ec5423790bbbec9a39bdc4528a069d3b54738456
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "ref-cache.h"
52 #include "compression.h"
53 #include "locking.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static 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_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_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 void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90                                    struct page *locked_page,
91                                    u64 start, u64 end, int *page_started,
92                                    unsigned long *nr_written, int unlock);
93
94 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
95 {
96         int err;
97
98         err = btrfs_init_acl(inode, dir);
99         if (!err)
100                 err = btrfs_xattr_security_init(inode, dir);
101         return err;
102 }
103
104 /*
105  * this does all the hard work for inserting an inline extent into
106  * the btree.  The caller should have done a btrfs_drop_extents so that
107  * no overlapping inline items exist in the btree
108  */
109 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
110                                 struct btrfs_root *root, struct inode *inode,
111                                 u64 start, size_t size, size_t compressed_size,
112                                 struct page **compressed_pages)
113 {
114         struct btrfs_key key;
115         struct btrfs_path *path;
116         struct extent_buffer *leaf;
117         struct page *page = NULL;
118         char *kaddr;
119         unsigned long ptr;
120         struct btrfs_file_extent_item *ei;
121         int err = 0;
122         int ret;
123         size_t cur_size = size;
124         size_t datasize;
125         unsigned long offset;
126         int use_compress = 0;
127
128         if (compressed_size && compressed_pages) {
129                 use_compress = 1;
130                 cur_size = compressed_size;
131         }
132
133         path = btrfs_alloc_path();
134         if (!path)
135                 return -ENOMEM;
136
137         btrfs_set_trans_block_group(trans, inode);
138
139         key.objectid = inode->i_ino;
140         key.offset = start;
141         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142         datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144         inode_add_bytes(inode, size);
145         ret = btrfs_insert_empty_item(trans, root, path, &key,
146                                       datasize);
147         BUG_ON(ret);
148         if (ret) {
149                 err = ret;
150                 goto fail;
151         }
152         leaf = path->nodes[0];
153         ei = btrfs_item_ptr(leaf, path->slots[0],
154                             struct btrfs_file_extent_item);
155         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157         btrfs_set_file_extent_encryption(leaf, ei, 0);
158         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160         ptr = btrfs_file_extent_inline_start(ei);
161
162         if (use_compress) {
163                 struct page *cpage;
164                 int i = 0;
165                 while (compressed_size > 0) {
166                         cpage = compressed_pages[i];
167                         cur_size = min_t(unsigned long, compressed_size,
168                                        PAGE_CACHE_SIZE);
169
170                         kaddr = kmap(cpage);
171                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
172                         kunmap(cpage);
173
174                         i++;
175                         ptr += cur_size;
176                         compressed_size -= cur_size;
177                 }
178                 btrfs_set_file_extent_compression(leaf, ei,
179                                                   BTRFS_COMPRESS_ZLIB);
180         } else {
181                 page = find_get_page(inode->i_mapping,
182                                      start >> PAGE_CACHE_SHIFT);
183                 btrfs_set_file_extent_compression(leaf, ei, 0);
184                 kaddr = kmap_atomic(page, KM_USER0);
185                 offset = start & (PAGE_CACHE_SIZE - 1);
186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187                 kunmap_atomic(kaddr, KM_USER0);
188                 page_cache_release(page);
189         }
190         btrfs_mark_buffer_dirty(leaf);
191         btrfs_free_path(path);
192
193         BTRFS_I(inode)->disk_i_size = inode->i_size;
194         btrfs_update_inode(trans, root, inode);
195         return 0;
196 fail:
197         btrfs_free_path(path);
198         return err;
199 }
200
201
202 /*
203  * conditionally insert an inline extent into the file.  This
204  * does the checks required to make sure the data is small enough
205  * to fit as an inline extent.
206  */
207 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
208                                  struct btrfs_root *root,
209                                  struct inode *inode, u64 start, u64 end,
210                                  size_t compressed_size,
211                                  struct page **compressed_pages)
212 {
213         u64 isize = i_size_read(inode);
214         u64 actual_end = min(end + 1, isize);
215         u64 inline_len = actual_end - start;
216         u64 aligned_end = (end + root->sectorsize - 1) &
217                         ~((u64)root->sectorsize - 1);
218         u64 hint_byte;
219         u64 data_len = inline_len;
220         int ret;
221
222         if (compressed_size)
223                 data_len = compressed_size;
224
225         if (start > 0 ||
226             actual_end >= PAGE_CACHE_SIZE ||
227             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
228             (!compressed_size &&
229             (actual_end & (root->sectorsize - 1)) == 0) ||
230             end + 1 < isize ||
231             data_len > root->fs_info->max_inline) {
232                 return 1;
233         }
234
235         ret = btrfs_drop_extents(trans, root, inode, start,
236                                  aligned_end, start, &hint_byte);
237         BUG_ON(ret);
238
239         if (isize > actual_end)
240                 inline_len = min_t(u64, isize, actual_end);
241         ret = insert_inline_extent(trans, root, inode, start,
242                                    inline_len, compressed_size,
243                                    compressed_pages);
244         BUG_ON(ret);
245         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
246         return 0;
247 }
248
249 struct async_extent {
250         u64 start;
251         u64 ram_size;
252         u64 compressed_size;
253         struct page **pages;
254         unsigned long nr_pages;
255         struct list_head list;
256 };
257
258 struct async_cow {
259         struct inode *inode;
260         struct btrfs_root *root;
261         struct page *locked_page;
262         u64 start;
263         u64 end;
264         struct list_head extents;
265         struct btrfs_work work;
266 };
267
268 static noinline int add_async_extent(struct async_cow *cow,
269                                      u64 start, u64 ram_size,
270                                      u64 compressed_size,
271                                      struct page **pages,
272                                      unsigned long nr_pages)
273 {
274         struct async_extent *async_extent;
275
276         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
277         async_extent->start = start;
278         async_extent->ram_size = ram_size;
279         async_extent->compressed_size = compressed_size;
280         async_extent->pages = pages;
281         async_extent->nr_pages = nr_pages;
282         list_add_tail(&async_extent->list, &cow->extents);
283         return 0;
284 }
285
286 /*
287  * we create compressed extents in two phases.  The first
288  * phase compresses a range of pages that have already been
289  * locked (both pages and state bits are locked).
290  *
291  * This is done inside an ordered work queue, and the compression
292  * is spread across many cpus.  The actual IO submission is step
293  * two, and the ordered work queue takes care of making sure that
294  * happens in the same order things were put onto the queue by
295  * writepages and friends.
296  *
297  * If this code finds it can't get good compression, it puts an
298  * entry onto the work queue to write the uncompressed bytes.  This
299  * makes sure that both compressed inodes and uncompressed inodes
300  * are written in the same order that pdflush sent them down.
301  */
302 static noinline int compress_file_range(struct inode *inode,
303                                         struct page *locked_page,
304                                         u64 start, u64 end,
305                                         struct async_cow *async_cow,
306                                         int *num_added)
307 {
308         struct btrfs_root *root = BTRFS_I(inode)->root;
309         struct btrfs_trans_handle *trans;
310         u64 num_bytes;
311         u64 orig_start;
312         u64 disk_num_bytes;
313         u64 blocksize = root->sectorsize;
314         u64 actual_end;
315         u64 isize = i_size_read(inode);
316         int ret = 0;
317         struct page **pages = NULL;
318         unsigned long nr_pages;
319         unsigned long nr_pages_ret = 0;
320         unsigned long total_compressed = 0;
321         unsigned long total_in = 0;
322         unsigned long max_compressed = 128 * 1024;
323         unsigned long max_uncompressed = 128 * 1024;
324         int i;
325         int will_compress;
326
327         orig_start = start;
328
329         actual_end = min_t(u64, isize, end + 1);
330 again:
331         will_compress = 0;
332         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
333         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
334
335         /*
336          * we don't want to send crud past the end of i_size through
337          * compression, that's just a waste of CPU time.  So, if the
338          * end of the file is before the start of our current
339          * requested range of bytes, we bail out to the uncompressed
340          * cleanup code that can deal with all of this.
341          *
342          * It isn't really the fastest way to fix things, but this is a
343          * very uncommon corner.
344          */
345         if (actual_end <= start)
346                 goto cleanup_and_bail_uncompressed;
347
348         total_compressed = actual_end - start;
349
350         /* we want to make sure that amount of ram required to uncompress
351          * an extent is reasonable, so we limit the total size in ram
352          * of a compressed extent to 128k.  This is a crucial number
353          * because it also controls how easily we can spread reads across
354          * cpus for decompression.
355          *
356          * We also want to make sure the amount of IO required to do
357          * a random read is reasonably small, so we limit the size of
358          * a compressed extent to 128k.
359          */
360         total_compressed = min(total_compressed, max_uncompressed);
361         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
362         num_bytes = max(blocksize,  num_bytes);
363         disk_num_bytes = num_bytes;
364         total_in = 0;
365         ret = 0;
366
367         /*
368          * we do compression for mount -o compress and when the
369          * inode has not been flagged as nocompress.  This flag can
370          * change at any time if we discover bad compression ratios.
371          */
372         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
373             btrfs_test_opt(root, COMPRESS)) {
374                 WARN_ON(pages);
375                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
376
377                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
378                                                 total_compressed, pages,
379                                                 nr_pages, &nr_pages_ret,
380                                                 &total_in,
381                                                 &total_compressed,
382                                                 max_compressed);
383
384                 if (!ret) {
385                         unsigned long offset = total_compressed &
386                                 (PAGE_CACHE_SIZE - 1);
387                         struct page *page = pages[nr_pages_ret - 1];
388                         char *kaddr;
389
390                         /* zero the tail end of the last page, we might be
391                          * sending it down to disk
392                          */
393                         if (offset) {
394                                 kaddr = kmap_atomic(page, KM_USER0);
395                                 memset(kaddr + offset, 0,
396                                        PAGE_CACHE_SIZE - offset);
397                                 kunmap_atomic(kaddr, KM_USER0);
398                         }
399                         will_compress = 1;
400                 }
401         }
402         if (start == 0) {
403                 trans = btrfs_join_transaction(root, 1);
404                 BUG_ON(!trans);
405                 btrfs_set_trans_block_group(trans, inode);
406
407                 /* lets try to make an inline extent */
408                 if (ret || total_in < (actual_end - start)) {
409                         /* we didn't compress the entire range, try
410                          * to make an uncompressed inline extent.
411                          */
412                         ret = cow_file_range_inline(trans, root, inode,
413                                                     start, end, 0, NULL);
414                 } else {
415                         /* try making a compressed inline extent */
416                         ret = cow_file_range_inline(trans, root, inode,
417                                                     start, end,
418                                                     total_compressed, pages);
419                 }
420                 btrfs_end_transaction(trans, root);
421                 if (ret == 0) {
422                         /*
423                          * inline extent creation worked, we don't need
424                          * to create any more async work items.  Unlock
425                          * and free up our temp pages.
426                          */
427                         extent_clear_unlock_delalloc(inode,
428                                                      &BTRFS_I(inode)->io_tree,
429                                                      start, end, NULL, 1, 0,
430                                                      0, 1, 1, 1);
431                         ret = 0;
432                         goto free_pages_out;
433                 }
434         }
435
436         if (will_compress) {
437                 /*
438                  * we aren't doing an inline extent round the compressed size
439                  * up to a block size boundary so the allocator does sane
440                  * things
441                  */
442                 total_compressed = (total_compressed + blocksize - 1) &
443                         ~(blocksize - 1);
444
445                 /*
446                  * one last check to make sure the compression is really a
447                  * win, compare the page count read with the blocks on disk
448                  */
449                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
450                         ~(PAGE_CACHE_SIZE - 1);
451                 if (total_compressed >= total_in) {
452                         will_compress = 0;
453                 } else {
454                         disk_num_bytes = total_compressed;
455                         num_bytes = total_in;
456                 }
457         }
458         if (!will_compress && pages) {
459                 /*
460                  * the compression code ran but failed to make things smaller,
461                  * free any pages it allocated and our page pointer array
462                  */
463                 for (i = 0; i < nr_pages_ret; i++) {
464                         WARN_ON(pages[i]->mapping);
465                         page_cache_release(pages[i]);
466                 }
467                 kfree(pages);
468                 pages = NULL;
469                 total_compressed = 0;
470                 nr_pages_ret = 0;
471
472                 /* flag the file so we don't compress in the future */
473                 btrfs_set_flag(inode, NOCOMPRESS);
474         }
475         if (will_compress) {
476                 *num_added += 1;
477
478                 /* the async work queues will take care of doing actual
479                  * allocation on disk for these compressed pages,
480                  * and will submit them to the elevator.
481                  */
482                 add_async_extent(async_cow, start, num_bytes,
483                                  total_compressed, pages, nr_pages_ret);
484
485                 if (start + num_bytes < end && start + num_bytes < actual_end) {
486                         start += num_bytes;
487                         pages = NULL;
488                         cond_resched();
489                         goto again;
490                 }
491         } else {
492 cleanup_and_bail_uncompressed:
493                 /*
494                  * No compression, but we still need to write the pages in
495                  * the file we've been given so far.  redirty the locked
496                  * page if it corresponds to our extent and set things up
497                  * for the async work queue to run cow_file_range to do
498                  * the normal delalloc dance
499                  */
500                 if (page_offset(locked_page) >= start &&
501                     page_offset(locked_page) <= end) {
502                         __set_page_dirty_nobuffers(locked_page);
503                         /* unlocked later on in the async handlers */
504                 }
505                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
506                 *num_added += 1;
507         }
508
509 out:
510         return 0;
511
512 free_pages_out:
513         for (i = 0; i < nr_pages_ret; i++) {
514                 WARN_ON(pages[i]->mapping);
515                 page_cache_release(pages[i]);
516         }
517         kfree(pages);
518
519         goto out;
520 }
521
522 /*
523  * phase two of compressed writeback.  This is the ordered portion
524  * of the code, which only gets called in the order the work was
525  * queued.  We walk all the async extents created by compress_file_range
526  * and send them down to the disk.
527  */
528 static noinline int submit_compressed_extents(struct inode *inode,
529                                               struct async_cow *async_cow)
530 {
531         struct async_extent *async_extent;
532         u64 alloc_hint = 0;
533         struct btrfs_trans_handle *trans;
534         struct btrfs_key ins;
535         struct extent_map *em;
536         struct btrfs_root *root = BTRFS_I(inode)->root;
537         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
538         struct extent_io_tree *io_tree;
539         int ret;
540
541         if (list_empty(&async_cow->extents))
542                 return 0;
543
544         trans = btrfs_join_transaction(root, 1);
545
546         while (!list_empty(&async_cow->extents)) {
547                 async_extent = list_entry(async_cow->extents.next,
548                                           struct async_extent, list);
549                 list_del(&async_extent->list);
550
551                 io_tree = &BTRFS_I(inode)->io_tree;
552
553                 /* did the compression code fall back to uncompressed IO? */
554                 if (!async_extent->pages) {
555                         int page_started = 0;
556                         unsigned long nr_written = 0;
557
558                         lock_extent(io_tree, async_extent->start,
559                                     async_extent->start +
560                                     async_extent->ram_size - 1, GFP_NOFS);
561
562                         /* allocate blocks */
563                         cow_file_range(inode, async_cow->locked_page,
564                                        async_extent->start,
565                                        async_extent->start +
566                                        async_extent->ram_size - 1,
567                                        &page_started, &nr_written, 0);
568
569                         /*
570                          * if page_started, cow_file_range inserted an
571                          * inline extent and took care of all the unlocking
572                          * and IO for us.  Otherwise, we need to submit
573                          * all those pages down to the drive.
574                          */
575                         if (!page_started)
576                                 extent_write_locked_range(io_tree,
577                                                   inode, async_extent->start,
578                                                   async_extent->start +
579                                                   async_extent->ram_size - 1,
580                                                   btrfs_get_extent,
581                                                   WB_SYNC_ALL);
582                         kfree(async_extent);
583                         cond_resched();
584                         continue;
585                 }
586
587                 lock_extent(io_tree, async_extent->start,
588                             async_extent->start + async_extent->ram_size - 1,
589                             GFP_NOFS);
590                 /*
591                  * here we're doing allocation and writeback of the
592                  * compressed pages
593                  */
594                 btrfs_drop_extent_cache(inode, async_extent->start,
595                                         async_extent->start +
596                                         async_extent->ram_size - 1, 0);
597
598                 ret = btrfs_reserve_extent(trans, root,
599                                            async_extent->compressed_size,
600                                            async_extent->compressed_size,
601                                            0, alloc_hint,
602                                            (u64)-1, &ins, 1);
603                 BUG_ON(ret);
604                 em = alloc_extent_map(GFP_NOFS);
605                 em->start = async_extent->start;
606                 em->len = async_extent->ram_size;
607                 em->orig_start = em->start;
608
609                 em->block_start = ins.objectid;
610                 em->block_len = ins.offset;
611                 em->bdev = root->fs_info->fs_devices->latest_bdev;
612                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
613                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
614
615                 while (1) {
616                         spin_lock(&em_tree->lock);
617                         ret = add_extent_mapping(em_tree, em);
618                         spin_unlock(&em_tree->lock);
619                         if (ret != -EEXIST) {
620                                 free_extent_map(em);
621                                 break;
622                         }
623                         btrfs_drop_extent_cache(inode, async_extent->start,
624                                                 async_extent->start +
625                                                 async_extent->ram_size - 1, 0);
626                 }
627
628                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
629                                                ins.objectid,
630                                                async_extent->ram_size,
631                                                ins.offset,
632                                                BTRFS_ORDERED_COMPRESSED);
633                 BUG_ON(ret);
634
635                 btrfs_end_transaction(trans, root);
636
637                 /*
638                  * clear dirty, set writeback and unlock the pages.
639                  */
640                 extent_clear_unlock_delalloc(inode,
641                                              &BTRFS_I(inode)->io_tree,
642                                              async_extent->start,
643                                              async_extent->start +
644                                              async_extent->ram_size - 1,
645                                              NULL, 1, 1, 0, 1, 1, 0);
646
647                 ret = btrfs_submit_compressed_write(inode,
648                                     async_extent->start,
649                                     async_extent->ram_size,
650                                     ins.objectid,
651                                     ins.offset, async_extent->pages,
652                                     async_extent->nr_pages);
653
654                 BUG_ON(ret);
655                 trans = btrfs_join_transaction(root, 1);
656                 alloc_hint = ins.objectid + ins.offset;
657                 kfree(async_extent);
658                 cond_resched();
659         }
660
661         btrfs_end_transaction(trans, root);
662         return 0;
663 }
664
665 /*
666  * when extent_io.c finds a delayed allocation range in the file,
667  * the call backs end up in this code.  The basic idea is to
668  * allocate extents on disk for the range, and create ordered data structs
669  * in ram to track those extents.
670  *
671  * locked_page is the page that writepage had locked already.  We use
672  * it to make sure we don't do extra locks or unlocks.
673  *
674  * *page_started is set to one if we unlock locked_page and do everything
675  * required to start IO on it.  It may be clean and already done with
676  * IO when we return.
