Btrfs: add new defrag-range ioctl.
[linux-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include "compat.h"
40 #include "ctree.h"
41 #include "disk-io.h"
42 #include "transaction.h"
43 #include "btrfs_inode.h"
44 #include "ioctl.h"
45 #include "print-tree.h"
46 #include "volumes.h"
47 #include "ordered-data.h"
48 #include "xattr.h"
49 #include "tree-log.h"
50 #include "compression.h"
51 #include "locking.h"
52
53 struct btrfs_iget_args {
54         u64 ino;
55         struct btrfs_root *root;
56 };
57
58 static const struct inode_operations btrfs_dir_inode_operations;
59 static const struct inode_operations btrfs_symlink_inode_operations;
60 static const struct inode_operations btrfs_dir_ro_inode_operations;
61 static const struct inode_operations btrfs_special_inode_operations;
62 static const struct inode_operations btrfs_file_inode_operations;
63 static const struct address_space_operations btrfs_aops;
64 static const struct address_space_operations btrfs_symlink_aops;
65 static const struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
67
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
72
73 #define S_SHIFT 12
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
76         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
77         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
78         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
79         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
80         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
81         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
82 };
83
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87                                    struct page *locked_page,
88                                    u64 start, u64 end, int *page_started,
89                                    unsigned long *nr_written, int unlock);
90
91 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
92                                      struct inode *inode,  struct inode *dir)
93 {
94         int err;
95
96         err = btrfs_init_acl(trans, inode, dir);
97         if (!err)
98                 err = btrfs_xattr_security_init(trans, inode, dir);
99         return err;
100 }
101
102 /*
103  * this does all the hard work for inserting an inline extent into
104  * the btree.  The caller should have done a btrfs_drop_extents so that
105  * no overlapping inline items exist in the btree
106  */
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108                                 struct btrfs_root *root, struct inode *inode,
109                                 u64 start, size_t size, size_t compressed_size,
110                                 struct page **compressed_pages)
111 {
112         struct btrfs_key key;
113         struct btrfs_path *path;
114         struct extent_buffer *leaf;
115         struct page *page = NULL;
116         char *kaddr;
117         unsigned long ptr;
118         struct btrfs_file_extent_item *ei;
119         int err = 0;
120         int ret;
121         size_t cur_size = size;
122         size_t datasize;
123         unsigned long offset;
124         int use_compress = 0;
125
126         if (compressed_size && compressed_pages) {
127                 use_compress = 1;
128                 cur_size = compressed_size;
129         }
130
131         path = btrfs_alloc_path();
132         if (!path)
133                 return -ENOMEM;
134
135         path->leave_spinning = 1;
136         btrfs_set_trans_block_group(trans, inode);
137
138         key.objectid = inode->i_ino;
139         key.offset = start;
140         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141         datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143         inode_add_bytes(inode, size);
144         ret = btrfs_insert_empty_item(trans, root, path, &key,
145                                       datasize);
146         BUG_ON(ret);
147         if (ret) {
148                 err = ret;
149                 goto fail;
150         }
151         leaf = path->nodes[0];
152         ei = btrfs_item_ptr(leaf, path->slots[0],
153                             struct btrfs_file_extent_item);
154         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156         btrfs_set_file_extent_encryption(leaf, ei, 0);
157         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159         ptr = btrfs_file_extent_inline_start(ei);
160
161         if (use_compress) {
162                 struct page *cpage;
163                 int i = 0;
164                 while (compressed_size > 0) {
165                         cpage = compressed_pages[i];
166                         cur_size = min_t(unsigned long, compressed_size,
167                                        PAGE_CACHE_SIZE);
168
169                         kaddr = kmap_atomic(cpage, KM_USER0);
170                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
171                         kunmap_atomic(kaddr, KM_USER0);
172
173                         i++;
174                         ptr += cur_size;
175                         compressed_size -= cur_size;
176                 }
177                 btrfs_set_file_extent_compression(leaf, ei,
178                                                   BTRFS_COMPRESS_ZLIB);
179         } else {
180                 page = find_get_page(inode->i_mapping,
181                                      start >> PAGE_CACHE_SHIFT);
182                 btrfs_set_file_extent_compression(leaf, ei, 0);
183                 kaddr = kmap_atomic(page, KM_USER0);
184                 offset = start & (PAGE_CACHE_SIZE - 1);
185                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186                 kunmap_atomic(kaddr, KM_USER0);
187                 page_cache_release(page);
188         }
189         btrfs_mark_buffer_dirty(leaf);
190         btrfs_free_path(path);
191
192         /*
193          * we're an inline extent, so nobody can
194          * extend the file past i_size without locking
195          * a page we already have locked.
196          *
197          * We must do any isize and inode updates
198          * before we unlock the pages.  Otherwise we
199          * could end up racing with unlink.
200          */
201         BTRFS_I(inode)->disk_i_size = inode->i_size;
202         btrfs_update_inode(trans, root, inode);
203
204         return 0;
205 fail:
206         btrfs_free_path(path);
207         return err;
208 }
209
210
211 /*
212  * conditionally insert an inline extent into the file.  This
213  * does the checks required to make sure the data is small enough
214  * to fit as an inline extent.
215  */
216 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
217                                  struct btrfs_root *root,
218                                  struct inode *inode, u64 start, u64 end,
219                                  size_t compressed_size,
220                                  struct page **compressed_pages)
221 {
222         u64 isize = i_size_read(inode);
223         u64 actual_end = min(end + 1, isize);
224         u64 inline_len = actual_end - start;
225         u64 aligned_end = (end + root->sectorsize - 1) &
226                         ~((u64)root->sectorsize - 1);
227         u64 hint_byte;
228         u64 data_len = inline_len;
229         int ret;
230
231         if (compressed_size)
232                 data_len = compressed_size;
233
234         if (start > 0 ||
235             actual_end >= PAGE_CACHE_SIZE ||
236             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
237             (!compressed_size &&
238             (actual_end & (root->sectorsize - 1)) == 0) ||
239             end + 1 < isize ||
240             data_len > root->fs_info->max_inline) {
241                 return 1;
242         }
243
244         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
245                                  &hint_byte, 1);
246         BUG_ON(ret);
247
248         if (isize > actual_end)
249                 inline_len = min_t(u64, isize, actual_end);
250         ret = insert_inline_extent(trans, root, inode, start,
251                                    inline_len, compressed_size,
252                                    compressed_pages);
253         BUG_ON(ret);
254         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
255         return 0;
256 }
257
258 struct async_extent {
259         u64 start;
260         u64 ram_size;
261         u64 compressed_size;
262         struct page **pages;
263         unsigned long nr_pages;
264         struct list_head list;
265 };
266
267 struct async_cow {
268         struct inode *inode;
269         struct btrfs_root *root;
270         struct page *locked_page;
271         u64 start;
272         u64 end;
273         struct list_head extents;
274         struct btrfs_work work;
275 };
276
277 static noinline int add_async_extent(struct async_cow *cow,
278                                      u64 start, u64 ram_size,
279                                      u64 compressed_size,
280                                      struct page **pages,
281                                      unsigned long nr_pages)
282 {
283         struct async_extent *async_extent;
284
285         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
286         async_extent->start = start;
287         async_extent->ram_size = ram_size;
288         async_extent->compressed_size = compressed_size;
289         async_extent->pages = pages;
290         async_extent->nr_pages = nr_pages;
291         list_add_tail(&async_extent->list, &cow->extents);
292         return 0;
293 }
294
295 /*
296  * we create compressed extents in two phases.  The first
297  * phase compresses a range of pages that have already been
298  * locked (both pages and state bits are locked).
299  *
300  * This is done inside an ordered work queue, and the compression
301  * is spread across many cpus.  The actual IO submission is step
302  * two, and the ordered work queue takes care of making sure that
303  * happens in the same order things were put onto the queue by
304  * writepages and friends.
305  *
306  * If this code finds it can't get good compression, it puts an
307  * entry onto the work queue to write the uncompressed bytes.  This
308  * makes sure that both compressed inodes and uncompressed inodes
309  * are written in the same order that pdflush sent them down.
310  */
311 static noinline int compress_file_range(struct inode *inode,
312                                         struct page *locked_page,
313                                         u64 start, u64 end,
314                                         struct async_cow *async_cow,
315                                         int *num_added)
316 {
317         struct btrfs_root *root = BTRFS_I(inode)->root;
318         struct btrfs_trans_handle *trans;
319         u64 num_bytes;
320         u64 orig_start;
321         u64 disk_num_bytes;
322         u64 blocksize = root->sectorsize;
323         u64 actual_end;
324         u64 isize = i_size_read(inode);
325         int ret = 0;
326         struct page **pages = NULL;
327         unsigned long nr_pages;
328         unsigned long nr_pages_ret = 0;
329         unsigned long total_compressed = 0;
330         unsigned long total_in = 0;
331         unsigned long max_compressed = 128 * 1024;
332         unsigned long max_uncompressed = 128 * 1024;
333         int i;
334         int will_compress;
335
336         orig_start = start;
337
338         actual_end = min_t(u64, isize, end + 1);
339 again:
340         will_compress = 0;
341         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
342         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
343
344         /*
345          * we don't want to send crud past the end of i_size through
346          * compression, that's just a waste of CPU time.  So, if the
347          * end of the file is before the start of our current
348          * requested range of bytes, we bail out to the uncompressed
349          * cleanup code that can deal with all of this.
350          *
351          * It isn't really the fastest way to fix things, but this is a
352          * very uncommon corner.
353          */
354         if (actual_end <= start)
355                 goto cleanup_and_bail_uncompressed;
356
357         total_compressed = actual_end - start;
358
359         /* we want to make sure that amount of ram required to uncompress
360          * an extent is reasonable, so we limit the total size in ram
361          * of a compressed extent to 128k.  This is a crucial number
362          * because it also controls how easily we can spread reads across
363          * cpus for decompression.
364          *
365          * We also want to make sure the amount of IO required to do
366          * a random read is reasonably small, so we limit the size of
367          * a compressed extent to 128k.
368          */
369         total_compressed = min(total_compressed, max_uncompressed);
370         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
371         num_bytes = max(blocksize,  num_bytes);
372         disk_num_bytes = num_bytes;
373         total_in = 0;
374         ret = 0;
375
376         /*
377          * we do compression for mount -o compress and when the
378          * inode has not been flagged as nocompress.  This flag can
379          * change at any time if we discover bad compression ratios.
