Btrfs: kill BTRFS_I(inode)->block_group
[pandora-kernel.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 #include "free-space-cache.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91                                    struct page *locked_page,
92                                    u64 start, u64 end, int *page_started,
93                                    unsigned long *nr_written, int unlock);
94
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96                                      struct inode *inode,  struct inode *dir,
97                                      const struct qstr *qstr)
98 {
99         int err;
100
101         err = btrfs_init_acl(trans, inode, dir);
102         if (!err)
103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
104         return err;
105 }
106
107 /*
108  * this does all the hard work for inserting an inline extent into
109  * the btree.  The caller should have done a btrfs_drop_extents so that
110  * no overlapping inline items exist in the btree
111  */
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113                                 struct btrfs_root *root, struct inode *inode,
114                                 u64 start, size_t size, size_t compressed_size,
115                                 int compress_type,
116                                 struct page **compressed_pages)
117 {
118         struct btrfs_key key;
119         struct btrfs_path *path;
120         struct extent_buffer *leaf;
121         struct page *page = NULL;
122         char *kaddr;
123         unsigned long ptr;
124         struct btrfs_file_extent_item *ei;
125         int err = 0;
126         int ret;
127         size_t cur_size = size;
128         size_t datasize;
129         unsigned long offset;
130
131         if (compressed_size && compressed_pages)
132                 cur_size = compressed_size;
133
134         path = btrfs_alloc_path();
135         if (!path)
136                 return -ENOMEM;
137
138         path->leave_spinning = 1;
139
140         key.objectid = inode->i_ino;
141         key.offset = start;
142         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143         datasize = btrfs_file_extent_calc_inline_size(cur_size);
144
145         inode_add_bytes(inode, size);
146         ret = btrfs_insert_empty_item(trans, root, path, &key,
147                                       datasize);
148         BUG_ON(ret);
149         if (ret) {
150                 err = ret;
151                 goto fail;
152         }
153         leaf = path->nodes[0];
154         ei = btrfs_item_ptr(leaf, path->slots[0],
155                             struct btrfs_file_extent_item);
156         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158         btrfs_set_file_extent_encryption(leaf, ei, 0);
159         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161         ptr = btrfs_file_extent_inline_start(ei);
162
163         if (compress_type != BTRFS_COMPRESS_NONE) {
164                 struct page *cpage;
165                 int i = 0;
166                 while (compressed_size > 0) {
167                         cpage = compressed_pages[i];
168                         cur_size = min_t(unsigned long, compressed_size,
169                                        PAGE_CACHE_SIZE);
170
171                         kaddr = kmap_atomic(cpage, KM_USER0);
172                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
173                         kunmap_atomic(kaddr, KM_USER0);
174
175                         i++;
176                         ptr += cur_size;
177                         compressed_size -= cur_size;
178                 }
179                 btrfs_set_file_extent_compression(leaf, ei,
180                                                   compress_type);
181         } else {
182                 page = find_get_page(inode->i_mapping,
183                                      start >> PAGE_CACHE_SHIFT);
184                 btrfs_set_file_extent_compression(leaf, ei, 0);
185                 kaddr = kmap_atomic(page, KM_USER0);
186                 offset = start & (PAGE_CACHE_SIZE - 1);
187                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188                 kunmap_atomic(kaddr, KM_USER0);
189                 page_cache_release(page);
190         }
191         btrfs_mark_buffer_dirty(leaf);
192         btrfs_free_path(path);
193
194         /*
195          * we're an inline extent, so nobody can
196          * extend the file past i_size without locking
197          * a page we already have locked.
198          *
199          * We must do any isize and inode updates
200          * before we unlock the pages.  Otherwise we
201          * could end up racing with unlink.
202          */
203         BTRFS_I(inode)->disk_i_size = inode->i_size;
204         btrfs_update_inode(trans, root, inode);
205
206         return 0;
207 fail:
208         btrfs_free_path(path);
209         return err;
210 }
211
212
213 /*
214  * conditionally insert an inline extent into the file.  This
215  * does the checks required to make sure the data is small enough
216  * to fit as an inline extent.
217  */
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219                                  struct btrfs_root *root,
220                                  struct inode *inode, u64 start, u64 end,
221                                  size_t compressed_size, int compress_type,
222                                  struct page **compressed_pages)
223 {
224         u64 isize = i_size_read(inode);
225         u64 actual_end = min(end + 1, isize);
226         u64 inline_len = actual_end - start;
227         u64 aligned_end = (end + root->sectorsize - 1) &
228                         ~((u64)root->sectorsize - 1);
229         u64 hint_byte;
230         u64 data_len = inline_len;
231         int ret;
232
233         if (compressed_size)
234                 data_len = compressed_size;
235
236         if (start > 0 ||
237             actual_end >= PAGE_CACHE_SIZE ||
238             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239             (!compressed_size &&
240             (actual_end & (root->sectorsize - 1)) == 0) ||
241             end + 1 < isize ||
242             data_len > root->fs_info->max_inline) {
243                 return 1;
244         }
245
246         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
247                                  &hint_byte, 1);
248         BUG_ON(ret);
249
250         if (isize > actual_end)
251                 inline_len = min_t(u64, isize, actual_end);
252         ret = insert_inline_extent(trans, root, inode, start,
253                                    inline_len, compressed_size,
254                                    compress_type, compressed_pages);
255         BUG_ON(ret);
256         btrfs_delalloc_release_metadata(inode, end + 1 - start);
257         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
258         return 0;
259 }
260
261 struct async_extent {
262         u64 start;
263         u64 ram_size;
264         u64 compressed_size;
265         struct page **pages;
266         unsigned long nr_pages;
267         int compress_type;
268         struct list_head list;
269 };
270
271 struct async_cow {
272         struct inode *inode;
273         struct btrfs_root *root;
274         struct page *locked_page;
275         u64 start;
276         u64 end;
277         struct list_head extents;
278         struct btrfs_work work;
279 };
280
281 static noinline int add_async_extent(struct async_cow *cow,
282                                      u64 start, u64 ram_size,
283                                      u64 compressed_size,
284                                      struct page **pages,
285                                      unsigned long nr_pages,
286                                      int compress_type)
287 {
288         struct async_extent *async_extent;
289
290         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291         BUG_ON(!async_extent);
292         async_extent->start = start;
293         async_extent->ram_size = ram_size;
294         async_extent->compressed_size = compressed_size;
295         async_extent->pages = pages;
296         async_extent->nr_pages = nr_pages;
297         async_extent->compress_type = compress_type;
298         list_add_tail(&async_extent->list, &cow->extents);
299         return 0;
300 }
301
302 /*
303  * we create compressed extents in two phases.  The first
304  * phase compresses a range of pages that have already been
305  * locked (both pages and state bits are locked).
306  *
307  * This is done inside an ordered work queue, and the compression
308  * is spread across many cpus.  The actual IO submission is step
309  * two, and the ordered work queue takes care of making sure that
310  * happens in the same order things were put onto the queue by
311  * writepages and friends.
312  *
313  * If this code finds it can't get good compression, it puts an
314  * entry onto the work queue to write the uncompressed bytes.  This
315  * makes sure that both compressed inodes and uncompressed inodes
316  * are written in the same order that pdflush sent them down.
317  */
318 static noinline int compress_file_range(struct inode *inode,
319                                         struct page *locked_page,
320                                         u64 start, u64 end,
321                                         struct async_cow *async_cow,
322                                         int *num_added)
323 {
324         struct btrfs_root *root = BTRFS_I(inode)->root;
325         struct btrfs_trans_handle *trans;
326         u64 num_bytes;
327         u64 blocksize = root->sectorsize;
328         u64 actual_end;
329         u64 isize = i_size_read(inode);
330         int ret = 0;
331         struct page **pages = NULL;
332         unsigned long nr_pages;
333         unsigned long nr_pages_ret = 0;
334         unsigned long total_compressed = 0;
335         unsigned long total_in = 0;
336         unsigned long max_compressed = 128 * 1024;
337         unsigned long max_uncompressed = 128 * 1024;
338         int i;
339         int will_compress;
340         int compress_type = root->fs_info->compress_type;
341
342         actual_end = min_t(u64, isize, end + 1);
343 again:
344         will_compress = 0;
345         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
346         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
347
348         /*
349          * we don't want to send crud past the end of i_size through
350          * compression, that's just a waste of CPU time.  So, if the
351          * end of the file is before the start of our current
352          * requested range of bytes, we bail out to the uncompressed
353          * cleanup code that can deal with all of this.
354          *
355          * It isn't really the fastest way to fix things, but this is a
356          * very uncommon corner.
357          */
358         if (actual_end <= start)
359                 goto cleanup_and_bail_uncompressed;
360
361         total_compressed = actual_end - start;
362
363         /* we want to make sure that amount of ram required to uncompress
364          * an extent is reasonable, so we limit the total size in ram
365          * of a compressed extent to 128k.  This is a crucial number
366          * because it also controls how easily we can spread reads across
367          * cpus for decompression.
368          *
369          * We also want to make sure the amount of IO required to do
370          * a random read is reasonably small, so we limit the size of
371          * a compressed extent to 128k.
372          */
373         total_compressed = min(total_compressed, max_uncompressed);
374         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
375         num_bytes = max(blocksize,  num_bytes);
376         total_in = 0;
377         ret = 0;
378
379         /*
380          * we do compression for mount -o compress and when the
381          * inode has not been flagged as nocompress.  This flag can
382          * change at any time if we discover bad compression ratios.
383          */
384         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
385             (btrfs_test_opt(root, COMPRESS) ||
386              (BTRFS_I(inode)->force_compress) ||
387              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
388                 WARN_ON(pages);
389                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
390                 BUG_ON(!pages);
391
392                 if (BTRFS_I(inode)->force_compress)
393                         compress_type = BTRFS_I(inode)->force_compress;
394
395                 ret = btrfs_compress_pages(compress_type,
396                                            inode->i_mapping, start,
397                                            total_compressed, pages,
398                                            nr_pages, &nr_pages_ret,
399                                            &total_in,
400                                            &total_compressed,
401                                            max_compressed);
402
403                 if (!ret) {
404                         unsigned long offset = total_compressed &
405                                 (PAGE_CACHE_SIZE - 1);
406                         struct page *page = pages[nr_pages_ret - 1];
407                         char *kaddr;
408
409                         /* zero the tail end of the last page, we might be
410                          * sending it down to disk
411                          */
412                         if (offset) {
413                                 kaddr = kmap_atomic(page, KM_USER0);
414                                 memset(kaddr + offset, 0,
415                                        PAGE_CACHE_SIZE - offset);
416                                 kunmap_atomic(kaddr, KM_USER0);
417                         }
418                         will_compress = 1;
419                 }
420         }
421         if (start == 0) {
422                 trans = btrfs_join_transaction(root);
423                 BUG_ON(IS_ERR(trans));
424                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
425
426                 /* lets try to make an inline extent */
427                 if (ret || total_in < (actual_end - start)) {
428                         /* we didn't compress the entire range, try
429                          * to make an uncompressed inline extent.
430                          */
431                         ret = cow_file_range_inline(trans, root, inode,
432                                                     start, end, 0, 0, NULL);
433                 } else {
434                         /* try making a compressed inline extent */
435                         ret = cow_file_range_inline(trans, root, inode,
436                                                     start, end,
437                                                     total_compressed,
438                                                     compress_type, pages);
439                 }
440                 if (ret == 0) {
441                         /*
442                          * inline extent creation worked, we don't need
443                          * to create any more async work items.  Unlock
444                          * and free up our temp pages.
445                          */
446                         extent_clear_unlock_delalloc(inode,
447                              &BTRFS_I(inode)->io_tree,
448                              start, end, NULL,
449                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
450                              EXTENT_CLEAR_DELALLOC |
451                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
452
453                         btrfs_end_transaction(trans, root);
454                         goto free_pages_out;
455                 }
456                 btrfs_end_transaction(trans, root);
457         }
458
459         if (will_compress) {
460                 /*
461                  * we aren't doing an inline extent round the compressed size
462                  * up to a block size boundary so the allocator does sane
463                  * things
464                  */
465                 total_compressed = (total_compressed + blocksize - 1) &
466                         ~(blocksize - 1);
467
468                 /*
469                  * one last check to make sure the compression is really a
470                  * win, compare the page count read with the blocks on disk
471                  */
472                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
473                         ~(PAGE_CACHE_SIZE - 1);
474                 if (total_compressed >= total_in) {
475                         will_compress = 0;
476                 } else {
477                         num_bytes = total_in;
478                 }
479         }
480         if (!will_compress && pages) {
481                 /*
482                  * the compression code ran but failed to make things smaller,
483                  * free any pages it allocated and our page pointer array
484                  */
485                 for (i = 0; i < nr_pages_ret; i++) {
486                         WARN_ON(pages[i]->mapping);
487                         page_cache_release(pages[i]);
488                 }
489                 kfree(pages);
490                 pages = NULL;
491                 total_compressed = 0;
492                 nr_pages_ret = 0;
493
494                 /* flag the file so we don't compress in the future */
495                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
496                     !(BTRFS_I(inode)->force_compress)) {
497                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
498                 }
499         }
500         if (will_compress) {
501                 *num_added += 1;
502
503                 /* the async work queues will take care of doing actual
504                  * allocation on disk for these compressed pages,
505                  * and will submit them to the elevator.
506                  */
507                 add_async_extent(async_cow, start, num_bytes,
508                                  total_compressed, pages, nr_pages_ret,
509                                  compress_type);
510
511                 if (start + num_bytes < end) {
512                         start += num_bytes;
513                         pages = NULL;
514                         cond_resched();
515                         goto again;
516                 }
517         } else {
518 cleanup_and_bail_uncompressed:
519                 /*
520                  * No compression, but we still need to write the pages in
521                  * the file we've been given so far.  redirty the locked
522                  * page if it corresponds to our extent and set things up
523                  * for the async work queue to run cow_file_range to do
524                  * the normal delalloc dance
525                  */
526                 if (page_offset(locked_page) >= start &&
527                     page_offset(locked_page) <= end) {
528                         __set_page_dirty_nobuffers(locked_page);
529                         /* unlocked later on in the async handlers */
530                 }
531                 add_async_extent(async_cow, start, end - start + 1,
532                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
533                 *num_added += 1;
534         }
535
536 out:
537         return 0;
538
539 free_pages_out:
540         for (i = 0; i < nr_pages_ret; i++) {
541                 WARN_ON(pages[i]->mapping);
542                 page_cache_release(pages[i]);
543         }
544         kfree(pages);
545
546         goto out;
547 }
548
549 /*
550  * phase two of compressed writeback.  This is the ordered portion
551  * of the code, which only gets called in the order the work was
552  * queued.  We walk all the async extents created by compress_file_range
553  * and send them down to the disk.
554  */
555 static noinline int submit_compressed_extents(struct inode *inode,
556                                               struct async_cow *async_cow)
557 {
558         struct async_extent *async_extent;
559         u64 alloc_hint = 0;
560         struct btrfs_trans_handle *trans;
561         struct btrfs_key ins;
562         struct extent_map *em;
563         struct btrfs_root *root = BTRFS_I(inode)->root;
564         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
565         struct extent_io_tree *io_tree;
566         int ret = 0;
567
568         if (list_empty(&async_cow->extents))
569                 return 0;
570
571
572         while (!list_empty(&async_cow->extents)) {
573                 async_extent = list_entry(async_cow->extents.next,
574                                           struct async_extent, list);
575                 list_del(&async_extent->list);
576
577                 io_tree = &BTRFS_I(inode)->io_tree;
578
579 retry:
580                 /* did the compression code fall back to uncompressed IO? */
581                 if (!async_extent->pages) {
582                         int page_started = 0;
583                         unsigned long nr_written = 0;
584
585                         lock_extent(io_tree, async_extent->start,
586                                          async_extent->start +
587                                          async_extent->ram_size - 1, GFP_NOFS);
588
589                         /* allocate blocks */
590                         ret = cow_file_range(inode, async_cow->locked_page,
591                                              async_extent->start,
592                                              async_extent->start +
593                                              async_extent->ram_size - 1,
594                                              &page_started, &nr_written, 0);
595
596                         /*
597                          * if page_started, cow_file_range inserted an
598                          * inline extent and took care of all the unlocking
599                          * and IO for us.  Otherwise, we need to submit
600                          * all those pages down to the drive.
601                          */
602                         if (!page_started && !ret)
603                                 extent_write_locked_range(io_tree,
604                                                   inode, async_extent->start,
605                                                   async_extent->start +
606                                                   async_extent->ram_size - 1,
607                                                   btrfs_get_extent,
608                                                   WB_SYNC_ALL);
609                         kfree(async_extent);
610                         cond_resched();
611                         continue;
612                 }
613
614                 lock_extent(io_tree, async_extent->start,
615                             async_extent->start + async_extent->ram_size - 1,
616                             GFP_NOFS);
617
618                 trans = btrfs_join_transaction(root);
619                 BUG_ON(IS_ERR(trans));
620                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
621                 ret = btrfs_reserve_extent(trans, root,
622                                            async_extent->compressed_size,
623                                            async_extent->compressed_size,
624                                            0, alloc_hint,
625                                            (u64)-1, &ins, 1);
626                 btrfs_end_transaction(trans, root);
627
628                 if (ret) {
629                         int i;
630                         for (i = 0; i < async_extent->nr_pages; i++) {
631                                 WARN_ON(async_extent->pages[i]->mapping);
632                                 page_cache_release(async_extent->pages[i]);
633                         }
634                         kfree(async_extent->pages);
635                         async_extent->nr_pages = 0;
636                         async_extent->pages = NULL;
637                         unlock_extent(io_tree, async_extent->start,
638                                       async_extent->start +
639                                       async_extent->ram_size - 1, GFP_NOFS);
640                         goto retry;
641                 }
642
643                 /*
644                  * here we're doing allocation and writeback of the
645                  * compressed pages
646                  */
647                 btrfs_drop_extent_cache(inode, async_extent->start,
648                                         async_extent->start +
649                                         async_extent->ram_size - 1, 0);
650
651                 em = alloc_extent_map(GFP_NOFS);
652                 BUG_ON(!em);
653                 em->start = async_extent->start;
654                 em->len = async_extent->ram_size;
655                 em->orig_start = em->start;
656
657                 em->block_start = ins.objectid;
658                 em->block_len = ins.offset;
659                 em->bdev = root->fs_info->fs_devices->latest_bdev;
660                 em->compress_type = async_extent->compress_type;
661                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
662                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
663
664                 while (1) {
665                         write_lock(&em_tree->lock);
666                         ret = add_extent_mapping(em_tree, em);
667                         write_unlock(&em_tree->lock);
668                         if (ret != -EEXIST) {
669                                 free_extent_map(em);
670                                 break;
671                         }
672                         btrfs_drop_extent_cache(inode, async_extent->start,
673                                                 async_extent->start +
674                                                 async_extent->ram_size - 1, 0);
675                 }
676
677                 ret = btrfs_add_ordered_extent_compress(inode,
678                                                 async_extent->start,
679                                                 ins.objectid,
680                                                 async_extent->ram_size,
681                                                 ins.offset,
682                                                 BTRFS_ORDERED_COMPRESSED,
683                                                 async_extent->compress_type);
684                 BUG_ON(ret);
685
686                 /*
687                  * clear dirty, set writeback and unlock the pages.
688                  */
689                 extent_clear_unlock_delalloc(inode,
690                                 &BTRFS_I(inode)->io_tree,
691                                 async_extent->start,
692                                 async_extent->start +
693                                 async_extent->ram_size - 1,
694                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
695                                 EXTENT_CLEAR_UNLOCK |
696                                 EXTENT_CLEAR_DELALLOC |
697                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
698
699                 ret = btrfs_submit_compressed_write(inode,
700                                     async_extent->start,
701                                     async_extent->ram_size,
702                                     ins.objectid,
703                                     ins.offset, async_extent->pages,
704                                     async_extent->nr_pages);
705
706                 BUG_ON(ret);
707                 alloc_hint = ins.objectid + ins.offset;
708                 kfree(async_extent);
709                 cond_resched();
710         }
711
712         return 0;
713 }
714
715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
716                                       u64 num_bytes)
717 {
718         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719         struct extent_map *em;
720         u64 alloc_hint = 0;
721
722         read_lock(&em_tree->lock);
723         em = search_extent_mapping(em_tree, start, num_bytes);
724         if (em) {
725                 /*
726                  * if block start isn't an actual block number then find the
727                  * first block in this inode and use that as a hint.  If that
728                  * block is also bogus then just don't worry about it.
729                  */
730                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
731                         free_extent_map(em);
732                         em = search_extent_mapping(em_tree, 0, 0);
733                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
734                                 alloc_hint = em->block_start;
735                         if (em)
736                                 free_extent_map(em);
737                 } else {
738                         alloc_hint = em->block_start;
739                         free_extent_map(em);
740                 }
741         }
742         read_unlock(&em_tree->lock);
743
744         return alloc_hint;
745 }
746
747 /*
748  * when extent_io.c finds a delayed allocation range in the file,
749  * the call backs end up in this code.  The basic idea is to
750  * allocate extents on disk for the range, and create ordered data structs
751  * in ram to track those extents.
752  *
753  * locked_page is the page that writepage had locked already.  We use
754  * it to make sure we don't do extra locks or unlocks.
755  *
756  * *page_started is set to one if we unlock locked_page and do everything
757  * required to start IO on it.  It may be clean and already done with
758  * IO when we return.
759  */
760 static noinline int cow_file_range(struct inode *inode,
761                                    struct page *locked_page,
762                                    u64 start, u64 end, int *page_started,
763                                    unsigned long *nr_written,
764                                    int unlock)
765 {
766         struct btrfs_root *root = BTRFS_I(inode)->root;
767         struct btrfs_trans_handle *trans;
768         u64 alloc_hint = 0;
769         u64 num_bytes;
770         unsigned long ram_size;
771         u64 disk_num_bytes;
772         u64 cur_alloc_size;
773         u64 blocksize = root->sectorsize;
774         struct btrfs_key ins;
775         struct extent_map *em;
776         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
777         int ret = 0;
778
779         BUG_ON(root == root->fs_info->tree_root);
780         trans = btrfs_join_transaction(root);
781         BUG_ON(IS_ERR(trans));
782         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
783
784         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
785         num_bytes = max(blocksize,  num_bytes);
786         disk_num_bytes = num_bytes;
787         ret = 0;
788
789         if (start == 0) {
790                 /* lets try to make an inline extent */
791                 ret = cow_file_range_inline(trans, root, inode,
792                                             start, end, 0, 0, NULL);
793                 if (ret == 0) {
794                         extent_clear_unlock_delalloc(inode,
795                                      &BTRFS_I(inode)->io_tree,
796                                      start, end, NULL,
797                                      EXTENT_CLEAR_UNLOCK_PAGE |
798                                      EXTENT_CLEAR_UNLOCK |
799                                      EXTENT_CLEAR_DELALLOC |
800                                      EXTENT_CLEAR_DIRTY |
801                                      EXTENT_SET_WRITEBACK |
802                                      EXTENT_END_WRITEBACK);
803
804                         *nr_written = *nr_written +
805                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
806                         *page_started = 1;
807                         ret = 0;
808                         goto out;
809                 }
810         }
811
812         BUG_ON(disk_num_bytes >
813                btrfs_super_total_bytes(&root->fs_info->super_copy));
814
815         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
816         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
817
818         while (disk_num_bytes > 0) {
819                 unsigned long op;
820
821                 cur_alloc_size = disk_num_bytes;
822                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
823                                            root->sectorsize, 0, alloc_hint,
824                                            (u64)-1, &ins, 1);
825                 BUG_ON(ret);
826
827                 em = alloc_extent_map(GFP_NOFS);
828                 BUG_ON(!em);
829                 em->start = start;
830                 em->orig_start = em->start;
831                 ram_size = ins.offset;
832                 em->len = ins.offset;
833
834                 em->block_start = ins.objectid;
835                 em->block_len = ins.offset;
836                 em->bdev = root->fs_info->fs_devices->latest_bdev;
837                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
838
839                 while (1) {
840                         write_lock(&em_tree->lock);
841                         ret = add_extent_mapping(em_tree, em);
842                         write_unlock(&em_tree->lock);
843                         if (ret != -EEXIST) {
844                                 free_extent_map(em);
845                                 break;
846                         }
847                         btrfs_drop_extent_cache(inode, start,
848                                                 start + ram_size - 1, 0);
849                 }
850
851                 cur_alloc_size = ins.offset;
852                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
853                                                ram_size, cur_alloc_size, 0);
854                 BUG_ON(ret);
855
856                 if (root->root_key.objectid ==
857                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
858                         ret = btrfs_reloc_clone_csums(inode, start,
859                                                       cur_alloc_size);
860                         BUG_ON(ret);
861                 }
862
863                 if (disk_num_bytes < cur_alloc_size)
864                         break;
865
866                 /* we're not doing compressed IO, don't unlock the first
867                  * page (which the caller expects to stay locked), don't
868                  * clear any dirty bits and don't set any writeback bits
869                  *
870                  * Do set the Private2 bit so we know this page was properly
871                  * setup for writepage
872                  */
873                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
874                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
875                         EXTENT_SET_PRIVATE2;
876
877                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
878                                              start, start + ram_size - 1,
879                                              locked_page, op);
880                 disk_num_bytes -= cur_alloc_size;
881                 num_bytes -= cur_alloc_size;
882                 alloc_hint = ins.objectid + ins.offset;
883                 start += cur_alloc_size;
884         }
885 out:
886         ret = 0;
887         btrfs_end_transaction(trans, root);
888
889         return ret;
890 }
891
892 /*
893  * work queue call back to started compression on a file and pages
894  */
895 static noinline void async_cow_start(struct btrfs_work *work)
896 {
897         struct async_cow *async_cow;
898         int num_added = 0;
899         async_cow = container_of(work, struct async_cow, work);
900
901         compress_file_range(async_cow->inode, async_cow->locked_page,
902                             async_cow->start, async_cow->end, async_cow,
903                             &num_added);
904         if (num_added == 0)
905                 async_cow->inode = NULL;
906 }
907
908 /*
909  * work queue call back to submit previously compressed pages
910  */
911 static noinline void async_cow_submit(struct btrfs_work *work)
912 {
913         struct async_cow *async_cow;
914         struct btrfs_root *root;
915         unsigned long nr_pages;
916
917         async_cow = container_of(work, struct async_cow, work);
918
919         root = async_cow->root;
920         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
921                 PAGE_CACHE_SHIFT;
922
923         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
924
925         if (atomic_read(&root->fs_info->async_delalloc_pages) <
926             5 * 1042 * 1024 &&
927             waitqueue_active(&root->fs_info->async_submit_wait))
928                 wake_up(&root->fs_info->async_submit_wait);
929
930         if (async_cow->inode)
931                 submit_compressed_extents(async_cow->inode, async_cow);
932 }
933
934 static noinline void async_cow_free(struct btrfs_work *work)
935 {
936         struct async_cow *async_cow;
937         async_cow = container_of(work, struct async_cow, work);
938         kfree(async_cow);
939 }
940
941 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
942                                 u64 start, u64 end, int *page_started,
943                                 unsigned long *nr_written)
944 {
945         struct async_cow *async_cow;
946         struct btrfs_root *root = BTRFS_I(inode)->root;
947         unsigned long nr_pages;
948         u64 cur_end;
949         int limit = 10 * 1024 * 1042;
950
951         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
952                          1, 0, NULL, GFP_NOFS);
953         while (start < end) {
954                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
955                 BUG_ON(!async_cow);
956                 async_cow->inode = inode;
957                 async_cow->root = root;
958                 async_cow->locked_page = locked_page;
959                 async_cow->start = start;
960
961                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
962                         cur_end = end;
963                 else
964                         cur_end = min(end, start + 512 * 1024 - 1);
965
966                 async_cow->end = cur_end;
967                 INIT_LIST_HEAD(&async_cow->extents);
968
969                 async_cow->work.func = async_cow_start;
970                 async_cow->work.ordered_func = async_cow_submit;
971                 async_cow->work.ordered_free = async_cow_free;
972                 async_cow->work.flags = 0;
973
974                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
975                         PAGE_CACHE_SHIFT;
976                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
977
978                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
979                                    &async_cow->work);
980
981                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
982                         wait_event(root->fs_info->async_submit_wait,
983                            (atomic_read(&root->fs_info->async_delalloc_pages) <
984                             limit));
985                 }
986
987                 while (atomic_read(&root->fs_info->async_submit_draining) &&
988                       atomic_read(&root->fs_info->async_delalloc_pages)) {
989                         wait_event(root->fs_info->async_submit_wait,
990                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
991                            0));
992                 }
993
994                 *nr_written += nr_pages;
995                 start = cur_end + 1;
996         }
997         *page_started = 1;
998         return 0;
999 }
1000
1001 static noinline int csum_exist_in_range(struct btrfs_root *root,
1002                                         u64 bytenr, u64 num_bytes)
1003 {
1004         int ret;
1005         struct btrfs_ordered_sum *sums;
1006         LIST_HEAD(list);
1007
1008         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1009                                        bytenr + num_bytes - 1, &list);
1010         if (ret == 0 && list_empty(&list))
1011                 return 0;
1012
1013         while (!list_empty(&list)) {
1014                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1015                 list_del(&sums->list);
1016                 kfree(sums);
1017         }
1018         return 1;
1019 }
1020
1021 /*
1022  * when nowcow writeback call back.  This checks for snapshots or COW copies
1023  * of the extents that exist in the file, and COWs the file as required.
