(GFP_NOFS);
5308         if (!em) {
5309                 em = ERR_PTR(-ENOMEM);
5310                 goto out;
5311         }
5312
5313         em->start = start;
5314         em->orig_start = em->start;
5315         em->len = ins.offset;
5316
5317         em->block_start = ins.objectid;
5318         em->block_len = ins.offset;
5319         em->bdev = root->fs_info->fs_devices->latest_bdev;
5320         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5321
5322         while (1) {
5323                 write_lock(&em_tree->lock);
5324                 ret = add_extent_mapping(em_tree, em);
5325                 write_unlock(&em_tree->lock);
5326                 if (ret != -EEXIST)
5327                         break;
5328                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5329         }
5330
5331         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5332                                            ins.offset, ins.offset, 0);
5333         if (ret) {
5334                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5335                 em = ERR_PTR(ret);
5336         }
5337 out:
5338         btrfs_end_transaction(trans, root);
5339         return em;
5340 }
5341
5342 /*
5343  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5344  * block must be cow'd
5345  */
5346 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5347                                       struct inode *inode, u64 offset, u64 len)
5348 {
5349         struct btrfs_path *path;
5350         int ret;
5351         struct extent_buffer *leaf;
5352         struct btrfs_root *root = BTRFS_I(inode)->root;
5353         struct btrfs_file_extent_item *fi;
5354         struct btrfs_key key;
5355         u64 disk_bytenr;
5356         u64 backref_offset;
5357         u64 extent_end;
5358         u64 num_bytes;
5359         int slot;
5360         int found_type;
5361
5362         path = btrfs_alloc_path();
5363         if (!path)
5364                 return -ENOMEM;
5365
5366         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5367                                        offset, 0);
5368         if (ret < 0)
5369                 goto out;
5370
5371         slot = path->slots[0];
5372         if (ret == 1) {
5373                 if (slot == 0) {
5374                         /* can't find the item, must cow */
5375                         ret = 0;
5376                         goto out;
5377                 }
5378                 slot--;
5379         }
5380         ret = 0;
5381         leaf = path->nodes[0];
5382         btrfs_item_key_to_cpu(leaf, &key, slot);
5383         if (key.objectid != inode->i_ino ||
5384             key.type != BTRFS_EXTENT_DATA_KEY) {
5385                 /* not our file or wrong item type, must cow */
5386                 goto out;
5387         }
5388
5389         if (key.offset > offset) {
5390                 /* Wrong offset, must cow */
5391                 goto out;
5392         }
5393
5394         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5395         found_type = btrfs_file_extent_type(leaf, fi);
5396         if (found_type != BTRFS_FILE_EXTENT_REG &&
5397             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5398                 /* not a regular extent, must cow */
5399                 goto out;
5400         }
5401         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5402         backref_offset = btrfs_file_extent_offset(leaf, fi);
5403
5404         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5405         if (extent_end < offset + len) {
5406                 /* extent doesn't include our full range, must cow */
5407                 goto out;
5408         }
5409
5410         if (btrfs_extent_readonly(root, disk_bytenr))
5411                 goto out;
5412
5413         /*
5414          * look for other files referencing this extent, if we
5415          * find any we must cow
5416          */
5417         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5418                                   key.offset - backref_offset, disk_bytenr))
5419                 goto out;
5420
5421         /*
5422          * adjust disk_bytenr and num_bytes to cover just the bytes
5423          * in this extent we are about to write.  If there
5424          * are any csums in that range we have to cow in order
5425          * to keep the csums correct
5426          */
5427         disk_bytenr += backref_offset;
5428         disk_bytenr += offset - key.offset;
5429         num_bytes = min(offset + len, extent_end) - offset;
5430         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5431                                 goto out;
5432         /*
5433          * all of the above have passed, it is safe to overwrite this extent
5434          * without cow
5435          */
5436         ret = 1;
5437 out:
5438         btrfs_free_path(path);
5439         return ret;
5440 }
5441
5442 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5443                                    struct buffer_head *bh_result, int create)
5444 {
5445         struct extent_map *em;
5446         struct btrfs_root *root = BTRFS_I(inode)->root;
5447         u64 start = iblock << inode->i_blkbits;
5448         u64 len = bh_result->b_size;
5449         struct btrfs_trans_handle *trans;
5450
5451         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5452         if (IS_ERR(em))
5453                 return PTR_ERR(em);
5454
5455         /*
5456          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5457          * io.  INLINE is special, and we could probably kludge it in here, but
5458          * it's still buffered so for safety lets just fall back to the generic
5459          * buffered path.
5460          *
5461          * For COMPRESSED we _have_ to read the entire extent in so we can
5462          * decompress it, so there will be buffering required no matter what we
5463          * do, so go ahead and fallback to buffered.
5464          *
5465          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5466          * to buffered IO.  Don't blame me, this is the price we pay for using
5467          * the generic code.
5468          */
5469         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5470             em->block_start == EXTENT_MAP_INLINE) {
5471                 free_extent_map(em);
5472                 return -ENOTBLK;
5473         }
5474
5475         /* Just a good old fashioned hole, return */
5476         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5477                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5478                 free_extent_map(em);
5479                 /* DIO will do one hole at a time, so just unlock a sector */
5480                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5481                               start + root->sectorsize - 1, GFP_NOFS);
5482                 return 0;
5483         }
5484
5485         /*
5486          * We don't allocate a new extent in the following cases
5487          *
5488          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5489          * existing extent.
