Btrfs: BUG_ON is deleted from the caller of btrfs_truncate_item & btrfs_extend_item
[sfrench/cifs-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53 #include "free-space-cache.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91                                    struct page *locked_page,
92                                    u64 start, u64 end, int *page_started,
93                                    unsigned long *nr_written, int unlock);
94
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96                                      struct inode *inode,  struct inode *dir,
97                                      const struct qstr *qstr)
98 {
99         int err;
100
101         err = btrfs_init_acl(trans, inode, dir);
102         if (!err)
103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
104         return err;
105 }
106
107 /*
108  * this does all the hard work for inserting an inline extent into
109  * the btree.  The caller should have done a btrfs_drop_extents so that
110  * no overlapping inline items exist in the btree
111  */
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113                                 struct btrfs_root *root, struct inode *inode,
114                                 u64 start, size_t size, size_t compressed_size,
115                                 int compress_type,
116                                 struct page **compressed_pages)
117 {
118         struct btrfs_key key;
119         struct btrfs_path *path;
120         struct extent_buffer *leaf;
121         struct page *page = NULL;
122         char *kaddr;
123         unsigned long ptr;
124         struct btrfs_file_extent_item *ei;
125         int err = 0;
126         int ret;
127         size_t cur_size = size;
128         size_t datasize;
129         unsigned long offset;
130
131         if (compressed_size && compressed_pages)
132                 cur_size = compressed_size;
133
134         path = btrfs_alloc_path();
135         if (!path)
136                 return -ENOMEM;
137
138         path->leave_spinning = 1;
139         btrfs_set_trans_block_group(trans, inode);
140
141         key.objectid = inode->i_ino;
142         key.offset = start;
143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
145
146         inode_add_bytes(inode, size);
147         ret = btrfs_insert_empty_item(trans, root, path, &key,
148                                       datasize);
149         BUG_ON(ret);
150         if (ret) {
151                 err = ret;
152                 goto fail;
153         }
154         leaf = path->nodes[0];
155         ei = btrfs_item_ptr(leaf, path->slots[0],
156                             struct btrfs_file_extent_item);
157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
159         btrfs_set_file_extent_encryption(leaf, ei, 0);
160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
162         ptr = btrfs_file_extent_inline_start(ei);
163
164         if (compress_type != BTRFS_COMPRESS_NONE) {
165                 struct page *cpage;
166                 int i = 0;
167                 while (compressed_size > 0) {
168                         cpage = compressed_pages[i];
169                         cur_size = min_t(unsigned long, compressed_size,
170                                        PAGE_CACHE_SIZE);
171
172                         kaddr = kmap_atomic(cpage, KM_USER0);
173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
174                         kunmap_atomic(kaddr, KM_USER0);
175
176                         i++;
177                         ptr += cur_size;
178                         compressed_size -= cur_size;
179                 }
180                 btrfs_set_file_extent_compression(leaf, ei,
181                                                   compress_type);
182         } else {
183                 page = find_get_page(inode->i_mapping,
184                                      start >> PAGE_CACHE_SHIFT);
185                 btrfs_set_file_extent_compression(leaf, ei, 0);
186                 kaddr = kmap_atomic(page, KM_USER0);
187                 offset = start & (PAGE_CACHE_SIZE - 1);
188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
189                 kunmap_atomic(kaddr, KM_USER0);
190                 page_cache_release(page);
191         }
192         btrfs_mark_buffer_dirty(leaf);
193         btrfs_free_path(path);
194
195         /*
196          * we're an inline extent, so nobody can
197          * extend the file past i_size without locking
198          * a page we already have locked.
199          *
200          * We must do any isize and inode updates
201          * before we unlock the pages.  Otherwise we
202          * could end up racing with unlink.
203          */
204         BTRFS_I(inode)->disk_i_size = inode->i_size;
205         btrfs_update_inode(trans, root, inode);
206
207         return 0;
208 fail:
209         btrfs_free_path(path);
210         return err;
211 }
212
213
214 /*
215  * conditionally insert an inline extent into the file.  This
216  * does the checks required to make sure the data is small enough
217  * to fit as an inline extent.
218  */
219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
220                                  struct btrfs_root *root,
221                                  struct inode *inode, u64 start, u64 end,
222                                  size_t compressed_size, int compress_type,
223                                  struct page **compressed_pages)
224 {
225         u64 isize = i_size_read(inode);
226         u64 actual_end = min(end + 1, isize);
227         u64 inline_len = actual_end - start;
228         u64 aligned_end = (end + root->sectorsize - 1) &
229                         ~((u64)root->sectorsize - 1);
230         u64 hint_byte;
231         u64 data_len = inline_len;
232         int ret;
233
234         if (compressed_size)
235                 data_len = compressed_size;
236
237         if (start > 0 ||
238             actual_end >= PAGE_CACHE_SIZE ||
239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240             (!compressed_size &&
241             (actual_end & (root->sectorsize - 1)) == 0) ||
242             end + 1 < isize ||
243             data_len > root->fs_info->max_inline) {
244                 return 1;
245         }
246
247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
248                                  &hint_byte, 1);
249         BUG_ON(ret);
250
251         if (isize > actual_end)
252                 inline_len = min_t(u64, isize, actual_end);
253         ret = insert_inline_extent(trans, root, inode, start,
254                                    inline_len, compressed_size,
255                                    compress_type, compressed_pages);
256         BUG_ON(ret);
257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
259         return 0;
260 }
261
262 struct async_extent {
263         u64 start;
264         u64 ram_size;
265         u64 compressed_size;
266         struct page **pages;
267         unsigned long nr_pages;
268         int compress_type;
269         struct list_head list;
270 };
271
272 struct async_cow {
273         struct inode *inode;
274         struct btrfs_root *root;
275         struct page *locked_page;
276         u64 start;
277         u64 end;
278         struct list_head extents;
279         struct btrfs_work work;
280 };
281
282 static noinline int add_async_extent(struct async_cow *cow,
283                                      u64 start, u64 ram_size,
284                                      u64 compressed_size,
285                                      struct page **pages,
286                                      unsigned long nr_pages,
287                                      int compress_type)
288 {
289         struct async_extent *async_extent;
290
291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
292         BUG_ON(!async_extent);
293         async_extent->start = start;
294         async_extent->ram_size = ram_size;
295         async_extent->compressed_size = compressed_size;
296         async_extent->pages = pages;
297         async_extent->nr_pages = nr_pages;
298         async_extent->compress_type = compress_type;
299         list_add_tail(&async_extent->list, &cow->extents);
300         return 0;
301 }
302
303 /*
304  * we create compressed extents in two phases.  The first
305  * phase compresses a range of pages that have already been
306  * locked (both pages and state bits are locked).
307  *
308  * This is done inside an ordered work queue, and the compression
309  * is spread across many cpus.  The actual IO submission is step
310  * two, and the ordered work queue takes care of making sure that
311  * happens in the same order things were put onto the queue by
312  * writepages and friends.
313  *
314  * If this code finds it can't get good compression, it puts an
315  * entry onto the work queue to write the uncompressed bytes.  This
316  * makes sure that both compressed inodes and uncompressed inodes
317  * are written in the same order that pdflush sent them down.
318  */
319 static noinline int compress_file_range(struct inode *inode,
320                                         struct page *locked_page,
321                                         u64 start, u64 end,
322                                         struct async_cow *async_cow,
323                                         int *num_added)
324 {
325         struct btrfs_root *root = BTRFS_I(inode)->root;
326         struct btrfs_trans_handle *trans;
327         u64 num_bytes;
328         u64 blocksize = root->sectorsize;
329         u64 actual_end;
330         u64 isize = i_size_read(inode);
331         int ret = 0;
332         struct page **pages = NULL;
333         unsigned long nr_pages;
334         unsigned long nr_pages_ret = 0;
335         unsigned long total_compressed = 0;
336         unsigned long total_in = 0;
337         unsigned long max_compressed = 128 * 1024;
338         unsigned long max_uncompressed = 128 * 1024;
339         int i;
340         int will_compress;
341         int compress_type = root->fs_info->compress_type;
342
343         actual_end = min_t(u64, isize, end + 1);
344 again:
345         will_compress = 0;
346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
348
349         /*
350          * we don't want to send crud past the end of i_size through
351          * compression, that's just a waste of CPU time.  So, if the
352          * end of the file is before the start of our current
353          * requested range of bytes, we bail out to the uncompressed
354          * cleanup code that can deal with all of this.
355          *
356          * It isn't really the fastest way to fix things, but this is a
357          * very uncommon corner.
358          */
359         if (actual_end <= start)
360                 goto cleanup_and_bail_uncompressed;
361
362         total_compressed = actual_end - start;
363
364         /* we want to make sure that amount of ram required to uncompress
365          * an extent is reasonable, so we limit the total size in ram
366          * of a compressed extent to 128k.  This is a crucial number
367          * because it also controls how easily we can spread reads across
368          * cpus for decompression.
369          *
370          * We also want to make sure the amount of IO required to do
371          * a random read is reasonably small, so we limit the size of
372          * a compressed extent to 128k.
373          */
374         total_compressed = min(total_compressed, max_uncompressed);
375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376         num_bytes = max(blocksize,  num_bytes);
377         total_in = 0;
378         ret = 0;
379
380         /*
381          * we do compression for mount -o compress and when the
382          * inode has not been flagged as nocompress.  This flag can
383          * change at any time if we discover bad compression ratios.
384          */
385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386             (btrfs_test_opt(root, COMPRESS) ||
387              (BTRFS_I(inode)->force_compress) ||
388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
389                 WARN_ON(pages);
390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
391                 BUG_ON(!pages);
392
393                 if (BTRFS_I(inode)->force_compress)
394                         compress_type = BTRFS_I(inode)->force_compress;
395
396                 ret = btrfs_compress_pages(compress_type,
397                                            inode->i_mapping, start,
398                                            total_compressed, pages,
399                                            nr_pages, &nr_pages_ret,
400                                            &total_in,
401                                            &total_compressed,
402                                            max_compressed);
403
404                 if (!ret) {
405                         unsigned long offset = total_compressed &
406                                 (PAGE_CACHE_SIZE - 1);
407                         struct page *page = pages[nr_pages_ret - 1];
408                         char *kaddr;
409
410                         /* zero the tail end of the last page, we might be
411                          * sending it down to disk
412                          */
413                         if (offset) {
414                                 kaddr = kmap_atomic(page, KM_USER0);
415                                 memset(kaddr + offset, 0,
416                                        PAGE_CACHE_SIZE - offset);
417                                 kunmap_atomic(kaddr, KM_USER0);
418                         }
419                         will_compress = 1;
420                 }
421         }
422         if (start == 0) {
423                 trans = btrfs_join_transaction(root, 1);
424                 BUG_ON(IS_ERR(trans));
425                 btrfs_set_trans_block_group(trans, inode);
426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427
428                 /* lets try to make an inline extent */
429                 if (ret || total_in < (actual_end - start)) {
430                         /* we didn't compress the entire range, try
431                          * to make an uncompressed inline extent.
432                          */
433                         ret = cow_file_range_inline(trans, root, inode,
434                                                     start, end, 0, 0, NULL);
435                 } else {
436                         /* try making a compressed inline extent */
437                         ret = cow_file_range_inline(trans, root, inode,
438                                                     start, end,
439                                                     total_compressed,
440                                                     compress_type, pages);
441                 }
442                 if (ret == 0) {
443                         /*
444                          * inline extent creation worked, we don't need
445                          * to create any more async work items.  Unlock
446                          * and free up our temp pages.
447                          */
448                         extent_clear_unlock_delalloc(inode,
449                              &BTRFS_I(inode)->io_tree,
450                              start, end, NULL,
451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
452                              EXTENT_CLEAR_DELALLOC |
453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
454
455                         btrfs_end_transaction(trans, root);
456                         goto free_pages_out;
457                 }
458                 btrfs_end_transaction(trans, root);
459         }
460
461         if (will_compress) {
462                 /*
463                  * we aren't doing an inline extent round the compressed size
464                  * up to a block size boundary so the allocator does sane
465                  * things
466                  */
467                 total_compressed = (total_compressed + blocksize - 1) &
468                         ~(blocksize - 1);
469
470                 /*
471                  * one last check to make sure the compression is really a
472                  * win, compare the page count read with the blocks on disk
473                  */
474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
475                         ~(PAGE_CACHE_SIZE - 1);
476                 if (total_compressed >= total_in) {
477                         will_compress = 0;
478                 } else {
479                         num_bytes = total_in;
480                 }
481         }
482         if (!will_compress && pages) {
483                 /*
484                  * the compression code ran but failed to make things smaller,
485                  * free any pages it allocated and our page pointer array
486                  */
487                 for (i = 0; i < nr_pages_ret; i++) {
488                         WARN_ON(pages[i]->mapping);
489                         page_cache_release(pages[i]);
490                 }
491                 kfree(pages);
492                 pages = NULL;
493                 total_compressed = 0;
494                 nr_pages_ret = 0;
495
496                 /* flag the file so we don't compress in the future */
497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
498                     !(BTRFS_I(inode)->force_compress)) {
499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
500                 }
501         }
502         if (will_compress) {
503                 *num_added += 1;
504
505                 /* the async work queues will take care of doing actual
506                  * allocation on disk for these compressed pages,
507                  * and will submit them to the elevator.
508                  */
509                 add_async_extent(async_cow, start, num_bytes,
510                                  total_compressed, pages, nr_pages_ret,
511                                  compress_type);
512
513                 if (start + num_bytes < end) {
514                         start += num_bytes;
515                         pages = NULL;
516                         cond_resched();
517                         goto again;
518                 }
519         } else {
520 cleanup_and_bail_uncompressed:
521                 /*
522                  * No compression, but we still need to write the pages in
523                  * the file we've been given so far.  redirty the locked
524                  * page if it corresponds to our extent and set things up
525                  * for the async work queue to run cow_file_range to do
526                  * the normal delalloc dance
527                  */
528                 if (page_offset(locked_page) >= start &&
529                     page_offset(locked_page) <= end) {
530                         __set_page_dirty_nobuffers(locked_page);
531                         /* unlocked later on in the async handlers */
532                 }
533                 add_async_extent(async_cow, start, end - start + 1,
534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
535                 *num_added += 1;
536         }
537
538 out:
539         return 0;
540
541 free_pages_out:
542         for (i = 0; i < nr_pages_ret; i++) {
543                 WARN_ON(pages[i]->mapping);
544                 page_cache_release(pages[i]);
545         }
546         kfree(pages);
547
548         goto out;
549 }
550
551 /*
552  * phase two of compressed writeback.  This is the ordered portion
553  * of the code, which only gets called in the order the work was
554  * queued.  We walk all the async extents created by compress_file_range
555  * and send them down to the disk.
556  */
557 static noinline int submit_compressed_extents(struct inode *inode,
558                                               struct async_cow *async_cow)
559 {
560         struct async_extent *async_extent;
561         u64 alloc_hint = 0;
562         struct btrfs_trans_handle *trans;
563         struct btrfs_key ins;
564         struct extent_map *em;
565         struct btrfs_root *root = BTRFS_I(inode)->root;
566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
567         struct extent_io_tree *io_tree;
568         int ret = 0;
569
570         if (list_empty(&async_cow->extents))
571                 return 0;
572
573
574         while (!list_empty(&async_cow->extents)) {
575                 async_extent = list_entry(async_cow->extents.next,
576                                           struct async_extent, list);
577                 list_del(&async_extent->list);
578
579                 io_tree = &BTRFS_I(inode)->io_tree;
580
581 retry:
582                 /* did the compression code fall back to uncompressed IO? */
583                 if (!async_extent->pages) {
584                         int page_started = 0;
585                         unsigned long nr_written = 0;
586
587                         lock_extent(io_tree, async_extent->start,
588                                          async_extent->start +
589                                          async_extent->ram_size - 1, GFP_NOFS);
590
591                         /* allocate blocks */
592                         ret = cow_file_range(inode, async_cow->locked_page,
593                                              async_extent->start,
594                                              async_extent->start +
595                                              async_extent->ram_size - 1,
596                                              &page_started, &nr_written, 0);
597
598                         /*
599                          * if page_started, cow_file_range inserted an
600                          * inline extent and took care of all the unlocking
601                          * and IO for us.  Otherwise, we need to submit
602                          * all those pages down to the drive.
603                          */
604                         if (!page_started && !ret)
605                                 extent_write_locked_range(io_tree,
606                                                   inode, async_extent->start,
607                                                   async_extent->start +
608                                                   async_extent->ram_size - 1,
609                                                   btrfs_get_extent,
610                                                   WB_SYNC_ALL);
611                         kfree(async_extent);
612                         cond_resched();
613                         continue;
614                 }
615
616                 lock_extent(io_tree, async_extent->start,
617                             async_extent->start + async_extent->ram_size - 1,
618                             GFP_NOFS);
619
620                 trans = btrfs_join_transaction(root, 1);
621                 BUG_ON(IS_ERR(trans));
622                 ret = btrfs_reserve_extent(trans, root,
623                                            async_extent->compressed_size,
624                                            async_extent->compressed_size,
625                                            0, alloc_hint,
626                                            (u64)-1, &ins, 1);
627                 btrfs_end_transaction(trans, root);
628
629                 if (ret) {
630                         int i;
631                         for (i = 0; i < async_extent->nr_pages; i++) {
632                                 WARN_ON(async_extent->pages[i]->mapping);
633                                 page_cache_release(async_extent->pages[i]);
634                         }
635                         kfree(async_extent->pages);
636                         async_extent->nr_pages = 0;
637                         async_extent->pages = NULL;
638                         unlock_extent(io_tree, async_extent->start,
639                                       async_extent->start +
640                                       async_extent->ram_size - 1, GFP_NOFS);
641                         goto retry;
642                 }
643
644                 /*
645                  * here we're doing allocation and writeback of the
646                  * compressed pages
647                  */
648                 btrfs_drop_extent_cache(inode, async_extent->start,
649                                         async_extent->start +
650                                         async_extent->ram_size - 1, 0);
651
652                 em = alloc_extent_map(GFP_NOFS);
653                 BUG_ON(!em);
654                 em->start = async_extent->start;
655                 em->len = async_extent->ram_size;
656                 em->orig_start = em->start;
657
658                 em->block_start = ins.objectid;
659                 em->block_len = ins.offset;
660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
661                 em->compress_type = async_extent->compress_type;
662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
664
665                 while (1) {
666                         write_lock(&em_tree->lock);
667                         ret = add_extent_mapping(em_tree, em);
668                         write_unlock(&em_tree->lock);
669                         if (ret != -EEXIST) {
670                                 free_extent_map(em);
671                                 break;
672                         }
673                         btrfs_drop_extent_cache(inode, async_extent->start,
674                                                 async_extent->start +
675                                                 async_extent->ram_size - 1, 0);
676                 }
677
678                 ret = btrfs_add_ordered_extent_compress(inode,
679                                                 async_extent->start,
680                                                 ins.objectid,
681                                                 async_extent->ram_size,
682                                                 ins.offset,
683                                                 BTRFS_ORDERED_COMPRESSED,
684                                                 async_extent->compress_type);
685                 BUG_ON(ret);
686
687                 /*
688                  * clear dirty, set writeback and unlock the pages.
689                  */
690                 extent_clear_unlock_delalloc(inode,
691                                 &BTRFS_I(inode)->io_tree,
692                                 async_extent->start,
693                                 async_extent->start +
694                                 async_extent->ram_size - 1,
695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
696                                 EXTENT_CLEAR_UNLOCK |
697                                 EXTENT_CLEAR_DELALLOC |
698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
699
700                 ret = btrfs_submit_compressed_write(inode,
701                                     async_extent->start,
702                                     async_extent->ram_size,
703                                     ins.objectid,
704                                     ins.offset, async_extent->pages,
705                                     async_extent->nr_pages);
706
707                 BUG_ON(ret);
708                 alloc_hint = ins.objectid + ins.offset;
709                 kfree(async_extent);
710                 cond_resched();
711         }
712
713         return 0;
714 }
715
716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
717                                       u64 num_bytes)
718 {
719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
720         struct extent_map *em;
721         u64 alloc_hint = 0;
722
723         read_lock(&em_tree->lock);
724         em = search_extent_mapping(em_tree, start, num_bytes);
725         if (em) {
726                 /*
727                  * if block start isn't an actual block number then find the
728                  * first block in this inode and use that as a hint.  If that
729                  * block is also bogus then just don't worry about it.
730                  */
731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
732                         free_extent_map(em);
733                         em = search_extent_mapping(em_tree, 0, 0);
734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
735                                 alloc_hint = em->block_start;
736                         if (em)
737                                 free_extent_map(em);
738                 } else {
739                         alloc_hint = em->block_start;
740                         free_extent_map(em);
741                 }
742         }
743         read_unlock(&em_tree->lock);
744
745         return alloc_hint;
746 }
747
748 /*
749  * when extent_io.c finds a delayed allocation range in the file,
750  * the call backs end up in this code.  The basic idea is to
751  * allocate extents on disk for the range, and create ordered data structs
752  * in ram to track those extents.
753  *
754  * locked_page is the page that writepage had locked already.  We use
755  * it to make sure we don't do extra locks or unlocks.
756  *
757  * *page_started is set to one if we unlock locked_page and do everything
758  * required to start IO on it.  It may be clean and already done with
759  * IO when we return.
760  */
761 static noinline int cow_file_range(struct inode *inode,
762                                    struct page *locked_page,
763                                    u64 start, u64 end, int *page_started,
764                                    unsigned long *nr_written,
765                                    int unlock)
766 {
767         struct btrfs_root *root = BTRFS_I(inode)->root;
768         struct btrfs_trans_handle *trans;
769         u64 alloc_hint = 0;
770         u64 num_bytes;
771         unsigned long ram_size;
772         u64 disk_num_bytes;
773         u64 cur_alloc_size;
774         u64 blocksize = root->sectorsize;
775         struct btrfs_key ins;
776         struct extent_map *em;
777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
778         int ret = 0;
779
780         BUG_ON(root == root->fs_info->tree_root);
781         trans = btrfs_join_transaction(root, 1);
782         BUG_ON(IS_ERR(trans));
783         btrfs_set_trans_block_group(trans, inode);
784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
785
786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
787         num_bytes = max(blocksize,  num_bytes);
788         disk_num_bytes = num_bytes;
789         ret = 0;
790
791         if (start == 0) {
792                 /* lets try to make an inline extent */
793                 ret = cow_file_range_inline(trans, root, inode,
794                                             start, end, 0, 0, NULL);
795                 if (ret == 0) {
796                         extent_clear_unlock_delalloc(inode,
797                                      &BTRFS_I(inode)->io_tree,
798                                      start, end, NULL,
799                                      EXTENT_CLEAR_UNLOCK_PAGE |
800                                      EXTENT_CLEAR_UNLOCK |
801                                      EXTENT_CLEAR_DELALLOC |
802                                      EXTENT_CLEAR_DIRTY |
803                                      EXTENT_SET_WRITEBACK |
804                                      EXTENT_END_WRITEBACK);
805
806                         *nr_written = *nr_written +
807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
808                         *page_started = 1;
809                         ret = 0;
810                         goto out;
811                 }
812         }
813
814         BUG_ON(disk_num_bytes >
815                btrfs_super_total_bytes(&root->fs_info->super_copy));
816
817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
819
820         while (disk_num_bytes > 0) {
821                 unsigned long op;
822
823                 cur_alloc_size = disk_num_bytes;
824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
825                                            root->sectorsize, 0, alloc_hint,
826                                            (u64)-1, &ins, 1);
827                 BUG_ON(ret);
828
829                 em = alloc_extent_map(GFP_NOFS);
830                 BUG_ON(!em);
831                 em->start = start;
832                 em->orig_start = em->start;
833                 ram_size = ins.offset;
834                 em->len = ins.offset;
835
836                 em->block_start = ins.objectid;
837                 em->block_len = ins.offset;
838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
840
841                 while (1) {
842                         write_lock(&em_tree->lock);
843                         ret = add_extent_mapping(em_tree, em);
844                         write_unlock(&em_tree->lock);
845                         if (ret != -EEXIST) {
846                                 free_extent_map(em);
847                                 break;
848                         }
849                         btrfs_drop_extent_cache(inode, start,
850                                                 start + ram_size - 1, 0);
851                 }
852
853                 cur_alloc_size = ins.offset;
854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
855                                                ram_size, cur_alloc_size, 0);
856                 BUG_ON(ret);
857
858                 if (root->root_key.objectid ==
859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
860                         ret = btrfs_reloc_clone_csums(inode, start,
861                                                       cur_alloc_size);
862                         BUG_ON(ret);
863                 }
864
865                 if (disk_num_bytes < cur_alloc_size)
866                         break;
867
868                 /* we're not doing compressed IO, don't unlock the first
869                  * page (which the caller expects to stay locked), don't
870                  * clear any dirty bits and don't set any writeback bits
871                  *
872                  * Do set the Private2 bit so we know this page was properly
873                  * setup for writepage
874                  */
875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
877                         EXTENT_SET_PRIVATE2;
878
879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
880                                              start, start + ram_size - 1,
881                                              locked_page, op);
882                 disk_num_bytes -= cur_alloc_size;
883                 num_bytes -= cur_alloc_size;
884                 alloc_hint = ins.objectid + ins.offset;
885                 start += cur_alloc_size;
886         }
887 out:
888         ret = 0;
889         btrfs_end_transaction(trans, root);
890
891         return ret;
892 }
893
894 /*
895  * work queue call back to started compression on a file and pages
896  */
897 static noinline void async_cow_start(struct btrfs_work *work)
898 {
899         struct async_cow *async_cow;
900         int num_added = 0;
901         async_cow = container_of(work, struct async_cow, work);
902
903         compress_file_range(async_cow->inode, async_cow->locked_page,
904                             async_cow->start, async_cow->end, async_cow,
905                             &num_added);
906         if (num_added == 0)
907                 async_cow->inode = NULL;
908 }
909
910 /*
911  * work queue call back to submit previously compressed pages
912  */
913 static noinline void async_cow_submit(struct btrfs_work *work)
914 {
915         struct async_cow *async_cow;
916         struct btrfs_root *root;
917         unsigned long nr_pages;
918
919         async_cow = container_of(work, struct async_cow, work);
920
921         root = async_cow->root;
922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
923                 PAGE_CACHE_SHIFT;
924
925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
926
927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
928             5 * 1042 * 1024 &&
929             waitqueue_active(&root->fs_info->async_submit_wait))
930                 wake_up(&root->fs_info->async_submit_wait);
931
932         if (async_cow->inode)
933                 submit_compressed_extents(async_cow->inode, async_cow);
934 }
935
936 static noinline void async_cow_free(struct btrfs_work *work)
937 {
938         struct async_cow *async_cow;
939         async_cow = container_of(work, struct async_cow, work);
940         kfree(async_cow);
941 }
942
943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
944                                 u64 start, u64 end, int *page_started,
945                                 unsigned long *nr_written)
946 {
947         struct async_cow *async_cow;
948         struct btrfs_root *root = BTRFS_I(inode)->root;
949         unsigned long nr_pages;
950         u64 cur_end;
951         int limit = 10 * 1024 * 1042;
952
953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
954                          1, 0, NULL, GFP_NOFS);
955         while (start < end) {
956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
957                 BUG_ON(!async_cow);
958                 async_cow->inode = inode;
959                 async_cow->root = root;
960                 async_cow->locked_page = locked_page;
961                 async_cow->start = start;
962
963                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
964                         cur_end = end;
965                 else
966                         cur_end = min(end, start + 512 * 1024 - 1);
967
968                 async_cow->end = cur_end;
969                 INIT_LIST_HEAD(&async_cow->extents);
970
971                 async_cow->work.func = async_cow_start;
972                 async_cow->work.ordered_func = async_cow_submit;
973                 async_cow->work.ordered_free = async_cow_free;
974                 async_cow->work.flags = 0;
975
976                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
977                         PAGE_CACHE_SHIFT;
978                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
979
980                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
981                                    &async_cow->work);
982
983                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
984                         wait_event(root->fs_info->async_submit_wait,
985                            (atomic_read(&root->fs_info->async_delalloc_pages) <
986                             limit));
987                 }
988
989                 while (atomic_read(&root->fs_info->async_submit_draining) &&
990                       atomic_read(&root->fs_info->async_delalloc_pages)) {
991                         wait_event(root->fs_info->async_submit_wait,
992                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
993                            0));
994                 }
995
996                 *nr_written += nr_pages;
997                 start = cur_end + 1;
998         }
999         *page_started = 1;
1000         return 0;
1001 }
1002
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004                                         u64 bytenr, u64 num_bytes)
1005 {
1006         int ret;
1007         struct btrfs_ordered_sum *sums;
1008         LIST_HEAD(list);
1009
1010         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011                                        bytenr + num_bytes - 1, &list);
1012         if (ret == 0 && list_empty(&list))
1013                 return 0;
1014
1015         while (!list_empty(&list)) {
1016                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017                 list_del(&sums->list);
1018                 kfree(sums);
1019         }
1020         return 1;
1021 }
1022
1023 /*
1024  * when nowcow writeback call back.  This checks for snapshots or COW copies
1025  * of the extents that exist in the file, and COWs the file as required.