1024  *
1025  * If no cow copies or snapshots exist, we write directly to the existing
1026  * blocks on disk
1027  */
1028 static noinline int run_delalloc_nocow(struct inode *inode,
1029                                        struct page *locked_page,
1030                               u64 start, u64 end, int *page_started, int force,
1031                               unsigned long *nr_written)
1032 {
1033         struct btrfs_root *root = BTRFS_I(inode)->root;
1034         struct btrfs_trans_handle *trans;
1035         struct extent_buffer *leaf;
1036         struct btrfs_path *path;
1037         struct btrfs_file_extent_item *fi;
1038         struct btrfs_key found_key;
1039         u64 cow_start;
1040         u64 cur_offset;
1041         u64 extent_end;
1042         u64 extent_offset;
1043         u64 disk_bytenr;
1044         u64 num_bytes;
1045         int extent_type;
1046         int ret;
1047         int type;
1048         int nocow;
1049         int check_prev = 1;
1050         bool nolock = false;
1051
1052         path = btrfs_alloc_path();
1053         BUG_ON(!path);
1054         if (root == root->fs_info->tree_root) {
1055                 nolock = true;
1056                 trans = btrfs_join_transaction_nolock(root);
1057         } else {
1058                 trans = btrfs_join_transaction(root);
1059         }
1060         BUG_ON(IS_ERR(trans));
1061         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1062
1063         cow_start = (u64)-1;
1064         cur_offset = start;
1065         while (1) {
1066                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1067                                                cur_offset, 0);
1068                 BUG_ON(ret < 0);
1069                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070                         leaf = path->nodes[0];
1071                         btrfs_item_key_to_cpu(leaf, &found_key,
1072                                               path->slots[0] - 1);
1073                         if (found_key.objectid == inode->i_ino &&
1074                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1075                                 path->slots[0]--;
1076                 }
1077                 check_prev = 0;
1078 next_slot:
1079                 leaf = path->nodes[0];
1080                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081                         ret = btrfs_next_leaf(root, path);
1082                         if (ret < 0)
1083                                 BUG_ON(1);
1084                         if (ret > 0)
1085                                 break;
1086                         leaf = path->nodes[0];
1087                 }
1088
1089                 nocow = 0;
1090                 disk_bytenr = 0;
1091                 num_bytes = 0;
1092                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093
1094                 if (found_key.objectid > inode->i_ino ||
1095                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096                     found_key.offset > end)
1097                         break;
1098
1099                 if (found_key.offset > cur_offset) {
1100                         extent_end = found_key.offset;
1101                         extent_type = 0;
1102                         goto out_check;
1103                 }
1104
1105                 fi = btrfs_item_ptr(leaf, path->slots[0],
1106                                     struct btrfs_file_extent_item);
1107                 extent_type = btrfs_file_extent_type(leaf, fi);
1108
1109                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1113                         extent_end = found_key.offset +
1114                                 btrfs_file_extent_num_bytes(leaf, fi);
1115                         if (extent_end <= start) {
1116                                 path->slots[0]++;
1117                                 goto next_slot;
1118                         }
1119                         if (disk_bytenr == 0)
1120                                 goto out_check;
1121                         if (btrfs_file_extent_compression(leaf, fi) ||
1122                             btrfs_file_extent_encryption(leaf, fi) ||
1123                             btrfs_file_extent_other_encoding(leaf, fi))
1124                                 goto out_check;
1125                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126                                 goto out_check;
1127                         if (btrfs_extent_readonly(root, disk_bytenr))
1128                                 goto out_check;
1129                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130                                                   found_key.offset -
1131                                                   extent_offset, disk_bytenr))
1132                                 goto out_check;
1133                         disk_bytenr += extent_offset;
1134                         disk_bytenr += cur_offset - found_key.offset;
1135                         num_bytes = min(end + 1, extent_end) - cur_offset;
1136                         /*
1137                          * force cow if csum exists in the range.
1138                          * this ensure that csum for a given extent are
1139                          * either valid or do not exist.
1140                          */
1141                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1142                                 goto out_check;
1143                         nocow = 1;
1144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145                         extent_end = found_key.offset +
1146                                 btrfs_file_extent_inline_len(leaf, fi);
1147                         extent_end = ALIGN(extent_end, root->sectorsize);
1148                 } else {
1149                         BUG_ON(1);
1150                 }
1151 out_check:
1152                 if (extent_end <= start) {
1153                         path->slots[0]++;
1154                         goto next_slot;
1155                 }
1156                 if (!nocow) {
1157                         if (cow_start == (u64)-1)
1158                                 cow_start = cur_offset;
1159                         cur_offset = extent_end;
1160                         if (cur_offset > end)
1161                                 break;
1162                         path->slots[0]++;
1163                         goto next_slot;
1164                 }
1165
1166                 btrfs_release_path(root, path);
1167                 if (cow_start != (u64)-1) {
1168                         ret = cow_file_range(inode, locked_page, cow_start,
1169                                         found_key.offset - 1, page_started,
1170                                         nr_written, 1);
1171                         BUG_ON(ret);
1172                         cow_start = (u64)-1;
1173                 }
1174
1175                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176                         struct extent_map *em;
1177                         struct extent_map_tree *em_tree;
1178                         em_tree = &BTRFS_I(inode)->extent_tree;
1179                         em = alloc_extent_map(GFP_NOFS);
1180                         BUG_ON(!em);
1181                         em->start = cur_offset;
1182                         em->orig_start = em->start;
1183                         em->len = num_bytes;
1184                         em->block_len = num_bytes;
1185                         em->block_start = disk_bytenr;
1186                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1187                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188                         while (1) {
1189                                 write_lock(&em_tree->lock);
1190                                 ret = add_extent_mapping(em_tree, em);
1191                                 write_unlock(&em_tree->lock);
1192                                 if (ret != -EEXIST) {
1193                                         free_extent_map(em);
1194                                         break;
1195                                 }
1196                                 btrfs_drop_extent_cache(inode, em->start,
1197                                                 em->start + em->len - 1, 0);
1198                         }
1199                         type = BTRFS_ORDERED_PREALLOC;
1200                 } else {
1201                         type = BTRFS_ORDERED_NOCOW;
1202                 }
1203
1204                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205                                                num_bytes, num_bytes, type);
1206                 BUG_ON(ret);
1207
1208                 if (root->root_key.objectid ==
1209                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1211                                                       num_bytes);
1212                         BUG_ON(ret);
1213                 }
1214
1215                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216                                 cur_offset, cur_offset + num_bytes - 1,
1217                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219                                 EXTENT_SET_PRIVATE2);
1220                 cur_offset = extent_end;
1221                 if (cur_offset > end)
1222                         break;
1223         }
1224         btrfs_release_path(root, path);
1225
1226         if (cur_offset <= end && cow_start == (u64)-1)
1227                 cow_start = cur_offset;
1228         if (cow_start != (u64)-1) {
1229                 ret = cow_file_range(inode, locked_page, cow_start, end,
1230                                      page_started, nr_written, 1);
1231                 BUG_ON(ret);
1232         }
1233
1234         if (nolock) {
1235                 ret = btrfs_end_transaction_nolock(trans, root);
1236                 BUG_ON(ret);
1237         } else {
1238                 ret = btrfs_end_transaction(trans, root);
1239                 BUG_ON(ret);
1240         }
1241         btrfs_free_path(path);
1242         return 0;
1243 }
1244
1245 /*
1246  * extent_io.c call back to do delayed allocation processing
1247  */
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249                               u64 start, u64 end, int *page_started,
1250                               unsigned long *nr_written)
1251 {
1252         int ret;
1253         struct btrfs_root *root = BTRFS_I(inode)->root;
1254
1255         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257                                          page_started, 1, nr_written);
1258         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260                                          page_started, 0, nr_written);
1261         else if (!btrfs_test_opt(root, COMPRESS) &&
1262                  !(BTRFS_I(inode)->force_compress) &&
1263                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264                 ret = cow_file_range(inode, locked_page, start, end,
1265                                       page_started, nr_written, 1);
1266         else
1267                 ret = cow_file_range_async(inode, locked_page, start, end,
1268                                            page_started, nr_written);
1269         return ret;
1270 }
1271
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273                                    struct extent_state *orig, u64 split)
1274 {
1275         /* not delalloc, ignore it */
1276         if (!(orig->state & EXTENT_DELALLOC))
1277                 return 0;
1278
1279         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1280         return 0;
1281 }
1282
1283 /*
1284  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285  * extents so we can keep track of new extents that are just merged onto old
1286  * extents, such as when we are doing sequential writes, so we can properly
1287  * account for the metadata space we'll need.
1288  */
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290                                    struct extent_state *new,
1291                                    struct extent_state *other)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(other->state & EXTENT_DELALLOC))
1295                 return 0;
1296
1297         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1298         return 0;
1299 }
1300
1301 /*
1302  * extent_io.c set_bit_hook, used to track delayed allocation
1303  * bytes in this file, and to maintain the list of inodes that
1304  * have pending delalloc work to be done.
1305  */
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307                               struct extent_state *state, int *bits)
1308 {
1309
1310         /*
1311          * set_bit and clear bit hooks normally require _irqsave/restore
1312          * but in this case, we are only testeing for the DELALLOC
1313          * bit, which is only set or cleared with irqs on
1314          */
1315         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316                 struct btrfs_root *root = BTRFS_I(inode)->root;
1317                 u64 len = state->end + 1 - state->start;
1318                 int do_list = (root->root_key.objectid !=
1319                                BTRFS_ROOT_TREE_OBJECTID);
1320
1321                 if (*bits & EXTENT_FIRST_DELALLOC)
1322                         *bits &= ~EXTENT_FIRST_DELALLOC;
1323                 else
1324                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325
1326                 spin_lock(&root->fs_info->delalloc_lock);
1327                 BTRFS_I(inode)->delalloc_bytes += len;
1328                 root->fs_info->delalloc_bytes += len;
1329                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331                                       &root->fs_info->delalloc_inodes);
1332                 }
1333                 spin_unlock(&root->fs_info->delalloc_lock);
1334         }
1335         return 0;
1336 }
1337
1338 /*
1339  * extent_io.c clear_bit_hook, see set_bit_hook for why
1340  */
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342                                 struct extent_state *state, int *bits)
1343 {
1344         /*
1345          * set_bit and clear bit hooks normally require _irqsave/restore
1346          * but in this case, we are only testeing for the DELALLOC
1347          * bit, which is only set or cleared with irqs on
1348          */
1349         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350                 struct btrfs_root *root = BTRFS_I(inode)->root;
1351                 u64 len = state->end + 1 - state->start;
1352                 int do_list = (root->root_key.objectid !=
1353                                BTRFS_ROOT_TREE_OBJECTID);
1354
1355                 if (*bits & EXTENT_FIRST_DELALLOC)
1356                         *bits &= ~EXTENT_FIRST_DELALLOC;
1357                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359
1360                 if (*bits & EXTENT_DO_ACCOUNTING)
1361                         btrfs_delalloc_release_metadata(inode, len);
1362
1363                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364                     && do_list)
1365                         btrfs_free_reserved_data_space(inode, len);
1366
1367                 spin_lock(&root->fs_info->delalloc_lock);
1368                 root->fs_info->delalloc_bytes -= len;
1369                 BTRFS_I(inode)->delalloc_bytes -= len;
1370
1371                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374                 }
1375                 spin_unlock(&root->fs_info->delalloc_lock);
1376         }
1377         return 0;
1378 }
1379
1380 /*
1381  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382  * we don't create bios that span stripes or chunks
1383  */
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385                          size_t size, struct bio *bio,
1386                          unsigned long bio_flags)
1387 {
1388         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389         struct btrfs_mapping_tree *map_tree;
1390         u64 logical = (u64)bio->bi_sector << 9;
1391         u64 length = 0;
1392         u64 map_length;
1393         int ret;
1394
1395         if (bio_flags & EXTENT_BIO_COMPRESSED)
1396                 return 0;
1397
1398         length = bio->bi_size;
1399         map_tree = &root->fs_info->mapping_tree;
1400         map_length = length;
1401         ret = btrfs_map_block(map_tree, READ, logical,
1402                               &map_length, NULL, 0);
1403
1404         if (map_length < length + size)
1405                 return 1;
1406         return ret;
1407 }
1408
1409 /*
1410  * in order to insert checksums into the metadata in large chunks,
1411  * we wait until bio submission time.   All the pages in the bio are
1412  * checksummed and sums are attached onto the ordered extent record.
1413  *
1414  * At IO completion time the cums attached on the ordered extent record
1415  * are inserted into the btree
1416  */
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418                                     struct bio *bio, int mirror_num,
1419                                     unsigned long bio_flags,
1420                                     u64 bio_offset)
1421 {
1422         struct btrfs_root *root = BTRFS_I(inode)->root;
1423         int ret = 0;
1424
1425         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1426         BUG_ON(ret);
1427         return 0;
1428 }
1429
1430 /*
1431  * in order to insert checksums into the metadata in large chunks,
1432  * we wait until bio submission time.   All the pages in the bio are
1433  * checksummed and sums are attached onto the ordered extent record.
1434  *
1435  * At IO completion time the cums attached on the ordered extent record
1436  * are inserted into the btree
1437  */
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439                           int mirror_num, unsigned long bio_flags,
1440                           u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1444 }
1445
1446 /*
1447  * extent_io.c submission hook. This does the right thing for csum calculation
1448  * on write, or reading the csums from the tree before a read
1449  */
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451                           int mirror_num, unsigned long bio_flags,
1452                           u64 bio_offset)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         int ret = 0;
1456         int skip_sum;
1457
1458         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459
1460         if (root == root->fs_info->tree_root)
1461                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462         else
1463                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1464         BUG_ON(ret);
1465
1466         if (!(rw & REQ_WRITE)) {
1467                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468                         return btrfs_submit_compressed_read(inode, bio,
1469                                                     mirror_num, bio_flags);
1470                 } else if (!skip_sum) {
1471                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1472                         if (ret)
1473                                 return ret;
1474                 }
1475                 goto mapit;
1476         } else if (!skip_sum) {
1477                 /* csum items have already been cloned */
1478                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479                         goto mapit;
1480                 /* we're doing a write, do the async checksumming */
1481                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482                                    inode, rw, bio, mirror_num,
1483                                    bio_flags, bio_offset,
1484                                    __btrfs_submit_bio_start,
1485                                    __btrfs_submit_bio_done);
1486         }
1487
1488 mapit:
1489         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1490 }
1491
1492 /*
1493  * given a list of ordered sums record them in the inode.  This happens
1494  * at IO completion time based on sums calculated at bio submission time.
1495  */
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497                              struct inode *inode, u64 file_offset,
1498                              struct list_head *list)
1499 {
1500         struct btrfs_ordered_sum *sum;
1501
1502         list_for_each_entry(sum, list, list) {
1503                 btrfs_csum_file_blocks(trans,
1504                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1505         }
1506         return 0;
1507 }
1508
1509 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1510                               struct extent_state **cached_state)
1511 {
1512         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1513                 WARN_ON(1);
1514         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1515                                    cached_state, GFP_NOFS);
1516 }
1517
1518 /* see btrfs_writepage_start_hook for details on why this is required */
1519 struct btrfs_writepage_fixup {
1520         struct page *page;
1521         struct btrfs_work work;
1522 };
1523
1524 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1525 {
1526         struct btrfs_writepage_fixup *fixup;
1527         struct btrfs_ordered_extent *ordered;
1528         struct extent_state *cached_state = NULL;
1529         struct page *page;
1530         struct inode *inode;
1531         u64 page_start;
1532         u64 page_end;
1533
1534         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1535         page = fixup->page;
1536 again:
1537         lock_page(page);
1538         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1539                 ClearPageChecked(page);
1540                 goto out_page;
1541         }
1542
1543         inode = page->mapping->host;
1544         page_start = page_offset(page);
1545         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1546
1547         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1548                          &cached_state, GFP_NOFS);
1549
1550         /* already ordered? We're done */
1551         if (PagePrivate2(page))
1552                 goto out;
1553
1554         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1555         if (ordered) {
1556                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1557                                      page_end, &cached_state, GFP_NOFS);
1558                 unlock_page(page);
1559                 btrfs_start_ordered_extent(inode, ordered, 1);
1560                 goto again;
1561         }
1562
1563         BUG();
1564         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1565         ClearPageChecked(page);
1566 out:
1567         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1568                              &cached_state, GFP_NOFS);
1569 out_page:
1570         unlock_page(page);
1571         page_cache_release(page);
1572         kfree(fixup);
1573 }
1574
1575 /*
1576  * There are a few paths in the higher layers of the kernel that directly
1577  * set the page dirty bit without asking the filesystem if it is a
1578  * good idea.  This causes problems because we want to make sure COW
1579  * properly happens and the data=ordered rules are followed.
1580  *
1581  * In our case any range that doesn't have the ORDERED bit set
1582  * hasn't been properly setup for IO.  We kick off an async process
1583  * to fix it up.  The async helper will wait for ordered extents, set
1584  * the delalloc bit and make it safe to write the page.
1585  */
1586 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1587 {
1588         struct inode *inode = page->mapping->host;
1589         struct btrfs_writepage_fixup *fixup;
1590         struct btrfs_root *root = BTRFS_I(inode)->root;
1591
1592         /* this page is properly in the ordered list */
1593         if (TestClearPagePrivate2(page))
1594                 return 0;
1595
1596         if (PageChecked(page))
1597                 return -EAGAIN;
1598
1599         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1600         if (!fixup)
1601                 return -EAGAIN;
1602
1603         SetPageChecked(page);
1604         page_cache_get(page);
1605         fixup->work.func = btrfs_writepage_fixup_worker;
1606         fixup->page = page;
1607         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1608         return -EAGAIN;
1609 }
1610
1611 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1612                                        struct inode *inode, u64 file_pos,
1613                                        u64 disk_bytenr, u64 disk_num_bytes,
1614                                        u64 num_bytes, u64 ram_bytes,
1615                                        u8 compression, u8 encryption,
1616                                        u16 other_encoding, int extent_type)
1617 {
1618         struct btrfs_root *root = BTRFS_I(inode)->root;
1619         struct btrfs_file_extent_item *fi;
1620         struct btrfs_path *path;
1621         struct extent_buffer *leaf;
1622         struct btrfs_key ins;
1623         u64 hint;
1624         int ret;
1625
1626         path = btrfs_alloc_path();
1627         BUG_ON(!path);
1628
1629         path->leave_spinning = 1;
1630
1631         /*
1632          * we may be replacing one extent in the tree with another.
1633          * The new extent is pinned in the extent map, and we don't want
1634          * to drop it from the cache until it is completely in the btree.
1635          *
1636          * So, tell btrfs_drop_extents to leave this extent in the cache.
1637          * the caller is expected to unpin it and allow it to be merged
1638          * with the others.
1639          */
1640         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1641                                  &hint, 0);
1642         BUG_ON(ret);
1643
1644         ins.objectid = inode->i_ino;
1645         ins.offset = file_pos;
1646         ins.type = BTRFS_EXTENT_DATA_KEY;
1647         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1648         BUG_ON(ret);
1649         leaf = path->nodes[0];
1650         fi = btrfs_item_ptr(leaf, path->slots[0],
1651                             struct btrfs_file_extent_item);
1652         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1653         btrfs_set_file_extent_type(leaf, fi, extent_type);
1654         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1655         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1656         btrfs_set_file_extent_offset(leaf, fi, 0);
1657         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1658         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1659         btrfs_set_file_extent_compression(leaf, fi, compression);
1660         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1661         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1662
1663         btrfs_unlock_up_safe(path, 1);
1664         btrfs_set_lock_blocking(leaf);
1665
1666         btrfs_mark_buffer_dirty(leaf);
1667
1668         inode_add_bytes(inode, num_bytes);
1669
1670         ins.objectid = disk_bytenr;
1671         ins.offset = disk_num_bytes;
1672         ins.type = BTRFS_EXTENT_ITEM_KEY;
1673         ret = btrfs_alloc_reserved_file_extent(trans, root,
1674                                         root->root_key.objectid,
1675                                         inode->i_ino, file_pos, &ins);
1676         BUG_ON(ret);
1677         btrfs_free_path(path);
1678
1679         return 0;
1680 }
1681
1682 /*
1683  * helper function for btrfs_finish_ordered_io, this
1684  * just reads in some of the csum leaves to prime them into ram
1685  * before we start the transaction.  It limits the amount of btree
1686  * reads required while inside the transaction.
1687  */
1688 /* as ordered data IO finishes, this gets called so we can finish
1689  * an ordered extent if the range of bytes in the file it covers are
1690  * fully written.
1691  */
1692 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1693 {
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         struct btrfs_trans_handle *trans = NULL;
1696         struct btrfs_ordered_extent *ordered_extent = NULL;
1697         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1698         struct extent_state *cached_state = NULL;
1699         int compress_type = 0;
1700         int ret;
1701         bool nolock = false;
1702
1703         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1704                                              end - start + 1);
1705         if (!ret)
1706                 return 0;
1707         BUG_ON(!ordered_extent);
1708
1709         nolock = (root == root->fs_info->tree_root);
1710
1711         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1712                 BUG_ON(!list_empty(&ordered_extent->list));
1713                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1714                 if (!ret) {
1715                         if (nolock)
1716                                 trans = btrfs_join_transaction_nolock(root);
1717                         else
1718                                 trans = btrfs_join_transaction(root);
1719                         BUG_ON(IS_ERR(trans));
1720                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1721                         ret = btrfs_update_inode(trans, root, inode);
1722                         BUG_ON(ret);
1723                 }
1724                 goto out;
1725         }
1726
1727         lock_extent_bits(io_tree, ordered_extent->file_offset,
1728                          ordered_extent->file_offset + ordered_extent->len - 1,
1729                          0, &cached_state, GFP_NOFS);
1730
1731         if (nolock)
1732                 trans = btrfs_join_transaction_nolock(root);
1733         else
1734                 trans = btrfs_join_transaction(root);
1735         BUG_ON(IS_ERR(trans));
1736         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1737
1738         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1739                 compress_type = ordered_extent->compress_type;
1740         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1741                 BUG_ON(compress_type);
1742                 ret = btrfs_mark_extent_written(trans, inode,
1743                                                 ordered_extent->file_offset,
1744                                                 ordered_extent->file_offset +
1745                                                 ordered_extent->len);
1746                 BUG_ON(ret);
1747         } else {
1748                 BUG_ON(root == root->fs_info->tree_root);
1749                 ret = insert_reserved_file_extent(trans, inode,
1750                                                 ordered_extent->file_offset,
1751                                                 ordered_extent->start,
1752                                                 ordered_extent->disk_len,
1753                                                 ordered_extent->len,
1754                                                 ordered_extent->len,
1755                                                 compress_type, 0, 0,
1756                                                 BTRFS_FILE_EXTENT_REG);
1757                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1758                                    ordered_extent->file_offset,
1759                                    ordered_extent->len);
1760                 BUG_ON(ret);
1761         }
1762         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1763                              ordered_extent->file_offset +
1764                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1765
1766         add_pending_csums(trans, inode, ordered_extent->file_offset,
1767                           &ordered_extent->list);
1768
1769         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1770         if (!ret) {
1771                 ret = btrfs_update_inode(trans, root, inode);
1772                 BUG_ON(ret);
1773         }
1774         ret = 0;
1775 out:
1776         if (nolock) {
1777                 if (trans)
1778                         btrfs_end_transaction_nolock(trans, root);
1779         } else {
1780                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1781                 if (trans)
1782                         btrfs_end_transaction(trans, root);
1783         }
1784
1785         /* once for us */
1786         btrfs_put_ordered_extent(ordered_extent);
1787         /* once for the tree */
1788         btrfs_put_ordered_extent(ordered_extent);
1789
1790         return 0;
1791 }
1792
1793 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1794                                 struct extent_state *state, int uptodate)
1795 {
1796         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1797
1798         ClearPagePrivate2(page);
1799         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1800 }
1801
1802 /*
1803  * When IO fails, either with EIO or csum verification fails, we
1804  * try other mirrors that might have a good copy of the data.  This
1805  * io_failure_record is used to record state as we go through all the
1806  * mirrors.  If another mirror has good data, the page is set up to date
1807  * and things continue.  If a good mirror can't be found, the original
1808  * bio end_io callback is called to indicate things have failed.