677  */
678 static noinline int cow_file_range(struct inode *inode,
679                                    struct page *locked_page,
680                                    u64 start, u64 end, int *page_started,
681                                    unsigned long *nr_written,
682                                    int unlock)
683 {
684         struct btrfs_root *root = BTRFS_I(inode)->root;
685         struct btrfs_trans_handle *trans;
686         u64 alloc_hint = 0;
687         u64 num_bytes;
688         unsigned long ram_size;
689         u64 disk_num_bytes;
690         u64 cur_alloc_size;
691         u64 blocksize = root->sectorsize;
692         u64 actual_end;
693         u64 isize = i_size_read(inode);
694         struct btrfs_key ins;
695         struct extent_map *em;
696         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
697         int ret = 0;
698
699         trans = btrfs_join_transaction(root, 1);
700         BUG_ON(!trans);
701         btrfs_set_trans_block_group(trans, inode);
702
703         actual_end = min_t(u64, isize, end + 1);
704
705         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
706         num_bytes = max(blocksize,  num_bytes);
707         disk_num_bytes = num_bytes;
708         ret = 0;
709
710         if (start == 0) {
711                 /* lets try to make an inline extent */
712                 ret = cow_file_range_inline(trans, root, inode,
713                                             start, end, 0, NULL);
714                 if (ret == 0) {
715                         extent_clear_unlock_delalloc(inode,
716                                                      &BTRFS_I(inode)->io_tree,
717                                                      start, end, NULL, 1, 1,
718                                                      1, 1, 1, 1);
719                         *nr_written = *nr_written +
720                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
721                         *page_started = 1;
722                         ret = 0;
723                         goto out;
724                 }
725         }
726
727         BUG_ON(disk_num_bytes >
728                btrfs_super_total_bytes(&root->fs_info->super_copy));
729
730         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
731
732         while (disk_num_bytes > 0) {
733                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
734                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
735                                            root->sectorsize, 0, alloc_hint,
736                                            (u64)-1, &ins, 1);
737                 BUG_ON(ret);
738
739                 em = alloc_extent_map(GFP_NOFS);
740                 em->start = start;
741                 em->orig_start = em->start;
742
743                 ram_size = ins.offset;
744                 em->len = ins.offset;
745
746                 em->block_start = ins.objectid;
747                 em->block_len = ins.offset;
748                 em->bdev = root->fs_info->fs_devices->latest_bdev;
749                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
750
751                 while (1) {
752                         spin_lock(&em_tree->lock);
753                         ret = add_extent_mapping(em_tree, em);
754                         spin_unlock(&em_tree->lock);
755                         if (ret != -EEXIST) {
756                                 free_extent_map(em);
757                                 break;
758                         }
759                         btrfs_drop_extent_cache(inode, start,
760                                                 start + ram_size - 1, 0);
761                 }
762
763                 cur_alloc_size = ins.offset;
764                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
765                                                ram_size, cur_alloc_size, 0);
766                 BUG_ON(ret);
767
768                 if (root->root_key.objectid ==
769                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
770                         ret = btrfs_reloc_clone_csums(inode, start,
771                                                       cur_alloc_size);
772                         BUG_ON(ret);
773                 }
774
775                 if (disk_num_bytes < cur_alloc_size)
776                         break;
777
778                 /* we're not doing compressed IO, don't unlock the first
779                  * page (which the caller expects to stay locked), don't
780                  * clear any dirty bits and don't set any writeback bits
781                  */
782                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
783                                              start, start + ram_size - 1,
784                                              locked_page, unlock, 1,
785                                              1, 0, 0, 0);
786                 disk_num_bytes -= cur_alloc_size;
787                 num_bytes -= cur_alloc_size;
788                 alloc_hint = ins.objectid + ins.offset;
789                 start += cur_alloc_size;
790         }
791 out:
792         ret = 0;
793         btrfs_end_transaction(trans, root);
794
795         return ret;
796 }
797
798 /*
799  * work queue call back to started compression on a file and pages
800  */
801 static noinline void async_cow_start(struct btrfs_work *work)
802 {
803         struct async_cow *async_cow;
804         int num_added = 0;
805         async_cow = container_of(work, struct async_cow, work);
806
807         compress_file_range(async_cow->inode, async_cow->locked_page,
808                             async_cow->start, async_cow->end, async_cow,
809                             &num_added);
810         if (num_added == 0)
811                 async_cow->inode = NULL;
812 }
813
814 /*
815  * work queue call back to submit previously compressed pages
816  */
817 static noinline void async_cow_submit(struct btrfs_work *work)
818 {
819         struct async_cow *async_cow;
820         struct btrfs_root *root;
821         unsigned long nr_pages;
822
823         async_cow = container_of(work, struct async_cow, work);
824
825         root = async_cow->root;
826         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
827                 PAGE_CACHE_SHIFT;
828
829         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
830
831         if (atomic_read(&root->fs_info->async_delalloc_pages) <
832             5 * 1042 * 1024 &&
833             waitqueue_active(&root->fs_info->async_submit_wait))
834                 wake_up(&root->fs_info->async_submit_wait);
835
836         if (async_cow->inode)
837                 submit_compressed_extents(async_cow->inode, async_cow);
838 }
839
840 static noinline void async_cow_free(struct btrfs_work *work)
841 {
842         struct async_cow *async_cow;
843         async_cow = container_of(work, struct async_cow, work);
844         kfree(async_cow);
845 }
846
847 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
848                                 u64 start, u64 end, int *page_started,
849                                 unsigned long *nr_written)
850 {
851         struct async_cow *async_cow;
852         struct btrfs_root *root = BTRFS_I(inode)->root;
853         unsigned long nr_pages;
854         u64 cur_end;
855         int limit = 10 * 1024 * 1042;
856
857         if (!btrfs_test_opt(root, COMPRESS)) {
858                 return cow_file_range(inode, locked_page, start, end,
859                                       page_started, nr_written, 1);
860         }
861
862         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
863                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
864         while (start < end) {
865                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
866                 async_cow->inode = inode;
867                 async_cow->root = root;
868                 async_cow->locked_page = locked_page;
869                 async_cow->start = start;
870
871                 if (btrfs_test_flag(inode, NOCOMPRESS))
872                         cur_end = end;
873                 else
874                         cur_end = min(end, start + 512 * 1024 - 1);
875
876                 async_cow->end = cur_end;
877                 INIT_LIST_HEAD(&async_cow->extents);
878
879                 async_cow->work.func = async_cow_start;
880                 async_cow->work.ordered_func = async_cow_submit;
881                 async_cow->work.ordered_free = async_cow_free;
882                 async_cow->work.flags = 0;
883
884                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
885                         PAGE_CACHE_SHIFT;
886                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
887
888                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
889                                    &async_cow->work);
890
891                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
892                         wait_event(root->fs_info->async_submit_wait,
893                            (atomic_read(&root->fs_info->async_delalloc_pages) <
894                             limit));
895                 }
896
897                 while (atomic_read(&root->fs_info->async_submit_draining) &&
898                       atomic_read(&root->fs_info->async_delalloc_pages)) {
899                         wait_event(root->fs_info->async_submit_wait,
900                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
901                            0));
902                 }
903
904                 *nr_written += nr_pages;
905                 start = cur_end + 1;
906         }
907         *page_started = 1;
908         return 0;
909 }
910
911 static noinline int csum_exist_in_range(struct btrfs_root *root,
912                                         u64 bytenr, u64 num_bytes)
913 {
914         int ret;
915         struct btrfs_ordered_sum *sums;
916         LIST_HEAD(list);
917
918         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
919                                        bytenr + num_bytes - 1, &list);
920         if (ret == 0 && list_empty(&list))
921                 return 0;
922
923         while (!list_empty(&list)) {
924                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
925                 list_del(&sums->list);
926                 kfree(sums);
927         }
928         return 1;
929 }
930
931 /*
932  * when nowcow writeback call back.  This checks for snapshots or COW copies
933  * of the extents that exist in the file, and COWs the file as required.
934  *
935  * If no cow copies or snapshots exist, we write directly to the existing
936  * blocks on disk
937  */
938 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
939                               u64 start, u64 end, int *page_started, int force,
940                               unsigned long *nr_written)
941 {
942         struct btrfs_root *root = BTRFS_I(inode)->root;
943         struct btrfs_trans_handle *trans;
944         struct extent_buffer *leaf;
945         struct btrfs_path *path;
946         struct btrfs_file_extent_item *fi;
947         struct btrfs_key found_key;
948         u64 cow_start;
949         u64 cur_offset;
950         u64 extent_end;
951         u64 disk_bytenr;
952         u64 num_bytes;
953         int extent_type;
954         int ret;
955         int type;
956         int nocow;
957         int check_prev = 1;
958
959         path = btrfs_alloc_path();
960         BUG_ON(!path);
961         trans = btrfs_join_transaction(root, 1);
962         BUG_ON(!trans);
963
964         cow_start = (u64)-1;
965         cur_offset = start;
966         while (1) {
967                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
968                                                cur_offset, 0);
969                 BUG_ON(ret < 0);
970                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
971                         leaf = path->nodes[0];
972                         btrfs_item_key_to_cpu(leaf, &found_key,
973                                               path->slots[0] - 1);
974                         if (found_key.objectid == inode->i_ino &&
975                             found_key.type == BTRFS_EXTENT_DATA_KEY)
976                                 path->slots[0]--;
977                 }
978                 check_prev = 0;
979 next_slot:
980                 leaf = path->nodes[0];
981                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982                         ret = btrfs_next_leaf(root, path);
983                         if (ret < 0)
984                                 BUG_ON(1);
985                         if (ret > 0)
986                                 break;
987                         leaf = path->nodes[0];
988                 }
989
990                 nocow = 0;
991                 disk_bytenr = 0;
992                 num_bytes = 0;
993                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
994
995                 if (found_key.objectid > inode->i_ino ||
996                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
997                     found_key.offset > end)
998                         break;
999
1000                 if (found_key.offset > cur_offset) {
1001                         extent_end = found_key.offset;
1002                         goto out_check;
1003                 }
1004
1005                 fi = btrfs_item_ptr(leaf, path->slots[0],
1006                                     struct btrfs_file_extent_item);
1007                 extent_type = btrfs_file_extent_type(leaf, fi);
1008
1009                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1010                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1011                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1012                         extent_end = found_key.offset +
1013                                 btrfs_file_extent_num_bytes(leaf, fi);
1014                         if (extent_end <= start) {
1015                                 path->slots[0]++;
1016                                 goto next_slot;
1017                         }
1018                         if (disk_bytenr == 0)
1019                                 goto out_check;
1020                         if (btrfs_file_extent_compression(leaf, fi) ||
1021                             btrfs_file_extent_encryption(leaf, fi) ||
1022                             btrfs_file_extent_other_encoding(leaf, fi))
1023                                 goto out_check;
1024                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1025                                 goto out_check;
1026                         if (btrfs_extent_readonly(root, disk_bytenr))
1027                                 goto out_check;
1028                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1029                                                   disk_bytenr))
1030                                 goto out_check;
1031                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1032                         disk_bytenr += cur_offset - found_key.offset;
1033                         num_bytes = min(end + 1, extent_end) - cur_offset;
1034                         /*
1035                          * force cow if csum exists in the range.
1036                          * this ensure that csum for a given extent are
1037                          * either valid or do not exist.
1038                          */
1039                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1040                                 goto out_check;
1041                         nocow = 1;
1042                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1043                         extent_end = found_key.offset +
1044                                 btrfs_file_extent_inline_len(leaf, fi);
1045                         extent_end = ALIGN(extent_end, root->sectorsize);
1046                 } else {
1047                         BUG_ON(1);
1048                 }
1049 out_check:
1050                 if (extent_end <= start) {
1051                         path->slots[0]++;
1052                         goto next_slot;
1053                 }
1054                 if (!nocow) {
1055                         if (cow_start == (u64)-1)
1056                                 cow_start = cur_offset;
1057                         cur_offset = extent_end;
1058                         if (cur_offset > end)
1059                                 break;
1060                         path->slots[0]++;
1061                         goto next_slot;
1062                 }
1063
1064                 btrfs_release_path(root, path);
1065                 if (cow_start != (u64)-1) {
1066                         ret = cow_file_range(inode, locked_page, cow_start,
1067                                         found_key.offset - 1, page_started,
1068                                         nr_written, 1);
1069                         BUG_ON(ret);
1070                         cow_start = (u64)-1;
1071                 }
1072
1073                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1074                         struct extent_map *em;
1075                         struct extent_map_tree *em_tree;
1076                         em_tree = &BTRFS_I(inode)->extent_tree;
1077                         em = alloc_extent_map(GFP_NOFS);
1078                         em->start = cur_offset;
1079                         em->orig_start = em->start;
1080                         em->len = num_bytes;
1081                         em->block_len = num_bytes;
1082                         em->block_start = disk_bytenr;
1083                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1084                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1085                         while (1) {
1086                                 spin_lock(&em_tree->lock);
1087                                 ret = add_extent_mapping(em_tree, em);
1088                                 spin_unlock(&em_tree->lock);
1089                                 if (ret != -EEXIST) {
1090                                         free_extent_map(em);
1091                                         break;
1092                                 }
1093                                 btrfs_drop_extent_cache(inode, em->start,
1094                                                 em->start + em->len - 1, 0);
1095                         }
1096                         type = BTRFS_ORDERED_PREALLOC;
1097                 } else {
1098                         type = BTRFS_ORDERED_NOCOW;
1099                 }
1100
1101                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1102                                                num_bytes, num_bytes, type);
1103                 BUG_ON(ret);
1104
1105                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1106                                         cur_offset, cur_offset + num_bytes - 1,
1107                                         locked_page, 1, 1, 1, 0, 0, 0);
1108                 cur_offset = extent_end;
1109                 if (cur_offset > end)
1110                         break;
1111         }
1112         btrfs_release_path(root, path);
1113
1114         if (cur_offset <= end && cow_start == (u64)-1)
1115                 cow_start = cur_offset;
1116         if (cow_start != (u64)-1) {
1117                 ret = cow_file_range(inode, locked_page, cow_start, end,
1118                                      page_started, nr_written, 1);
1119                 BUG_ON(ret);
1120         }
1121
1122         ret = btrfs_end_transaction(trans, root);
1123         BUG_ON(ret);
1124         btrfs_free_path(path);
1125         return 0;
1126 }
1127
1128 /*
1129  * extent_io.c call back to do delayed allocation processing
1130  */
1131 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1132                               u64 start, u64 end, int *page_started,
1133                               unsigned long *nr_written)
1134 {
1135         int ret;
1136
1137         if (btrfs_test_flag(inode, NODATACOW))
1138                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1139                                          page_started, 1, nr_written);
1140         else if (btrfs_test_flag(inode, PREALLOC))
1141                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1142                                          page_started, 0, nr_written);
1143         else
1144                 ret = cow_file_range_async(inode, locked_page, start, end,
1145                                            page_started, nr_written);
1146
1147         return ret;
1148 }
1149
1150 /*
1151  * extent_io.c set_bit_hook, used to track delayed allocation
1152  * bytes in this file, and to maintain the list of inodes that
1153  * have pending delalloc work to be done.
1154  */
1155 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156                        unsigned long old, unsigned long bits)
1157 {
1158         /*
1159          * set_bit and clear bit hooks normally require _irqsave/restore
1160          * but in this case, we are only testeing for the DELALLOC
1161          * bit, which is only set or cleared with irqs on
1162          */
1163         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164                 struct btrfs_root *root = BTRFS_I(inode)->root;
1165                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166                 spin_lock(&root->fs_info->delalloc_lock);
1167                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1168                 root->fs_info->delalloc_bytes += end - start + 1;
1169                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171                                       &root->fs_info->delalloc_inodes);
1172                 }
1173                 spin_unlock(&root->fs_info->delalloc_lock);
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * extent_io.c clear_bit_hook, see set_bit_hook for why
1180  */
1181 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1182                          unsigned long old, unsigned long bits)
1183 {
1184         /*
1185          * set_bit and clear bit hooks normally require _irqsave/restore
1186          * but in this case, we are only testeing for the DELALLOC
1187          * bit, which is only set or cleared with irqs on
1188          */
1189         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190                 struct btrfs_root *root = BTRFS_I(inode)->root;
1191
1192                 spin_lock(&root->fs_info->delalloc_lock);
1193                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1194                         printk(KERN_INFO "btrfs warning: delalloc account "
1195                                "%llu %llu\n",
1196                                (unsigned long long)end - start + 1,
1197                                (unsigned long long)
1198                                root->fs_info->delalloc_bytes);
1199                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1200                         root->fs_info->delalloc_bytes = 0;
1201                         BTRFS_I(inode)->delalloc_bytes = 0;
1202                 } else {
1203                         btrfs_delalloc_free_space(root, inode,
1204                                                   end - start + 1);
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct btrfs_ordered_sum *sum;
1328
1329         btrfs_set_trans_block_group(trans, inode);
1330
1331         list_for_each_entry(sum, list, list) {
1332                 btrfs_csum_file_blocks(trans,
1333                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1334         }
1335         return 0;
1336 }
1337
1338 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339 {
1340         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1341                 WARN_ON(1);
1342         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343                                    GFP_NOFS);
1344 }
1345
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup {
1348         struct page *page;
1349         struct btrfs_work work;
1350 };
1351
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1353 {
1354         struct btrfs_writepage_fixup *fixup;
1355         struct btrfs_ordered_extent *ordered;
1356         struct page *page;
1357         struct inode *inode;
1358         u64 page_start;
1359         u64 page_end;
1360
1361         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362         page = fixup->page;
1363 again:
1364         lock_page(page);
1365         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366                 ClearPageChecked(page);
1367                 goto out_page;
1368         }
1369
1370         inode = page->mapping->host;
1371         page_start = page_offset(page);
1372         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375
1376         /* already ordered? We're done */
1377         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378                              EXTENT_ORDERED, 0)) {
1379                 goto out;
1380         }
1381
1382         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383         if (ordered) {
1384                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385                               page_end, GFP_NOFS);
1386                 unlock_page(page);
1387                 btrfs_start_ordered_extent(inode, ordered, 1);
1388                 goto again;
1389         }
1390
1391         btrfs_set_extent_delalloc(inode, page_start, page_end);
1392         ClearPageChecked(page);
1393 out:
1394         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395 out_page:
1396         unlock_page(page);
1397         page_cache_release(page);
1398 }
1399
1400 /*
1401  * There are a few paths in the higher layers of the kernel that directly
1402  * set the page dirty bit without asking the filesystem if it is a
1403  * good idea.  This causes problems because we want to make sure COW
1404  * properly happens and the data=ordered rules are followed.
1405  *
1406  * In our case any range that doesn't have the ORDERED bit set
1407  * hasn't been properly setup for IO.  We kick off an async process
1408  * to fix it up.  The async helper will wait for ordered extents, set
1409  * the delalloc bit and make it safe to write the page.