1026  *
1027  * If no cow copies or snapshots exist, we write directly to the existing
1028  * blocks on disk
1029  */
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031                                        struct page *locked_page,
1032                               u64 start, u64 end, int *page_started, int force,
1033                               unsigned long *nr_written)
1034 {
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         struct btrfs_trans_handle *trans;
1037         struct extent_buffer *leaf;
1038         struct btrfs_path *path;
1039         struct btrfs_file_extent_item *fi;
1040         struct btrfs_key found_key;
1041         u64 cow_start;
1042         u64 cur_offset;
1043         u64 extent_end;
1044         u64 extent_offset;
1045         u64 disk_bytenr;
1046         u64 num_bytes;
1047         int extent_type;
1048         int ret;
1049         int type;
1050         int nocow;
1051         int check_prev = 1;
1052         bool nolock = false;
1053
1054         path = btrfs_alloc_path();
1055         BUG_ON(!path);
1056         if (root == root->fs_info->tree_root) {
1057                 nolock = true;
1058                 trans = btrfs_join_transaction_nolock(root, 1);
1059         } else {
1060                 trans = btrfs_join_transaction(root, 1);
1061         }
1062         BUG_ON(IS_ERR(trans));
1063
1064         cow_start = (u64)-1;
1065         cur_offset = start;
1066         while (1) {
1067                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068                                                cur_offset, 0);
1069                 BUG_ON(ret < 0);
1070                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071                         leaf = path->nodes[0];
1072                         btrfs_item_key_to_cpu(leaf, &found_key,
1073                                               path->slots[0] - 1);
1074                         if (found_key.objectid == inode->i_ino &&
1075                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1076                                 path->slots[0]--;
1077                 }
1078                 check_prev = 0;
1079 next_slot:
1080                 leaf = path->nodes[0];
1081                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082                         ret = btrfs_next_leaf(root, path);
1083                         if (ret < 0)
1084                                 BUG_ON(1);
1085                         if (ret > 0)
1086                                 break;
1087                         leaf = path->nodes[0];
1088                 }
1089
1090                 nocow = 0;
1091                 disk_bytenr = 0;
1092                 num_bytes = 0;
1093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094
1095                 if (found_key.objectid > inode->i_ino ||
1096                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097                     found_key.offset > end)
1098                         break;
1099
1100                 if (found_key.offset > cur_offset) {
1101                         extent_end = found_key.offset;
1102                         extent_type = 0;
1103                         goto out_check;
1104                 }
1105
1106                 fi = btrfs_item_ptr(leaf, path->slots[0],
1107                                     struct btrfs_file_extent_item);
1108                 extent_type = btrfs_file_extent_type(leaf, fi);
1109
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_num_bytes(leaf, fi);
1116                         if (extent_end <= start) {
1117                                 path->slots[0]++;
1118                                 goto next_slot;
1119                         }
1120                         if (disk_bytenr == 0)
1121                                 goto out_check;
1122                         if (btrfs_file_extent_compression(leaf, fi) ||
1123                             btrfs_file_extent_encryption(leaf, fi) ||
1124                             btrfs_file_extent_other_encoding(leaf, fi))
1125                                 goto out_check;
1126                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127                                 goto out_check;
1128                         if (btrfs_extent_readonly(root, disk_bytenr))
1129                                 goto out_check;
1130                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131                                                   found_key.offset -
1132                                                   extent_offset, disk_bytenr))
1133                                 goto out_check;
1134                         disk_bytenr += extent_offset;
1135                         disk_bytenr += cur_offset - found_key.offset;
1136                         num_bytes = min(end + 1, extent_end) - cur_offset;
1137                         /*
1138                          * force cow if csum exists in the range.
1139                          * this ensure that csum for a given extent are
1140                          * either valid or do not exist.
1141                          */
1142                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143                                 goto out_check;
1144                         nocow = 1;
1145                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146                         extent_end = found_key.offset +
1147                                 btrfs_file_extent_inline_len(leaf, fi);
1148                         extent_end = ALIGN(extent_end, root->sectorsize);
1149                 } else {
1150                         BUG_ON(1);
1151                 }
1152 out_check:
1153                 if (extent_end <= start) {
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157                 if (!nocow) {
1158                         if (cow_start == (u64)-1)
1159                                 cow_start = cur_offset;
1160                         cur_offset = extent_end;
1161                         if (cur_offset > end)
1162                                 break;
1163                         path->slots[0]++;
1164                         goto next_slot;
1165                 }
1166
1167                 btrfs_release_path(root, path);
1168                 if (cow_start != (u64)-1) {
1169                         ret = cow_file_range(inode, locked_page, cow_start,
1170                                         found_key.offset - 1, page_started,
1171                                         nr_written, 1);
1172                         BUG_ON(ret);
1173                         cow_start = (u64)-1;
1174                 }
1175
1176                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177                         struct extent_map *em;
1178                         struct extent_map_tree *em_tree;
1179                         em_tree = &BTRFS_I(inode)->extent_tree;
1180                         em = alloc_extent_map(GFP_NOFS);
1181                         BUG_ON(!em);
1182                         em->start = cur_offset;
1183                         em->orig_start = em->start;
1184                         em->len = num_bytes;
1185                         em->block_len = num_bytes;
1186                         em->block_start = disk_bytenr;
1187                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1188                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189                         while (1) {
1190                                 write_lock(&em_tree->lock);
1191                                 ret = add_extent_mapping(em_tree, em);
1192                                 write_unlock(&em_tree->lock);
1193                                 if (ret != -EEXIST) {
1194                                         free_extent_map(em);
1195                                         break;
1196                                 }
1197                                 btrfs_drop_extent_cache(inode, em->start,
1198                                                 em->start + em->len - 1, 0);
1199                         }
1200                         type = BTRFS_ORDERED_PREALLOC;
1201                 } else {
1202                         type = BTRFS_ORDERED_NOCOW;
1203                 }
1204
1205                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206                                                num_bytes, num_bytes, type);
1207                 BUG_ON(ret);
1208
1209                 if (root->root_key.objectid ==
1210                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212                                                       num_bytes);
1213                         BUG_ON(ret);
1214                 }
1215
1216                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217                                 cur_offset, cur_offset + num_bytes - 1,
1218                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220                                 EXTENT_SET_PRIVATE2);
1221                 cur_offset = extent_end;
1222                 if (cur_offset > end)
1223                         break;
1224         }
1225         btrfs_release_path(root, path);
1226
1227         if (cur_offset <= end && cow_start == (u64)-1)
1228                 cow_start = cur_offset;
1229         if (cow_start != (u64)-1) {
1230                 ret = cow_file_range(inode, locked_page, cow_start, end,
1231                                      page_started, nr_written, 1);
1232                 BUG_ON(ret);
1233         }
1234
1235         if (nolock) {
1236                 ret = btrfs_end_transaction_nolock(trans, root);
1237                 BUG_ON(ret);
1238         } else {
1239                 ret = btrfs_end_transaction(trans, root);
1240                 BUG_ON(ret);
1241         }
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * extent_io.c call back to do delayed allocation processing
1248  */
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250                               u64 start, u64 end, int *page_started,
1251                               unsigned long *nr_written)
1252 {
1253         int ret;
1254         struct btrfs_root *root = BTRFS_I(inode)->root;
1255
1256         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258                                          page_started, 1, nr_written);
1259         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261                                          page_started, 0, nr_written);
1262         else if (!btrfs_test_opt(root, COMPRESS) &&
1263                  !(BTRFS_I(inode)->force_compress) &&
1264                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265                 ret = cow_file_range(inode, locked_page, start, end,
1266                                       page_started, nr_written, 1);
1267         else
1268                 ret = cow_file_range_async(inode, locked_page, start, end,
1269                                            page_started, nr_written);
1270         return ret;
1271 }
1272
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274                                    struct extent_state *orig, u64 split)
1275 {
1276         /* not delalloc, ignore it */
1277         if (!(orig->state & EXTENT_DELALLOC))
1278                 return 0;
1279
1280         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281         return 0;
1282 }
1283
1284 /*
1285  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286  * extents so we can keep track of new extents that are just merged onto old
1287  * extents, such as when we are doing sequential writes, so we can properly
1288  * account for the metadata space we'll need.
1289  */
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291                                    struct extent_state *new,
1292                                    struct extent_state *other)
1293 {
1294         /* not delalloc, ignore it */
1295         if (!(other->state & EXTENT_DELALLOC))
1296                 return 0;
1297
1298         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1299         return 0;
1300 }
1301
1302 /*
1303  * extent_io.c set_bit_hook, used to track delayed allocation
1304  * bytes in this file, and to maintain the list of inodes that
1305  * have pending delalloc work to be done.
1306  */
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308                               struct extent_state *state, int *bits)
1309 {
1310
1311         /*
1312          * set_bit and clear bit hooks normally require _irqsave/restore
1313          * but in this case, we are only testing for the DELALLOC
1314          * bit, which is only set or cleared with irqs on
1315          */
1316         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317                 struct btrfs_root *root = BTRFS_I(inode)->root;
1318                 u64 len = state->end + 1 - state->start;
1319                 int do_list = (root->root_key.objectid !=
1320                                BTRFS_ROOT_TREE_OBJECTID);
1321
1322                 if (*bits & EXTENT_FIRST_DELALLOC)
1323                         *bits &= ~EXTENT_FIRST_DELALLOC;
1324                 else
1325                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326
1327                 spin_lock(&root->fs_info->delalloc_lock);
1328                 BTRFS_I(inode)->delalloc_bytes += len;
1329                 root->fs_info->delalloc_bytes += len;
1330                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332                                       &root->fs_info->delalloc_inodes);
1333                 }
1334                 spin_unlock(&root->fs_info->delalloc_lock);
1335         }
1336         return 0;
1337 }
1338
1339 /*
1340  * extent_io.c clear_bit_hook, see set_bit_hook for why
1341  */
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343                                 struct extent_state *state, int *bits)
1344 {
1345         /*
1346          * set_bit and clear bit hooks normally require _irqsave/restore
1347          * but in this case, we are only testing for the DELALLOC
1348          * bit, which is only set or cleared with irqs on
1349          */
1350         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351                 struct btrfs_root *root = BTRFS_I(inode)->root;
1352                 u64 len = state->end + 1 - state->start;
1353                 int do_list = (root->root_key.objectid !=
1354                                BTRFS_ROOT_TREE_OBJECTID);
1355
1356                 if (*bits & EXTENT_FIRST_DELALLOC)
1357                         *bits &= ~EXTENT_FIRST_DELALLOC;
1358                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360
1361                 if (*bits & EXTENT_DO_ACCOUNTING)
1362                         btrfs_delalloc_release_metadata(inode, len);
1363
1364                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365                     && do_list)
1366                         btrfs_free_reserved_data_space(inode, len);
1367
1368                 spin_lock(&root->fs_info->delalloc_lock);
1369                 root->fs_info->delalloc_bytes -= len;
1370                 BTRFS_I(inode)->delalloc_bytes -= len;
1371
1372                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375                 }
1376                 spin_unlock(&root->fs_info->delalloc_lock);
1377         }
1378         return 0;
1379 }
1380
1381 /*
1382  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383  * we don't create bios that span stripes or chunks
1384  */
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386                          size_t size, struct bio *bio,
1387                          unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390         struct btrfs_mapping_tree *map_tree;
1391         u64 logical = (u64)bio->bi_sector << 9;
1392         u64 length = 0;
1393         u64 map_length;
1394         int ret;
1395
1396         if (bio_flags & EXTENT_BIO_COMPRESSED)
1397                 return 0;
1398
1399         length = bio->bi_size;
1400         map_tree = &root->fs_info->mapping_tree;
1401         map_length = length;
1402         ret = btrfs_map_block(map_tree, READ, logical,
1403                               &map_length, NULL, 0);
1404
1405         if (map_length < length + size)
1406                 return 1;
1407         return ret;
1408 }
1409
1410 /*
1411  * in order to insert checksums into the metadata in large chunks,
1412  * we wait until bio submission time.   All the pages in the bio are
1413  * checksummed and sums are attached onto the ordered extent record.
1414  *
1415  * At IO completion time the cums attached on the ordered extent record
1416  * are inserted into the btree
1417  */
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419                                     struct bio *bio, int mirror_num,
1420                                     unsigned long bio_flags,
1421                                     u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425
1426         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427         BUG_ON(ret);
1428         return 0;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440                           int mirror_num, unsigned long bio_flags,
1441                           u64 bio_offset)
1442 {
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 }
1446
1447 /*
1448  * extent_io.c submission hook. This does the right thing for csum calculation
1449  * on write, or reading the csums from the tree before a read
1450  */
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags,
1453                           u64 bio_offset)
1454 {
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456         int ret = 0;
1457         int skip_sum;
1458
1459         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460
1461         if (root == root->fs_info->tree_root)
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463         else
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465         BUG_ON(ret);
1466
1467         if (!(rw & REQ_WRITE)) {
1468                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469                         return btrfs_submit_compressed_read(inode, bio,
1470                                                     mirror_num, bio_flags);
1471                 } else if (!skip_sum) {
1472                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473                         if (ret)
1474                                 return ret;
1475                 }
1476                 goto mapit;
1477         } else if (!skip_sum) {
1478                 /* csum items have already been cloned */
1479                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480                         goto mapit;
1481                 /* we're doing a write, do the async checksumming */
1482                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483                                    inode, rw, bio, mirror_num,
1484                                    bio_flags, bio_offset,
1485                                    __btrfs_submit_bio_start,
1486                                    __btrfs_submit_bio_done);
1487         }
1488
1489 mapit:
1490         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1491 }
1492
1493 /*
1494  * given a list of ordered sums record them in the inode.  This happens
1495  * at IO completion time based on sums calculated at bio submission time.
1496  */
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498                              struct inode *inode, u64 file_offset,
1499                              struct list_head *list)
1500 {
1501         struct btrfs_ordered_sum *sum;
1502
1503         btrfs_set_trans_block_group(trans, inode);
1504
1505         list_for_each_entry(sum, list, list) {
1506                 btrfs_csum_file_blocks(trans,
1507                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508         }
1509         return 0;
1510 }
1511
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513                               struct extent_state **cached_state)
1514 {
1515         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516                 WARN_ON(1);
1517         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518                                    cached_state, GFP_NOFS);
1519 }
1520
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1523         struct page *page;
1524         struct btrfs_work work;
1525 };
1526
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 {
1529         struct btrfs_writepage_fixup *fixup;
1530         struct btrfs_ordered_extent *ordered;
1531         struct extent_state *cached_state = NULL;
1532         struct page *page;
1533         struct inode *inode;
1534         u64 page_start;
1535         u64 page_end;
1536
1537         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538         page = fixup->page;
1539 again:
1540         lock_page(page);
1541         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542                 ClearPageChecked(page);
1543                 goto out_page;
1544         }
1545
1546         inode = page->mapping->host;
1547         page_start = page_offset(page);
1548         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549
1550         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551                          &cached_state, GFP_NOFS);
1552
1553         /* already ordered? We're done */
1554         if (PagePrivate2(page))
1555                 goto out;
1556
1557         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558         if (ordered) {
1559                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560                                      page_end, &cached_state, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         BUG();
1567         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568         ClearPageChecked(page);
1569 out:
1570         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571                              &cached_state, GFP_NOFS);
1572 out_page:
1573         unlock_page(page);
1574         page_cache_release(page);
1575         kfree(fixup);
1576 }
1577
1578 /*
1579  * There are a few paths in the higher layers of the kernel that directly
1580  * set the page dirty bit without asking the filesystem if it is a
1581  * good idea.  This causes problems because we want to make sure COW
1582  * properly happens and the data=ordered rules are followed.
1583  *
1584  * In our case any range that doesn't have the ORDERED bit set
1585  * hasn't been properly setup for IO.  We kick off an async process
1586  * to fix it up.  The async helper will wait for ordered extents, set
1587  * the delalloc bit and make it safe to write the page.
1588  */
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         struct btrfs_writepage_fixup *fixup;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594
1595         /* this page is properly in the ordered list */
1596         if (TestClearPagePrivate2(page))
1597                 return 0;
1598
1599         if (PageChecked(page))
1600                 return -EAGAIN;
1601
1602         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603         if (!fixup)
1604                 return -EAGAIN;
1605
1606         SetPageChecked(page);
1607         page_cache_get(page);
1608         fixup->work.func = btrfs_writepage_fixup_worker;
1609         fixup->page = page;
1610         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611         return -EAGAIN;
1612 }
1613
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615                                        struct inode *inode, u64 file_pos,
1616                                        u64 disk_bytenr, u64 disk_num_bytes,
1617                                        u64 num_bytes, u64 ram_bytes,
1618                                        u8 compression, u8 encryption,
1619                                        u16 other_encoding, int extent_type)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         struct btrfs_file_extent_item *fi;
1623         struct btrfs_path *path;
1624         struct extent_buffer *leaf;
1625         struct btrfs_key ins;
1626         u64 hint;
1627         int ret;
1628
1629         path = btrfs_alloc_path();
1630         BUG_ON(!path);
1631
1632         path->leave_spinning = 1;
1633
1634         /*
1635          * we may be replacing one extent in the tree with another.
1636          * The new extent is pinned in the extent map, and we don't want
1637          * to drop it from the cache until it is completely in the btree.
1638          *
1639          * So, tell btrfs_drop_extents to leave this extent in the cache.
1640          * the caller is expected to unpin it and allow it to be merged
1641          * with the others.
1642          */
1643         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644                                  &hint, 0);
1645         BUG_ON(ret);
1646
1647         ins.objectid = inode->i_ino;
1648         ins.offset = file_pos;
1649         ins.type = BTRFS_EXTENT_DATA_KEY;
1650         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651         BUG_ON(ret);
1652         leaf = path->nodes[0];
1653         fi = btrfs_item_ptr(leaf, path->slots[0],
1654                             struct btrfs_file_extent_item);
1655         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656         btrfs_set_file_extent_type(leaf, fi, extent_type);
1657         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659         btrfs_set_file_extent_offset(leaf, fi, 0);
1660         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662         btrfs_set_file_extent_compression(leaf, fi, compression);
1663         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665
1666         btrfs_unlock_up_safe(path, 1);
1667         btrfs_set_lock_blocking(leaf);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670
1671         inode_add_bytes(inode, num_bytes);
1672
1673         ins.objectid = disk_bytenr;
1674         ins.offset = disk_num_bytes;
1675         ins.type = BTRFS_EXTENT_ITEM_KEY;
1676         ret = btrfs_alloc_reserved_file_extent(trans, root,
1677                                         root->root_key.objectid,
1678                                         inode->i_ino, file_pos, &ins);
1679         BUG_ON(ret);
1680         btrfs_free_path(path);
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * helper function for btrfs_finish_ordered_io, this
1687  * just reads in some of the csum leaves to prime them into ram
1688  * before we start the transaction.  It limits the amount of btree
1689  * reads required while inside the transaction.
1690  */
1691 /* as ordered data IO finishes, this gets called so we can finish
1692  * an ordered extent if the range of bytes in the file it covers are
1693  * fully written.
1694  */
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_trans_handle *trans = NULL;
1699         struct btrfs_ordered_extent *ordered_extent = NULL;
1700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701         struct extent_state *cached_state = NULL;
1702         int compress_type = 0;
1703         int ret;
1704         bool nolock = false;
1705
1706         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707                                              end - start + 1);
1708         if (!ret)
1709                 return 0;
1710         BUG_ON(!ordered_extent);
1711
1712         nolock = (root == root->fs_info->tree_root);
1713
1714         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715                 BUG_ON(!list_empty(&ordered_extent->list));
1716                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717                 if (!ret) {
1718                         if (nolock)
1719                                 trans = btrfs_join_transaction_nolock(root, 1);
1720                         else
1721                                 trans = btrfs_join_transaction(root, 1);
1722                         BUG_ON(IS_ERR(trans));
1723                         btrfs_set_trans_block_group(trans, inode);
1724                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725                         ret = btrfs_update_inode(trans, root, inode);
1726                         BUG_ON(ret);
1727                 }
1728                 goto out;
1729         }
1730
1731         lock_extent_bits(io_tree, ordered_extent->file_offset,
1732                          ordered_extent->file_offset + ordered_extent->len - 1,
1733                          0, &cached_state, GFP_NOFS);
1734
1735         if (nolock)
1736                 trans = btrfs_join_transaction_nolock(root, 1);
1737         else
1738                 trans = btrfs_join_transaction(root, 1);
1739         BUG_ON(IS_ERR(trans));
1740         btrfs_set_trans_block_group(trans, inode);
1741         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742
1743         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744                 compress_type = ordered_extent->compress_type;
1745         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746                 BUG_ON(compress_type);
1747                 ret = btrfs_mark_extent_written(trans, inode,
1748                                                 ordered_extent->file_offset,
1749                                                 ordered_extent->file_offset +
1750                                                 ordered_extent->len);
1751                 BUG_ON(ret);
1752         } else {
1753                 BUG_ON(root == root->fs_info->tree_root);
1754                 ret = insert_reserved_file_extent(trans, inode,
1755                                                 ordered_extent->file_offset,
1756                                                 ordered_extent->start,
1757                                                 ordered_extent->disk_len,
1758                                                 ordered_extent->len,
1759                                                 ordered_extent->len,
1760                                                 compress_type, 0, 0,
1761                                                 BTRFS_FILE_EXTENT_REG);
1762                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763                                    ordered_extent->file_offset,
1764                                    ordered_extent->len);
1765                 BUG_ON(ret);
1766         }
1767         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768                              ordered_extent->file_offset +
1769                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770
1771         add_pending_csums(trans, inode, ordered_extent->file_offset,
1772                           &ordered_extent->list);
1773
1774         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775         if (!ret) {
1776                 ret = btrfs_update_inode(trans, root, inode);
1777                 BUG_ON(ret);
1778         }
1779         ret = 0;
1780 out:
1781         if (nolock) {
1782                 if (trans)
1783                         btrfs_end_transaction_nolock(trans, root);
1784         } else {
1785                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1786                 if (trans)
1787                         btrfs_end_transaction(trans, root);
1788         }
1789
1790         /* once for us */
1791         btrfs_put_ordered_extent(ordered_extent);
1792         /* once for the tree */
1793         btrfs_put_ordered_extent(ordered_extent);
1794
1795         return 0;
1796 }
1797
1798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1799                                 struct extent_state *state, int uptodate)
1800 {
1801         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1802
1803         ClearPagePrivate2(page);
1804         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1805 }
1806
1807 /*
1808  * When IO fails, either with EIO or csum verification fails, we
1809  * try other mirrors that might have a good copy of the data.  This
1810  * io_failure_record is used to record state as we go through all the
1811  * mirrors.  If another mirror has good data, the page is set up to date
1812  * and things continue.  If a good mirror can't be found, the original
1813  * bio end_io callback is called to indicate things have failed.
1814  */
1815 struct io_failure_record {
1816         struct page *page;
1817         u64 start;
1818         u64 len;
1819         u64 logical;
1820         unsigned long bio_flags;
1821         int last_mirror;
1822 };
1823
1824 static int btrfs_io_failed_hook(struct bio *failed_bio,
1825                          struct page *page, u64 start, u64 end,
1826                          struct extent_state *state)
1827 {
1828         struct io_failure_record *failrec = NULL;
1829         u64 private;
1830         struct extent_map *em;
1831         struct inode *inode = page->mapping->host;
1832         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1833         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1834         struct bio *bio;
1835         int num_copies;
1836         int ret;
1837         int rw;
1838         u64 logical;
1839
1840         ret = get_state_private(failure_tree, start, &private);
1841         if (ret) {
1842                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1843                 if (!failrec)
1844                         return -ENOMEM;
1845                 failrec->start = start;
1846                 failrec->len = end - start + 1;
1847                 failrec->last_mirror = 0;
1848                 failrec->bio_flags = 0;
1849
1850                 read_lock(&em_tree->lock);
1851                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1852                 if (em->start > start || em->start + em->len < start) {
1853                         free_extent_map(em);
1854                         em = NULL;
1855                 }
1856                 read_unlock(&em_tree->lock);
1857
1858                 if (!em || IS_ERR(em)) {
1859                         kfree(failrec);
1860                         return -EIO;
1861                 }
1862                 logical = start - em->start;
1863                 logical = em->block_start + logical;
1864                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1865                         logical = em->block_start;
1866                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1867                         extent_set_compress_type(&failrec->bio_flags,
1868                                                  em->compress_type);
1869                 }
1870                 failrec->logical = logical;
1871                 free_extent_map(em);
1872                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1873                                 EXTENT_DIRTY, GFP_NOFS);
1874                 set_state_private(failure_tree, start,
1875                                  (u64)(unsigned long)failrec);
1876         } else {
1877                 failrec = (struct io_failure_record *)(unsigned long)private;
1878         }
1879         num_copies = btrfs_num_copies(
1880                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1881                               failrec->logical, failrec->len);
1882         failrec->last_mirror++;
1883         if (!state) {
1884                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1885                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1886                                                     failrec->start,
1887                                                     EXTENT_LOCKED);
1888                 if (state && state->start != failrec->start)
1889                         state = NULL;
1890                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1891         }
1892         if (!state || failrec->last_mirror > num_copies) {
1893                 set_state_private(failure_tree, failrec->start, 0);
1894                 clear_extent_bits(failure_tree, failrec->start,
1895                                   failrec->start + failrec->len - 1,
1896                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1897                 kfree(failrec);
1898                 return -EIO;
1899         }
1900         bio = bio_alloc(GFP_NOFS, 1);
1901         bio->bi_private = state;
1902         bio->bi_end_io = failed_bio->bi_end_io;
1903         bio->bi_sector = failrec->logical >> 9;
1904         bio->bi_bdev = failed_bio->bi_bdev;
1905         bio->bi_size = 0;
1906
1907         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1908         if (failed_bio->bi_rw & REQ_WRITE)
1909                 rw = WRITE;
1910         else
1911                 rw = READ;
1912
1913         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1914                                                       failrec->last_mirror,
1915                                                       failrec->bio_flags, 0);
1916         return ret;
1917 }
1918
1919 /*
1920  * each time an IO finishes, we do a fast check in the IO failure tree
1921  * to see if we need to process or clean up an io_failure_record
1922  */
1923 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1924 {
1925         u64 private;
1926         u64 private_failure;
1927         struct io_failure_record *failure;
1928         int ret;
1929
1930         private = 0;
1931         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1932                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1933                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1934                                         start, &private_failure);
1935                 if (ret == 0) {
1936                         failure = (struct io_failure_record *)(unsigned long)
1937                                    private_failure;
1938                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1939                                           failure->start, 0);
1940                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1941                                           failure->start,
1942                                           failure->start + failure->len - 1,
1943                                           EXTENT_DIRTY | EXTENT_LOCKED,
1944                                           GFP_NOFS);
1945                         kfree(failure);
1946                 }
1947         }
1948         return 0;
1949 }
1950
1951 /*
1952  * when reads are done, we need to check csums to verify the data is correct
1953  * if there's a match, we allow the bio to finish.  If not, we go through
1954  * the io_failure_record routines to find good copies
1955  */
1956 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1957                                struct extent_state *state)
1958 {
1959         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1960         struct inode *inode = page->mapping->host;
1961         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1962         char *kaddr;
1963         u64 private = ~(u32)0;
1964         int ret;
1965         struct btrfs_root *root = BTRFS_I(inode)->root;
1966         u32 csum = ~(u32)0;
1967
1968         if (PageChecked(page)) {
1969                 ClearPageChecked(page);
1970                 goto good;
1971         }
1972
1973         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1974                 return 0;
1975
1976         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1977             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1978                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1979                                   GFP_NOFS);
1980                 return 0;
1981         }
1982
1983         if (state && state->start == start) {
1984                 private = state->private;
1985                 ret = 0;
1986         } else {
1987                 ret = get_state_private(io_tree, start, &private);
1988         }
1989         kaddr = kmap_atomic(page, KM_USER0);
1990         if (ret)
1991                 goto zeroit;
1992
1993         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1994         btrfs_csum_final(csum, (char *)&csum);
1995         if (csum != private)
1996                 goto zeroit;
1997
1998         kunmap_atomic(kaddr, KM_USER0);
1999 good:
2000         /* if the io failure tree for this inode is non-empty,
2001          * check to see if we've recovered from a failed IO
2002          */
2003         btrfs_clean_io_failures(inode, start);
2004         return 0;
2005
2006 zeroit:
2007         if (printk_ratelimit()) {
2008                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2009                        "private %llu\n", page->mapping->host->i_ino,
2010                        (unsigned long long)start, csum,
2011                        (unsigned long long)private);
2012         }
2013         memset(kaddr + offset, 1, end - start + 1);
2014         flush_dcache_page(page);
2015         kunmap_atomic(kaddr, KM_USER0);
2016         if (private == 0)
2017                 return 0;
2018         return -EIO;
2019 }
2020
2021 struct delayed_iput {
2022         struct list_head list;
2023         struct inode *inode;
2024 };
2025
2026 void btrfs_add_delayed_iput(struct inode *inode)
2027 {
2028         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2029         struct delayed_iput *delayed;
2030
2031         if (atomic_add_unless(&inode->i_count, -1, 1))
2032                 return;
2033
2034         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2035         delayed->inode = inode;
2036
2037         spin_lock(&fs_info->delayed_iput_lock);
2038         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2039         spin_unlock(&fs_info->delayed_iput_lock);
2040 }
2041
2042 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2043 {
2044         LIST_HEAD(list);
2045         struct btrfs_fs_info *fs_info = root->fs_info;
2046         struct delayed_iput *delayed;
2047         int empty;
2048
2049         spin_lock(&fs_info->delayed_iput_lock);
2050         empty = list_empty(&fs_info->delayed_iputs);
2051         spin_unlock(&fs_info->delayed_iput_lock);
2052         if (empty)
2053                 return;
2054
2055         down_read(&root->fs_info->cleanup_work_sem);
2056         spin_lock(&fs_info->delayed_iput_lock);
2057         list_splice_init(&fs_info->delayed_iputs, &list);
2058         spin_unlock(&fs_info->delayed_iput_lock);
2059
2060         while (!list_empty(&list)) {
2061                 delayed = list_entry(list.next, struct delayed_iput, list);
2062                 list_del(&delayed->list);
2063                 iput(delayed->inode);
2064                 kfree(delayed);
2065         }
2066         up_read(&root->fs_info->cleanup_work_sem);
2067 }
2068
2069 /*
2070  * calculate extra metadata reservation when snapshotting a subvolume
2071  * contains orphan files.