380          */
381         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
382             (btrfs_test_opt(root, COMPRESS) ||
383              (BTRFS_I(inode)->force_compress))) {
384                 WARN_ON(pages);
385                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
386
387                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
388                                                 total_compressed, pages,
389                                                 nr_pages, &nr_pages_ret,
390                                                 &total_in,
391                                                 &total_compressed,
392                                                 max_compressed);
393
394                 if (!ret) {
395                         unsigned long offset = total_compressed &
396                                 (PAGE_CACHE_SIZE - 1);
397                         struct page *page = pages[nr_pages_ret - 1];
398                         char *kaddr;
399
400                         /* zero the tail end of the last page, we might be
401                          * sending it down to disk
402                          */
403                         if (offset) {
404                                 kaddr = kmap_atomic(page, KM_USER0);
405                                 memset(kaddr + offset, 0,
406                                        PAGE_CACHE_SIZE - offset);
407                                 kunmap_atomic(kaddr, KM_USER0);
408                         }
409                         will_compress = 1;
410                 }
411         }
412         if (start == 0) {
413                 trans = btrfs_join_transaction(root, 1);
414                 BUG_ON(!trans);
415                 btrfs_set_trans_block_group(trans, inode);
416
417                 /* lets try to make an inline extent */
418                 if (ret || total_in < (actual_end - start)) {
419                         /* we didn't compress the entire range, try
420                          * to make an uncompressed inline extent.
421                          */
422                         ret = cow_file_range_inline(trans, root, inode,
423                                                     start, end, 0, NULL);
424                 } else {
425                         /* try making a compressed inline extent */
426                         ret = cow_file_range_inline(trans, root, inode,
427                                                     start, end,
428                                                     total_compressed, pages);
429                 }
430                 if (ret == 0) {
431                         /*
432                          * inline extent creation worked, we don't need
433                          * to create any more async work items.  Unlock
434                          * and free up our temp pages.
435                          */
436                         extent_clear_unlock_delalloc(inode,
437                              &BTRFS_I(inode)->io_tree,
438                              start, end, NULL,
439                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
440                              EXTENT_CLEAR_DELALLOC |
441                              EXTENT_CLEAR_ACCOUNTING |
442                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
443
444                         btrfs_end_transaction(trans, root);
445                         goto free_pages_out;
446                 }
447                 btrfs_end_transaction(trans, root);
448         }
449
450         if (will_compress) {
451                 /*
452                  * we aren't doing an inline extent round the compressed size
453                  * up to a block size boundary so the allocator does sane
454                  * things
455                  */
456                 total_compressed = (total_compressed + blocksize - 1) &
457                         ~(blocksize - 1);
458
459                 /*
460                  * one last check to make sure the compression is really a
461                  * win, compare the page count read with the blocks on disk
462                  */
463                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
464                         ~(PAGE_CACHE_SIZE - 1);
465                 if (total_compressed >= total_in) {
466                         will_compress = 0;
467                 } else {
468                         disk_num_bytes = total_compressed;
469                         num_bytes = total_in;
470                 }
471         }
472         if (!will_compress && pages) {
473                 /*
474                  * the compression code ran but failed to make things smaller,
475                  * free any pages it allocated and our page pointer array
476                  */
477                 for (i = 0; i < nr_pages_ret; i++) {
478                         WARN_ON(pages[i]->mapping);
479                         page_cache_release(pages[i]);
480                 }
481                 kfree(pages);
482                 pages = NULL;
483                 total_compressed = 0;
484                 nr_pages_ret = 0;
485
486                 /* flag the file so we don't compress in the future */
487                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
488                     !(BTRFS_I(inode)->force_compress)) {
489                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
490                 }
491         }
492         if (will_compress) {
493                 *num_added += 1;
494
495                 /* the async work queues will take care of doing actual
496                  * allocation on disk for these compressed pages,
497                  * and will submit them to the elevator.
498                  */
499                 add_async_extent(async_cow, start, num_bytes,
500                                  total_compressed, pages, nr_pages_ret);
501
502                 if (start + num_bytes < end && start + num_bytes < actual_end) {
503                         start += num_bytes;
504                         pages = NULL;
505                         cond_resched();
506                         goto again;
507                 }
508         } else {
509 cleanup_and_bail_uncompressed:
510                 /*
511                  * No compression, but we still need to write the pages in
512                  * the file we've been given so far.  redirty the locked
513                  * page if it corresponds to our extent and set things up
514                  * for the async work queue to run cow_file_range to do
515                  * the normal delalloc dance
516                  */
517                 if (page_offset(locked_page) >= start &&
518                     page_offset(locked_page) <= end) {
519                         __set_page_dirty_nobuffers(locked_page);
520                         /* unlocked later on in the async handlers */
521                 }
522                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
523                 *num_added += 1;
524         }
525
526 out:
527         return 0;
528
529 free_pages_out:
530         for (i = 0; i < nr_pages_ret; i++) {
531                 WARN_ON(pages[i]->mapping);
532                 page_cache_release(pages[i]);
533         }
534         kfree(pages);
535
536         goto out;
537 }
538
539 /*
540  * phase two of compressed writeback.  This is the ordered portion
541  * of the code, which only gets called in the order the work was
542  * queued.  We walk all the async extents created by compress_file_range
543  * and send them down to the disk.
544  */
545 static noinline int submit_compressed_extents(struct inode *inode,
546                                               struct async_cow *async_cow)
547 {
548         struct async_extent *async_extent;
549         u64 alloc_hint = 0;
550         struct btrfs_trans_handle *trans;
551         struct btrfs_key ins;
552         struct extent_map *em;
553         struct btrfs_root *root = BTRFS_I(inode)->root;
554         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
555         struct extent_io_tree *io_tree;
556         int ret = 0;
557
558         if (list_empty(&async_cow->extents))
559                 return 0;
560
561
562         while (!list_empty(&async_cow->extents)) {
563                 async_extent = list_entry(async_cow->extents.next,
564                                           struct async_extent, list);
565                 list_del(&async_extent->list);
566
567                 io_tree = &BTRFS_I(inode)->io_tree;
568
569 retry:
570                 /* did the compression code fall back to uncompressed IO? */
571                 if (!async_extent->pages) {
572                         int page_started = 0;
573                         unsigned long nr_written = 0;
574
575                         lock_extent(io_tree, async_extent->start,
576                                     async_extent->start +
577                                     async_extent->ram_size - 1, GFP_NOFS);
578
579                         /* allocate blocks */
580                         ret = cow_file_range(inode, async_cow->locked_page,
581                                              async_extent->start,
582                                              async_extent->start +
583                                              async_extent->ram_size - 1,
584                                              &page_started, &nr_written, 0);
585
586                         /*
587                          * if page_started, cow_file_range inserted an
588                          * inline extent and took care of all the unlocking
589                          * and IO for us.  Otherwise, we need to submit
590                          * all those pages down to the drive.
591                          */
592                         if (!page_started && !ret)
593                                 extent_write_locked_range(io_tree,
594                                                   inode, async_extent->start,
595                                                   async_extent->start +
596                                                   async_extent->ram_size - 1,
597                                                   btrfs_get_extent,
598                                                   WB_SYNC_ALL);
599                         kfree(async_extent);
600                         cond_resched();
601                         continue;
602                 }
603
604                 lock_extent(io_tree, async_extent->start,
605                             async_extent->start + async_extent->ram_size - 1,
606                             GFP_NOFS);
607
608                 trans = btrfs_join_transaction(root, 1);
609                 ret = btrfs_reserve_extent(trans, root,
610                                            async_extent->compressed_size,
611                                            async_extent->compressed_size,
612                                            0, alloc_hint,
613                                            (u64)-1, &ins, 1);
614                 btrfs_end_transaction(trans, root);
615
616                 if (ret) {
617                         int i;
618                         for (i = 0; i < async_extent->nr_pages; i++) {
619                                 WARN_ON(async_extent->pages[i]->mapping);
620                                 page_cache_release(async_extent->pages[i]);
621                         }
622                         kfree(async_extent->pages);
623                         async_extent->nr_pages = 0;
624                         async_extent->pages = NULL;
625                         unlock_extent(io_tree, async_extent->start,
626                                       async_extent->start +
627                                       async_extent->ram_size - 1, GFP_NOFS);
628                         goto retry;
629                 }
630
631                 /*
632                  * here we're doing allocation and writeback of the
633                  * compressed pages
634                  */
635                 btrfs_drop_extent_cache(inode, async_extent->start,
636                                         async_extent->start +
637                                         async_extent->ram_size - 1, 0);
638
639                 em = alloc_extent_map(GFP_NOFS);
640                 em->start = async_extent->start;
641                 em->len = async_extent->ram_size;
642                 em->orig_start = em->start;
643
644                 em->block_start = ins.objectid;
645                 em->block_len = ins.offset;
646                 em->bdev = root->fs_info->fs_devices->latest_bdev;
647                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
648                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
649
650                 while (1) {
651                         write_lock(&em_tree->lock);
652                         ret = add_extent_mapping(em_tree, em);
653                         write_unlock(&em_tree->lock);
654                         if (ret != -EEXIST) {
655                                 free_extent_map(em);
656                                 break;
657                         }
658                         btrfs_drop_extent_cache(inode, async_extent->start,
659                                                 async_extent->start +
660                                                 async_extent->ram_size - 1, 0);
661                 }
662
663                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
664                                                ins.objectid,
665                                                async_extent->ram_size,
666                                                ins.offset,
667                                                BTRFS_ORDERED_COMPRESSED);
668                 BUG_ON(ret);
669
670                 /*
671                  * clear dirty, set writeback and unlock the pages.
672                  */
673                 extent_clear_unlock_delalloc(inode,
674                                 &BTRFS_I(inode)->io_tree,
675                                 async_extent->start,
676                                 async_extent->start +
677                                 async_extent->ram_size - 1,
678                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
679                                 EXTENT_CLEAR_UNLOCK |
680                                 EXTENT_CLEAR_DELALLOC |
681                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
682
683                 ret = btrfs_submit_compressed_write(inode,
684                                     async_extent->start,
685                                     async_extent->ram_size,
686                                     ins.objectid,
687                                     ins.offset, async_extent->pages,
688                                     async_extent->nr_pages);
689
690                 BUG_ON(ret);
691                 alloc_hint = ins.objectid + ins.offset;
692                 kfree(async_extent);
693                 cond_resched();
694         }
695
696         return 0;
697 }
698
699 /*
700  * when extent_io.c finds a delayed allocation range in the file,
701  * the call backs end up in this code.  The basic idea is to
702  * allocate extents on disk for the range, and create ordered data structs
703  * in ram to track those extents.