1809  */
1810 struct io_failure_record {
1811         struct page *page;
1812         u64 start;
1813         u64 len;
1814         u64 logical;
1815         unsigned long bio_flags;
1816         int last_mirror;
1817 };
1818
1819 static int btrfs_io_failed_hook(struct bio *failed_bio,
1820                          struct page *page, u64 start, u64 end,
1821                          struct extent_state *state)
1822 {
1823         struct io_failure_record *failrec = NULL;
1824         u64 private;
1825         struct extent_map *em;
1826         struct inode *inode = page->mapping->host;
1827         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1828         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1829         struct bio *bio;
1830         int num_copies;
1831         int ret;
1832         int rw;
1833         u64 logical;
1834
1835         ret = get_state_private(failure_tree, start, &private);
1836         if (ret) {
1837                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1838                 if (!failrec)
1839                         return -ENOMEM;
1840                 failrec->start = start;
1841                 failrec->len = end - start + 1;
1842                 failrec->last_mirror = 0;
1843                 failrec->bio_flags = 0;
1844
1845                 read_lock(&em_tree->lock);
1846                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1847                 if (em->start > start || em->start + em->len < start) {
1848                         free_extent_map(em);
1849                         em = NULL;
1850                 }
1851                 read_unlock(&em_tree->lock);
1852
1853                 if (!em || IS_ERR(em)) {
1854                         kfree(failrec);
1855                         return -EIO;
1856                 }
1857                 logical = start - em->start;
1858                 logical = em->block_start + logical;
1859                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1860                         logical = em->block_start;
1861                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1862                         extent_set_compress_type(&failrec->bio_flags,
1863                                                  em->compress_type);
1864                 }
1865                 failrec->logical = logical;
1866                 free_extent_map(em);
1867                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1868                                 EXTENT_DIRTY, GFP_NOFS);
1869                 set_state_private(failure_tree, start,
1870                                  (u64)(unsigned long)failrec);
1871         } else {
1872                 failrec = (struct io_failure_record *)(unsigned long)private;
1873         }
1874         num_copies = btrfs_num_copies(
1875                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1876                               failrec->logical, failrec->len);
1877         failrec->last_mirror++;
1878         if (!state) {
1879                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1880                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1881                                                     failrec->start,
1882                                                     EXTENT_LOCKED);
1883                 if (state && state->start != failrec->start)
1884                         state = NULL;
1885                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1886         }
1887         if (!state || failrec->last_mirror > num_copies) {
1888                 set_state_private(failure_tree, failrec->start, 0);
1889                 clear_extent_bits(failure_tree, failrec->start,
1890                                   failrec->start + failrec->len - 1,
1891                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1892                 kfree(failrec);
1893                 return -EIO;
1894         }
1895         bio = bio_alloc(GFP_NOFS, 1);
1896         bio->bi_private = state;
1897         bio->bi_end_io = failed_bio->bi_end_io;
1898         bio->bi_sector = failrec->logical >> 9;
1899         bio->bi_bdev = failed_bio->bi_bdev;
1900         bio->bi_size = 0;
1901
1902         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1903         if (failed_bio->bi_rw & REQ_WRITE)
1904                 rw = WRITE;
1905         else
1906                 rw = READ;
1907
1908         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1909                                                       failrec->last_mirror,
1910                                                       failrec->bio_flags, 0);
1911         return ret;
1912 }
1913
1914 /*
1915  * each time an IO finishes, we do a fast check in the IO failure tree
1916  * to see if we need to process or clean up an io_failure_record
1917  */
1918 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1919 {
1920         u64 private;
1921         u64 private_failure;
1922         struct io_failure_record *failure;
1923         int ret;
1924
1925         private = 0;
1926         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1927                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1928                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1929                                         start, &private_failure);
1930                 if (ret == 0) {
1931                         failure = (struct io_failure_record *)(unsigned long)
1932                                    private_failure;
1933                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1934                                           failure->start, 0);
1935                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1936                                           failure->start,
1937                                           failure->start + failure->len - 1,
1938                                           EXTENT_DIRTY | EXTENT_LOCKED,
1939                                           GFP_NOFS);
1940                         kfree(failure);
1941                 }
1942         }
1943         return 0;
1944 }
1945
1946 /*
1947  * when reads are done, we need to check csums to verify the data is correct
1948  * if there's a match, we allow the bio to finish.  If not, we go through
1949  * the io_failure_record routines to find good copies
1950  */
1951 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1952                                struct extent_state *state)
1953 {
1954         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1955         struct inode *inode = page->mapping->host;
1956         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1957         char *kaddr;
1958         u64 private = ~(u32)0;
1959         int ret;
1960         struct btrfs_root *root = BTRFS_I(inode)->root;
1961         u32 csum = ~(u32)0;
1962
1963         if (PageChecked(page)) {
1964                 ClearPageChecked(page);
1965                 goto good;
1966         }
1967
1968         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1969                 return 0;
1970
1971         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1972             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1973                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1974                                   GFP_NOFS);
1975                 return 0;
1976         }
1977
1978         if (state && state->start == start) {
1979                 private = state->private;
1980                 ret = 0;
1981         } else {
1982                 ret = get_state_private(io_tree, start, &private);
1983         }
1984         kaddr = kmap_atomic(page, KM_USER0);
1985         if (ret)
1986                 goto zeroit;
1987
1988         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1989         btrfs_csum_final(csum, (char *)&csum);
1990         if (csum != private)
1991                 goto zeroit;
1992
1993         kunmap_atomic(kaddr, KM_USER0);
1994 good:
1995         /* if the io failure tree for this inode is non-empty,
1996          * check to see if we've recovered from a failed IO
1997          */
1998         btrfs_clean_io_failures(inode, start);
1999         return 0;
2000
2001 zeroit:
2002         if (printk_ratelimit()) {
2003                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2004                        "private %llu\n", page->mapping->host->i_ino,
2005                        (unsigned long long)start, csum,
2006                        (unsigned long long)private);
2007         }
2008         memset(kaddr + offset, 1, end - start + 1);
2009         flush_dcache_page(page);
2010         kunmap_atomic(kaddr, KM_USER0);
2011         if (private == 0)
2012                 return 0;
2013         return -EIO;
2014 }
2015
2016 struct delayed_iput {
2017         struct list_head list;
2018         struct inode *inode;
2019 };
2020
2021 void btrfs_add_delayed_iput(struct inode *inode)
2022 {
2023         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2024         struct delayed_iput *delayed;
2025
2026         if (atomic_add_unless(&inode->i_count, -1, 1))
2027                 return;
2028
2029         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2030         delayed->inode = inode;
2031
2032         spin_lock(&fs_info->delayed_iput_lock);
2033         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2034         spin_unlock(&fs_info->delayed_iput_lock);
2035 }
2036
2037 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2038 {
2039         LIST_HEAD(list);
2040         struct btrfs_fs_info *fs_info = root->fs_info;
2041         struct delayed_iput *delayed;
2042         int empty;
2043
2044         spin_lock(&fs_info->delayed_iput_lock);
2045         empty = list_empty(&fs_info->delayed_iputs);
2046         spin_unlock(&fs_info->delayed_iput_lock);
2047         if (empty)
2048                 return;
2049
2050         down_read(&root->fs_info->cleanup_work_sem);
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         list_splice_init(&fs_info->delayed_iputs, &list);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054
2055         while (!list_empty(&list)) {
2056                 delayed = list_entry(list.next, struct delayed_iput, list);
2057                 list_del(&delayed->list);
2058                 iput(delayed->inode);
2059                 kfree(delayed);
2060         }
2061         up_read(&root->fs_info->cleanup_work_sem);
2062 }
2063
2064 /*
2065  * calculate extra metadata reservation when snapshotting a subvolume
2066  * contains orphan files.
2067  */
2068 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2069                                 struct btrfs_pending_snapshot *pending,
2070                                 u64 *bytes_to_reserve)
2071 {
2072         struct btrfs_root *root;
2073         struct btrfs_block_rsv *block_rsv;
2074         u64 num_bytes;
2075         int index;
2076
2077         root = pending->root;
2078         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2079                 return;
2080
2081         block_rsv = root->orphan_block_rsv;
2082
2083         /* orphan block reservation for the snapshot */
2084         num_bytes = block_rsv->size;
2085
2086         /*
2087          * after the snapshot is created, COWing tree blocks may use more
2088          * space than it frees. So we should make sure there is enough
2089          * reserved space.
2090          */
2091         index = trans->transid & 0x1;
2092         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2093                 num_bytes += block_rsv->size -
2094                              (block_rsv->reserved + block_rsv->freed[index]);
2095         }
2096
2097         *bytes_to_reserve += num_bytes;
2098 }
2099
2100 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2101                                 struct btrfs_pending_snapshot *pending)
2102 {
2103         struct btrfs_root *root = pending->root;
2104         struct btrfs_root *snap = pending->snap;
2105         struct btrfs_block_rsv *block_rsv;
2106         u64 num_bytes;
2107         int index;
2108         int ret;
2109
2110         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2111                 return;
2112
2113         /* refill source subvolume's orphan block reservation */
2114         block_rsv = root->orphan_block_rsv;
2115         index = trans->transid & 0x1;
2116         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2117                 num_bytes = block_rsv->size -
2118                             (block_rsv->reserved + block_rsv->freed[index]);
2119                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2120                                               root->orphan_block_rsv,
2121                                               num_bytes);
2122                 BUG_ON(ret);
2123         }
2124
2125         /* setup orphan block reservation for the snapshot */
2126         block_rsv = btrfs_alloc_block_rsv(snap);
2127         BUG_ON(!block_rsv);
2128
2129         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2130         snap->orphan_block_rsv = block_rsv;
2131
2132         num_bytes = root->orphan_block_rsv->size;
2133         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2134                                       block_rsv, num_bytes);
2135         BUG_ON(ret);
2136
2137 #if 0
2138         /* insert orphan item for the snapshot */
2139         WARN_ON(!root->orphan_item_inserted);
2140         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2141                                        snap->root_key.objectid);
2142         BUG_ON(ret);
2143         snap->orphan_item_inserted = 1;
2144 #endif
2145 }
2146
2147 enum btrfs_orphan_cleanup_state {
2148         ORPHAN_CLEANUP_STARTED  = 1,
2149         ORPHAN_CLEANUP_DONE     = 2,
2150 };
2151
2152 /*
2153  * This is called in transaction commmit time. If there are no orphan
2154  * files in the subvolume, it removes orphan item and frees block_rsv
2155  * structure.
2156  */
2157 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2158                               struct btrfs_root *root)
2159 {
2160         int ret;
2161
2162         if (!list_empty(&root->orphan_list) ||
2163             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2164                 return;
2165
2166         if (root->orphan_item_inserted &&
2167             btrfs_root_refs(&root->root_item) > 0) {
2168                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2169                                             root->root_key.objectid);
2170                 BUG_ON(ret);
2171                 root->orphan_item_inserted = 0;
2172         }
2173
2174         if (root->orphan_block_rsv) {
2175                 WARN_ON(root->orphan_block_rsv->size > 0);
2176                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2177                 root->orphan_block_rsv = NULL;
2178         }
2179 }
2180
2181 /*
2182  * This creates an orphan entry for the given inode in case something goes
2183  * wrong in the middle of an unlink/truncate.
2184  *
2185  * NOTE: caller of this function should reserve 5 units of metadata for
2186  *       this function.
2187  */
2188 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2189 {
2190         struct btrfs_root *root = BTRFS_I(inode)->root;
2191         struct btrfs_block_rsv *block_rsv = NULL;
2192         int reserve = 0;
2193         int insert = 0;
2194         int ret;
2195
2196         if (!root->orphan_block_rsv) {
2197                 block_rsv = btrfs_alloc_block_rsv(root);
2198                 BUG_ON(!block_rsv);
2199         }
2200
2201         spin_lock(&root->orphan_lock);
2202         if (!root->orphan_block_rsv) {
2203                 root->orphan_block_rsv = block_rsv;
2204         } else if (block_rsv) {
2205                 btrfs_free_block_rsv(root, block_rsv);
2206                 block_rsv = NULL;
2207         }
2208
2209         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2210                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2211 #if 0
2212                 /*
2213                  * For proper ENOSPC handling, we should do orphan
2214                  * cleanup when mounting. But this introduces backward
2215                  * compatibility issue.
2216                  */
2217                 if (!xchg(&root->orphan_item_inserted, 1))
2218                         insert = 2;
2219                 else
2220                         insert = 1;
2221 #endif
2222                 insert = 1;
2223         }
2224
2225         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2226                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2227                 reserve = 1;
2228         }
2229         spin_unlock(&root->orphan_lock);
2230
2231         if (block_rsv)
2232                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2233
2234         /* grab metadata reservation from transaction handle */
2235         if (reserve) {
2236                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2237                 BUG_ON(ret);
2238         }
2239
2240         /* insert an orphan item to track this unlinked/truncated file */
2241         if (insert >= 1) {
2242                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2243                 BUG_ON(ret);
2244         }
2245
2246         /* insert an orphan item to track subvolume contains orphan files */
2247         if (insert >= 2) {
2248                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2249                                                root->root_key.objectid);
2250                 BUG_ON(ret);
2251         }
2252         return 0;
2253 }
2254
2255 /*
2256  * We have done the truncate/delete so we can go ahead and remove the orphan
2257  * item for this particular inode.
2258  */
2259 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2260 {
2261         struct btrfs_root *root = BTRFS_I(inode)->root;
2262         int delete_item = 0;
2263         int release_rsv = 0;
2264         int ret = 0;
2265
2266         spin_lock(&root->orphan_lock);
2267         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2268                 list_del_init(&BTRFS_I(inode)->i_orphan);
2269                 delete_item = 1;
2270         }
2271
2272         if (BTRFS_I(inode)->orphan_meta_reserved) {
2273                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2274                 release_rsv = 1;
2275         }
2276         spin_unlock(&root->orphan_lock);
2277
2278         if (trans && delete_item) {
2279                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2280                 BUG_ON(ret);
2281         }
2282
2283         if (release_rsv)
2284                 btrfs_orphan_release_metadata(inode);
2285
2286         return 0;
2287 }
2288
2289 /*
2290  * this cleans up any orphans that may be left on the list from the last use
2291  * of this root.
2292  */
2293 int btrfs_orphan_cleanup(struct btrfs_root *root)
2294 {
2295         struct btrfs_path *path;
2296         struct extent_buffer *leaf;
2297         struct btrfs_key key, found_key;
2298         struct btrfs_trans_handle *trans;
2299         struct inode *inode;
2300         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2301
2302         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2303                 return 0;
2304
2305         path = btrfs_alloc_path();
2306         if (!path) {
2307                 ret = -ENOMEM;
2308                 goto out;
2309         }
2310         path->reada = -1;
2311
2312         key.objectid = BTRFS_ORPHAN_OBJECTID;
2313         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2314         key.offset = (u64)-1;
2315
2316         while (1) {
2317                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2318                 if (ret < 0)
2319                         goto out;
2320
2321                 /*
2322                  * if ret == 0 means we found what we were searching for, which
2323                  * is weird, but possible, so only screw with path if we didn't
2324                  * find the key and see if we have stuff that matches
2325                  */
2326                 if (ret > 0) {
2327                         ret = 0;
2328                         if (path->slots[0] == 0)
2329                                 break;
2330                         path->slots[0]--;
2331                 }
2332
2333                 /* pull out the item */
2334                 leaf = path->nodes[0];
2335                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2336
2337                 /* make sure the item matches what we want */
2338                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2339                         break;
2340                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2341                         break;
2342
2343                 /* release the path since we're done with it */
2344                 btrfs_release_path(root, path);
2345
2346                 /*
2347                  * this is where we are basically btrfs_lookup, without the
2348                  * crossing root thing.  we store the inode number in the
2349                  * offset of the orphan item.
2350                  */
2351                 found_key.objectid = found_key.offset;
2352                 found_key.type = BTRFS_INODE_ITEM_KEY;
2353                 found_key.offset = 0;
2354                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2355                 if (IS_ERR(inode)) {
2356                         ret = PTR_ERR(inode);
2357                         goto out;
2358                 }
2359
2360                 /*
2361                  * add this inode to the orphan list so btrfs_orphan_del does
2362                  * the proper thing when we hit it
2363                  */
2364                 spin_lock(&root->orphan_lock);
2365                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2366                 spin_unlock(&root->orphan_lock);
2367
2368                 /*
2369                  * if this is a bad inode, means we actually succeeded in
2370                  * removing the inode, but not the orphan record, which means
2371                  * we need to manually delete the orphan since iput will just
2372                  * do a destroy_inode
2373                  */
2374                 if (is_bad_inode(inode)) {
2375                         trans = btrfs_start_transaction(root, 0);
2376                         if (IS_ERR(trans)) {
2377                                 ret = PTR_ERR(trans);
2378                                 goto out;
2379                         }
2380                         btrfs_orphan_del(trans, inode);
2381                         btrfs_end_transaction(trans, root);
2382                         iput(inode);
2383                         continue;
2384                 }
2385
2386                 /* if we have links, this was a truncate, lets do that */
2387                 if (inode->i_nlink) {
2388                         if (!S_ISREG(inode->i_mode)) {
2389                                 WARN_ON(1);
2390                                 iput(inode);
2391                                 continue;
2392                         }
2393                         nr_truncate++;
2394                         ret = btrfs_truncate(inode);
2395                 } else {
2396                         nr_unlink++;
2397                 }
2398
2399                 /* this will do delete_inode and everything for us */
2400                 iput(inode);
2401                 if (ret)
2402                         goto out;
2403         }
2404         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2405
2406         if (root->orphan_block_rsv)
2407                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2408                                         (u64)-1);
2409
2410         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2411                 trans = btrfs_join_transaction(root);
2412                 if (!IS_ERR(trans))
2413                         btrfs_end_transaction(trans, root);
2414         }
2415
2416         if (nr_unlink)
2417                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2418         if (nr_truncate)
2419                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2420
2421 out:
2422         if (ret)
2423                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2424         btrfs_free_path(path);
2425         return ret;
2426 }
2427
2428 /*
2429  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2430  * don't find any xattrs, we know there can't be any acls.
2431  *
2432  * slot is the slot the inode is in, objectid is the objectid of the inode
2433  */
2434 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2435                                           int slot, u64 objectid)
2436 {
2437         u32 nritems = btrfs_header_nritems(leaf);
2438         struct btrfs_key found_key;
2439         int scanned = 0;
2440
2441         slot++;
2442         while (slot < nritems) {
2443                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2444
2445                 /* we found a different objectid, there must not be acls */
2446                 if (found_key.objectid != objectid)
2447                         return 0;
2448
2449                 /* we found an xattr, assume we've got an acl */
2450                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2451                         return 1;
2452
2453                 /*
2454                  * we found a key greater than an xattr key, there can't
2455                  * be any acls later on
2456                  */
2457                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2458                         return 0;
2459
2460                 slot++;
2461                 scanned++;
2462
2463                 /*
2464                  * it goes inode, inode backrefs, xattrs, extents,
2465                  * so if there are a ton of hard links to an inode there can
2466                  * be a lot of backrefs.  Don't waste time searching too hard,
2467                  * this is just an optimization
2468                  */
2469                 if (scanned >= 8)
2470                         break;
2471         }
2472         /* we hit the end of the leaf before we found an xattr or
2473          * something larger than an xattr.  We have to assume the inode
2474          * has acls
2475          */
2476         return 1;
2477 }
2478
2479 /*
2480  * read an inode from the btree into the in-memory inode
2481  */
2482 static void btrfs_read_locked_inode(struct inode *inode)
2483 {
2484         struct btrfs_path *path;
2485         struct extent_buffer *leaf;
2486         struct btrfs_inode_item *inode_item;
2487         struct btrfs_timespec *tspec;
2488         struct btrfs_root *root = BTRFS_I(inode)->root;
2489         struct btrfs_key location;
2490         int maybe_acls;
2491         u32 rdev;
2492         int ret;
2493
2494         path = btrfs_alloc_path();
2495         BUG_ON(!path);
2496         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2497
2498         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2499         if (ret)
2500                 goto make_bad;
2501
2502         leaf = path->nodes[0];
2503         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2504                                     struct btrfs_inode_item);
2505
2506         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2507         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2508         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2509         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2510         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2511
2512         tspec = btrfs_inode_atime(inode_item);
2513         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2514         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2515
2516         tspec = btrfs_inode_mtime(inode_item);
2517         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2518         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2519
2520         tspec = btrfs_inode_ctime(inode_item);
2521         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523
2524         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2525         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2526         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2527         inode->i_generation = BTRFS_I(inode)->generation;
2528         inode->i_rdev = 0;
2529         rdev = btrfs_inode_rdev(leaf, inode_item);
2530
2531         BTRFS_I(inode)->index_cnt = (u64)-1;
2532         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2533
2534         /*
2535          * try to precache a NULL acl entry for files that don't have
2536          * any xattrs or acls
2537          */
2538         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2539         if (!maybe_acls)
2540                 cache_no_acl(inode);
2541
2542         btrfs_free_path(path);
2543         inode_item = NULL;
2544
2545         switch (inode->i_mode & S_IFMT) {
2546         case S_IFREG:
2547                 inode->i_mapping->a_ops = &btrfs_aops;
2548                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2549                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2550                 inode->i_fop = &btrfs_file_operations;
2551                 inode->i_op = &btrfs_file_inode_operations;
2552                 break;
2553         case S_IFDIR:
2554                 inode->i_fop = &btrfs_dir_file_operations;
2555                 if (root == root->fs_info->tree_root)
2556                         inode->i_op = &btrfs_dir_ro_inode_operations;
2557                 else
2558                         inode->i_op = &btrfs_dir_inode_operations;
2559                 break;
2560         case S_IFLNK:
2561                 inode->i_op = &btrfs_symlink_inode_operations;
2562                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2563                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2564                 break;
2565         default:
2566                 inode->i_op = &btrfs_special_inode_operations;
2567                 init_special_inode(inode, inode->i_mode, rdev);
2568                 break;
2569         }
2570
2571         btrfs_update_iflags(inode);
2572         return;
2573
2574 make_bad:
2575         btrfs_free_path(path);
2576         make_bad_inode(inode);
2577 }
2578
2579 /*
2580  * given a leaf and an inode, copy the inode fields into the leaf
2581  */
2582 static void fill_inode_item(struct btrfs_trans_handle *trans,
2583                             struct extent_buffer *leaf,
2584                             struct btrfs_inode_item *item,
2585                             struct inode *inode)
2586 {
2587         if (!leaf->map_token)
2588                 map_private_extent_buffer(leaf, (unsigned long)item,
2589                                           sizeof(struct btrfs_inode_item),
2590                                           &leaf->map_token, &leaf->kaddr,
2591                                           &leaf->map_start, &leaf->map_len,
2592                                           KM_USER1);
2593
2594         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2595         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2596         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2597         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2598         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2599
2600         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2601                                inode->i_atime.tv_sec);
2602         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2603                                 inode->i_atime.tv_nsec);
2604
2605         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2606                                inode->i_mtime.tv_sec);
2607         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2608                                 inode->i_mtime.tv_nsec);
2609
2610         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2611                                inode->i_ctime.tv_sec);
2612         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2613                                 inode->i_ctime.tv_nsec);
2614
2615         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2616         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2617         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2618         btrfs_set_inode_transid(leaf, item, trans->transid);
2619         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2620         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2621         btrfs_set_inode_block_group(leaf, item, 0);
2622
2623         if (leaf->map_token) {
2624                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2625                 leaf->map_token = NULL;
2626         }
2627 }
2628
2629 /*
2630  * copy everything in the in-memory inode into the btree.
2631  */
2632 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2633                                 struct btrfs_root *root, struct inode *inode)
2634 {
2635         struct btrfs_inode_item *inode_item;
2636         struct btrfs_path *path;
2637         struct extent_buffer *leaf;
2638         int ret;
2639
2640         path = btrfs_alloc_path();
2641         BUG_ON(!path);
2642         path->leave_spinning = 1;
2643         ret = btrfs_lookup_inode(trans, root, path,
2644                                  &BTRFS_I(inode)->location, 1);
2645         if (ret) {
2646                 if (ret > 0)
2647                         ret = -ENOENT;
2648                 goto failed;
2649         }
2650
2651         btrfs_unlock_up_safe(path, 1);
2652         leaf = path->nodes[0];
2653         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2654                                   struct btrfs_inode_item);
2655
2656         fill_inode_item(trans, leaf, inode_item, inode);
2657         btrfs_mark_buffer_dirty(leaf);
2658         btrfs_set_inode_last_trans(trans, inode);
2659         ret = 0;
2660 failed:
2661         btrfs_free_path(path);
2662         return ret;
2663 }
2664
2665
2666 /*
2667  * unlink helper that gets used here in inode.c and in the tree logging
2668  * recovery code.  It remove a link in a directory with a given name, and
2669  * also drops the back refs in the inode to the directory
2670  */
2671 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2672                                 struct btrfs_root *root,
2673                                 struct inode *dir, struct inode *inode,
2674                                 const char *name, int name_len)
2675 {
2676         struct btrfs_path *path;
2677         int ret = 0;
2678         struct extent_buffer *leaf;
2679         struct btrfs_dir_item *di;
2680         struct btrfs_key key;
2681         u64 index;
2682
2683         path = btrfs_alloc_path();
2684         if (!path) {
2685                 ret = -ENOMEM;
2686                 goto out;
2687         }
2688
2689         path->leave_spinning = 1;
2690         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2691                                     name, name_len, -1);
2692         if (IS_ERR(di)) {
2693                 ret = PTR_ERR(di);
2694                 goto err;
2695         }
2696         if (!di) {
2697                 ret = -ENOENT;
2698                 goto err;
2699         }
2700         leaf = path->nodes[0];
2701         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2702         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2703         if (ret)
2704                 goto err;
2705         btrfs_release_path(root, path);
2706
2707         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2708                                   inode->i_ino,
2709                                   dir->i_ino, &index);
2710         if (ret) {
2711                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2712                        "inode %lu parent %lu\n", name_len, name,
2713                        inode->i_ino, dir->i_ino);
2714                 goto err;
2715         }
2716
2717         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2718                                          index, name, name_len, -1);
2719         if (IS_ERR(di)) {
2720                 ret = PTR_ERR(di);
2721                 goto err;
2722         }
2723         if (!di) {
2724                 ret = -ENOENT;
2725                 goto err;
2726         }
2727         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2728         btrfs_release_path(root, path);
2729
2730         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2731                                          inode, dir->i_ino);
2732         BUG_ON(ret != 0 && ret != -ENOENT);
2733
2734         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2735                                            dir, index);
2736         if (ret == -ENOENT)
2737                 ret = 0;
2738 err:
2739         btrfs_free_path(path);
2740         if (ret)
2741                 goto out;
2742
2743         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2744         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2745         btrfs_update_inode(trans, root, dir);
2746 out:
2747         return ret;
2748 }
2749
2750 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2751                        struct btrfs_root *root,
2752                        struct inode *dir, struct inode *inode,
2753                        const char *name, int name_len)
2754 {
2755         int ret;
2756         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2757         if (!ret) {
2758                 btrfs_drop_nlink(inode);
2759                 ret = btrfs_update_inode(trans, root, inode);
2760         }
2761         return ret;
2762 }
2763                 
2764
2765 /* helper to check if there is any shared block in the path */
2766 static int check_path_shared(struct btrfs_root *root,
2767                              struct btrfs_path *path)
2768 {
2769         struct extent_buffer *eb;
2770         int level;
2771         u64 refs = 1;
2772
2773         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2774                 int ret;
2775
2776                 if (!path->nodes[level])
2777                         break;
2778                 eb = path->nodes[level];
2779                 if (!btrfs_block_can_be_shared(root, eb))
2780                         continue;
2781                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2782                                                &refs, NULL);
2783                 if (refs > 1)
2784                         return 1;
2785         }
2786         return 0;
2787 }
2788
2789 /*
2790  * helper to start transaction for unlink and rmdir.
2791  *
2792  * unlink and rmdir are special in btrfs, they do not always free space.
2793  * so in enospc case, we should make sure they will free space before
2794  * allowing them to use the global metadata reservation.