1410  */
1411 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1412 {
1413         struct inode *inode = page->mapping->host;
1414         struct btrfs_writepage_fixup *fixup;
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret;
1417
1418         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419                              EXTENT_ORDERED, 0);
1420         if (ret)
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u8 compression, u8 encryption,
1443                                        u16 other_encoding, int extent_type)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         struct btrfs_file_extent_item *fi;
1447         struct btrfs_path *path;
1448         struct extent_buffer *leaf;
1449         struct btrfs_key ins;
1450         u64 hint;
1451         int ret;
1452
1453         path = btrfs_alloc_path();
1454         BUG_ON(!path);
1455
1456         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1457                                  file_pos + num_bytes, file_pos, &hint);
1458         BUG_ON(ret);
1459
1460         ins.objectid = inode->i_ino;
1461         ins.offset = file_pos;
1462         ins.type = BTRFS_EXTENT_DATA_KEY;
1463         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1464         BUG_ON(ret);
1465         leaf = path->nodes[0];
1466         fi = btrfs_item_ptr(leaf, path->slots[0],
1467                             struct btrfs_file_extent_item);
1468         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1469         btrfs_set_file_extent_type(leaf, fi, extent_type);
1470         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1471         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1472         btrfs_set_file_extent_offset(leaf, fi, 0);
1473         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1474         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1475         btrfs_set_file_extent_compression(leaf, fi, compression);
1476         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1477         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1478         btrfs_mark_buffer_dirty(leaf);
1479
1480         inode_add_bytes(inode, num_bytes);
1481         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1482
1483         ins.objectid = disk_bytenr;
1484         ins.offset = disk_num_bytes;
1485         ins.type = BTRFS_EXTENT_ITEM_KEY;
1486         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1487                                           root->root_key.objectid,
1488                                           trans->transid, inode->i_ino, &ins);
1489         BUG_ON(ret);
1490
1491         btrfs_free_path(path);
1492         return 0;
1493 }
1494
1495 /* as ordered data IO finishes, this gets called so we can finish
1496  * an ordered extent if the range of bytes in the file it covers are
1497  * fully written.
1498  */
1499 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1500 {
1501         struct btrfs_root *root = BTRFS_I(inode)->root;
1502         struct btrfs_trans_handle *trans;
1503         struct btrfs_ordered_extent *ordered_extent;
1504         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1505         int compressed = 0;
1506         int ret;
1507
1508         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1509         if (!ret)
1510                 return 0;
1511
1512         trans = btrfs_join_transaction(root, 1);
1513
1514         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1515         BUG_ON(!ordered_extent);
1516         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1517                 goto nocow;
1518
1519         lock_extent(io_tree, ordered_extent->file_offset,
1520                     ordered_extent->file_offset + ordered_extent->len - 1,
1521                     GFP_NOFS);
1522
1523         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1524                 compressed = 1;
1525         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1526                 BUG_ON(compressed);
1527                 ret = btrfs_mark_extent_written(trans, root, inode,
1528                                                 ordered_extent->file_offset,
1529                                                 ordered_extent->file_offset +
1530                                                 ordered_extent->len);
1531                 BUG_ON(ret);
1532         } else {
1533                 ret = insert_reserved_file_extent(trans, inode,
1534                                                 ordered_extent->file_offset,
1535                                                 ordered_extent->start,
1536                                                 ordered_extent->disk_len,
1537                                                 ordered_extent->len,
1538                                                 ordered_extent->len,
1539                                                 compressed, 0, 0,
1540                                                 BTRFS_FILE_EXTENT_REG);
1541                 BUG_ON(ret);
1542         }
1543         unlock_extent(io_tree, ordered_extent->file_offset,
1544                     ordered_extent->file_offset + ordered_extent->len - 1,
1545                     GFP_NOFS);
1546 nocow:
1547         add_pending_csums(trans, inode, ordered_extent->file_offset,
1548                           &ordered_extent->list);
1549
1550         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1551         btrfs_ordered_update_i_size(inode, ordered_extent);
1552         btrfs_update_inode(trans, root, inode);
1553         btrfs_remove_ordered_extent(inode, ordered_extent);
1554         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1555
1556         /* once for us */
1557         btrfs_put_ordered_extent(ordered_extent);
1558         /* once for the tree */
1559         btrfs_put_ordered_extent(ordered_extent);
1560
1561         btrfs_end_transaction(trans, root);
1562         return 0;
1563 }
1564
1565 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1566                                 struct extent_state *state, int uptodate)
1567 {
1568         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1569 }
1570
1571 /*
1572  * When IO fails, either with EIO or csum verification fails, we
1573  * try other mirrors that might have a good copy of the data.  This
1574  * io_failure_record is used to record state as we go through all the
1575  * mirrors.  If another mirror has good data, the page is set up to date
1576  * and things continue.  If a good mirror can't be found, the original
1577  * bio end_io callback is called to indicate things have failed.
1578  */
1579 struct io_failure_record {
1580         struct page *page;
1581         u64 start;
1582         u64 len;
1583         u64 logical;
1584         unsigned long bio_flags;
1585         int last_mirror;
1586 };
1587
1588 static int btrfs_io_failed_hook(struct bio *failed_bio,
1589                          struct page *page, u64 start, u64 end,
1590                          struct extent_state *state)
1591 {
1592         struct io_failure_record *failrec = NULL;
1593         u64 private;
1594         struct extent_map *em;
1595         struct inode *inode = page->mapping->host;
1596         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1597         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1598         struct bio *bio;
1599         int num_copies;
1600         int ret;
1601         int rw;
1602         u64 logical;
1603
1604         ret = get_state_private(failure_tree, start, &private);
1605         if (ret) {
1606                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1607                 if (!failrec)
1608                         return -ENOMEM;
1609                 failrec->start = start;
1610                 failrec->len = end - start + 1;
1611                 failrec->last_mirror = 0;
1612                 failrec->bio_flags = 0;
1613
1614                 spin_lock(&em_tree->lock);
1615                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1616                 if (em->start > start || em->start + em->len < start) {
1617                         free_extent_map(em);
1618                         em = NULL;
1619                 }
1620                 spin_unlock(&em_tree->lock);
1621
1622                 if (!em || IS_ERR(em)) {
1623                         kfree(failrec);
1624                         return -EIO;
1625                 }
1626                 logical = start - em->start;
1627                 logical = em->block_start + logical;
1628                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1629                         logical = em->block_start;
1630                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1631                 }
1632                 failrec->logical = logical;
1633                 free_extent_map(em);
1634                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1635                                 EXTENT_DIRTY, GFP_NOFS);
1636                 set_state_private(failure_tree, start,
1637                                  (u64)(unsigned long)failrec);
1638         } else {
1639                 failrec = (struct io_failure_record *)(unsigned long)private;
1640         }
1641         num_copies = btrfs_num_copies(
1642                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1643                               failrec->logical, failrec->len);
1644         failrec->last_mirror++;
1645         if (!state) {
1646                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1647                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1648                                                     failrec->start,
1649                                                     EXTENT_LOCKED);
1650                 if (state && state->start != failrec->start)
1651                         state = NULL;
1652                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1653         }
1654         if (!state || failrec->last_mirror > num_copies) {
1655                 set_state_private(failure_tree, failrec->start, 0);
1656                 clear_extent_bits(failure_tree, failrec->start,
1657                                   failrec->start + failrec->len - 1,
1658                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1659                 kfree(failrec);
1660                 return -EIO;
1661         }
1662         bio = bio_alloc(GFP_NOFS, 1);
1663         bio->bi_private = state;
1664         bio->bi_end_io = failed_bio->bi_end_io;
1665         bio->bi_sector = failrec->logical >> 9;
1666         bio->bi_bdev = failed_bio->bi_bdev;
1667         bio->bi_size = 0;
1668
1669         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1670         if (failed_bio->bi_rw & (1 << BIO_RW))
1671                 rw = WRITE;
1672         else
1673                 rw = READ;
1674
1675         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1676                                                       failrec->last_mirror,
1677                                                       failrec->bio_flags);
1678         return 0;
1679 }
1680
1681 /*
1682  * each time an IO finishes, we do a fast check in the IO failure tree
1683  * to see if we need to process or clean up an io_failure_record
1684  */
1685 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1686 {
1687         u64 private;
1688         u64 private_failure;
1689         struct io_failure_record *failure;
1690         int ret;
1691
1692         private = 0;
1693         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1694                              (u64)-1, 1, EXTENT_DIRTY)) {
1695                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1696                                         start, &private_failure);
1697                 if (ret == 0) {
1698                         failure = (struct io_failure_record *)(unsigned long)
1699                                    private_failure;
1700                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1701                                           failure->start, 0);
1702                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1703                                           failure->start,
1704                                           failure->start + failure->len - 1,
1705                                           EXTENT_DIRTY | EXTENT_LOCKED,
1706                                           GFP_NOFS);
1707                         kfree(failure);
1708                 }
1709         }
1710         return 0;
1711 }
1712
1713 /*
1714  * when reads are done, we need to check csums to verify the data is correct
1715  * if there's a match, we allow the bio to finish.  If not, we go through
1716  * the io_failure_record routines to find good copies
1717  */
1718 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1719                                struct extent_state *state)
1720 {
1721         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1722         struct inode *inode = page->mapping->host;
1723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1724         char *kaddr;
1725         u64 private = ~(u32)0;
1726         int ret;
1727         struct btrfs_root *root = BTRFS_I(inode)->root;
1728         u32 csum = ~(u32)0;
1729
1730         if (PageChecked(page)) {
1731                 ClearPageChecked(page);
1732                 goto good;
1733         }
1734         if (btrfs_test_flag(inode, NODATASUM))
1735                 return 0;
1736
1737         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1738             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1739                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1740                                   GFP_NOFS);
1741                 return 0;
1742         }
1743
1744         if (state && state->start == start) {
1745                 private = state->private;
1746                 ret = 0;
1747         } else {
1748                 ret = get_state_private(io_tree, start, &private);
1749         }
1750         kaddr = kmap_atomic(page, KM_USER0);
1751         if (ret)
1752                 goto zeroit;
1753
1754         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1755         btrfs_csum_final(csum, (char *)&csum);
1756         if (csum != private)
1757                 goto zeroit;
1758
1759         kunmap_atomic(kaddr, KM_USER0);
1760 good:
1761         /* if the io failure tree for this inode is non-empty,
1762          * check to see if we've recovered from a failed IO
1763          */
1764         btrfs_clean_io_failures(inode, start);
1765         return 0;
1766
1767 zeroit:
1768         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1769                "private %llu\n", page->mapping->host->i_ino,
1770                (unsigned long long)start, csum,
1771                (unsigned long long)private);
1772         memset(kaddr + offset, 1, end - start + 1);
1773         flush_dcache_page(page);
1774         kunmap_atomic(kaddr, KM_USER0);
1775         if (private == 0)
1776                 return 0;
1777         return -EIO;
1778 }
1779
1780 /*
1781  * This creates an orphan entry for the given inode in case something goes
1782  * wrong in the middle of an unlink/truncate.
1783  */
1784 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1785 {
1786         struct btrfs_root *root = BTRFS_I(inode)->root;
1787         int ret = 0;
1788
1789         spin_lock(&root->list_lock);
1790
1791         /* already on the orphan list, we're good */
1792         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1793                 spin_unlock(&root->list_lock);
1794                 return 0;
1795         }
1796
1797         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1798
1799         spin_unlock(&root->list_lock);
1800
1801         /*
1802          * insert an orphan item to track this unlinked/truncated file
1803          */
1804         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1805
1806         return ret;
1807 }
1808
1809 /*
1810  * We have done the truncate/delete so we can go ahead and remove the orphan
1811  * item for this particular inode.
1812  */
1813 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1814 {
1815         struct btrfs_root *root = BTRFS_I(inode)->root;
1816         int ret = 0;
1817
1818         spin_lock(&root->list_lock);
1819
1820         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1821                 spin_unlock(&root->list_lock);
1822                 return 0;
1823         }
1824
1825         list_del_init(&BTRFS_I(inode)->i_orphan);
1826         if (!trans) {
1827                 spin_unlock(&root->list_lock);
1828                 return 0;
1829         }
1830
1831         spin_unlock(&root->list_lock);
1832
1833         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1834
1835         return ret;
1836 }
1837
1838 /*
1839  * this cleans up any orphans that may be left on the list from the last use
1840  * of this root.
1841  */
1842 void btrfs_orphan_cleanup(struct btrfs_root *root)
1843 {
1844         struct btrfs_path *path;
1845         struct extent_buffer *leaf;
1846         struct btrfs_item *item;
1847         struct btrfs_key key, found_key;
1848         struct btrfs_trans_handle *trans;
1849         struct inode *inode;
1850         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1851
1852         path = btrfs_alloc_path();
1853         if (!path)
1854                 return;
1855         path->reada = -1;
1856
1857         key.objectid = BTRFS_ORPHAN_OBJECTID;
1858         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1859         key.offset = (u64)-1;
1860
1861
1862         while (1) {
1863                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1864                 if (ret < 0) {
1865                         printk(KERN_ERR "Error searching slot for orphan: %d"
1866                                "\n", ret);
1867                         break;
1868                 }
1869
1870                 /*
1871                  * if ret == 0 means we found what we were searching for, which
1872                  * is weird, but possible, so only screw with path if we didnt
1873                  * find the key and see if we have stuff that matches
1874                  */
1875                 if (ret > 0) {
1876                         if (path->slots[0] == 0)
1877                                 break;
1878                         path->slots[0]--;
1879                 }
1880
1881                 /* pull out the item */
1882                 leaf = path->nodes[0];
1883                 item = btrfs_item_nr(leaf, path->slots[0]);
1884                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1885
1886                 /* make sure the item matches what we want */
1887                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1888                         break;
1889                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1890                         break;
1891
1892                 /* release the path since we're done with it */
1893                 btrfs_release_path(root, path);
1894
1895                 /*
1896                  * this is where we are basically btrfs_lookup, without the
1897                  * crossing root thing.  we store the inode number in the
1898                  * offset of the orphan item.
1899                  */
1900                 inode = btrfs_iget_locked(root->fs_info->sb,
1901                                           found_key.offset, root);
1902                 if (!inode)
1903                         break;
1904
1905                 if (inode->i_state & I_NEW) {
1906                         BTRFS_I(inode)->root = root;
1907
1908                         /* have to set the location manually */
1909                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1910                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1911                         BTRFS_I(inode)->location.offset = 0;
1912
1913                         btrfs_read_locked_inode(inode);
1914                         unlock_new_inode(inode);
1915                 }
1916
1917                 /*
1918                  * add this inode to the orphan list so btrfs_orphan_del does
1919                  * the proper thing when we hit it
1920                  */
1921                 spin_lock(&root->list_lock);
1922                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1923                 spin_unlock(&root->list_lock);
1924
1925                 /*
1926                  * if this is a bad inode, means we actually succeeded in
1927                  * removing the inode, but not the orphan record, which means
1928                  * we need to manually delete the orphan since iput will just
1929                  * do a destroy_inode
1930                  */
1931                 if (is_bad_inode(inode)) {
1932                         trans = btrfs_start_transaction(root, 1);
1933                         btrfs_orphan_del(trans, inode);
1934                         btrfs_end_transaction(trans, root);
1935                         iput(inode);
1936                         continue;
1937                 }
1938
1939                 /* if we have links, this was a truncate, lets do that */
1940                 if (inode->i_nlink) {
1941                         nr_truncate++;
1942                         btrfs_truncate(inode);
1943                 } else {
1944                         nr_unlink++;
1945                 }
1946
1947                 /* this will do delete_inode and everything for us */
1948                 iput(inode);
1949         }
1950
1951         if (nr_unlink)
1952                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1953         if (nr_truncate)
1954                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1955
1956         btrfs_free_path(path);
1957 }
1958
1959 /*
1960  * read an inode from the btree into the in-memory inode
1961  */
1962 void btrfs_read_locked_inode(struct inode *inode)
1963 {
1964         struct btrfs_path *path;
1965         struct extent_buffer *leaf;
1966         struct btrfs_inode_item *inode_item;
1967         struct btrfs_timespec *tspec;
1968         struct btrfs_root *root = BTRFS_I(inode)->root;
1969         struct btrfs_key location;
1970         u64 alloc_group_block;
1971         u32 rdev;
1972         int ret;
1973
1974         path = btrfs_alloc_path();
1975         BUG_ON(!path);
1976         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1977
1978         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1979         if (ret)
1980                 goto make_bad;
1981
1982         leaf = path->nodes[0];
1983         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1984                                     struct btrfs_inode_item);
1985
1986         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1987         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1988         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1989         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1990         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1991
1992         tspec = btrfs_inode_atime(inode_item);
1993         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1994         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1995
1996         tspec = btrfs_inode_mtime(inode_item);
1997         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1998         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1999
2000         tspec = btrfs_inode_ctime(inode_item);
2001         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2002         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2003
2004         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2005         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2006         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2007         inode->i_generation = BTRFS_I(inode)->generation;
2008         inode->i_rdev = 0;
2009         rdev = btrfs_inode_rdev(leaf, inode_item);
2010
2011         BTRFS_I(inode)->index_cnt = (u64)-1;
2012         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2013
2014         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2015
2016         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2017                                                 alloc_group_block, 0);
2018         btrfs_free_path(path);
2019         inode_item = NULL;
2020
2021         switch (inode->i_mode & S_IFMT) {
2022         case S_IFREG:
2023                 inode->i_mapping->a_ops = &btrfs_aops;
2024                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2025                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2026                 inode->i_fop = &btrfs_file_operations;
2027                 inode->i_op = &btrfs_file_inode_operations;
2028                 break;
2029         case S_IFDIR:
2030                 inode->i_fop = &btrfs_dir_file_operations;
2031                 if (root == root->fs_info->tree_root)
2032                         inode->i_op = &btrfs_dir_ro_inode_operations;
2033                 else
2034                         inode->i_op = &btrfs_dir_inode_operations;
2035                 break;
2036         case S_IFLNK:
2037                 inode->i_op = &btrfs_symlink_inode_operations;
2038                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2039                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2040                 break;
2041         default:
2042                 inode->i_op = &btrfs_special_inode_operations;
2043                 init_special_inode(inode, inode->i_mode, rdev);
2044                 break;
2045         }
2046         return;
2047
2048 make_bad:
2049         btrfs_free_path(path);
2050         make_bad_inode(inode);
2051 }
2052
2053 /*
2054  * given a leaf and an inode, copy the inode fields into the leaf
2055  */
2056 static void fill_inode_item(struct btrfs_trans_handle *trans,
2057                             struct extent_buffer *leaf,
2058                             struct btrfs_inode_item *item,
2059                             struct inode *inode)
2060 {
2061         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2062         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2063         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2064         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2065         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2066
2067         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2068                                inode->i_atime.tv_sec);
2069         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2070                                 inode->i_atime.tv_nsec);
2071
2072         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2073                                inode->i_mtime.tv_sec);
2074         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2075                                 inode->i_mtime.tv_nsec);
2076
2077         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2078                                inode->i_ctime.tv_sec);
2079         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2080                                 inode->i_ctime.tv_nsec);
2081
2082         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2083         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2084         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2085         btrfs_set_inode_transid(leaf, item, trans->transid);
2086         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2087         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2088         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2089 }
2090
2091 /*
2092  * copy everything in the in-memory inode into the btree.