704  *
705  * locked_page is the page that writepage had locked already.  We use
706  * it to make sure we don't do extra locks or unlocks.
707  *
708  * *page_started is set to one if we unlock locked_page and do everything
709  * required to start IO on it.  It may be clean and already done with
710  * IO when we return.
711  */
712 static noinline int cow_file_range(struct inode *inode,
713                                    struct page *locked_page,
714                                    u64 start, u64 end, int *page_started,
715                                    unsigned long *nr_written,
716                                    int unlock)
717 {
718         struct btrfs_root *root = BTRFS_I(inode)->root;
719         struct btrfs_trans_handle *trans;
720         u64 alloc_hint = 0;
721         u64 num_bytes;
722         unsigned long ram_size;
723         u64 disk_num_bytes;
724         u64 cur_alloc_size;
725         u64 blocksize = root->sectorsize;
726         u64 actual_end;
727         u64 isize = i_size_read(inode);
728         struct btrfs_key ins;
729         struct extent_map *em;
730         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731         int ret = 0;
732
733         trans = btrfs_join_transaction(root, 1);
734         BUG_ON(!trans);
735         btrfs_set_trans_block_group(trans, inode);
736
737         actual_end = min_t(u64, isize, end + 1);
738
739         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
740         num_bytes = max(blocksize,  num_bytes);
741         disk_num_bytes = num_bytes;
742         ret = 0;
743
744         if (start == 0) {
745                 /* lets try to make an inline extent */
746                 ret = cow_file_range_inline(trans, root, inode,
747                                             start, end, 0, NULL);
748                 if (ret == 0) {
749                         extent_clear_unlock_delalloc(inode,
750                                      &BTRFS_I(inode)->io_tree,
751                                      start, end, NULL,
752                                      EXTENT_CLEAR_UNLOCK_PAGE |
753                                      EXTENT_CLEAR_UNLOCK |
754                                      EXTENT_CLEAR_DELALLOC |
755                                      EXTENT_CLEAR_ACCOUNTING |
756                                      EXTENT_CLEAR_DIRTY |
757                                      EXTENT_SET_WRITEBACK |
758                                      EXTENT_END_WRITEBACK);
759
760                         *nr_written = *nr_written +
761                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
762                         *page_started = 1;
763                         ret = 0;
764                         goto out;
765                 }
766         }
767
768         BUG_ON(disk_num_bytes >
769                btrfs_super_total_bytes(&root->fs_info->super_copy));
770
771
772         read_lock(&BTRFS_I(inode)->extent_tree.lock);
773         em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
774                                    start, num_bytes);
775         if (em) {
776                 /*
777                  * if block start isn't an actual block number then find the
778                  * first block in this inode and use that as a hint.  If that
779                  * block is also bogus then just don't worry about it.
780                  */
781                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
782                         free_extent_map(em);
783                         em = search_extent_mapping(em_tree, 0, 0);
784                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
785                                 alloc_hint = em->block_start;
786                         if (em)
787                                 free_extent_map(em);
788                 } else {
789                         alloc_hint = em->block_start;
790                         free_extent_map(em);
791                 }
792         }
793         read_unlock(&BTRFS_I(inode)->extent_tree.lock);
794         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
795
796         while (disk_num_bytes > 0) {
797                 unsigned long op;
798
799                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
800                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
801                                            root->sectorsize, 0, alloc_hint,
802                                            (u64)-1, &ins, 1);
803                 BUG_ON(ret);
804
805                 em = alloc_extent_map(GFP_NOFS);
806                 em->start = start;
807                 em->orig_start = em->start;
808                 ram_size = ins.offset;
809                 em->len = ins.offset;
810
811                 em->block_start = ins.objectid;
812                 em->block_len = ins.offset;
813                 em->bdev = root->fs_info->fs_devices->latest_bdev;
814                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
815
816                 while (1) {
817                         write_lock(&em_tree->lock);
818                         ret = add_extent_mapping(em_tree, em);
819                         write_unlock(&em_tree->lock);
820                         if (ret != -EEXIST) {
821                                 free_extent_map(em);
822                                 break;
823                         }
824                         btrfs_drop_extent_cache(inode, start,
825                                                 start + ram_size - 1, 0);
826                 }
827
828                 cur_alloc_size = ins.offset;
829                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
830                                                ram_size, cur_alloc_size, 0);
831                 BUG_ON(ret);
832
833                 if (root->root_key.objectid ==
834                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
835                         ret = btrfs_reloc_clone_csums(inode, start,
836                                                       cur_alloc_size);
837                         BUG_ON(ret);
838                 }
839
840                 if (disk_num_bytes < cur_alloc_size)
841                         break;
842
843                 /* we're not doing compressed IO, don't unlock the first
844                  * page (which the caller expects to stay locked), don't
845                  * clear any dirty bits and don't set any writeback bits
846                  *
847                  * Do set the Private2 bit so we know this page was properly
848                  * setup for writepage
849                  */
850                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
851                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
852                         EXTENT_SET_PRIVATE2;
853
854                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
855                                              start, start + ram_size - 1,
856                                              locked_page, op);
857                 disk_num_bytes -= cur_alloc_size;
858                 num_bytes -= cur_alloc_size;
859                 alloc_hint = ins.objectid + ins.offset;
860                 start += cur_alloc_size;
861         }
862 out:
863         ret = 0;
864         btrfs_end_transaction(trans, root);
865
866         return ret;
867 }
868
869 /*
870  * work queue call back to started compression on a file and pages
871  */
872 static noinline void async_cow_start(struct btrfs_work *work)
873 {
874         struct async_cow *async_cow;
875         int num_added = 0;
876         async_cow = container_of(work, struct async_cow, work);
877
878         compress_file_range(async_cow->inode, async_cow->locked_page,
879                             async_cow->start, async_cow->end, async_cow,
880                             &num_added);
881         if (num_added == 0)
882                 async_cow->inode = NULL;
883 }
884
885 /*
886  * work queue call back to submit previously compressed pages
887  */
888 static noinline void async_cow_submit(struct btrfs_work *work)
889 {
890         struct async_cow *async_cow;
891         struct btrfs_root *root;
892         unsigned long nr_pages;
893
894         async_cow = container_of(work, struct async_cow, work);
895
896         root = async_cow->root;
897         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
898                 PAGE_CACHE_SHIFT;
899
900         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
901
902         if (atomic_read(&root->fs_info->async_delalloc_pages) <
903             5 * 1042 * 1024 &&
904             waitqueue_active(&root->fs_info->async_submit_wait))
905                 wake_up(&root->fs_info->async_submit_wait);
906
907         if (async_cow->inode)
908                 submit_compressed_extents(async_cow->inode, async_cow);
909 }
910
911 static noinline void async_cow_free(struct btrfs_work *work)
912 {
913         struct async_cow *async_cow;
914         async_cow = container_of(work, struct async_cow, work);
915         kfree(async_cow);
916 }
917
918 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
919                                 u64 start, u64 end, int *page_started,
920                                 unsigned long *nr_written)
921 {
922         struct async_cow *async_cow;
923         struct btrfs_root *root = BTRFS_I(inode)->root;
924         unsigned long nr_pages;
925         u64 cur_end;
926         int limit = 10 * 1024 * 1042;
927
928         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
929                          1, 0, NULL, GFP_NOFS);
930         while (start < end) {
931                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
932                 async_cow->inode = inode;
933                 async_cow->root = root;
934                 async_cow->locked_page = locked_page;
935                 async_cow->start = start;
936
937                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
938                         cur_end = end;
939                 else
940                         cur_end = min(end, start + 512 * 1024 - 1);
941
942                 async_cow->end = cur_end;
943                 INIT_LIST_HEAD(&async_cow->extents);
944
945                 async_cow->work.func = async_cow_start;
946                 async_cow->work.ordered_func = async_cow_submit;
947                 async_cow->work.ordered_free = async_cow_free;
948                 async_cow->work.flags = 0;
949
950                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
951                         PAGE_CACHE_SHIFT;
952                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
953
954                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
955                                    &async_cow->work);
956
957                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
958                         wait_event(root->fs_info->async_submit_wait,
959                            (atomic_read(&root->fs_info->async_delalloc_pages) <
960                             limit));
961                 }
962
963                 while (atomic_read(&root->fs_info->async_submit_draining) &&
964                       atomic_read(&root->fs_info->async_delalloc_pages)) {
965                         wait_event(root->fs_info->async_submit_wait,
966                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
967                            0));
968                 }
969
970                 *nr_written += nr_pages;
971                 start = cur_end + 1;
972         }
973         *page_started = 1;
974         return 0;
975 }
976
977 static noinline int csum_exist_in_range(struct btrfs_root *root,
978                                         u64 bytenr, u64 num_bytes)
979 {
980         int ret;
981         struct btrfs_ordered_sum *sums;
982         LIST_HEAD(list);
983
984         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
985                                        bytenr + num_bytes - 1, &list);
986         if (ret == 0 && list_empty(&list))
987                 return 0;
988
989         while (!list_empty(&list)) {
990                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
991                 list_del(&sums->list);
992                 kfree(sums);
993         }
994         return 1;
995 }
996
997 /*
998  * when nowcow writeback call back.  This checks for snapshots or COW copies
999  * of the extents that exist in the file, and COWs the file as required.