2072  */
2073 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2074                                 struct btrfs_pending_snapshot *pending,
2075                                 u64 *bytes_to_reserve)
2076 {
2077         struct btrfs_root *root;
2078         struct btrfs_block_rsv *block_rsv;
2079         u64 num_bytes;
2080         int index;
2081
2082         root = pending->root;
2083         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2084                 return;
2085
2086         block_rsv = root->orphan_block_rsv;
2087
2088         /* orphan block reservation for the snapshot */
2089         num_bytes = block_rsv->size;
2090
2091         /*
2092          * after the snapshot is created, COWing tree blocks may use more
2093          * space than it frees. So we should make sure there is enough
2094          * reserved space.
2095          */
2096         index = trans->transid & 0x1;
2097         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2098                 num_bytes += block_rsv->size -
2099                              (block_rsv->reserved + block_rsv->freed[index]);
2100         }
2101
2102         *bytes_to_reserve += num_bytes;
2103 }
2104
2105 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2106                                 struct btrfs_pending_snapshot *pending)
2107 {
2108         struct btrfs_root *root = pending->root;
2109         struct btrfs_root *snap = pending->snap;
2110         struct btrfs_block_rsv *block_rsv;
2111         u64 num_bytes;
2112         int index;
2113         int ret;
2114
2115         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2116                 return;
2117
2118         /* refill source subvolume's orphan block reservation */
2119         block_rsv = root->orphan_block_rsv;
2120         index = trans->transid & 0x1;
2121         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122                 num_bytes = block_rsv->size -
2123                             (block_rsv->reserved + block_rsv->freed[index]);
2124                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2125                                               root->orphan_block_rsv,
2126                                               num_bytes);
2127                 BUG_ON(ret);
2128         }
2129
2130         /* setup orphan block reservation for the snapshot */
2131         block_rsv = btrfs_alloc_block_rsv(snap);
2132         BUG_ON(!block_rsv);
2133
2134         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2135         snap->orphan_block_rsv = block_rsv;
2136
2137         num_bytes = root->orphan_block_rsv->size;
2138         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                       block_rsv, num_bytes);
2140         BUG_ON(ret);
2141
2142 #if 0
2143         /* insert orphan item for the snapshot */
2144         WARN_ON(!root->orphan_item_inserted);
2145         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2146                                        snap->root_key.objectid);
2147         BUG_ON(ret);
2148         snap->orphan_item_inserted = 1;
2149 #endif
2150 }
2151
2152 enum btrfs_orphan_cleanup_state {
2153         ORPHAN_CLEANUP_STARTED  = 1,
2154         ORPHAN_CLEANUP_DONE     = 2,
2155 };
2156
2157 /*
2158  * This is called in transaction commmit time. If there are no orphan
2159  * files in the subvolume, it removes orphan item and frees block_rsv
2160  * structure.
2161  */
2162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2163                               struct btrfs_root *root)
2164 {
2165         int ret;
2166
2167         if (!list_empty(&root->orphan_list) ||
2168             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2169                 return;
2170
2171         if (root->orphan_item_inserted &&
2172             btrfs_root_refs(&root->root_item) > 0) {
2173                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2174                                             root->root_key.objectid);
2175                 BUG_ON(ret);
2176                 root->orphan_item_inserted = 0;
2177         }
2178
2179         if (root->orphan_block_rsv) {
2180                 WARN_ON(root->orphan_block_rsv->size > 0);
2181                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2182                 root->orphan_block_rsv = NULL;
2183         }
2184 }
2185
2186 /*
2187  * This creates an orphan entry for the given inode in case something goes
2188  * wrong in the middle of an unlink/truncate.
2189  *
2190  * NOTE: caller of this function should reserve 5 units of metadata for
2191  *       this function.
2192  */
2193 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2194 {
2195         struct btrfs_root *root = BTRFS_I(inode)->root;
2196         struct btrfs_block_rsv *block_rsv = NULL;
2197         int reserve = 0;
2198         int insert = 0;
2199         int ret;
2200
2201         if (!root->orphan_block_rsv) {
2202                 block_rsv = btrfs_alloc_block_rsv(root);
2203                 BUG_ON(!block_rsv);
2204         }
2205
2206         spin_lock(&root->orphan_lock);
2207         if (!root->orphan_block_rsv) {
2208                 root->orphan_block_rsv = block_rsv;
2209         } else if (block_rsv) {
2210                 btrfs_free_block_rsv(root, block_rsv);
2211                 block_rsv = NULL;
2212         }
2213
2214         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2215                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2216 #if 0
2217                 /*
2218                  * For proper ENOSPC handling, we should do orphan
2219                  * cleanup when mounting. But this introduces backward
2220                  * compatibility issue.
2221                  */
2222                 if (!xchg(&root->orphan_item_inserted, 1))
2223                         insert = 2;
2224                 else
2225                         insert = 1;
2226 #endif
2227                 insert = 1;
2228         }
2229
2230         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2231                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2232                 reserve = 1;
2233         }
2234         spin_unlock(&root->orphan_lock);
2235
2236         if (block_rsv)
2237                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2238
2239         /* grab metadata reservation from transaction handle */
2240         if (reserve) {
2241                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2242                 BUG_ON(ret);
2243         }
2244
2245         /* insert an orphan item to track this unlinked/truncated file */
2246         if (insert >= 1) {
2247                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2248                 BUG_ON(ret);
2249         }
2250
2251         /* insert an orphan item to track subvolume contains orphan files */
2252         if (insert >= 2) {
2253                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2254                                                root->root_key.objectid);
2255                 BUG_ON(ret);
2256         }
2257         return 0;
2258 }
2259
2260 /*
2261  * We have done the truncate/delete so we can go ahead and remove the orphan
2262  * item for this particular inode.
2263  */
2264 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2265 {
2266         struct btrfs_root *root = BTRFS_I(inode)->root;
2267         int delete_item = 0;
2268         int release_rsv = 0;
2269         int ret = 0;
2270
2271         spin_lock(&root->orphan_lock);
2272         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2273                 list_del_init(&BTRFS_I(inode)->i_orphan);
2274                 delete_item = 1;
2275         }
2276
2277         if (BTRFS_I(inode)->orphan_meta_reserved) {
2278                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2279                 release_rsv = 1;
2280         }
2281         spin_unlock(&root->orphan_lock);
2282
2283         if (trans && delete_item) {
2284                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2285                 BUG_ON(ret);
2286         }
2287
2288         if (release_rsv)
2289                 btrfs_orphan_release_metadata(inode);
2290
2291         return 0;
2292 }
2293
2294 /*
2295  * this cleans up any orphans that may be left on the list from the last use
2296  * of this root.
2297  */
2298 int btrfs_orphan_cleanup(struct btrfs_root *root)
2299 {
2300         struct btrfs_path *path;
2301         struct extent_buffer *leaf;
2302         struct btrfs_key key, found_key;
2303         struct btrfs_trans_handle *trans;
2304         struct inode *inode;
2305         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2306
2307         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2308                 return 0;
2309
2310         path = btrfs_alloc_path();
2311         if (!path) {
2312                 ret = -ENOMEM;
2313                 goto out;
2314         }
2315         path->reada = -1;
2316
2317         key.objectid = BTRFS_ORPHAN_OBJECTID;
2318         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2319         key.offset = (u64)-1;
2320
2321         while (1) {
2322                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2323                 if (ret < 0)
2324                         goto out;
2325
2326                 /*
2327                  * if ret == 0 means we found what we were searching for, which
2328                  * is weird, but possible, so only screw with path if we didn't
2329                  * find the key and see if we have stuff that matches
2330                  */
2331                 if (ret > 0) {
2332                         ret = 0;
2333                         if (path->slots[0] == 0)
2334                                 break;
2335                         path->slots[0]--;
2336                 }
2337
2338                 /* pull out the item */
2339                 leaf = path->nodes[0];
2340                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2341
2342                 /* make sure the item matches what we want */
2343                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2344                         break;
2345                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2346                         break;
2347
2348                 /* release the path since we're done with it */
2349                 btrfs_release_path(root, path);
2350
2351                 /*
2352                  * this is where we are basically btrfs_lookup, without the
2353                  * crossing root thing.  we store the inode number in the
2354                  * offset of the orphan item.
2355                  */
2356                 found_key.objectid = found_key.offset;
2357                 found_key.type = BTRFS_INODE_ITEM_KEY;
2358                 found_key.offset = 0;
2359                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2360                 if (IS_ERR(inode)) {
2361                         ret = PTR_ERR(inode);
2362                         goto out;
2363                 }
2364
2365                 /*
2366                  * add this inode to the orphan list so btrfs_orphan_del does
2367                  * the proper thing when we hit it
2368                  */
2369                 spin_lock(&root->orphan_lock);
2370                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2371                 spin_unlock(&root->orphan_lock);
2372
2373                 /*
2374                  * if this is a bad inode, means we actually succeeded in
2375                  * removing the inode, but not the orphan record, which means
2376                  * we need to manually delete the orphan since iput will just
2377                  * do a destroy_inode
2378                  */
2379                 if (is_bad_inode(inode)) {
2380                         trans = btrfs_start_transaction(root, 0);
2381                         if (IS_ERR(trans)) {
2382                                 ret = PTR_ERR(trans);
2383                                 goto out;
2384                         }
2385                         btrfs_orphan_del(trans, inode);
2386                         btrfs_end_transaction(trans, root);
2387                         iput(inode);
2388                         continue;
2389                 }
2390
2391                 /* if we have links, this was a truncate, lets do that */
2392                 if (inode->i_nlink) {
2393                         if (!S_ISREG(inode->i_mode)) {
2394                                 WARN_ON(1);
2395                                 iput(inode);
2396                                 continue;
2397                         }
2398                         nr_truncate++;
2399                         ret = btrfs_truncate(inode);
2400                 } else {
2401                         nr_unlink++;
2402                 }
2403
2404                 /* this will do delete_inode and everything for us */
2405                 iput(inode);
2406                 if (ret)
2407                         goto out;
2408         }
2409         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2410
2411         if (root->orphan_block_rsv)
2412                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2413                                         (u64)-1);
2414
2415         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2416                 trans = btrfs_join_transaction(root, 1);
2417                 if (!IS_ERR(trans))
2418                         btrfs_end_transaction(trans, root);
2419         }
2420
2421         if (nr_unlink)
2422                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2423         if (nr_truncate)
2424                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2425
2426 out:
2427         if (ret)
2428                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2429         btrfs_free_path(path);
2430         return ret;
2431 }
2432
2433 /*
2434  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2435  * don't find any xattrs, we know there can't be any acls.
2436  *
2437  * slot is the slot the inode is in, objectid is the objectid of the inode
2438  */
2439 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2440                                           int slot, u64 objectid)
2441 {
2442         u32 nritems = btrfs_header_nritems(leaf);
2443         struct btrfs_key found_key;
2444         int scanned = 0;
2445
2446         slot++;
2447         while (slot < nritems) {
2448                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2449
2450                 /* we found a different objectid, there must not be acls */
2451                 if (found_key.objectid != objectid)
2452                         return 0;
2453
2454                 /* we found an xattr, assume we've got an acl */
2455                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2456                         return 1;
2457
2458                 /*
2459                  * we found a key greater than an xattr key, there can't
2460                  * be any acls later on
2461                  */
2462                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2463                         return 0;
2464
2465                 slot++;
2466                 scanned++;
2467
2468                 /*
2469                  * it goes inode, inode backrefs, xattrs, extents,
2470                  * so if there are a ton of hard links to an inode there can
2471                  * be a lot of backrefs.  Don't waste time searching too hard,
2472                  * this is just an optimization
2473                  */
2474                 if (scanned >= 8)
2475                         break;
2476         }
2477         /* we hit the end of the leaf before we found an xattr or
2478          * something larger than an xattr.  We have to assume the inode
2479          * has acls
2480          */
2481         return 1;
2482 }
2483
2484 /*
2485  * read an inode from the btree into the in-memory inode
2486  */
2487 static void btrfs_read_locked_inode(struct inode *inode)
2488 {
2489         struct btrfs_path *path;
2490         struct extent_buffer *leaf;
2491         struct btrfs_inode_item *inode_item;
2492         struct btrfs_timespec *tspec;
2493         struct btrfs_root *root = BTRFS_I(inode)->root;
2494         struct btrfs_key location;
2495         int maybe_acls;
2496         u64 alloc_group_block;
2497         u32 rdev;
2498         int ret;
2499
2500         path = btrfs_alloc_path();
2501         BUG_ON(!path);
2502         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2503
2504         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2505         if (ret)
2506                 goto make_bad;
2507
2508         leaf = path->nodes[0];
2509         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2510                                     struct btrfs_inode_item);
2511
2512         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2513         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2514         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2515         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2516         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2517
2518         tspec = btrfs_inode_atime(inode_item);
2519         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2520         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521
2522         tspec = btrfs_inode_mtime(inode_item);
2523         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2524         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525
2526         tspec = btrfs_inode_ctime(inode_item);
2527         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2528         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2529
2530         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2531         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2532         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2533         inode->i_generation = BTRFS_I(inode)->generation;
2534         inode->i_rdev = 0;
2535         rdev = btrfs_inode_rdev(leaf, inode_item);
2536
2537         BTRFS_I(inode)->index_cnt = (u64)-1;
2538         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2539
2540         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2541
2542         /*
2543          * try to precache a NULL acl entry for files that don't have
2544          * any xattrs or acls
2545          */
2546         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547         if (!maybe_acls)
2548                 cache_no_acl(inode);
2549
2550         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2551                                                 alloc_group_block, 0);
2552         btrfs_free_path(path);
2553         inode_item = NULL;
2554
2555         switch (inode->i_mode & S_IFMT) {
2556         case S_IFREG:
2557                 inode->i_mapping->a_ops = &btrfs_aops;
2558                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2559                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2560                 inode->i_fop = &btrfs_file_operations;
2561                 inode->i_op = &btrfs_file_inode_operations;
2562                 break;
2563         case S_IFDIR:
2564                 inode->i_fop = &btrfs_dir_file_operations;
2565                 if (root == root->fs_info->tree_root)
2566                         inode->i_op = &btrfs_dir_ro_inode_operations;
2567                 else
2568                         inode->i_op = &btrfs_dir_inode_operations;
2569                 break;
2570         case S_IFLNK:
2571                 inode->i_op = &btrfs_symlink_inode_operations;
2572                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574                 break;
2575         default:
2576                 inode->i_op = &btrfs_special_inode_operations;
2577                 init_special_inode(inode, inode->i_mode, rdev);
2578                 break;
2579         }
2580
2581         btrfs_update_iflags(inode);
2582         return;
2583
2584 make_bad:
2585         btrfs_free_path(path);
2586         make_bad_inode(inode);
2587 }
2588
2589 /*
2590  * given a leaf and an inode, copy the inode fields into the leaf
2591  */
2592 static void fill_inode_item(struct btrfs_trans_handle *trans,
2593                             struct extent_buffer *leaf,
2594                             struct btrfs_inode_item *item,
2595                             struct inode *inode)
2596 {
2597         if (!leaf->map_token)
2598                 map_private_extent_buffer(leaf, (unsigned long)item,
2599                                           sizeof(struct btrfs_inode_item),
2600                                           &leaf->map_token, &leaf->kaddr,
2601                                           &leaf->map_start, &leaf->map_len,
2602                                           KM_USER1);
2603
2604         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2605         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2606         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2607         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2608         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2609
2610         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2611                                inode->i_atime.tv_sec);
2612         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2613                                 inode->i_atime.tv_nsec);
2614
2615         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2616                                inode->i_mtime.tv_sec);
2617         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2618                                 inode->i_mtime.tv_nsec);
2619
2620         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2621                                inode->i_ctime.tv_sec);
2622         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2623                                 inode->i_ctime.tv_nsec);
2624
2625         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2626         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2627         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2628         btrfs_set_inode_transid(leaf, item, trans->transid);
2629         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2630         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2631         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2632
2633         if (leaf->map_token) {
2634                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2635                 leaf->map_token = NULL;
2636         }
2637 }
2638
2639 /*
2640  * copy everything in the in-memory inode into the btree.
2641  */
2642 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2643                                 struct btrfs_root *root, struct inode *inode)
2644 {
2645         struct btrfs_inode_item *inode_item;
2646         struct btrfs_path *path;
2647         struct extent_buffer *leaf;
2648         int ret;
2649
2650         path = btrfs_alloc_path();
2651         BUG_ON(!path);
2652         path->leave_spinning = 1;
2653         ret = btrfs_lookup_inode(trans, root, path,
2654                                  &BTRFS_I(inode)->location, 1);
2655         if (ret) {
2656                 if (ret > 0)
2657                         ret = -ENOENT;
2658                 goto failed;
2659         }
2660
2661         btrfs_unlock_up_safe(path, 1);
2662         leaf = path->nodes[0];
2663         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2664                                   struct btrfs_inode_item);
2665
2666         fill_inode_item(trans, leaf, inode_item, inode);
2667         btrfs_mark_buffer_dirty(leaf);
2668         btrfs_set_inode_last_trans(trans, inode);
2669         ret = 0;
2670 failed:
2671         btrfs_free_path(path);
2672         return ret;
2673 }
2674
2675
2676 /*
2677  * unlink helper that gets used here in inode.c and in the tree logging
2678  * recovery code.  It remove a link in a directory with a given name, and
2679  * also drops the back refs in the inode to the directory
2680  */
2681 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2682                                 struct btrfs_root *root,
2683                                 struct inode *dir, struct inode *inode,
2684                                 const char *name, int name_len)
2685 {
2686         struct btrfs_path *path;
2687         int ret = 0;
2688         struct extent_buffer *leaf;
2689         struct btrfs_dir_item *di;
2690         struct btrfs_key key;
2691         u64 index;
2692
2693         path = btrfs_alloc_path();
2694         if (!path) {
2695                 ret = -ENOMEM;
2696                 goto out;
2697         }
2698
2699         path->leave_spinning = 1;
2700         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2701                                     name, name_len, -1);
2702         if (IS_ERR(di)) {
2703                 ret = PTR_ERR(di);
2704                 goto err;
2705         }
2706         if (!di) {
2707                 ret = -ENOENT;
2708                 goto err;
2709         }
2710         leaf = path->nodes[0];
2711         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2712         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2713         if (ret)
2714                 goto err;
2715         btrfs_release_path(root, path);
2716
2717         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2718                                   inode->i_ino,
2719                                   dir->i_ino, &index);
2720         if (ret) {
2721                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2722                        "inode %lu parent %lu\n", name_len, name,
2723                        inode->i_ino, dir->i_ino);
2724                 goto err;
2725         }
2726
2727         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2728                                          index, name, name_len, -1);
2729         if (IS_ERR(di)) {
2730                 ret = PTR_ERR(di);
2731                 goto err;
2732         }
2733         if (!di) {
2734                 ret = -ENOENT;
2735                 goto err;
2736         }
2737         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2738         btrfs_release_path(root, path);
2739
2740         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2741                                          inode, dir->i_ino);
2742         BUG_ON(ret != 0 && ret != -ENOENT);
2743
2744         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2745                                            dir, index);
2746         if (ret == -ENOENT)
2747                 ret = 0;
2748 err:
2749         btrfs_free_path(path);
2750         if (ret)
2751                 goto out;
2752
2753         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2754         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2755         btrfs_update_inode(trans, root, dir);
2756 out:
2757         return ret;
2758 }
2759
2760 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2761                        struct btrfs_root *root,
2762                        struct inode *dir, struct inode *inode,
2763                        const char *name, int name_len)
2764 {
2765         int ret;
2766         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2767         if (!ret) {
2768                 btrfs_drop_nlink(inode);
2769                 ret = btrfs_update_inode(trans, root, inode);
2770         }
2771         return ret;
2772 }
2773                 
2774
2775 /* helper to check if there is any shared block in the path */
2776 static int check_path_shared(struct btrfs_root *root,
2777                              struct btrfs_path *path)
2778 {
2779         struct extent_buffer *eb;
2780         int level;
2781         u64 refs = 1;
2782
2783         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2784                 int ret;
2785
2786                 if (!path->nodes[level])
2787                         break;
2788                 eb = path->nodes[level];
2789                 if (!btrfs_block_can_be_shared(root, eb))
2790                         continue;
2791                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2792                                                &refs, NULL);
2793                 if (refs > 1)
2794                         return 1;
2795         }
2796         return 0;
2797 }
2798
2799 /*
2800  * helper to start transaction for unlink and rmdir.
2801  *
2802  * unlink and rmdir are special in btrfs, they do not always free space.
2803  * so in enospc case, we should make sure they will free space before
2804  * allowing them to use the global metadata reservation.
2805  */
2806 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2807                                                        struct dentry *dentry)
2808 {
2809         struct btrfs_trans_handle *trans;
2810         struct btrfs_root *root = BTRFS_I(dir)->root;
2811         struct btrfs_path *path;
2812         struct btrfs_inode_ref *ref;
2813         struct btrfs_dir_item *di;
2814         struct inode *inode = dentry->d_inode;
2815         u64 index;
2816         int check_link = 1;
2817         int err = -ENOSPC;
2818         int ret;
2819
2820         trans = btrfs_start_transaction(root, 10);
2821         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2822                 return trans;
2823
2824         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2825                 return ERR_PTR(-ENOSPC);
2826
2827         /* check if there is someone else holds reference */
2828         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2829                 return ERR_PTR(-ENOSPC);
2830
2831         if (atomic_read(&inode->i_count) > 2)
2832                 return ERR_PTR(-ENOSPC);
2833
2834         if (xchg(&root->fs_info->enospc_unlink, 1))
2835                 return ERR_PTR(-ENOSPC);
2836
2837         path = btrfs_alloc_path();
2838         if (!path) {
2839                 root->fs_info->enospc_unlink = 0;
2840                 return ERR_PTR(-ENOMEM);
2841         }
2842
2843         trans = btrfs_start_transaction(root, 0);
2844         if (IS_ERR(trans)) {
2845                 btrfs_free_path(path);
2846                 root->fs_info->enospc_unlink = 0;
2847                 return trans;
2848         }
2849
2850         path->skip_locking = 1;
2851         path->search_commit_root = 1;
2852
2853         ret = btrfs_lookup_inode(trans, root, path,
2854                                 &BTRFS_I(dir)->location, 0);
2855         if (ret < 0) {
2856                 err = ret;
2857                 goto out;
2858         }
2859         if (ret == 0) {
2860                 if (check_path_shared(root, path))
2861                         goto out;
2862         } else {
2863                 check_link = 0;
2864         }
2865         btrfs_release_path(root, path);
2866
2867         ret = btrfs_lookup_inode(trans, root, path,
2868                                 &BTRFS_I(inode)->location, 0);
2869         if (ret < 0) {
2870                 err = ret;
2871                 goto out;
2872         }
2873         if (ret == 0) {
2874                 if (check_path_shared(root, path))
2875                         goto out;
2876         } else {
2877                 check_link = 0;
2878         }
2879         btrfs_release_path(root, path);
2880
2881         if (ret == 0 && S_ISREG(inode->i_mode)) {
2882                 ret = btrfs_lookup_file_extent(trans, root, path,
2883                                                inode->i_ino, (u64)-1, 0);
2884                 if (ret < 0) {
2885                         err = ret;
2886                         goto out;
2887                 }
2888                 BUG_ON(ret == 0);
2889                 if (check_path_shared(root, path))
2890                         goto out;
2891                 btrfs_release_path(root, path);
2892         }
2893
2894         if (!check_link) {
2895                 err = 0;
2896                 goto out;
2897         }
2898
2899         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2900                                 dentry->d_name.name, dentry->d_name.len, 0);
2901         if (IS_ERR(di)) {
2902                 err = PTR_ERR(di);
2903                 goto out;
2904         }
2905         if (di) {
2906                 if (check_path_shared(root, path))
2907                         goto out;
2908         } else {
2909                 err = 0;
2910                 goto out;
2911         }
2912         btrfs_release_path(root, path);
2913
2914         ref = btrfs_lookup_inode_ref(trans, root, path,
2915                                 dentry->d_name.name, dentry->d_name.len,
2916                                 inode->i_ino, dir->i_ino, 0);
2917         if (IS_ERR(ref)) {
2918                 err = PTR_ERR(ref);
2919                 goto out;
2920         }
2921         BUG_ON(!ref);
2922         if (check_path_shared(root, path))
2923                 goto out;
2924         index = btrfs_inode_ref_index(path->nodes[0], ref);
2925         btrfs_release_path(root, path);
2926
2927         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2928                                 dentry->d_name.name, dentry->d_name.len, 0);
2929         if (IS_ERR(di)) {
2930                 err = PTR_ERR(di);
2931                 goto out;
2932         }
2933         BUG_ON(ret == -ENOENT);
2934         if (check_path_shared(root, path))
2935                 goto out;
2936
2937         err = 0;
2938 out:
2939         btrfs_free_path(path);
2940         if (err) {
2941                 btrfs_end_transaction(trans, root);
2942                 root->fs_info->enospc_unlink = 0;
2943                 return ERR_PTR(err);
2944         }
2945
2946         trans->block_rsv = &root->fs_info->global_block_rsv;
2947         return trans;
2948 }
2949
2950 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2951                                struct btrfs_root *root)
2952 {
2953         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2954                 BUG_ON(!root->fs_info->enospc_unlink);
2955                 root->fs_info->enospc_unlink = 0;
2956         }
2957         btrfs_end_transaction_throttle(trans, root);
2958 }
2959
2960 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2961 {
2962         struct btrfs_root *root = BTRFS_I(dir)->root;
2963         struct btrfs_trans_handle *trans;
2964         struct inode *inode = dentry->d_inode;
2965         int ret;
2966         unsigned long nr = 0;
2967
2968         trans = __unlink_start_trans(dir, dentry);
2969         if (IS_ERR(trans))
2970                 return PTR_ERR(trans);
2971
2972         btrfs_set_trans_block_group(trans, dir);
2973
2974         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2975
2976         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2977                                  dentry->d_name.name, dentry->d_name.len);
2978         BUG_ON(ret);
2979
2980         if (inode->i_nlink == 0) {
2981                 ret = btrfs_orphan_add(trans, inode);
2982                 BUG_ON(ret);
2983         }
2984
2985         nr = trans->blocks_used;
2986         __unlink_end_trans(trans, root);
2987         btrfs_btree_balance_dirty(root, nr);
2988         return ret;
2989 }
2990
2991 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2992                         struct btrfs_root *root,
2993                         struct inode *dir, u64 objectid,
2994                         const char *name, int name_len)
2995 {
2996         struct btrfs_path *path;
2997         struct extent_buffer *leaf;
2998         struct btrfs_dir_item *di;
2999         struct btrfs_key key;
3000         u64 index;
3001         int ret;
3002
3003         path = btrfs_alloc_path();
3004         if (!path)
3005                 return -ENOMEM;
3006
3007         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3008                                    name, name_len, -1);
3009         BUG_ON(!di || IS_ERR(di));
3010
3011         leaf = path->nodes[0];
3012         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3013         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3014         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3015         BUG_ON(ret);
3016         btrfs_release_path(root, path);
3017
3018         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3019                                  objectid, root->root_key.objectid,
3020                                  dir->i_ino, &index, name, name_len);
3021         if (ret < 0) {
3022                 BUG_ON(ret != -ENOENT);
3023                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3024                                                  name, name_len);
3025                 BUG_ON(!di || IS_ERR(di));
3026
3027                 leaf = path->nodes[0];
3028                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3029                 btrfs_release_path(root, path);
3030                 index = key.offset;
3031         }
3032
3033         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3034                                          index, name, name_len, -1);
3035         BUG_ON(!di || IS_ERR(di));
3036
3037         leaf = path->nodes[0];
3038         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3039         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3040         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3041         BUG_ON(ret);
3042         btrfs_release_path(root, path);
3043
3044         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3045         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3046         ret = btrfs_update_inode(trans, root, dir);
3047         BUG_ON(ret);
3048
3049         btrfs_free_path(path);
3050         return 0;
3051 }
3052
3053 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3054 {
3055         struct inode *inode = dentry->d_inode;
3056         int err = 0;
3057         struct btrfs_root *root = BTRFS_I(dir)->root;
3058         struct btrfs_trans_handle *trans;
3059         unsigned long nr = 0;
3060
3061         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3062             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3063                 return -ENOTEMPTY;
3064
3065         trans = __unlink_start_trans(dir, dentry);
3066         if (IS_ERR(trans))
3067                 return PTR_ERR(trans);
3068
3069         btrfs_set_trans_block_group(trans, dir);
3070
3071         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3072                 err = btrfs_unlink_subvol(trans, root, dir,
3073                                           BTRFS_I(inode)->location.objectid,
3074                                           dentry->d_name.name,
3075                                           dentry->d_name.len);
3076                 goto out;
3077         }
3078
3079         err = btrfs_orphan_add(trans, inode);
3080         if (err)
3081                 goto out;
3082
3083         /* now the directory is empty */
3084         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085                                  dentry->d_name.name, dentry->d_name.len);
3086         if (!err)
3087                 btrfs_i_size_write(inode, 0);
3088 out:
3089         nr = trans->blocks_used;
3090         __unlink_end_trans(trans, root);
3091         btrfs_btree_balance_dirty(root, nr);
3092
3093         return err;
3094 }
3095
3096 #if 0
3097 /*
3098  * when truncating bytes in a file, it is possible to avoid reading
3099  * the leaves that contain only checksum items.  This can be the
3100  * majority of the IO required to delete a large file, but it must
3101  * be done carefully.