2093  */
2094 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2095                                 struct btrfs_root *root, struct inode *inode)
2096 {
2097         struct btrfs_inode_item *inode_item;
2098         struct btrfs_path *path;
2099         struct extent_buffer *leaf;
2100         int ret;
2101
2102         path = btrfs_alloc_path();
2103         BUG_ON(!path);
2104         ret = btrfs_lookup_inode(trans, root, path,
2105                                  &BTRFS_I(inode)->location, 1);
2106         if (ret) {
2107                 if (ret > 0)
2108                         ret = -ENOENT;
2109                 goto failed;
2110         }
2111
2112         btrfs_unlock_up_safe(path, 1);
2113         leaf = path->nodes[0];
2114         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2115                                   struct btrfs_inode_item);
2116
2117         fill_inode_item(trans, leaf, inode_item, inode);
2118         btrfs_mark_buffer_dirty(leaf);
2119         btrfs_set_inode_last_trans(trans, inode);
2120         ret = 0;
2121 failed:
2122         btrfs_free_path(path);
2123         return ret;
2124 }
2125
2126
2127 /*
2128  * unlink helper that gets used here in inode.c and in the tree logging
2129  * recovery code.  It remove a link in a directory with a given name, and
2130  * also drops the back refs in the inode to the directory
2131  */
2132 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2133                        struct btrfs_root *root,
2134                        struct inode *dir, struct inode *inode,
2135                        const char *name, int name_len)
2136 {
2137         struct btrfs_path *path;
2138         int ret = 0;
2139         struct extent_buffer *leaf;
2140         struct btrfs_dir_item *di;
2141         struct btrfs_key key;
2142         u64 index;
2143
2144         path = btrfs_alloc_path();
2145         if (!path) {
2146                 ret = -ENOMEM;
2147                 goto err;
2148         }
2149
2150         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2151                                     name, name_len, -1);
2152         if (IS_ERR(di)) {
2153                 ret = PTR_ERR(di);
2154                 goto err;
2155         }
2156         if (!di) {
2157                 ret = -ENOENT;
2158                 goto err;
2159         }
2160         leaf = path->nodes[0];
2161         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2162         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2163         if (ret)
2164                 goto err;
2165         btrfs_release_path(root, path);
2166
2167         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2168                                   inode->i_ino,
2169                                   dir->i_ino, &index);
2170         if (ret) {
2171                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2172                        "inode %lu parent %lu\n", name_len, name,
2173                        inode->i_ino, dir->i_ino);
2174                 goto err;
2175         }
2176
2177         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2178                                          index, name, name_len, -1);
2179         if (IS_ERR(di)) {
2180                 ret = PTR_ERR(di);
2181                 goto err;
2182         }
2183         if (!di) {
2184                 ret = -ENOENT;
2185                 goto err;
2186         }
2187         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2188         btrfs_release_path(root, path);
2189
2190         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2191                                          inode, dir->i_ino);
2192         BUG_ON(ret != 0 && ret != -ENOENT);
2193         if (ret != -ENOENT)
2194                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2195
2196         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2197                                            dir, index);
2198         BUG_ON(ret);
2199 err:
2200         btrfs_free_path(path);
2201         if (ret)
2202                 goto out;
2203
2204         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2205         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2206         btrfs_update_inode(trans, root, dir);
2207         btrfs_drop_nlink(inode);
2208         ret = btrfs_update_inode(trans, root, inode);
2209         dir->i_sb->s_dirt = 1;
2210 out:
2211         return ret;
2212 }
2213
2214 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2215 {
2216         struct btrfs_root *root;
2217         struct btrfs_trans_handle *trans;
2218         struct inode *inode = dentry->d_inode;
2219         int ret;
2220         unsigned long nr = 0;
2221
2222         root = BTRFS_I(dir)->root;
2223
2224         trans = btrfs_start_transaction(root, 1);
2225
2226         btrfs_set_trans_block_group(trans, dir);
2227         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2228                                  dentry->d_name.name, dentry->d_name.len);
2229
2230         if (inode->i_nlink == 0)
2231                 ret = btrfs_orphan_add(trans, inode);
2232
2233         nr = trans->blocks_used;
2234
2235         btrfs_end_transaction_throttle(trans, root);
2236         btrfs_btree_balance_dirty(root, nr);
2237         return ret;
2238 }
2239
2240 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2241 {
2242         struct inode *inode = dentry->d_inode;
2243         int err = 0;
2244         int ret;
2245         struct btrfs_root *root = BTRFS_I(dir)->root;
2246         struct btrfs_trans_handle *trans;
2247         unsigned long nr = 0;
2248
2249         /*
2250          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2251          * the root of a subvolume or snapshot
2252          */
2253         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2254             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2255                 return -ENOTEMPTY;
2256         }
2257
2258         trans = btrfs_start_transaction(root, 1);
2259         btrfs_set_trans_block_group(trans, dir);
2260
2261         err = btrfs_orphan_add(trans, inode);
2262         if (err)
2263                 goto fail_trans;
2264
2265         /* now the directory is empty */
2266         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2267                                  dentry->d_name.name, dentry->d_name.len);
2268         if (!err)
2269                 btrfs_i_size_write(inode, 0);
2270
2271 fail_trans:
2272         nr = trans->blocks_used;
2273         ret = btrfs_end_transaction_throttle(trans, root);
2274         btrfs_btree_balance_dirty(root, nr);
2275
2276         if (ret && !err)
2277                 err = ret;
2278         return err;
2279 }
2280
2281 #if 0
2282 /*
2283  * when truncating bytes in a file, it is possible to avoid reading
2284  * the leaves that contain only checksum items.  This can be the
2285  * majority of the IO required to delete a large file, but it must
2286  * be done carefully.
2287  *
2288  * The keys in the level just above the leaves are checked to make sure
2289  * the lowest key in a given leaf is a csum key, and starts at an offset
2290  * after the new  size.
2291  *
2292  * Then the key for the next leaf is checked to make sure it also has
2293  * a checksum item for the same file.  If it does, we know our target leaf
2294  * contains only checksum items, and it can be safely freed without reading
2295  * it.
2296  *
2297  * This is just an optimization targeted at large files.  It may do
2298  * nothing.  It will return 0 unless things went badly.
2299  */
2300 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2301                                      struct btrfs_root *root,
2302                                      struct btrfs_path *path,
2303                                      struct inode *inode, u64 new_size)
2304 {
2305         struct btrfs_key key;
2306         int ret;
2307         int nritems;
2308         struct btrfs_key found_key;
2309         struct btrfs_key other_key;
2310         struct btrfs_leaf_ref *ref;
2311         u64 leaf_gen;
2312         u64 leaf_start;
2313
2314         path->lowest_level = 1;
2315         key.objectid = inode->i_ino;
2316         key.type = BTRFS_CSUM_ITEM_KEY;
2317         key.offset = new_size;
2318 again:
2319         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2320         if (ret < 0)
2321                 goto out;
2322
2323         if (path->nodes[1] == NULL) {
2324                 ret = 0;
2325                 goto out;
2326         }
2327         ret = 0;
2328         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2329         nritems = btrfs_header_nritems(path->nodes[1]);
2330
2331         if (!nritems)
2332                 goto out;
2333
2334         if (path->slots[1] >= nritems)
2335                 goto next_node;
2336
2337         /* did we find a key greater than anything we want to delete? */
2338         if (found_key.objectid > inode->i_ino ||
2339            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2340                 goto out;
2341
2342         /* we check the next key in the node to make sure the leave contains
2343          * only checksum items.  This comparison doesn't work if our
2344          * leaf is the last one in the node
2345          */
2346         if (path->slots[1] + 1 >= nritems) {
2347 next_node:
2348                 /* search forward from the last key in the node, this
2349                  * will bring us into the next node in the tree
2350                  */
2351                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2352
2353                 /* unlikely, but we inc below, so check to be safe */
2354                 if (found_key.offset == (u64)-1)
2355                         goto out;
2356
2357                 /* search_forward needs a path with locks held, do the
2358                  * search again for the original key.  It is possible
2359                  * this will race with a balance and return a path that
2360                  * we could modify, but this drop is just an optimization
2361                  * and is allowed to miss some leaves.
2362                  */
2363                 btrfs_release_path(root, path);
2364                 found_key.offset++;
2365
2366                 /* setup a max key for search_forward */
2367                 other_key.offset = (u64)-1;
2368                 other_key.type = key.type;
2369                 other_key.objectid = key.objectid;
2370
2371                 path->keep_locks = 1;
2372                 ret = btrfs_search_forward(root, &found_key, &other_key,
2373                                            path, 0, 0);
2374                 path->keep_locks = 0;
2375                 if (ret || found_key.objectid != key.objectid ||
2376                     found_key.type != key.type) {
2377                         ret = 0;
2378                         goto out;
2379                 }
2380
2381                 key.offset = found_key.offset;
2382                 btrfs_release_path(root, path);
2383                 cond_resched();
2384                 goto again;
2385         }
2386
2387         /* we know there's one more slot after us in the tree,
2388          * read that key so we can verify it is also a checksum item
2389          */
2390         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2391
2392         if (found_key.objectid < inode->i_ino)
2393                 goto next_key;
2394
2395         if (found_key.type != key.type || found_key.offset < new_size)
2396                 goto next_key;
2397
2398         /*
2399          * if the key for the next leaf isn't a csum key from this objectid,
2400          * we can't be sure there aren't good items inside this leaf.
2401          * Bail out
2402          */
2403         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2404                 goto out;
2405
2406         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2407         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2408         /*
2409          * it is safe to delete this leaf, it contains only
2410          * csum items from this inode at an offset >= new_size
2411          */
2412         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2413         BUG_ON(ret);
2414
2415         if (root->ref_cows && leaf_gen < trans->transid) {
2416                 ref = btrfs_alloc_leaf_ref(root, 0);
2417                 if (ref) {
2418                         ref->root_gen = root->root_key.offset;
2419                         ref->bytenr = leaf_start;
2420                         ref->owner = 0;
2421                         ref->generation = leaf_gen;
2422                         ref->nritems = 0;
2423
2424                         btrfs_sort_leaf_ref(ref);
2425
2426                         ret = btrfs_add_leaf_ref(root, ref, 0);
2427                         WARN_ON(ret);
2428                         btrfs_free_leaf_ref(root, ref);
2429                 } else {
2430                         WARN_ON(1);
2431                 }
2432         }
2433 next_key:
2434         btrfs_release_path(root, path);
2435
2436         if (other_key.objectid == inode->i_ino &&
2437             other_key.type == key.type && other_key.offset > key.offset) {
2438                 key.offset = other_key.offset;
2439                 cond_resched();
2440                 goto again;
2441         }
2442         ret = 0;
2443 out:
2444         /* fixup any changes we've made to the path */
2445         path->lowest_level = 0;
2446         path->keep_locks = 0;
2447         btrfs_release_path(root, path);
2448         return ret;
2449 }
2450
2451 #endif
2452
2453 /*
2454  * this can truncate away extent items, csum items and directory items.
2455  * It starts at a high offset and removes keys until it can't find
2456  * any higher than new_size
2457  *
2458  * csum items that cross the new i_size are truncated to the new size
2459  * as well.
2460  *
2461  * min_type is the minimum key type to truncate down to.  If set to 0, this
2462  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2463  */
2464 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2465                                         struct btrfs_root *root,
2466                                         struct inode *inode,
2467                                         u64 new_size, u32 min_type)
2468 {
2469         int ret;
2470         struct btrfs_path *path;
2471         struct btrfs_key key;
2472         struct btrfs_key found_key;
2473         u32 found_type = (u8)-1;
2474         struct extent_buffer *leaf;
2475         struct btrfs_file_extent_item *fi;
2476         u64 extent_start = 0;
2477         u64 extent_num_bytes = 0;
2478         u64 item_end = 0;
2479         u64 root_gen = 0;
2480         u64 root_owner = 0;
2481         int found_extent;
2482         int del_item;
2483         int pending_del_nr = 0;
2484         int pending_del_slot = 0;
2485         int extent_type = -1;
2486         int encoding;
2487         u64 mask = root->sectorsize - 1;
2488
2489         if (root->ref_cows)
2490                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2491         path = btrfs_alloc_path();
2492         path->reada = -1;
2493         BUG_ON(!path);
2494
2495         /* FIXME, add redo link to tree so we don't leak on crash */
2496         key.objectid = inode->i_ino;
2497         key.offset = (u64)-1;
2498         key.type = (u8)-1;
2499
2500 search_again:
2501         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2502         if (ret < 0)
2503                 goto error;
2504
2505         if (ret > 0) {
2506                 /* there are no items in the tree for us to truncate, we're
2507                  * done
2508                  */
2509                 if (path->slots[0] == 0) {
2510                         ret = 0;
2511                         goto error;
2512                 }
2513                 path->slots[0]--;
2514         }
2515
2516         while (1) {
2517                 fi = NULL;
2518                 leaf = path->nodes[0];
2519                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2520                 found_type = btrfs_key_type(&found_key);
2521                 encoding = 0;
2522
2523                 if (found_key.objectid != inode->i_ino)
2524                         break;
2525
2526                 if (found_type < min_type)
2527                         break;
2528
2529                 item_end = found_key.offset;
2530                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2531                         fi = btrfs_item_ptr(leaf, path->slots[0],
2532                                             struct btrfs_file_extent_item);
2533                         extent_type = btrfs_file_extent_type(leaf, fi);
2534                         encoding = btrfs_file_extent_compression(leaf, fi);
2535                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2536                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2537
2538                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2539                                 item_end +=
2540                                     btrfs_file_extent_num_bytes(leaf, fi);
2541                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2542                                 item_end += btrfs_file_extent_inline_len(leaf,
2543                                                                          fi);
2544                         }
2545                         item_end--;
2546                 }
2547                 if (item_end < new_size) {
2548                         if (found_type == BTRFS_DIR_ITEM_KEY)
2549                                 found_type = BTRFS_INODE_ITEM_KEY;
2550                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2551                                 found_type = BTRFS_EXTENT_DATA_KEY;
2552                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2553                                 found_type = BTRFS_XATTR_ITEM_KEY;
2554                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2555                                 found_type = BTRFS_INODE_REF_KEY;
2556                         else if (found_type)
2557                                 found_type--;
2558                         else
2559                                 break;
2560                         btrfs_set_key_type(&key, found_type);
2561                         goto next;
2562                 }
2563                 if (found_key.offset >= new_size)
2564                         del_item = 1;
2565                 else
2566                         del_item = 0;
2567                 found_extent = 0;
2568
2569                 /* FIXME, shrink the extent if the ref count is only 1 */
2570                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2571                         goto delete;
2572
2573                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2574                         u64 num_dec;
2575                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2576                         if (!del_item && !encoding) {
2577                                 u64 orig_num_bytes =
2578                                         btrfs_file_extent_num_bytes(leaf, fi);
2579                                 extent_num_bytes = new_size -
2580                                         found_key.offset + root->sectorsize - 1;
2581                                 extent_num_bytes = extent_num_bytes &
2582                                         ~((u64)root->sectorsize - 1);
2583                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2584                                                          extent_num_bytes);
2585                                 num_dec = (orig_num_bytes -
2586                                            extent_num_bytes);
2587                                 if (root->ref_cows && extent_start != 0)
2588                                         inode_sub_bytes(inode, num_dec);
2589                                 btrfs_mark_buffer_dirty(leaf);
2590                         } else {
2591                                 extent_num_bytes =
2592                                         btrfs_file_extent_disk_num_bytes(leaf,
2593                                                                          fi);
2594                                 /* FIXME blocksize != 4096 */
2595                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2596                                 if (extent_start != 0) {
2597                                         found_extent = 1;
2598                                         if (root->ref_cows)
2599                                                 inode_sub_bytes(inode, num_dec);
2600                                 }
2601                                 root_gen = btrfs_header_generation(leaf);
2602                                 root_owner = btrfs_header_owner(leaf);
2603                         }
2604                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2605                         /*
2606                          * we can't truncate inline items that have had
2607                          * special encodings
2608                          */
2609                         if (!del_item &&
2610                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2611                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2612                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2613                                 u32 size = new_size - found_key.offset;
2614
2615                                 if (root->ref_cows) {
2616                                         inode_sub_bytes(inode, item_end + 1 -
2617                                                         new_size);
2618                                 }
2619                                 size =
2620                                     btrfs_file_extent_calc_inline_size(size);
2621                                 ret = btrfs_truncate_item(trans, root, path,
2622                                                           size, 1);
2623                                 BUG_ON(ret);
2624                         } else if (root->ref_cows) {
2625                                 inode_sub_bytes(inode, item_end + 1 -
2626                                                 found_key.offset);
2627                         }
2628                 }
2629 delete:
2630                 if (del_item) {
2631                         if (!pending_del_nr) {
2632                                 /* no pending yet, add ourselves */
2633                                 pending_del_slot = path->slots[0];
2634                                 pending_del_nr = 1;
2635                         } else if (pending_del_nr &&
2636                                    path->slots[0] + 1 == pending_del_slot) {
2637                                 /* hop on the pending chunk */
2638                                 pending_del_nr++;
2639                                 pending_del_slot = path->slots[0];
2640                         } else {
2641                                 BUG();
2642                         }
2643                 } else {
2644                         break;
2645                 }
2646                 if (found_extent) {
2647                         ret = btrfs_free_extent(trans, root, extent_start,
2648                                                 extent_num_bytes,
2649                                                 leaf->start, root_owner,
2650                                                 root_gen, inode->i_ino, 0);
2651                         BUG_ON(ret);
2652                 }
2653 next:
2654                 if (path->slots[0] == 0) {
2655                         if (pending_del_nr)
2656                                 goto del_pending;
2657                         btrfs_release_path(root, path);
2658                         if (found_type == BTRFS_INODE_ITEM_KEY)
2659                                 break;
2660                         goto search_again;
2661                 }
2662
2663                 path->slots[0]--;
2664                 if (pending_del_nr &&
2665                     path->slots[0] + 1 != pending_del_slot) {
2666                         struct btrfs_key debug;
2667 del_pending:
2668                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2669                                               pending_del_slot);
2670                         ret = btrfs_del_items(trans, root, path,
2671                                               pending_del_slot,
2672                                               pending_del_nr);
2673                         BUG_ON(ret);
2674                         pending_del_nr = 0;
2675                         btrfs_release_path(root, path);
2676                         if (found_type == BTRFS_INODE_ITEM_KEY)
2677                                 break;
2678                         goto search_again;
2679                 }
2680         }
2681         ret = 0;
2682 error:
2683         if (pending_del_nr) {
2684                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2685                                       pending_del_nr);
2686         }
2687         btrfs_free_path(path);
2688         inode->i_sb->s_dirt = 1;
2689         return ret;
2690 }
2691
2692 /*
2693  * taken from block_truncate_page, but does cow as it zeros out
2694  * any bytes left in the last page in the file.