2795  */
2796 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2797                                                        struct dentry *dentry)
2798 {
2799         struct btrfs_trans_handle *trans;
2800         struct btrfs_root *root = BTRFS_I(dir)->root;
2801         struct btrfs_path *path;
2802         struct btrfs_inode_ref *ref;
2803         struct btrfs_dir_item *di;
2804         struct inode *inode = dentry->d_inode;
2805         u64 index;
2806         int check_link = 1;
2807         int err = -ENOSPC;
2808         int ret;
2809
2810         trans = btrfs_start_transaction(root, 10);
2811         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2812                 return trans;
2813
2814         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2815                 return ERR_PTR(-ENOSPC);
2816
2817         /* check if there is someone else holds reference */
2818         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2819                 return ERR_PTR(-ENOSPC);
2820
2821         if (atomic_read(&inode->i_count) > 2)
2822                 return ERR_PTR(-ENOSPC);
2823
2824         if (xchg(&root->fs_info->enospc_unlink, 1))
2825                 return ERR_PTR(-ENOSPC);
2826
2827         path = btrfs_alloc_path();
2828         if (!path) {
2829                 root->fs_info->enospc_unlink = 0;
2830                 return ERR_PTR(-ENOMEM);
2831         }
2832
2833         trans = btrfs_start_transaction(root, 0);
2834         if (IS_ERR(trans)) {
2835                 btrfs_free_path(path);
2836                 root->fs_info->enospc_unlink = 0;
2837                 return trans;
2838         }
2839
2840         path->skip_locking = 1;
2841         path->search_commit_root = 1;
2842
2843         ret = btrfs_lookup_inode(trans, root, path,
2844                                 &BTRFS_I(dir)->location, 0);
2845         if (ret < 0) {
2846                 err = ret;
2847                 goto out;
2848         }
2849         if (ret == 0) {
2850                 if (check_path_shared(root, path))
2851                         goto out;
2852         } else {
2853                 check_link = 0;
2854         }
2855         btrfs_release_path(root, path);
2856
2857         ret = btrfs_lookup_inode(trans, root, path,
2858                                 &BTRFS_I(inode)->location, 0);
2859         if (ret < 0) {
2860                 err = ret;
2861                 goto out;
2862         }
2863         if (ret == 0) {
2864                 if (check_path_shared(root, path))
2865                         goto out;
2866         } else {
2867                 check_link = 0;
2868         }
2869         btrfs_release_path(root, path);
2870
2871         if (ret == 0 && S_ISREG(inode->i_mode)) {
2872                 ret = btrfs_lookup_file_extent(trans, root, path,
2873                                                inode->i_ino, (u64)-1, 0);
2874                 if (ret < 0) {
2875                         err = ret;
2876                         goto out;
2877                 }
2878                 BUG_ON(ret == 0);
2879                 if (check_path_shared(root, path))
2880                         goto out;
2881                 btrfs_release_path(root, path);
2882         }
2883
2884         if (!check_link) {
2885                 err = 0;
2886                 goto out;
2887         }
2888
2889         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2890                                 dentry->d_name.name, dentry->d_name.len, 0);
2891         if (IS_ERR(di)) {
2892                 err = PTR_ERR(di);
2893                 goto out;
2894         }
2895         if (di) {
2896                 if (check_path_shared(root, path))
2897                         goto out;
2898         } else {
2899                 err = 0;
2900                 goto out;
2901         }
2902         btrfs_release_path(root, path);
2903
2904         ref = btrfs_lookup_inode_ref(trans, root, path,
2905                                 dentry->d_name.name, dentry->d_name.len,
2906                                 inode->i_ino, dir->i_ino, 0);
2907         if (IS_ERR(ref)) {
2908                 err = PTR_ERR(ref);
2909                 goto out;
2910         }
2911         BUG_ON(!ref);
2912         if (check_path_shared(root, path))
2913                 goto out;
2914         index = btrfs_inode_ref_index(path->nodes[0], ref);
2915         btrfs_release_path(root, path);
2916
2917         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2918                                 dentry->d_name.name, dentry->d_name.len, 0);
2919         if (IS_ERR(di)) {
2920                 err = PTR_ERR(di);
2921                 goto out;
2922         }
2923         BUG_ON(ret == -ENOENT);
2924         if (check_path_shared(root, path))
2925                 goto out;
2926
2927         err = 0;
2928 out:
2929         btrfs_free_path(path);
2930         if (err) {
2931                 btrfs_end_transaction(trans, root);
2932                 root->fs_info->enospc_unlink = 0;
2933                 return ERR_PTR(err);
2934         }
2935
2936         trans->block_rsv = &root->fs_info->global_block_rsv;
2937         return trans;
2938 }
2939
2940 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2941                                struct btrfs_root *root)
2942 {
2943         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2944                 BUG_ON(!root->fs_info->enospc_unlink);
2945                 root->fs_info->enospc_unlink = 0;
2946         }
2947         btrfs_end_transaction_throttle(trans, root);
2948 }
2949
2950 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2951 {
2952         struct btrfs_root *root = BTRFS_I(dir)->root;
2953         struct btrfs_trans_handle *trans;
2954         struct inode *inode = dentry->d_inode;
2955         int ret;
2956         unsigned long nr = 0;
2957
2958         trans = __unlink_start_trans(dir, dentry);
2959         if (IS_ERR(trans))
2960                 return PTR_ERR(trans);
2961
2962         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2963
2964         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2965                                  dentry->d_name.name, dentry->d_name.len);
2966         BUG_ON(ret);
2967
2968         if (inode->i_nlink == 0) {
2969                 ret = btrfs_orphan_add(trans, inode);
2970                 BUG_ON(ret);
2971         }
2972
2973         nr = trans->blocks_used;
2974         __unlink_end_trans(trans, root);
2975         btrfs_btree_balance_dirty(root, nr);
2976         return ret;
2977 }
2978
2979 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2980                         struct btrfs_root *root,
2981                         struct inode *dir, u64 objectid,
2982                         const char *name, int name_len)
2983 {
2984         struct btrfs_path *path;
2985         struct extent_buffer *leaf;
2986         struct btrfs_dir_item *di;
2987         struct btrfs_key key;
2988         u64 index;
2989         int ret;
2990
2991         path = btrfs_alloc_path();
2992         if (!path)
2993                 return -ENOMEM;
2994
2995         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2996                                    name, name_len, -1);
2997         BUG_ON(!di || IS_ERR(di));
2998
2999         leaf = path->nodes[0];
3000         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3001         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3002         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3003         BUG_ON(ret);
3004         btrfs_release_path(root, path);
3005
3006         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3007                                  objectid, root->root_key.objectid,
3008                                  dir->i_ino, &index, name, name_len);
3009         if (ret < 0) {
3010                 BUG_ON(ret != -ENOENT);
3011                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3012                                                  name, name_len);
3013                 BUG_ON(!di || IS_ERR(di));
3014
3015                 leaf = path->nodes[0];
3016                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3017                 btrfs_release_path(root, path);
3018                 index = key.offset;
3019         }
3020
3021         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3022                                          index, name, name_len, -1);
3023         BUG_ON(!di || IS_ERR(di));
3024
3025         leaf = path->nodes[0];
3026         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3027         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3028         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3029         BUG_ON(ret);
3030         btrfs_release_path(root, path);
3031
3032         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3033         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3034         ret = btrfs_update_inode(trans, root, dir);
3035         BUG_ON(ret);
3036
3037         btrfs_free_path(path);
3038         return 0;
3039 }
3040
3041 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3042 {
3043         struct inode *inode = dentry->d_inode;
3044         int err = 0;
3045         struct btrfs_root *root = BTRFS_I(dir)->root;
3046         struct btrfs_trans_handle *trans;
3047         unsigned long nr = 0;
3048
3049         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3050             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3051                 return -ENOTEMPTY;
3052
3053         trans = __unlink_start_trans(dir, dentry);
3054         if (IS_ERR(trans))
3055                 return PTR_ERR(trans);
3056
3057         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3058                 err = btrfs_unlink_subvol(trans, root, dir,
3059                                           BTRFS_I(inode)->location.objectid,
3060                                           dentry->d_name.name,
3061                                           dentry->d_name.len);
3062                 goto out;
3063         }
3064
3065         err = btrfs_orphan_add(trans, inode);
3066         if (err)
3067                 goto out;
3068
3069         /* now the directory is empty */
3070         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3071                                  dentry->d_name.name, dentry->d_name.len);
3072         if (!err)
3073                 btrfs_i_size_write(inode, 0);
3074 out:
3075         nr = trans->blocks_used;
3076         __unlink_end_trans(trans, root);
3077         btrfs_btree_balance_dirty(root, nr);
3078
3079         return err;
3080 }
3081
3082 #if 0
3083 /*
3084  * when truncating bytes in a file, it is possible to avoid reading
3085  * the leaves that contain only checksum items.  This can be the
3086  * majority of the IO required to delete a large file, but it must
3087  * be done carefully.
3088  *
3089  * The keys in the level just above the leaves are checked to make sure
3090  * the lowest key in a given leaf is a csum key, and starts at an offset
3091  * after the new  size.
3092  *
3093  * Then the key for the next leaf is checked to make sure it also has
3094  * a checksum item for the same file.  If it does, we know our target leaf
3095  * contains only checksum items, and it can be safely freed without reading
3096  * it.
3097  *
3098  * This is just an optimization targeted at large files.  It may do
3099  * nothing.  It will return 0 unless things went badly.
3100  */
3101 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3102                                      struct btrfs_root *root,
3103                                      struct btrfs_path *path,
3104                                      struct inode *inode, u64 new_size)
3105 {
3106         struct btrfs_key key;
3107         int ret;
3108         int nritems;
3109         struct btrfs_key found_key;
3110         struct btrfs_key other_key;
3111         struct btrfs_leaf_ref *ref;
3112         u64 leaf_gen;
3113         u64 leaf_start;
3114
3115         path->lowest_level = 1;
3116         key.objectid = inode->i_ino;
3117         key.type = BTRFS_CSUM_ITEM_KEY;
3118         key.offset = new_size;
3119 again:
3120         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3121         if (ret < 0)
3122                 goto out;
3123
3124         if (path->nodes[1] == NULL) {
3125                 ret = 0;
3126                 goto out;
3127         }
3128         ret = 0;
3129         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3130         nritems = btrfs_header_nritems(path->nodes[1]);
3131
3132         if (!nritems)
3133                 goto out;
3134
3135         if (path->slots[1] >= nritems)
3136                 goto next_node;
3137
3138         /* did we find a key greater than anything we want to delete? */
3139         if (found_key.objectid > inode->i_ino ||
3140            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3141                 goto out;
3142
3143         /* we check the next key in the node to make sure the leave contains
3144          * only checksum items.  This comparison doesn't work if our
3145          * leaf is the last one in the node
3146          */
3147         if (path->slots[1] + 1 >= nritems) {
3148 next_node:
3149                 /* search forward from the last key in the node, this
3150                  * will bring us into the next node in the tree
3151                  */
3152                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3153
3154                 /* unlikely, but we inc below, so check to be safe */
3155                 if (found_key.offset == (u64)-1)
3156                         goto out;
3157
3158                 /* search_forward needs a path with locks held, do the
3159                  * search again for the original key.  It is possible
3160                  * this will race with a balance and return a path that
3161                  * we could modify, but this drop is just an optimization
3162                  * and is allowed to miss some leaves.
3163                  */
3164                 btrfs_release_path(root, path);
3165                 found_key.offset++;
3166
3167                 /* setup a max key for search_forward */
3168                 other_key.offset = (u64)-1;
3169                 other_key.type = key.type;
3170                 other_key.objectid = key.objectid;
3171
3172                 path->keep_locks = 1;
3173                 ret = btrfs_search_forward(root, &found_key, &other_key,
3174                                            path, 0, 0);
3175                 path->keep_locks = 0;
3176                 if (ret || found_key.objectid != key.objectid ||
3177                     found_key.type != key.type) {
3178                         ret = 0;
3179                         goto out;
3180                 }
3181
3182                 key.offset = found_key.offset;
3183                 btrfs_release_path(root, path);
3184                 cond_resched();
3185                 goto again;
3186         }
3187
3188         /* we know there's one more slot after us in the tree,
3189          * read that key so we can verify it is also a checksum item
3190          */
3191         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3192
3193         if (found_key.objectid < inode->i_ino)
3194                 goto next_key;
3195
3196         if (found_key.type != key.type || found_key.offset < new_size)
3197                 goto next_key;
3198
3199         /*
3200          * if the key for the next leaf isn't a csum key from this objectid,
3201          * we can't be sure there aren't good items inside this leaf.
3202          * Bail out
3203          */
3204         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3205                 goto out;
3206
3207         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3208         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3209         /*
3210          * it is safe to delete this leaf, it contains only
3211          * csum items from this inode at an offset >= new_size
3212          */
3213         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3214         BUG_ON(ret);
3215
3216         if (root->ref_cows && leaf_gen < trans->transid) {
3217                 ref = btrfs_alloc_leaf_ref(root, 0);
3218                 if (ref) {
3219                         ref->root_gen = root->root_key.offset;
3220                         ref->bytenr = leaf_start;
3221                         ref->owner = 0;
3222                         ref->generation = leaf_gen;
3223                         ref->nritems = 0;
3224
3225                         btrfs_sort_leaf_ref(ref);
3226
3227                         ret = btrfs_add_leaf_ref(root, ref, 0);
3228                         WARN_ON(ret);
3229                         btrfs_free_leaf_ref(root, ref);
3230                 } else {
3231                         WARN_ON(1);
3232                 }
3233         }
3234 next_key:
3235         btrfs_release_path(root, path);
3236
3237         if (other_key.objectid == inode->i_ino &&
3238             other_key.type == key.type && other_key.offset > key.offset) {
3239                 key.offset = other_key.offset;
3240                 cond_resched();
3241                 goto again;
3242         }
3243         ret = 0;
3244 out:
3245         /* fixup any changes we've made to the path */
3246         path->lowest_level = 0;
3247         path->keep_locks = 0;
3248         btrfs_release_path(root, path);
3249         return ret;
3250 }
3251
3252 #endif
3253
3254 /*
3255  * this can truncate away extent items, csum items and directory items.
3256  * It starts at a high offset and removes keys until it can't find
3257  * any higher than new_size
3258  *
3259  * csum items that cross the new i_size are truncated to the new size
3260  * as well.
3261  *
3262  * min_type is the minimum key type to truncate down to.  If set to 0, this
3263  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3264  */
3265 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3266                                struct btrfs_root *root,
3267                                struct inode *inode,
3268                                u64 new_size, u32 min_type)
3269 {
3270         struct btrfs_path *path;
3271         struct extent_buffer *leaf;
3272         struct btrfs_file_extent_item *fi;
3273         struct btrfs_key key;
3274         struct btrfs_key found_key;
3275         u64 extent_start = 0;
3276         u64 extent_num_bytes = 0;
3277         u64 extent_offset = 0;
3278         u64 item_end = 0;
3279         u64 mask = root->sectorsize - 1;
3280         u32 found_type = (u8)-1;
3281         int found_extent;
3282         int del_item;
3283         int pending_del_nr = 0;
3284         int pending_del_slot = 0;
3285         int extent_type = -1;
3286         int encoding;
3287         int ret;
3288         int err = 0;
3289
3290         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3291
3292         if (root->ref_cows || root == root->fs_info->tree_root)
3293                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3294
3295         path = btrfs_alloc_path();
3296         BUG_ON(!path);
3297         path->reada = -1;
3298
3299         key.objectid = inode->i_ino;
3300         key.offset = (u64)-1;
3301         key.type = (u8)-1;
3302
3303 search_again:
3304         path->leave_spinning = 1;
3305         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3306         if (ret < 0) {
3307                 err = ret;
3308                 goto out;
3309         }
3310
3311         if (ret > 0) {
3312                 /* there are no items in the tree for us to truncate, we're
3313                  * done
3314                  */
3315                 if (path->slots[0] == 0)
3316                         goto out;
3317                 path->slots[0]--;
3318         }
3319
3320         while (1) {
3321                 fi = NULL;
3322                 leaf = path->nodes[0];
3323                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3324                 found_type = btrfs_key_type(&found_key);
3325                 encoding = 0;
3326
3327                 if (found_key.objectid != inode->i_ino)
3328                         break;
3329
3330                 if (found_type < min_type)
3331                         break;
3332
3333                 item_end = found_key.offset;
3334                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3335                         fi = btrfs_item_ptr(leaf, path->slots[0],
3336                                             struct btrfs_file_extent_item);
3337                         extent_type = btrfs_file_extent_type(leaf, fi);
3338                         encoding = btrfs_file_extent_compression(leaf, fi);
3339                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3340                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3341
3342                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3343                                 item_end +=
3344                                     btrfs_file_extent_num_bytes(leaf, fi);
3345                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3346                                 item_end += btrfs_file_extent_inline_len(leaf,
3347                                                                          fi);
3348                         }
3349                         item_end--;
3350                 }
3351                 if (found_type > min_type) {
3352                         del_item = 1;
3353                 } else {
3354                         if (item_end < new_size)
3355                                 break;
3356                         if (found_key.offset >= new_size)
3357                                 del_item = 1;
3358                         else
3359                                 del_item = 0;
3360                 }
3361                 found_extent = 0;
3362                 /* FIXME, shrink the extent if the ref count is only 1 */
3363                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3364                         goto delete;
3365
3366                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3367                         u64 num_dec;
3368                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3369                         if (!del_item && !encoding) {
3370                                 u64 orig_num_bytes =
3371                                         btrfs_file_extent_num_bytes(leaf, fi);
3372                                 extent_num_bytes = new_size -
3373                                         found_key.offset + root->sectorsize - 1;
3374                                 extent_num_bytes = extent_num_bytes &
3375                                         ~((u64)root->sectorsize - 1);
3376                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3377                                                          extent_num_bytes);
3378                                 num_dec = (orig_num_bytes -
3379                                            extent_num_bytes);
3380                                 if (root->ref_cows && extent_start != 0)
3381                                         inode_sub_bytes(inode, num_dec);
3382                                 btrfs_mark_buffer_dirty(leaf);
3383                         } else {
3384                                 extent_num_bytes =
3385                                         btrfs_file_extent_disk_num_bytes(leaf,
3386                                                                          fi);
3387                                 extent_offset = found_key.offset -
3388                                         btrfs_file_extent_offset(leaf, fi);
3389
3390                                 /* FIXME blocksize != 4096 */
3391                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3392                                 if (extent_start != 0) {
3393                                         found_extent = 1;
3394                                         if (root->ref_cows)
3395                                                 inode_sub_bytes(inode, num_dec);
3396                                 }
3397                         }
3398                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3399                         /*
3400                          * we can't truncate inline items that have had
3401                          * special encodings
3402                          */
3403                         if (!del_item &&
3404                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3405                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3406                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3407                                 u32 size = new_size - found_key.offset;
3408
3409                                 if (root->ref_cows) {
3410                                         inode_sub_bytes(inode, item_end + 1 -
3411                                                         new_size);
3412                                 }
3413                                 size =
3414                                     btrfs_file_extent_calc_inline_size(size);
3415                                 ret = btrfs_truncate_item(trans, root, path,
3416                                                           size, 1);
3417                                 BUG_ON(ret);
3418                         } else if (root->ref_cows) {
3419                                 inode_sub_bytes(inode, item_end + 1 -
3420                                                 found_key.offset);
3421                         }
3422                 }
3423 delete:
3424                 if (del_item) {
3425                         if (!pending_del_nr) {
3426                                 /* no pending yet, add ourselves */
3427                                 pending_del_slot = path->slots[0];
3428                                 pending_del_nr = 1;
3429                         } else if (pending_del_nr &&
3430                                    path->slots[0] + 1 == pending_del_slot) {
3431                                 /* hop on the pending chunk */
3432                                 pending_del_nr++;
3433                                 pending_del_slot = path->slots[0];
3434                         } else {
3435                                 BUG();
3436                         }
3437                 } else {
3438                         break;
3439                 }
3440                 if (found_extent && (root->ref_cows ||
3441                                      root == root->fs_info->tree_root)) {
3442                         btrfs_set_path_blocking(path);
3443                         ret = btrfs_free_extent(trans, root, extent_start,
3444                                                 extent_num_bytes, 0,
3445                                                 btrfs_header_owner(leaf),
3446                                                 inode->i_ino, extent_offset);
3447                         BUG_ON(ret);
3448                 }
3449
3450                 if (found_type == BTRFS_INODE_ITEM_KEY)
3451                         break;
3452
3453                 if (path->slots[0] == 0 ||
3454                     path->slots[0] != pending_del_slot) {
3455                         if (root->ref_cows) {
3456                                 err = -EAGAIN;
3457                                 goto out;
3458                         }
3459                         if (pending_del_nr) {
3460                                 ret = btrfs_del_items(trans, root, path,
3461                                                 pending_del_slot,
3462                                                 pending_del_nr);
3463                                 BUG_ON(ret);
3464                                 pending_del_nr = 0;
3465                         }
3466                         btrfs_release_path(root, path);
3467                         goto search_again;
3468                 } else {
3469                         path->slots[0]--;
3470                 }
3471         }
3472 out:
3473         if (pending_del_nr) {
3474                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3475                                       pending_del_nr);
3476                 BUG_ON(ret);
3477         }
3478         btrfs_free_path(path);
3479         return err;
3480 }
3481
3482 /*
3483  * taken from block_truncate_page, but does cow as it zeros out
3484  * any bytes left in the last page in the file.
3485  */
3486 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3487 {
3488         struct inode *inode = mapping->host;
3489         struct btrfs_root *root = BTRFS_I(inode)->root;
3490         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3491         struct btrfs_ordered_extent *ordered;
3492         struct extent_state *cached_state = NULL;
3493         char *kaddr;
3494         u32 blocksize = root->sectorsize;
3495         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3496         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3497         struct page *page;
3498         int ret = 0;
3499         u64 page_start;
3500         u64 page_end;
3501
3502         if ((offset & (blocksize - 1)) == 0)
3503                 goto out;
3504         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3505         if (ret)
3506                 goto out;
3507
3508         ret = -ENOMEM;
3509 again:
3510         page = grab_cache_page(mapping, index);
3511         if (!page) {
3512                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3513                 goto out;
3514         }
3515
3516         page_start = page_offset(page);
3517         page_end = page_start + PAGE_CACHE_SIZE - 1;
3518
3519         if (!PageUptodate(page)) {
3520                 ret = btrfs_readpage(NULL, page);
3521                 lock_page(page);
3522                 if (page->mapping != mapping) {
3523                         unlock_page(page);
3524                         page_cache_release(page);
3525                         goto again;
3526                 }
3527                 if (!PageUptodate(page)) {
3528                         ret = -EIO;
3529                         goto out_unlock;
3530                 }
3531         }
3532         wait_on_page_writeback(page);
3533
3534         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3535                          GFP_NOFS);
3536         set_page_extent_mapped(page);
3537
3538         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3539         if (ordered) {
3540                 unlock_extent_cached(io_tree, page_start, page_end,
3541                                      &cached_state, GFP_NOFS);
3542                 unlock_page(page);
3543                 page_cache_release(page);
3544                 btrfs_start_ordered_extent(inode, ordered, 1);
3545                 btrfs_put_ordered_extent(ordered);
3546                 goto again;
3547         }
3548
3549         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3550                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3551                           0, 0, &cached_state, GFP_NOFS);
3552
3553         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3554                                         &cached_state);
3555         if (ret) {
3556                 unlock_extent_cached(io_tree, page_start, page_end,
3557                                      &cached_state, GFP_NOFS);
3558                 goto out_unlock;
3559         }
3560
3561         ret = 0;
3562         if (offset != PAGE_CACHE_SIZE) {
3563                 kaddr = kmap(page);
3564                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3565                 flush_dcache_page(page);
3566                 kunmap(page);
3567         }
3568         ClearPageChecked(page);
3569         set_page_dirty(page);
3570         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3571                              GFP_NOFS);
3572
3573 out_unlock:
3574         if (ret)
3575                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3576         unlock_page(page);
3577         page_cache_release(page);
3578 out:
3579         return ret;
3580 }
3581
3582 /*
3583  * This function puts in dummy file extents for the area we're creating a hole
3584  * for.  So if we are truncating this file to a larger size we need to insert
3585  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3586  * the range between oldsize and size
3587  */
3588 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3589 {
3590         struct btrfs_trans_handle *trans;
3591         struct btrfs_root *root = BTRFS_I(inode)->root;
3592         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3593         struct extent_map *em = NULL;
3594         struct extent_state *cached_state = NULL;
3595         u64 mask = root->sectorsize - 1;
3596         u64 hole_start = (oldsize + mask) & ~mask;
3597         u64 block_end = (size + mask) & ~mask;
3598         u64 last_byte;
3599         u64 cur_offset;
3600         u64 hole_size;
3601         int err = 0;
3602
3603         if (size <= hole_start)
3604                 return 0;
3605
3606         while (1) {
3607                 struct btrfs_ordered_extent *ordered;
3608                 btrfs_wait_ordered_range(inode, hole_start,
3609                                          block_end - hole_start);
3610                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3611                                  &cached_state, GFP_NOFS);
3612                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3613                 if (!ordered)
3614                         break;
3615                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3616                                      &cached_state, GFP_NOFS);
3617                 btrfs_put_ordered_extent(ordered);
3618         }
3619
3620         cur_offset = hole_start;
3621         while (1) {
3622                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3623                                 block_end - cur_offset, 0);
3624                 BUG_ON(IS_ERR(em) || !em);
3625                 last_byte = min(extent_map_end(em), block_end);
3626                 last_byte = (last_byte + mask) & ~mask;
3627                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3628                         u64 hint_byte = 0;
3629                         hole_size = last_byte - cur_offset;
3630
3631                         trans = btrfs_start_transaction(root, 2);
3632                         if (IS_ERR(trans)) {
3633                                 err = PTR_ERR(trans);
3634                                 break;
3635                         }
3636
3637                         err = btrfs_drop_extents(trans, inode, cur_offset,
3638                                                  cur_offset + hole_size,
3639                                                  &hint_byte, 1);
3640                         if (err)
3641                                 break;
3642
3643                         err = btrfs_insert_file_extent(trans, root,
3644                                         inode->i_ino, cur_offset, 0,
3645                                         0, hole_size, 0, hole_size,
3646                                         0, 0, 0);
3647                         if (err)
3648                                 break;
3649
3650                         btrfs_drop_extent_cache(inode, hole_start,
3651                                         last_byte - 1, 0);
3652
3653                         btrfs_end_transaction(trans, root);
3654                 }
3655                 free_extent_map(em);
3656                 em = NULL;
3657                 cur_offset = last_byte;
3658                 if (cur_offset >= block_end)
3659                         break;
3660         }
3661
3662         free_extent_map(em);
3663         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3664                              GFP_NOFS);
3665         return err;
3666 }
3667
3668 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3669 {
3670         loff_t oldsize = i_size_read(inode);
3671         int ret;
3672
3673         if (newsize == oldsize)
3674                 return 0;
3675
3676         if (newsize > oldsize) {
3677                 i_size_write(inode, newsize);
3678                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3679                 truncate_pagecache(inode, oldsize, newsize);
3680                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3681                 if (ret) {
3682                         btrfs_setsize(inode, oldsize);
3683                         return ret;
3684                 }
3685
3686                 mark_inode_dirty(inode);
3687         } else {
3688
3689                 /*
3690                  * We're truncating a file that used to have good data down to
3691                  * zero. Make sure it gets into the ordered flush list so that
3692                  * any new writes get down to disk quickly.
3693                  */
3694                 if (newsize == 0)
3695                         BTRFS_I(inode)->ordered_data_close = 1;
3696
3697                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3698                 truncate_setsize(inode, newsize);
3699                 ret = btrfs_truncate(inode);
3700         }
3701
3702         return ret;
3703 }
3704
3705 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3706 {
3707         struct inode *inode = dentry->d_inode;
3708         struct btrfs_root *root = BTRFS_I(inode)->root;
3709         int err;
3710
3711         if (btrfs_root_readonly(root))
3712                 return -EROFS;
3713
3714         err = inode_change_ok(inode, attr);
3715         if (err)
3716                 return err;
3717
3718         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3719                 err = btrfs_setsize(inode, attr->ia_size);
3720                 if (err)
3721                         return err;
3722         }
3723
3724         if (attr->ia_valid) {
3725                 setattr_copy(inode, attr);
3726                 mark_inode_dirty(inode);
3727
3728                 if (attr->ia_valid & ATTR_MODE)
3729                         err = btrfs_acl_chmod(inode);
3730         }
3731
3732         return err;
3733 }
3734
3735 void btrfs_evict_inode(struct inode *inode)
3736 {
3737         struct btrfs_trans_handle *trans;
3738         struct btrfs_root *root = BTRFS_I(inode)->root;
3739         unsigned long nr;
3740         int ret;
3741
3742         trace_btrfs_inode_evict(inode);
3743
3744         truncate_inode_pages(&inode->i_data, 0);
3745         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3746                                root == root->fs_info->tree_root))
3747                 goto no_delete;
3748
3749         if (is_bad_inode(inode)) {
3750                 btrfs_orphan_del(NULL, inode);
3751                 goto no_delete;
3752         }
3753         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3754         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3755
3756         if (root->fs_info->log_root_recovering) {
3757                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3758                 goto no_delete;
3759         }
3760
3761         if (inode->i_nlink > 0) {
3762                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3763                 goto no_delete;
3764         }
3765
3766         btrfs_i_size_write(inode, 0);
3767
3768         while (1) {
3769                 trans = btrfs_start_transaction(root, 0);
3770                 BUG_ON(IS_ERR(trans));
3771                 trans->block_rsv = root->orphan_block_rsv;
3772
3773                 ret = btrfs_block_rsv_check(trans, root,
3774                                             root->orphan_block_rsv, 0, 5);
3775                 if (ret) {
3776                         BUG_ON(ret != -EAGAIN);
3777                         ret = btrfs_commit_transaction(trans, root);
3778                         BUG_ON(ret);
3779                         continue;
3780                 }
3781
3782                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3783                 if (ret != -EAGAIN)
3784                         break;
3785
3786                 nr = trans->blocks_used;
3787                 btrfs_end_transaction(trans, root);
3788                 trans = NULL;
3789                 btrfs_btree_balance_dirty(root, nr);
3790
3791         }
3792
3793         if (ret == 0) {
3794                 ret = btrfs_orphan_del(trans, inode);
3795                 BUG_ON(ret);
3796         }
3797
3798         nr = trans->blocks_used;
3799         btrfs_end_transaction(trans, root);
3800         btrfs_btree_balance_dirty(root, nr);
3801 no_delete:
3802         end_writeback(inode);
3803         return;
3804 }
3805
3806 /*
3807  * this returns the key found in the dir entry in the location pointer.
3808  * If no dir entries were found, location->objectid is 0.
3809  */
3810 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3811                                struct btrfs_key *location)
3812 {
3813         const char *name = dentry->d_name.name;
3814         int namelen = dentry->d_name.len;
3815         struct btrfs_dir_item *di;
3816         struct btrfs_path *path;
3817         struct btrfs_root *root = BTRFS_I(dir)->root;
3818         int ret = 0;
3819
3820         path = btrfs_alloc_path();
3821         BUG_ON(!path);
3822
3823         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3824                                     namelen, 0);
3825         if (IS_ERR(di))
3826                 ret = PTR_ERR(di);
3827
3828         if (!di || IS_ERR(di))
3829                 goto out_err;
3830
3831         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3832 out:
3833         btrfs_free_path(path);
3834         return ret;
3835 out_err:
3836         location->objectid = 0;
3837         goto out;
3838 }
3839
3840 /*
3841  * when we hit a tree root in a directory, the btrfs part of the inode
3842  * needs to be changed to reflect the root directory of the tree root.  This
3843  * is kind of like crossing a mount point.