3102  *
3103  * The keys in the level just above the leaves are checked to make sure
3104  * the lowest key in a given leaf is a csum key, and starts at an offset
3105  * after the new  size.
3106  *
3107  * Then the key for the next leaf is checked to make sure it also has
3108  * a checksum item for the same file.  If it does, we know our target leaf
3109  * contains only checksum items, and it can be safely freed without reading
3110  * it.
3111  *
3112  * This is just an optimization targeted at large files.  It may do
3113  * nothing.  It will return 0 unless things went badly.
3114  */
3115 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3116                                      struct btrfs_root *root,
3117                                      struct btrfs_path *path,
3118                                      struct inode *inode, u64 new_size)
3119 {
3120         struct btrfs_key key;
3121         int ret;
3122         int nritems;
3123         struct btrfs_key found_key;
3124         struct btrfs_key other_key;
3125         struct btrfs_leaf_ref *ref;
3126         u64 leaf_gen;
3127         u64 leaf_start;
3128
3129         path->lowest_level = 1;
3130         key.objectid = inode->i_ino;
3131         key.type = BTRFS_CSUM_ITEM_KEY;
3132         key.offset = new_size;
3133 again:
3134         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3135         if (ret < 0)
3136                 goto out;
3137
3138         if (path->nodes[1] == NULL) {
3139                 ret = 0;
3140                 goto out;
3141         }
3142         ret = 0;
3143         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3144         nritems = btrfs_header_nritems(path->nodes[1]);
3145
3146         if (!nritems)
3147                 goto out;
3148
3149         if (path->slots[1] >= nritems)
3150                 goto next_node;
3151
3152         /* did we find a key greater than anything we want to delete? */
3153         if (found_key.objectid > inode->i_ino ||
3154            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3155                 goto out;
3156
3157         /* we check the next key in the node to make sure the leave contains
3158          * only checksum items.  This comparison doesn't work if our
3159          * leaf is the last one in the node
3160          */
3161         if (path->slots[1] + 1 >= nritems) {
3162 next_node:
3163                 /* search forward from the last key in the node, this
3164                  * will bring us into the next node in the tree
3165                  */
3166                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3167
3168                 /* unlikely, but we inc below, so check to be safe */
3169                 if (found_key.offset == (u64)-1)
3170                         goto out;
3171
3172                 /* search_forward needs a path with locks held, do the
3173                  * search again for the original key.  It is possible
3174                  * this will race with a balance and return a path that
3175                  * we could modify, but this drop is just an optimization
3176                  * and is allowed to miss some leaves.
3177                  */
3178                 btrfs_release_path(root, path);
3179                 found_key.offset++;
3180
3181                 /* setup a max key for search_forward */
3182                 other_key.offset = (u64)-1;
3183                 other_key.type = key.type;
3184                 other_key.objectid = key.objectid;
3185
3186                 path->keep_locks = 1;
3187                 ret = btrfs_search_forward(root, &found_key, &other_key,
3188                                            path, 0, 0);
3189                 path->keep_locks = 0;
3190                 if (ret || found_key.objectid != key.objectid ||
3191                     found_key.type != key.type) {
3192                         ret = 0;
3193                         goto out;
3194                 }
3195
3196                 key.offset = found_key.offset;
3197                 btrfs_release_path(root, path);
3198                 cond_resched();
3199                 goto again;
3200         }
3201
3202         /* we know there's one more slot after us in the tree,
3203          * read that key so we can verify it is also a checksum item
3204          */
3205         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3206
3207         if (found_key.objectid < inode->i_ino)
3208                 goto next_key;
3209
3210         if (found_key.type != key.type || found_key.offset < new_size)
3211                 goto next_key;
3212
3213         /*
3214          * if the key for the next leaf isn't a csum key from this objectid,
3215          * we can't be sure there aren't good items inside this leaf.
3216          * Bail out
3217          */
3218         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3219                 goto out;
3220
3221         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3222         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3223         /*
3224          * it is safe to delete this leaf, it contains only
3225          * csum items from this inode at an offset >= new_size
3226          */
3227         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3228         BUG_ON(ret);
3229
3230         if (root->ref_cows && leaf_gen < trans->transid) {
3231                 ref = btrfs_alloc_leaf_ref(root, 0);
3232                 if (ref) {
3233                         ref->root_gen = root->root_key.offset;
3234                         ref->bytenr = leaf_start;
3235                         ref->owner = 0;
3236                         ref->generation = leaf_gen;
3237                         ref->nritems = 0;
3238
3239                         btrfs_sort_leaf_ref(ref);
3240
3241                         ret = btrfs_add_leaf_ref(root, ref, 0);
3242                         WARN_ON(ret);
3243                         btrfs_free_leaf_ref(root, ref);
3244                 } else {
3245                         WARN_ON(1);
3246                 }
3247         }
3248 next_key:
3249         btrfs_release_path(root, path);
3250
3251         if (other_key.objectid == inode->i_ino &&
3252             other_key.type == key.type && other_key.offset > key.offset) {
3253                 key.offset = other_key.offset;
3254                 cond_resched();
3255                 goto again;
3256         }
3257         ret = 0;
3258 out:
3259         /* fixup any changes we've made to the path */
3260         path->lowest_level = 0;
3261         path->keep_locks = 0;
3262         btrfs_release_path(root, path);
3263         return ret;
3264 }
3265
3266 #endif
3267
3268 /*
3269  * this can truncate away extent items, csum items and directory items.
3270  * It starts at a high offset and removes keys until it can't find
3271  * any higher than new_size
3272  *
3273  * csum items that cross the new i_size are truncated to the new size
3274  * as well.
3275  *
3276  * min_type is the minimum key type to truncate down to.  If set to 0, this
3277  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3278  */
3279 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3280                                struct btrfs_root *root,
3281                                struct inode *inode,
3282                                u64 new_size, u32 min_type)
3283 {
3284         struct btrfs_path *path;
3285         struct extent_buffer *leaf;
3286         struct btrfs_file_extent_item *fi;
3287         struct btrfs_key key;
3288         struct btrfs_key found_key;
3289         u64 extent_start = 0;
3290         u64 extent_num_bytes = 0;
3291         u64 extent_offset = 0;
3292         u64 item_end = 0;
3293         u64 mask = root->sectorsize - 1;
3294         u32 found_type = (u8)-1;
3295         int found_extent;
3296         int del_item;
3297         int pending_del_nr = 0;
3298         int pending_del_slot = 0;
3299         int extent_type = -1;
3300         int encoding;
3301         int ret;
3302         int err = 0;
3303
3304         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3305
3306         if (root->ref_cows || root == root->fs_info->tree_root)
3307                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3308
3309         path = btrfs_alloc_path();
3310         BUG_ON(!path);
3311         path->reada = -1;
3312
3313         key.objectid = inode->i_ino;
3314         key.offset = (u64)-1;
3315         key.type = (u8)-1;
3316
3317 search_again:
3318         path->leave_spinning = 1;
3319         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3320         if (ret < 0) {
3321                 err = ret;
3322                 goto out;
3323         }
3324
3325         if (ret > 0) {
3326                 /* there are no items in the tree for us to truncate, we're
3327                  * done
3328                  */
3329                 if (path->slots[0] == 0)
3330                         goto out;
3331                 path->slots[0]--;
3332         }
3333
3334         while (1) {
3335                 fi = NULL;
3336                 leaf = path->nodes[0];
3337                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3338                 found_type = btrfs_key_type(&found_key);
3339                 encoding = 0;
3340
3341                 if (found_key.objectid != inode->i_ino)
3342                         break;
3343
3344                 if (found_type < min_type)
3345                         break;
3346
3347                 item_end = found_key.offset;
3348                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3349                         fi = btrfs_item_ptr(leaf, path->slots[0],
3350                                             struct btrfs_file_extent_item);
3351                         extent_type = btrfs_file_extent_type(leaf, fi);
3352                         encoding = btrfs_file_extent_compression(leaf, fi);
3353                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3354                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3355
3356                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3357                                 item_end +=
3358                                     btrfs_file_extent_num_bytes(leaf, fi);
3359                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3360                                 item_end += btrfs_file_extent_inline_len(leaf,
3361                                                                          fi);
3362                         }
3363                         item_end--;
3364                 }
3365                 if (found_type > min_type) {
3366                         del_item = 1;
3367                 } else {
3368                         if (item_end < new_size)
3369                                 break;
3370                         if (found_key.offset >= new_size)
3371                                 del_item = 1;
3372                         else
3373                                 del_item = 0;
3374                 }
3375                 found_extent = 0;
3376                 /* FIXME, shrink the extent if the ref count is only 1 */
3377                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3378                         goto delete;
3379
3380                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3381                         u64 num_dec;
3382                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3383                         if (!del_item && !encoding) {
3384                                 u64 orig_num_bytes =
3385                                         btrfs_file_extent_num_bytes(leaf, fi);
3386                                 extent_num_bytes = new_size -
3387                                         found_key.offset + root->sectorsize - 1;
3388                                 extent_num_bytes = extent_num_bytes &
3389                                         ~((u64)root->sectorsize - 1);
3390                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3391                                                          extent_num_bytes);
3392                                 num_dec = (orig_num_bytes -
3393                                            extent_num_bytes);
3394                                 if (root->ref_cows && extent_start != 0)
3395                                         inode_sub_bytes(inode, num_dec);
3396                                 btrfs_mark_buffer_dirty(leaf);
3397                         } else {
3398                                 extent_num_bytes =
3399                                         btrfs_file_extent_disk_num_bytes(leaf,
3400                                                                          fi);
3401                                 extent_offset = found_key.offset -
3402                                         btrfs_file_extent_offset(leaf, fi);
3403
3404                                 /* FIXME blocksize != 4096 */
3405                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3406                                 if (extent_start != 0) {
3407                                         found_extent = 1;
3408                                         if (root->ref_cows)
3409                                                 inode_sub_bytes(inode, num_dec);
3410                                 }
3411                         }
3412                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3413                         /*
3414                          * we can't truncate inline items that have had
3415                          * special encodings
3416                          */
3417                         if (!del_item &&
3418                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3419                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3420                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3421                                 u32 size = new_size - found_key.offset;
3422
3423                                 if (root->ref_cows) {
3424                                         inode_sub_bytes(inode, item_end + 1 -
3425                                                         new_size);
3426                                 }
3427                                 size =
3428                                     btrfs_file_extent_calc_inline_size(size);
3429                                 ret = btrfs_truncate_item(trans, root, path,
3430                                                           size, 1);
3431                         } else if (root->ref_cows) {
3432                                 inode_sub_bytes(inode, item_end + 1 -
3433                                                 found_key.offset);
3434                         }
3435                 }
3436 delete:
3437                 if (del_item) {
3438                         if (!pending_del_nr) {
3439                                 /* no pending yet, add ourselves */
3440                                 pending_del_slot = path->slots[0];
3441                                 pending_del_nr = 1;
3442                         } else if (pending_del_nr &&
3443                                    path->slots[0] + 1 == pending_del_slot) {
3444                                 /* hop on the pending chunk */
3445                                 pending_del_nr++;
3446                                 pending_del_slot = path->slots[0];
3447                         } else {
3448                                 BUG();
3449                         }
3450                 } else {
3451                         break;
3452                 }
3453                 if (found_extent && (root->ref_cows ||
3454                                      root == root->fs_info->tree_root)) {
3455                         btrfs_set_path_blocking(path);
3456                         ret = btrfs_free_extent(trans, root, extent_start,
3457                                                 extent_num_bytes, 0,
3458                                                 btrfs_header_owner(leaf),
3459                                                 inode->i_ino, extent_offset);
3460                         BUG_ON(ret);
3461                 }
3462
3463                 if (found_type == BTRFS_INODE_ITEM_KEY)
3464                         break;
3465
3466                 if (path->slots[0] == 0 ||
3467                     path->slots[0] != pending_del_slot) {
3468                         if (root->ref_cows) {
3469                                 err = -EAGAIN;
3470                                 goto out;
3471                         }
3472                         if (pending_del_nr) {
3473                                 ret = btrfs_del_items(trans, root, path,
3474                                                 pending_del_slot,
3475                                                 pending_del_nr);
3476                                 BUG_ON(ret);
3477                                 pending_del_nr = 0;
3478                         }
3479                         btrfs_release_path(root, path);
3480                         goto search_again;
3481                 } else {
3482                         path->slots[0]--;
3483                 }
3484         }
3485 out:
3486         if (pending_del_nr) {
3487                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3488                                       pending_del_nr);
3489                 BUG_ON(ret);
3490         }
3491         btrfs_free_path(path);
3492         return err;
3493 }
3494
3495 /*
3496  * taken from block_truncate_page, but does cow as it zeros out
3497  * any bytes left in the last page in the file.
3498  */
3499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3500 {
3501         struct inode *inode = mapping->host;
3502         struct btrfs_root *root = BTRFS_I(inode)->root;
3503         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3504         struct btrfs_ordered_extent *ordered;
3505         struct extent_state *cached_state = NULL;
3506         char *kaddr;
3507         u32 blocksize = root->sectorsize;
3508         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3509         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3510         struct page *page;
3511         int ret = 0;
3512         u64 page_start;
3513         u64 page_end;
3514
3515         if ((offset & (blocksize - 1)) == 0)
3516                 goto out;
3517         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3518         if (ret)
3519                 goto out;
3520
3521         ret = -ENOMEM;
3522 again:
3523         page = grab_cache_page(mapping, index);
3524         if (!page) {
3525                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3526                 goto out;
3527         }
3528
3529         page_start = page_offset(page);
3530         page_end = page_start + PAGE_CACHE_SIZE - 1;
3531
3532         if (!PageUptodate(page)) {
3533                 ret = btrfs_readpage(NULL, page);
3534                 lock_page(page);
3535                 if (page->mapping != mapping) {
3536                         unlock_page(page);
3537                         page_cache_release(page);
3538                         goto again;
3539                 }
3540                 if (!PageUptodate(page)) {
3541                         ret = -EIO;
3542                         goto out_unlock;
3543                 }
3544         }
3545         wait_on_page_writeback(page);
3546
3547         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3548                          GFP_NOFS);
3549         set_page_extent_mapped(page);
3550
3551         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3552         if (ordered) {
3553                 unlock_extent_cached(io_tree, page_start, page_end,
3554                                      &cached_state, GFP_NOFS);
3555                 unlock_page(page);
3556                 page_cache_release(page);
3557                 btrfs_start_ordered_extent(inode, ordered, 1);
3558                 btrfs_put_ordered_extent(ordered);
3559                 goto again;
3560         }
3561
3562         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3563                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3564                           0, 0, &cached_state, GFP_NOFS);
3565
3566         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3567                                         &cached_state);
3568         if (ret) {
3569                 unlock_extent_cached(io_tree, page_start, page_end,
3570                                      &cached_state, GFP_NOFS);
3571                 goto out_unlock;
3572         }
3573
3574         ret = 0;
3575         if (offset != PAGE_CACHE_SIZE) {
3576                 kaddr = kmap(page);
3577                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3578                 flush_dcache_page(page);
3579                 kunmap(page);
3580         }
3581         ClearPageChecked(page);
3582         set_page_dirty(page);
3583         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3584                              GFP_NOFS);
3585
3586 out_unlock:
3587         if (ret)
3588                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3589         unlock_page(page);
3590         page_cache_release(page);
3591 out:
3592         return ret;
3593 }
3594
3595 /*
3596  * This function puts in dummy file extents for the area we're creating a hole
3597  * for.  So if we are truncating this file to a larger size we need to insert
3598  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3599  * the range between oldsize and size
3600  */
3601 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3602 {
3603         struct btrfs_trans_handle *trans;
3604         struct btrfs_root *root = BTRFS_I(inode)->root;
3605         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3606         struct extent_map *em = NULL;
3607         struct extent_state *cached_state = NULL;
3608         u64 mask = root->sectorsize - 1;
3609         u64 hole_start = (oldsize + mask) & ~mask;
3610         u64 block_end = (size + mask) & ~mask;
3611         u64 last_byte;
3612         u64 cur_offset;
3613         u64 hole_size;
3614         int err = 0;
3615
3616         if (size <= hole_start)
3617                 return 0;
3618
3619         while (1) {
3620                 struct btrfs_ordered_extent *ordered;
3621                 btrfs_wait_ordered_range(inode, hole_start,
3622                                          block_end - hole_start);
3623                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3624                                  &cached_state, GFP_NOFS);
3625                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3626                 if (!ordered)
3627                         break;
3628                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3629                                      &cached_state, GFP_NOFS);
3630                 btrfs_put_ordered_extent(ordered);
3631         }
3632
3633         cur_offset = hole_start;
3634         while (1) {
3635                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3636                                 block_end - cur_offset, 0);
3637                 BUG_ON(IS_ERR(em) || !em);
3638                 last_byte = min(extent_map_end(em), block_end);
3639                 last_byte = (last_byte + mask) & ~mask;
3640                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3641                         u64 hint_byte = 0;
3642                         hole_size = last_byte - cur_offset;
3643
3644                         trans = btrfs_start_transaction(root, 2);
3645                         if (IS_ERR(trans)) {
3646                                 err = PTR_ERR(trans);
3647                                 break;
3648                         }
3649                         btrfs_set_trans_block_group(trans, inode);
3650
3651                         err = btrfs_drop_extents(trans, inode, cur_offset,
3652                                                  cur_offset + hole_size,
3653                                                  &hint_byte, 1);
3654                         if (err)
3655                                 break;
3656
3657                         err = btrfs_insert_file_extent(trans, root,
3658                                         inode->i_ino, cur_offset, 0,
3659                                         0, hole_size, 0, hole_size,
3660                                         0, 0, 0);
3661                         if (err)
3662                                 break;
3663
3664                         btrfs_drop_extent_cache(inode, hole_start,
3665                                         last_byte - 1, 0);
3666
3667                         btrfs_end_transaction(trans, root);
3668                 }
3669                 free_extent_map(em);
3670                 em = NULL;
3671                 cur_offset = last_byte;
3672                 if (cur_offset >= block_end)
3673                         break;
3674         }
3675
3676         free_extent_map(em);
3677         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3678                              GFP_NOFS);
3679         return err;
3680 }
3681
3682 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3683 {
3684         loff_t oldsize = i_size_read(inode);
3685         int ret;
3686
3687         if (newsize == oldsize)
3688                 return 0;
3689
3690         if (newsize > oldsize) {
3691                 i_size_write(inode, newsize);
3692                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3693                 truncate_pagecache(inode, oldsize, newsize);
3694                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3695                 if (ret) {
3696                         btrfs_setsize(inode, oldsize);
3697                         return ret;
3698                 }
3699
3700                 mark_inode_dirty(inode);
3701         } else {
3702
3703                 /*
3704                  * We're truncating a file that used to have good data down to
3705                  * zero. Make sure it gets into the ordered flush list so that
3706                  * any new writes get down to disk quickly.
3707                  */
3708                 if (newsize == 0)
3709                         BTRFS_I(inode)->ordered_data_close = 1;
3710
3711                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3712                 truncate_setsize(inode, newsize);
3713                 ret = btrfs_truncate(inode);
3714         }
3715
3716         return ret;
3717 }
3718
3719 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3720 {
3721         struct inode *inode = dentry->d_inode;
3722         struct btrfs_root *root = BTRFS_I(inode)->root;
3723         int err;
3724
3725         if (btrfs_root_readonly(root))
3726                 return -EROFS;
3727
3728         err = inode_change_ok(inode, attr);
3729         if (err)
3730                 return err;
3731
3732         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3733                 err = btrfs_setsize(inode, attr->ia_size);
3734                 if (err)
3735                         return err;
3736         }
3737
3738         if (attr->ia_valid) {
3739                 setattr_copy(inode, attr);
3740                 mark_inode_dirty(inode);
3741
3742                 if (attr->ia_valid & ATTR_MODE)
3743                         err = btrfs_acl_chmod(inode);
3744         }
3745
3746         return err;
3747 }
3748
3749 void btrfs_evict_inode(struct inode *inode)
3750 {
3751         struct btrfs_trans_handle *trans;
3752         struct btrfs_root *root = BTRFS_I(inode)->root;
3753         unsigned long nr;
3754         int ret;
3755
3756         trace_btrfs_inode_evict(inode);
3757
3758         truncate_inode_pages(&inode->i_data, 0);
3759         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3760                                root == root->fs_info->tree_root))
3761                 goto no_delete;
3762
3763         if (is_bad_inode(inode)) {
3764                 btrfs_orphan_del(NULL, inode);
3765                 goto no_delete;
3766         }
3767         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3768         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3769
3770         if (root->fs_info->log_root_recovering) {
3771                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3772                 goto no_delete;
3773         }
3774
3775         if (inode->i_nlink > 0) {
3776                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3777                 goto no_delete;
3778         }
3779
3780         btrfs_i_size_write(inode, 0);
3781
3782         while (1) {
3783                 trans = btrfs_start_transaction(root, 0);
3784                 BUG_ON(IS_ERR(trans));
3785                 btrfs_set_trans_block_group(trans, inode);
3786                 trans->block_rsv = root->orphan_block_rsv;
3787
3788                 ret = btrfs_block_rsv_check(trans, root,
3789                                             root->orphan_block_rsv, 0, 5);
3790                 if (ret) {
3791                         BUG_ON(ret != -EAGAIN);
3792                         ret = btrfs_commit_transaction(trans, root);
3793                         BUG_ON(ret);
3794                         continue;
3795                 }
3796
3797                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3798                 if (ret != -EAGAIN)
3799                         break;
3800
3801                 nr = trans->blocks_used;
3802                 btrfs_end_transaction(trans, root);
3803                 trans = NULL;
3804                 btrfs_btree_balance_dirty(root, nr);
3805
3806         }
3807
3808         if (ret == 0) {
3809                 ret = btrfs_orphan_del(trans, inode);
3810                 BUG_ON(ret);
3811         }
3812
3813         nr = trans->blocks_used;
3814         btrfs_end_transaction(trans, root);
3815         btrfs_btree_balance_dirty(root, nr);
3816 no_delete:
3817         end_writeback(inode);
3818         return;
3819 }
3820
3821 /*
3822  * this returns the key found in the dir entry in the location pointer.
3823  * If no dir entries were found, location->objectid is 0.
3824  */
3825 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3826                                struct btrfs_key *location)
3827 {
3828         const char *name = dentry->d_name.name;
3829         int namelen = dentry->d_name.len;
3830         struct btrfs_dir_item *di;
3831         struct btrfs_path *path;
3832         struct btrfs_root *root = BTRFS_I(dir)->root;
3833         int ret = 0;
3834
3835         path = btrfs_alloc_path();
3836         BUG_ON(!path);
3837
3838         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3839                                     namelen, 0);
3840         if (IS_ERR(di))
3841                 ret = PTR_ERR(di);
3842
3843         if (!di || IS_ERR(di))
3844                 goto out_err;
3845
3846         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3847 out:
3848         btrfs_free_path(path);
3849         return ret;
3850 out_err:
3851         location->objectid = 0;
3852         goto out;
3853 }
3854
3855 /*
3856  * when we hit a tree root in a directory, the btrfs part of the inode
3857  * needs to be changed to reflect the root directory of the tree root.  This
3858  * is kind of like crossing a mount point.
3859  */
3860 static int fixup_tree_root_location(struct btrfs_root *root,
3861                                     struct inode *dir,
3862                                     struct dentry *dentry,
3863                                     struct btrfs_key *location,
3864                                     struct btrfs_root **sub_root)
3865 {
3866         struct btrfs_path *path;
3867         struct btrfs_root *new_root;
3868         struct btrfs_root_ref *ref;
3869         struct extent_buffer *leaf;
3870         int ret;
3871         int err = 0;
3872
3873         path = btrfs_alloc_path();
3874         if (!path) {
3875                 err = -ENOMEM;
3876                 goto out;
3877         }
3878
3879         err = -ENOENT;
3880         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3881                                   BTRFS_I(dir)->root->root_key.objectid,
3882                                   location->objectid);
3883         if (ret) {
3884                 if (ret < 0)
3885                         err = ret;
3886                 goto out;
3887         }
3888
3889         leaf = path->nodes[0];
3890         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3891         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3892             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3893                 goto out;
3894
3895         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3896                                    (unsigned long)(ref + 1),
3897                                    dentry->d_name.len);
3898         if (ret)
3899                 goto out;
3900
3901         btrfs_release_path(root->fs_info->tree_root, path);
3902
3903         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3904         if (IS_ERR(new_root)) {
3905                 err = PTR_ERR(new_root);
3906                 goto out;
3907         }
3908
3909         if (btrfs_root_refs(&new_root->root_item) == 0) {
3910                 err = -ENOENT;
3911                 goto out;
3912         }
3913
3914         *sub_root = new_root;
3915         location->objectid = btrfs_root_dirid(&new_root->root_item);
3916         location->type = BTRFS_INODE_ITEM_KEY;
3917         location->offset = 0;
3918         err = 0;
3919 out:
3920         btrfs_free_path(path);
3921         return err;
3922 }
3923
3924 static void inode_tree_add(struct inode *inode)
3925 {
3926         struct btrfs_root *root = BTRFS_I(inode)->root;
3927         struct btrfs_inode *entry;
3928         struct rb_node **p;
3929         struct rb_node *parent;
3930 again:
3931         p = &root->inode_tree.rb_node;
3932         parent = NULL;
3933
3934         if (inode_unhashed(inode))
3935                 return;
3936
3937         spin_lock(&root->inode_lock);
3938         while (*p) {
3939                 parent = *p;
3940                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3941
3942                 if (inode->i_ino < entry->vfs_inode.i_ino)
3943                         p = &parent->rb_left;
3944                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3945                         p = &parent->rb_right;
3946                 else {
3947                         WARN_ON(!(entry->vfs_inode.i_state &
3948                                   (I_WILL_FREE | I_FREEING)));
3949                         rb_erase(parent, &root->inode_tree);
3950                         RB_CLEAR_NODE(parent);
3951                         spin_unlock(&root->inode_lock);
3952                         goto again;
3953                 }
3954         }
3955         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3956         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3957         spin_unlock(&root->inode_lock);
3958 }
3959
3960 static void inode_tree_del(struct inode *inode)
3961 {
3962         struct btrfs_root *root = BTRFS_I(inode)->root;
3963         int empty = 0;
3964
3965         spin_lock(&root->inode_lock);
3966         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3967                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3968                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3969                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3970         }
3971         spin_unlock(&root->inode_lock);
3972
3973         /*
3974          * Free space cache has inodes in the tree root, but the tree root has a
3975          * root_refs of 0, so this could end up dropping the tree root as a
3976          * snapshot, so we need the extra !root->fs_info->tree_root check to
3977          * make sure we don't drop it.