1000  *
1001  * If no cow copies or snapshots exist, we write directly to the existing
1002  * blocks on disk
1003  */
1004 static noinline int run_delalloc_nocow(struct inode *inode,
1005                                        struct page *locked_page,
1006                               u64 start, u64 end, int *page_started, int force,
1007                               unsigned long *nr_written)
1008 {
1009         struct btrfs_root *root = BTRFS_I(inode)->root;
1010         struct btrfs_trans_handle *trans;
1011         struct extent_buffer *leaf;
1012         struct btrfs_path *path;
1013         struct btrfs_file_extent_item *fi;
1014         struct btrfs_key found_key;
1015         u64 cow_start;
1016         u64 cur_offset;
1017         u64 extent_end;
1018         u64 extent_offset;
1019         u64 disk_bytenr;
1020         u64 num_bytes;
1021         int extent_type;
1022         int ret;
1023         int type;
1024         int nocow;
1025         int check_prev = 1;
1026
1027         path = btrfs_alloc_path();
1028         BUG_ON(!path);
1029         trans = btrfs_join_transaction(root, 1);
1030         BUG_ON(!trans);
1031
1032         cow_start = (u64)-1;
1033         cur_offset = start;
1034         while (1) {
1035                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1036                                                cur_offset, 0);
1037                 BUG_ON(ret < 0);
1038                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1039                         leaf = path->nodes[0];
1040                         btrfs_item_key_to_cpu(leaf, &found_key,
1041                                               path->slots[0] - 1);
1042                         if (found_key.objectid == inode->i_ino &&
1043                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1044                                 path->slots[0]--;
1045                 }
1046                 check_prev = 0;
1047 next_slot:
1048                 leaf = path->nodes[0];
1049                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1050                         ret = btrfs_next_leaf(root, path);
1051                         if (ret < 0)
1052                                 BUG_ON(1);
1053                         if (ret > 0)
1054                                 break;
1055                         leaf = path->nodes[0];
1056                 }
1057
1058                 nocow = 0;
1059                 disk_bytenr = 0;
1060                 num_bytes = 0;
1061                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1062
1063                 if (found_key.objectid > inode->i_ino ||
1064                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1065                     found_key.offset > end)
1066                         break;
1067
1068                 if (found_key.offset > cur_offset) {
1069                         extent_end = found_key.offset;
1070                         extent_type = 0;
1071                         goto out_check;
1072                 }
1073
1074                 fi = btrfs_item_ptr(leaf, path->slots[0],
1075                                     struct btrfs_file_extent_item);
1076                 extent_type = btrfs_file_extent_type(leaf, fi);
1077
1078                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1079                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1080                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1081                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1082                         extent_end = found_key.offset +
1083                                 btrfs_file_extent_num_bytes(leaf, fi);
1084                         if (extent_end <= start) {
1085                                 path->slots[0]++;
1086                                 goto next_slot;
1087                         }
1088                         if (disk_bytenr == 0)
1089                                 goto out_check;
1090                         if (btrfs_file_extent_compression(leaf, fi) ||
1091                             btrfs_file_extent_encryption(leaf, fi) ||
1092                             btrfs_file_extent_other_encoding(leaf, fi))
1093                                 goto out_check;
1094                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1095                                 goto out_check;
1096                         if (btrfs_extent_readonly(root, disk_bytenr))
1097                                 goto out_check;
1098                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1099                                                   found_key.offset -
1100                                                   extent_offset, disk_bytenr))
1101                                 goto out_check;
1102                         disk_bytenr += extent_offset;
1103                         disk_bytenr += cur_offset - found_key.offset;
1104                         num_bytes = min(end + 1, extent_end) - cur_offset;
1105                         /*
1106                          * force cow if csum exists in the range.
1107                          * this ensure that csum for a given extent are
1108                          * either valid or do not exist.
1109                          */
1110                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1111                                 goto out_check;
1112                         nocow = 1;
1113                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_inline_len(leaf, fi);
1116                         extent_end = ALIGN(extent_end, root->sectorsize);
1117                 } else {
1118                         BUG_ON(1);
1119                 }
1120 out_check:
1121                 if (extent_end <= start) {
1122                         path->slots[0]++;
1123                         goto next_slot;
1124                 }
1125                 if (!nocow) {
1126                         if (cow_start == (u64)-1)
1127                                 cow_start = cur_offset;
1128                         cur_offset = extent_end;
1129                         if (cur_offset > end)
1130                                 break;
1131                         path->slots[0]++;
1132                         goto next_slot;
1133                 }
1134
1135                 btrfs_release_path(root, path);
1136                 if (cow_start != (u64)-1) {
1137                         ret = cow_file_range(inode, locked_page, cow_start,
1138                                         found_key.offset - 1, page_started,
1139                                         nr_written, 1);
1140                         BUG_ON(ret);
1141                         cow_start = (u64)-1;
1142                 }
1143
1144                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1145                         struct extent_map *em;
1146                         struct extent_map_tree *em_tree;
1147                         em_tree = &BTRFS_I(inode)->extent_tree;
1148                         em = alloc_extent_map(GFP_NOFS);
1149                         em->start = cur_offset;
1150                         em->orig_start = em->start;
1151                         em->len = num_bytes;
1152                         em->block_len = num_bytes;
1153                         em->block_start = disk_bytenr;
1154                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1155                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1156                         while (1) {
1157                                 write_lock(&em_tree->lock);
1158                                 ret = add_extent_mapping(em_tree, em);
1159                                 write_unlock(&em_tree->lock);
1160                                 if (ret != -EEXIST) {
1161                                         free_extent_map(em);
1162                                         break;
1163                                 }
1164                                 btrfs_drop_extent_cache(inode, em->start,
1165                                                 em->start + em->len - 1, 0);
1166                         }
1167                         type = BTRFS_ORDERED_PREALLOC;
1168                 } else {
1169                         type = BTRFS_ORDERED_NOCOW;
1170                 }
1171
1172                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1173                                                num_bytes, num_bytes, type);
1174                 BUG_ON(ret);
1175
1176                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1177                                 cur_offset, cur_offset + num_bytes - 1,
1178                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1179                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1180                                 EXTENT_SET_PRIVATE2);
1181                 cur_offset = extent_end;
1182                 if (cur_offset > end)
1183                         break;
1184         }
1185         btrfs_release_path(root, path);
1186
1187         if (cur_offset <= end && cow_start == (u64)-1)
1188                 cow_start = cur_offset;
1189         if (cow_start != (u64)-1) {
1190                 ret = cow_file_range(inode, locked_page, cow_start, end,
1191                                      page_started, nr_written, 1);
1192                 BUG_ON(ret);
1193         }
1194
1195         ret = btrfs_end_transaction(trans, root);
1196         BUG_ON(ret);
1197         btrfs_free_path(path);
1198         return 0;
1199 }
1200
1201 /*
1202  * extent_io.c call back to do delayed allocation processing
1203  */
1204 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1205                               u64 start, u64 end, int *page_started,
1206                               unsigned long *nr_written)
1207 {
1208         int ret;
1209         struct btrfs_root *root = BTRFS_I(inode)->root;
1210
1211         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1212                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1213                                          page_started, 1, nr_written);
1214         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1215                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1216                                          page_started, 0, nr_written);
1217         else if (!btrfs_test_opt(root, COMPRESS) &&
1218                  !(BTRFS_I(inode)->force_compress))
1219                 ret = cow_file_range(inode, locked_page, start, end,
1220                                       page_started, nr_written, 1);
1221         else
1222                 ret = cow_file_range_async(inode, locked_page, start, end,
1223                                            page_started, nr_written);
1224         return ret;
1225 }
1226
1227 static int btrfs_split_extent_hook(struct inode *inode,
1228                                     struct extent_state *orig, u64 split)
1229 {
1230         struct btrfs_root *root = BTRFS_I(inode)->root;
1231         u64 size;
1232
1233         if (!(orig->state & EXTENT_DELALLOC))
1234                 return 0;
1235
1236         size = orig->end - orig->start + 1;
1237         if (size > root->fs_info->max_extent) {
1238                 u64 num_extents;
1239                 u64 new_size;
1240
1241                 new_size = orig->end - split + 1;
1242                 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1243                                         root->fs_info->max_extent);
1244
1245                 /*
1246                  * if we break a large extent up then leave oustanding_extents
1247                  * be, since we've already accounted for the large extent.
1248                  */
1249                 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1250                               root->fs_info->max_extent) < num_extents)
1251                         return 0;
1252         }
1253
1254         spin_lock(&BTRFS_I(inode)->accounting_lock);
1255         BTRFS_I(inode)->outstanding_extents++;
1256         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1257
1258         return 0;
1259 }
1260
1261 /*
1262  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1263  * extents so we can keep track of new extents that are just merged onto old
1264  * extents, such as when we are doing sequential writes, so we can properly
1265  * account for the metadata space we'll need.
1266  */
1267 static int btrfs_merge_extent_hook(struct inode *inode,
1268                                    struct extent_state *new,
1269                                    struct extent_state *other)
1270 {
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272         u64 new_size, old_size;
1273         u64 num_extents;
1274
1275         /* not delalloc, ignore it */
1276         if (!(other->state & EXTENT_DELALLOC))
1277                 return 0;
1278
1279         old_size = other->end - other->start + 1;
1280         if (new->start < other->start)
1281                 new_size = other->end - new->start + 1;
1282         else
1283                 new_size = new->end - other->start + 1;
1284
1285         /* we're not bigger than the max, unreserve the space and go */
1286         if (new_size <= root->fs_info->max_extent) {
1287                 spin_lock(&BTRFS_I(inode)->accounting_lock);
1288                 BTRFS_I(inode)->outstanding_extents--;
1289                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1290                 return 0;
1291         }
1292
1293         /*
1294          * If we grew by another max_extent, just return, we want to keep that
1295          * reserved amount.
1296          */
1297         num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1298                                 root->fs_info->max_extent);
1299         if (div64_u64(new_size + root->fs_info->max_extent - 1,
1300                       root->fs_info->max_extent) > num_extents)
1301                 return 0;
1302
1303         spin_lock(&BTRFS_I(inode)->accounting_lock);
1304         BTRFS_I(inode)->outstanding_extents--;
1305         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1306
1307         return 0;
1308 }
1309
1310 /*
1311  * extent_io.c set_bit_hook, used to track delayed allocation
1312  * bytes in this file, and to maintain the list of inodes that
1313  * have pending delalloc work to be done.