3844  */
3845 static int fixup_tree_root_location(struct btrfs_root *root,
3846                                     struct inode *dir,
3847                                     struct dentry *dentry,
3848                                     struct btrfs_key *location,
3849                                     struct btrfs_root **sub_root)
3850 {
3851         struct btrfs_path *path;
3852         struct btrfs_root *new_root;
3853         struct btrfs_root_ref *ref;
3854         struct extent_buffer *leaf;
3855         int ret;
3856         int err = 0;
3857
3858         path = btrfs_alloc_path();
3859         if (!path) {
3860                 err = -ENOMEM;
3861                 goto out;
3862         }
3863
3864         err = -ENOENT;
3865         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3866                                   BTRFS_I(dir)->root->root_key.objectid,
3867                                   location->objectid);
3868         if (ret) {
3869                 if (ret < 0)
3870                         err = ret;
3871                 goto out;
3872         }
3873
3874         leaf = path->nodes[0];
3875         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3876         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3877             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3878                 goto out;
3879
3880         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3881                                    (unsigned long)(ref + 1),
3882                                    dentry->d_name.len);
3883         if (ret)
3884                 goto out;
3885
3886         btrfs_release_path(root->fs_info->tree_root, path);
3887
3888         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3889         if (IS_ERR(new_root)) {
3890                 err = PTR_ERR(new_root);
3891                 goto out;
3892         }
3893
3894         if (btrfs_root_refs(&new_root->root_item) == 0) {
3895                 err = -ENOENT;
3896                 goto out;
3897         }
3898
3899         *sub_root = new_root;
3900         location->objectid = btrfs_root_dirid(&new_root->root_item);
3901         location->type = BTRFS_INODE_ITEM_KEY;
3902         location->offset = 0;
3903         err = 0;
3904 out:
3905         btrfs_free_path(path);
3906         return err;
3907 }
3908
3909 static void inode_tree_add(struct inode *inode)
3910 {
3911         struct btrfs_root *root = BTRFS_I(inode)->root;
3912         struct btrfs_inode *entry;
3913         struct rb_node **p;
3914         struct rb_node *parent;
3915 again:
3916         p = &root->inode_tree.rb_node;
3917         parent = NULL;
3918
3919         if (inode_unhashed(inode))
3920                 return;
3921
3922         spin_lock(&root->inode_lock);
3923         while (*p) {
3924                 parent = *p;
3925                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3926
3927                 if (inode->i_ino < entry->vfs_inode.i_ino)
3928                         p = &parent->rb_left;
3929                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3930                         p = &parent->rb_right;
3931                 else {
3932                         WARN_ON(!(entry->vfs_inode.i_state &
3933                                   (I_WILL_FREE | I_FREEING)));
3934                         rb_erase(parent, &root->inode_tree);
3935                         RB_CLEAR_NODE(parent);
3936                         spin_unlock(&root->inode_lock);
3937                         goto again;
3938                 }
3939         }
3940         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3941         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3942         spin_unlock(&root->inode_lock);
3943 }
3944
3945 static void inode_tree_del(struct inode *inode)
3946 {
3947         struct btrfs_root *root = BTRFS_I(inode)->root;
3948         int empty = 0;
3949
3950         spin_lock(&root->inode_lock);
3951         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3952                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3953                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3954                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3955         }
3956         spin_unlock(&root->inode_lock);
3957
3958         /*
3959          * Free space cache has inodes in the tree root, but the tree root has a
3960          * root_refs of 0, so this could end up dropping the tree root as a
3961          * snapshot, so we need the extra !root->fs_info->tree_root check to
3962          * make sure we don't drop it.
3963          */
3964         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3965             root != root->fs_info->tree_root) {
3966                 synchronize_srcu(&root->fs_info->subvol_srcu);
3967                 spin_lock(&root->inode_lock);
3968                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3969                 spin_unlock(&root->inode_lock);
3970                 if (empty)
3971                         btrfs_add_dead_root(root);
3972         }
3973 }
3974
3975 int btrfs_invalidate_inodes(struct btrfs_root *root)
3976 {
3977         struct rb_node *node;
3978         struct rb_node *prev;
3979         struct btrfs_inode *entry;
3980         struct inode *inode;
3981         u64 objectid = 0;
3982
3983         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3984
3985         spin_lock(&root->inode_lock);
3986 again:
3987         node = root->inode_tree.rb_node;
3988         prev = NULL;
3989         while (node) {
3990                 prev = node;
3991                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3992
3993                 if (objectid < entry->vfs_inode.i_ino)
3994                         node = node->rb_left;
3995                 else if (objectid > entry->vfs_inode.i_ino)
3996                         node = node->rb_right;
3997                 else
3998                         break;
3999         }
4000         if (!node) {
4001                 while (prev) {
4002                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4003                         if (objectid <= entry->vfs_inode.i_ino) {
4004                                 node = prev;
4005                                 break;
4006                         }
4007                         prev = rb_next(prev);
4008                 }
4009         }
4010         while (node) {
4011                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4012                 objectid = entry->vfs_inode.i_ino + 1;
4013                 inode = igrab(&entry->vfs_inode);
4014                 if (inode) {
4015                         spin_unlock(&root->inode_lock);
4016                         if (atomic_read(&inode->i_count) > 1)
4017                                 d_prune_aliases(inode);
4018                         /*
4019                          * btrfs_drop_inode will have it removed from
4020                          * the inode cache when its usage count
4021                          * hits zero.
4022                          */
4023                         iput(inode);
4024                         cond_resched();
4025                         spin_lock(&root->inode_lock);
4026                         goto again;
4027                 }
4028
4029                 if (cond_resched_lock(&root->inode_lock))
4030                         goto again;
4031
4032                 node = rb_next(node);
4033         }
4034         spin_unlock(&root->inode_lock);
4035         return 0;
4036 }
4037
4038 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4039 {
4040         struct btrfs_iget_args *args = p;
4041         inode->i_ino = args->ino;
4042         BTRFS_I(inode)->root = args->root;
4043         btrfs_set_inode_space_info(args->root, inode);
4044         return 0;
4045 }
4046
4047 static int btrfs_find_actor(struct inode *inode, void *opaque)
4048 {
4049         struct btrfs_iget_args *args = opaque;
4050         return args->ino == inode->i_ino &&
4051                 args->root == BTRFS_I(inode)->root;
4052 }
4053
4054 static struct inode *btrfs_iget_locked(struct super_block *s,
4055                                        u64 objectid,
4056                                        struct btrfs_root *root)
4057 {
4058         struct inode *inode;
4059         struct btrfs_iget_args args;
4060         args.ino = objectid;
4061         args.root = root;
4062
4063         inode = iget5_locked(s, objectid, btrfs_find_actor,
4064                              btrfs_init_locked_inode,
4065                              (void *)&args);
4066         return inode;
4067 }
4068
4069 /* Get an inode object given its location and corresponding root.
4070  * Returns in *is_new if the inode was read from disk
4071  */
4072 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4073                          struct btrfs_root *root, int *new)
4074 {
4075         struct inode *inode;
4076
4077         inode = btrfs_iget_locked(s, location->objectid, root);
4078         if (!inode)
4079                 return ERR_PTR(-ENOMEM);
4080
4081         if (inode->i_state & I_NEW) {
4082                 BTRFS_I(inode)->root = root;
4083                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4084                 btrfs_read_locked_inode(inode);
4085                 inode_tree_add(inode);
4086                 unlock_new_inode(inode);
4087                 if (new)
4088                         *new = 1;
4089         }
4090
4091         return inode;
4092 }
4093
4094 static struct inode *new_simple_dir(struct super_block *s,
4095                                     struct btrfs_key *key,
4096                                     struct btrfs_root *root)
4097 {
4098         struct inode *inode = new_inode(s);
4099
4100         if (!inode)
4101                 return ERR_PTR(-ENOMEM);
4102
4103         BTRFS_I(inode)->root = root;
4104         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4105         BTRFS_I(inode)->dummy_inode = 1;
4106
4107         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4108         inode->i_op = &simple_dir_inode_operations;
4109         inode->i_fop = &simple_dir_operations;
4110         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4111         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4112
4113         return inode;
4114 }
4115
4116 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4117 {
4118         struct inode *inode;
4119         struct btrfs_root *root = BTRFS_I(dir)->root;
4120         struct btrfs_root *sub_root = root;
4121         struct btrfs_key location;
4122         int index;
4123         int ret;
4124
4125         if (dentry->d_name.len > BTRFS_NAME_LEN)
4126                 return ERR_PTR(-ENAMETOOLONG);
4127
4128         ret = btrfs_inode_by_name(dir, dentry, &location);
4129
4130         if (ret < 0)
4131                 return ERR_PTR(ret);
4132
4133         if (location.objectid == 0)
4134                 return NULL;
4135
4136         if (location.type == BTRFS_INODE_ITEM_KEY) {
4137                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4138                 return inode;
4139         }
4140
4141         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4142
4143         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4144         ret = fixup_tree_root_location(root, dir, dentry,
4145                                        &location, &sub_root);
4146         if (ret < 0) {
4147                 if (ret != -ENOENT)
4148                         inode = ERR_PTR(ret);
4149                 else
4150                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4151         } else {
4152                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4153         }
4154         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4155
4156         if (!IS_ERR(inode) && root != sub_root) {
4157                 down_read(&root->fs_info->cleanup_work_sem);
4158                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4159                         ret = btrfs_orphan_cleanup(sub_root);
4160                 up_read(&root->fs_info->cleanup_work_sem);
4161                 if (ret)
4162                         inode = ERR_PTR(ret);
4163         }
4164
4165         return inode;
4166 }
4167
4168 static int btrfs_dentry_delete(const struct dentry *dentry)
4169 {
4170         struct btrfs_root *root;
4171
4172         if (!dentry->d_inode && !IS_ROOT(dentry))
4173                 dentry = dentry->d_parent;
4174
4175         if (dentry->d_inode) {
4176                 root = BTRFS_I(dentry->d_inode)->root;
4177                 if (btrfs_root_refs(&root->root_item) == 0)
4178                         return 1;
4179         }
4180         return 0;
4181 }
4182
4183 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4184                                    struct nameidata *nd)
4185 {
4186         struct inode *inode;
4187
4188         inode = btrfs_lookup_dentry(dir, dentry);
4189         if (IS_ERR(inode))
4190                 return ERR_CAST(inode);
4191
4192         return d_splice_alias(inode, dentry);
4193 }
4194
4195 static unsigned char btrfs_filetype_table[] = {
4196         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4197 };
4198
4199 static int btrfs_real_readdir(struct file *filp, void *dirent,
4200                               filldir_t filldir)
4201 {
4202         struct inode *inode = filp->f_dentry->d_inode;
4203         struct btrfs_root *root = BTRFS_I(inode)->root;
4204         struct btrfs_item *item;
4205         struct btrfs_dir_item *di;
4206         struct btrfs_key key;
4207         struct btrfs_key found_key;
4208         struct btrfs_path *path;
4209         int ret;
4210         struct extent_buffer *leaf;
4211         int slot;
4212         unsigned char d_type;
4213         int over = 0;
4214         u32 di_cur;
4215         u32 di_total;
4216         u32 di_len;
4217         int key_type = BTRFS_DIR_INDEX_KEY;
4218         char tmp_name[32];
4219         char *name_ptr;
4220         int name_len;
4221
4222         /* FIXME, use a real flag for deciding about the key type */
4223         if (root->fs_info->tree_root == root)
4224                 key_type = BTRFS_DIR_ITEM_KEY;
4225
4226         /* special case for "." */
4227         if (filp->f_pos == 0) {
4228                 over = filldir(dirent, ".", 1,
4229                                1, inode->i_ino,
4230                                DT_DIR);
4231                 if (over)
4232                         return 0;
4233                 filp->f_pos = 1;
4234         }
4235         /* special case for .., just use the back ref */
4236         if (filp->f_pos == 1) {
4237                 u64 pino = parent_ino(filp->f_path.dentry);
4238                 over = filldir(dirent, "..", 2,
4239                                2, pino, DT_DIR);
4240                 if (over)
4241                         return 0;
4242                 filp->f_pos = 2;
4243         }
4244         path = btrfs_alloc_path();
4245         path->reada = 2;
4246
4247         btrfs_set_key_type(&key, key_type);
4248         key.offset = filp->f_pos;
4249         key.objectid = inode->i_ino;
4250
4251         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4252         if (ret < 0)
4253                 goto err;
4254
4255         while (1) {
4256                 leaf = path->nodes[0];
4257                 slot = path->slots[0];
4258                 if (slot >= btrfs_header_nritems(leaf)) {
4259                         ret = btrfs_next_leaf(root, path);
4260                         if (ret < 0)
4261                                 goto err;
4262                         else if (ret > 0)
4263                                 break;
4264                         continue;
4265                 }
4266
4267                 item = btrfs_item_nr(leaf, slot);
4268                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4269
4270                 if (found_key.objectid != key.objectid)
4271                         break;
4272                 if (btrfs_key_type(&found_key) != key_type)
4273                         break;
4274                 if (found_key.offset < filp->f_pos)
4275                         goto next;
4276
4277                 filp->f_pos = found_key.offset;
4278
4279                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4280                 di_cur = 0;
4281                 di_total = btrfs_item_size(leaf, item);
4282
4283                 while (di_cur < di_total) {
4284                         struct btrfs_key location;
4285
4286                         if (verify_dir_item(root, leaf, di))
4287                                 break;
4288
4289                         name_len = btrfs_dir_name_len(leaf, di);
4290                         if (name_len <= sizeof(tmp_name)) {
4291                                 name_ptr = tmp_name;
4292                         } else {
4293                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4294                                 if (!name_ptr) {
4295                                         ret = -ENOMEM;
4296                                         goto err;
4297                                 }
4298                         }
4299                         read_extent_buffer(leaf, name_ptr,
4300                                            (unsigned long)(di + 1), name_len);
4301
4302                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4303                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4304
4305                         /* is this a reference to our own snapshot? If so
4306                          * skip it
4307                          */
4308                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4309                             location.objectid == root->root_key.objectid) {
4310                                 over = 0;
4311                                 goto skip;
4312                         }
4313                         over = filldir(dirent, name_ptr, name_len,
4314                                        found_key.offset, location.objectid,
4315                                        d_type);
4316
4317 skip:
4318                         if (name_ptr != tmp_name)
4319                                 kfree(name_ptr);
4320
4321                         if (over)
4322                                 goto nopos;
4323                         di_len = btrfs_dir_name_len(leaf, di) +
4324                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4325                         di_cur += di_len;
4326                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4327                 }
4328 next:
4329                 path->slots[0]++;
4330         }
4331
4332         /* Reached end of directory/root. Bump pos past the last item. */
4333         if (key_type == BTRFS_DIR_INDEX_KEY)
4334                 /*
4335                  * 32-bit glibc will use getdents64, but then strtol -
4336                  * so the last number we can serve is this.
4337                  */
4338                 filp->f_pos = 0x7fffffff;
4339         else
4340                 filp->f_pos++;
4341 nopos:
4342         ret = 0;
4343 err:
4344         btrfs_free_path(path);
4345         return ret;
4346 }
4347
4348 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4349 {
4350         struct btrfs_root *root = BTRFS_I(inode)->root;
4351         struct btrfs_trans_handle *trans;
4352         int ret = 0;
4353         bool nolock = false;
4354
4355         if (BTRFS_I(inode)->dummy_inode)
4356                 return 0;
4357
4358         smp_mb();
4359         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4360
4361         if (wbc->sync_mode == WB_SYNC_ALL) {
4362                 if (nolock)
4363                         trans = btrfs_join_transaction_nolock(root);
4364                 else
4365                         trans = btrfs_join_transaction(root);
4366                 if (IS_ERR(trans))
4367                         return PTR_ERR(trans);
4368                 if (nolock)
4369                         ret = btrfs_end_transaction_nolock(trans, root);
4370                 else
4371                         ret = btrfs_commit_transaction(trans, root);
4372         }
4373         return ret;
4374 }
4375
4376 /*
4377  * This is somewhat expensive, updating the tree every time the
4378  * inode changes.  But, it is most likely to find the inode in cache.
4379  * FIXME, needs more benchmarking...there are no reasons other than performance
4380  * to keep or drop this code.
4381  */
4382 void btrfs_dirty_inode(struct inode *inode)
4383 {
4384         struct btrfs_root *root = BTRFS_I(inode)->root;
4385         struct btrfs_trans_handle *trans;
4386         int ret;
4387
4388         if (BTRFS_I(inode)->dummy_inode)
4389                 return;
4390
4391         trans = btrfs_join_transaction(root);
4392         BUG_ON(IS_ERR(trans));
4393
4394         ret = btrfs_update_inode(trans, root, inode);
4395         if (ret && ret == -ENOSPC) {
4396                 /* whoops, lets try again with the full transaction */
4397                 btrfs_end_transaction(trans, root);
4398                 trans = btrfs_start_transaction(root, 1);
4399                 if (IS_ERR(trans)) {
4400                         if (printk_ratelimit()) {
4401                                 printk(KERN_ERR "btrfs: fail to "
4402                                        "dirty  inode %lu error %ld\n",
4403                                        inode->i_ino, PTR_ERR(trans));
4404                         }
4405                         return;
4406                 }
4407
4408                 ret = btrfs_update_inode(trans, root, inode);
4409                 if (ret) {
4410                         if (printk_ratelimit()) {
4411                                 printk(KERN_ERR "btrfs: fail to "
4412                                        "dirty  inode %lu error %d\n",
4413                                        inode->i_ino, ret);
4414                         }
4415                 }
4416         }
4417         btrfs_end_transaction(trans, root);
4418 }
4419
4420 /*
4421  * find the highest existing sequence number in a directory
4422  * and then set the in-memory index_cnt variable to reflect
4423  * free sequence numbers
4424  */
4425 static int btrfs_set_inode_index_count(struct inode *inode)
4426 {
4427         struct btrfs_root *root = BTRFS_I(inode)->root;
4428         struct btrfs_key key, found_key;
4429         struct btrfs_path *path;
4430         struct extent_buffer *leaf;
4431         int ret;
4432
4433         key.objectid = inode->i_ino;
4434         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4435         key.offset = (u64)-1;
4436
4437         path = btrfs_alloc_path();
4438         if (!path)
4439                 return -ENOMEM;
4440
4441         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4442         if (ret < 0)
4443                 goto out;
4444         /* FIXME: we should be able to handle this */
4445         if (ret == 0)
4446                 goto out;
4447         ret = 0;
4448
4449         /*
4450          * MAGIC NUMBER EXPLANATION:
4451          * since we search a directory based on f_pos we have to start at 2
4452          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4453          * else has to start at 2
4454          */
4455         if (path->slots[0] == 0) {
4456                 BTRFS_I(inode)->index_cnt = 2;
4457                 goto out;
4458         }
4459
4460         path->slots[0]--;
4461
4462         leaf = path->nodes[0];
4463         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4464
4465         if (found_key.objectid != inode->i_ino ||
4466             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4467                 BTRFS_I(inode)->index_cnt = 2;
4468                 goto out;
4469         }
4470
4471         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4472 out:
4473         btrfs_free_path(path);
4474         return ret;
4475 }
4476
4477 /*
4478  * helper to find a free sequence number in a given directory.  This current
4479  * code is very simple, later versions will do smarter things in the btree
4480  */
4481 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4482 {
4483         int ret = 0;
4484
4485         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4486                 ret = btrfs_set_inode_index_count(dir);
4487                 if (ret)
4488                         return ret;
4489         }
4490
4491         *index = BTRFS_I(dir)->index_cnt;
4492         BTRFS_I(dir)->index_cnt++;
4493
4494         return ret;
4495 }
4496
4497 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4498                                      struct btrfs_root *root,
4499                                      struct inode *dir,
4500                                      const char *name, int name_len,
4501                                      u64 ref_objectid, u64 objectid, int mode,
4502                                      u64 *index)
4503 {
4504         struct inode *inode;
4505         struct btrfs_inode_item *inode_item;
4506         struct btrfs_key *location;
4507         struct btrfs_path *path;
4508         struct btrfs_inode_ref *ref;
4509         struct btrfs_key key[2];
4510         u32 sizes[2];
4511         unsigned long ptr;
4512         int ret;
4513         int owner;
4514
4515         path = btrfs_alloc_path();
4516         BUG_ON(!path);
4517
4518         inode = new_inode(root->fs_info->sb);
4519         if (!inode) {
4520                 btrfs_free_path(path);
4521                 return ERR_PTR(-ENOMEM);
4522         }
4523
4524         if (dir) {
4525                 trace_btrfs_inode_request(dir);
4526
4527                 ret = btrfs_set_inode_index(dir, index);
4528                 if (ret) {
4529                         btrfs_free_path(path);
4530                         iput(inode);
4531                         return ERR_PTR(ret);
4532                 }
4533         }
4534         /*
4535          * index_cnt is ignored for everything but a dir,
4536          * btrfs_get_inode_index_count has an explanation for the magic
4537          * number
4538          */
4539         BTRFS_I(inode)->index_cnt = 2;
4540         BTRFS_I(inode)->root = root;
4541         BTRFS_I(inode)->generation = trans->transid;
4542         inode->i_generation = BTRFS_I(inode)->generation;
4543         btrfs_set_inode_space_info(root, inode);
4544
4545         if (mode & S_IFDIR)
4546                 owner = 0;
4547         else
4548                 owner = 1;
4549
4550         key[0].objectid = objectid;
4551         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4552         key[0].offset = 0;
4553
4554         key[1].objectid = objectid;
4555         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4556         key[1].offset = ref_objectid;
4557
4558         sizes[0] = sizeof(struct btrfs_inode_item);
4559         sizes[1] = name_len + sizeof(*ref);
4560
4561         path->leave_spinning = 1;
4562         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4563         if (ret != 0)
4564                 goto fail;
4565
4566         inode_init_owner(inode, dir, mode);
4567         inode->i_ino = objectid;
4568         inode_set_bytes(inode, 0);
4569         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4570         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4571                                   struct btrfs_inode_item);
4572         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4573
4574         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4575                              struct btrfs_inode_ref);
4576         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4577         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4578         ptr = (unsigned long)(ref + 1);
4579         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4580
4581         btrfs_mark_buffer_dirty(path->nodes[0]);
4582         btrfs_free_path(path);
4583
4584         location = &BTRFS_I(inode)->location;
4585         location->objectid = objectid;
4586         location->offset = 0;
4587         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4588
4589         btrfs_inherit_iflags(inode, dir);
4590
4591         if ((mode & S_IFREG)) {
4592                 if (btrfs_test_opt(root, NODATASUM))
4593                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4594                 if (btrfs_test_opt(root, NODATACOW) ||
4595                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4596                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4597         }
4598
4599         insert_inode_hash(inode);
4600         inode_tree_add(inode);
4601
4602         trace_btrfs_inode_new(inode);
4603
4604         return inode;
4605 fail:
4606         if (dir)
4607                 BTRFS_I(dir)->index_cnt--;
4608         btrfs_free_path(path);
4609         iput(inode);
4610         return ERR_PTR(ret);
4611 }
4612
4613 static inline u8 btrfs_inode_type(struct inode *inode)
4614 {
4615         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4616 }
4617
4618 /*
4619  * utility function to add 'inode' into 'parent_inode' with
4620  * a give name and a given sequence number.
4621  * if 'add_backref' is true, also insert a backref from the
4622  * inode to the parent directory.
4623  */
4624 int btrfs_add_link(struct btrfs_trans_handle *trans,
4625                    struct inode *parent_inode, struct inode *inode,
4626                    const char *name, int name_len, int add_backref, u64 index)
4627 {
4628         int ret = 0;
4629         struct btrfs_key key;
4630         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4631
4632         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4633                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4634         } else {
4635                 key.objectid = inode->i_ino;
4636                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4637                 key.offset = 0;
4638         }
4639
4640         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4641                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4642                                          key.objectid, root->root_key.objectid,
4643                                          parent_inode->i_ino,
4644                                          index, name, name_len);
4645         } else if (add_backref) {
4646                 ret = btrfs_insert_inode_ref(trans, root,
4647                                              name, name_len, inode->i_ino,
4648                                              parent_inode->i_ino, index);
4649         }
4650
4651         if (ret == 0) {
4652                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4653                                             parent_inode->i_ino, &key,
4654                                             btrfs_inode_type(inode), index);
4655                 BUG_ON(ret);
4656
4657                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4658                                    name_len * 2);
4659                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4660                 ret = btrfs_update_inode(trans, root, parent_inode);
4661         }
4662         return ret;
4663 }
4664
4665 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4666                             struct inode *dir, struct dentry *dentry,
4667                             struct inode *inode, int backref, u64 index)
4668 {
4669         int err = btrfs_add_link(trans, dir, inode,
4670                                  dentry->d_name.name, dentry->d_name.len,
4671                                  backref, index);
4672         if (!err) {
4673                 d_instantiate(dentry, inode);
4674                 return 0;
4675         }
4676         if (err > 0)
4677                 err = -EEXIST;
4678         return err;
4679 }
4680
4681 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4682                         int mode, dev_t rdev)
4683 {
4684         struct btrfs_trans_handle *trans;
4685         struct btrfs_root *root = BTRFS_I(dir)->root;
4686         struct inode *inode = NULL;
4687         int err;
4688         int drop_inode = 0;
4689         u64 objectid;
4690         unsigned long nr = 0;
4691         u64 index = 0;
4692
4693         if (!new_valid_dev(rdev))
4694                 return -EINVAL;
4695
4696         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4697         if (err)
4698                 return err;
4699
4700         /*
4701          * 2 for inode item and ref
4702          * 2 for dir items
4703          * 1 for xattr if selinux is on
4704          */
4705         trans = btrfs_start_transaction(root, 5);
4706         if (IS_ERR(trans))
4707                 return PTR_ERR(trans);
4708
4709         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4710                                 dentry->d_name.len, dir->i_ino, objectid,
4711                                 mode, &index);
4712         if (IS_ERR(inode)) {
4713                 err = PTR_ERR(inode);
4714                 goto out_unlock;
4715         }
4716
4717         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4718         if (err) {
4719                 drop_inode = 1;
4720                 goto out_unlock;
4721         }
4722
4723         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4724         if (err)
4725                 drop_inode = 1;
4726         else {
4727                 inode->i_op = &btrfs_special_inode_operations;
4728                 init_special_inode(inode, inode->i_mode, rdev);
4729                 btrfs_update_inode(trans, root, inode);
4730         }
4731 out_unlock:
4732         nr = trans->blocks_used;
4733         btrfs_end_transaction_throttle(trans, root);
4734         btrfs_btree_balance_dirty(root, nr);
4735         if (drop_inode) {
4736                 inode_dec_link_count(inode);
4737                 iput(inode);
4738         }
4739         return err;
4740 }
4741
4742 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4743                         int mode, struct nameidata *nd)
4744 {
4745         struct btrfs_trans_handle *trans;
4746         struct btrfs_root *root = BTRFS_I(dir)->root;
4747         struct inode *inode = NULL;
4748         int drop_inode = 0;
4749         int err;
4750         unsigned long nr = 0;
4751         u64 objectid;
4752         u64 index = 0;
4753
4754         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4755         if (err)
4756                 return err;
4757         /*
4758          * 2 for inode item and ref
4759          * 2 for dir items
4760          * 1 for xattr if selinux is on
4761          */
4762         trans = btrfs_start_transaction(root, 5);
4763         if (IS_ERR(trans))
4764                 return PTR_ERR(trans);
4765
4766         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4767                                 dentry->d_name.len, dir->i_ino, objectid,
4768                                 mode, &index);
4769         if (IS_ERR(inode)) {
4770                 err = PTR_ERR(inode);
4771                 goto out_unlock;
4772         }
4773
4774         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4775         if (err) {
4776                 drop_inode = 1;
4777                 goto out_unlock;
4778         }
4779
4780         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4781         if (err)
4782                 drop_inode = 1;
4783         else {
4784                 inode->i_mapping->a_ops = &btrfs_aops;
4785                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4786                 inode->i_fop = &btrfs_file_operations;
4787                 inode->i_op = &btrfs_file_inode_operations;
4788                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4789         }
4790 out_unlock:
4791         nr = trans->blocks_used;
4792         btrfs_end_transaction_throttle(trans, root);
4793         if (drop_inode) {
4794                 inode_dec_link_count(inode);
4795                 iput(inode);
4796         }
4797         btrfs_btree_balance_dirty(root, nr);
4798         return err;
4799 }
4800
4801 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4802                       struct dentry *dentry)
4803 {
4804         struct btrfs_trans_handle *trans;
4805         struct btrfs_root *root = BTRFS_I(dir)->root;
4806         struct inode *inode = old_dentry->d_inode;
4807         u64 index;
4808         unsigned long nr = 0;
4809         int err;
4810         int drop_inode = 0;
4811
4812         /* do not allow sys_link's with other subvols of the same device */
4813         if (root->objectid != BTRFS_I(inode)->root->objectid)
4814                 return -EXDEV;
4815
4816         if (inode->i_nlink == ~0U)
4817                 return -EMLINK;
4818
4819         err = btrfs_set_inode_index(dir, &index);
4820         if (err)
4821                 goto fail;
4822
4823         /*
4824          * 2 items for inode and inode ref
4825          * 2 items for dir items
4826          * 1 item for parent inode
4827          */
4828         trans = btrfs_start_transaction(root, 5);
4829         if (IS_ERR(trans)) {
4830                 err = PTR_ERR(trans);
4831                 goto fail;
4832         }
4833
4834         btrfs_inc_nlink(inode);
4835         inode->i_ctime = CURRENT_TIME;
4836         ihold(inode);
4837
4838         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4839
4840         if (err) {
4841                 drop_inode = 1;
4842         } else {
4843                 struct dentry *parent = dget_parent(dentry);
4844                 err = btrfs_update_inode(trans, root, inode);
4845                 BUG_ON(err);
4846                 btrfs_log_new_name(trans, inode, NULL, parent);
4847                 dput(parent);
4848         }
4849
4850         nr = trans->blocks_used;
4851         btrfs_end_transaction_throttle(trans, root);
4852 fail:
4853         if (drop_inode) {
4854                 inode_dec_link_count(inode);
4855                 iput(inode);
4856         }
4857         btrfs_btree_balance_dirty(root, nr);
4858         return err;
4859 }
4860
4861 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4862 {
4863         struct inode *inode = NULL;
4864         struct btrfs_trans_handle *trans;
4865         struct btrfs_root *root = BTRFS_I(dir)->root;
4866         int err = 0;
4867         int drop_on_err = 0;
4868         u64 objectid = 0;
4869         u64 index = 0;
4870         unsigned long nr = 1;
4871
4872         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4873         if (err)
4874                 return err;
4875
4876         /*
4877          * 2 items for inode and ref
4878          * 2 items for dir items
4879          * 1 for xattr if selinux is on
4880          */
4881         trans = btrfs_start_transaction(root, 5);
4882         if (IS_ERR(trans))
4883                 return PTR_ERR(trans);
4884
4885         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4886                                 dentry->d_name.len, dir->i_ino, objectid,
4887                                 S_IFDIR | mode, &index);
4888         if (IS_ERR(inode)) {
4889                 err = PTR_ERR(inode);
4890                 goto out_fail;
4891         }
4892
4893         drop_on_err = 1;
4894
4895         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4896         if (err)
4897                 goto out_fail;
4898
4899         inode->i_op = &btrfs_dir_inode_operations;
4900         inode->i_fop = &btrfs_dir_file_operations;
4901
4902         btrfs_i_size_write(inode, 0);
4903         err = btrfs_update_inode(trans, root, inode);
4904         if (err)
4905                 goto out_fail;
4906
4907         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4908                              dentry->d_name.len, 0, index);
4909         if (err)
4910                 goto out_fail;
4911
4912         d_instantiate(dentry, inode);
4913         drop_on_err = 0;
4914
4915 out_fail:
4916         nr = trans->blocks_used;
4917         btrfs_end_transaction_throttle(trans, root);
4918         if (drop_on_err)
4919                 iput(inode);
4920         btrfs_btree_balance_dirty(root, nr);
4921         return err;
4922 }
4923
4924 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4925  * and an extent that you want to insert, deal with overlap and insert
4926  * the new extent into the tree.