2695  */
2696 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2697 {
2698         struct inode *inode = mapping->host;
2699         struct btrfs_root *root = BTRFS_I(inode)->root;
2700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2701         struct btrfs_ordered_extent *ordered;
2702         char *kaddr;
2703         u32 blocksize = root->sectorsize;
2704         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2705         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2706         struct page *page;
2707         int ret = 0;
2708         u64 page_start;
2709         u64 page_end;
2710
2711         if ((offset & (blocksize - 1)) == 0)
2712                 goto out;
2713
2714         ret = -ENOMEM;
2715 again:
2716         page = grab_cache_page(mapping, index);
2717         if (!page)
2718                 goto out;
2719
2720         page_start = page_offset(page);
2721         page_end = page_start + PAGE_CACHE_SIZE - 1;
2722
2723         if (!PageUptodate(page)) {
2724                 ret = btrfs_readpage(NULL, page);
2725                 lock_page(page);
2726                 if (page->mapping != mapping) {
2727                         unlock_page(page);
2728                         page_cache_release(page);
2729                         goto again;
2730                 }
2731                 if (!PageUptodate(page)) {
2732                         ret = -EIO;
2733                         goto out_unlock;
2734                 }
2735         }
2736         wait_on_page_writeback(page);
2737
2738         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2739         set_page_extent_mapped(page);
2740
2741         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2742         if (ordered) {
2743                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2744                 unlock_page(page);
2745                 page_cache_release(page);
2746                 btrfs_start_ordered_extent(inode, ordered, 1);
2747                 btrfs_put_ordered_extent(ordered);
2748                 goto again;
2749         }
2750
2751         btrfs_set_extent_delalloc(inode, page_start, page_end);
2752         ret = 0;
2753         if (offset != PAGE_CACHE_SIZE) {
2754                 kaddr = kmap(page);
2755                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2756                 flush_dcache_page(page);
2757                 kunmap(page);
2758         }
2759         ClearPageChecked(page);
2760         set_page_dirty(page);
2761         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2762
2763 out_unlock:
2764         unlock_page(page);
2765         page_cache_release(page);
2766 out:
2767         return ret;
2768 }
2769
2770 int btrfs_cont_expand(struct inode *inode, loff_t size)
2771 {
2772         struct btrfs_trans_handle *trans;
2773         struct btrfs_root *root = BTRFS_I(inode)->root;
2774         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2775         struct extent_map *em;
2776         u64 mask = root->sectorsize - 1;
2777         u64 hole_start = (inode->i_size + mask) & ~mask;
2778         u64 block_end = (size + mask) & ~mask;
2779         u64 last_byte;
2780         u64 cur_offset;
2781         u64 hole_size;
2782         int err;
2783
2784         if (size <= hole_start)
2785                 return 0;
2786
2787         err = btrfs_check_metadata_free_space(root);
2788         if (err)
2789                 return err;
2790
2791         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2792
2793         while (1) {
2794                 struct btrfs_ordered_extent *ordered;
2795                 btrfs_wait_ordered_range(inode, hole_start,
2796                                          block_end - hole_start);
2797                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2798                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2799                 if (!ordered)
2800                         break;
2801                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2802                 btrfs_put_ordered_extent(ordered);
2803         }
2804
2805         trans = btrfs_start_transaction(root, 1);
2806         btrfs_set_trans_block_group(trans, inode);
2807
2808         cur_offset = hole_start;
2809         while (1) {
2810                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2811                                 block_end - cur_offset, 0);
2812                 BUG_ON(IS_ERR(em) || !em);
2813                 last_byte = min(extent_map_end(em), block_end);
2814                 last_byte = (last_byte + mask) & ~mask;
2815                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2816                         u64 hint_byte = 0;
2817                         hole_size = last_byte - cur_offset;
2818                         err = btrfs_drop_extents(trans, root, inode,
2819                                                  cur_offset,
2820                                                  cur_offset + hole_size,
2821                                                  cur_offset, &hint_byte);
2822                         if (err)
2823                                 break;
2824                         err = btrfs_insert_file_extent(trans, root,
2825                                         inode->i_ino, cur_offset, 0,
2826                                         0, hole_size, 0, hole_size,
2827                                         0, 0, 0);
2828                         btrfs_drop_extent_cache(inode, hole_start,
2829                                         last_byte - 1, 0);
2830                 }
2831                 free_extent_map(em);
2832                 cur_offset = last_byte;
2833                 if (err || cur_offset >= block_end)
2834                         break;
2835         }
2836
2837         btrfs_end_transaction(trans, root);
2838         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2839         return err;
2840 }
2841
2842 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2843 {
2844         struct inode *inode = dentry->d_inode;
2845         int err;
2846
2847         err = inode_change_ok(inode, attr);
2848         if (err)
2849                 return err;
2850
2851         if (S_ISREG(inode->i_mode) &&
2852             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2853                 err = btrfs_cont_expand(inode, attr->ia_size);
2854                 if (err)
2855                         return err;
2856         }
2857
2858         err = inode_setattr(inode, attr);
2859
2860         if (!err && ((attr->ia_valid & ATTR_MODE)))
2861                 err = btrfs_acl_chmod(inode);
2862         return err;
2863 }
2864
2865 void btrfs_delete_inode(struct inode *inode)
2866 {
2867         struct btrfs_trans_handle *trans;
2868         struct btrfs_root *root = BTRFS_I(inode)->root;
2869         unsigned long nr;
2870         int ret;
2871
2872         truncate_inode_pages(&inode->i_data, 0);
2873         if (is_bad_inode(inode)) {
2874                 btrfs_orphan_del(NULL, inode);
2875                 goto no_delete;
2876         }
2877         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2878
2879         btrfs_i_size_write(inode, 0);
2880         trans = btrfs_join_transaction(root, 1);
2881
2882         btrfs_set_trans_block_group(trans, inode);
2883         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2884         if (ret) {
2885                 btrfs_orphan_del(NULL, inode);
2886                 goto no_delete_lock;
2887         }
2888
2889         btrfs_orphan_del(trans, inode);
2890
2891         nr = trans->blocks_used;
2892         clear_inode(inode);
2893
2894         btrfs_end_transaction(trans, root);
2895         btrfs_btree_balance_dirty(root, nr);
2896         return;
2897
2898 no_delete_lock:
2899         nr = trans->blocks_used;
2900         btrfs_end_transaction(trans, root);
2901         btrfs_btree_balance_dirty(root, nr);
2902 no_delete:
2903         clear_inode(inode);
2904 }
2905
2906 /*
2907  * this returns the key found in the dir entry in the location pointer.
2908  * If no dir entries were found, location->objectid is 0.
2909  */
2910 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2911                                struct btrfs_key *location)
2912 {
2913         const char *name = dentry->d_name.name;
2914         int namelen = dentry->d_name.len;
2915         struct btrfs_dir_item *di;
2916         struct btrfs_path *path;
2917         struct btrfs_root *root = BTRFS_I(dir)->root;
2918         int ret = 0;
2919
2920         path = btrfs_alloc_path();
2921         BUG_ON(!path);
2922
2923         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2924                                     namelen, 0);
2925         if (IS_ERR(di))
2926                 ret = PTR_ERR(di);
2927
2928         if (!di || IS_ERR(di))
2929                 goto out_err;
2930
2931         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2932 out:
2933         btrfs_free_path(path);
2934         return ret;
2935 out_err:
2936         location->objectid = 0;
2937         goto out;
2938 }
2939
2940 /*
2941  * when we hit a tree root in a directory, the btrfs part of the inode
2942  * needs to be changed to reflect the root directory of the tree root.  This
2943  * is kind of like crossing a mount point.
2944  */
2945 static int fixup_tree_root_location(struct btrfs_root *root,
2946                              struct btrfs_key *location,
2947                              struct btrfs_root **sub_root,
2948                              struct dentry *dentry)
2949 {
2950         struct btrfs_root_item *ri;
2951
2952         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2953                 return 0;
2954         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2955                 return 0;
2956
2957         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2958                                         dentry->d_name.name,
2959                                         dentry->d_name.len);
2960         if (IS_ERR(*sub_root))
2961                 return PTR_ERR(*sub_root);
2962
2963         ri = &(*sub_root)->root_item;
2964         location->objectid = btrfs_root_dirid(ri);
2965         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2966         location->offset = 0;
2967
2968         return 0;
2969 }
2970
2971 static noinline void init_btrfs_i(struct inode *inode)
2972 {
2973         struct btrfs_inode *bi = BTRFS_I(inode);
2974
2975         bi->i_acl = NULL;
2976         bi->i_default_acl = NULL;
2977
2978         bi->generation = 0;
2979         bi->sequence = 0;
2980         bi->last_trans = 0;
2981         bi->logged_trans = 0;
2982         bi->delalloc_bytes = 0;
2983         bi->reserved_bytes = 0;
2984         bi->disk_i_size = 0;
2985         bi->flags = 0;
2986         bi->index_cnt = (u64)-1;
2987         bi->log_dirty_trans = 0;
2988         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2989         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2990                              inode->i_mapping, GFP_NOFS);
2991         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2992                              inode->i_mapping, GFP_NOFS);
2993         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2994         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2995         mutex_init(&BTRFS_I(inode)->extent_mutex);
2996         mutex_init(&BTRFS_I(inode)->log_mutex);
2997 }
2998
2999 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3000 {
3001         struct btrfs_iget_args *args = p;
3002         inode->i_ino = args->ino;
3003         init_btrfs_i(inode);
3004         BTRFS_I(inode)->root = args->root;
3005         btrfs_set_inode_space_info(args->root, inode);
3006         return 0;
3007 }
3008
3009 static int btrfs_find_actor(struct inode *inode, void *opaque)
3010 {
3011         struct btrfs_iget_args *args = opaque;
3012         return args->ino == inode->i_ino &&
3013                 args->root == BTRFS_I(inode)->root;
3014 }
3015
3016 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3017                             struct btrfs_root *root, int wait)
3018 {
3019         struct inode *inode;
3020         struct btrfs_iget_args args;
3021         args.ino = objectid;
3022         args.root = root;
3023
3024         if (wait) {
3025                 inode = ilookup5(s, objectid, btrfs_find_actor,
3026                                  (void *)&args);
3027         } else {
3028                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3029                                         (void *)&args);
3030         }
3031         return inode;
3032 }
3033
3034 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3035                                 struct btrfs_root *root)
3036 {
3037         struct inode *inode;
3038         struct btrfs_iget_args args;
3039         args.ino = objectid;
3040         args.root = root;
3041
3042         inode = iget5_locked(s, objectid, btrfs_find_actor,
3043                              btrfs_init_locked_inode,
3044                              (void *)&args);
3045         return inode;
3046 }
3047
3048 /* Get an inode object given its location and corresponding root.
3049  * Returns in *is_new if the inode was read from disk
3050  */
3051 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3052                          struct btrfs_root *root, int *is_new)
3053 {
3054         struct inode *inode;
3055
3056         inode = btrfs_iget_locked(s, location->objectid, root);
3057         if (!inode)
3058                 return ERR_PTR(-EACCES);
3059
3060         if (inode->i_state & I_NEW) {
3061                 BTRFS_I(inode)->root = root;
3062                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3063                 btrfs_read_locked_inode(inode);
3064                 unlock_new_inode(inode);
3065                 if (is_new)
3066                         *is_new = 1;
3067         } else {
3068                 if (is_new)
3069                         *is_new = 0;
3070         }
3071
3072         return inode;
3073 }
3074
3075 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3076 {
3077         struct inode *inode;
3078         struct btrfs_inode *bi = BTRFS_I(dir);
3079         struct btrfs_root *root = bi->root;
3080         struct btrfs_root *sub_root = root;
3081         struct btrfs_key location;
3082         int ret, new;
3083
3084         if (dentry->d_name.len > BTRFS_NAME_LEN)
3085                 return ERR_PTR(-ENAMETOOLONG);
3086
3087         ret = btrfs_inode_by_name(dir, dentry, &location);
3088
3089         if (ret < 0)
3090                 return ERR_PTR(ret);
3091
3092         inode = NULL;
3093         if (location.objectid) {
3094                 ret = fixup_tree_root_location(root, &location, &sub_root,
3095                                                 dentry);
3096                 if (ret < 0)
3097                         return ERR_PTR(ret);
3098                 if (ret > 0)
3099                         return ERR_PTR(-ENOENT);
3100                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3101                 if (IS_ERR(inode))
3102                         return ERR_CAST(inode);
3103         }
3104         return inode;
3105 }
3106
3107 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3108                                    struct nameidata *nd)
3109 {
3110         struct inode *inode;
3111
3112         if (dentry->d_name.len > BTRFS_NAME_LEN)
3113                 return ERR_PTR(-ENAMETOOLONG);
3114
3115         inode = btrfs_lookup_dentry(dir, dentry);
3116         if (IS_ERR(inode))
3117                 return ERR_CAST(inode);
3118
3119         return d_splice_alias(inode, dentry);
3120 }
3121
3122 static unsigned char btrfs_filetype_table[] = {
3123         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3124 };
3125
3126 static int btrfs_real_readdir(struct file *filp, void *dirent,
3127                               filldir_t filldir)
3128 {
3129         struct inode *inode = filp->f_dentry->d_inode;
3130         struct btrfs_root *root = BTRFS_I(inode)->root;
3131         struct btrfs_item *item;
3132         struct btrfs_dir_item *di;
3133         struct btrfs_key key;
3134         struct btrfs_key found_key;
3135         struct btrfs_path *path;
3136         int ret;
3137         u32 nritems;
3138         struct extent_buffer *leaf;
3139         int slot;
3140         int advance;
3141         unsigned char d_type;
3142         int over = 0;
3143         u32 di_cur;
3144         u32 di_total;
3145         u32 di_len;
3146         int key_type = BTRFS_DIR_INDEX_KEY;
3147         char tmp_name[32];
3148         char *name_ptr;
3149         int name_len;
3150
3151         /* FIXME, use a real flag for deciding about the key type */
3152         if (root->fs_info->tree_root == root)
3153                 key_type = BTRFS_DIR_ITEM_KEY;
3154
3155         /* special case for "." */
3156         if (filp->f_pos == 0) {
3157                 over = filldir(dirent, ".", 1,
3158                                1, inode->i_ino,
3159                                DT_DIR);
3160                 if (over)
3161                         return 0;
3162                 filp->f_pos = 1;
3163         }
3164         /* special case for .., just use the back ref */
3165         if (filp->f_pos == 1) {
3166                 u64 pino = parent_ino(filp->f_path.dentry);
3167                 over = filldir(dirent, "..", 2,
3168                                2, pino, DT_DIR);
3169                 if (over)
3170                         return 0;
3171                 filp->f_pos = 2;
3172         }
3173         path = btrfs_alloc_path();
3174         path->reada = 2;
3175
3176         btrfs_set_key_type(&key, key_type);
3177         key.offset = filp->f_pos;
3178         key.objectid = inode->i_ino;
3179
3180         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3181         if (ret < 0)
3182                 goto err;
3183         advance = 0;
3184
3185         while (1) {
3186                 leaf = path->nodes[0];
3187                 nritems = btrfs_header_nritems(leaf);
3188                 slot = path->slots[0];
3189                 if (advance || slot >= nritems) {
3190                         if (slot >= nritems - 1) {
3191                                 ret = btrfs_next_leaf(root, path);
3192                                 if (ret)
3193                                         break;
3194                                 leaf = path->nodes[0];
3195                                 nritems = btrfs_header_nritems(leaf);
3196                                 slot = path->slots[0];
3197                         } else {
3198                                 slot++;
3199                                 path->slots[0]++;
3200                         }
3201                 }
3202
3203                 advance = 1;
3204                 item = btrfs_item_nr(leaf, slot);
3205                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3206
3207                 if (found_key.objectid != key.objectid)
3208                         break;
3209                 if (btrfs_key_type(&found_key) != key_type)
3210                         break;
3211                 if (found_key.offset < filp->f_pos)
3212                         continue;
3213
3214                 filp->f_pos = found_key.offset;
3215
3216                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3217                 di_cur = 0;
3218                 di_total = btrfs_item_size(leaf, item);
3219
3220                 while (di_cur < di_total) {
3221                         struct btrfs_key location;
3222
3223                         name_len = btrfs_dir_name_len(leaf, di);
3224                         if (name_len <= sizeof(tmp_name)) {
3225                                 name_ptr = tmp_name;
3226                         } else {
3227                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3228                                 if (!name_ptr) {
3229                                         ret = -ENOMEM;
3230                                         goto err;
3231                                 }
3232                         }
3233                         read_extent_buffer(leaf, name_ptr,
3234                                            (unsigned long)(di + 1), name_len);
3235
3236                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3237                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3238
3239                         /* is this a reference to our own snapshot? If so
3240                          * skip it
3241                          */
3242                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3243                             location.objectid == root->root_key.objectid) {
3244                                 over = 0;
3245                                 goto skip;
3246                         }
3247                         over = filldir(dirent, name_ptr, name_len,
3248                                        found_key.offset, location.objectid,
3249                                        d_type);
3250
3251 skip:
3252                         if (name_ptr != tmp_name)
3253                                 kfree(name_ptr);
3254
3255                         if (over)
3256                                 goto nopos;
3257                         di_len = btrfs_dir_name_len(leaf, di) +
3258                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3259                         di_cur += di_len;
3260                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3261                 }
3262         }
3263
3264         /* Reached end of directory/root. Bump pos past the last item. */
3265         if (key_type == BTRFS_DIR_INDEX_KEY)
3266                 filp->f_pos = INT_LIMIT(off_t);
3267         else
3268                 filp->f_pos++;
3269 nopos:
3270         ret = 0;
3271 err:
3272         btrfs_free_path(path);
3273         return ret;
3274 }
3275
3276 int btrfs_write_inode(struct inode *inode, int wait)
3277 {
3278         struct btrfs_root *root = BTRFS_I(inode)->root;
3279         struct btrfs_trans_handle *trans;
3280         int ret = 0;
3281
3282         if (root->fs_info->btree_inode == inode)
3283                 return 0;
3284
3285         if (wait) {
3286                 trans = btrfs_join_transaction(root, 1);
3287                 btrfs_set_trans_block_group(trans, inode);
3288                 ret = btrfs_commit_transaction(trans, root);
3289         }
3290         return ret;
3291 }
3292
3293 /*
3294  * This is somewhat expensive, updating the tree every time the
3295  * inode changes.  But, it is most likely to find the inode in cache.
3296  * FIXME, needs more benchmarking...there are no reasons other than performance
3297  * to keep or drop this code.