1314  */
1315 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1316                        unsigned long old, unsigned long bits)
1317 {
1318
1319         /*
1320          * set_bit and clear bit hooks normally require _irqsave/restore
1321          * but in this case, we are only testeing for the DELALLOC
1322          * bit, which is only set or cleared with irqs on
1323          */
1324         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1325                 struct btrfs_root *root = BTRFS_I(inode)->root;
1326
1327                 spin_lock(&BTRFS_I(inode)->accounting_lock);
1328                 BTRFS_I(inode)->outstanding_extents++;
1329                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1330                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1331                 spin_lock(&root->fs_info->delalloc_lock);
1332                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1333                 root->fs_info->delalloc_bytes += end - start + 1;
1334                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1335                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1336                                       &root->fs_info->delalloc_inodes);
1337                 }
1338                 spin_unlock(&root->fs_info->delalloc_lock);
1339         }
1340         return 0;
1341 }
1342
1343 /*
1344  * extent_io.c clear_bit_hook, see set_bit_hook for why
1345  */
1346 static int btrfs_clear_bit_hook(struct inode *inode,
1347                                 struct extent_state *state, unsigned long bits)
1348 {
1349         /*
1350          * set_bit and clear bit hooks normally require _irqsave/restore
1351          * but in this case, we are only testeing for the DELALLOC
1352          * bit, which is only set or cleared with irqs on
1353          */
1354         if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1355                 struct btrfs_root *root = BTRFS_I(inode)->root;
1356
1357                 if (bits & EXTENT_DO_ACCOUNTING) {
1358                         spin_lock(&BTRFS_I(inode)->accounting_lock);
1359                         BTRFS_I(inode)->outstanding_extents--;
1360                         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1361                         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1362                 }
1363
1364                 spin_lock(&root->fs_info->delalloc_lock);
1365                 if (state->end - state->start + 1 >
1366                     root->fs_info->delalloc_bytes) {
1367                         printk(KERN_INFO "btrfs warning: delalloc account "
1368                                "%llu %llu\n",
1369                                (unsigned long long)
1370                                state->end - state->start + 1,
1371                                (unsigned long long)
1372                                root->fs_info->delalloc_bytes);
1373                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1374                         root->fs_info->delalloc_bytes = 0;
1375                         BTRFS_I(inode)->delalloc_bytes = 0;
1376                 } else {
1377                         btrfs_delalloc_free_space(root, inode,
1378                                                   state->end -
1379                                                   state->start + 1);
1380                         root->fs_info->delalloc_bytes -= state->end -
1381                                 state->start + 1;
1382                         BTRFS_I(inode)->delalloc_bytes -= state->end -
1383                                 state->start + 1;
1384                 }
1385                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1386                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1387                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1388                 }
1389                 spin_unlock(&root->fs_info->delalloc_lock);
1390         }
1391         return 0;
1392 }
1393
1394 /*
1395  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1396  * we don't create bios that span stripes or chunks
1397  */
1398 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1399                          size_t size, struct bio *bio,
1400                          unsigned long bio_flags)
1401 {
1402         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1403         struct btrfs_mapping_tree *map_tree;
1404         u64 logical = (u64)bio->bi_sector << 9;
1405         u64 length = 0;
1406         u64 map_length;
1407         int ret;
1408
1409         if (bio_flags & EXTENT_BIO_COMPRESSED)
1410                 return 0;
1411
1412         length = bio->bi_size;
1413         map_tree = &root->fs_info->mapping_tree;
1414         map_length = length;
1415         ret = btrfs_map_block(map_tree, READ, logical,
1416                               &map_length, NULL, 0);
1417
1418         if (map_length < length + size)
1419                 return 1;
1420         return 0;
1421 }
1422
1423 /*
1424  * in order to insert checksums into the metadata in large chunks,
1425  * we wait until bio submission time.   All the pages in the bio are
1426  * checksummed and sums are attached onto the ordered extent record.
1427  *
1428  * At IO completion time the cums attached on the ordered extent record
1429  * are inserted into the btree
1430  */
1431 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1432                                     struct bio *bio, int mirror_num,
1433                                     unsigned long bio_flags)
1434 {
1435         struct btrfs_root *root = BTRFS_I(inode)->root;
1436         int ret = 0;
1437
1438         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1439         BUG_ON(ret);
1440         return 0;
1441 }
1442
1443 /*
1444  * in order to insert checksums into the metadata in large chunks,
1445  * we wait until bio submission time.   All the pages in the bio are
1446  * checksummed and sums are attached onto the ordered extent record.
1447  *
1448  * At IO completion time the cums attached on the ordered extent record
1449  * are inserted into the btree
1450  */
1451 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1456 }
1457
1458 /*
1459  * extent_io.c submission hook. This does the right thing for csum calculation
1460  * on write, or reading the csums from the tree before a read
1461  */
1462 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1463                           int mirror_num, unsigned long bio_flags)
1464 {
1465         struct btrfs_root *root = BTRFS_I(inode)->root;
1466         int ret = 0;
1467         int skip_sum;
1468
1469         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1470
1471         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1472         BUG_ON(ret);
1473
1474         if (!(rw & (1 << BIO_RW))) {
1475                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1476                         return btrfs_submit_compressed_read(inode, bio,
1477                                                     mirror_num, bio_flags);
1478                 } else if (!skip_sum)
1479                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1480                 goto mapit;
1481         } else if (!skip_sum) {
1482                 /* csum items have already been cloned */
1483                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1484                         goto mapit;
1485                 /* we're doing a write, do the async checksumming */
1486                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1487                                    inode, rw, bio, mirror_num,
1488                                    bio_flags, __btrfs_submit_bio_start,
1489                                    __btrfs_submit_bio_done);
1490         }
1491
1492 mapit:
1493         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1494 }
1495
1496 /*
1497  * given a list of ordered sums record them in the inode.  This happens
1498  * at IO completion time based on sums calculated at bio submission time.
1499  */
1500 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1501                              struct inode *inode, u64 file_offset,
1502                              struct list_head *list)
1503 {
1504         struct btrfs_ordered_sum *sum;
1505
1506         btrfs_set_trans_block_group(trans, inode);
1507
1508         list_for_each_entry(sum, list, list) {
1509                 btrfs_csum_file_blocks(trans,
1510                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1511         }
1512         return 0;
1513 }
1514
1515 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1516 {
1517         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1518                 WARN_ON(1);
1519         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1520                                    GFP_NOFS);
1521 }
1522
1523 /* see btrfs_writepage_start_hook for details on why this is required */
1524 struct btrfs_writepage_fixup {
1525         struct page *page;
1526         struct btrfs_work work;
1527 };
1528
1529 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1530 {
1531         struct btrfs_writepage_fixup *fixup;
1532         struct btrfs_ordered_extent *ordered;
1533         struct page *page;
1534         struct inode *inode;
1535         u64 page_start;
1536         u64 page_end;
1537
1538         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1539         page = fixup->page;
1540 again:
1541         lock_page(page);
1542         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1543                 ClearPageChecked(page);
1544                 goto out_page;
1545         }
1546
1547         inode = page->mapping->host;
1548         page_start = page_offset(page);
1549         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1550
1551         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, 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(&BTRFS_I(inode)->io_tree, page_start,
1560                               page_end, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         btrfs_set_extent_delalloc(inode, page_start, page_end);
1567         ClearPageChecked(page);
1568 out:
1569         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1570 out_page:
1571         unlock_page(page);
1572         page_cache_release(page);
1573 }
1574
1575 /*
1576  * There are a few paths in the higher layers of the kernel that directly
1577  * set the page dirty bit without asking the filesystem if it is a
1578  * good idea.  This causes problems because we want to make sure COW
1579  * properly happens and the data=ordered rules are followed.
1580  *
1581  * In our case any range that doesn't have the ORDERED bit set
1582  * hasn't been properly setup for IO.  We kick off an async process
1583  * to fix it up.  The async helper will wait for ordered extents, set
1584  * the delalloc bit and make it safe to write the page.
1585  */
1586 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1587 {
1588         struct inode *inode = page->mapping->host;
1589         struct btrfs_writepage_fixup *fixup;
1590         struct btrfs_root *root = BTRFS_I(inode)->root;
1591
1592         /* this page is properly in the ordered list */
1593         if (TestClearPagePrivate2(page))
1594                 return 0;
1595
1596         if (PageChecked(page))
1597                 return -EAGAIN;
1598
1599         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1600         if (!fixup)
1601                 return -EAGAIN;
1602
1603         SetPageChecked(page);
1604         page_cache_get(page);
1605         fixup->work.func = btrfs_writepage_fixup_worker;
1606         fixup->page = page;
1607         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1608         return -EAGAIN;
1609 }
1610
1611 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1612                                        struct inode *inode, u64 file_pos,
1613                                        u64 disk_bytenr, u64 disk_num_bytes,
1614                                        u64 num_bytes, u64 ram_bytes,
1615                                        u8 compression, u8 encryption,
1616                                        u16 other_encoding, int extent_type)
1617 {
1618         struct btrfs_root *root = BTRFS_I(inode)->root;
1619         struct btrfs_file_extent_item *fi;
1620         struct btrfs_path *path;
1621         struct extent_buffer *leaf;
1622         struct btrfs_key ins;
1623         u64 hint;
1624         int ret;
1625
1626         path = btrfs_alloc_path();
1627         BUG_ON(!path);
1628
1629         path->leave_spinning = 1;
1630
1631         /*
1632          * we may be replacing one extent in the tree with another.
1633          * The new extent is pinned in the extent map, and we don't want
1634          * to drop it from the cache until it is completely in the btree.
1635          *
1636          * So, tell btrfs_drop_extents to leave this extent in the cache.
1637          * the caller is expected to unpin it and allow it to be merged
1638          * with the others.
1639          */
1640         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1641                                  &hint, 0);
1642         BUG_ON(ret);
1643
1644         ins.objectid = inode->i_ino;
1645         ins.offset = file_pos;
1646         ins.type = BTRFS_EXTENT_DATA_KEY;
1647         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1648         BUG_ON(ret);
1649         leaf = path->nodes[0];
1650         fi = btrfs_item_ptr(leaf, path->slots[0],
1651                             struct btrfs_file_extent_item);
1652         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1653         btrfs_set_file_extent_type(leaf, fi, extent_type);
1654         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1655         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1656         btrfs_set_file_extent_offset(leaf, fi, 0);
1657         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1658         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1659         btrfs_set_file_extent_compression(leaf, fi, compression);
1660         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1661         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1662
1663         btrfs_unlock_up_safe(path, 1);
1664         btrfs_set_lock_blocking(leaf);
1665
1666         btrfs_mark_buffer_dirty(leaf);
1667
1668         inode_add_bytes(inode, num_bytes);
1669
1670         ins.objectid = disk_bytenr;
1671         ins.offset = disk_num_bytes;
1672         ins.type = BTRFS_EXTENT_ITEM_KEY;
1673         ret = btrfs_alloc_reserved_file_extent(trans, root,
1674                                         root->root_key.objectid,
1675                                         inode->i_ino, file_pos, &ins);
1676         BUG_ON(ret);
1677         btrfs_free_path(path);
1678
1679         return 0;
1680 }
1681
1682 /*
1683  * helper function for btrfs_finish_ordered_io, this
1684  * just reads in some of the csum leaves to prime them into ram
1685  * before we start the transaction.  It limits the amount of btree
1686  * reads required while inside the transaction.
1687  */
1688 /* as ordered data IO finishes, this gets called so we can finish
1689  * an ordered extent if the range of bytes in the file it covers are
1690  * fully written.