4927  */
4928 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4929                                 struct extent_map *existing,
4930                                 struct extent_map *em,
4931                                 u64 map_start, u64 map_len)
4932 {
4933         u64 start_diff;
4934
4935         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4936         start_diff = map_start - em->start;
4937         em->start = map_start;
4938         em->len = map_len;
4939         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4940             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4941                 em->block_start += start_diff;
4942                 em->block_len -= start_diff;
4943         }
4944         return add_extent_mapping(em_tree, em);
4945 }
4946
4947 static noinline int uncompress_inline(struct btrfs_path *path,
4948                                       struct inode *inode, struct page *page,
4949                                       size_t pg_offset, u64 extent_offset,
4950                                       struct btrfs_file_extent_item *item)
4951 {
4952         int ret;
4953         struct extent_buffer *leaf = path->nodes[0];
4954         char *tmp;
4955         size_t max_size;
4956         unsigned long inline_size;
4957         unsigned long ptr;
4958         int compress_type;
4959
4960         WARN_ON(pg_offset != 0);
4961         compress_type = btrfs_file_extent_compression(leaf, item);
4962         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4963         inline_size = btrfs_file_extent_inline_item_len(leaf,
4964                                         btrfs_item_nr(leaf, path->slots[0]));
4965         tmp = kmalloc(inline_size, GFP_NOFS);
4966         if (!tmp)
4967                 return -ENOMEM;
4968         ptr = btrfs_file_extent_inline_start(item);
4969
4970         read_extent_buffer(leaf, tmp, ptr, inline_size);
4971
4972         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4973         ret = btrfs_decompress(compress_type, tmp, page,
4974                                extent_offset, inline_size, max_size);
4975         if (ret) {
4976                 char *kaddr = kmap_atomic(page, KM_USER0);
4977                 unsigned long copy_size = min_t(u64,
4978                                   PAGE_CACHE_SIZE - pg_offset,
4979                                   max_size - extent_offset);
4980                 memset(kaddr + pg_offset, 0, copy_size);
4981                 kunmap_atomic(kaddr, KM_USER0);
4982         }
4983         kfree(tmp);
4984         return 0;
4985 }
4986
4987 /*
4988  * a bit scary, this does extent mapping from logical file offset to the disk.
4989  * the ugly parts come from merging extents from the disk with the in-ram
4990  * representation.  This gets more complex because of the data=ordered code,
4991  * where the in-ram extents might be locked pending data=ordered completion.
4992  *
4993  * This also copies inline extents directly into the page.
4994  */
4995
4996 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4997                                     size_t pg_offset, u64 start, u64 len,
4998                                     int create)
4999 {
5000         int ret;
5001         int err = 0;
5002         u64 bytenr;
5003         u64 extent_start = 0;
5004         u64 extent_end = 0;
5005         u64 objectid = inode->i_ino;
5006         u32 found_type;
5007         struct btrfs_path *path = NULL;
5008         struct btrfs_root *root = BTRFS_I(inode)->root;
5009         struct btrfs_file_extent_item *item;
5010         struct extent_buffer *leaf;
5011         struct btrfs_key found_key;
5012         struct extent_map *em = NULL;
5013         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5014         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5015         struct btrfs_trans_handle *trans = NULL;
5016         int compress_type;
5017
5018 again:
5019         read_lock(&em_tree->lock);
5020         em = lookup_extent_mapping(em_tree, start, len);
5021         if (em)
5022                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5023         read_unlock(&em_tree->lock);
5024
5025         if (em) {
5026                 if (em->start > start || em->start + em->len <= start)
5027                         free_extent_map(em);
5028                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5029                         free_extent_map(em);
5030                 else
5031                         goto out;
5032         }
5033         em = alloc_extent_map(GFP_NOFS);
5034         if (!em) {
5035                 err = -ENOMEM;
5036                 goto out;
5037         }
5038         em->bdev = root->fs_info->fs_devices->latest_bdev;
5039         em->start = EXTENT_MAP_HOLE;
5040         em->orig_start = EXTENT_MAP_HOLE;
5041         em->len = (u64)-1;
5042         em->block_len = (u64)-1;
5043
5044         if (!path) {
5045                 path = btrfs_alloc_path();
5046                 BUG_ON(!path);
5047         }
5048
5049         ret = btrfs_lookup_file_extent(trans, root, path,
5050                                        objectid, start, trans != NULL);
5051         if (ret < 0) {
5052                 err = ret;
5053                 goto out;
5054         }
5055
5056         if (ret != 0) {
5057                 if (path->slots[0] == 0)
5058                         goto not_found;
5059                 path->slots[0]--;
5060         }
5061
5062         leaf = path->nodes[0];
5063         item = btrfs_item_ptr(leaf, path->slots[0],
5064                               struct btrfs_file_extent_item);
5065         /* are we inside the extent that was found? */
5066         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5067         found_type = btrfs_key_type(&found_key);
5068         if (found_key.objectid != objectid ||
5069             found_type != BTRFS_EXTENT_DATA_KEY) {
5070                 goto not_found;
5071         }
5072
5073         found_type = btrfs_file_extent_type(leaf, item);
5074         extent_start = found_key.offset;
5075         compress_type = btrfs_file_extent_compression(leaf, item);
5076         if (found_type == BTRFS_FILE_EXTENT_REG ||
5077             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5078                 extent_end = extent_start +
5079                        btrfs_file_extent_num_bytes(leaf, item);
5080         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5081                 size_t size;
5082                 size = btrfs_file_extent_inline_len(leaf, item);
5083                 extent_end = (extent_start + size + root->sectorsize - 1) &
5084                         ~((u64)root->sectorsize - 1);
5085         }
5086
5087         if (start >= extent_end) {
5088                 path->slots[0]++;
5089                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5090                         ret = btrfs_next_leaf(root, path);
5091                         if (ret < 0) {
5092                                 err = ret;
5093                                 goto out;
5094                         }
5095                         if (ret > 0)
5096                                 goto not_found;
5097                         leaf = path->nodes[0];
5098                 }
5099                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5100                 if (found_key.objectid != objectid ||
5101                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5102                         goto not_found;
5103                 if (start + len <= found_key.offset)
5104                         goto not_found;
5105                 em->start = start;
5106                 em->len = found_key.offset - start;
5107                 goto not_found_em;
5108         }
5109
5110         if (found_type == BTRFS_FILE_EXTENT_REG ||
5111             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5112                 em->start = extent_start;
5113                 em->len = extent_end - extent_start;
5114                 em->orig_start = extent_start -
5115                                  btrfs_file_extent_offset(leaf, item);
5116                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5117                 if (bytenr == 0) {
5118                         em->block_start = EXTENT_MAP_HOLE;
5119                         goto insert;
5120                 }
5121                 if (compress_type != BTRFS_COMPRESS_NONE) {
5122                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5123                         em->compress_type = compress_type;
5124                         em->block_start = bytenr;
5125                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5126                                                                          item);
5127                 } else {
5128                         bytenr += btrfs_file_extent_offset(leaf, item);
5129                         em->block_start = bytenr;
5130                         em->block_len = em->len;
5131                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5132                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5133                 }
5134                 goto insert;
5135         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5136                 unsigned long ptr;
5137                 char *map;
5138                 size_t size;
5139                 size_t extent_offset;
5140                 size_t copy_size;
5141
5142                 em->block_start = EXTENT_MAP_INLINE;
5143                 if (!page || create) {
5144                         em->start = extent_start;
5145                         em->len = extent_end - extent_start;
5146                         goto out;
5147                 }
5148
5149                 size = btrfs_file_extent_inline_len(leaf, item);
5150                 extent_offset = page_offset(page) + pg_offset - extent_start;
5151                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5152                                 size - extent_offset);
5153                 em->start = extent_start + extent_offset;
5154                 em->len = (copy_size + root->sectorsize - 1) &
5155                         ~((u64)root->sectorsize - 1);
5156                 em->orig_start = EXTENT_MAP_INLINE;
5157                 if (compress_type) {
5158                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5159                         em->compress_type = compress_type;
5160                 }
5161                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5162                 if (create == 0 && !PageUptodate(page)) {
5163                         if (btrfs_file_extent_compression(leaf, item) !=
5164                             BTRFS_COMPRESS_NONE) {
5165                                 ret = uncompress_inline(path, inode, page,
5166                                                         pg_offset,
5167                                                         extent_offset, item);
5168                                 BUG_ON(ret);
5169                         } else {
5170                                 map = kmap(page);
5171                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5172                                                    copy_size);
5173                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5174                                         memset(map + pg_offset + copy_size, 0,
5175                                                PAGE_CACHE_SIZE - pg_offset -
5176                                                copy_size);
5177                                 }
5178                                 kunmap(page);
5179                         }
5180                         flush_dcache_page(page);
5181                 } else if (create && PageUptodate(page)) {
5182                         WARN_ON(1);
5183                         if (!trans) {
5184                                 kunmap(page);
5185                                 free_extent_map(em);
5186                                 em = NULL;
5187                                 btrfs_release_path(root, path);
5188                                 trans = btrfs_join_transaction(root);
5189                                 if (IS_ERR(trans))
5190                                         return ERR_CAST(trans);
5191                                 goto again;
5192                         }
5193                         map = kmap(page);
5194                         write_extent_buffer(leaf, map + pg_offset, ptr,
5195                                             copy_size);
5196                         kunmap(page);
5197                         btrfs_mark_buffer_dirty(leaf);
5198                 }
5199                 set_extent_uptodate(io_tree, em->start,
5200                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5201                 goto insert;
5202         } else {
5203                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5204                 WARN_ON(1);
5205         }
5206 not_found:
5207         em->start = start;
5208         em->len = len;
5209 not_found_em:
5210         em->block_start = EXTENT_MAP_HOLE;
5211         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5212 insert:
5213         btrfs_release_path(root, path);
5214         if (em->start > start || extent_map_end(em) <= start) {
5215                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5216                        "[%llu %llu]\n", (unsigned long long)em->start,
5217                        (unsigned long long)em->len,
5218                        (unsigned long long)start,
5219                        (unsigned long long)len);
5220                 err = -EIO;
5221                 goto out;
5222         }
5223
5224         err = 0;
5225         write_lock(&em_tree->lock);
5226         ret = add_extent_mapping(em_tree, em);
5227         /* it is possible that someone inserted the extent into the tree
5228          * while we had the lock dropped.  It is also possible that
5229          * an overlapping map exists in the tree
5230          */
5231         if (ret == -EEXIST) {
5232                 struct extent_map *existing;
5233
5234                 ret = 0;
5235
5236                 existing = lookup_extent_mapping(em_tree, start, len);
5237                 if (existing && (existing->start > start ||
5238                     existing->start + existing->len <= start)) {
5239                         free_extent_map(existing);
5240                         existing = NULL;
5241                 }
5242                 if (!existing) {
5243                         existing = lookup_extent_mapping(em_tree, em->start,
5244                                                          em->len);
5245                         if (existing) {
5246                                 err = merge_extent_mapping(em_tree, existing,
5247                                                            em, start,
5248                                                            root->sectorsize);
5249                                 free_extent_map(existing);
5250                                 if (err) {
5251                                         free_extent_map(em);
5252                                         em = NULL;
5253                                 }
5254                         } else {
5255                                 err = -EIO;
5256                                 free_extent_map(em);
5257                                 em = NULL;
5258                         }
5259                 } else {
5260                         free_extent_map(em);
5261                         em = existing;
5262                         err = 0;
5263                 }
5264         }
5265         write_unlock(&em_tree->lock);
5266 out:
5267
5268         trace_btrfs_get_extent(root, em);
5269
5270         if (path)
5271                 btrfs_free_path(path);
5272         if (trans) {
5273                 ret = btrfs_end_transaction(trans, root);
5274                 if (!err)
5275                         err = ret;
5276         }
5277         if (err) {
5278                 free_extent_map(em);
5279                 return ERR_PTR(err);
5280         }
5281         return em;
5282 }
5283
5284 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5285                                            size_t pg_offset, u64 start, u64 len,
5286                                            int create)
5287 {
5288         struct extent_map *em;
5289         struct extent_map *hole_em = NULL;
5290         u64 range_start = start;
5291         u64 end;
5292         u64 found;
5293         u64 found_end;
5294         int err = 0;
5295
5296         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5297         if (IS_ERR(em))
5298                 return em;
5299         if (em) {
5300                 /*
5301                  * if our em maps to a hole, there might
5302                  * actually be delalloc bytes behind it
5303                  */
5304                 if (em->block_start != EXTENT_MAP_HOLE)
5305                         return em;
5306                 else
5307                         hole_em = em;
5308         }
5309
5310         /* check to see if we've wrapped (len == -1 or similar) */
5311         end = start + len;
5312         if (end < start)
5313                 end = (u64)-1;
5314         else
5315                 end -= 1;
5316
5317         em = NULL;
5318
5319         /* ok, we didn't find anything, lets look for delalloc */
5320         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5321                                  end, len, EXTENT_DELALLOC, 1);
5322         found_end = range_start + found;
5323         if (found_end < range_start)
5324                 found_end = (u64)-1;
5325
5326         /*
5327          * we didn't find anything useful, return
5328          * the original results from get_extent()
5329          */
5330         if (range_start > end || found_end <= start) {
5331                 em = hole_em;
5332                 hole_em = NULL;
5333                 goto out;
5334         }
5335
5336         /* adjust the range_start to make sure it doesn't
5337          * go backwards from the start they passed in
5338          */
5339         range_start = max(start,range_start);
5340         found = found_end - range_start;
5341
5342         if (found > 0) {
5343                 u64 hole_start = start;
5344                 u64 hole_len = len;
5345
5346                 em = alloc_extent_map(GFP_NOFS);
5347                 if (!em) {
5348                         err = -ENOMEM;
5349                         goto out;
5350                 }
5351                 /*
5352                  * when btrfs_get_extent can't find anything it
5353                  * returns one huge hole
5354                  *
5355                  * make sure what it found really fits our range, and
5356                  * adjust to make sure it is based on the start from
5357                  * the caller
5358                  */
5359                 if (hole_em) {
5360                         u64 calc_end = extent_map_end(hole_em);
5361
5362                         if (calc_end <= start || (hole_em->start > end)) {
5363                                 free_extent_map(hole_em);
5364                                 hole_em = NULL;
5365                         } else {
5366                                 hole_start = max(hole_em->start, start);
5367                                 hole_len = calc_end - hole_start;
5368                         }
5369                 }
5370                 em->bdev = NULL;
5371                 if (hole_em && range_start > hole_start) {
5372                         /* our hole starts before our delalloc, so we
5373                          * have to return just the parts of the hole
5374                          * that go until  the delalloc starts
5375                          */
5376                         em->len = min(hole_len,
5377                                       range_start - hole_start);
5378                         em->start = hole_start;
5379                         em->orig_start = hole_start;
5380                         /*
5381                          * don't adjust block start at all,
5382                          * it is fixed at EXTENT_MAP_HOLE
5383                          */
5384                         em->block_start = hole_em->block_start;
5385                         em->block_len = hole_len;
5386                 } else {
5387                         em->start = range_start;
5388                         em->len = found;
5389                         em->orig_start = range_start;
5390                         em->block_start = EXTENT_MAP_DELALLOC;
5391                         em->block_len = found;
5392                 }
5393         } else if (hole_em) {
5394                 return hole_em;
5395         }
5396 out:
5397
5398         free_extent_map(hole_em);
5399         if (err) {
5400                 free_extent_map(em);
5401                 return ERR_PTR(err);
5402         }
5403         return em;
5404 }
5405
5406 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5407                                                   struct extent_map *em,
5408                                                   u64 start, u64 len)
5409 {
5410         struct btrfs_root *root = BTRFS_I(inode)->root;
5411         struct btrfs_trans_handle *trans;
5412         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5413         struct btrfs_key ins;
5414         u64 alloc_hint;
5415         int ret;
5416         bool insert = false;
5417
5418         /*
5419          * Ok if the extent map we looked up is a hole and is for the exact
5420          * range we want, there is no reason to allocate a new one, however if
5421          * it is not right then we need to free this one and drop the cache for
5422          * our range.
5423          */
5424         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5425             em->len != len) {
5426                 free_extent_map(em);
5427                 em = NULL;
5428                 insert = true;
5429                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5430         }
5431
5432         trans = btrfs_join_transaction(root);
5433         if (IS_ERR(trans))
5434                 return ERR_CAST(trans);
5435
5436         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5437
5438         alloc_hint = get_extent_allocation_hint(inode, start, len);
5439         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5440                                    alloc_hint, (u64)-1, &ins, 1);
5441         if (ret) {
5442                 em = ERR_PTR(ret);
5443                 goto out;
5444         }
5445
5446         if (!em) {
5447                 em = alloc_extent_map(GFP_NOFS);
5448                 if (!em) {
5449                         em = ERR_PTR(-ENOMEM);
5450                         goto out;
5451                 }
5452         }
5453
5454         em->start = start;
5455         em->orig_start = em->start;
5456         em->len = ins.offset;
5457
5458         em->block_start = ins.objectid;
5459         em->block_len = ins.offset;
5460         em->bdev = root->fs_info->fs_devices->latest_bdev;
5461
5462         /*
5463          * We need to do this because if we're using the original em we searched
5464          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5465          */
5466         em->flags = 0;
5467         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5468
5469         while (insert) {
5470                 write_lock(&em_tree->lock);
5471                 ret = add_extent_mapping(em_tree, em);
5472                 write_unlock(&em_tree->lock);
5473                 if (ret != -EEXIST)
5474                         break;
5475                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5476         }
5477
5478         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5479                                            ins.offset, ins.offset, 0);
5480         if (ret) {
5481                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5482                 em = ERR_PTR(ret);
5483         }
5484 out:
5485         btrfs_end_transaction(trans, root);
5486         return em;
5487 }
5488
5489 /*
5490  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5491  * block must be cow'd
5492  */
5493 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5494                                       struct inode *inode, u64 offset, u64 len)
5495 {
5496         struct btrfs_path *path;
5497         int ret;
5498         struct extent_buffer *leaf;
5499         struct btrfs_root *root = BTRFS_I(inode)->root;
5500         struct btrfs_file_extent_item *fi;
5501         struct btrfs_key key;
5502         u64 disk_bytenr;
5503         u64 backref_offset;
5504         u64 extent_end;
5505         u64 num_bytes;
5506         int slot;
5507         int found_type;
5508
5509         path = btrfs_alloc_path();
5510         if (!path)
5511                 return -ENOMEM;
5512
5513         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5514                                        offset, 0);
5515         if (ret < 0)
5516                 goto out;
5517
5518         slot = path->slots[0];
5519         if (ret == 1) {
5520                 if (slot == 0) {
5521                         /* can't find the item, must cow */
5522                         ret = 0;
5523                         goto out;
5524                 }
5525                 slot--;
5526         }
5527         ret = 0;
5528         leaf = path->nodes[0];
5529         btrfs_item_key_to_cpu(leaf, &key, slot);
5530         if (key.objectid != inode->i_ino ||
5531             key.type != BTRFS_EXTENT_DATA_KEY) {
5532                 /* not our file or wrong item type, must cow */
5533                 goto out;
5534         }
5535
5536         if (key.offset > offset) {
5537                 /* Wrong offset, must cow */
5538                 goto out;
5539         }
5540
5541         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5542         found_type = btrfs_file_extent_type(leaf, fi);
5543         if (found_type != BTRFS_FILE_EXTENT_REG &&
5544             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5545                 /* not a regular extent, must cow */
5546                 goto out;
5547         }
5548         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5549         backref_offset = btrfs_file_extent_offset(leaf, fi);
5550
5551         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5552         if (extent_end < offset + len) {
5553                 /* extent doesn't include our full range, must cow */
5554                 goto out;
5555         }
5556
5557         if (btrfs_extent_readonly(root, disk_bytenr))
5558                 goto out;
5559
5560         /*
5561          * look for other files referencing this extent, if we
5562          * find any we must cow
5563          */
5564         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5565                                   key.offset - backref_offset, disk_bytenr))
5566                 goto out;
5567
5568         /*
5569          * adjust disk_bytenr and num_bytes to cover just the bytes
5570          * in this extent we are about to write.  If there
5571          * are any csums in that range we have to cow in order
5572          * to keep the csums correct
5573          */
5574         disk_bytenr += backref_offset;
5575         disk_bytenr += offset - key.offset;
5576         num_bytes = min(offset + len, extent_end) - offset;
5577         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5578                                 goto out;
5579         /*
5580          * all of the above have passed, it is safe to overwrite this extent
5581          * without cow
5582          */
5583         ret = 1;
5584 out:
5585         btrfs_free_path(path);
5586         return ret;
5587 }
5588
5589 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5590                                    struct buffer_head *bh_result, int create)
5591 {
5592         struct extent_map *em;
5593         struct btrfs_root *root = BTRFS_I(inode)->root;
5594         u64 start = iblock << inode->i_blkbits;
5595         u64 len = bh_result->b_size;
5596         struct btrfs_trans_handle *trans;
5597
5598         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5599         if (IS_ERR(em))
5600                 return PTR_ERR(em);
5601
5602         /*
5603          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5604          * io.  INLINE is special, and we could probably kludge it in here, but
5605          * it's still buffered so for safety lets just fall back to the generic
5606          * buffered path.
5607          *
5608          * For COMPRESSED we _have_ to read the entire extent in so we can
5609          * decompress it, so there will be buffering required no matter what we
5610          * do, so go ahead and fallback to buffered.
5611          *
5612          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5613          * to buffered IO.  Don't blame me, this is the price we pay for using
5614          * the generic code.
5615          */
5616         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5617             em->block_start == EXTENT_MAP_INLINE) {
5618                 free_extent_map(em);
5619                 return -ENOTBLK;
5620         }
5621
5622         /* Just a good old fashioned hole, return */
5623         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5624                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5625                 free_extent_map(em);
5626                 /* DIO will do one hole at a time, so just unlock a sector */
5627                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5628                               start + root->sectorsize - 1, GFP_NOFS);
5629                 return 0;
5630         }
5631
5632         /*
5633          * We don't allocate a new extent in the following cases
5634          *
5635          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5636          * existing extent.
5637          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5638          * just use the extent.