3978          */
3979         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3980             root != root->fs_info->tree_root) {
3981                 synchronize_srcu(&root->fs_info->subvol_srcu);
3982                 spin_lock(&root->inode_lock);
3983                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3984                 spin_unlock(&root->inode_lock);
3985                 if (empty)
3986                         btrfs_add_dead_root(root);
3987         }
3988 }
3989
3990 int btrfs_invalidate_inodes(struct btrfs_root *root)
3991 {
3992         struct rb_node *node;
3993         struct rb_node *prev;
3994         struct btrfs_inode *entry;
3995         struct inode *inode;
3996         u64 objectid = 0;
3997
3998         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3999
4000         spin_lock(&root->inode_lock);
4001 again:
4002         node = root->inode_tree.rb_node;
4003         prev = NULL;
4004         while (node) {
4005                 prev = node;
4006                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4007
4008                 if (objectid < entry->vfs_inode.i_ino)
4009                         node = node->rb_left;
4010                 else if (objectid > entry->vfs_inode.i_ino)
4011                         node = node->rb_right;
4012                 else
4013                         break;
4014         }
4015         if (!node) {
4016                 while (prev) {
4017                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4018                         if (objectid <= entry->vfs_inode.i_ino) {
4019                                 node = prev;
4020                                 break;
4021                         }
4022                         prev = rb_next(prev);
4023                 }
4024         }
4025         while (node) {
4026                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4027                 objectid = entry->vfs_inode.i_ino + 1;
4028                 inode = igrab(&entry->vfs_inode);
4029                 if (inode) {
4030                         spin_unlock(&root->inode_lock);
4031                         if (atomic_read(&inode->i_count) > 1)
4032                                 d_prune_aliases(inode);
4033                         /*
4034                          * btrfs_drop_inode will have it removed from
4035                          * the inode cache when its usage count
4036                          * hits zero.
4037                          */
4038                         iput(inode);
4039                         cond_resched();
4040                         spin_lock(&root->inode_lock);
4041                         goto again;
4042                 }
4043
4044                 if (cond_resched_lock(&root->inode_lock))
4045                         goto again;
4046
4047                 node = rb_next(node);
4048         }
4049         spin_unlock(&root->inode_lock);
4050         return 0;
4051 }
4052
4053 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4054 {
4055         struct btrfs_iget_args *args = p;
4056         inode->i_ino = args->ino;
4057         BTRFS_I(inode)->root = args->root;
4058         btrfs_set_inode_space_info(args->root, inode);
4059         return 0;
4060 }
4061
4062 static int btrfs_find_actor(struct inode *inode, void *opaque)
4063 {
4064         struct btrfs_iget_args *args = opaque;
4065         return args->ino == inode->i_ino &&
4066                 args->root == BTRFS_I(inode)->root;
4067 }
4068
4069 static struct inode *btrfs_iget_locked(struct super_block *s,
4070                                        u64 objectid,
4071                                        struct btrfs_root *root)
4072 {
4073         struct inode *inode;
4074         struct btrfs_iget_args args;
4075         args.ino = objectid;
4076         args.root = root;
4077
4078         inode = iget5_locked(s, objectid, btrfs_find_actor,
4079                              btrfs_init_locked_inode,
4080                              (void *)&args);
4081         return inode;
4082 }
4083
4084 /* Get an inode object given its location and corresponding root.
4085  * Returns in *is_new if the inode was read from disk
4086  */
4087 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4088                          struct btrfs_root *root, int *new)
4089 {
4090         struct inode *inode;
4091
4092         inode = btrfs_iget_locked(s, location->objectid, root);
4093         if (!inode)
4094                 return ERR_PTR(-ENOMEM);
4095
4096         if (inode->i_state & I_NEW) {
4097                 BTRFS_I(inode)->root = root;
4098                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4099                 btrfs_read_locked_inode(inode);
4100                 inode_tree_add(inode);
4101                 unlock_new_inode(inode);
4102                 if (new)
4103                         *new = 1;
4104         }
4105
4106         return inode;
4107 }
4108
4109 static struct inode *new_simple_dir(struct super_block *s,
4110                                     struct btrfs_key *key,
4111                                     struct btrfs_root *root)
4112 {
4113         struct inode *inode = new_inode(s);
4114
4115         if (!inode)
4116                 return ERR_PTR(-ENOMEM);
4117
4118         BTRFS_I(inode)->root = root;
4119         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4120         BTRFS_I(inode)->dummy_inode = 1;
4121
4122         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4123         inode->i_op = &simple_dir_inode_operations;
4124         inode->i_fop = &simple_dir_operations;
4125         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4126         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4127
4128         return inode;
4129 }
4130
4131 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4132 {
4133         struct inode *inode;
4134         struct btrfs_root *root = BTRFS_I(dir)->root;
4135         struct btrfs_root *sub_root = root;
4136         struct btrfs_key location;
4137         int index;
4138         int ret;
4139
4140         if (dentry->d_name.len > BTRFS_NAME_LEN)
4141                 return ERR_PTR(-ENAMETOOLONG);
4142
4143         ret = btrfs_inode_by_name(dir, dentry, &location);
4144
4145         if (ret < 0)
4146                 return ERR_PTR(ret);
4147
4148         if (location.objectid == 0)
4149                 return NULL;
4150
4151         if (location.type == BTRFS_INODE_ITEM_KEY) {
4152                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4153                 return inode;
4154         }
4155
4156         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4157
4158         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4159         ret = fixup_tree_root_location(root, dir, dentry,
4160                                        &location, &sub_root);
4161         if (ret < 0) {
4162                 if (ret != -ENOENT)
4163                         inode = ERR_PTR(ret);
4164                 else
4165                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4166         } else {
4167                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4168         }
4169         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4170
4171         if (!IS_ERR(inode) && root != sub_root) {
4172                 down_read(&root->fs_info->cleanup_work_sem);
4173                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4174                         ret = btrfs_orphan_cleanup(sub_root);
4175                 up_read(&root->fs_info->cleanup_work_sem);
4176                 if (ret)
4177                         inode = ERR_PTR(ret);
4178         }
4179
4180         return inode;
4181 }
4182
4183 static int btrfs_dentry_delete(const struct dentry *dentry)
4184 {
4185         struct btrfs_root *root;
4186
4187         if (!dentry->d_inode && !IS_ROOT(dentry))
4188                 dentry = dentry->d_parent;
4189
4190         if (dentry->d_inode) {
4191                 root = BTRFS_I(dentry->d_inode)->root;
4192                 if (btrfs_root_refs(&root->root_item) == 0)
4193                         return 1;
4194         }
4195         return 0;
4196 }
4197
4198 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4199                                    struct nameidata *nd)
4200 {
4201         struct inode *inode;
4202
4203         inode = btrfs_lookup_dentry(dir, dentry);
4204         if (IS_ERR(inode))
4205                 return ERR_CAST(inode);
4206
4207         return d_splice_alias(inode, dentry);
4208 }
4209
4210 static unsigned char btrfs_filetype_table[] = {
4211         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4212 };
4213
4214 static int btrfs_real_readdir(struct file *filp, void *dirent,
4215                               filldir_t filldir)
4216 {
4217         struct inode *inode = filp->f_dentry->d_inode;
4218         struct btrfs_root *root = BTRFS_I(inode)->root;
4219         struct btrfs_item *item;
4220         struct btrfs_dir_item *di;
4221         struct btrfs_key key;
4222         struct btrfs_key found_key;
4223         struct btrfs_path *path;
4224         int ret;
4225         struct extent_buffer *leaf;
4226         int slot;
4227         unsigned char d_type;
4228         int over = 0;
4229         u32 di_cur;
4230         u32 di_total;
4231         u32 di_len;
4232         int key_type = BTRFS_DIR_INDEX_KEY;
4233         char tmp_name[32];
4234         char *name_ptr;
4235         int name_len;
4236
4237         /* FIXME, use a real flag for deciding about the key type */
4238         if (root->fs_info->tree_root == root)
4239                 key_type = BTRFS_DIR_ITEM_KEY;
4240
4241         /* special case for "." */
4242         if (filp->f_pos == 0) {
4243                 over = filldir(dirent, ".", 1,
4244                                1, inode->i_ino,
4245                                DT_DIR);
4246                 if (over)
4247                         return 0;
4248                 filp->f_pos = 1;
4249         }
4250         /* special case for .., just use the back ref */
4251         if (filp->f_pos == 1) {
4252                 u64 pino = parent_ino(filp->f_path.dentry);
4253                 over = filldir(dirent, "..", 2,
4254                                2, pino, DT_DIR);
4255                 if (over)
4256                         return 0;
4257                 filp->f_pos = 2;
4258         }
4259         path = btrfs_alloc_path();
4260         path->reada = 2;
4261
4262         btrfs_set_key_type(&key, key_type);
4263         key.offset = filp->f_pos;
4264         key.objectid = inode->i_ino;
4265
4266         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4267         if (ret < 0)
4268                 goto err;
4269
4270         while (1) {
4271                 leaf = path->nodes[0];
4272                 slot = path->slots[0];
4273                 if (slot >= btrfs_header_nritems(leaf)) {
4274                         ret = btrfs_next_leaf(root, path);
4275                         if (ret < 0)
4276                                 goto err;
4277                         else if (ret > 0)
4278                                 break;
4279                         continue;
4280                 }
4281
4282                 item = btrfs_item_nr(leaf, slot);
4283                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4284
4285                 if (found_key.objectid != key.objectid)
4286                         break;
4287                 if (btrfs_key_type(&found_key) != key_type)
4288                         break;
4289                 if (found_key.offset < filp->f_pos)
4290                         goto next;
4291
4292                 filp->f_pos = found_key.offset;
4293
4294                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4295                 di_cur = 0;
4296                 di_total = btrfs_item_size(leaf, item);
4297
4298                 while (di_cur < di_total) {
4299                         struct btrfs_key location;
4300
4301                         if (verify_dir_item(root, leaf, di))
4302                                 break;
4303
4304                         name_len = btrfs_dir_name_len(leaf, di);
4305                         if (name_len <= sizeof(tmp_name)) {
4306                                 name_ptr = tmp_name;
4307                         } else {
4308                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4309                                 if (!name_ptr) {
4310                                         ret = -ENOMEM;
4311                                         goto err;
4312                                 }
4313                         }
4314                         read_extent_buffer(leaf, name_ptr,
4315                                            (unsigned long)(di + 1), name_len);
4316
4317                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4318                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4319
4320                         /* is this a reference to our own snapshot? If so
4321                          * skip it
4322                          */
4323                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4324                             location.objectid == root->root_key.objectid) {
4325                                 over = 0;
4326                                 goto skip;
4327                         }
4328                         over = filldir(dirent, name_ptr, name_len,
4329                                        found_key.offset, location.objectid,
4330                                        d_type);
4331
4332 skip:
4333                         if (name_ptr != tmp_name)
4334                                 kfree(name_ptr);
4335
4336                         if (over)
4337                                 goto nopos;
4338                         di_len = btrfs_dir_name_len(leaf, di) +
4339                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4340                         di_cur += di_len;
4341                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4342                 }
4343 next:
4344                 path->slots[0]++;
4345         }
4346
4347         /* Reached end of directory/root. Bump pos past the last item. */
4348         if (key_type == BTRFS_DIR_INDEX_KEY)
4349                 /*
4350                  * 32-bit glibc will use getdents64, but then strtol -
4351                  * so the last number we can serve is this.
4352                  */
4353                 filp->f_pos = 0x7fffffff;
4354         else
4355                 filp->f_pos++;
4356 nopos:
4357         ret = 0;
4358 err:
4359         btrfs_free_path(path);
4360         return ret;
4361 }
4362
4363 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4364 {
4365         struct btrfs_root *root = BTRFS_I(inode)->root;
4366         struct btrfs_trans_handle *trans;
4367         int ret = 0;
4368         bool nolock = false;
4369
4370         if (BTRFS_I(inode)->dummy_inode)
4371                 return 0;
4372
4373         smp_mb();
4374         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4375
4376         if (wbc->sync_mode == WB_SYNC_ALL) {
4377                 if (nolock)
4378                         trans = btrfs_join_transaction_nolock(root, 1);
4379                 else
4380                         trans = btrfs_join_transaction(root, 1);
4381                 if (IS_ERR(trans))
4382                         return PTR_ERR(trans);
4383                 btrfs_set_trans_block_group(trans, inode);
4384                 if (nolock)
4385                         ret = btrfs_end_transaction_nolock(trans, root);
4386                 else
4387                         ret = btrfs_commit_transaction(trans, root);
4388         }
4389         return ret;
4390 }
4391
4392 /*
4393  * This is somewhat expensive, updating the tree every time the
4394  * inode changes.  But, it is most likely to find the inode in cache.
4395  * FIXME, needs more benchmarking...there are no reasons other than performance
4396  * to keep or drop this code.
4397  */
4398 void btrfs_dirty_inode(struct inode *inode)
4399 {
4400         struct btrfs_root *root = BTRFS_I(inode)->root;
4401         struct btrfs_trans_handle *trans;
4402         int ret;
4403
4404         if (BTRFS_I(inode)->dummy_inode)
4405                 return;
4406
4407         trans = btrfs_join_transaction(root, 1);
4408         BUG_ON(IS_ERR(trans));
4409         btrfs_set_trans_block_group(trans, inode);
4410
4411         ret = btrfs_update_inode(trans, root, inode);
4412         if (ret && ret == -ENOSPC) {
4413                 /* whoops, lets try again with the full transaction */
4414                 btrfs_end_transaction(trans, root);
4415                 trans = btrfs_start_transaction(root, 1);
4416                 if (IS_ERR(trans)) {
4417                         if (printk_ratelimit()) {
4418                                 printk(KERN_ERR "btrfs: fail to "
4419                                        "dirty  inode %lu error %ld\n",
4420                                        inode->i_ino, PTR_ERR(trans));
4421                         }
4422                         return;
4423                 }
4424                 btrfs_set_trans_block_group(trans, inode);
4425
4426                 ret = btrfs_update_inode(trans, root, inode);
4427                 if (ret) {
4428                         if (printk_ratelimit()) {
4429                                 printk(KERN_ERR "btrfs: fail to "
4430                                        "dirty  inode %lu error %d\n",
4431                                        inode->i_ino, ret);
4432                         }
4433                 }
4434         }
4435         btrfs_end_transaction(trans, root);
4436 }
4437
4438 /*
4439  * find the highest existing sequence number in a directory
4440  * and then set the in-memory index_cnt variable to reflect
4441  * free sequence numbers
4442  */
4443 static int btrfs_set_inode_index_count(struct inode *inode)
4444 {
4445         struct btrfs_root *root = BTRFS_I(inode)->root;
4446         struct btrfs_key key, found_key;
4447         struct btrfs_path *path;
4448         struct extent_buffer *leaf;
4449         int ret;
4450
4451         key.objectid = inode->i_ino;
4452         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4453         key.offset = (u64)-1;
4454
4455         path = btrfs_alloc_path();
4456         if (!path)
4457                 return -ENOMEM;
4458
4459         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4460         if (ret < 0)
4461                 goto out;
4462         /* FIXME: we should be able to handle this */
4463         if (ret == 0)
4464                 goto out;
4465         ret = 0;
4466
4467         /*
4468          * MAGIC NUMBER EXPLANATION:
4469          * since we search a directory based on f_pos we have to start at 2
4470          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4471          * else has to start at 2
4472          */
4473         if (path->slots[0] == 0) {
4474                 BTRFS_I(inode)->index_cnt = 2;
4475                 goto out;
4476         }
4477
4478         path->slots[0]--;
4479
4480         leaf = path->nodes[0];
4481         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4482
4483         if (found_key.objectid != inode->i_ino ||
4484             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4485                 BTRFS_I(inode)->index_cnt = 2;
4486                 goto out;
4487         }
4488
4489         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4490 out:
4491         btrfs_free_path(path);
4492         return ret;
4493 }
4494
4495 /*
4496  * helper to find a free sequence number in a given directory.  This current
4497  * code is very simple, later versions will do smarter things in the btree
4498  */
4499 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4500 {
4501         int ret = 0;
4502
4503         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4504                 ret = btrfs_set_inode_index_count(dir);
4505                 if (ret)
4506                         return ret;
4507         }
4508
4509         *index = BTRFS_I(dir)->index_cnt;
4510         BTRFS_I(dir)->index_cnt++;
4511
4512         return ret;
4513 }
4514
4515 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4516                                      struct btrfs_root *root,
4517                                      struct inode *dir,
4518                                      const char *name, int name_len,
4519                                      u64 ref_objectid, u64 objectid,
4520                                      u64 alloc_hint, int mode, u64 *index)
4521 {
4522         struct inode *inode;
4523         struct btrfs_inode_item *inode_item;
4524         struct btrfs_key *location;
4525         struct btrfs_path *path;
4526         struct btrfs_inode_ref *ref;
4527         struct btrfs_key key[2];
4528         u32 sizes[2];
4529         unsigned long ptr;
4530         int ret;
4531         int owner;
4532
4533         path = btrfs_alloc_path();
4534         BUG_ON(!path);
4535
4536         inode = new_inode(root->fs_info->sb);
4537         if (!inode) {
4538                 btrfs_free_path(path);
4539                 return ERR_PTR(-ENOMEM);
4540         }
4541
4542         if (dir) {
4543                 trace_btrfs_inode_request(dir);
4544
4545                 ret = btrfs_set_inode_index(dir, index);
4546                 if (ret) {
4547                         btrfs_free_path(path);
4548                         iput(inode);
4549                         return ERR_PTR(ret);
4550                 }
4551         }
4552         /*
4553          * index_cnt is ignored for everything but a dir,
4554          * btrfs_get_inode_index_count has an explanation for the magic
4555          * number
4556          */
4557         BTRFS_I(inode)->index_cnt = 2;
4558         BTRFS_I(inode)->root = root;
4559         BTRFS_I(inode)->generation = trans->transid;
4560         inode->i_generation = BTRFS_I(inode)->generation;
4561         btrfs_set_inode_space_info(root, inode);
4562
4563         if (mode & S_IFDIR)
4564                 owner = 0;
4565         else
4566                 owner = 1;
4567         BTRFS_I(inode)->block_group =
4568                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4569
4570         key[0].objectid = objectid;
4571         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4572         key[0].offset = 0;
4573
4574         key[1].objectid = objectid;
4575         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4576         key[1].offset = ref_objectid;
4577
4578         sizes[0] = sizeof(struct btrfs_inode_item);
4579         sizes[1] = name_len + sizeof(*ref);
4580
4581         path->leave_spinning = 1;
4582         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4583         if (ret != 0)
4584                 goto fail;
4585
4586         inode_init_owner(inode, dir, mode);
4587         inode->i_ino = objectid;
4588         inode_set_bytes(inode, 0);
4589         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4590         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4591                                   struct btrfs_inode_item);
4592         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4593
4594         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4595                              struct btrfs_inode_ref);
4596         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4597         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4598         ptr = (unsigned long)(ref + 1);
4599         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4600
4601         btrfs_mark_buffer_dirty(path->nodes[0]);
4602         btrfs_free_path(path);
4603
4604         location = &BTRFS_I(inode)->location;
4605         location->objectid = objectid;
4606         location->offset = 0;
4607         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4608
4609         btrfs_inherit_iflags(inode, dir);
4610
4611         if ((mode & S_IFREG)) {
4612                 if (btrfs_test_opt(root, NODATASUM))
4613                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4614                 if (btrfs_test_opt(root, NODATACOW) ||
4615                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4616                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4617         }
4618
4619         insert_inode_hash(inode);
4620         inode_tree_add(inode);
4621
4622         trace_btrfs_inode_new(inode);
4623
4624         return inode;
4625 fail:
4626         if (dir)
4627                 BTRFS_I(dir)->index_cnt--;
4628         btrfs_free_path(path);
4629         iput(inode);
4630         return ERR_PTR(ret);
4631 }
4632
4633 static inline u8 btrfs_inode_type(struct inode *inode)
4634 {
4635         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4636 }
4637
4638 /*
4639  * utility function to add 'inode' into 'parent_inode' with
4640  * a give name and a given sequence number.
4641  * if 'add_backref' is true, also insert a backref from the
4642  * inode to the parent directory.
4643  */
4644 int btrfs_add_link(struct btrfs_trans_handle *trans,
4645                    struct inode *parent_inode, struct inode *inode,
4646                    const char *name, int name_len, int add_backref, u64 index)
4647 {
4648         int ret = 0;
4649         struct btrfs_key key;
4650         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4651
4652         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4653                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4654         } else {
4655                 key.objectid = inode->i_ino;
4656                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4657                 key.offset = 0;
4658         }
4659
4660         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4661                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4662                                          key.objectid, root->root_key.objectid,
4663                                          parent_inode->i_ino,
4664                                          index, name, name_len);
4665         } else if (add_backref) {
4666                 ret = btrfs_insert_inode_ref(trans, root,
4667                                              name, name_len, inode->i_ino,
4668                                              parent_inode->i_ino, index);
4669         }
4670
4671         if (ret == 0) {
4672                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4673                                             parent_inode->i_ino, &key,
4674                                             btrfs_inode_type(inode), index);
4675                 BUG_ON(ret);
4676
4677                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4678                                    name_len * 2);
4679                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4680                 ret = btrfs_update_inode(trans, root, parent_inode);
4681         }
4682         return ret;
4683 }
4684
4685 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4686                             struct inode *dir, struct dentry *dentry,
4687                             struct inode *inode, int backref, u64 index)
4688 {
4689         int err = btrfs_add_link(trans, dir, inode,
4690                                  dentry->d_name.name, dentry->d_name.len,
4691                                  backref, index);
4692         if (!err) {
4693                 d_instantiate(dentry, inode);
4694                 return 0;
4695         }
4696         if (err > 0)
4697                 err = -EEXIST;
4698         return err;
4699 }
4700
4701 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4702                         int mode, dev_t rdev)
4703 {
4704         struct btrfs_trans_handle *trans;
4705         struct btrfs_root *root = BTRFS_I(dir)->root;
4706         struct inode *inode = NULL;
4707         int err;
4708         int drop_inode = 0;
4709         u64 objectid;
4710         unsigned long nr = 0;
4711         u64 index = 0;
4712
4713         if (!new_valid_dev(rdev))
4714                 return -EINVAL;
4715
4716         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4717         if (err)
4718                 return err;
4719
4720         /*
4721          * 2 for inode item and ref
4722          * 2 for dir items
4723          * 1 for xattr if selinux is on
4724          */
4725         trans = btrfs_start_transaction(root, 5);
4726         if (IS_ERR(trans))
4727                 return PTR_ERR(trans);
4728
4729         btrfs_set_trans_block_group(trans, dir);
4730
4731         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4732                                 dentry->d_name.len, dir->i_ino, objectid,
4733                                 BTRFS_I(dir)->block_group, mode, &index);
4734         if (IS_ERR(inode)) {
4735                 err = PTR_ERR(inode);
4736                 goto out_unlock;
4737         }
4738
4739         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4740         if (err) {
4741                 drop_inode = 1;
4742                 goto out_unlock;
4743         }
4744
4745         btrfs_set_trans_block_group(trans, inode);
4746         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4747         if (err)
4748                 drop_inode = 1;
4749         else {
4750                 inode->i_op = &btrfs_special_inode_operations;
4751                 init_special_inode(inode, inode->i_mode, rdev);
4752                 btrfs_update_inode(trans, root, inode);
4753         }
4754         btrfs_update_inode_block_group(trans, inode);
4755         btrfs_update_inode_block_group(trans, dir);
4756 out_unlock:
4757         nr = trans->blocks_used;
4758         btrfs_end_transaction_throttle(trans, root);
4759         btrfs_btree_balance_dirty(root, nr);
4760         if (drop_inode) {
4761                 inode_dec_link_count(inode);
4762                 iput(inode);
4763         }
4764         return err;
4765 }
4766
4767 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4768                         int mode, struct nameidata *nd)
4769 {
4770         struct btrfs_trans_handle *trans;
4771         struct btrfs_root *root = BTRFS_I(dir)->root;
4772         struct inode *inode = NULL;
4773         int drop_inode = 0;
4774         int err;
4775         unsigned long nr = 0;
4776         u64 objectid;
4777         u64 index = 0;
4778
4779         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4780         if (err)
4781                 return err;
4782         /*
4783          * 2 for inode item and ref
4784          * 2 for dir items
4785          * 1 for xattr if selinux is on
4786          */
4787         trans = btrfs_start_transaction(root, 5);
4788         if (IS_ERR(trans))
4789                 return PTR_ERR(trans);
4790
4791         btrfs_set_trans_block_group(trans, dir);
4792
4793         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4794                                 dentry->d_name.len, dir->i_ino, objectid,
4795                                 BTRFS_I(dir)->block_group, mode, &index);
4796         if (IS_ERR(inode)) {
4797                 err = PTR_ERR(inode);
4798                 goto out_unlock;
4799         }
4800
4801         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4802         if (err) {
4803                 drop_inode = 1;
4804                 goto out_unlock;
4805         }
4806
4807         btrfs_set_trans_block_group(trans, inode);
4808         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4809         if (err)
4810                 drop_inode = 1;
4811         else {
4812                 inode->i_mapping->a_ops = &btrfs_aops;
4813                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4814                 inode->i_fop = &btrfs_file_operations;
4815                 inode->i_op = &btrfs_file_inode_operations;
4816                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4817         }
4818         btrfs_update_inode_block_group(trans, inode);
4819         btrfs_update_inode_block_group(trans, dir);
4820 out_unlock:
4821         nr = trans->blocks_used;
4822         btrfs_end_transaction_throttle(trans, root);
4823         if (drop_inode) {
4824                 inode_dec_link_count(inode);
4825                 iput(inode);
4826         }
4827         btrfs_btree_balance_dirty(root, nr);
4828         return err;
4829 }
4830
4831 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4832                       struct dentry *dentry)
4833 {
4834         struct btrfs_trans_handle *trans;
4835         struct btrfs_root *root = BTRFS_I(dir)->root;
4836         struct inode *inode = old_dentry->d_inode;
4837         u64 index;
4838         unsigned long nr = 0;
4839         int err;
4840         int drop_inode = 0;
4841
4842         /* do not allow sys_link's with other subvols of the same device */
4843         if (root->objectid != BTRFS_I(inode)->root->objectid)
4844                 return -EXDEV;
4845
4846         if (inode->i_nlink == ~0U)
4847                 return -EMLINK;
4848
4849         err = btrfs_set_inode_index(dir, &index);
4850         if (err)
4851                 goto fail;
4852
4853         /*
4854          * 2 items for inode and inode ref
4855          * 2 items for dir items
4856          * 1 item for parent inode
4857          */
4858         trans = btrfs_start_transaction(root, 5);
4859         if (IS_ERR(trans)) {
4860                 err = PTR_ERR(trans);
4861                 goto fail;
4862         }
4863
4864         btrfs_inc_nlink(inode);
4865         inode->i_ctime = CURRENT_TIME;
4866
4867         btrfs_set_trans_block_group(trans, dir);
4868         ihold(inode);
4869
4870         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4871
4872         if (err) {
4873                 drop_inode = 1;
4874         } else {
4875                 struct dentry *parent = dget_parent(dentry);
4876                 btrfs_update_inode_block_group(trans, dir);
4877                 err = btrfs_update_inode(trans, root, inode);
4878                 BUG_ON(err);
4879                 btrfs_log_new_name(trans, inode, NULL, parent);
4880                 dput(parent);
4881         }
4882
4883         nr = trans->blocks_used;
4884         btrfs_end_transaction_throttle(trans, root);
4885 fail:
4886         if (drop_inode) {
4887                 inode_dec_link_count(inode);
4888                 iput(inode);
4889         }
4890         btrfs_btree_balance_dirty(root, nr);
4891         return err;
4892 }
4893
4894 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4895 {
4896         struct inode *inode = NULL;
4897         struct btrfs_trans_handle *trans;
4898         struct btrfs_root *root = BTRFS_I(dir)->root;
4899         int err = 0;
4900         int drop_on_err = 0;
4901         u64 objectid = 0;
4902         u64 index = 0;
4903         unsigned long nr = 1;
4904
4905         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4906         if (err)
4907                 return err;
4908
4909         /*
4910          * 2 items for inode and ref
4911          * 2 items for dir items
4912          * 1 for xattr if selinux is on
4913          */
4914         trans = btrfs_start_transaction(root, 5);
4915         if (IS_ERR(trans))
4916                 return PTR_ERR(trans);
4917         btrfs_set_trans_block_group(trans, dir);
4918
4919         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4920                                 dentry->d_name.len, dir->i_ino, objectid,
4921                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4922                                 &index);
4923         if (IS_ERR(inode)) {
4924                 err = PTR_ERR(inode);
4925                 goto out_fail;
4926         }
4927
4928         drop_on_err = 1;
4929
4930         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4931         if (err)
4932                 goto out_fail;
4933
4934         inode->i_op = &btrfs_dir_inode_operations;
4935         inode->i_fop = &btrfs_dir_file_operations;
4936         btrfs_set_trans_block_group(trans, inode);
4937
4938         btrfs_i_size_write(inode, 0);
4939         err = btrfs_update_inode(trans, root, inode);
4940         if (err)
4941                 goto out_fail;
4942
4943         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4944                              dentry->d_name.len, 0, index);
4945         if (err)
4946                 goto out_fail;
4947
4948         d_instantiate(dentry, inode);
4949         drop_on_err = 0;
4950         btrfs_update_inode_block_group(trans, inode);
4951         btrfs_update_inode_block_group(trans, dir);
4952
4953 out_fail:
4954         nr = trans->blocks_used;
4955         btrfs_end_transaction_throttle(trans, root);
4956         if (drop_on_err)
4957                 iput(inode);
4958         btrfs_btree_balance_dirty(root, nr);
4959         return err;
4960 }
4961
4962 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4963  * and an extent that you want to insert, deal with overlap and insert
4964  * the new extent into the tree.