1691  */
1692 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1693 {
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         struct btrfs_trans_handle *trans;
1696         struct btrfs_ordered_extent *ordered_extent = NULL;
1697         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1698         int compressed = 0;
1699         int ret;
1700
1701         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1702         if (!ret)
1703                 return 0;
1704
1705         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1706         BUG_ON(!ordered_extent);
1707
1708         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1709                 BUG_ON(!list_empty(&ordered_extent->list));
1710                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1711                 if (!ret) {
1712                         trans = btrfs_join_transaction(root, 1);
1713                         ret = btrfs_update_inode(trans, root, inode);
1714                         BUG_ON(ret);
1715                         btrfs_end_transaction(trans, root);
1716                 }
1717                 goto out;
1718         }
1719
1720         lock_extent(io_tree, ordered_extent->file_offset,
1721                     ordered_extent->file_offset + ordered_extent->len - 1,
1722                     GFP_NOFS);
1723
1724         trans = btrfs_join_transaction(root, 1);
1725
1726         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1727                 compressed = 1;
1728         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1729                 BUG_ON(compressed);
1730                 ret = btrfs_mark_extent_written(trans, inode,
1731                                                 ordered_extent->file_offset,
1732                                                 ordered_extent->file_offset +
1733                                                 ordered_extent->len);
1734                 BUG_ON(ret);
1735         } else {
1736                 ret = insert_reserved_file_extent(trans, inode,
1737                                                 ordered_extent->file_offset,
1738                                                 ordered_extent->start,
1739                                                 ordered_extent->disk_len,
1740                                                 ordered_extent->len,
1741                                                 ordered_extent->len,
1742                                                 compressed, 0, 0,
1743                                                 BTRFS_FILE_EXTENT_REG);
1744                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1745                                    ordered_extent->file_offset,
1746                                    ordered_extent->len);
1747                 BUG_ON(ret);
1748         }
1749         unlock_extent(io_tree, ordered_extent->file_offset,
1750                     ordered_extent->file_offset + ordered_extent->len - 1,
1751                     GFP_NOFS);
1752         add_pending_csums(trans, inode, ordered_extent->file_offset,
1753                           &ordered_extent->list);
1754
1755         /* this also removes the ordered extent from the tree */
1756         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1757         ret = btrfs_update_inode(trans, root, inode);
1758         BUG_ON(ret);
1759         btrfs_end_transaction(trans, root);
1760 out:
1761         /* once for us */
1762         btrfs_put_ordered_extent(ordered_extent);
1763         /* once for the tree */
1764         btrfs_put_ordered_extent(ordered_extent);
1765
1766         return 0;
1767 }
1768
1769 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1770                                 struct extent_state *state, int uptodate)
1771 {
1772         ClearPagePrivate2(page);
1773         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1774 }
1775
1776 /*
1777  * When IO fails, either with EIO or csum verification fails, we
1778  * try other mirrors that might have a good copy of the data.  This
1779  * io_failure_record is used to record state as we go through all the
1780  * mirrors.  If another mirror has good data, the page is set up to date
1781  * and things continue.  If a good mirror can't be found, the original
1782  * bio end_io callback is called to indicate things have failed.
1783  */
1784 struct io_failure_record {
1785         struct page *page;
1786         u64 start;
1787         u64 len;
1788         u64 logical;
1789         unsigned long bio_flags;
1790         int last_mirror;
1791 };
1792
1793 static int btrfs_io_failed_hook(struct bio *failed_bio,
1794                          struct page *page, u64 start, u64 end,
1795                          struct extent_state *state)
1796 {
1797         struct io_failure_record *failrec = NULL;
1798         u64 private;
1799         struct extent_map *em;
1800         struct inode *inode = page->mapping->host;
1801         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1802         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1803         struct bio *bio;
1804         int num_copies;
1805         int ret;
1806         int rw;
1807         u64 logical;
1808
1809         ret = get_state_private(failure_tree, start, &private);
1810         if (ret) {
1811                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1812                 if (!failrec)
1813                         return -ENOMEM;
1814                 failrec->start = start;
1815                 failrec->len = end - start + 1;
1816                 failrec->last_mirror = 0;
1817                 failrec->bio_flags = 0;
1818
1819                 read_lock(&em_tree->lock);
1820                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1821                 if (em->start > start || em->start + em->len < start) {
1822                         free_extent_map(em);
1823                         em = NULL;
1824                 }
1825                 read_unlock(&em_tree->lock);
1826
1827                 if (!em || IS_ERR(em)) {
1828                         kfree(failrec);
1829                         return -EIO;
1830                 }
1831                 logical = start - em->start;
1832                 logical = em->block_start + logical;
1833                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1834                         logical = em->block_start;
1835                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1836                 }
1837                 failrec->logical = logical;
1838                 free_extent_map(em);
1839                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1840                                 EXTENT_DIRTY, GFP_NOFS);
1841                 set_state_private(failure_tree, start,
1842                                  (u64)(unsigned long)failrec);
1843         } else {
1844                 failrec = (struct io_failure_record *)(unsigned long)private;
1845         }
1846         num_copies = btrfs_num_copies(
1847                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1848                               failrec->logical, failrec->len);
1849         failrec->last_mirror++;
1850         if (!state) {
1851                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1852                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1853                                                     failrec->start,
1854                                                     EXTENT_LOCKED);
1855                 if (state && state->start != failrec->start)
1856                         state = NULL;
1857                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1858         }
1859         if (!state || failrec->last_mirror > num_copies) {
1860                 set_state_private(failure_tree, failrec->start, 0);
1861                 clear_extent_bits(failure_tree, failrec->start,
1862                                   failrec->start + failrec->len - 1,
1863                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1864                 kfree(failrec);
1865                 return -EIO;
1866         }
1867         bio = bio_alloc(GFP_NOFS, 1);
1868         bio->bi_private = state;
1869         bio->bi_end_io = failed_bio->bi_end_io;
1870         bio->bi_sector = failrec->logical >> 9;
1871         bio->bi_bdev = failed_bio->bi_bdev;
1872         bio->bi_size = 0;
1873
1874         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1875         if (failed_bio->bi_rw & (1 << BIO_RW))
1876                 rw = WRITE;
1877         else
1878                 rw = READ;
1879
1880         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1881                                                       failrec->last_mirror,
1882                                                       failrec->bio_flags);
1883         return 0;
1884 }
1885
1886 /*
1887  * each time an IO finishes, we do a fast check in the IO failure tree
1888  * to see if we need to process or clean up an io_failure_record
1889  */
1890 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1891 {
1892         u64 private;
1893         u64 private_failure;
1894         struct io_failure_record *failure;
1895         int ret;
1896
1897         private = 0;
1898         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1899                              (u64)-1, 1, EXTENT_DIRTY)) {
1900                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1901                                         start, &private_failure);
1902                 if (ret == 0) {
1903                         failure = (struct io_failure_record *)(unsigned long)
1904                                    private_failure;
1905                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1906                                           failure->start, 0);
1907                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1908                                           failure->start,
1909                                           failure->start + failure->len - 1,
1910                                           EXTENT_DIRTY | EXTENT_LOCKED,
1911                                           GFP_NOFS);
1912                         kfree(failure);
1913                 }
1914         }
1915         return 0;
1916 }
1917
1918 /*
1919  * when reads are done, we need to check csums to verify the data is correct
1920  * if there's a match, we allow the bio to finish.  If not, we go through
1921  * the io_failure_record routines to find good copies
1922  */
1923 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1924                                struct extent_state *state)
1925 {
1926         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1927         struct inode *inode = page->mapping->host;
1928         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1929         char *kaddr;
1930         u64 private = ~(u32)0;
1931         int ret;
1932         struct btrfs_root *root = BTRFS_I(inode)->root;
1933         u32 csum = ~(u32)0;
1934
1935         if (PageChecked(page)) {
1936                 ClearPageChecked(page);
1937                 goto good;
1938         }
1939
1940         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1941                 return 0;
1942
1943         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1944             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1945                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1946                                   GFP_NOFS);
1947                 return 0;
1948         }
1949
1950         if (state && state->start == start) {
1951                 private = state->private;
1952                 ret = 0;
1953         } else {
1954                 ret = get_state_private(io_tree, start, &private);
1955         }
1956         kaddr = kmap_atomic(page, KM_USER0);
1957         if (ret)
1958                 goto zeroit;
1959
1960         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1961         btrfs_csum_final(csum, (char *)&csum);
1962         if (csum != private)
1963                 goto zeroit;
1964
1965         kunmap_atomic(kaddr, KM_USER0);
1966 good:
1967         /* if the io failure tree for this inode is non-empty,
1968          * check to see if we've recovered from a failed IO
1969          */
1970         btrfs_clean_io_failures(inode, start);
1971         return 0;
1972
1973 zeroit:
1974         if (printk_ratelimit()) {
1975                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1976                        "private %llu\n", page->mapping->host->i_ino,
1977                        (unsigned long long)start, csum,
1978                        (unsigned long long)private);
1979         }
1980         memset(kaddr + offset, 1, end - start + 1);
1981         flush_dcache_page(page);
1982         kunmap_atomic(kaddr, KM_USER0);
1983         if (private == 0)
1984                 return 0;
1985         return -EIO;
1986 }
1987
1988 struct delayed_iput {
1989         struct list_head list;
1990         struct inode *inode;
1991 };
1992
1993 void btrfs_add_delayed_iput(struct inode *inode)
1994 {
1995         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1996         struct delayed_iput *delayed;
1997
1998         if (atomic_add_unless(&inode->i_count, -1, 1))
1999                 return;
2000
2001         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2002         delayed->inode = inode;
2003
2004         spin_lock(&fs_info->delayed_iput_lock);
2005         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2006         spin_unlock(&fs_info->delayed_iput_lock);
2007 }
2008
2009 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2010 {
2011         LIST_HEAD(list);
2012         struct btrfs_fs_info *fs_info = root->fs_info;
2013         struct delayed_iput *delayed;
2014         int empty;
2015
2016         spin_lock(&fs_info->delayed_iput_lock);
2017         empty = list_empty(&fs_info->delayed_iputs);
2018         spin_unlock(&fs_info->delayed_iput_lock);
2019         if (empty)
2020                 return;
2021
2022         down_read(&root->fs_info->cleanup_work_sem);
2023         spin_lock(&fs_info->delayed_iput_lock);
2024         list_splice_init(&fs_info->delayed_iputs, &list);
2025         spin_unlock(&fs_info->delayed_iput_lock);
2026
2027         while (!list_empty(&list)) {
2028                 delayed = list_entry(list.next, struct delayed_iput, list);
2029                 list_del(&delayed->list);
2030                 iput(delayed->inode);
2031                 kfree(delayed);
2032         }
2033         up_read(&root->fs_info->cleanup_work_sem);
2034 }
2035
2036 /*
2037  * This creates an orphan entry for the given inode in case something goes
2038  * wrong in the middle of an unlink/truncate.