5639          *
5640          */
5641         if (!create) {
5642                 len = em->len - (start - em->start);
5643                 goto map;
5644         }
5645
5646         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5647             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5648              em->block_start != EXTENT_MAP_HOLE)) {
5649                 int type;
5650                 int ret;
5651                 u64 block_start;
5652
5653                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5654                         type = BTRFS_ORDERED_PREALLOC;
5655                 else
5656                         type = BTRFS_ORDERED_NOCOW;
5657                 len = min(len, em->len - (start - em->start));
5658                 block_start = em->block_start + (start - em->start);
5659
5660                 /*
5661                  * we're not going to log anything, but we do need
5662                  * to make sure the current transaction stays open
5663                  * while we look for nocow cross refs
5664                  */
5665                 trans = btrfs_join_transaction(root);
5666                 if (IS_ERR(trans))
5667                         goto must_cow;
5668
5669                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5670                         ret = btrfs_add_ordered_extent_dio(inode, start,
5671                                            block_start, len, len, type);
5672                         btrfs_end_transaction(trans, root);
5673                         if (ret) {
5674                                 free_extent_map(em);
5675                                 return ret;
5676                         }
5677                         goto unlock;
5678                 }
5679                 btrfs_end_transaction(trans, root);
5680         }
5681 must_cow:
5682         /*
5683          * this will cow the extent, reset the len in case we changed
5684          * it above
5685          */
5686         len = bh_result->b_size;
5687         em = btrfs_new_extent_direct(inode, em, start, len);
5688         if (IS_ERR(em))
5689                 return PTR_ERR(em);
5690         len = min(len, em->len - (start - em->start));
5691 unlock:
5692         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5693                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5694                           0, NULL, GFP_NOFS);
5695 map:
5696         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5697                 inode->i_blkbits;
5698         bh_result->b_size = len;
5699         bh_result->b_bdev = em->bdev;
5700         set_buffer_mapped(bh_result);
5701         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5702                 set_buffer_new(bh_result);
5703
5704         free_extent_map(em);
5705
5706         return 0;
5707 }
5708
5709 struct btrfs_dio_private {
5710         struct inode *inode;
5711         u64 logical_offset;
5712         u64 disk_bytenr;
5713         u64 bytes;
5714         u32 *csums;
5715         void *private;
5716
5717         /* number of bios pending for this dio */
5718         atomic_t pending_bios;
5719
5720         /* IO errors */
5721         int errors;
5722
5723         struct bio *orig_bio;
5724 };
5725
5726 static void btrfs_endio_direct_read(struct bio *bio, int err)
5727 {
5728         struct btrfs_dio_private *dip = bio->bi_private;
5729         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5730         struct bio_vec *bvec = bio->bi_io_vec;
5731         struct inode *inode = dip->inode;
5732         struct btrfs_root *root = BTRFS_I(inode)->root;
5733         u64 start;
5734         u32 *private = dip->csums;
5735
5736         start = dip->logical_offset;
5737         do {
5738                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5739                         struct page *page = bvec->bv_page;
5740                         char *kaddr;
5741                         u32 csum = ~(u32)0;
5742                         unsigned long flags;
5743
5744                         local_irq_save(flags);
5745                         kaddr = kmap_atomic(page, KM_IRQ0);
5746                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5747                                                csum, bvec->bv_len);
5748                         btrfs_csum_final(csum, (char *)&csum);
5749                         kunmap_atomic(kaddr, KM_IRQ0);
5750                         local_irq_restore(flags);
5751
5752                         flush_dcache_page(bvec->bv_page);
5753                         if (csum != *private) {
5754                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5755                                       " %llu csum %u private %u\n",
5756                                       inode->i_ino, (unsigned long long)start,
5757                                       csum, *private);
5758                                 err = -EIO;
5759                         }
5760                 }
5761
5762                 start += bvec->bv_len;
5763                 private++;
5764                 bvec++;
5765         } while (bvec <= bvec_end);
5766
5767         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5768                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5769         bio->bi_private = dip->private;
5770
5771         kfree(dip->csums);
5772         kfree(dip);
5773
5774         /* If we had a csum failure make sure to clear the uptodate flag */
5775         if (err)
5776                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5777         dio_end_io(bio, err);
5778 }
5779
5780 static void btrfs_endio_direct_write(struct bio *bio, int err)
5781 {
5782         struct btrfs_dio_private *dip = bio->bi_private;
5783         struct inode *inode = dip->inode;
5784         struct btrfs_root *root = BTRFS_I(inode)->root;
5785         struct btrfs_trans_handle *trans;
5786         struct btrfs_ordered_extent *ordered = NULL;
5787         struct extent_state *cached_state = NULL;
5788         u64 ordered_offset = dip->logical_offset;
5789         u64 ordered_bytes = dip->bytes;
5790         int ret;
5791
5792         if (err)
5793                 goto out_done;
5794 again:
5795         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5796                                                    &ordered_offset,
5797                                                    ordered_bytes);
5798         if (!ret)
5799                 goto out_test;
5800
5801         BUG_ON(!ordered);
5802
5803         trans = btrfs_join_transaction(root);
5804         if (IS_ERR(trans)) {
5805                 err = -ENOMEM;
5806                 goto out;
5807         }
5808         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5809
5810         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5811                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5812                 if (!ret)
5813                         ret = btrfs_update_inode(trans, root, inode);
5814                 err = ret;
5815                 goto out;
5816         }
5817
5818         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5819                          ordered->file_offset + ordered->len - 1, 0,
5820                          &cached_state, GFP_NOFS);
5821
5822         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5823                 ret = btrfs_mark_extent_written(trans, inode,
5824                                                 ordered->file_offset,
5825                                                 ordered->file_offset +
5826                                                 ordered->len);
5827                 if (ret) {
5828                         err = ret;
5829                         goto out_unlock;
5830                 }
5831         } else {
5832                 ret = insert_reserved_file_extent(trans, inode,
5833                                                   ordered->file_offset,
5834                                                   ordered->start,
5835                                                   ordered->disk_len,
5836                                                   ordered->len,
5837                                                   ordered->len,
5838                                                   0, 0, 0,
5839                                                   BTRFS_FILE_EXTENT_REG);
5840                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5841                                    ordered->file_offset, ordered->len);
5842                 if (ret) {
5843                         err = ret;
5844                         WARN_ON(1);
5845                         goto out_unlock;
5846                 }
5847         }
5848
5849         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5850         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5851         if (!ret)
5852                 btrfs_update_inode(trans, root, inode);
5853         ret = 0;
5854 out_unlock:
5855         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5856                              ordered->file_offset + ordered->len - 1,
5857                              &cached_state, GFP_NOFS);
5858 out:
5859         btrfs_delalloc_release_metadata(inode, ordered->len);
5860         btrfs_end_transaction(trans, root);
5861         ordered_offset = ordered->file_offset + ordered->len;
5862         btrfs_put_ordered_extent(ordered);
5863         btrfs_put_ordered_extent(ordered);
5864
5865 out_test:
5866         /*
5867          * our bio might span multiple ordered extents.  If we haven't
5868          * completed the accounting for the whole dio, go back and try again
5869          */
5870         if (ordered_offset < dip->logical_offset + dip->bytes) {
5871                 ordered_bytes = dip->logical_offset + dip->bytes -
5872                         ordered_offset;
5873                 goto again;
5874         }
5875 out_done:
5876         bio->bi_private = dip->private;
5877
5878         kfree(dip->csums);
5879         kfree(dip);
5880
5881         /* If we had an error make sure to clear the uptodate flag */
5882         if (err)
5883                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5884         dio_end_io(bio, err);
5885 }
5886
5887 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5888                                     struct bio *bio, int mirror_num,
5889                                     unsigned long bio_flags, u64 offset)
5890 {
5891         int ret;
5892         struct btrfs_root *root = BTRFS_I(inode)->root;
5893         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5894         BUG_ON(ret);
5895         return 0;
5896 }
5897
5898 static void btrfs_end_dio_bio(struct bio *bio, int err)
5899 {
5900         struct btrfs_dio_private *dip = bio->bi_private;
5901
5902         if (err) {
5903                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5904                       "sector %#Lx len %u err no %d\n",
5905                       dip->inode->i_ino, bio->bi_rw,
5906                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5907                 dip->errors = 1;
5908
5909                 /*
5910                  * before atomic variable goto zero, we must make sure
5911                  * dip->errors is perceived to be set.
5912                  */
5913                 smp_mb__before_atomic_dec();
5914         }
5915
5916         /* if there are more bios still pending for this dio, just exit */
5917         if (!atomic_dec_and_test(&dip->pending_bios))
5918                 goto out;
5919
5920         if (dip->errors)
5921                 bio_io_error(dip->orig_bio);
5922         else {
5923                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5924                 bio_endio(dip->orig_bio, 0);
5925         }
5926 out:
5927         bio_put(bio);
5928 }
5929
5930 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5931                                        u64 first_sector, gfp_t gfp_flags)
5932 {
5933         int nr_vecs = bio_get_nr_vecs(bdev);
5934         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5935 }
5936
5937 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5938                                          int rw, u64 file_offset, int skip_sum,
5939                                          u32 *csums, int async_submit)
5940 {
5941         int write = rw & REQ_WRITE;
5942         struct btrfs_root *root = BTRFS_I(inode)->root;
5943         int ret;
5944
5945         bio_get(bio);
5946         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5947         if (ret)
5948                 goto err;
5949
5950         if (skip_sum)
5951                 goto map;
5952
5953         if (write && async_submit) {
5954                 ret = btrfs_wq_submit_bio(root->fs_info,
5955                                    inode, rw, bio, 0, 0,
5956                                    file_offset,
5957                                    __btrfs_submit_bio_start_direct_io,
5958                                    __btrfs_submit_bio_done);
5959                 goto err;
5960         } else if (write) {
5961                 /*
5962                  * If we aren't doing async submit, calculate the csum of the
5963                  * bio now.
5964                  */
5965                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5966                 if (ret)
5967                         goto err;
5968         } else if (!skip_sum) {
5969                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5970                                           file_offset, csums);
5971                 if (ret)
5972                         goto err;
5973         }
5974
5975 map:
5976         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5977 err:
5978         bio_put(bio);
5979         return ret;
5980 }
5981
5982 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5983                                     int skip_sum)
5984 {
5985         struct inode *inode = dip->inode;
5986         struct btrfs_root *root = BTRFS_I(inode)->root;
5987         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5988         struct bio *bio;
5989         struct bio *orig_bio = dip->orig_bio;
5990         struct bio_vec *bvec = orig_bio->bi_io_vec;
5991         u64 start_sector = orig_bio->bi_sector;
5992         u64 file_offset = dip->logical_offset;
5993         u64 submit_len = 0;
5994         u64 map_length;
5995         int nr_pages = 0;
5996         u32 *csums = dip->csums;
5997         int ret = 0;
5998         int async_submit = 0;
5999         int write = rw & REQ_WRITE;
6000
6001         map_length = orig_bio->bi_size;
6002         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6003                               &map_length, NULL, 0);
6004         if (ret) {
6005                 bio_put(orig_bio);
6006                 return -EIO;
6007         }
6008
6009         if (map_length >= orig_bio->bi_size) {
6010                 bio = orig_bio;
6011                 goto submit;
6012         }
6013
6014         async_submit = 1;
6015         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6016         if (!bio)
6017                 return -ENOMEM;
6018         bio->bi_private = dip;
6019         bio->bi_end_io = btrfs_end_dio_bio;
6020         atomic_inc(&dip->pending_bios);
6021
6022         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6023                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6024                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6025                                  bvec->bv_offset) < bvec->bv_len)) {
6026                         /*
6027                          * inc the count before we submit the bio so
6028                          * we know the end IO handler won't happen before
6029                          * we inc the count. Otherwise, the dip might get freed
6030                          * before we're done setting it up
6031                          */
6032                         atomic_inc(&dip->pending_bios);
6033                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6034                                                      file_offset, skip_sum,
6035                                                      csums, async_submit);
6036                         if (ret) {
6037                                 bio_put(bio);
6038                                 atomic_dec(&dip->pending_bios);
6039                                 goto out_err;
6040                         }
6041
6042                         /* Write's use the ordered csums */
6043                         if (!write && !skip_sum)
6044                                 csums = csums + nr_pages;
6045                         start_sector += submit_len >> 9;
6046                         file_offset += submit_len;
6047
6048                         submit_len = 0;
6049                         nr_pages = 0;
6050
6051                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6052                                                   start_sector, GFP_NOFS);
6053                         if (!bio)
6054                                 goto out_err;
6055                         bio->bi_private = dip;
6056                         bio->bi_end_io = btrfs_end_dio_bio;
6057
6058                         map_length = orig_bio->bi_size;
6059                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6060                                               &map_length, NULL, 0);
6061                         if (ret) {
6062                                 bio_put(bio);
6063                                 goto out_err;
6064                         }
6065                 } else {
6066                         submit_len += bvec->bv_len;
6067                         nr_pages ++;
6068                         bvec++;
6069                 }
6070         }
6071
6072 submit:
6073         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6074                                      csums, async_submit);
6075         if (!ret)
6076                 return 0;
6077
6078         bio_put(bio);
6079 out_err:
6080         dip->errors = 1;
6081         /*
6082          * before atomic variable goto zero, we must
6083          * make sure dip->errors is perceived to be set.
6084          */
6085         smp_mb__before_atomic_dec();
6086         if (atomic_dec_and_test(&dip->pending_bios))
6087                 bio_io_error(dip->orig_bio);
6088
6089         /* bio_end_io() will handle error, so we needn't return it */
6090         return 0;
6091 }
6092
6093 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6094                                 loff_t file_offset)
6095 {
6096         struct btrfs_root *root = BTRFS_I(inode)->root;
6097         struct btrfs_dio_private *dip;
6098         struct bio_vec *bvec = bio->bi_io_vec;
6099         int skip_sum;
6100         int write = rw & REQ_WRITE;
6101         int ret = 0;
6102
6103         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6104
6105         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6106         if (!dip) {
6107                 ret = -ENOMEM;
6108                 goto free_ordered;
6109         }
6110         dip->csums = NULL;
6111
6112         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6113         if (!write && !skip_sum) {
6114                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6115                 if (!dip->csums) {
6116                         kfree(dip);
6117                         ret = -ENOMEM;
6118                         goto free_ordered;
6119                 }
6120         }
6121
6122         dip->private = bio->bi_private;
6123         dip->inode = inode;
6124         dip->logical_offset = file_offset;
6125
6126         dip->bytes = 0;
6127         do {
6128                 dip->bytes += bvec->bv_len;
6129                 bvec++;
6130         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6131
6132         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6133         bio->bi_private = dip;
6134         dip->errors = 0;
6135         dip->orig_bio = bio;
6136         atomic_set(&dip->pending_bios, 0);
6137
6138         if (write)
6139                 bio->bi_end_io = btrfs_endio_direct_write;
6140         else
6141                 bio->bi_end_io = btrfs_endio_direct_read;
6142
6143         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6144         if (!ret)
6145                 return;
6146 free_ordered:
6147         /*
6148          * If this is a write, we need to clean up the reserved space and kill
6149          * the ordered extent.
6150          */
6151         if (write) {
6152                 struct btrfs_ordered_extent *ordered;
6153                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6154                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6155                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6156                         btrfs_free_reserved_extent(root, ordered->start,
6157                                                    ordered->disk_len);
6158                 btrfs_put_ordered_extent(ordered);
6159                 btrfs_put_ordered_extent(ordered);
6160         }
6161         bio_endio(bio, ret);
6162 }
6163
6164 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6165                         const struct iovec *iov, loff_t offset,
6166                         unsigned long nr_segs)
6167 {
6168         int seg;
6169         int i;
6170         size_t size;
6171         unsigned long addr;
6172         unsigned blocksize_mask = root->sectorsize - 1;
6173         ssize_t retval = -EINVAL;
6174         loff_t end = offset;
6175
6176         if (offset & blocksize_mask)
6177                 goto out;
6178
6179         /* Check the memory alignment.  Blocks cannot straddle pages */
6180         for (seg = 0; seg < nr_segs; seg++) {
6181                 addr = (unsigned long)iov[seg].iov_base;
6182                 size = iov[seg].iov_len;
6183                 end += size;
6184                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6185                         goto out;
6186
6187                 /* If this is a write we don't need to check anymore */
6188                 if (rw & WRITE)
6189                         continue;
6190
6191                 /*
6192                  * Check to make sure we don't have duplicate iov_base's in this
6193                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6194                  * when reading back.
6195                  */
6196                 for (i = seg + 1; i < nr_segs; i++) {
6197                         if (iov[seg].iov_base == iov[i].iov_base)
6198                                 goto out;
6199                 }
6200         }
6201         retval = 0;
6202 out:
6203         return retval;
6204 }
6205 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6206                         const struct iovec *iov, loff_t offset,
6207                         unsigned long nr_segs)
6208 {
6209         struct file *file = iocb->ki_filp;
6210         struct inode *inode = file->f_mapping->host;
6211         struct btrfs_ordered_extent *ordered;
6212         struct extent_state *cached_state = NULL;
6213         u64 lockstart, lockend;
6214         ssize_t ret;
6215         int writing = rw & WRITE;
6216         int write_bits = 0;
6217         size_t count = iov_length(iov, nr_segs);
6218
6219         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6220                             offset, nr_segs)) {
6221                 return 0;
6222         }
6223
6224         lockstart = offset;
6225         lockend = offset + count - 1;
6226
6227         if (writing) {
6228                 ret = btrfs_delalloc_reserve_space(inode, count);
6229                 if (ret)
6230                         goto out;
6231         }
6232
6233         while (1) {
6234                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6235                                  0, &cached_state, GFP_NOFS);
6236                 /*
6237                  * We're concerned with the entire range that we're going to be
6238                  * doing DIO to, so we need to make sure theres no ordered
6239                  * extents in this range.
6240                  */
6241                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6242                                                      lockend - lockstart + 1);
6243                 if (!ordered)
6244                         break;
6245                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6246                                      &cached_state, GFP_NOFS);
6247                 btrfs_start_ordered_extent(inode, ordered, 1);
6248                 btrfs_put_ordered_extent(ordered);
6249                 cond_resched();
6250         }
6251
6252         /*
6253          * we don't use btrfs_set_extent_delalloc because we don't want
6254          * the dirty or uptodate bits
6255          */
6256         if (writing) {
6257                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6258                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6259                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6260                                      GFP_NOFS);
6261                 if (ret) {
6262                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6263                                          lockend, EXTENT_LOCKED | write_bits,
6264                                          1, 0, &cached_state, GFP_NOFS);
6265                         goto out;
6266                 }
6267         }
6268
6269         free_extent_state(cached_state);
6270         cached_state = NULL;
6271
6272         ret = __blockdev_direct_IO(rw, iocb, inode,
6273                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6274                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6275                    btrfs_submit_direct, 0);
6276
6277         if (ret < 0 && ret != -EIOCBQUEUED) {
6278                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6279                               offset + iov_length(iov, nr_segs) - 1,
6280                               EXTENT_LOCKED | write_bits, 1, 0,
6281                               &cached_state, GFP_NOFS);
6282         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6283                 /*
6284                  * We're falling back to buffered, unlock the section we didn't
6285                  * do IO on.
6286                  */
6287                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6288                               offset + iov_length(iov, nr_segs) - 1,
6289                               EXTENT_LOCKED | write_bits, 1, 0,
6290                               &cached_state, GFP_NOFS);
6291         }
6292 out:
6293         free_extent_state(cached_state);
6294         return ret;
6295 }
6296
6297 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6298                 __u64 start, __u64 len)
6299 {
6300         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6301 }
6302
6303 int btrfs_readpage(struct file *file, struct page *page)
6304 {
6305         struct extent_io_tree *tree;
6306         tree = &BTRFS_I(page->mapping->host)->io_tree;
6307         return extent_read_full_page(tree, page, btrfs_get_extent);
6308 }
6309
6310 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6311 {
6312         struct extent_io_tree *tree;
6313
6314
6315         if (current->flags & PF_MEMALLOC) {
6316                 redirty_page_for_writepage(wbc, page);
6317                 unlock_page(page);
6318                 return 0;
6319         }
6320         tree = &BTRFS_I(page->mapping->host)->io_tree;
6321         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6322 }
6323
6324 int btrfs_writepages(struct address_space *mapping,
6325                      struct writeback_control *wbc)
6326 {
6327         struct extent_io_tree *tree;
6328
6329         tree = &BTRFS_I(mapping->host)->io_tree;
6330         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6331 }
6332
6333 static int
6334 btrfs_readpages(struct file *file, struct address_space *mapping,
6335                 struct list_head *pages, unsigned nr_pages)
6336 {
6337         struct extent_io_tree *tree;
6338         tree = &BTRFS_I(mapping->host)->io_tree;
6339         return extent_readpages(tree, mapping, pages, nr_pages,
6340                                 btrfs_get_extent);
6341 }
6342 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6343 {
6344         struct extent_io_tree *tree;
6345         struct extent_map_tree *map;
6346         int ret;
6347
6348         tree = &BTRFS_I(page->mapping->host)->io_tree;
6349         map = &BTRFS_I(page->mapping->host)->extent_tree;
6350         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6351         if (ret == 1) {
6352                 ClearPagePrivate(page);
6353                 set_page_private(page, 0);
6354                 page_cache_release(page);
6355         }
6356         return ret;
6357 }
6358
6359 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6360 {
6361         if (PageWriteback(page) || PageDirty(page))
6362                 return 0;
6363         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6364 }
6365
6366 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6367 {
6368         struct extent_io_tree *tree;
6369         struct btrfs_ordered_extent *ordered;
6370         struct extent_state *cached_state = NULL;
6371         u64 page_start = page_offset(page);
6372         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6373
6374
6375         /*
6376          * we have the page locked, so new writeback can't start,
6377          * and the dirty bit won't be cleared while we are here.
6378          *
6379          * Wait for IO on this page so that we can safely clear
6380          * the PagePrivate2 bit and do ordered accounting
6381          */
6382         wait_on_page_writeback(page);
6383
6384         tree = &BTRFS_I(page->mapping->host)->io_tree;
6385         if (offset) {
6386                 btrfs_releasepage(page, GFP_NOFS);
6387                 return;
6388         }
6389         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6390                          GFP_NOFS);
6391         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6392                                            page_offset(page));
6393         if (ordered) {
6394                 /*
6395                  * IO on this page will never be started, so we need
6396                  * to account for any ordered extents now
6397                  */
6398                 clear_extent_bit(tree, page_start, page_end,
6399                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6400                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6401                                  &cached_state, GFP_NOFS);
6402                 /*
6403                  * whoever cleared the private bit is responsible
6404                  * for the finish_ordered_io
6405                  */
6406                 if (TestClearPagePrivate2(page)) {
6407                         btrfs_finish_ordered_io(page->mapping->host,
6408                                                 page_start, page_end);
6409                 }
6410                 btrfs_put_ordered_extent(ordered);
6411                 cached_state = NULL;
6412                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6413                                  GFP_NOFS);
6414         }
6415         clear_extent_bit(tree, page_start, page_end,
6416                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6417                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6418         __btrfs_releasepage(page, GFP_NOFS);
6419
6420         ClearPageChecked(page);
6421         if (PagePrivate(page)) {
6422                 ClearPagePrivate(page);
6423                 set_page_private(page, 0);
6424                 page_cache_release(page);
6425         }
6426 }
6427
6428 /*
6429  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6430  * called from a page fault handler when a page is first dirtied. Hence we must
6431  * be careful to check for EOF conditions here. We set the page up correctly
6432  * for a written page which means we get ENOSPC checking when writing into
6433  * holes and correct delalloc and unwritten extent mapping on filesystems that
6434  * support these features.
6435  *
6436  * We are not allowed to take the i_mutex here so we have to play games to
6437  * protect against truncate races as the page could now be beyond EOF.  Because
6438  * vmtruncate() writes the inode size before removing pages, once we have the
6439  * page lock we can determine safely if the page is beyond EOF. If it is not
6440  * beyond EOF, then the page is guaranteed safe against truncation until we
6441  * unlock the page.
6442  */
6443 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6444 {
6445         struct page *page = vmf->page;
6446         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6447         struct btrfs_root *root = BTRFS_I(inode)->root;
6448         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6449         struct btrfs_ordered_extent *ordered;
6450         struct extent_state *cached_state = NULL;
6451         char *kaddr;
6452         unsigned long zero_start;
6453         loff_t size;
6454         int ret;
6455         u64 page_start;
6456         u64 page_end;
6457
6458         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6459         if (ret) {
6460                 if (ret == -ENOMEM)
6461                         ret = VM_FAULT_OOM;
6462                 else /* -ENOSPC, -EIO, etc */
6463                         ret = VM_FAULT_SIGBUS;
6464                 goto out;
6465         }
6466
6467         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6468 again:
6469         lock_page(page);
6470         size = i_size_read(inode);
6471         page_start = page_offset(page);
6472         page_end = page_start + PAGE_CACHE_SIZE - 1;
6473
6474         if ((page->mapping != inode->i_mapping) ||
6475             (page_start >= size)) {
6476                 /* page got truncated out from underneath us */
6477                 goto out_unlock;
6478         }
6479         wait_on_page_writeback(page);
6480
6481         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6482                          GFP_NOFS);
6483         set_page_extent_mapped(page);
6484
6485         /*
6486          * we can't set the delalloc bits if there are pending ordered
6487          * extents.  Drop our locks and wait for them to finish
6488          */
6489         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6490         if (ordered) {
6491                 unlock_extent_cached(io_tree, page_start, page_end,
6492                                      &cached_state, GFP_NOFS);
6493                 unlock_page(page);
6494                 btrfs_start_ordered_extent(inode, ordered, 1);
6495                 btrfs_put_ordered_extent(ordered);
6496                 goto again;
6497         }
6498
6499         /*
6500          * XXX - page_mkwrite gets called every time the page is dirtied, even
6501          * if it was already dirty, so for space accounting reasons we need to
6502          * clear any delalloc bits for the range we are fixing to save.  There
6503          * is probably a better way to do this, but for now keep consistent with
6504          * prepare_pages in the normal write path.
6505          */
6506         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6507                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6508                           0, 0, &cached_state, GFP_NOFS);
6509
6510         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6511                                         &cached_state);
6512         if (ret) {
6513                 unlock_extent_cached(io_tree, page_start, page_end,
6514                                      &cached_state, GFP_NOFS);
6515                 ret = VM_FAULT_SIGBUS;
6516                 goto out_unlock;
6517         }
6518         ret = 0;
6519
6520         /* page is wholly or partially inside EOF */
6521         if (page_start + PAGE_CACHE_SIZE > size)
6522                 zero_start = size & ~PAGE_CACHE_MASK;
6523         else
6524                 zero_start = PAGE_CACHE_SIZE;
6525
6526         if (zero_start != PAGE_CACHE_SIZE) {
6527                 kaddr = kmap(page);
6528                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6529                 flush_dcache_page(page);
6530                 kunmap(page);
6531         }
6532         ClearPageChecked(page);
6533         set_page_dirty(page);
6534         SetPageUptodate(page);
6535
6536         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6537         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6538
6539         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6540
6541 out_unlock:
6542         if (!ret)
6543                 return VM_FAULT_LOCKED;
6544         unlock_page(page);
6545         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6546 out:
6547         return ret;
6548 }
6549
6550 static int btrfs_truncate(struct inode *inode)
6551 {
6552         struct btrfs_root *root = BTRFS_I(inode)->root;
6553         struct btrfs_block_rsv *rsv;
6554         int ret;
6555         int err = 0;
6556         struct btrfs_trans_handle *trans;
6557         unsigned long nr;
6558         u64 mask = root->sectorsize - 1;
6559
6560         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6561         if (ret)
6562                 return ret;
6563
6564         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6565         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6566
6567         /*
6568          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6569          * 3 things going on here
6570          *
6571          * 1) We need to reserve space for our orphan item and the space to
6572          * delete our orphan item.  Lord knows we don't want to have a dangling
6573          * orphan item because we didn't reserve space to remove it.
6574          *
6575          * 2) We need to reserve space to update our inode.
6576          *
6577          * 3) We need to have something to cache all the space that is going to
6578          * be free'd up by the truncate operation, but also have some slack
6579          * space reserved in case it uses space during the truncate (thank you
6580          * very much snapshotting).
6581          *
6582          * And we need these to all be seperate.  The fact is we can use alot of
6583          * space doing the truncate, and we have no earthly idea how much space
6584          * we will use, so we need the truncate reservation to be seperate so it
6585          * doesn't end up using space reserved for updating the inode or
6586          * removing the orphan item.  We also need to be able to stop the
6587          * transaction and start a new one, which means we need to be able to
6588          * update the inode several times, and we have no idea of knowing how
6589          * many times that will be, so we can't just reserve 1 item for the
6590          * entirety of the opration, so that has to be done seperately as well.
6591          * Then there is the orphan item, which does indeed need to be held on
6592          * to for the whole operation, and we need nobody to touch this reserved
6593          * space except the orphan code.
6594          *
6595          * So that leaves us with
6596          *
6597          * 1) root->orphan_block_rsv - for the orphan deletion.
6598          * 2) rsv - for the truncate reservation, which we will steal from the
6599          * transaction reservation.
6600          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6601          * updating the inode.
6602          */
6603         rsv = btrfs_alloc_block_rsv(root);
6604         if (!rsv)
6605                 return -ENOMEM;
6606         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6607
6608         trans = btrfs_start_transaction(root, 4);
6609         if (IS_ERR(trans)) {
6610                 err = PTR_ERR(trans);
6611                 goto out;
6612         }
6613
6614         /*
6615          * Reserve space for the truncate process.  Truncate should be adding
6616          * space, but if there are snapshots it may end up using space.
6617          */
6618         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6619         BUG_ON(ret);
6620
6621         ret = btrfs_orphan_add(trans, inode);
6622         if (ret) {
6623                 btrfs_end_transaction(trans, root);
6624                 goto out;
6625         }
6626
6627         nr = trans->blocks_used;
6628         btrfs_end_transaction(trans, root);
6629         btrfs_btree_balance_dirty(root, nr);
6630
6631         /*
6632          * Ok so we've already migrated our bytes over for the truncate, so here
6633          * just reserve the one slot we need for updating the inode.
6634          */
6635         trans = btrfs_start_transaction(root, 1);
6636         if (IS_ERR(trans)) {
6637                 err = PTR_ERR(trans);
6638                 goto out;
6639         }
6640         trans->block_rsv = rsv;
6641
6642         /*
6643          * setattr is responsible for setting the ordered_data_close flag,
6644          * but that is only tested during the last file release.  That
6645          * could happen well after the next commit, leaving a great big
6646          * window where new writes may get lost if someone chooses to write
6647          * to this file after truncating to zero
6648          *
6649          * The inode doesn't have any dirty data here, and so if we commit
6650          * this is a noop.  If someone immediately starts writing to the inode
6651          * it is very likely we'll catch some of their writes in this
6652          * transaction, and the commit will find this file on the ordered
6653          * data list with good things to send down.
6654          *
6655          * This is a best effort solution, there is still a window where
6656          * using truncate to replace the contents of the file will
6657          * end up with a zero length file after a crash.
6658          */
6659         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6660                 btrfs_add_ordered_operation(trans, root, inode);
6661
6662         while (1) {
6663                 if (!trans) {
6664                         trans = btrfs_start_transaction(root, 3);
6665                         if (IS_ERR(trans)) {
6666                                 err = PTR_ERR(trans);
6667                                 goto out;
6668                         }
6669
6670                         ret = btrfs_truncate_reserve_metadata(trans, root,
6671                                                               rsv);
6672                         BUG_ON(ret);
6673
6674                         trans->block_rsv = rsv;
6675                 }
6676
6677                 ret = btrfs_truncate_inode_items(trans, root, inode,
6678                                                  inode->i_size,
6679                                                  BTRFS_EXTENT_DATA_KEY);
6680                 if (ret != -EAGAIN) {
6681                         err = ret;
6682                         break;
6683                 }
6684
6685                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6686                 ret = btrfs_update_inode(trans, root, inode);
6687                 if (ret) {
6688                         err = ret;
6689                         break;
6690                 }
6691
6692                 nr = trans->blocks_used;
6693                 btrfs_end_transaction(trans, root);
6694                 trans = NULL;
6695                 btrfs_btree_balance_dirty(root, nr);
6696         }
6697
6698         if (ret == 0 && inode->i_nlink > 0) {
6699                 trans->block_rsv = root->orphan_block_rsv;
6700                 ret = btrfs_orphan_del(trans, inode);
6701                 if (ret)
6702                         err = ret;
6703         } else if (ret && inode->i_nlink > 0) {
6704                 /*
6705                  * Failed to do the truncate, remove us from the in memory
6706                  * orphan list.