4965  */
4966 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4967                                 struct extent_map *existing,
4968                                 struct extent_map *em,
4969                                 u64 map_start, u64 map_len)
4970 {
4971         u64 start_diff;
4972
4973         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4974         start_diff = map_start - em->start;
4975         em->start = map_start;
4976         em->len = map_len;
4977         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4978             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4979                 em->block_start += start_diff;
4980                 em->block_len -= start_diff;
4981         }
4982         return add_extent_mapping(em_tree, em);
4983 }
4984
4985 static noinline int uncompress_inline(struct btrfs_path *path,
4986                                       struct inode *inode, struct page *page,
4987                                       size_t pg_offset, u64 extent_offset,
4988                                       struct btrfs_file_extent_item *item)
4989 {
4990         int ret;
4991         struct extent_buffer *leaf = path->nodes[0];
4992         char *tmp;
4993         size_t max_size;
4994         unsigned long inline_size;
4995         unsigned long ptr;
4996         int compress_type;
4997
4998         WARN_ON(pg_offset != 0);
4999         compress_type = btrfs_file_extent_compression(leaf, item);
5000         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5001         inline_size = btrfs_file_extent_inline_item_len(leaf,
5002                                         btrfs_item_nr(leaf, path->slots[0]));
5003         tmp = kmalloc(inline_size, GFP_NOFS);
5004         if (!tmp)
5005                 return -ENOMEM;
5006         ptr = btrfs_file_extent_inline_start(item);
5007
5008         read_extent_buffer(leaf, tmp, ptr, inline_size);
5009
5010         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5011         ret = btrfs_decompress(compress_type, tmp, page,
5012                                extent_offset, inline_size, max_size);
5013         if (ret) {
5014                 char *kaddr = kmap_atomic(page, KM_USER0);
5015                 unsigned long copy_size = min_t(u64,
5016                                   PAGE_CACHE_SIZE - pg_offset,
5017                                   max_size - extent_offset);
5018                 memset(kaddr + pg_offset, 0, copy_size);
5019                 kunmap_atomic(kaddr, KM_USER0);
5020         }
5021         kfree(tmp);
5022         return 0;
5023 }
5024
5025 /*
5026  * a bit scary, this does extent mapping from logical file offset to the disk.
5027  * the ugly parts come from merging extents from the disk with the in-ram
5028  * representation.  This gets more complex because of the data=ordered code,
5029  * where the in-ram extents might be locked pending data=ordered completion.
5030  *
5031  * This also copies inline extents directly into the page.
5032  */
5033
5034 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5035                                     size_t pg_offset, u64 start, u64 len,
5036                                     int create)
5037 {
5038         int ret;
5039         int err = 0;
5040         u64 bytenr;
5041         u64 extent_start = 0;
5042         u64 extent_end = 0;
5043         u64 objectid = inode->i_ino;
5044         u32 found_type;
5045         struct btrfs_path *path = NULL;
5046         struct btrfs_root *root = BTRFS_I(inode)->root;
5047         struct btrfs_file_extent_item *item;
5048         struct extent_buffer *leaf;
5049         struct btrfs_key found_key;
5050         struct extent_map *em = NULL;
5051         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5052         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5053         struct btrfs_trans_handle *trans = NULL;
5054         int compress_type;
5055
5056 again:
5057         read_lock(&em_tree->lock);
5058         em = lookup_extent_mapping(em_tree, start, len);
5059         if (em)
5060                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5061         read_unlock(&em_tree->lock);
5062
5063         if (em) {
5064                 if (em->start > start || em->start + em->len <= start)
5065                         free_extent_map(em);
5066                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5067                         free_extent_map(em);
5068                 else
5069                         goto out;
5070         }
5071         em = alloc_extent_map(GFP_NOFS);
5072         if (!em) {
5073                 err = -ENOMEM;
5074                 goto out;
5075         }
5076         em->bdev = root->fs_info->fs_devices->latest_bdev;
5077         em->start = EXTENT_MAP_HOLE;
5078         em->orig_start = EXTENT_MAP_HOLE;
5079         em->len = (u64)-1;
5080         em->block_len = (u64)-1;
5081
5082         if (!path) {
5083                 path = btrfs_alloc_path();
5084                 BUG_ON(!path);
5085         }
5086
5087         ret = btrfs_lookup_file_extent(trans, root, path,
5088                                        objectid, start, trans != NULL);
5089         if (ret < 0) {
5090                 err = ret;
5091                 goto out;
5092         }
5093
5094         if (ret != 0) {
5095                 if (path->slots[0] == 0)
5096                         goto not_found;
5097                 path->slots[0]--;
5098         }
5099
5100         leaf = path->nodes[0];
5101         item = btrfs_item_ptr(leaf, path->slots[0],
5102                               struct btrfs_file_extent_item);
5103         /* are we inside the extent that was found? */
5104         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5105         found_type = btrfs_key_type(&found_key);
5106         if (found_key.objectid != objectid ||
5107             found_type != BTRFS_EXTENT_DATA_KEY) {
5108                 goto not_found;
5109         }
5110
5111         found_type = btrfs_file_extent_type(leaf, item);
5112         extent_start = found_key.offset;
5113         compress_type = btrfs_file_extent_compression(leaf, item);
5114         if (found_type == BTRFS_FILE_EXTENT_REG ||
5115             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5116                 extent_end = extent_start +
5117                        btrfs_file_extent_num_bytes(leaf, item);
5118         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5119                 size_t size;
5120                 size = btrfs_file_extent_inline_len(leaf, item);
5121                 extent_end = (extent_start + size + root->sectorsize - 1) &
5122                         ~((u64)root->sectorsize - 1);
5123         }
5124
5125         if (start >= extent_end) {
5126                 path->slots[0]++;
5127                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5128                         ret = btrfs_next_leaf(root, path);
5129                         if (ret < 0) {
5130                                 err = ret;
5131                                 goto out;
5132                         }
5133                         if (ret > 0)
5134                                 goto not_found;
5135                         leaf = path->nodes[0];
5136                 }
5137                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5138                 if (found_key.objectid != objectid ||
5139                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5140                         goto not_found;
5141                 if (start + len <= found_key.offset)
5142                         goto not_found;
5143                 em->start = start;
5144                 em->len = found_key.offset - start;
5145                 goto not_found_em;
5146         }
5147
5148         if (found_type == BTRFS_FILE_EXTENT_REG ||
5149             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5150                 em->start = extent_start;
5151                 em->len = extent_end - extent_start;
5152                 em->orig_start = extent_start -
5153                                  btrfs_file_extent_offset(leaf, item);
5154                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5155                 if (bytenr == 0) {
5156                         em->block_start = EXTENT_MAP_HOLE;
5157                         goto insert;
5158                 }
5159                 if (compress_type != BTRFS_COMPRESS_NONE) {
5160                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5161                         em->compress_type = compress_type;
5162                         em->block_start = bytenr;
5163                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5164                                                                          item);
5165                 } else {
5166                         bytenr += btrfs_file_extent_offset(leaf, item);
5167                         em->block_start = bytenr;
5168                         em->block_len = em->len;
5169                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5170                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5171                 }
5172                 goto insert;
5173         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5174                 unsigned long ptr;
5175                 char *map;
5176                 size_t size;
5177                 size_t extent_offset;
5178                 size_t copy_size;
5179
5180                 em->block_start = EXTENT_MAP_INLINE;
5181                 if (!page || create) {
5182                         em->start = extent_start;
5183                         em->len = extent_end - extent_start;
5184                         goto out;
5185                 }
5186
5187                 size = btrfs_file_extent_inline_len(leaf, item);
5188                 extent_offset = page_offset(page) + pg_offset - extent_start;
5189                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5190                                 size - extent_offset);
5191                 em->start = extent_start + extent_offset;
5192                 em->len = (copy_size + root->sectorsize - 1) &
5193                         ~((u64)root->sectorsize - 1);
5194                 em->orig_start = EXTENT_MAP_INLINE;
5195                 if (compress_type) {
5196                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5197                         em->compress_type = compress_type;
5198                 }
5199                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5200                 if (create == 0 && !PageUptodate(page)) {
5201                         if (btrfs_file_extent_compression(leaf, item) !=
5202                             BTRFS_COMPRESS_NONE) {
5203                                 ret = uncompress_inline(path, inode, page,
5204                                                         pg_offset,
5205                                                         extent_offset, item);
5206                                 BUG_ON(ret);
5207                         } else {
5208                                 map = kmap(page);
5209                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5210                                                    copy_size);
5211                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5212                                         memset(map + pg_offset + copy_size, 0,
5213                                                PAGE_CACHE_SIZE - pg_offset -
5214                                                copy_size);
5215                                 }
5216                                 kunmap(page);
5217                         }
5218                         flush_dcache_page(page);
5219                 } else if (create && PageUptodate(page)) {
5220                         WARN_ON(1);
5221                         if (!trans) {
5222                                 kunmap(page);
5223                                 free_extent_map(em);
5224                                 em = NULL;
5225                                 btrfs_release_path(root, path);
5226                                 trans = btrfs_join_transaction(root, 1);
5227                                 if (IS_ERR(trans))
5228                                         return ERR_CAST(trans);
5229                                 goto again;
5230                         }
5231                         map = kmap(page);
5232                         write_extent_buffer(leaf, map + pg_offset, ptr,
5233                                             copy_size);
5234                         kunmap(page);
5235                         btrfs_mark_buffer_dirty(leaf);
5236                 }
5237                 set_extent_uptodate(io_tree, em->start,
5238                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5239                 goto insert;
5240         } else {
5241                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5242                 WARN_ON(1);
5243         }
5244 not_found:
5245         em->start = start;
5246         em->len = len;
5247 not_found_em:
5248         em->block_start = EXTENT_MAP_HOLE;
5249         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5250 insert:
5251         btrfs_release_path(root, path);
5252         if (em->start > start || extent_map_end(em) <= start) {
5253                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5254                        "[%llu %llu]\n", (unsigned long long)em->start,
5255                        (unsigned long long)em->len,
5256                        (unsigned long long)start,
5257                        (unsigned long long)len);
5258                 err = -EIO;
5259                 goto out;
5260         }
5261
5262         err = 0;
5263         write_lock(&em_tree->lock);
5264         ret = add_extent_mapping(em_tree, em);
5265         /* it is possible that someone inserted the extent into the tree
5266          * while we had the lock dropped.  It is also possible that
5267          * an overlapping map exists in the tree
5268          */
5269         if (ret == -EEXIST) {
5270                 struct extent_map *existing;
5271
5272                 ret = 0;
5273
5274                 existing = lookup_extent_mapping(em_tree, start, len);
5275                 if (existing && (existing->start > start ||
5276                     existing->start + existing->len <= start)) {
5277                         free_extent_map(existing);
5278                         existing = NULL;
5279                 }
5280                 if (!existing) {
5281                         existing = lookup_extent_mapping(em_tree, em->start,
5282                                                          em->len);
5283                         if (existing) {
5284                                 err = merge_extent_mapping(em_tree, existing,
5285                                                            em, start,
5286                                                            root->sectorsize);
5287                                 free_extent_map(existing);
5288                                 if (err) {
5289                                         free_extent_map(em);
5290                                         em = NULL;
5291                                 }
5292                         } else {
5293                                 err = -EIO;
5294                                 free_extent_map(em);
5295                                 em = NULL;
5296                         }
5297                 } else {
5298                         free_extent_map(em);
5299                         em = existing;
5300                         err = 0;
5301                 }
5302         }
5303         write_unlock(&em_tree->lock);
5304 out:
5305
5306         trace_btrfs_get_extent(root, em);
5307
5308         if (path)
5309                 btrfs_free_path(path);
5310         if (trans) {
5311                 ret = btrfs_end_transaction(trans, root);
5312                 if (!err)
5313                         err = ret;
5314         }
5315         if (err) {
5316                 free_extent_map(em);
5317                 return ERR_PTR(err);
5318         }
5319         return em;
5320 }
5321
5322 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5323                                            size_t pg_offset, u64 start, u64 len,
5324                                            int create)
5325 {
5326         struct extent_map *em;
5327         struct extent_map *hole_em = NULL;
5328         u64 range_start = start;
5329         u64 end;
5330         u64 found;
5331         u64 found_end;
5332         int err = 0;
5333
5334         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5335         if (IS_ERR(em))
5336                 return em;
5337         if (em) {
5338                 /*
5339                  * if our em maps to a hole, there might
5340                  * actually be delalloc bytes behind it
5341                  */
5342                 if (em->block_start != EXTENT_MAP_HOLE)
5343                         return em;
5344                 else
5345                         hole_em = em;
5346         }
5347
5348         /* check to see if we've wrapped (len == -1 or similar) */
5349         end = start + len;
5350         if (end < start)
5351                 end = (u64)-1;
5352         else
5353                 end -= 1;
5354
5355         em = NULL;
5356
5357         /* ok, we didn't find anything, lets look for delalloc */
5358         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5359                                  end, len, EXTENT_DELALLOC, 1);
5360         found_end = range_start + found;
5361         if (found_end < range_start)
5362                 found_end = (u64)-1;
5363
5364         /*
5365          * we didn't find anything useful, return
5366          * the original results from get_extent()
5367          */
5368         if (range_start > end || found_end <= start) {
5369                 em = hole_em;
5370                 hole_em = NULL;
5371                 goto out;
5372         }
5373
5374         /* adjust the range_start to make sure it doesn't
5375          * go backwards from the start they passed in
5376          */
5377         range_start = max(start,range_start);
5378         found = found_end - range_start;
5379
5380         if (found > 0) {
5381                 u64 hole_start = start;
5382                 u64 hole_len = len;
5383
5384                 em = alloc_extent_map(GFP_NOFS);
5385                 if (!em) {
5386                         err = -ENOMEM;
5387                         goto out;
5388                 }
5389                 /*
5390                  * when btrfs_get_extent can't find anything it
5391                  * returns one huge hole
5392                  *
5393                  * make sure what it found really fits our range, and
5394                  * adjust to make sure it is based on the start from
5395                  * the caller
5396                  */
5397                 if (hole_em) {
5398                         u64 calc_end = extent_map_end(hole_em);
5399
5400                         if (calc_end <= start || (hole_em->start > end)) {
5401                                 free_extent_map(hole_em);
5402                                 hole_em = NULL;
5403                         } else {
5404                                 hole_start = max(hole_em->start, start);
5405                                 hole_len = calc_end - hole_start;
5406                         }
5407                 }
5408                 em->bdev = NULL;
5409                 if (hole_em && range_start > hole_start) {
5410                         /* our hole starts before our delalloc, so we
5411                          * have to return just the parts of the hole
5412                          * that go until  the delalloc starts
5413                          */
5414                         em->len = min(hole_len,
5415                                       range_start - hole_start);
5416                         em->start = hole_start;
5417                         em->orig_start = hole_start;
5418                         /*
5419                          * don't adjust block start at all,
5420                          * it is fixed at EXTENT_MAP_HOLE
5421                          */
5422                         em->block_start = hole_em->block_start;
5423                         em->block_len = hole_len;
5424                 } else {
5425                         em->start = range_start;
5426                         em->len = found;
5427                         em->orig_start = range_start;
5428                         em->block_start = EXTENT_MAP_DELALLOC;
5429                         em->block_len = found;
5430                 }
5431         } else if (hole_em) {
5432                 return hole_em;
5433         }
5434 out:
5435
5436         free_extent_map(hole_em);
5437         if (err) {
5438                 free_extent_map(em);
5439                 return ERR_PTR(err);
5440         }
5441         return em;
5442 }
5443
5444 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5445                                                   struct extent_map *em,
5446                                                   u64 start, u64 len)
5447 {
5448         struct btrfs_root *root = BTRFS_I(inode)->root;
5449         struct btrfs_trans_handle *trans;
5450         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5451         struct btrfs_key ins;
5452         u64 alloc_hint;
5453         int ret;
5454         bool insert = false;
5455
5456         /*
5457          * Ok if the extent map we looked up is a hole and is for the exact
5458          * range we want, there is no reason to allocate a new one, however if
5459          * it is not right then we need to free this one and drop the cache for
5460          * our range.
5461          */
5462         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5463             em->len != len) {
5464                 free_extent_map(em);
5465                 em = NULL;
5466                 insert = true;
5467                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5468         }
5469
5470         trans = btrfs_join_transaction(root, 0);
5471         if (IS_ERR(trans))
5472                 return ERR_CAST(trans);
5473
5474         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5475
5476         alloc_hint = get_extent_allocation_hint(inode, start, len);
5477         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5478                                    alloc_hint, (u64)-1, &ins, 1);
5479         if (ret) {
5480                 em = ERR_PTR(ret);
5481                 goto out;
5482         }
5483
5484         if (!em) {
5485                 em = alloc_extent_map(GFP_NOFS);
5486                 if (!em) {
5487                         em = ERR_PTR(-ENOMEM);
5488                         goto out;
5489                 }
5490         }
5491
5492         em->start = start;
5493         em->orig_start = em->start;
5494         em->len = ins.offset;
5495
5496         em->block_start = ins.objectid;
5497         em->block_len = ins.offset;
5498         em->bdev = root->fs_info->fs_devices->latest_bdev;
5499
5500         /*
5501          * We need to do this because if we're using the original em we searched
5502          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5503          */
5504         em->flags = 0;
5505         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5506
5507         while (insert) {
5508                 write_lock(&em_tree->lock);
5509                 ret = add_extent_mapping(em_tree, em);
5510                 write_unlock(&em_tree->lock);
5511                 if (ret != -EEXIST)
5512                         break;
5513                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5514         }
5515
5516         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5517                                            ins.offset, ins.offset, 0);
5518         if (ret) {
5519                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5520                 em = ERR_PTR(ret);
5521         }
5522 out:
5523         btrfs_end_transaction(trans, root);
5524         return em;
5525 }
5526
5527 /*
5528  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5529  * block must be cow'd
5530  */
5531 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5532                                       struct inode *inode, u64 offset, u64 len)
5533 {
5534         struct btrfs_path *path;
5535         int ret;
5536         struct extent_buffer *leaf;
5537         struct btrfs_root *root = BTRFS_I(inode)->root;
5538         struct btrfs_file_extent_item *fi;
5539         struct btrfs_key key;
5540         u64 disk_bytenr;
5541         u64 backref_offset;
5542         u64 extent_end;
5543         u64 num_bytes;
5544         int slot;
5545         int found_type;
5546
5547         path = btrfs_alloc_path();
5548         if (!path)
5549                 return -ENOMEM;
5550
5551         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5552                                        offset, 0);
5553         if (ret < 0)
5554                 goto out;
5555
5556         slot = path->slots[0];
5557         if (ret == 1) {
5558                 if (slot == 0) {
5559                         /* can't find the item, must cow */
5560                         ret = 0;
5561                         goto out;
5562                 }
5563                 slot--;
5564         }
5565         ret = 0;
5566         leaf = path->nodes[0];
5567         btrfs_item_key_to_cpu(leaf, &key, slot);
5568         if (key.objectid != inode->i_ino ||
5569             key.type != BTRFS_EXTENT_DATA_KEY) {
5570                 /* not our file or wrong item type, must cow */
5571                 goto out;
5572         }
5573
5574         if (key.offset > offset) {
5575                 /* Wrong offset, must cow */
5576                 goto out;
5577         }
5578
5579         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5580         found_type = btrfs_file_extent_type(leaf, fi);
5581         if (found_type != BTRFS_FILE_EXTENT_REG &&
5582             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5583                 /* not a regular extent, must cow */
5584                 goto out;
5585         }
5586         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5587         backref_offset = btrfs_file_extent_offset(leaf, fi);
5588
5589         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5590         if (extent_end < offset + len) {
5591                 /* extent doesn't include our full range, must cow */
5592                 goto out;
5593         }
5594
5595         if (btrfs_extent_readonly(root, disk_bytenr))
5596                 goto out;
5597
5598         /*
5599          * look for other files referencing this extent, if we
5600          * find any we must cow
5601          */
5602         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5603                                   key.offset - backref_offset, disk_bytenr))
5604                 goto out;
5605
5606         /*
5607          * adjust disk_bytenr and num_bytes to cover just the bytes
5608          * in this extent we are about to write.  If there
5609          * are any csums in that range we have to cow in order
5610          * to keep the csums correct
5611          */
5612         disk_bytenr += backref_offset;
5613         disk_bytenr += offset - key.offset;
5614         num_bytes = min(offset + len, extent_end) - offset;
5615         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5616                                 goto out;
5617         /*
5618          * all of the above have passed, it is safe to overwrite this extent
5619          * without cow
5620          */
5621         ret = 1;
5622 out:
5623         btrfs_free_path(path);
5624         return ret;
5625 }
5626
5627 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5628                                    struct buffer_head *bh_result, int create)
5629 {
5630         struct extent_map *em;
5631         struct btrfs_root *root = BTRFS_I(inode)->root;
5632         u64 start = iblock << inode->i_blkbits;
5633         u64 len = bh_result->b_size;
5634         struct btrfs_trans_handle *trans;
5635
5636         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5637         if (IS_ERR(em))
5638                 return PTR_ERR(em);
5639
5640         /*
5641          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5642          * io.  INLINE is special, and we could probably kludge it in here, but
5643          * it's still buffered so for safety lets just fall back to the generic
5644          * buffered path.
5645          *
5646          * For COMPRESSED we _have_ to read the entire extent in so we can
5647          * decompress it, so there will be buffering required no matter what we
5648          * do, so go ahead and fallback to buffered.
5649          *
5650          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5651          * to buffered IO.  Don't blame me, this is the price we pay for using
5652          * the generic code.
5653          */
5654         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5655             em->block_start == EXTENT_MAP_INLINE) {
5656                 free_extent_map(em);
5657                 return -ENOTBLK;
5658         }
5659
5660         /* Just a good old fashioned hole, return */
5661         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5662                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5663                 free_extent_map(em);
5664                 /* DIO will do one hole at a time, so just unlock a sector */
5665                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5666                               start + root->sectorsize - 1, GFP_NOFS);
5667                 return 0;
5668         }
5669
5670         /*
5671          * We don't allocate a new extent in the following cases
5672          *
5673          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5674          * existing extent.
5675          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5676          * just use the extent.