2039  */
2040 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2041 {
2042         struct btrfs_root *root = BTRFS_I(inode)->root;
2043         int ret = 0;
2044
2045         spin_lock(&root->list_lock);
2046
2047         /* already on the orphan list, we're good */
2048         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2049                 spin_unlock(&root->list_lock);
2050                 return 0;
2051         }
2052
2053         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2054
2055         spin_unlock(&root->list_lock);
2056
2057         /*
2058          * insert an orphan item to track this unlinked/truncated file
2059          */
2060         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2061
2062         return ret;
2063 }
2064
2065 /*
2066  * We have done the truncate/delete so we can go ahead and remove the orphan
2067  * item for this particular inode.
2068  */
2069 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2070 {
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         int ret = 0;
2073
2074         spin_lock(&root->list_lock);
2075
2076         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2077                 spin_unlock(&root->list_lock);
2078                 return 0;
2079         }
2080
2081         list_del_init(&BTRFS_I(inode)->i_orphan);
2082         if (!trans) {
2083                 spin_unlock(&root->list_lock);
2084                 return 0;
2085         }
2086
2087         spin_unlock(&root->list_lock);
2088
2089         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2090
2091         return ret;
2092 }
2093
2094 /*
2095  * this cleans up any orphans that may be left on the list from the last use
2096  * of this root.
2097  */
2098 void btrfs_orphan_cleanup(struct btrfs_root *root)
2099 {
2100         struct btrfs_path *path;
2101         struct extent_buffer *leaf;
2102         struct btrfs_item *item;
2103         struct btrfs_key key, found_key;
2104         struct btrfs_trans_handle *trans;
2105         struct inode *inode;
2106         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2107
2108         if (!xchg(&root->clean_orphans, 0))
2109                 return;
2110
2111         path = btrfs_alloc_path();
2112         BUG_ON(!path);
2113         path->reada = -1;
2114
2115         key.objectid = BTRFS_ORPHAN_OBJECTID;
2116         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2117         key.offset = (u64)-1;
2118
2119         while (1) {
2120                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2121                 if (ret < 0) {
2122                         printk(KERN_ERR "Error searching slot for orphan: %d"
2123                                "\n", ret);
2124                         break;
2125                 }
2126
2127                 /*
2128                  * if ret == 0 means we found what we were searching for, which
2129                  * is weird, but possible, so only screw with path if we didnt
2130                  * find the key and see if we have stuff that matches
2131                  */
2132                 if (ret > 0) {
2133                         if (path->slots[0] == 0)
2134                                 break;
2135                         path->slots[0]--;
2136                 }
2137
2138                 /* pull out the item */
2139                 leaf = path->nodes[0];
2140                 item = btrfs_item_nr(leaf, path->slots[0]);
2141                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2142
2143                 /* make sure the item matches what we want */
2144                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2145                         break;
2146                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2147                         break;
2148
2149                 /* release the path since we're done with it */
2150                 btrfs_release_path(root, path);
2151
2152                 /*
2153                  * this is where we are basically btrfs_lookup, without the
2154                  * crossing root thing.  we store the inode number in the
2155                  * offset of the orphan item.
2156                  */
2157                 found_key.objectid = found_key.offset;
2158                 found_key.type = BTRFS_INODE_ITEM_KEY;
2159                 found_key.offset = 0;
2160                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2161                 if (IS_ERR(inode))
2162                         break;
2163
2164                 /*
2165                  * add this inode to the orphan list so btrfs_orphan_del does
2166                  * the proper thing when we hit it
2167                  */
2168                 spin_lock(&root->list_lock);
2169                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2170                 spin_unlock(&root->list_lock);
2171
2172                 /*
2173                  * if this is a bad inode, means we actually succeeded in
2174                  * removing the inode, but not the orphan record, which means
2175                  * we need to manually delete the orphan since iput will just
2176                  * do a destroy_inode
2177                  */
2178                 if (is_bad_inode(inode)) {
2179                         trans = btrfs_start_transaction(root, 1);
2180                         btrfs_orphan_del(trans, inode);
2181                         btrfs_end_transaction(trans, root);
2182                         iput(inode);
2183                         continue;
2184                 }
2185
2186                 /* if we have links, this was a truncate, lets do that */
2187                 if (inode->i_nlink) {
2188                         nr_truncate++;
2189                         btrfs_truncate(inode);
2190                 } else {
2191                         nr_unlink++;
2192                 }
2193
2194                 /* this will do delete_inode and everything for us */
2195                 iput(inode);
2196         }
2197
2198         if (nr_unlink)
2199                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2200         if (nr_truncate)
2201                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2202
2203         btrfs_free_path(path);
2204 }
2205
2206 /*
2207  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2208  * don't find any xattrs, we know there can't be any acls.
2209  *
2210  * slot is the slot the inode is in, objectid is the objectid of the inode
2211  */
2212 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2213                                           int slot, u64 objectid)
2214 {
2215         u32 nritems = btrfs_header_nritems(leaf);
2216         struct btrfs_key found_key;
2217         int scanned = 0;
2218
2219         slot++;
2220         while (slot < nritems) {
2221                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2222
2223                 /* we found a different objectid, there must not be acls */
2224                 if (found_key.objectid != objectid)
2225                         return 0;
2226
2227                 /* we found an xattr, assume we've got an acl */
2228                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2229                         return 1;
2230
2231                 /*
2232                  * we found a key greater than an xattr key, there can't
2233                  * be any acls later on
2234                  */
2235                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2236                         return 0;
2237
2238                 slot++;
2239                 scanned++;
2240
2241                 /*
2242                  * it goes inode, inode backrefs, xattrs, extents,
2243                  * so if there are a ton of hard links to an inode there can
2244                  * be a lot of backrefs.  Don't waste time searching too hard,
2245                  * this is just an optimization
2246                  */
2247                 if (scanned >= 8)
2248                         break;
2249         }
2250         /* we hit the end of the leaf before we found an xattr or
2251          * something larger than an xattr.  We have to assume the inode
2252          * has acls
2253          */
2254         return 1;
2255 }
2256
2257 /*
2258  * read an inode from the btree into the in-memory inode
2259  */
2260 static void btrfs_read_locked_inode(struct inode *inode)
2261 {
2262         struct btrfs_path *path;
2263         struct extent_buffer *leaf;
2264         struct btrfs_inode_item *inode_item;
2265         struct btrfs_timespec *tspec;
2266         struct btrfs_root *root = BTRFS_I(inode)->root;
2267         struct btrfs_key location;
2268         int maybe_acls;
2269         u64 alloc_group_block;
2270         u32 rdev;
2271         int ret;
2272
2273         path = btrfs_alloc_path();
2274         BUG_ON(!path);
2275         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2276
2277         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2278         if (ret)
2279                 goto make_bad;
2280
2281         leaf = path->nodes[0];
2282         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2283                                     struct btrfs_inode_item);
2284
2285         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2286         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2287         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2288         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2289         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2290
2291         tspec = btrfs_inode_atime(inode_item);
2292         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2293         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2294
2295         tspec = btrfs_inode_mtime(inode_item);
2296         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2297         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2298
2299         tspec = btrfs_inode_ctime(inode_item);
2300         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2301         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2302
2303         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2304         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2305         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2306         inode->i_generation = BTRFS_I(inode)->generation;
2307         inode->i_rdev = 0;
2308         rdev = btrfs_inode_rdev(leaf, inode_item);
2309
2310         BTRFS_I(inode)->index_cnt = (u64)-1;
2311         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2312
2313         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2314
2315         /*
2316          * try to precache a NULL acl entry for files that don't have
2317          * any xattrs or acls
2318          */
2319         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2320         if (!maybe_acls)
2321                 cache_no_acl(inode);
2322
2323         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2324                                                 alloc_group_block, 0);
2325         btrfs_free_path(path);
2326         inode_item = NULL;
2327
2328         switch (inode->i_mode & S_IFMT) {
2329         case S_IFREG:
2330                 inode->i_mapping->a_ops = &btrfs_aops;
2331                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2332                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2333                 inode->i_fop = &btrfs_file_operations;
2334                 inode->i_op = &btrfs_file_inode_operations;
2335                 break;
2336         case S_IFDIR:
2337                 inode->i_fop = &btrfs_dir_file_operations;
2338                 if (root == root->fs_info->tree_root)
2339                         inode->i_op = &btrfs_dir_ro_inode_operations;
2340                 else
2341                         inode->i_op = &btrfs_dir_inode_operations;
2342                 break;
2343         case S_IFLNK:
2344                 inode->i_op = &btrfs_symlink_inode_operations;
2345                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2346                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2347                 break;
2348         default:
2349                 inode->i_op = &btrfs_special_inode_operations;
2350                 init_special_inode(inode, inode->i_mode, rdev);
2351                 break;
2352         }
2353
2354         btrfs_update_iflags(inode);
2355         return;
2356
2357 make_bad:
2358         btrfs_free_path(path);
2359         make_bad_inode(inode);
2360 }
2361
2362 /*
2363  * given a leaf and an inode, copy the inode fields into the leaf
2364  */
2365 static void fill_inode_item(struct btrfs_trans_handle *trans,
2366                             struct extent_buffer *leaf,
2367                             struct btrfs_inode_item *item,
2368                             struct inode *inode)
2369 {
2370         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2371         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2372         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2373         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2374         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2375
2376         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2377                                inode->i_atime.tv_sec);
2378         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2379                                 inode->i_atime.tv_nsec);
2380
2381         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2382                                inode->i_mtime.tv_sec);
2383         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2384                                 inode->i_mtime.tv_nsec);
2385
2386         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2387                                inode->i_ctime.tv_sec);
2388         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2389                                 inode->i_ctime.tv_nsec);
2390
2391         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2392         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2393         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2394         btrfs_set_inode_transid(leaf, item, trans->transid);
2395         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2396         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2397         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2398 }
2399
2400 /*
2401  * copy everything in the in-memory inode into the btree.