6707                  */
6708                 ret = btrfs_orphan_del(NULL, inode);
6709         }
6710
6711         trans->block_rsv = &root->fs_info->trans_block_rsv;
6712         ret = btrfs_update_inode(trans, root, inode);
6713         if (ret && !err)
6714                 err = ret;
6715
6716         nr = trans->blocks_used;
6717         ret = btrfs_end_transaction_throttle(trans, root);
6718         btrfs_btree_balance_dirty(root, nr);
6719
6720 out:
6721         btrfs_free_block_rsv(root, rsv);
6722
6723         if (ret && !err)
6724                 err = ret;
6725
6726         return err;
6727 }
6728
6729 /*
6730  * create a new subvolume directory/inode (helper for the ioctl).
6731  */
6732 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6733                              struct btrfs_root *new_root, u64 new_dirid)
6734 {
6735         struct inode *inode;
6736         int err;
6737         u64 index = 0;
6738
6739         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6740                                 new_dirid, S_IFDIR | 0700, &index);
6741         if (IS_ERR(inode))
6742                 return PTR_ERR(inode);
6743         inode->i_op = &btrfs_dir_inode_operations;
6744         inode->i_fop = &btrfs_dir_file_operations;
6745
6746         inode->i_nlink = 1;
6747         btrfs_i_size_write(inode, 0);
6748
6749         err = btrfs_update_inode(trans, new_root, inode);
6750         BUG_ON(err);
6751
6752         iput(inode);
6753         return 0;
6754 }
6755
6756 /* helper function for file defrag and space balancing.  This
6757  * forces readahead on a given range of bytes in an inode
6758  */
6759 unsigned long btrfs_force_ra(struct address_space *mapping,
6760                               struct file_ra_state *ra, struct file *file,
6761                               pgoff_t offset, pgoff_t last_index)
6762 {
6763         pgoff_t req_size = last_index - offset + 1;
6764
6765         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6766         return offset + req_size;
6767 }
6768
6769 struct inode *btrfs_alloc_inode(struct super_block *sb)
6770 {
6771         struct btrfs_inode *ei;
6772         struct inode *inode;
6773
6774         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6775         if (!ei)
6776                 return NULL;
6777
6778         ei->root = NULL;
6779         ei->space_info = NULL;
6780         ei->generation = 0;
6781         ei->sequence = 0;
6782         ei->last_trans = 0;
6783         ei->last_sub_trans = 0;
6784         ei->logged_trans = 0;
6785         ei->delalloc_bytes = 0;
6786         ei->reserved_bytes = 0;
6787         ei->disk_i_size = 0;
6788         ei->flags = 0;
6789         ei->index_cnt = (u64)-1;
6790         ei->last_unlink_trans = 0;
6791
6792         atomic_set(&ei->outstanding_extents, 0);
6793         atomic_set(&ei->reserved_extents, 0);
6794
6795         ei->ordered_data_close = 0;
6796         ei->orphan_meta_reserved = 0;
6797         ei->dummy_inode = 0;
6798         ei->force_compress = BTRFS_COMPRESS_NONE;
6799
6800         inode = &ei->vfs_inode;
6801         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6802         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6803         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6804         mutex_init(&ei->log_mutex);
6805         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6806         INIT_LIST_HEAD(&ei->i_orphan);
6807         INIT_LIST_HEAD(&ei->delalloc_inodes);
6808         INIT_LIST_HEAD(&ei->ordered_operations);
6809         RB_CLEAR_NODE(&ei->rb_node);
6810
6811         return inode;
6812 }
6813
6814 static void btrfs_i_callback(struct rcu_head *head)
6815 {
6816         struct inode *inode = container_of(head, struct inode, i_rcu);
6817         INIT_LIST_HEAD(&inode->i_dentry);
6818         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6819 }
6820
6821 void btrfs_destroy_inode(struct inode *inode)
6822 {
6823         struct btrfs_ordered_extent *ordered;
6824         struct btrfs_root *root = BTRFS_I(inode)->root;
6825
6826         WARN_ON(!list_empty(&inode->i_dentry));
6827         WARN_ON(inode->i_data.nrpages);
6828         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6829         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6830
6831         /*
6832          * This can happen where we create an inode, but somebody else also
6833          * created the same inode and we need to destroy the one we already
6834          * created.
6835          */
6836         if (!root)
6837                 goto free;
6838
6839         /*
6840          * Make sure we're properly removed from the ordered operation
6841          * lists.
6842          */
6843         smp_mb();
6844         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6845                 spin_lock(&root->fs_info->ordered_extent_lock);
6846                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6847                 spin_unlock(&root->fs_info->ordered_extent_lock);
6848         }
6849
6850         spin_lock(&root->orphan_lock);
6851         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6852                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6853                        inode->i_ino);
6854                 list_del_init(&BTRFS_I(inode)->i_orphan);
6855         }
6856         spin_unlock(&root->orphan_lock);
6857
6858         while (1) {
6859                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6860                 if (!ordered)
6861                         break;
6862                 else {
6863                         printk(KERN_ERR "btrfs found ordered "
6864                                "extent %llu %llu on inode cleanup\n",
6865                                (unsigned long long)ordered->file_offset,
6866                                (unsigned long long)ordered->len);
6867                         btrfs_remove_ordered_extent(inode, ordered);
6868                         btrfs_put_ordered_extent(ordered);
6869                         btrfs_put_ordered_extent(ordered);
6870                 }
6871         }
6872         inode_tree_del(inode);
6873         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6874 free:
6875         call_rcu(&inode->i_rcu, btrfs_i_callback);
6876 }
6877
6878 int btrfs_drop_inode(struct inode *inode)
6879 {
6880         struct btrfs_root *root = BTRFS_I(inode)->root;
6881
6882         if (btrfs_root_refs(&root->root_item) == 0 &&
6883             root != root->fs_info->tree_root)
6884                 return 1;
6885         else
6886                 return generic_drop_inode(inode);
6887 }
6888
6889 static void init_once(void *foo)
6890 {
6891         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6892
6893         inode_init_once(&ei->vfs_inode);
6894 }
6895
6896 void btrfs_destroy_cachep(void)
6897 {
6898         if (btrfs_inode_cachep)
6899                 kmem_cache_destroy(btrfs_inode_cachep);
6900         if (btrfs_trans_handle_cachep)
6901                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6902         if (btrfs_transaction_cachep)
6903                 kmem_cache_destroy(btrfs_transaction_cachep);
6904         if (btrfs_path_cachep)
6905                 kmem_cache_destroy(btrfs_path_cachep);
6906         if (btrfs_free_space_cachep)
6907                 kmem_cache_destroy(btrfs_free_space_cachep);
6908 }
6909
6910 int btrfs_init_cachep(void)
6911 {
6912         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6913                         sizeof(struct btrfs_inode), 0,
6914                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6915         if (!btrfs_inode_cachep)
6916                 goto fail;
6917
6918         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6919                         sizeof(struct btrfs_trans_handle), 0,
6920                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6921         if (!btrfs_trans_handle_cachep)
6922                 goto fail;
6923
6924         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6925                         sizeof(struct btrfs_transaction), 0,
6926                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6927         if (!btrfs_transaction_cachep)
6928                 goto fail;
6929
6930         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6931                         sizeof(struct btrfs_path), 0,
6932                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6933         if (!btrfs_path_cachep)
6934                 goto fail;
6935
6936         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6937                         sizeof(struct btrfs_free_space), 0,
6938                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6939         if (!btrfs_free_space_cachep)
6940                 goto fail;
6941
6942         return 0;
6943 fail:
6944         btrfs_destroy_cachep();
6945         return -ENOMEM;
6946 }
6947
6948 static int btrfs_getattr(struct vfsmount *mnt,
6949                          struct dentry *dentry, struct kstat *stat)
6950 {
6951         struct inode *inode = dentry->d_inode;
6952         generic_fillattr(inode, stat);
6953         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6954         stat->blksize = PAGE_CACHE_SIZE;
6955         stat->blocks = (inode_get_bytes(inode) +
6956                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6957         return 0;
6958 }
6959
6960 /*
6961  * If a file is moved, it will inherit the cow and compression flags of the new
6962  * directory.
6963  */
6964 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6965 {
6966         struct btrfs_inode *b_dir = BTRFS_I(dir);
6967         struct btrfs_inode *b_inode = BTRFS_I(inode);
6968
6969         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6970                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6971         else
6972                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6973
6974         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6975                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6976         else
6977                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6978 }
6979
6980 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6981                            struct inode *new_dir, struct dentry *new_dentry)
6982 {
6983         struct btrfs_trans_handle *trans;
6984         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6985         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6986         struct inode *new_inode = new_dentry->d_inode;
6987         struct inode *old_inode = old_dentry->d_inode;
6988         struct timespec ctime = CURRENT_TIME;
6989         u64 index = 0;
6990         u64 root_objectid;
6991         int ret;
6992
6993         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6994                 return -EPERM;
6995
6996         /* we only allow rename subvolume link between subvolumes */
6997         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6998                 return -EXDEV;
6999
7000         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7001             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7002                 return -ENOTEMPTY;
7003
7004         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7005             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7006                 return -ENOTEMPTY;
7007         /*
7008          * we're using rename to replace one file with another.
7009          * and the replacement file is large.  Start IO on it now so
7010          * we don't add too much work to the end of the transaction
7011          */
7012         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7013             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7014                 filemap_flush(old_inode->i_mapping);
7015
7016         /* close the racy window with snapshot create/destroy ioctl */
7017         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7018                 down_read(&root->fs_info->subvol_sem);
7019         /*
7020          * We want to reserve the absolute worst case amount of items.  So if
7021          * both inodes are subvols and we need to unlink them then that would
7022          * require 4 item modifications, but if they are both normal inodes it
7023          * would require 5 item modifications, so we'll assume their normal
7024          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7025          * should cover the worst case number of items we'll modify.
7026          */
7027         trans = btrfs_start_transaction(root, 20);
7028         if (IS_ERR(trans)) {
7029                 ret = PTR_ERR(trans);
7030                 goto out_notrans;
7031         }
7032
7033         if (dest != root)
7034                 btrfs_record_root_in_trans(trans, dest);
7035
7036         ret = btrfs_set_inode_index(new_dir, &index);
7037         if (ret)
7038                 goto out_fail;
7039
7040         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7041                 /* force full log commit if subvolume involved. */
7042                 root->fs_info->last_trans_log_full_commit = trans->transid;
7043         } else {
7044                 ret = btrfs_insert_inode_ref(trans, dest,
7045                                              new_dentry->d_name.name,
7046                                              new_dentry->d_name.len,
7047                                              old_inode->i_ino,
7048                                              new_dir->i_ino, index);
7049                 if (ret)
7050                         goto out_fail;
7051                 /*
7052                  * this is an ugly little race, but the rename is required
7053                  * to make sure that if we crash, the inode is either at the
7054                  * old name or the new one.  pinning the log transaction lets
7055                  * us make sure we don't allow a log commit to come in after
7056                  * we unlink the name but before we add the new name back in.
7057                  */
7058                 btrfs_pin_log_trans(root);
7059         }
7060         /*
7061          * make sure the inode gets flushed if it is replacing
7062          * something.
7063          */
7064         if (new_inode && new_inode->i_size &&
7065             old_inode && S_ISREG(old_inode->i_mode)) {
7066                 btrfs_add_ordered_operation(trans, root, old_inode);
7067         }
7068
7069         old_dir->i_ctime = old_dir->i_mtime = ctime;
7070         new_dir->i_ctime = new_dir->i_mtime = ctime;
7071         old_inode->i_ctime = ctime;
7072
7073         if (old_dentry->d_parent != new_dentry->d_parent)
7074                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7075
7076         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7077                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7078                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7079                                         old_dentry->d_name.name,
7080                                         old_dentry->d_name.len);
7081         } else {
7082                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7083                                         old_dentry->d_inode,
7084                                         old_dentry->d_name.name,
7085                                         old_dentry->d_name.len);
7086                 if (!ret)
7087                         ret = btrfs_update_inode(trans, root, old_inode);
7088         }
7089         BUG_ON(ret);
7090
7091         if (new_inode) {
7092                 new_inode->i_ctime = CURRENT_TIME;
7093                 if (unlikely(new_inode->i_ino ==
7094                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7095                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7096                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7097                                                 root_objectid,
7098                                                 new_dentry->d_name.name,
7099                                                 new_dentry->d_name.len);
7100                         BUG_ON(new_inode->i_nlink == 0);
7101                 } else {
7102                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7103                                                  new_dentry->d_inode,
7104                                                  new_dentry->d_name.name,
7105                                                  new_dentry->d_name.len);
7106                 }
7107                 BUG_ON(ret);
7108                 if (new_inode->i_nlink == 0) {
7109                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7110                         BUG_ON(ret);
7111                 }
7112         }
7113
7114         fixup_inode_flags(new_dir, old_inode);
7115
7116         ret = btrfs_add_link(trans, new_dir, old_inode,
7117                              new_dentry->d_name.name,
7118                              new_dentry->d_name.len, 0, index);
7119         BUG_ON(ret);
7120
7121         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7122                 struct dentry *parent = dget_parent(new_dentry);
7123                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7124                 dput(parent);
7125                 btrfs_end_log_trans(root);
7126         }
7127 out_fail:
7128         btrfs_end_transaction_throttle(trans, root);
7129 out_notrans:
7130         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7131                 up_read(&root->fs_info->subvol_sem);
7132
7133         return ret;
7134 }
7135
7136 /*
7137  * some fairly slow code that needs optimization. This walks the list
7138  * of all the inodes with pending delalloc and forces them to disk.
7139  */
7140 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7141 {
7142         struct list_head *head = &root->fs_info->delalloc_inodes;
7143         struct btrfs_inode *binode;
7144         struct inode *inode;
7145
7146         if (root->fs_info->sb->s_flags & MS_RDONLY)
7147                 return -EROFS;
7148
7149         spin_lock(&root->fs_info->delalloc_lock);
7150         while (!list_empty(head)) {
7151                 binode = list_entry(head->next, struct btrfs_inode,
7152                                     delalloc_inodes);
7153                 inode = igrab(&binode->vfs_inode);
7154                 if (!inode)
7155                         list_del_init(&binode->delalloc_inodes);
7156                 spin_unlock(&root->fs_info->delalloc_lock);
7157                 if (inode) {
7158                         filemap_flush(inode->i_mapping);
7159                         if (delay_iput)
7160                                 btrfs_add_delayed_iput(inode);
7161                         else
7162                                 iput(inode);
7163                 }
7164                 cond_resched();
7165                 spin_lock(&root->fs_info->delalloc_lock);
7166         }
7167         spin_unlock(&root->fs_info->delalloc_lock);
7168
7169         /* the filemap_flush will queue IO into the worker threads, but
7170          * we have to make sure the IO is actually started and that
7171          * ordered extents get created before we return
7172          */
7173         atomic_inc(&root->fs_info->async_submit_draining);
7174         while (atomic_read(&root->fs_info->nr_async_submits) ||
7175               atomic_read(&root->fs_info->async_delalloc_pages)) {
7176                 wait_event(root->fs_info->async_submit_wait,
7177                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7178                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7179         }
7180         atomic_dec(&root->fs_info->async_submit_draining);
7181         return 0;
7182 }
7183
7184 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7185                                    int sync)
7186 {
7187         struct btrfs_inode *binode;
7188         struct inode *inode = NULL;
7189
7190         spin_lock(&root->fs_info->delalloc_lock);
7191         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7192                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7193                                     struct btrfs_inode, delalloc_inodes);
7194                 inode = igrab(&binode->vfs_inode);
7195                 if (inode) {
7196                         list_move_tail(&binode->delalloc_inodes,
7197                                        &root->fs_info->delalloc_inodes);
7198                         break;
7199                 }
7200
7201                 list_del_init(&binode->delalloc_inodes);
7202                 cond_resched_lock(&root->fs_info->delalloc_lock);
7203         }
7204         spin_unlock(&root->fs_info->delalloc_lock);
7205
7206         if (inode) {
7207                 if (sync) {
7208                         filemap_write_and_wait(inode->i_mapping);
7209                         /*
7210                          * We have to do this because compression doesn't
7211                          * actually set PG_writeback until it submits the pages
7212                          * for IO, which happens in an async thread, so we could
7213                          * race and not actually wait for any writeback pages
7214                          * because they've not been submitted yet.  Technically
7215                          * this could still be the case for the ordered stuff
7216                          * since the async thread may not have started to do its
7217                          * work yet.  If this becomes the case then we need to
7218                          * figure out a way to make sure that in writepage we
7219                          * wait for any async pages to be submitted before
7220                          * returning so that fdatawait does what its supposed to
7221                          * do.
7222                          */
7223                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7224                 } else {
7225                         filemap_flush(inode->i_mapping);
7226                 }
7227                 if (delay_iput)
7228                         btrfs_add_delayed_iput(inode);
7229                 else
7230                         iput(inode);
7231                 return 1;
7232         }
7233         return 0;
7234 }
7235
7236 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7237                          const char *symname)
7238 {
7239         struct btrfs_trans_handle *trans;
7240         struct btrfs_root *root = BTRFS_I(dir)->root;
7241         struct btrfs_path *path;
7242         struct btrfs_key key;
7243         struct inode *inode = NULL;
7244         int err;
7245         int drop_inode = 0;
7246         u64 objectid;
7247         u64 index = 0 ;
7248         int name_len;
7249         int datasize;
7250         unsigned long ptr;
7251         struct btrfs_file_extent_item *ei;
7252         struct extent_buffer *leaf;
7253         unsigned long nr = 0;
7254
7255         name_len = strlen(symname) + 1;
7256         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7257                 return -ENAMETOOLONG;
7258
7259         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7260         if (err)
7261                 return err;
7262         /*
7263          * 2 items for inode item and ref
7264          * 2 items for dir items
7265          * 1 item for xattr if selinux is on
7266          */
7267         trans = btrfs_start_transaction(root, 5);
7268         if (IS_ERR(trans))
7269                 return PTR_ERR(trans);
7270
7271         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7272                                 dentry->d_name.len, dir->i_ino, objectid,
7273                                 S_IFLNK|S_IRWXUGO, &index);
7274         if (IS_ERR(inode)) {
7275                 err = PTR_ERR(inode);
7276                 goto out_unlock;
7277         }
7278
7279         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7280         if (err) {
7281                 drop_inode = 1;
7282                 goto out_unlock;
7283         }
7284
7285         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7286         if (err)
7287                 drop_inode = 1;
7288         else {
7289                 inode->i_mapping->a_ops = &btrfs_aops;
7290                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7291                 inode->i_fop = &btrfs_file_operations;
7292                 inode->i_op = &btrfs_file_inode_operations;
7293                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7294         }
7295         if (drop_inode)
7296                 goto out_unlock;
7297
7298         path = btrfs_alloc_path();
7299         BUG_ON(!path);
7300         key.objectid = inode->i_ino;
7301         key.offset = 0;
7302         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7303         datasize = btrfs_file_extent_calc_inline_size(name_len);
7304         err = btrfs_insert_empty_item(trans, root, path, &key,
7305                                       datasize);
7306         if (err) {
7307                 drop_inode = 1;
7308                 goto out_unlock;
7309         }
7310         leaf = path->nodes[0];
7311         ei = btrfs_item_ptr(leaf, path->slots[0],
7312                             struct btrfs_file_extent_item);
7313         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7314         btrfs_set_file_extent_type(leaf, ei,
7315                                    BTRFS_FILE_EXTENT_INLINE);
7316         btrfs_set_file_extent_encryption(leaf, ei, 0);
7317         btrfs_set_file_extent_compression(leaf, ei, 0);
7318         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7319         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7320
7321         ptr = btrfs_file_extent_inline_start(ei);
7322         write_extent_buffer(leaf, symname, ptr, name_len);
7323         btrfs_mark_buffer_dirty(leaf);
7324         btrfs_free_path(path);
7325
7326         inode->i_op = &btrfs_symlink_inode_operations;
7327         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7328         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7329         inode_set_bytes(inode, name_len);
7330         btrfs_i_size_write(inode, name_len - 1);
7331         err = btrfs_update_inode(trans, root, inode);
7332         if (err)
7333                 drop_inode = 1;
7334
7335 out_unlock:
7336         nr = trans->blocks_used;
7337         btrfs_end_transaction_throttle(trans, root);
7338         if (drop_inode) {
7339                 inode_dec_link_count(inode);
7340                 iput(inode);
7341         }
7342         btrfs_btree_balance_dirty(root, nr);
7343         return err;
7344 }
7345
7346 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7347                                        u64 start, u64 num_bytes, u64 min_size,
7348                                        loff_t actual_len, u64 *alloc_hint,
7349                                        struct btrfs_trans_handle *trans)
7350 {
7351         struct btrfs_root *root = BTRFS_I(inode)->root;
7352         struct btrfs_key ins;
7353         u64 cur_offset = start;
7354         u64 i_size;
7355         int ret = 0;
7356         bool own_trans = true;
7357
7358         if (trans)
7359                 own_trans = false;
7360         while (num_bytes > 0) {
7361                 if (own_trans) {
7362                         trans = btrfs_start_transaction(root, 3);
7363                         if (IS_ERR(trans)) {
7364                                 ret = PTR_ERR(trans);
7365                                 break;
7366                         }
7367                 }
7368
7369                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7370                                            0, *alloc_hint, (u64)-1, &ins, 1);
7371                 if (ret) {
7372                         if (own_trans)
7373                                 btrfs_end_transaction(trans, root);
7374                         break;
7375                 }
7376
7377                 ret = insert_reserved_file_extent(trans, inode,
7378                                                   cur_offset, ins.objectid,
7379                                                   ins.offset, ins.offset,
7380                                                   ins.offset, 0, 0, 0,
7381                                                   BTRFS_FILE_EXTENT_PREALLOC);
7382                 BUG_ON(ret);
7383                 btrfs_drop_extent_cache(inode, cur_offset,
7384                                         cur_offset + ins.offset -1, 0);
7385
7386                 num_bytes -= ins.offset;
7387                 cur_offset += ins.offset;
7388                 *alloc_hint = ins.objectid + ins.offset;
7389
7390                 inode->i_ctime = CURRENT_TIME;
7391                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7392                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7393                     (actual_len > inode->i_size) &&
7394                     (cur_offset > inode->i_size)) {
7395                         if (cur_offset > actual_len)
7396                                 i_size = actual_len;
7397                         else
7398                                 i_size = cur_offset;
7399                         i_size_write(inode, i_size);
7400                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7401                 }
7402
7403                 ret = btrfs_update_inode(trans, root, inode);
7404                 BUG_ON(ret);
7405
7406                 if (own_trans)
7407                         btrfs_end_transaction(trans, root);
7408         }
7409         return ret;
7410 }
7411
7412 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7413                               u64 start, u64 num_bytes, u64 min_size,
7414                               loff_t actual_len, u64 *alloc_hint)
7415 {
7416         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7417                                            min_size, actual_len, alloc_hint,
7418                                            NULL);
7419 }
7420
7421 int btrfs_prealloc_file_range_trans(struct inode *inode,
7422                                     struct btrfs_trans_handle *trans, int mode,
7423                                     u64 start, u64 num_bytes, u64 min_size,
7424                                     loff_t actual_len, u64 *alloc_hint)
7425 {
7426         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7427                                            min_size, actual_len, alloc_hint, trans);
7428 }
7429
7430 static int btrfs_set_page_dirty(struct page *page)
7431 {
7432         return __set_page_dirty_nobuffers(page);
7433 }
7434
7435 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7436 {
7437         struct btrfs_root *root = BTRFS_I(inode)->root;
7438
7439         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7440                 return -EROFS;
7441         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7442                 return -EACCES;
7443         return generic_permission(inode, mask, flags, btrfs_check_acl);
7444 }
7445
7446 static const struct inode_operations btrfs_dir_inode_operations = {
7447         .getattr        = btrfs_getattr,
7448         .lookup         = btrfs_lookup,
7449         .create         = btrfs_create,
7450         .unlink         = btrfs_unlink,
7451         .link           = btrfs_link,
7452         .mkdir          = btrfs_mkdir,
7453         .rmdir          = btrfs_rmdir,
7454         .rename         = btrfs_rename,
7455         .symlink        = btrfs_symlink,
7456         .setattr        = btrfs_setattr,
7457         .mknod          = btrfs_mknod,
7458         .setxattr       = btrfs_setxattr,
7459         .getxattr       = btrfs_getxattr,
7460         .listxattr      = btrfs_listxattr,
7461         .removexattr    = btrfs_removexattr,
7462         .permission     = btrfs_permission,
7463 };
7464 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7465         .lookup         = btrfs_lookup,
7466         .permission     = btrfs_permission,
7467 };
7468
7469 static const struct file_operations btrfs_dir_file_operations = {
7470         .llseek         = generic_file_llseek,
7471         .read           = generic_read_dir,
7472         .readdir        = btrfs_real_readdir,
7473         .unlocked_ioctl = btrfs_ioctl,
7474 #ifdef CONFIG_COMPAT
7475         .compat_ioctl   = btrfs_ioctl,
7476 #endif
7477         .release        = btrfs_release_file,
7478         .fsync          = btrfs_sync_file,
7479 };
7480
7481 static struct extent_io_ops btrfs_extent_io_ops = {
7482         .fill_delalloc = run_delalloc_range,
7483         .submit_bio_hook = btrfs_submit_bio_hook,
7484         .merge_bio_hook = btrfs_merge_bio_hook,
7485         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7486         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7487         .writepage_start_hook = btrfs_writepage_start_hook,
7488         .readpage_io_failed_hook = btrfs_io_failed_hook,
7489         .set_bit_hook = btrfs_set_bit_hook,
7490         .clear_bit_hook = btrfs_clear_bit_hook,
7491         .merge_extent_hook = btrfs_merge_extent_hook,
7492         .split_extent_hook = btrfs_split_extent_hook,
7493 };
7494
7495 /*
7496  * btrfs doesn't support the bmap operation because swapfiles
7497  * use bmap to make a mapping of extents in the file.  They assume
7498  * these extents won't change over the life of the file and they
7499  * use the bmap result to do IO directly to the drive.
7500  *
7501  * the btrfs bmap call would return logical addresses that aren't
7502  * suitable for IO and they also will change frequently as COW
7503  * operations happen.  So, swapfile + btrfs == corruption.
7504  *
7505  * For now we're avoiding this by dropping bmap.
7506  */
7507 static const struct address_space_operations btrfs_aops = {
7508         .readpage       = btrfs_readpage,
7509         .writepage      = btrfs_writepage,
7510         .writepages     = btrfs_writepages,
7511         .readpages      = btrfs_readpages,
7512         .direct_IO      = btrfs_direct_IO,
7513         .invalidatepage = btrfs_invalidatepage,
7514         .releasepage    = btrfs_releasepage,
7515         .set_page_dirty = btrfs_set_page_dirty,
7516         .error_remove_page = generic_error_remove_page,
7517 };
7518
7519 static const struct address_space_operations btrfs_symlink_aops = {
7520         .readpage       = btrfs_readpage,
7521         .writepage      = btrfs_writepage,
7522         .invalidatepage = btrfs_invalidatepage,
7523         .releasepage    = btrfs_releasepage,
7524 };
7525
7526 static const struct inode_operations btrfs_file_inode_operations = {
7527         .getattr        = btrfs_getattr,
7528         .setattr        = btrfs_setattr,
7529         .setxattr       = btrfs_setxattr,
7530         .getxattr       = btrfs_getxattr,
7531         .listxattr      = btrfs_listxattr,
7532         .removexattr    = btrfs_removexattr,
7533         .permission     = btrfs_permission,
7534         .fiemap         = btrfs_fiemap,
7535 };
7536 static const struct inode_operations btrfs_special_inode_operations = {
7537         .getattr        = btrfs_getattr,
7538         .setattr        = btrfs_setattr,
7539         .permission     = btrfs_permission,
7540         .setxattr       = btrfs_setxattr,
7541         .getxattr       = btrfs_getxattr,
7542         .listxattr      = btrfs_listxattr,
7543         .removexattr    = btrfs_removexattr,
7544 };
7545 static const struct inode_operations btrfs_symlink_inode_operations = {
7546         .readlink       = generic_readlink,
7547         .follow_link    = page_follow_link_light,
7548         .put_link       = page_put_link,
7549         .getattr        = btrfs_getattr,
7550         .permission     = btrfs_permission,
7551         .setxattr       = btrfs_setxattr,
7552         .getxattr       = btrfs_getxattr,
7553         .listxattr      = btrfs_listxattr,
7554         .removexattr    = btrfs_removexattr,
7555 };
7556
7557 const struct dentry_operations btrfs_dentry_operations = {
7558         .d_delete       = btrfs_dentry_delete,
7559 };