5677          *
5678          */
5679         if (!create) {
5680                 len = em->len - (start - em->start);
5681                 goto map;
5682         }
5683
5684         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5685             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5686              em->block_start != EXTENT_MAP_HOLE)) {
5687                 int type;
5688                 int ret;
5689                 u64 block_start;
5690
5691                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5692                         type = BTRFS_ORDERED_PREALLOC;
5693                 else
5694                         type = BTRFS_ORDERED_NOCOW;
5695                 len = min(len, em->len - (start - em->start));
5696                 block_start = em->block_start + (start - em->start);
5697
5698                 /*
5699                  * we're not going to log anything, but we do need
5700                  * to make sure the current transaction stays open
5701                  * while we look for nocow cross refs
5702                  */
5703                 trans = btrfs_join_transaction(root, 0);
5704                 if (IS_ERR(trans))
5705                         goto must_cow;
5706
5707                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5708                         ret = btrfs_add_ordered_extent_dio(inode, start,
5709                                            block_start, len, len, type);
5710                         btrfs_end_transaction(trans, root);
5711                         if (ret) {
5712                                 free_extent_map(em);
5713                                 return ret;
5714                         }
5715                         goto unlock;
5716                 }
5717                 btrfs_end_transaction(trans, root);
5718         }
5719 must_cow:
5720         /*
5721          * this will cow the extent, reset the len in case we changed
5722          * it above
5723          */
5724         len = bh_result->b_size;
5725         em = btrfs_new_extent_direct(inode, em, start, len);
5726         if (IS_ERR(em))
5727                 return PTR_ERR(em);
5728         len = min(len, em->len - (start - em->start));
5729 unlock:
5730         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5731                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5732                           0, NULL, GFP_NOFS);
5733 map:
5734         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5735                 inode->i_blkbits;
5736         bh_result->b_size = len;
5737         bh_result->b_bdev = em->bdev;
5738         set_buffer_mapped(bh_result);
5739         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5740                 set_buffer_new(bh_result);
5741
5742         free_extent_map(em);
5743
5744         return 0;
5745 }
5746
5747 struct btrfs_dio_private {
5748         struct inode *inode;
5749         u64 logical_offset;
5750         u64 disk_bytenr;
5751         u64 bytes;
5752         u32 *csums;
5753         void *private;
5754
5755         /* number of bios pending for this dio */
5756         atomic_t pending_bios;
5757
5758         /* IO errors */
5759         int errors;
5760
5761         struct bio *orig_bio;
5762 };
5763
5764 static void btrfs_endio_direct_read(struct bio *bio, int err)
5765 {
5766         struct btrfs_dio_private *dip = bio->bi_private;
5767         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5768         struct bio_vec *bvec = bio->bi_io_vec;
5769         struct inode *inode = dip->inode;
5770         struct btrfs_root *root = BTRFS_I(inode)->root;
5771         u64 start;
5772         u32 *private = dip->csums;
5773
5774         start = dip->logical_offset;
5775         do {
5776                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5777                         struct page *page = bvec->bv_page;
5778                         char *kaddr;
5779                         u32 csum = ~(u32)0;
5780                         unsigned long flags;
5781
5782                         local_irq_save(flags);
5783                         kaddr = kmap_atomic(page, KM_IRQ0);
5784                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5785                                                csum, bvec->bv_len);
5786                         btrfs_csum_final(csum, (char *)&csum);
5787                         kunmap_atomic(kaddr, KM_IRQ0);
5788                         local_irq_restore(flags);
5789
5790                         flush_dcache_page(bvec->bv_page);
5791                         if (csum != *private) {
5792                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5793                                       " %llu csum %u private %u\n",
5794                                       inode->i_ino, (unsigned long long)start,
5795                                       csum, *private);
5796                                 err = -EIO;
5797                         }
5798                 }
5799
5800                 start += bvec->bv_len;
5801                 private++;
5802                 bvec++;
5803         } while (bvec <= bvec_end);
5804
5805         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5806                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5807         bio->bi_private = dip->private;
5808
5809         kfree(dip->csums);
5810         kfree(dip);
5811
5812         /* If we had a csum failure make sure to clear the uptodate flag */
5813         if (err)
5814                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5815         dio_end_io(bio, err);
5816 }
5817
5818 static void btrfs_endio_direct_write(struct bio *bio, int err)
5819 {
5820         struct btrfs_dio_private *dip = bio->bi_private;
5821         struct inode *inode = dip->inode;
5822         struct btrfs_root *root = BTRFS_I(inode)->root;
5823         struct btrfs_trans_handle *trans;
5824         struct btrfs_ordered_extent *ordered = NULL;
5825         struct extent_state *cached_state = NULL;
5826         u64 ordered_offset = dip->logical_offset;
5827         u64 ordered_bytes = dip->bytes;
5828         int ret;
5829
5830         if (err)
5831                 goto out_done;
5832 again:
5833         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5834                                                    &ordered_offset,
5835                                                    ordered_bytes);
5836         if (!ret)
5837                 goto out_test;
5838
5839         BUG_ON(!ordered);
5840
5841         trans = btrfs_join_transaction(root, 1);
5842         if (IS_ERR(trans)) {
5843                 err = -ENOMEM;
5844                 goto out;
5845         }
5846         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5847
5848         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5849                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5850                 if (!ret)
5851                         ret = btrfs_update_inode(trans, root, inode);
5852                 err = ret;
5853                 goto out;
5854         }
5855
5856         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5857                          ordered->file_offset + ordered->len - 1, 0,
5858                          &cached_state, GFP_NOFS);
5859
5860         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5861                 ret = btrfs_mark_extent_written(trans, inode,
5862                                                 ordered->file_offset,
5863                                                 ordered->file_offset +
5864                                                 ordered->len);
5865                 if (ret) {
5866                         err = ret;
5867                         goto out_unlock;
5868                 }
5869         } else {
5870                 ret = insert_reserved_file_extent(trans, inode,
5871                                                   ordered->file_offset,
5872                                                   ordered->start,
5873                                                   ordered->disk_len,
5874                                                   ordered->len,
5875                                                   ordered->len,
5876                                                   0, 0, 0,
5877                                                   BTRFS_FILE_EXTENT_REG);
5878                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5879                                    ordered->file_offset, ordered->len);
5880                 if (ret) {
5881                         err = ret;
5882                         WARN_ON(1);
5883                         goto out_unlock;
5884                 }
5885         }
5886
5887         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5888         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5889         if (!ret)
5890                 btrfs_update_inode(trans, root, inode);
5891         ret = 0;
5892 out_unlock:
5893         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5894                              ordered->file_offset + ordered->len - 1,
5895                              &cached_state, GFP_NOFS);
5896 out:
5897         btrfs_delalloc_release_metadata(inode, ordered->len);
5898         btrfs_end_transaction(trans, root);
5899         ordered_offset = ordered->file_offset + ordered->len;
5900         btrfs_put_ordered_extent(ordered);
5901         btrfs_put_ordered_extent(ordered);
5902
5903 out_test:
5904         /*
5905          * our bio might span multiple ordered extents.  If we haven't
5906          * completed the accounting for the whole dio, go back and try again
5907          */
5908         if (ordered_offset < dip->logical_offset + dip->bytes) {
5909                 ordered_bytes = dip->logical_offset + dip->bytes -
5910                         ordered_offset;
5911                 goto again;
5912         }
5913 out_done:
5914         bio->bi_private = dip->private;
5915
5916         kfree(dip->csums);
5917         kfree(dip);
5918
5919         /* If we had an error make sure to clear the uptodate flag */
5920         if (err)
5921                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5922         dio_end_io(bio, err);
5923 }
5924
5925 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5926                                     struct bio *bio, int mirror_num,
5927                                     unsigned long bio_flags, u64 offset)
5928 {
5929         int ret;
5930         struct btrfs_root *root = BTRFS_I(inode)->root;
5931         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5932         BUG_ON(ret);
5933         return 0;
5934 }
5935
5936 static void btrfs_end_dio_bio(struct bio *bio, int err)
5937 {
5938         struct btrfs_dio_private *dip = bio->bi_private;
5939
5940         if (err) {
5941                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5942                       "sector %#Lx len %u err no %d\n",
5943                       dip->inode->i_ino, bio->bi_rw,
5944                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5945                 dip->errors = 1;
5946
5947                 /*
5948                  * before atomic variable goto zero, we must make sure
5949                  * dip->errors is perceived to be set.
5950                  */
5951                 smp_mb__before_atomic_dec();
5952         }
5953
5954         /* if there are more bios still pending for this dio, just exit */
5955         if (!atomic_dec_and_test(&dip->pending_bios))
5956                 goto out;
5957
5958         if (dip->errors)
5959                 bio_io_error(dip->orig_bio);
5960         else {
5961                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5962                 bio_endio(dip->orig_bio, 0);
5963         }
5964 out:
5965         bio_put(bio);
5966 }
5967
5968 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5969                                        u64 first_sector, gfp_t gfp_flags)
5970 {
5971         int nr_vecs = bio_get_nr_vecs(bdev);
5972         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5973 }
5974
5975 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5976                                          int rw, u64 file_offset, int skip_sum,
5977                                          u32 *csums, int async_submit)
5978 {
5979         int write = rw & REQ_WRITE;
5980         struct btrfs_root *root = BTRFS_I(inode)->root;
5981         int ret;
5982
5983         bio_get(bio);
5984         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5985         if (ret)
5986                 goto err;
5987
5988         if (skip_sum)
5989                 goto map;
5990
5991         if (write && async_submit) {
5992                 ret = btrfs_wq_submit_bio(root->fs_info,
5993                                    inode, rw, bio, 0, 0,
5994                                    file_offset,
5995                                    __btrfs_submit_bio_start_direct_io,
5996                                    __btrfs_submit_bio_done);
5997                 goto err;
5998         } else if (write) {
5999                 /*
6000                  * If we aren't doing async submit, calculate the csum of the
6001                  * bio now.
6002                  */
6003                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6004                 if (ret)
6005                         goto err;
6006         } else if (!skip_sum) {
6007                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6008                                           file_offset, csums);
6009                 if (ret)
6010                         goto err;
6011         }
6012
6013 map:
6014         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6015 err:
6016         bio_put(bio);
6017         return ret;
6018 }
6019
6020 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6021                                     int skip_sum)
6022 {
6023         struct inode *inode = dip->inode;
6024         struct btrfs_root *root = BTRFS_I(inode)->root;
6025         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6026         struct bio *bio;
6027         struct bio *orig_bio = dip->orig_bio;
6028         struct bio_vec *bvec = orig_bio->bi_io_vec;
6029         u64 start_sector = orig_bio->bi_sector;
6030         u64 file_offset = dip->logical_offset;
6031         u64 submit_len = 0;
6032         u64 map_length;
6033         int nr_pages = 0;
6034         u32 *csums = dip->csums;
6035         int ret = 0;
6036         int async_submit = 0;
6037         int write = rw & REQ_WRITE;
6038
6039         map_length = orig_bio->bi_size;
6040         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6041                               &map_length, NULL, 0);
6042         if (ret) {
6043                 bio_put(orig_bio);
6044                 return -EIO;
6045         }
6046
6047         if (map_length >= orig_bio->bi_size) {
6048                 bio = orig_bio;
6049                 goto submit;
6050         }
6051
6052         async_submit = 1;
6053         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6054         if (!bio)
6055                 return -ENOMEM;
6056         bio->bi_private = dip;
6057         bio->bi_end_io = btrfs_end_dio_bio;
6058         atomic_inc(&dip->pending_bios);
6059
6060         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6061                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6062                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6063                                  bvec->bv_offset) < bvec->bv_len)) {
6064                         /*
6065                          * inc the count before we submit the bio so
6066                          * we know the end IO handler won't happen before
6067                          * we inc the count. Otherwise, the dip might get freed
6068                          * before we're done setting it up
6069                          */
6070                         atomic_inc(&dip->pending_bios);
6071                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6072                                                      file_offset, skip_sum,
6073                                                      csums, async_submit);
6074                         if (ret) {
6075                                 bio_put(bio);
6076                                 atomic_dec(&dip->pending_bios);
6077                                 goto out_err;
6078                         }
6079
6080                         /* Write's use the ordered csums */
6081                         if (!write && !skip_sum)
6082                                 csums = csums + nr_pages;
6083                         start_sector += submit_len >> 9;
6084                         file_offset += submit_len;
6085
6086                         submit_len = 0;
6087                         nr_pages = 0;
6088
6089                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6090                                                   start_sector, GFP_NOFS);
6091                         if (!bio)
6092                                 goto out_err;
6093                         bio->bi_private = dip;
6094                         bio->bi_end_io = btrfs_end_dio_bio;
6095
6096                         map_length = orig_bio->bi_size;
6097                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6098                                               &map_length, NULL, 0);
6099                         if (ret) {
6100                                 bio_put(bio);
6101                                 goto out_err;
6102                         }
6103                 } else {
6104                         submit_len += bvec->bv_len;
6105                         nr_pages ++;
6106                         bvec++;
6107                 }
6108         }
6109
6110 submit:
6111         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6112                                      csums, async_submit);
6113         if (!ret)
6114                 return 0;
6115
6116         bio_put(bio);
6117 out_err:
6118         dip->errors = 1;
6119         /*
6120          * before atomic variable goto zero, we must
6121          * make sure dip->errors is perceived to be set.
6122          */
6123         smp_mb__before_atomic_dec();
6124         if (atomic_dec_and_test(&dip->pending_bios))
6125                 bio_io_error(dip->orig_bio);
6126
6127         /* bio_end_io() will handle error, so we needn't return it */
6128         return 0;
6129 }
6130
6131 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6132                                 loff_t file_offset)
6133 {
6134         struct btrfs_root *root = BTRFS_I(inode)->root;
6135         struct btrfs_dio_private *dip;
6136         struct bio_vec *bvec = bio->bi_io_vec;
6137         int skip_sum;
6138         int write = rw & REQ_WRITE;
6139         int ret = 0;
6140
6141         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6142
6143         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6144         if (!dip) {
6145                 ret = -ENOMEM;
6146                 goto free_ordered;
6147         }
6148         dip->csums = NULL;
6149
6150         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6151         if (!write && !skip_sum) {
6152                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6153                 if (!dip->csums) {
6154                         kfree(dip);
6155                         ret = -ENOMEM;
6156                         goto free_ordered;
6157                 }
6158         }
6159
6160         dip->private = bio->bi_private;
6161         dip->inode = inode;
6162         dip->logical_offset = file_offset;
6163
6164         dip->bytes = 0;
6165         do {
6166                 dip->bytes += bvec->bv_len;
6167                 bvec++;
6168         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6169
6170         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6171         bio->bi_private = dip;
6172         dip->errors = 0;
6173         dip->orig_bio = bio;
6174         atomic_set(&dip->pending_bios, 0);
6175
6176         if (write)
6177                 bio->bi_end_io = btrfs_endio_direct_write;
6178         else
6179                 bio->bi_end_io = btrfs_endio_direct_read;
6180
6181         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6182         if (!ret)
6183                 return;
6184 free_ordered:
6185         /*
6186          * If this is a write, we need to clean up the reserved space and kill
6187          * the ordered extent.
6188          */
6189         if (write) {
6190                 struct btrfs_ordered_extent *ordered;
6191                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6192                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6193                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6194                         btrfs_free_reserved_extent(root, ordered->start,
6195                                                    ordered->disk_len);
6196                 btrfs_put_ordered_extent(ordered);
6197                 btrfs_put_ordered_extent(ordered);
6198         }
6199         bio_endio(bio, ret);
6200 }
6201
6202 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6203                         const struct iovec *iov, loff_t offset,
6204                         unsigned long nr_segs)
6205 {
6206         int seg;
6207         int i;
6208         size_t size;
6209         unsigned long addr;
6210         unsigned blocksize_mask = root->sectorsize - 1;
6211         ssize_t retval = -EINVAL;
6212         loff_t end = offset;
6213
6214         if (offset & blocksize_mask)
6215                 goto out;
6216
6217         /* Check the memory alignment.  Blocks cannot straddle pages */
6218         for (seg = 0; seg < nr_segs; seg++) {
6219                 addr = (unsigned long)iov[seg].iov_base;
6220                 size = iov[seg].iov_len;
6221                 end += size;
6222                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6223                         goto out;
6224
6225                 /* If this is a write we don't need to check anymore */
6226                 if (rw & WRITE)
6227                         continue;
6228
6229                 /*
6230                  * Check to make sure we don't have duplicate iov_base's in this
6231                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6232                  * when reading back.
6233                  */
6234                 for (i = seg + 1; i < nr_segs; i++) {
6235                         if (iov[seg].iov_base == iov[i].iov_base)
6236                                 goto out;
6237                 }
6238         }
6239         retval = 0;
6240 out:
6241         return retval;
6242 }
6243 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6244                         const struct iovec *iov, loff_t offset,
6245                         unsigned long nr_segs)
6246 {
6247         struct file *file = iocb->ki_filp;
6248         struct inode *inode = file->f_mapping->host;
6249         struct btrfs_ordered_extent *ordered;
6250         struct extent_state *cached_state = NULL;
6251         u64 lockstart, lockend;
6252         ssize_t ret;
6253         int writing = rw & WRITE;
6254         int write_bits = 0;
6255         size_t count = iov_length(iov, nr_segs);
6256
6257         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6258                             offset, nr_segs)) {
6259                 return 0;
6260         }
6261
6262         lockstart = offset;
6263         lockend = offset + count - 1;
6264
6265         if (writing) {
6266                 ret = btrfs_delalloc_reserve_space(inode, count);
6267                 if (ret)
6268                         goto out;
6269         }
6270
6271         while (1) {
6272                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6273                                  0, &cached_state, GFP_NOFS);
6274                 /*
6275                  * We're concerned with the entire range that we're going to be
6276                  * doing DIO to, so we need to make sure theres no ordered
6277                  * extents in this range.
6278                  */
6279                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6280                                                      lockend - lockstart + 1);
6281                 if (!ordered)
6282                         break;
6283                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6284                                      &cached_state, GFP_NOFS);
6285                 btrfs_start_ordered_extent(inode, ordered, 1);
6286                 btrfs_put_ordered_extent(ordered);
6287                 cond_resched();
6288         }
6289
6290         /*
6291          * we don't use btrfs_set_extent_delalloc because we don't want
6292          * the dirty or uptodate bits
6293          */
6294         if (writing) {
6295                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6296                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6297                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6298                                      GFP_NOFS);
6299                 if (ret) {
6300                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6301                                          lockend, EXTENT_LOCKED | write_bits,
6302                                          1, 0, &cached_state, GFP_NOFS);
6303                         goto out;
6304                 }
6305         }
6306
6307         free_extent_state(cached_state);
6308         cached_state = NULL;
6309
6310         ret = __blockdev_direct_IO(rw, iocb, inode,
6311                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6312                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6313                    btrfs_submit_direct, 0);
6314
6315         if (ret < 0 && ret != -EIOCBQUEUED) {
6316                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6317                               offset + iov_length(iov, nr_segs) - 1,
6318                               EXTENT_LOCKED | write_bits, 1, 0,
6319                               &cached_state, GFP_NOFS);
6320         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6321                 /*
6322                  * We're falling back to buffered, unlock the section we didn't
6323                  * do IO on.
6324                  */
6325                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6326                               offset + iov_length(iov, nr_segs) - 1,
6327                               EXTENT_LOCKED | write_bits, 1, 0,
6328                               &cached_state, GFP_NOFS);
6329         }
6330 out:
6331         free_extent_state(cached_state);
6332         return ret;
6333 }
6334
6335 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6336                 __u64 start, __u64 len)
6337 {
6338         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6339 }
6340
6341 int btrfs_readpage(struct file *file, struct page *page)
6342 {
6343         struct extent_io_tree *tree;
6344         tree = &BTRFS_I(page->mapping->host)->io_tree;
6345         return extent_read_full_page(tree, page, btrfs_get_extent);
6346 }
6347
6348 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6349 {
6350         struct extent_io_tree *tree;
6351
6352
6353         if (current->flags & PF_MEMALLOC) {
6354                 redirty_page_for_writepage(wbc, page);
6355                 unlock_page(page);
6356                 return 0;
6357         }
6358         tree = &BTRFS_I(page->mapping->host)->io_tree;
6359         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6360 }
6361
6362 int btrfs_writepages(struct address_space *mapping,
6363                      struct writeback_control *wbc)
6364 {
6365         struct extent_io_tree *tree;
6366
6367         tree = &BTRFS_I(mapping->host)->io_tree;
6368         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6369 }
6370
6371 static int
6372 btrfs_readpages(struct file *file, struct address_space *mapping,
6373                 struct list_head *pages, unsigned nr_pages)
6374 {
6375         struct extent_io_tree *tree;
6376         tree = &BTRFS_I(mapping->host)->io_tree;
6377         return extent_readpages(tree, mapping, pages, nr_pages,
6378                                 btrfs_get_extent);
6379 }
6380 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6381 {
6382         struct extent_io_tree *tree;
6383         struct extent_map_tree *map;
6384         int ret;
6385
6386         tree = &BTRFS_I(page->mapping->host)->io_tree;
6387         map = &BTRFS_I(page->mapping->host)->extent_tree;
6388         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6389         if (ret == 1) {
6390                 ClearPagePrivate(page);
6391                 set_page_private(page, 0);
6392                 page_cache_release(page);
6393         }
6394         return ret;
6395 }
6396
6397 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6398 {
6399         if (PageWriteback(page) || PageDirty(page))
6400                 return 0;
6401         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6402 }
6403
6404 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6405 {
6406         struct extent_io_tree *tree;
6407         struct btrfs_ordered_extent *ordered;
6408         struct extent_state *cached_state = NULL;
6409         u64 page_start = page_offset(page);
6410         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6411
6412
6413         /*
6414          * we have the page locked, so new writeback can't start,
6415          * and the dirty bit won't be cleared while we are here.
6416          *
6417          * Wait for IO on this page so that we can safely clear
6418          * the PagePrivate2 bit and do ordered accounting
6419          */
6420         wait_on_page_writeback(page);
6421
6422         tree = &BTRFS_I(page->mapping->host)->io_tree;
6423         if (offset) {
6424                 btrfs_releasepage(page, GFP_NOFS);
6425                 return;
6426         }
6427         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6428                          GFP_NOFS);
6429         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6430                                            page_offset(page));
6431         if (ordered) {
6432                 /*
6433                  * IO on this page will never be started, so we need
6434                  * to account for any ordered extents now
6435                  */
6436                 clear_extent_bit(tree, page_start, page_end,
6437                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6438                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6439                                  &cached_state, GFP_NOFS);
6440                 /*
6441                  * whoever cleared the private bit is responsible
6442                  * for the finish_ordered_io
6443                  */
6444                 if (TestClearPagePrivate2(page)) {
6445                         btrfs_finish_ordered_io(page->mapping->host,
6446                                                 page_start, page_end);
6447                 }
6448                 btrfs_put_ordered_extent(ordered);
6449                 cached_state = NULL;
6450                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6451                                  GFP_NOFS);
6452         }
6453         clear_extent_bit(tree, page_start, page_end,
6454                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6455                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6456         __btrfs_releasepage(page, GFP_NOFS);
6457
6458         ClearPageChecked(page);
6459         if (PagePrivate(page)) {
6460                 ClearPagePrivate(page);
6461                 set_page_private(page, 0);
6462                 page_cache_release(page);
6463         }
6464 }
6465
6466 /*
6467  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6468  * called from a page fault handler when a page is first dirtied. Hence we must
6469  * be careful to check for EOF conditions here. We set the page up correctly
6470  * for a written page which means we get ENOSPC checking when writing into
6471  * holes and correct delalloc and unwritten extent mapping on filesystems that
6472  * support these features.
6473  *
6474  * We are not allowed to take the i_mutex here so we have to play games to
6475  * protect against truncate races as the page could now be beyond EOF.  Because
6476  * vmtruncate() writes the inode size before removing pages, once we have the
6477  * page lock we can determine safely if the page is beyond EOF. If it is not
6478  * beyond EOF, then the page is guaranteed safe against truncation until we
6479  * unlock the page.
6480  */
6481 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6482 {
6483         struct page *page = vmf->page;
6484         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6485         struct btrfs_root *root = BTRFS_I(inode)->root;
6486         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6487         struct btrfs_ordered_extent *ordered;
6488         struct extent_state *cached_state = NULL;
6489         char *kaddr;
6490         unsigned long zero_start;
6491         loff_t size;
6492         int ret;
6493         u64 page_start;
6494         u64 page_end;
6495
6496         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6497         if (ret) {
6498                 if (ret == -ENOMEM)
6499                         ret = VM_FAULT_OOM;
6500                 else /* -ENOSPC, -EIO, etc */
6501                         ret = VM_FAULT_SIGBUS;
6502                 goto out;
6503         }
6504
6505         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6506 again:
6507         lock_page(page);
6508         size = i_size_read(inode);
6509         page_start = page_offset(page);
6510         page_end = page_start + PAGE_CACHE_SIZE - 1;
6511
6512         if ((page->mapping != inode->i_mapping) ||
6513             (page_start >= size)) {
6514                 /* page got truncated out from underneath us */
6515                 goto out_unlock;
6516         }
6517         wait_on_page_writeback(page);
6518
6519         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6520                          GFP_NOFS);
6521         set_page_extent_mapped(page);
6522
6523         /*
6524          * we can't set the delalloc bits if there are pending ordered
6525          * extents.  Drop our locks and wait for them to finish
6526          */
6527         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6528         if (ordered) {
6529                 unlock_extent_cached(io_tree, page_start, page_end,
6530                                      &cached_state, GFP_NOFS);
6531                 unlock_page(page);
6532                 btrfs_start_ordered_extent(inode, ordered, 1);
6533                 btrfs_put_ordered_extent(ordered);
6534                 goto again;
6535         }
6536
6537         /*
6538          * XXX - page_mkwrite gets called every time the page is dirtied, even
6539          * if it was already dirty, so for space accounting reasons we need to
6540          * clear any delalloc bits for the range we are fixing to save.  There
6541          * is probably a better way to do this, but for now keep consistent with
6542          * prepare_pages in the normal write path.
6543          */
6544         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6545                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6546                           0, 0, &cached_state, GFP_NOFS);
6547
6548         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6549                                         &cached_state);
6550         if (ret) {
6551                 unlock_extent_cached(io_tree, page_start, page_end,
6552                                      &cached_state, GFP_NOFS);
6553                 ret = VM_FAULT_SIGBUS;
6554                 goto out_unlock;
6555         }
6556         ret = 0;
6557
6558         /* page is wholly or partially inside EOF */
6559         if (page_start + PAGE_CACHE_SIZE > size)
6560                 zero_start = size & ~PAGE_CACHE_MASK;
6561         else
6562                 zero_start = PAGE_CACHE_SIZE;
6563
6564         if (zero_start != PAGE_CACHE_SIZE) {
6565                 kaddr = kmap(page);
6566                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6567                 flush_dcache_page(page);
6568                 kunmap(page);
6569         }
6570         ClearPageChecked(page);
6571         set_page_dirty(page);
6572         SetPageUptodate(page);
6573
6574         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6575         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6576
6577         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6578
6579 out_unlock:
6580         if (!ret)
6581                 return VM_FAULT_LOCKED;
6582         unlock_page(page);
6583         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6584 out:
6585         return ret;
6586 }
6587
6588 static int btrfs_truncate(struct inode *inode)
6589 {
6590         struct btrfs_root *root = BTRFS_I(inode)->root;
6591         int ret;
6592         int err = 0;
6593         struct btrfs_trans_handle *trans;
6594         unsigned long nr;
6595         u64 mask = root->sectorsize - 1;
6596
6597         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6598         if (ret)
6599                 return ret;
6600
6601         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6602         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6603
6604         trans = btrfs_start_transaction(root, 5);
6605         if (IS_ERR(trans))
6606                 return PTR_ERR(trans);
6607
6608         btrfs_set_trans_block_group(trans, inode);
6609
6610         ret = btrfs_orphan_add(trans, inode);
6611         if (ret) {
6612                 btrfs_end_transaction(trans, root);
6613                 return ret;
6614         }
6615
6616         nr = trans->blocks_used;
6617         btrfs_end_transaction(trans, root);
6618         btrfs_btree_balance_dirty(root, nr);
6619
6620         /* Now start a transaction for the truncate */
6621         trans = btrfs_start_transaction(root, 0);
6622         if (IS_ERR(trans))
6623                 return PTR_ERR(trans);
6624         btrfs_set_trans_block_group(trans, inode);
6625         trans->block_rsv = root->orphan_block_rsv;
6626
6627         /*
6628          * setattr is responsible for setting the ordered_data_close flag,
6629          * but that is only tested during the last file release.  That
6630          * could happen well after the next commit, leaving a great big
6631          * window where new writes may get lost if someone chooses to write
6632          * to this file after truncating to zero
6633          *
6634          * The inode doesn't have any dirty data here, and so if we commit
6635          * this is a noop.  If someone immediately starts writing to the inode
6636          * it is very likely we'll catch some of their writes in this
6637          * transaction, and the commit will find this file on the ordered
6638          * data list with good things to send down.
6639          *
6640          * This is a best effort solution, there is still a window where
6641          * using truncate to replace the contents of the file will
6642          * end up with a zero length file after a crash.
6643          */
6644         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6645                 btrfs_add_ordered_operation(trans, root, inode);
6646
6647         while (1) {
6648                 if (!trans) {
6649                         trans = btrfs_start_transaction(root, 0);
6650                         if (IS_ERR(trans))
6651                                 return PTR_ERR(trans);
6652                         btrfs_set_trans_block_group(trans, inode);
6653                         trans->block_rsv = root->orphan_block_rsv;
6654                 }
6655
6656                 ret = btrfs_block_rsv_check(trans, root,
6657                                             root->orphan_block_rsv, 0, 5);
6658                 if (ret == -EAGAIN) {
6659                         ret = btrfs_commit_transaction(trans, root);
6660                         if (ret)
6661                                 return ret;
6662                         trans = NULL;
6663                         continue;
6664                 } else if (ret) {
6665                         err = ret;
6666                         break;
6667                 }
6668
6669                 ret = btrfs_truncate_inode_items(trans, root, inode,
6670                                                  inode->i_size,
6671                                                  BTRFS_EXTENT_DATA_KEY);
6672                 if (ret != -EAGAIN) {
6673                         err = ret;
6674                         break;
6675                 }
6676
6677                 ret = btrfs_update_inode(trans, root, inode);
6678                 if (ret) {
6679                         err = ret;
6680                         break;
6681                 }
6682
6683                 nr = trans->blocks_used;
6684                 btrfs_end_transaction(trans, root);
6685                 trans = NULL;
6686                 btrfs_btree_balance_dirty(root, nr);
6687         }
6688
6689         if (ret == 0 && inode->i_nlink > 0) {
6690                 ret = btrfs_orphan_del(trans, inode);
6691                 if (ret)
6692                         err = ret;
6693         } else if (ret && inode->i_nlink > 0) {
6694                 /*
6695                  * Failed to do the truncate, remove us from the in memory
6696                  * orphan list.
6697                  */
6698                 ret = btrfs_orphan_del(NULL, inode);
6699         }
6700
6701         ret = btrfs_update_inode(trans, root, inode);
6702         if (ret && !err)
6703                 err = ret;
6704
6705         nr = trans->blocks_used;
6706         ret = btrfs_end_transaction_throttle(trans, root);
6707         if (ret && !err)
6708                 err = ret;
6709         btrfs_btree_balance_dirty(root, nr);
6710
6711         return err;
6712 }
6713
6714 /*
6715  * create a new subvolume directory/inode (helper for the ioctl).