3298  */
3299 void btrfs_dirty_inode(struct inode *inode)
3300 {
3301         struct btrfs_root *root = BTRFS_I(inode)->root;
3302         struct btrfs_trans_handle *trans;
3303
3304         trans = btrfs_join_transaction(root, 1);
3305         btrfs_set_trans_block_group(trans, inode);
3306         btrfs_update_inode(trans, root, inode);
3307         btrfs_end_transaction(trans, root);
3308 }
3309
3310 /*
3311  * find the highest existing sequence number in a directory
3312  * and then set the in-memory index_cnt variable to reflect
3313  * free sequence numbers
3314  */
3315 static int btrfs_set_inode_index_count(struct inode *inode)
3316 {
3317         struct btrfs_root *root = BTRFS_I(inode)->root;
3318         struct btrfs_key key, found_key;
3319         struct btrfs_path *path;
3320         struct extent_buffer *leaf;
3321         int ret;
3322
3323         key.objectid = inode->i_ino;
3324         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3325         key.offset = (u64)-1;
3326
3327         path = btrfs_alloc_path();
3328         if (!path)
3329                 return -ENOMEM;
3330
3331         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3332         if (ret < 0)
3333                 goto out;
3334         /* FIXME: we should be able to handle this */
3335         if (ret == 0)
3336                 goto out;
3337         ret = 0;
3338
3339         /*
3340          * MAGIC NUMBER EXPLANATION:
3341          * since we search a directory based on f_pos we have to start at 2
3342          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3343          * else has to start at 2
3344          */
3345         if (path->slots[0] == 0) {
3346                 BTRFS_I(inode)->index_cnt = 2;
3347                 goto out;
3348         }
3349
3350         path->slots[0]--;
3351
3352         leaf = path->nodes[0];
3353         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3354
3355         if (found_key.objectid != inode->i_ino ||
3356             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3357                 BTRFS_I(inode)->index_cnt = 2;
3358                 goto out;
3359         }
3360
3361         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3362 out:
3363         btrfs_free_path(path);
3364         return ret;
3365 }
3366
3367 /*
3368  * helper to find a free sequence number in a given directory.  This current
3369  * code is very simple, later versions will do smarter things in the btree
3370  */
3371 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3372 {
3373         int ret = 0;
3374
3375         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3376                 ret = btrfs_set_inode_index_count(dir);
3377                 if (ret)
3378                         return ret;
3379         }
3380
3381         *index = BTRFS_I(dir)->index_cnt;
3382         BTRFS_I(dir)->index_cnt++;
3383
3384         return ret;
3385 }
3386
3387 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3388                                      struct btrfs_root *root,
3389                                      struct inode *dir,
3390                                      const char *name, int name_len,
3391                                      u64 ref_objectid, u64 objectid,
3392                                      u64 alloc_hint, int mode, u64 *index)
3393 {
3394         struct inode *inode;
3395         struct btrfs_inode_item *inode_item;
3396         struct btrfs_key *location;
3397         struct btrfs_path *path;
3398         struct btrfs_inode_ref *ref;
3399         struct btrfs_key key[2];
3400         u32 sizes[2];
3401         unsigned long ptr;
3402         int ret;
3403         int owner;
3404
3405         path = btrfs_alloc_path();
3406         BUG_ON(!path);
3407
3408         inode = new_inode(root->fs_info->sb);
3409         if (!inode)
3410                 return ERR_PTR(-ENOMEM);
3411
3412         if (dir) {
3413                 ret = btrfs_set_inode_index(dir, index);
3414                 if (ret)
3415                         return ERR_PTR(ret);
3416         }
3417         /*
3418          * index_cnt is ignored for everything but a dir,
3419          * btrfs_get_inode_index_count has an explanation for the magic
3420          * number
3421          */
3422         init_btrfs_i(inode);
3423         BTRFS_I(inode)->index_cnt = 2;
3424         BTRFS_I(inode)->root = root;
3425         BTRFS_I(inode)->generation = trans->transid;
3426         btrfs_set_inode_space_info(root, inode);
3427
3428         if (mode & S_IFDIR)
3429                 owner = 0;
3430         else
3431                 owner = 1;
3432         BTRFS_I(inode)->block_group =
3433                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3434         if ((mode & S_IFREG)) {
3435                 if (btrfs_test_opt(root, NODATASUM))
3436                         btrfs_set_flag(inode, NODATASUM);
3437                 if (btrfs_test_opt(root, NODATACOW))
3438                         btrfs_set_flag(inode, NODATACOW);
3439         }
3440
3441         key[0].objectid = objectid;
3442         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3443         key[0].offset = 0;
3444
3445         key[1].objectid = objectid;
3446         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3447         key[1].offset = ref_objectid;
3448
3449         sizes[0] = sizeof(struct btrfs_inode_item);
3450         sizes[1] = name_len + sizeof(*ref);
3451
3452         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3453         if (ret != 0)
3454                 goto fail;
3455
3456         if (objectid > root->highest_inode)
3457                 root->highest_inode = objectid;
3458
3459         inode->i_uid = current_fsuid();
3460
3461         if (dir && (dir->i_mode & S_ISGID)) {
3462                 inode->i_gid = dir->i_gid;
3463                 if (S_ISDIR(mode))
3464                         mode |= S_ISGID;
3465         } else
3466                 inode->i_gid = current_fsgid();
3467
3468         inode->i_mode = mode;
3469         inode->i_ino = objectid;
3470         inode_set_bytes(inode, 0);
3471         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3472         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3473                                   struct btrfs_inode_item);
3474         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3475
3476         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3477                              struct btrfs_inode_ref);
3478         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3479         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3480         ptr = (unsigned long)(ref + 1);
3481         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3482
3483         btrfs_mark_buffer_dirty(path->nodes[0]);
3484         btrfs_free_path(path);
3485
3486         location = &BTRFS_I(inode)->location;
3487         location->objectid = objectid;
3488         location->offset = 0;
3489         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3490
3491         insert_inode_hash(inode);
3492         return inode;
3493 fail:
3494         if (dir)
3495                 BTRFS_I(dir)->index_cnt--;
3496         btrfs_free_path(path);
3497         return ERR_PTR(ret);
3498 }
3499
3500 static inline u8 btrfs_inode_type(struct inode *inode)
3501 {
3502         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3503 }
3504
3505 /*
3506  * utility function to add 'inode' into 'parent_inode' with
3507  * a give name and a given sequence number.
3508  * if 'add_backref' is true, also insert a backref from the
3509  * inode to the parent directory.
3510  */
3511 int btrfs_add_link(struct btrfs_trans_handle *trans,
3512                    struct inode *parent_inode, struct inode *inode,
3513                    const char *name, int name_len, int add_backref, u64 index)
3514 {
3515         int ret;
3516         struct btrfs_key key;
3517         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3518
3519         key.objectid = inode->i_ino;
3520         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3521         key.offset = 0;
3522
3523         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3524                                     parent_inode->i_ino,
3525                                     &key, btrfs_inode_type(inode),
3526                                     index);
3527         if (ret == 0) {
3528                 if (add_backref) {
3529                         ret = btrfs_insert_inode_ref(trans, root,
3530                                                      name, name_len,
3531                                                      inode->i_ino,
3532                                                      parent_inode->i_ino,
3533                                                      index);
3534                 }
3535                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3536                                    name_len * 2);
3537                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3538                 ret = btrfs_update_inode(trans, root, parent_inode);
3539         }
3540         return ret;
3541 }
3542
3543 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3544                             struct dentry *dentry, struct inode *inode,
3545                             int backref, u64 index)
3546 {
3547         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3548                                  inode, dentry->d_name.name,
3549                                  dentry->d_name.len, backref, index);
3550         if (!err) {
3551                 d_instantiate(dentry, inode);
3552                 return 0;
3553         }
3554         if (err > 0)
3555                 err = -EEXIST;
3556         return err;
3557 }
3558
3559 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3560                         int mode, dev_t rdev)
3561 {
3562         struct btrfs_trans_handle *trans;
3563         struct btrfs_root *root = BTRFS_I(dir)->root;
3564         struct inode *inode = NULL;
3565         int err;
3566         int drop_inode = 0;
3567         u64 objectid;
3568         unsigned long nr = 0;
3569         u64 index = 0;
3570
3571         if (!new_valid_dev(rdev))
3572                 return -EINVAL;
3573
3574         err = btrfs_check_metadata_free_space(root);
3575         if (err)
3576                 goto fail;
3577
3578         trans = btrfs_start_transaction(root, 1);
3579         btrfs_set_trans_block_group(trans, dir);
3580
3581         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3582         if (err) {
3583                 err = -ENOSPC;
3584                 goto out_unlock;
3585         }
3586
3587         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3588                                 dentry->d_name.len,
3589                                 dentry->d_parent->d_inode->i_ino, objectid,
3590                                 BTRFS_I(dir)->block_group, mode, &index);
3591         err = PTR_ERR(inode);
3592         if (IS_ERR(inode))
3593                 goto out_unlock;
3594
3595         err = btrfs_init_inode_security(inode, dir);
3596         if (err) {
3597                 drop_inode = 1;
3598                 goto out_unlock;
3599         }
3600
3601         btrfs_set_trans_block_group(trans, inode);
3602         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3603         if (err)
3604                 drop_inode = 1;
3605         else {
3606                 inode->i_op = &btrfs_special_inode_operations;
3607                 init_special_inode(inode, inode->i_mode, rdev);
3608                 btrfs_update_inode(trans, root, inode);
3609         }
3610         dir->i_sb->s_dirt = 1;
3611         btrfs_update_inode_block_group(trans, inode);
3612         btrfs_update_inode_block_group(trans, dir);
3613 out_unlock:
3614         nr = trans->blocks_used;
3615         btrfs_end_transaction_throttle(trans, root);
3616 fail:
3617         if (drop_inode) {
3618                 inode_dec_link_count(inode);
3619                 iput(inode);
3620         }
3621         btrfs_btree_balance_dirty(root, nr);
3622         return err;
3623 }
3624
3625 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3626                         int mode, struct nameidata *nd)
3627 {
3628         struct btrfs_trans_handle *trans;
3629         struct btrfs_root *root = BTRFS_I(dir)->root;
3630         struct inode *inode = NULL;
3631         int err;
3632         int drop_inode = 0;
3633         unsigned long nr = 0;
3634         u64 objectid;
3635         u64 index = 0;
3636
3637         err = btrfs_check_metadata_free_space(root);
3638         if (err)
3639                 goto fail;
3640         trans = btrfs_start_transaction(root, 1);
3641         btrfs_set_trans_block_group(trans, dir);
3642
3643         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3644         if (err) {
3645                 err = -ENOSPC;
3646                 goto out_unlock;
3647         }
3648
3649         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3650                                 dentry->d_name.len,
3651                                 dentry->d_parent->d_inode->i_ino,
3652                                 objectid, BTRFS_I(dir)->block_group, mode,
3653                                 &index);
3654         err = PTR_ERR(inode);
3655         if (IS_ERR(inode))
3656                 goto out_unlock;
3657
3658         err = btrfs_init_inode_security(inode, dir);
3659         if (err) {
3660                 drop_inode = 1;
3661                 goto out_unlock;
3662         }
3663
3664         btrfs_set_trans_block_group(trans, inode);
3665         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3666         if (err)
3667                 drop_inode = 1;
3668         else {
3669                 inode->i_mapping->a_ops = &btrfs_aops;
3670                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3671                 inode->i_fop = &btrfs_file_operations;
3672                 inode->i_op = &btrfs_file_inode_operations;
3673                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3674         }
3675         dir->i_sb->s_dirt = 1;
3676         btrfs_update_inode_block_group(trans, inode);
3677         btrfs_update_inode_block_group(trans, dir);
3678 out_unlock:
3679         nr = trans->blocks_used;
3680         btrfs_end_transaction_throttle(trans, root);
3681 fail:
3682         if (drop_inode) {
3683                 inode_dec_link_count(inode);
3684                 iput(inode);
3685         }
3686         btrfs_btree_balance_dirty(root, nr);
3687         return err;
3688 }
3689
3690 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3691                       struct dentry *dentry)
3692 {
3693         struct btrfs_trans_handle *trans;
3694         struct btrfs_root *root = BTRFS_I(dir)->root;
3695         struct inode *inode = old_dentry->d_inode;
3696         u64 index;
3697         unsigned long nr = 0;
3698         int err;
3699         int drop_inode = 0;
3700
3701         if (inode->i_nlink == 0)
3702                 return -ENOENT;
3703
3704         btrfs_inc_nlink(inode);
3705         err = btrfs_check_metadata_free_space(root);
3706         if (err)
3707                 goto fail;
3708         err = btrfs_set_inode_index(dir, &index);
3709         if (err)
3710                 goto fail;
3711
3712         trans = btrfs_start_transaction(root, 1);
3713
3714         btrfs_set_trans_block_group(trans, dir);
3715         atomic_inc(&inode->i_count);
3716
3717         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3718
3719         if (err)
3720                 drop_inode = 1;
3721
3722         dir->i_sb->s_dirt = 1;
3723         btrfs_update_inode_block_group(trans, dir);
3724         err = btrfs_update_inode(trans, root, inode);
3725
3726         if (err)
3727                 drop_inode = 1;
3728
3729         nr = trans->blocks_used;
3730         btrfs_end_transaction_throttle(trans, root);
3731 fail:
3732         if (drop_inode) {
3733                 inode_dec_link_count(inode);
3734                 iput(inode);
3735         }
3736         btrfs_btree_balance_dirty(root, nr);
3737         return err;
3738 }
3739
3740 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3741 {
3742         struct inode *inode = NULL;
3743         struct btrfs_trans_handle *trans;
3744         struct btrfs_root *root = BTRFS_I(dir)->root;
3745         int err = 0;
3746         int drop_on_err = 0;
3747         u64 objectid = 0;
3748         u64 index = 0;
3749         unsigned long nr = 1;
3750
3751         err = btrfs_check_metadata_free_space(root);
3752         if (err)
3753                 goto out_unlock;
3754
3755         trans = btrfs_start_transaction(root, 1);
3756         btrfs_set_trans_block_group(trans, dir);
3757
3758         if (IS_ERR(trans)) {
3759                 err = PTR_ERR(trans);
3760                 goto out_unlock;
3761         }
3762
3763         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3764         if (err) {
3765                 err = -ENOSPC;
3766                 goto out_unlock;
3767         }
3768
3769         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3770                                 dentry->d_name.len,
3771                                 dentry->d_parent->d_inode->i_ino, objectid,
3772                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3773                                 &index);
3774         if (IS_ERR(inode)) {
3775                 err = PTR_ERR(inode);
3776                 goto out_fail;
3777         }
3778
3779         drop_on_err = 1;
3780
3781         err = btrfs_init_inode_security(inode, dir);
3782         if (err)
3783                 goto out_fail;
3784
3785         inode->i_op = &btrfs_dir_inode_operations;
3786         inode->i_fop = &btrfs_dir_file_operations;
3787         btrfs_set_trans_block_group(trans, inode);
3788
3789         btrfs_i_size_write(inode, 0);
3790         err = btrfs_update_inode(trans, root, inode);
3791         if (err)
3792                 goto out_fail;
3793
3794         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3795                                  inode, dentry->d_name.name,
3796                                  dentry->d_name.len, 0, index);
3797         if (err)
3798                 goto out_fail;
3799
3800         d_instantiate(dentry, inode);
3801         drop_on_err = 0;
3802         dir->i_sb->s_dirt = 1;
3803         btrfs_update_inode_block_group(trans, inode);
3804         btrfs_update_inode_block_group(trans, dir);
3805
3806 out_fail:
3807         nr = trans->blocks_used;
3808         btrfs_end_transaction_throttle(trans, root);
3809
3810 out_unlock:
3811         if (drop_on_err)
3812                 iput(inode);
3813         btrfs_btree_balance_dirty(root, nr);
3814         return err;
3815 }
3816
3817 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3818  * and an extent that you want to insert, deal with overlap and insert
3819  * the new extent into the tree.
3820  */
3821 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3822                                 struct extent_map *existing,
3823                                 struct extent_map *em,
3824                                 u64 map_start, u64 map_len)
3825 {
3826         u64 start_diff;
3827
3828         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3829         start_diff = map_start - em->start;
3830         em->start = map_start;
3831         em->len = map_len;
3832         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3833             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3834                 em->block_start += start_diff;
3835                 em->block_len -= start_diff;
3836         }
3837         return add_extent_mapping(em_tree, em);
3838 }
3839
3840 static noinline int uncompress_inline(struct btrfs_path *path,
3841                                       struct inode *inode, struct page *page,
3842                                       size_t pg_offset, u64 extent_offset,
3843                                       struct btrfs_file_extent_item *item)
3844 {
3845         int ret;
3846         struct extent_buffer *leaf = path->nodes[0];
3847         char *tmp;
3848         size_t max_size;
3849         unsigned long inline_size;
3850         unsigned long ptr;
3851
3852         WARN_ON(pg_offset != 0);
3853         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3854         inline_size = btrfs_file_extent_inline_item_len(leaf,
3855                                         btrfs_item_nr(leaf, path->slots[0]));
3856         tmp = kmalloc(inline_size, GFP_NOFS);
3857         ptr = btrfs_file_extent_inline_start(item);
3858
3859         read_extent_buffer(leaf, tmp, ptr, inline_size);
3860
3861         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3862         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3863                                     inline_size, max_size);
3864         if (ret) {
3865                 char *kaddr = kmap_atomic(page, KM_USER0);
3866                 unsigned long copy_size = min_t(u64,
3867                                   PAGE_CACHE_SIZE - pg_offset,
3868                                   max_size - extent_offset);
3869                 memset(kaddr + pg_offset, 0, copy_size);
3870                 kunmap_atomic(kaddr, KM_USER0);
3871         }
3872         kfree(tmp);
3873         return 0;
3874 }
3875
3876 /*
3877  * a bit scary, this does extent mapping from logical file offset to the disk.
3878  * the ugly parts come from merging extents from the disk with the in-ram
3879  * representation.  This gets more complex because of the data=ordered code,
3880  * where the in-ram extents might be locked pending data=ordered completion.
3881  *
3882  * This also copies inline extents directly into the page.