5490          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5491          * just use the extent.
5492          *
5493          */
5494         if (!create) {
5495                 len = em->len - (start - em->start);
5496                 goto map;
5497         }
5498
5499         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5500             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5501              em->block_start != EXTENT_MAP_HOLE)) {
5502                 int type;
5503                 int ret;
5504                 u64 block_start;
5505
5506                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5507                         type = BTRFS_ORDERED_PREALLOC;
5508                 else
5509                         type = BTRFS_ORDERED_NOCOW;
5510                 len = min(len, em->len - (start - em->start));
5511                 block_start = em->block_start + (start - em->start);
5512
5513                 /*
5514                  * we're not going to log anything, but we do need
5515                  * to make sure the current transaction stays open
5516                  * while we look for nocow cross refs
5517                  */
5518                 trans = btrfs_join_transaction(root, 0);
5519                 if (IS_ERR(trans))
5520                         goto must_cow;
5521
5522                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5523                         ret = btrfs_add_ordered_extent_dio(inode, start,
5524                                            block_start, len, len, type);
5525                         btrfs_end_transaction(trans, root);
5526                         if (ret) {
5527                                 free_extent_map(em);
5528                                 return ret;
5529                         }
5530                         goto unlock;
5531                 }
5532                 btrfs_end_transaction(trans, root);
5533         }
5534 must_cow:
5535         /*
5536          * this will cow the extent, reset the len in case we changed
5537          * it above
5538          */
5539         len = bh_result->b_size;
5540         free_extent_map(em);
5541         em = btrfs_new_extent_direct(inode, start, len);
5542         if (IS_ERR(em))
5543                 return PTR_ERR(em);
5544         len = min(len, em->len - (start - em->start));
5545 unlock:
5546         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5547                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5548                           0, NULL, GFP_NOFS);
5549 map:
5550         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5551                 inode->i_blkbits;
5552         bh_result->b_size = len;
5553         bh_result->b_bdev = em->bdev;
5554         set_buffer_mapped(bh_result);
5555         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5556                 set_buffer_new(bh_result);
5557
5558         free_extent_map(em);
5559
5560         return 0;
5561 }
5562
5563 struct btrfs_dio_private {
5564         struct inode *inode;
5565         u64 logical_offset;
5566         u64 disk_bytenr;
5567         u64 bytes;
5568         u32 *csums;
5569         void *private;
5570
5571         /* number of bios pending for this dio */
5572         atomic_t pending_bios;
5573
5574         /* IO errors */
5575         int errors;
5576
5577         struct bio *orig_bio;
5578 };
5579
5580 static void btrfs_endio_direct_read(struct bio *bio, int err)
5581 {
5582         struct btrfs_dio_private *dip = bio->bi_private;
5583         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5584         struct bio_vec *bvec = bio->bi_io_vec;
5585         struct inode *inode = dip->inode;
5586         struct btrfs_root *root = BTRFS_I(inode)->root;
5587         u64 start;
5588         u32 *private = dip->csums;
5589
5590         start = dip->logical_offset;
5591         do {
5592                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5593                         struct page *page = bvec->bv_page;
5594                         char *kaddr;
5595                         u32 csum = ~(u32)0;
5596                         unsigned long flags;
5597
5598                         local_irq_save(flags);
5599                         kaddr = kmap_atomic(page, KM_IRQ0);
5600                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5601                                                csum, bvec->bv_len);
5602                         btrfs_csum_final(csum, (char *)&csum);
5603                         kunmap_atomic(kaddr, KM_IRQ0);
5604                         local_irq_restore(flags);
5605
5606                         flush_dcache_page(bvec->bv_page);
5607                         if (csum != *private) {
5608                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5609                                       " %llu csum %u private %u\n",
5610                                       inode->i_ino, (unsigned long long)start,
5611                                       csum, *private);
5612                                 err = -EIO;
5613                         }
5614                 }
5615
5616                 start += bvec->bv_len;
5617                 private++;
5618                 bvec++;
5619         } while (bvec <= bvec_end);
5620
5621         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5622                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5623         bio->bi_private = dip->private;
5624
5625         kfree(dip->csums);
5626         kfree(dip);
5627         dio_end_io(bio, err);
5628 }
5629
5630 static void btrfs_endio_direct_write(struct bio *bio, int err)
5631 {
5632         struct btrfs_dio_private *dip = bio->bi_private;
5633         struct inode *inode = dip->inode;
5634         struct btrfs_root *root = BTRFS_I(inode)->root;
5635         struct btrfs_trans_handle *trans;
5636         struct btrfs_ordered_extent *ordered = NULL;
5637         struct extent_state *cached_state = NULL;
5638         u64 ordered_offset = dip->logical_offset;
5639         u64 ordered_bytes = dip->bytes;
5640         int ret;
5641
5642         if (err)
5643                 goto out_done;
5644 again:
5645         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5646                                                    &ordered_offset,
5647                                                    ordered_bytes);
5648         if (!ret)
5649                 goto out_test;
5650
5651         BUG_ON(!ordered);
5652
5653         trans = btrfs_join_transaction(root, 1);
5654         if (IS_ERR(trans)) {
5655                 err = -ENOMEM;
5656                 goto out;
5657         }
5658         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5659
5660         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5661                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5662                 if (!ret)
5663                         ret = btrfs_update_inode(trans, root, inode);
5664                 err =