6716  */
6717 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6718                              struct btrfs_root *new_root,
6719                              u64 new_dirid, u64 alloc_hint)
6720 {
6721         struct inode *inode;
6722         int err;
6723         u64 index = 0;
6724
6725         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6726                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6727         if (IS_ERR(inode))
6728                 return PTR_ERR(inode);
6729         inode->i_op = &btrfs_dir_inode_operations;
6730         inode->i_fop = &btrfs_dir_file_operations;
6731
6732         inode->i_nlink = 1;
6733         btrfs_i_size_write(inode, 0);
6734
6735         err = btrfs_update_inode(trans, new_root, inode);
6736         BUG_ON(err);
6737
6738         iput(inode);
6739         return 0;
6740 }
6741
6742 /* helper function for file defrag and space balancing.  This
6743  * forces readahead on a given range of bytes in an inode
6744  */
6745 unsigned long btrfs_force_ra(struct address_space *mapping,
6746                               struct file_ra_state *ra, struct file *file,
6747                               pgoff_t offset, pgoff_t last_index)
6748 {
6749         pgoff_t req_size = last_index - offset + 1;
6750
6751         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6752         return offset + req_size;
6753 }
6754
6755 struct inode *btrfs_alloc_inode(struct super_block *sb)
6756 {
6757         struct btrfs_inode *ei;
6758         struct inode *inode;
6759
6760         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6761         if (!ei)
6762                 return NULL;
6763
6764         ei->root = NULL;
6765         ei->space_info = NULL;
6766         ei->generation = 0;
6767         ei->sequence = 0;
6768         ei->last_trans = 0;
6769         ei->last_sub_trans = 0;
6770         ei->logged_trans = 0;
6771         ei->delalloc_bytes = 0;
6772         ei->reserved_bytes = 0;
6773         ei->disk_i_size = 0;
6774         ei->flags = 0;
6775         ei->index_cnt = (u64)-1;
6776         ei->last_unlink_trans = 0;
6777
6778         atomic_set(&ei->outstanding_extents, 0);
6779         atomic_set(&ei->reserved_extents, 0);
6780
6781         ei->ordered_data_close = 0;
6782         ei->orphan_meta_reserved = 0;
6783         ei->dummy_inode = 0;
6784         ei->force_compress = BTRFS_COMPRESS_NONE;
6785
6786         inode = &ei->vfs_inode;
6787         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6788         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6789         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6790         mutex_init(&ei->log_mutex);
6791         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6792         INIT_LIST_HEAD(&ei->i_orphan);
6793         INIT_LIST_HEAD(&ei->delalloc_inodes);
6794         INIT_LIST_HEAD(&ei->ordered_operations);
6795         RB_CLEAR_NODE(&ei->rb_node);
6796
6797         return inode;
6798 }
6799
6800 static void btrfs_i_callback(struct rcu_head *head)
6801 {
6802         struct inode *inode = container_of(head, struct inode, i_rcu);
6803         INIT_LIST_HEAD(&inode->i_dentry);
6804         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6805 }
6806
6807 void btrfs_destroy_inode(struct inode *inode)
6808 {
6809         struct btrfs_ordered_extent *ordered;
6810         struct btrfs_root *root = BTRFS_I(inode)->root;
6811
6812         WARN_ON(!list_empty(&inode->i_dentry));
6813         WARN_ON(inode->i_data.nrpages);
6814         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6815         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6816
6817         /*
6818          * This can happen where we create an inode, but somebody else also
6819          * created the same inode and we need to destroy the one we already
6820          * created.
6821          */
6822         if (!root)
6823                 goto free;
6824
6825         /*
6826          * Make sure we're properly removed from the ordered operation
6827          * lists.
6828          */
6829         smp_mb();
6830         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6831                 spin_lock(&root->fs_info->ordered_extent_lock);
6832                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6833                 spin_unlock(&root->fs_info->ordered_extent_lock);
6834         }
6835
6836         if (root == root->fs_info->tree_root) {
6837                 struct btrfs_block_group_cache *block_group;
6838
6839                 block_group = btrfs_lookup_block_group(root->fs_info,
6840                                                 BTRFS_I(inode)->block_group);
6841                 if (block_group && block_group->inode == inode) {
6842                         spin_lock(&block_group->lock);
6843                         block_group->inode = NULL;
6844                         spin_unlock(&block_group->lock);
6845                         btrfs_put_block_group(block_group);
6846                 } else if (block_group) {
6847                         btrfs_put_block_group(block_group);
6848                 }
6849         }
6850
6851         spin_lock(&root->orphan_lock);
6852         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6853                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6854                        inode->i_ino);
6855                 list_del_init(&BTRFS_I(inode)->i_orphan);
6856         }
6857         spin_unlock(&root->orphan_lock);
6858
6859         while (1) {
6860                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6861                 if (!ordered)
6862                         break;
6863                 else {
6864                         printk(KERN_ERR "btrfs found ordered "
6865                                "extent %llu %llu on inode cleanup\n",
6866                                (unsigned long long)ordered->file_offset,
6867                                (unsigned long long)ordered->len);
6868                         btrfs_remove_ordered_extent(inode, ordered);
6869                         btrfs_put_ordered_extent(ordered);
6870                         btrfs_put_ordered_extent(ordered);
6871                 }
6872         }
6873         inode_tree_del(inode);
6874         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6875 free:
6876         call_rcu(&inode->i_rcu, btrfs_i_callback);
6877 }
6878
6879 int btrfs_drop_inode(struct inode *inode)
6880 {
6881         struct btrfs_root *root = BTRFS_I(inode)->root;
6882
6883         if (btrfs_root_refs(&root->root_item) == 0 &&
6884             root != root->fs_info->tree_root)
6885                 return 1;
6886         else
6887                 return generic_drop_inode(inode);
6888 }
6889
6890 static void init_once(void *foo)
6891 {
6892         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6893
6894         inode_init_once(&ei->vfs_inode);
6895 }
6896
6897 void btrfs_destroy_cachep(void)
6898 {
6899         if (btrfs_inode_cachep)
6900                 kmem_cache_destroy(btrfs_inode_cachep);
6901         if (btrfs_trans_handle_cachep)
6902                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6903         if (btrfs_transaction_cachep)
6904                 kmem_cache_destroy(btrfs_transaction_cachep);
6905         if (btrfs_path_cachep)
6906                 kmem_cache_destroy(btrfs_path_cachep);
6907         if (btrfs_free_space_cachep)
6908                 kmem_cache_destroy(btrfs_free_space_cachep);
6909 }
6910
6911 int btrfs_init_cachep(void)
6912 {
6913         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6914                         sizeof(struct btrfs_inode), 0,
6915                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6916         if (!btrfs_inode_cachep)
6917                 goto fail;
6918
6919         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6920                         sizeof(struct btrfs_trans_handle), 0,
6921                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6922         if (!btrfs_trans_handle_cachep)
6923                 goto fail;
6924
6925         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6926                         sizeof(struct btrfs_transaction), 0,
6927                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6928         if (!btrfs_transaction_cachep)
6929                 goto fail;
6930
6931         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6932                         sizeof(struct btrfs_path), 0,
6933                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6934         if (!btrfs_path_cachep)
6935                 goto fail;
6936
6937         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6938                         sizeof(struct btrfs_free_space), 0,
6939                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6940         if (!btrfs_free_space_cachep)
6941                 goto fail;
6942
6943         return 0;
6944 fail:
6945         btrfs_destroy_cachep();
6946         return -ENOMEM;
6947 }
6948
6949 static int btrfs_getattr(struct vfsmount *mnt,
6950                          struct dentry *dentry, struct kstat *stat)
6951 {
6952         struct inode *inode = dentry->d_inode;
6953         generic_fillattr(inode, stat);
6954         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6955         stat->blksize = PAGE_CACHE_SIZE;
6956         stat->blocks = (inode_get_bytes(inode) +
6957                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6958         return 0;
6959 }
6960
6961 /*
6962  * If a file is moved, it will inherit the cow and compression flags of the new
6963  * directory.
6964  */
6965 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6966 {
6967         struct btrfs_inode *b_dir = BTRFS_I(dir);
6968         struct btrfs_inode *b_inode = BTRFS_I(inode);
6969
6970         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6971                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6972         else
6973                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6974
6975         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6976                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6977         else
6978                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6979 }
6980
6981 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6982                            struct inode *new_dir, struct dentry *new_dentry)
6983 {
6984         struct btrfs_trans_handle *trans;
6985         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6986         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6987         struct inode *new_inode = new_dentry->d_inode;
6988         struct inode *old_inode = old_dentry->d_inode;
6989         struct timespec ctime = CURRENT_TIME;
6990         u64 index = 0;
6991         u64 root_objectid;
6992         int ret;
6993
6994         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6995                 return -EPERM;
6996
6997         /* we only allow rename subvolume link between subvolumes */
6998         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6999                 return -EXDEV;
7000
7001         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7002             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7003                 return -ENOTEMPTY;
7004
7005         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7006             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7007                 return -ENOTEMPTY;
7008         /*
7009          * we're using rename to replace one file with another.
7010          * and the replacement file is large.  Start IO on it now so
7011          * we don't add too much work to the end of the transaction
7012          */
7013         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7014             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7015                 filemap_flush(old_inode->i_mapping);
7016
7017         /* close the racy window with snapshot create/destroy ioctl */
7018         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7019                 down_read(&root->fs_info->subvol_sem);
7020         /*
7021          * We want to reserve the absolute worst case amount of items.  So if
7022          * both inodes are subvols and we need to unlink them then that would
7023          * require 4 item modifications, but if they are both normal inodes it
7024          * would require 5 item modifications, so we'll assume their normal
7025          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7026          * should cover the worst case number of items we'll modify.
7027          */
7028         trans = btrfs_start_transaction(root, 20);
7029         if (IS_ERR(trans)) {
7030                 ret = PTR_ERR(trans);
7031                 goto out_notrans;
7032         }
7033
7034         btrfs_set_trans_block_group(trans, new_dir);
7035
7036         if (dest != root)
7037                 btrfs_record_root_in_trans(trans, dest);
7038
7039         ret = btrfs_set_inode_index(new_dir, &index);
7040         if (ret)
7041                 goto out_fail;
7042
7043         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7044                 /* force full log commit if subvolume involved. */
7045                 root->fs_info->last_trans_log_full_commit = trans->transid;
7046         } else {
7047                 ret = btrfs_insert_inode_ref(trans, dest,
7048                                              new_dentry->d_name.name,
7049                                              new_dentry->d_name.len,
7050                                              old_inode->i_ino,
7051                                              new_dir->i_ino, index);
7052                 if (ret)
7053                         goto out_fail;
7054                 /*
7055                  * this is an ugly little race, but the rename is required
7056                  * to make sure that if we crash, the inode is either at the
7057                  * old name or the new one.  pinning the log transaction lets
7058                  * us make sure we don't allow a log commit to come in after
7059                  * we unlink the name but before we add the new name back in.
7060                  */
7061                 btrfs_pin_log_trans(root);
7062         }
7063         /*
7064          * make sure the inode gets flushed if it is replacing
7065          * something.
7066          */
7067         if (new_inode && new_inode->i_size &&
7068             old_inode && S_ISREG(old_inode->i_mode)) {
7069                 btrfs_add_ordered_operation(trans, root, old_inode);
7070         }
7071
7072         old_dir->i_ctime = old_dir->i_mtime = ctime;
7073         new_dir->i_ctime = new_dir->i_mtime = ctime;
7074         old_inode->i_ctime = ctime;
7075
7076         if (old_dentry->d_parent != new_dentry->d_parent)
7077                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7078
7079         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7080                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7081                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7082                                         old_dentry->d_name.name,
7083                                         old_dentry->d_name.len);
7084         } else {
7085                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7086                                         old_dentry->d_inode,
7087                                         old_dentry->d_name.name,
7088                                         old_dentry->d_name.len);
7089                 if (!ret)
7090                         ret = btrfs_update_inode(trans, root, old_inode);
7091         }
7092         BUG_ON(ret);
7093
7094         if (new_inode) {
7095                 new_inode->i_ctime = CURRENT_TIME;
7096                 if (unlikely(new_inode->i_ino ==
7097                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7098                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7099                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7100                                                 root_objectid,
7101                                                 new_dentry->d_name.name,
7102                                                 new_dentry->d_name.len);
7103                         BUG_ON(new_inode->i_nlink == 0);
7104                 } else {
7105                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7106                                                  new_dentry->d_inode,
7107                                                  new_dentry->d_name.name,
7108                                                  new_dentry->d_name.len);
7109                 }
7110                 BUG_ON(ret);
7111                 if (new_inode->i_nlink == 0) {
7112                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7113                         BUG_ON(ret);
7114                 }
7115         }
7116
7117         fixup_inode_flags(new_dir, old_inode);
7118
7119         ret = btrfs_add_link(trans, new_dir, old_inode,
7120                              new_dentry->d_name.name,
7121                              new_dentry->d_name.len, 0, index);
7122         BUG_ON(ret);
7123
7124         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7125                 struct dentry *parent = dget_parent(new_dentry);
7126                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7127                 dput(parent);
7128                 btrfs_end_log_trans(root);
7129         }
7130 out_fail:
7131         btrfs_end_transaction_throttle(trans, root);
7132 out_notrans:
7133         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7134                 up_read(&root->fs_info->subvol_sem);
7135
7136         return ret;
7137 }
7138
7139 /*
7140  * some fairly slow code that needs optimization. This walks the list
7141  * of all the inodes with pending delalloc and forces them to disk.
7142  */
7143 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7144 {
7145         struct list_head *head = &root->fs_info->delalloc_inodes;
7146         struct btrfs_inode *binode;
7147         struct inode *inode;
7148
7149         if (root->fs_info->sb->s_flags & MS_RDONLY)
7150                 return -EROFS;
7151
7152         spin_lock(&root->fs_info->delalloc_lock);
7153         while (!list_empty(head)) {
7154                 binode = list_entry(head->next, struct btrfs_inode,
7155                                     delalloc_inodes);
7156                 inode = igrab(&binode->vfs_inode);
7157                 if (!inode)
7158                         list_del_init(&binode->delalloc_inodes);
7159                 spin_unlock(&root->fs_info->delalloc_lock);
7160                 if (inode) {
7161                         filemap_flush(inode->i_mapping);
7162                         if (delay_iput)
7163                                 btrfs_add_delayed_iput(inode);
7164                         else
7165                                 iput(inode);
7166                 }
7167                 cond_resched();
7168                 spin_lock(&root->fs_info->delalloc_lock);
7169         }
7170         spin_unlock(&root->fs_info->delalloc_lock);
7171
7172         /* the filemap_flush will queue IO into the worker threads, but
7173          * we have to make sure the IO is actually started and that
7174          * ordered extents get created before we return
7175          */
7176         atomic_inc(&root->fs_info->async_submit_draining);
7177         while (atomic_read(&root->fs_info->nr_async_submits) ||
7178               atomic_read(&root->fs_info->async_delalloc_pages)) {
7179                 wait_event(root->fs_info->async_submit_wait,
7180                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7181                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7182         }
7183         atomic_dec(&root->fs_info->async_submit_draining);
7184         return 0;
7185 }
7186
7187 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7188                                    int sync)
7189 {
7190         struct btrfs_inode *binode;
7191         struct inode *inode = NULL;
7192
7193         spin_lock(&root->fs_info->delalloc_lock);
7194         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7195                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7196                                     struct btrfs_inode, delalloc_inodes);
7197                 inode = igrab(&binode->vfs_inode);
7198                 if (inode) {
7199                         list_move_tail(&binode->delalloc_inodes,
7200                                        &root->fs_info->delalloc_inodes);
7201                         break;
7202                 }
7203
7204                 list_del_init(&binode->delalloc_inodes);
7205                 cond_resched_lock(&root->fs_info->delalloc_lock);
7206         }
7207         spin_unlock(&root->fs_info->delalloc_lock);
7208
7209         if (inode) {
7210                 if (sync) {
7211                         filemap_write_and_wait(inode->i_mapping);
7212                         /*
7213                          * We have to do this because compression doesn't
7214                          * actually set PG_writeback until it submits the pages
7215                          * for IO, which happens in an async thread, so we could
7216                          * race and not actually wait for any writeback pages
7217                          * because they've not been submitted yet.  Technically
7218                          * this could still be the case for the ordered stuff
7219                          * since the async thread may not have started to do its
7220                          * work yet.  If this becomes the case then we need to
7221                          * figure out a way to make sure that in writepage we
7222                          * wait for any async pages to be submitted before
7223                          * returning so that fdatawait does what its supposed to
7224                          * do.
7225                          */
7226                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7227                 } else {
7228                         filemap_flush(inode->i_mapping);
7229                 }
7230                 if (delay_iput)
7231                         btrfs_add_delayed_iput(inode);
7232                 else
7233                         iput(inode);
7234                 return 1;
7235         }
7236         return 0;
7237 }
7238
7239 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7240                          const char *symname)
7241 {
7242         struct btrfs_trans_handle *trans;
7243         struct btrfs_root *root = BTRFS_I(dir)->root;
7244         struct btrfs_path *path;
7245         struct btrfs_key key;
7246         struct inode *inode = NULL;
7247         int err;
7248         int drop_inode = 0;
7249         u64 objectid;
7250         u64 index = 0 ;
7251         int name_len;
7252         int datasize;
7253         unsigned long ptr;
7254         struct btrfs_file_extent_item *ei;
7255         struct extent_buffer *leaf;
7256         unsigned long nr = 0;
7257
7258         name_len = strlen(symname) + 1;
7259         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7260                 return -ENAMETOOLONG;
7261
7262         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7263         if (err)
7264                 return err;
7265         /*
7266          * 2 items for inode item and ref
7267          * 2 items for dir items
7268          * 1 item for xattr if selinux is on
7269          */
7270         trans = btrfs_start_transaction(root, 5);
7271         if (IS_ERR(trans))
7272                 return PTR_ERR(trans);
7273
7274         btrfs_set_trans_block_group(trans, dir);
7275
7276         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7277                                 dentry->d_name.len, dir->i_ino, objectid,
7278                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7279                                 &index);
7280         if (IS_ERR(inode)) {
7281                 err = PTR_ERR(inode);
7282                 goto out_unlock;
7283         }
7284
7285         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7286         if (err) {
7287                 drop_inode = 1;
7288                 goto out_unlock;
7289         }
7290
7291         btrfs_set_trans_block_group(trans, inode);
7292         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7293         if (err)
7294                 drop_inode = 1;
7295         else {
7296                 inode->i_mapping->a_ops = &btrfs_aops;
7297                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7298                 inode->i_fop = &btrfs_file_operations;
7299                 inode->i_op = &btrfs_file_inode_operations;
7300                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7301         }
7302         btrfs_update_inode_block_group(trans, inode);
7303         btrfs_update_inode_block_group(trans, dir);
7304         if (drop_inode)
7305                 goto out_unlock;
7306
7307         path = btrfs_alloc_path();
7308         BUG_ON(!path);
7309         key.objectid = inode->i_ino;
7310         key.offset = 0;
7311         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7312         datasize = btrfs_file_extent_calc_inline_size(name_len);
7313         err = btrfs_insert_empty_item(trans, root, path, &key,
7314                                       datasize);
7315         if (err) {
7316                 drop_inode = 1;
7317                 goto out_unlock;
7318         }
7319         leaf = path->nodes[0];
7320         ei = btrfs_item_ptr(leaf, path->slots[0],
7321                             struct btrfs_file_extent_item);
7322         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7323         btrfs_set_file_extent_type(leaf, ei,
7324                                    BTRFS_FILE_EXTENT_INLINE);
7325         btrfs_set_file_extent_encryption(leaf, ei, 0);
7326         btrfs_set_file_extent_compression(leaf, ei, 0);
7327         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7328         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7329
7330         ptr = btrfs_file_extent_inline_start(ei);
7331         write_extent_buffer(leaf, symname, ptr, name_len);
7332         btrfs_mark_buffer_dirty(leaf);
7333         btrfs_free_path(path);
7334
7335         inode->i_op = &btrfs_symlink_inode_operations;
7336         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7337         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7338         inode_set_bytes(inode, name_len);
7339         btrfs_i_size_write(inode, name_len - 1);
7340         err = btrfs_update_inode(trans, root, inode);
7341         if (err)
7342                 drop_inode = 1;
7343
7344 out_unlock:
7345         nr = trans->blocks_used;
7346         btrfs_end_transaction_throttle(trans, root);
7347         if (drop_inode) {
7348                 inode_dec_link_count(inode);
7349                 iput(inode);
7350         }
7351         btrfs_btree_balance_dirty(root, nr);
7352         return err;
7353 }
7354
7355 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7356                                        u64 start, u64 num_bytes, u64 min_size,
7357                                        loff_t actual_len, u64 *alloc_hint,
7358                                        struct btrfs_trans_handle *trans)
7359 {
7360         struct btrfs_root *root = BTRFS_I(inode)->root;
7361         struct btrfs_key ins;
7362         u64 cur_offset = start;
7363         u64 i_size;
7364         int ret = 0;
7365         bool own_trans = true;
7366
7367         if (trans)
7368                 own_trans = false;
7369         while (num_bytes > 0) {
7370                 if (own_trans) {
7371                         trans = btrfs_start_transaction(root, 3);
7372                         if (IS_ERR(trans)) {
7373                                 ret = PTR_ERR(trans);
7374                                 break;
7375                         }
7376                 }
7377
7378                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7379                                            0, *alloc_hint, (u64)-1, &ins, 1);
7380                 if (ret) {
7381                         if (own_trans)
7382                                 btrfs_end_transaction(trans, root);
7383                         break;
7384                 }
7385
7386                 ret = insert_reserved_file_extent(trans, inode,
7387                                                   cur_offset, ins.objectid,
7388                                                   ins.offset, ins.offset,
7389                                                   ins.offset, 0, 0, 0,
7390                                                   BTRFS_FILE_EXTENT_PREALLOC);
7391                 BUG_ON(ret);
7392                 btrfs_drop_extent_cache(inode, cur_offset,
7393                                         cur_offset + ins.offset -1, 0);
7394
7395                 num_bytes -= ins.offset;
7396                 cur_offset += ins.offset;
7397                 *alloc_hint = ins.objectid + ins.offset;
7398
7399                 inode->i_ctime = CURRENT_TIME;
7400                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7401                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7402                     (actual_len > inode->i_size) &&
7403                     (cur_offset > inode->i_size)) {
7404                         if (cur_offset > actual_len)
7405                                 i_size = actual_len;
7406                         else
7407                                 i_size = cur_offset;
7408                         i_size_write(inode, i_size);
7409                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7410                 }
7411
7412                 ret = btrfs_update_inode(trans, root, inode);
7413                 BUG_ON(ret);
7414
7415                 if (own_trans)
7416                         btrfs_end_transaction(trans, root);
7417         }
7418         return ret;
7419 }
7420
7421 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7422                               u64 start, u64 num_bytes, u64 min_size,
7423                               loff_t actual_len, u64 *alloc_hint)
7424 {
7425         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7426                                            min_size, actual_len, alloc_hint,
7427                                            NULL);
7428 }
7429
7430 int btrfs_prealloc_file_range_trans(struct inode *inode,
7431                                     struct btrfs_trans_handle *trans, int mode,
7432                                     u64 start, u64 num_bytes, u64 min_size,
7433                                     loff_t actual_len, u64 *alloc_hint)
7434 {
7435         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7436                                            min_size, actual_len, alloc_hint, trans);
7437 }
7438
7439 static int btrfs_set_page_dirty(struct page *page)
7440 {
7441         return __set_page_dirty_nobuffers(page);
7442 }
7443
7444 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7445 {
7446         struct btrfs_root *root = BTRFS_I(inode)->root;
7447
7448         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7449                 return -EROFS;
7450         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7451                 return -EACCES;
7452         return generic_permission(inode, mask, flags, btrfs_check_acl);
7453 }
7454
7455 static const struct inode_operations btrfs_dir_inode_operations = {
7456         .getattr        = btrfs_getattr,
7457         .lookup         = btrfs_lookup,
7458         .create         = btrfs_create,
7459         .unlink         = btrfs_unlink,
7460         .link           = btrfs_link,
7461         .mkdir          = btrfs_mkdir,
7462         .rmdir          = btrfs_rmdir,
7463         .rename         = btrfs_rename,
7464         .symlink        = btrfs_symlink,
7465         .setattr        = btrfs_setattr,
7466         .mknod          = btrfs_mknod,
7467         .setxattr       = btrfs_setxattr,
7468         .getxattr       = btrfs_getxattr,
7469         .listxattr      = btrfs_listxattr,
7470         .removexattr    = btrfs_removexattr,
7471         .permission     = btrfs_permission,
7472 };
7473 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7474         .lookup         = btrfs_lookup,
7475         .permission     = btrfs_permission,
7476 };
7477
7478 static const struct file_operations btrfs_dir_file_operations = {
7479         .llseek         = generic_file_llseek,
7480         .read           = generic_read_dir,
7481         .readdir        = btrfs_real_readdir,
7482         .unlocked_ioctl = btrfs_ioctl,
7483 #ifdef CONFIG_COMPAT
7484         .compat_ioctl   = btrfs_ioctl,
7485 #endif
7486         .release        = btrfs_release_file,
7487         .fsync          = btrfs_sync_file,
7488 };
7489
7490 static struct extent_io_ops btrfs_extent_io_ops = {
7491         .fill_delalloc = run_delalloc_range,
7492         .submit_bio_hook = btrfs_submit_bio_hook,
7493         .merge_bio_hook = btrfs_merge_bio_hook,
7494         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7495         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7496         .writepage_start_hook = btrfs_writepage_start_hook,
7497         .readpage_io_failed_hook = btrfs_io_failed_hook,
7498         .set_bit_hook = btrfs_set_bit_hook,
7499         .clear_bit_hook = btrfs_clear_bit_hook,
7500         .merge_extent_hook = btrfs_merge_extent_hook,
7501         .split_extent_hook = btrfs_split_extent_hook,
7502 };
7503
7504 /*
7505  * btrfs doesn't support the bmap operation because swapfiles
7506  * use bmap to make a mapping of extents in the file.  They assume
7507  * these extents won't change over the life of the file and they
7508  * use the bmap result to do IO directly to the drive.
7509  *
7510  * the btrfs bmap call would return logical addresses that aren't
7511  * suitable for IO and they also will change frequently as COW
7512  * operations happen.  So, swapfile + btrfs == corruption.
7513  *
7514  * For now we're avoiding this by dropping bmap.
7515  */
7516 static const struct address_space_operations btrfs_aops = {
7517         .readpage       = btrfs_readpage,
7518         .writepage      = btrfs_writepage,
7519         .writepages     = btrfs_writepages,
7520         .readpages      = btrfs_readpages,
7521         .direct_IO      = btrfs_direct_IO,
7522         .invalidatepage = btrfs_invalidatepage,
7523         .releasepage    = btrfs_releasepage,
7524         .set_page_dirty = btrfs_set_page_dirty,
7525         .error_remove_page = generic_error_remove_page,
7526 };
7527
7528 static const struct address_space_operations btrfs_symlink_aops = {
7529         .readpage       = btrfs_readpage,
7530         .writepage      = btrfs_writepage,
7531         .invalidatepage = btrfs_invalidatepage,
7532         .releasepage    = btrfs_releasepage,
7533 };
7534
7535 static const struct inode_operations btrfs_file_inode_operations = {
7536         .getattr        = btrfs_getattr,
7537         .setattr        = btrfs_setattr,
7538         .setxattr       = btrfs_setxattr,
7539         .getxattr       = btrfs_getxattr,
7540         .listxattr      = btrfs_listxattr,
7541         .removexattr    = btrfs_removexattr,
7542         .permission     = btrfs_permission,
7543         .fiemap         = btrfs_fiemap,
7544 };
7545 static const struct inode_operations btrfs_special_inode_operations = {
7546         .getattr        = btrfs_getattr,
7547         .setattr        = btrfs_setattr,
7548         .permission     = btrfs_permission,
7549         .setxattr       = btrfs_setxattr,
7550         .getxattr       = btrfs_getxattr,
7551         .listxattr      = btrfs_listxattr,
7552         .removexattr    = btrfs_removexattr,
7553 };
7554 static const struct inode_operations btrfs_symlink_inode_operations = {
7555         .readlink       = generic_readlink,
7556         .follow_link    = page_follow_link_light,
7557         .put_link       = page_put_link,
7558         .getattr        = btrfs_getattr,
7559         .permission     = btrfs_permission,
7560         .setxattr       = btrfs_setxattr,
7561         .getxattr       = btrfs_getxattr,
7562         .listxattr      = btrfs_listxattr,
7563         .removexattr    = btrfs_removexattr,
7564 };
7565
7566 const struct dentry_operations btrfs_dentry_operations = {
7567         .d_delete       = btrfs_dentry_delete,
7568 };