3883  */
3884
3885 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3886                                     size_t pg_offset, u64 start, u64 len,
3887                                     int create)
3888 {
3889         int ret;
3890         int err = 0;
3891         u64 bytenr;
3892         u64 extent_start = 0;
3893         u64 extent_end = 0;
3894         u64 objectid = inode->i_ino;
3895         u32 found_type;
3896         struct btrfs_path *path = NULL;
3897         struct btrfs_root *root = BTRFS_I(inode)->root;
3898         struct btrfs_file_extent_item *item;
3899         struct extent_buffer *leaf;
3900         struct btrfs_key found_key;
3901         struct extent_map *em = NULL;
3902         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3903         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3904         struct btrfs_trans_handle *trans = NULL;
3905         int compressed;
3906
3907 again:
3908         spin_lock(&em_tree->lock);
3909         em = lookup_extent_mapping(em_tree, start, len);
3910         if (em)
3911                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3912         spin_unlock(&em_tree->lock);
3913
3914         if (em) {
3915                 if (em->start > start || em->start + em->len <= start)
3916                         free_extent_map(em);
3917                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3918                         free_extent_map(em);
3919                 else
3920                         goto out;
3921         }
3922         em = alloc_extent_map(GFP_NOFS);
3923         if (!em) {
3924                 err = -ENOMEM;
3925                 goto out;
3926         }
3927         em->bdev = root->fs_info->fs_devices->latest_bdev;
3928         em->start = EXTENT_MAP_HOLE;
3929         em->orig_start = EXTENT_MAP_HOLE;
3930         em->len = (u64)-1;
3931         em->block_len = (u64)-1;
3932
3933         if (!path) {
3934                 path = btrfs_alloc_path();
3935                 BUG_ON(!path);
3936         }
3937
3938         ret = btrfs_lookup_file_extent(trans, root, path,
3939                                        objectid, start, trans != NULL);
3940         if (ret < 0) {
3941                 err = ret;
3942                 goto out;
3943         }
3944
3945         if (ret != 0) {
3946                 if (path->slots[0] == 0)
3947                         goto not_found;
3948                 path->slots[0]--;
3949         }
3950
3951         leaf = path->nodes[0];
3952         item = btrfs_item_ptr(leaf, path->slots[0],
3953                               struct btrfs_file_extent_item);
3954         /* are we inside the extent that was found? */
3955         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3956         found_type = btrfs_key_type(&found_key);
3957         if (found_key.objectid != objectid ||
3958             found_type != BTRFS_EXTENT_DATA_KEY) {
3959                 goto not_found;
3960         }
3961
3962         found_type = btrfs_file_extent_type(leaf, item);
3963         extent_start = found_key.offset;
3964         compressed = btrfs_file_extent_compression(leaf, item);
3965         if (found_type == BTRFS_FILE_EXTENT_REG ||
3966             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3967                 extent_end = extent_start +
3968                        btrfs_file_extent_num_bytes(leaf, item);
3969         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3970                 size_t size;
3971                 size = btrfs_file_extent_inline_len(leaf, item);
3972                 extent_end = (extent_start + size + root->sectorsize - 1) &
3973                         ~((u64)root->sectorsize - 1);
3974         }
3975
3976         if (start >= extent_end) {
3977                 path->slots[0]++;
3978                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3979                         ret = btrfs_next_leaf(root, path);
3980                         if (ret < 0) {
3981                                 err = ret;
3982                                 goto out;
3983                         }
3984                         if (ret > 0)
3985                                 goto not_found;
3986                         leaf = path->nodes[0];
3987                 }
3988                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3989                 if (found_key.objectid != objectid ||
3990                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3991                         goto not_found;
3992                 if (start + len <= found_key.offset)
3993                         goto not_found;
3994                 em->start = start;
3995                 em->len = found_key.offset - start;
3996                 goto not_found_em;
3997         }
3998
3999         if (found_type == BTRFS_FILE_EXTENT_REG ||
4000             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4001                 em->start = extent_start;
4002                 em->len = extent_end - extent_start;
4003                 em->orig_start = extent_start -
4004                                  btrfs_file_extent_offset(leaf, item);
4005                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4006                 if (bytenr == 0) {
4007                         em->block_start = EXTENT_MAP_HOLE;
4008                         goto insert;
4009                 }
4010                 if (compressed) {
4011                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4012                         em->block_start = bytenr;
4013                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4014                                                                          item);
4015                 } else {
4016                         bytenr += btrfs_file_extent_offset(leaf, item);
4017                         em->block_start = bytenr;
4018                         em->block_len = em->len;
4019                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4020                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4021                 }
4022                 goto insert;
4023         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4024                 unsigned long ptr;
4025                 char *map;
4026                 size_t size;
4027                 size_t extent_offset;
4028                 size_t copy_size;
4029
4030                 em->block_start = EXTENT_MAP_INLINE;
4031                 if (!page || create) {
4032                         em->start = extent_start;
4033                         em->len = extent_end - extent_start;
4034                         goto out;
4035                 }
4036
4037                 size = btrfs_file_extent_inline_len(leaf, item);
4038                 extent_offset = page_offset(page) + pg_offset - extent_start;
4039                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4040                                 size - extent_offset);
4041                 em->start = extent_start + extent_offset;
4042                 em->len = (copy_size + root->sectorsize - 1) &
4043                         ~((u64)root->sectorsize - 1);
4044                 em->orig_start = EXTENT_MAP_INLINE;
4045                 if (compressed)
4046                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4047                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4048                 if (create == 0 && !PageUptodate(page)) {
4049                         if (btrfs_file_extent_compression(leaf, item) ==
4050                             BTRFS_COMPRESS_ZLIB) {
4051                                 ret = uncompress_inline(path, inode, page,
4052                                                         pg_offset,
4053                                                         extent_offset, item);
4054                                 BUG_ON(ret);
4055                         } else {
4056                                 map = kmap(page);
4057                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4058                                                    copy_size);
4059                                 kunmap(page);
4060                         }
4061                         flush_dcache_page(page);
4062                 } else if (create && PageUptodate(page)) {
4063                         if (!trans) {
4064                                 kunmap(page);
4065                                 free_extent_map(em);
4066                                 em = NULL;
4067                                 btrfs_release_path(root, path);
4068                                 trans = btrfs_join_transaction(root, 1);
4069                                 goto again;
4070                         }
4071                         map = kmap(page);
4072                         write_extent_buffer(leaf, map + pg_offset, ptr,
4073                                             copy_size);
4074                         kunmap(page);
4075                         btrfs_mark_buffer_dirty(leaf);
4076                 }
4077                 set_extent_uptodate(io_tree, em->start,
4078                                     extent_map_end(em) - 1, GFP_NOFS);
4079                 goto insert;
4080         } else {
4081                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4082                 WARN_ON(1);
4083         }
4084 not_found:
4085         em->start = start;
4086         em->len = len;
4087 not_found_em:
4088         em->block_start = EXTENT_MAP_HOLE;
4089         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4090 insert:
4091         btrfs_release_path(root, path);
4092         if (em->start > start || extent_map_end(em) <= start) {
4093                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4094                        "[%llu %llu]\n", (unsigned long long)em->start,
4095                        (unsigned long long)em->len,
4096                        (unsigned long long)start,
4097                        (unsigned long long)len);
4098                 err = -EIO;
4099                 goto out;
4100         }
4101
4102         err = 0;
4103         spin_lock(&em_tree->lock);
4104         ret = add_extent_mapping(em_tree, em);
4105         /* it is possible that someone inserted the extent into the tree
4106          * while we had the lock dropped.  It is also possible that
4107          * an overlapping map exists in the tree
4108          */
4109         if (ret == -EEXIST) {
4110                 struct extent_map *existing;
4111
4112                 ret = 0;
4113
4114                 existing = lookup_extent_mapping(em_tree, start, len);
4115                 if (existing && (existing->start > start ||
4116                     existing->start + existing->len <= start)) {
4117                         free_extent_map(existing);
4118                         existing = NULL;
4119                 }
4120                 if (!existing) {
4121                         existing = lookup_extent_mapping(em_tree, em->start,
4122                                                          em->len);
4123                         if (existing) {
4124                                 err = merge_extent_mapping(em_tree, existing,
4125                                                            em, start,
4126                                                            root->sectorsize);
4127                                 free_extent_map(existing);
4128                                 if (err) {
4129                                         free_extent_map(em);
4130                                         em = NULL;
4131                                 }
4132                         } else {
4133                                 err = -EIO;
4134                                 free_extent_map(em);
4135                                 em = NULL;
4136                         }
4137                 } else {
4138                         free_extent_map(em);
4139                         em = existing;
4140                         err = 0;
4141                 }
4142         }
4143         spin_unlock(&em_tree->lock);
4144 out:
4145         if (path)
4146                 btrfs_free_path(path);
4147         if (trans) {
4148                 ret = btrfs_end_transaction(trans, root);
4149                 if (!err)
4150                         err = ret;
4151         }
4152         if (err) {
4153                 free_extent_map(em);
4154                 WARN_ON(1);
4155                 return ERR_PTR(err);
4156         }
4157         return em;
4158 }
4159
4160 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4161                         const struct iovec *iov, loff_t offset,
4162                         unsigned long nr_segs)
4163 {
4164         return -EINVAL;
4165 }
4166
4167 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4168                 __u64 start, __u64 len)
4169 {
4170         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4171 }
4172
4173 int btrfs_readpage(struct file *file, struct page *page)
4174 {
4175         struct extent_io_tree *tree;
4176         tree = &BTRFS_I(page->mapping->host)->io_tree;
4177         return extent_read_full_page(tree, page, btrfs_get_extent);
4178 }
4179
4180 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4181 {
4182         struct extent_io_tree *tree;
4183
4184
4185         if (current->flags & PF_MEMALLOC) {
4186                 redirty_page_for_writepage(wbc, page);
4187                 unlock_page(page);
4188                 return 0;
4189         }
4190         tree = &BTRFS_I(page->mapping->host)->io_tree;
4191         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4192 }
4193
4194 int btrfs_writepages(struct address_space *mapping,
4195                      struct writeback_control *wbc)
4196 {
4197         struct extent_io_tree *tree;
4198
4199         tree = &BTRFS_I(mapping->host)->io_tree;
4200         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4201 }
4202
4203 static int
4204 btrfs_readpages(struct file *file, struct address_space *mapping,
4205                 struct list_head *pages, unsigned nr_pages)
4206 {
4207         struct extent_io_tree *tree;
4208         tree = &BTRFS_I(mapping->host)->io_tree;
4209         return extent_readpages(tree, mapping, pages, nr_pages,
4210                                 btrfs_get_extent);
4211 }
4212 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4213 {
4214         struct extent_io_tree *tree;
4215         struct extent_map_tree *map;
4216         int ret;
4217
4218         tree = &BTRFS_I(page->mapping->host)->io_tree;
4219         map = &BTRFS_I(page->mapping->host)->extent_tree;
4220         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4221         if (ret == 1) {
4222                 ClearPagePrivate(page);
4223                 set_page_private(page, 0);
4224                 page_cache_release(page);
4225         }
4226         return ret;
4227 }
4228
4229 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4230 {
4231         if (PageWriteback(page) || PageDirty(page))
4232                 return 0;
4233         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4234 }
4235
4236 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4237 {
4238         struct extent_io_tree *tree;
4239         struct btrfs_ordered_extent *ordered;
4240         u64 page_start = page_offset(page);
4241         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4242
4243         wait_on_page_writeback(page);
4244         tree = &BTRFS_I(page->mapping->host)->io_tree;
4245         if (offset) {
4246                 btrfs_releasepage(page, GFP_NOFS);
4247                 return;
4248         }
4249
4250         lock_extent(tree, page_start, page_end, GFP_NOFS);
4251         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4252                                            page_offset(page));
4253         if (ordered) {
4254                 /*
4255                  * IO on this page will never be started, so we need
4256                  * to account for any ordered extents now
4257                  */
4258                 clear_extent_bit(tree, page_start, page_end,
4259                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4260                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4261                 btrfs_finish_ordered_io(page->mapping->host,
4262                                         page_start, page_end);
4263                 btrfs_put_ordered_extent(ordered);
4264                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4265         }
4266         clear_extent_bit(tree, page_start, page_end,
4267                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4268                  EXTENT_ORDERED,
4269                  1, 1, GFP_NOFS);
4270         __btrfs_releasepage(page, GFP_NOFS);
4271
4272         ClearPageChecked(page);
4273         if (PagePrivate(page)) {
4274                 ClearPagePrivate(page);
4275                 set_page_private(page, 0);
4276                 page_cache_release(page);
4277         }
4278 }
4279
4280 /*
4281  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4282  * called from a page fault handler when a page is first dirtied. Hence we must
4283  * be careful to check for EOF conditions here. We set the page up correctly
4284  * for a written page which means we get ENOSPC checking when writing into
4285  * holes and correct delalloc and unwritten extent mapping on filesystems that
4286  * support these features.
4287  *
4288  * We are not allowed to take the i_mutex here so we have to play games to
4289  * protect against truncate races as the page could now be beyond EOF.  Because
4290  * vmtruncate() writes the inode size before removing pages, once we have the
4291  * page lock we can determine safely if the page is beyond EOF. If it is not
4292  * beyond EOF, then the page is guaranteed safe against truncation until we
4293  * unlock the page.
4294  */
4295 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4296 {
4297         struct page *page = vmf->page;
4298         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4299         struct btrfs_root *root = BTRFS_I(inode)->root;
4300         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4301         struct btrfs_ordered_extent *ordered;
4302         char *kaddr;
4303         unsigned long zero_start;
4304         loff_t size;
4305         int ret;
4306         u64 page_start;
4307         u64 page_end;
4308
4309         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4310         if (ret)
4311                 goto out;
4312
4313         ret = -EINVAL;
4314 again:
4315         lock_page(page);
4316         size = i_size_read(inode);
4317         page_start = page_offset(page);
4318         page_end = page_start + PAGE_CACHE_SIZE - 1;
4319
4320         if ((page->mapping != inode->i_mapping) ||
4321             (page_start >= size)) {
4322                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4323                 /* page got truncated out from underneath us */
4324                 goto out_unlock;
4325         }
4326         wait_on_page_writeback(page);
4327
4328         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4329         set_page_extent_mapped(page);
4330
4331         /*
4332          * we can't set the delalloc bits if there are pending ordered
4333          * extents.  Drop our locks and wait for them to finish
4334          */
4335         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4336         if (ordered) {
4337                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4338                 unlock_page(page);
4339                 btrfs_start_ordered_extent(inode, ordered, 1);
4340                 btrfs_put_ordered_extent(ordered);
4341                 goto again;
4342         }
4343
4344         btrfs_set_extent_delalloc(inode, page_start, page_end);
4345         ret = 0;
4346
4347         /* page is wholly or partially inside EOF */
4348         if (page_start + PAGE_CACHE_SIZE > size)
4349                 zero_start = size & ~PAGE_CACHE_MASK;
4350         else
4351                 zero_start = PAGE_CACHE_SIZE;
4352
4353         if (zero_start != PAGE_CACHE_SIZE) {
4354                 kaddr = kmap(page);
4355                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4356                 flush_dcache_page(page);
4357                 kunmap(page);
4358         }
4359         ClearPageChecked(page);
4360         set_page_dirty(page);
4361         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4362
4363 out_unlock:
4364         unlock_page(page);
4365 out:
4366         if (ret)
4367                 ret = VM_FAULT_SIGBUS;
4368         return ret;
4369 }
4370
4371 static void btrfs_truncate(struct inode *inode)
4372 {
4373         struct btrfs_root *root = BTRFS_I(inode)->root;
4374         int ret;
4375         struct btrfs_trans_handle *trans;
4376         unsigned long nr;
4377         u64 mask = root->sectorsize - 1;
4378
4379         if (!S_ISREG(inode->i_mode))
4380                 return;
4381         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4382                 return;
4383
4384         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4385         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4386
4387         trans = btrfs_start_transaction(root, 1);
4388         btrfs_set_trans_block_group(trans, inode);
4389         btrfs_i_size_write(inode, inode->i_size);
4390
4391         ret = btrfs_orphan_add(trans, inode);
4392         if (ret)
4393                 goto out;
4394         /* FIXME, add redo link to tree so we don't leak on crash */
4395         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4396                                       BTRFS_EXTENT_DATA_KEY);
4397         btrfs_update_inode(trans, root, inode);
4398
4399         ret = btrfs_orphan_del(trans, inode);
4400         BUG_ON(ret);
4401
4402 out:
4403         nr = trans->blocks_used;
4404         ret = btrfs_end_transaction_throttle(trans, root);
4405         BUG_ON(ret);
4406         btrfs_btree_balance_dirty(root, nr);
4407 }
4408
4409 /*
4410  * create a new subvolume directory/inode (helper for the ioctl).
4411  */
4412 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4413                              struct btrfs_root *new_root, struct dentry *dentry,
4414                              u64 new_dirid, u64 alloc_hint)
4415 {
4416         struct inode *inode;
4417         int error;
4418         u64 index = 0;
4419
4420         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4421                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4422         if (IS_ERR(inode))
4423                 return PTR_ERR(inode);
4424         inode->i_op = &btrfs_dir_inode_operations;
4425         inode->i_fop = &btrfs_dir_file_operations;
4426
4427         inode->i_nlink = 1;
4428         btrfs_i_size_write(inode, 0);
4429
4430         error = btrfs_update_inode(trans, new_root, inode);
4431         if (error)
4432                 return error;
4433
4434         d_instantiate(dentry, inode);
4435         return 0;
4436 }
4437
4438 /* helper function for file defrag and space balancing.  This
4439  * forces readahead on a given range of bytes in an inode
4440  */
4441 unsigned long btrfs_force_ra(struct address_space *mapping,
4442                               struct file_ra_state *ra, struct file *file,
4443                               pgoff_t offset, pgoff_t last_index)
4444 {
4445         pgoff_t req_size = last_index - offset + 1;
4446
4447         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4448         return offset + req_size;
4449 }
4450
4451 struct inode *btrfs_alloc_inode(struct super_block *sb)
4452 {
4453         struct btrfs_inode *ei;
4454
4455         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4456         if (!ei)
4457                 return NULL;
4458         ei->last_trans = 0;
4459         ei->logged_trans = 0;
4460         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4461         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4462         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4463         INIT_LIST_HEAD(&ei->i_orphan);
4464         return &ei->vfs_inode;
4465 }
4466
4467 void btrfs_destroy_inode(struct inode *inode)
4468 {
4469         struct btrfs_ordered_extent *ordered;
4470         WARN_ON(!list_empty(&inode->i_dentry));
4471         WARN_ON(inode->i_data.nrpages);
4472
4473         if (BTRFS_I(inode)->i_acl &&
4474             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4475                 posix_acl_release(BTRFS_I(inode)->i_acl);
4476         if (BTRFS_I(inode)->i_default_acl &&
4477             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4478                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4479
4480         spin_lock(&BTRFS_I(inode)->root->list_lock);
4481         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4482                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4483                        " list\n", inode->i_ino);
4484                 dump_stack();
4485         }
4486         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4487
4488         while (1) {
4489                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4490                 if (!ordered)
4491                         break;
4492                 else {
4493                         printk(KERN_ERR "btrfs found ordered "
4494                                "extent %llu %llu on inode cleanup\n",
4495                                (unsigned long long)ordered->file_offset,
4496                                (unsigned long long)ordered->len);
4497                         btrfs_remove_ordered_extent(inode, ordered);
4498                         btrfs_put_ordered_extent(ordered);
4499                         btrfs_put_ordered_extent(ordered);
4500                 }
4501         }
4502         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4503         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4504 }
4505
4506 static void init_once(void *foo)
4507 {
4508         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4509
4510         inode_init_once(&ei->vfs_inode);
4511 }
4512
4513 void btrfs_destroy_cachep(void)
4514 {
4515         if (btrfs_inode_cachep)
4516                 kmem_cache_destroy(btrfs_inode_cachep);
4517         if (btrfs_trans_handle_cachep)
4518                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4519         if (btrfs_transaction_cachep)
4520                 kmem_cache_destroy(btrfs_transaction_cachep);
4521         if (btrfs_bit_radix_cachep)
4522                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4523         if (btrfs_path_cachep)
4524                 kmem_cache_destroy(btrfs_path_cachep);
4525 }
4526
4527 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4528                                        unsigned long extra_flags,
4529                                        void (*ctor)(void *))
4530 {
4531         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4532                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4533 }
4534
4535 int btrfs_init_cachep(void)
4536 {
4537         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4538                                           sizeof(struct btrfs_inode),
4539                                           0, init_once);
4540         if (!btrfs_inode_cachep)
4541                 goto fail;
4542         btrfs_trans_handle_cachep =
4543                         btrfs_cache_create("btrfs_trans_handle_cache",
4544                                            sizeof(struct btrfs_trans_handle),
4545                                            0, NULL);
4546         if (!btrfs_trans_handle_cachep)
4547                 goto fail;
4548         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4549                                              sizeof(struct btrfs_transaction),
4550                                              0, NULL);
4551         if (!btrfs_transaction_cachep)
4552                 goto fail;
4553         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4554                                          sizeof(struct btrfs_path),
4555                                          0, NULL);
4556         if (!btrfs_path_cachep)
4557                 goto fail;
4558         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4559                                               SLAB_DESTROY_BY_RCU, NULL);
4560         if (!btrfs_bit_radix_cachep)
4561                 goto fail;
4562         return 0;
4563 fail:
4564         btrfs_destroy_cachep();
4565         return -ENOMEM;
4566 }
4567
4568 static int btrfs_getattr(struct vfsmount *mnt,
4569                          struct dentry *dentry, struct kstat *stat)
4570 {
4571         struct inode *inode = dentry->d_inode;
4572         generic_fillattr(inode, stat);
4573         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4574         stat->blksize = PAGE_CACHE_SIZE;
4575         stat->blocks = (inode_get_bytes(inode) +
4576                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4577         return 0;
4578 }
4579
4580 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4581                            struct inode *new_dir, struct dentry *new_dentry)
4582 {
4583         struct btrfs_trans_handle *trans;
4584         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4585         struct inode *new_inode = new_dentry->d_inode;
4586         struct inode *old_inode = old_dentry->d_inode;
4587         struct timespec ctime = CURRENT_TIME;
4588         u64 index = 0;
4589         int ret;
4590
4591         /* we're not allowed to rename between subvolumes */
4592         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4593             BTRFS_I(new_dir)->root->root_key.objectid)
4594                 return -EXDEV;
4595
4596         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4597             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4598                 return -ENOTEMPTY;
4599         }
4600
4601         /* to rename a snapshot or subvolume, we need to juggle the
4602          * backrefs.  This isn't coded yet
4603          */
4604         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4605                 return -EXDEV;
4606
4607         ret = btrfs_check_metadata_free_space(root);
4608         if (ret)
4609                 goto out_unlock;
4610
4611         trans = btrfs_start_transaction(root, 1);
4612
4613         btrfs_set_trans_block_group(trans, new_dir);
4614
4615         btrfs_inc_nlink(old_dentry->d_inode);
4616         old_dir->i_ctime = old_dir->i_mtime = ctime;
4617         new_dir->i_ctime = new_dir->i_mtime = ctime;
4618         old_inode->i_ctime = ctime;
4619
4620         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4621                                  old_dentry->d_name.name,
4622                                  old_dentry->d_name.len);
4623         if (ret)
4624                 goto out_fail;
4625
4626         if (new_inode) {
4627                 new_inode->i_ctime = CURRENT_TIME;
4628                 ret = btrfs_unlink_inode(trans, root, new_dir,
4629                                          new_dentry->d_inode,
4630                                          new_dentry->d_name.name,
4631                                          new_dentry->d_name.len);
4632                 if (ret)
4633                         goto out_fail;
4634                 if (new_inode->i_nlink == 0) {
4635                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4636                         if (ret)
4637                                 goto out_fail;
4638                 }
4639
4640         }
4641         ret = btrfs_set_inode_index(new_dir, &index);
4642         if (ret)
4643                 goto out_fail;
4644
4645         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4646                              old_inode, new_dentry->d_name.name,
4647                              new_dentry->d_name.len, 1, index);
4648         if (ret)
4649                 goto out_fail;
4650
4651 out_fail:
4652         btrfs_end_transaction_throttle(trans, root);
4653 out_unlock:
4654         return ret;
4655 }
4656
4657 /*
4658  * some fairly slow code that needs optimization. This walks the list
4659  * of all the inodes with pending delalloc and forces them to disk.
4660  */
4661 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4662 {
4663         struct list_head *head = &root->fs_info->delalloc_inodes;
4664         struct btrfs_inode *binode;
4665         struct inode *inode;
4666
4667         if (root->fs_info->sb->s_flags & MS_RDONLY)
4668                 return -EROFS;
4669
4670         spin_lock(&root->fs_info->delalloc_lock);
4671         while (!list_empty(head)) {
4672                 binode = list_entry(head->next, struct btrfs_inode,
4673                                     delalloc_inodes);
4674                 inode = igrab(&binode->vfs_inode);
4675                 if (!inode)
4676                         list_del_init(&binode->delalloc_inodes);
4677                 spin_unlock(&root->fs_info->delalloc_lock);
4678                 if (inode) {
4679                         filemap_flush(inode->i_mapping);
4680                         iput(inode);
4681                 }
4682                 cond_resched();
4683                 spin_lock(&root->fs_info->delalloc_lock);
4684         }
4685         spin_unlock(&root->fs_info->delalloc_lock);
4686
4687         /* the filemap_flush will queue IO into the worker threads, but
4688          * we have to make sure the IO is actually started and that
4689          * ordered extents get created before we return
4690          */
4691         atomic_inc(&root->fs_info->async_submit_draining);
4692         while (atomic_read(&root->fs_info->nr_async_submits) ||
4693               atomic_read(&root->fs_info->async_delalloc_pages)) {
4694                 wait_event(root->fs_info->async_submit_wait,
4695                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4696                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4697         }
4698         atomic_dec(&root->fs_info->async_submit_draining);
4699         return 0;
4700 }
4701
4702 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4703                          const char *symname)
4704 {
4705         struct btrfs_trans_handle *trans;
4706         struct btrfs_root *root = BTRFS_I(dir)->root;
4707         struct btrfs_path *path;
4708         struct btrfs_key key;
4709         struct inode *inode = NULL;
4710         int err;
4711         int drop_inode = 0;
4712         u64 objectid;
4713         u64 index = 0 ;
4714         int name_len;
4715         int datasize;
4716         unsigned long ptr;
4717         struct btrfs_file_extent_item *ei;
4718         struct extent_buffer *leaf;
4719         unsigned long nr = 0;
4720
4721         name_len = strlen(symname) + 1;
4722         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4723                 return -ENAMETOOLONG;
4724
4725         err = btrfs_check_metadata_free_space(root);
4726         if (err)
4727                 goto out_fail;
4728
4729         trans = btrfs_start_transaction(root, 1);
4730         btrfs_set_trans_block_group(trans, dir);
4731
4732         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4733         if (err) {
4734                 err = -ENOSPC;
4735                 goto out_unlock;
4736         }
4737
4738         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4739                                 dentry->d_name.len,
4740                                 dentry->d_parent->d_inode->i_ino, objectid,
4741                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4742                                 &index);
4743         err = PTR_ERR(inode);
4744         if (IS_ERR(inode))
4745                 goto out_unlock;
4746
4747         err = btrfs_init_inode_security(inode, dir);
4748         if (err) {
4749                 drop_inode = 1;
4750                 goto out_unlock;
4751         }
4752
4753         btrfs_set_trans_block_group(trans, inode);
4754         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4755         if (err)
4756                 drop_inode = 1;
4757         else {
4758                 inode->i_mapping->a_ops = &btrfs_aops;
4759                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4760                 inode->i_fop = &btrfs_file_operations;
4761                 inode->i_op = &btrfs_file_inode_operations;
4762                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4763         }
4764         dir->i_sb->s_dirt = 1;
4765         btrfs_update_inode_block_group(trans, inode);
4766         btrfs_update_inode_block_group(trans, dir);
4767         if (drop_inode)
4768                 goto out_unlock;
4769
4770         path = btrfs_alloc_path();
4771         BUG_ON(!path);
4772         key.objectid = inode->i_ino;
4773         key.offset = 0;
4774         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4775         datasize = btrfs_file_extent_calc_inline_size(name_len);
4776         err = btrfs_insert_empty_item(trans, root, path, &key,
4777                                       datasize);
4778         if (err) {
4779                 drop_inode = 1;
4780                 goto out_unlock;
4781         }
4782         leaf = path->nodes[0];
4783         ei = btrfs_item_ptr(leaf, path->slots[0],
4784                             struct btrfs_file_extent_item);
4785         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4786         btrfs_set_file_extent_type(leaf, ei,
4787                                    BTRFS_FILE_EXTENT_INLINE);
4788         btrfs_set_file_extent_encryption(leaf, ei, 0);
4789         btrfs_set_file_extent_compression(leaf, ei, 0);
4790         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4791         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4792
4793         ptr = btrfs_file_extent_inline_start(ei);
4794         write_extent_buffer(leaf, symname, ptr, name_len);
4795         btrfs_mark_buffer_dirty(leaf);
4796         btrfs_free_path(path);
4797
4798         inode->i_op = &btrfs_symlink_inode_operations;
4799         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4800         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4801         inode_set_bytes(inode, name_len);
4802         btrfs_i_size_write(inode, name_len - 1);
4803         err = btrfs_update_inode(trans, root, inode);
4804         if (err)
4805                 drop_inode = 1;
4806
4807 out_unlock:
4808         nr = trans->blocks_used;
4809         btrfs_end_transaction_throttle(trans, root);
4810 out_fail:
4811         if (drop_inode) {
4812                 inode_dec_link_count(inode);
4813                 iput(inode);
4814         }
4815         btrfs_btree_balance_dirty(root, nr);
4816         return err;
4817 }
4818
4819 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4820                                u64 alloc_hint, int mode)
4821 {
4822         struct btrfs_trans_handle *trans;
4823         struct btrfs_root *root = BTRFS_I(inode)->root;
4824         struct btrfs_key ins;
4825         u64 alloc_size;
4826         u64 cur_offset = start;
4827         u64 num_bytes = end - start;
4828         int ret = 0;
4829
4830         trans = btrfs_join_transaction(root, 1);
4831         BUG_ON(!trans);
4832         btrfs_set_trans_block_group(trans, inode);
4833
4834         while (num_bytes > 0) {
4835                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4836                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4837                                            root->sectorsize, 0, alloc_hint,
4838                                            (u64)-1, &ins, 1);
4839                 if (ret) {
4840                         WARN_ON(1);
4841                         goto out;
4842                 }
4843                 ret = insert_reserved_file_extent(trans, inode,
4844                                                   cur_offset, ins.objectid,
4845                                                   ins.offset, ins.offset,
4846                                                   ins.offset, 0, 0, 0,
4847                                                   BTRFS_FILE_EXTENT_PREALLOC);
4848                 BUG_ON(ret);
4849                 num_bytes -= ins.offset;
4850                 cur_offset += ins.offset;
4851                 alloc_hint = ins.objectid + ins.offset;
4852         }
4853 out:
4854         if (cur_offset > start) {
4855                 inode->i_ctime = CURRENT_TIME;
4856                 btrfs_set_flag(inode, PREALLOC);
4857                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4858                     cur_offset > i_size_read(inode))
4859                         btrfs_i_size_write(inode, cur_offset);
4860                 ret = btrfs_update_inode(trans, root, inode);
4861                 BUG_ON(ret);
4862         }
4863
4864         btrfs_end_transaction(trans, root);
4865         return ret;
4866 }
4867
4868 static long btrfs_fallocate(struct inode *inode, int mode,
4869                             loff_t offset, loff_t len)
4870 {
4871         u64 cur_offset;
4872         u64 last_byte;
4873         u64 alloc_start;
4874         u64 alloc_end;
4875         u64 alloc_hint = 0;
4876         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4877         struct extent_map *em;
4878         int ret;
4879
4880         alloc_start = offset & ~mask;
4881         alloc_end =  (offset + len + mask) & ~mask;
4882
4883         mutex_lock(&inode->i_mutex);
4884         if (alloc_start > inode->i_size) {
4885                 ret = btrfs_cont_expand(inode, alloc_start);
4886                 if (ret)
4887                         goto out;
4888         }
4889
4890         while (1) {
4891                 struct btrfs_ordered_extent *ordered;
4892                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4893                             alloc_end - 1, GFP_NOFS);
4894                 ordered = btrfs_lookup_first_ordered_extent(inode,
4895                                                             alloc_end - 1);
4896                 if (ordered &&
4897                     ordered->file_offset + ordered->len > alloc_start &&
4898                     ordered->file_offset < alloc_end) {
4899                         btrfs_put_ordered_extent(ordered);
4900                         unlock_extent(&BTRFS_I(inode)->io_tree,
4901                                       alloc_start, alloc_end - 1, GFP_NOFS);
4902                         btrfs_wait_ordered_range(inode, alloc_start,
4903                                                  alloc_end - alloc_start);
4904                 } else {
4905                         if (ordered)
4906                                 btrfs_put_ordered_extent(ordered);
4907                         break;
4908                 }
4909         }
4910
4911         cur_offset = alloc_start;
4912         while (1) {
4913                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4914                                       alloc_end - cur_offset, 0);
4915                 BUG_ON(IS_ERR(em) || !em);
4916                 last_byte = min(extent_map_end(em), alloc_end);
4917                 last_byte = (last_byte + mask) & ~mask;
4918                 if (em->block_start == EXTENT_MAP_HOLE) {
4919                         ret = prealloc_file_range(inode, cur_offset,
4920                                         last_byte, alloc_hint, mode);
4921                         if (ret < 0) {
4922                                 free_extent_map(em);
4923                                 break;
4924                         }
4925                 }
4926                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4927                         alloc_hint = em->block_start;
4928                 free_extent_map(em);
4929
4930                 cur_offset = last_byte;
4931                 if (cur_offset >= alloc_end) {
4932                         ret = 0;
4933                         break;
4934                 }
4935         }
4936         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4937                       GFP_NOFS);
4938 out:
4939         mutex_unlock(&inode->i_mutex);
4940         return ret;
4941 }
4942
4943 static int btrfs_set_page_dirty(struct page *page)
4944 {
4945         return __set_page_dirty_nobuffers(page);
4946 }
4947
4948 static int btrfs_permission(struct inode *inode, int mask)
4949 {
4950         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4951                 return -EACCES;
4952         return generic_permission(inode, mask, btrfs_check_acl);
4953 }
4954
4955 static struct inode_operations btrfs_dir_inode_operations = {
4956         .getattr        = btrfs_getattr,
4957         .lookup         = btrfs_lookup,
4958         .create         = btrfs_create,
4959         .unlink         = btrfs_unlink,
4960         .link           = btrfs_link,
4961         .mkdir          = btrfs_mkdir,
4962         .rmdir          = btrfs_rmdir,
4963         .rename         = btrfs_rename,
4964         .symlink        = btrfs_symlink,
4965         .setattr        = btrfs_setattr,
4966         .mknod          = btrfs_mknod,
4967         .setxattr       = btrfs_setxattr,
4968         .getxattr       = btrfs_getxattr,
4969         .listxattr      = btrfs_listxattr,
4970         .removexattr    = btrfs_removexattr,
4971         .permission     = btrfs_permission,
4972 };
4973 static struct inode_operations btrfs_dir_ro_inode_operations = {
4974         .lookup         = btrfs_lookup,
4975         .permission     = btrfs_permission,
4976 };
4977 static struct file_operations btrfs_dir_file_operations = {
4978         .llseek         = generic_file_llseek,
4979         .read           = generic_read_dir,
4980         .readdir        = btrfs_real_readdir,
4981         .unlocked_ioctl = btrfs_ioctl,
4982 #ifdef CONFIG_COMPAT
4983         .compat_ioctl   = btrfs_ioctl,
4984 #endif
4985         .release        = btrfs_release_file,
4986         .fsync          = btrfs_sync_file,
4987 };
4988
4989 static struct extent_io_ops btrfs_extent_io_ops = {
4990         .fill_delalloc = run_delalloc_range,
4991         .submit_bio_hook = btrfs_submit_bio_hook,
4992         .merge_bio_hook = btrfs_merge_bio_hook,
4993         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4994         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4995         .writepage_start_hook = btrfs_writepage_start_hook,
4996         .readpage_io_failed_hook = btrfs_io_failed_hook,
4997         .set_bit_hook = btrfs_set_bit_hook,
4998         .clear_bit_hook = btrfs_clear_bit_hook,
4999 };
5000
5001 /*
5002  * btrfs doesn't support the bmap operation because swapfiles
5003  * use bmap to make a mapping of extents in the file.  They assume
5004  * these extents won't change over the life of the file and they
5005  * use the bmap result to do IO directly to the drive.
5006  *
5007  * the btrfs bmap call would return logical addresses that aren't
5008  * suitable for IO and they also will change frequently as COW
5009  * operations happen.  So, swapfile + btrfs == corruption.
5010  *
5011  * For now we're avoiding this by dropping bmap.
5012  */
5013 static struct address_space_operations btrfs_aops = {
5014         .readpage       = btrfs_readpage,
5015         .writepage      = btrfs_writepage,
5016         .writepages     = btrfs_writepages,
5017         .readpages      = btrfs_readpages,
5018         .sync_page      = block_sync_page,
5019         .direct_IO      = btrfs_direct_IO,
5020         .invalidatepage = btrfs_invalidatepage,
5021         .releasepage    = btrfs_releasepage,
5022         .set_page_dirty = btrfs_set_page_dirty,
5023 };
5024
5025 static struct address_space_operations btrfs_symlink_aops = {
5026         .readpage       = btrfs_readpage,
5027         .writepage      = btrfs_writepage,
5028         .invalidatepage = btrfs_invalidatepage,
5029         .releasepage    = btrfs_releasepage,
5030 };
5031
5032 static struct inode_operations btrfs_file_inode_operations = {
5033         .truncate       = btrfs_truncate,
5034         .getattr        = btrfs_getattr,
5035         .setattr        = btrfs_setattr,
5036         .setxattr       = btrfs_setxattr,
5037         .getxattr       = btrfs_getxattr,
5038         .listxattr      = btrfs_listxattr,
5039         .removexattr    = btrfs_removexattr,
5040         .permission     = btrfs_permission,
5041         .fallocate      = btrfs_fallocate,
5042         .fiemap         = btrfs_fiemap,
5043 };
5044 static struct inode_operations btrfs_special_inode_operations = {
5045         .getattr        = btrfs_getattr,
5046         .setattr        = btrfs_setattr,
5047         .permission     = btrfs_permission,
5048         .setxattr       = btrfs_setxattr,
5049         .getxattr       = btrfs_getxattr,
5050         .listxattr      = btrfs_listxattr,
5051         .removexattr    = btrfs_removexattr,
5052 };
5053 static struct inode_operations btrfs_symlink_inode_operations = {
5054         .readlink       = generic_readlink,
5055         .follow_link    = page_follow_link_light,
5056         .put_link       = page_put_link,
5057         .permission     = btrfs_permission,
5058         .setxattr       = btrfs_setxattr,
5059         .getxattr       = btrfs_getxattr,
5060         .listxattr      = btrfs_listxattr,
5061         .removexattr    = btrfs_removexattr,
5062 };