2402  */
2403 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2404                                 struct btrfs_root *root, struct inode *inode)
2405 {
2406         struct btrfs_inode_item *inode_item;
2407         struct btrfs_path *path;
2408         struct extent_buffer *leaf;
2409         int ret;
2410
2411         path = btrfs_alloc_path();
2412         BUG_ON(!path);
2413         path->leave_spinning = 1;
2414         ret = btrfs_lookup_inode(trans, root, path,
2415                                  &BTRFS_I(inode)->location, 1);
2416         if (ret) {
2417                 if (ret > 0)
2418                         ret = -ENOENT;
2419                 goto failed;
2420         }
2421
2422         btrfs_unlock_up_safe(path, 1);
2423         leaf = path->nodes[0];
2424         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2425                                   struct btrfs_inode_item);
2426
2427         fill_inode_item(trans, leaf, inode_item, inode);
2428         btrfs_mark_buffer_dirty(leaf);
2429         btrfs_set_inode_last_trans(trans, inode);
2430         ret = 0;
2431 failed:
2432         btrfs_free_path(path);
2433         return ret;
2434 }
2435
2436
2437 /*
2438  * unlink helper that gets used here in inode.c and in the tree logging
2439  * recovery code.  It remove a link in a directory with a given name, and
2440  * also drops the back refs in the inode to the directory
2441  */
2442 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2443                        struct btrfs_root *root,
2444                        struct inode *dir, struct inode *inode,
2445                        const char *name, int name_len)
2446 {
2447         struct btrfs_path *path;
2448         int ret = 0;
2449         struct extent_buffer *leaf;
2450         struct btrfs_dir_item *di;
2451         struct btrfs_key key;
2452         u64 index;
2453
2454         path = btrfs_alloc_path();
2455         if (!path) {
2456                 ret = -ENOMEM;
2457                 goto err;
2458         }
2459
2460         path->leave_spinning = 1;
2461         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2462                                     name, name_len, -1);
2463         if (IS_ERR(di)) {
2464                 ret = PTR_ERR(di);
2465                 goto err;
2466         }
2467         if (!di) {
2468                 ret = -ENOENT;
2469                 goto err;
2470         }
2471         leaf = path->nodes[0];
2472         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2473         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2474         if (ret)
2475                 goto err;
2476         btrfs_release_path(root, path);
2477
2478         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2479                                   inode->i_ino,
2480                                   dir->i_ino, &index);
2481         if (ret) {
2482                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2483                        "inode %lu parent %lu\n", name_len, name,
2484                        inode->i_ino, dir->i_ino);
2485                 goto err;
2486         }
2487
2488         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2489                                          index, name, name_len, -1);
2490         if (IS_ERR(di)) {
2491                 ret = PTR_ERR(di);
2492                 goto err;
2493         }
2494         if (!di) {
2495                 ret = -ENOENT;
2496                 goto err;
2497         }
2498         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2499         btrfs_release_path(root, path);
2500
2501         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2502                                          inode, dir->i_ino);
2503         BUG_ON(ret != 0 && ret != -ENOENT);
2504
2505         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2506                                            dir, index);
2507         BUG_ON(ret);
2508 err:
2509         btrfs_free_path(path);
2510         if (ret)
2511                 goto out;
2512
2513         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2514         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2515         btrfs_update_inode(trans, root, dir);
2516         btrfs_drop_nlink(inode);
2517         ret = btrfs_update_inode(trans, root, inode);
2518 out:
2519         return ret;
2520 }
2521
2522 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2523 {
2524         struct btrfs_root *root;
2525         struct btrfs_trans_handle *trans;
2526         struct inode *inode = dentry->d_inode;
2527         int ret;
2528         unsigned long nr = 0;
2529
2530         root = BTRFS_I(dir)->root;
2531
2532         /*
2533          * 5 items for unlink inode
2534          * 1 for orphan
2535          */
2536         ret = btrfs_reserve_metadata_space(root, 6);
2537         if (ret)
2538                 return ret;
2539
2540         trans = btrfs_start_transaction(root, 1);
2541         if (IS_ERR(trans)) {
2542                 btrfs_unreserve_metadata_space(root, 6);
2543                 return PTR_ERR(trans);
2544         }
2545
2546         btrfs_set_trans_block_group(trans, dir);
2547
2548         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2549
2550         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2551                                  dentry->d_name.name, dentry->d_name.len);
2552
2553         if (inode->i_nlink == 0)
2554                 ret = btrfs_orphan_add(trans, inode);
2555
2556         nr = trans->blocks_used;
2557
2558         btrfs_end_transaction_throttle(trans, root);
2559         btrfs_unreserve_metadata_space(root, 6);
2560         btrfs_btree_balance_dirty(root, nr);
2561         return ret;
2562 }
2563
2564 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2565                         struct btrfs_root *root,
2566                         struct inode *dir, u64 objectid,
2567                         const char *name, int name_len)
2568 {
2569         struct btrfs_path *path;
2570         struct extent_buffer *leaf;
2571         struct btrfs_dir_item *di;
2572         struct btrfs_key key;
2573         u64 index;
2574         int ret;
2575
2576         path = btrfs_alloc_path();
2577         if (!path)
2578                 return -ENOMEM;
2579
2580         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2581                                    name, name_len, -1);
2582         BUG_ON(!di || IS_ERR(di));
2583
2584         leaf = path->nodes[0];
2585         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2586         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2587         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2588         BUG_ON(ret);
2589         btrfs_release_path(root, path);
2590
2591         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2592                                  objectid, root->root_key.objectid,
2593                                  dir->i_ino, &index, name, name_len);
2594         if (ret < 0) {
2595                 BUG_ON(ret != -ENOENT);
2596                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2597                                                  name, name_len);
2598                 BUG_ON(!di || IS_ERR(di));
2599
2600                 leaf = path->nodes[0];
2601                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2602                 btrfs_release_path(root, path);
2603                 index = key.offset;
2604         }
2605
2606         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2607                                          index, name, name_len, -1);
2608         BUG_ON(!di || IS_ERR(di));
2609
2610         leaf = path->nodes[0];
2611         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2612         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2613         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2614         BUG_ON(ret);
2615         btrfs_release_path(root, path);
2616
2617         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2618         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2619         ret = btrfs_update_inode(trans, root, dir);
2620         BUG_ON(ret);
2621         dir->i_sb->s_dirt = 1;
2622
2623         btrfs_free_path(path);
2624         return 0;
2625 }
2626
2627 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2628 {
2629         struct inode *inode = dentry->d_inode;
2630         int err = 0;
2631         int ret;
2632         struct btrfs_root *root = BTRFS_I(dir)->root;
2633         struct btrfs_trans_handle *trans;
2634         unsigned long nr = 0;
2635
2636         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2637             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2638                 return -ENOTEMPTY;
2639
2640         ret = btrfs_reserve_metadata_space(root, 5);
2641         if (ret)
2642                 return ret;
2643
2644         trans = btrfs_start_transaction(root, 1);
2645         if (IS_ERR(trans)) {
2646                 btrfs_unreserve_metadata_space(root, 5);
2647                 return PTR_ERR(trans);
2648         }
2649
2650         btrfs_set_trans_block_group(trans, dir);
2651
2652         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2653                 err = btrfs_unlink_subvol(trans, root, dir,
2654                                           BTRFS_I(inode)->location.objectid,
2655                                           dentry->d_name.name,
2656                                           dentry->d_name.len);
2657                 goto out;
2658         }
2659
2660         err = btrfs_orphan_add(trans, inode);
2661         if (err)
2662                 goto out;
2663
2664         /* now the directory is empty */
2665         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2666                                  dentry->d_name.name, dentry->d_name.len);
2667         if (!err)
2668                 btrfs_i_size_write(inode, 0);
2669 out:
2670         nr = trans->blocks_used;
2671         ret = btrfs_end_transaction_throttle(trans, root);
2672         btrfs_unreserve_metadata_space(root, 5);
2673         btrfs_btree_balance_dirty(root, nr);
2674
2675         if (ret && !err)
2676                 err = ret;
2677         return err;
2678 }
2679
2680 #if 0
2681 /*
2682  * when truncating bytes in a file, it is possible to avoid reading
2683  * the leaves that contain only checksum items.  This can be the
2684  * majority of the IO required to delete a large file, but it must
2685  * be done carefully.
2686  *
2687  * The keys in the level just above the leaves are checked to make sure
2688  * the lowest key in a given leaf is a csum key, and starts at an offset
2689  * after the new  size.
2690  *
2691  * Then the key for the next leaf is checked to make sure it also has
2692  * a checksum item for the same file.  If it does, we know our target leaf
2693  * contains only checksum items, and it can be safely freed without reading
2694  * it.
2695  *
2696  * This is just an optimization targeted at large files.  It may do
2697  * nothing.  It will return 0 unless things went badly.
2698  */
2699 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2700                                      struct btrfs_root *root,
2701                                      struct btrfs_path *path,
2702                                      struct inode *inode, u64 new_size)
2703 {
2704         struct btrfs_key key;
2705         int ret;
2706         int nritems;
2707         struct btrfs_key found_key;
2708         struct btrfs_key other_key;
2709         struct btrfs_leaf_ref *ref;
2710         u64 leaf_gen;
2711         u64 leaf_start;
2712
2713         path->lowest_level = 1;
2714         key.objectid = inode->i_ino;
2715         key.type = BTRFS_CSUM_ITEM_KEY;
2716         key.offset = new_size;
2717 again:
2718         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2719         if (ret < 0)
2720                 goto out;
2721
2722         if (path->nodes[1] == NULL) {
2723                 ret = 0;
2724                 goto out;
2725         }
2726         ret = 0;
2727         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2728         nritems = btrfs_header_nritems(path->nodes[1]);
2729
2730         if (!nritems)
2731                 goto out;
2732
2733         if (path->slots[1] >= nritems)
2734                 goto next_node;
2735
2736         /* did we find a key greater than anything we want to delete? */
2737         if (found_key.objectid > inode->i_ino ||
2738            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2739                 goto out;
2740
2741         /* we check the next key in the node to make sure the leave contains
2742          * only checksum items.  This comparison doesn't work if our
2743          * leaf is the last one in the node
2744          */
2745         if (path->slots[1] + 1 >= nritems) {
2746 next_node:
2747                 /* search forward from the last key in the node, this
2748                  * will bring us into the next node in the tree
2749                  */
2750                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2751
2752                 /* unlikely, but we inc below, so check to be safe */
2753                 if (found_key.offset == (u64)-1)
2754                         goto out;
2755
2756                 /* search_forward needs a path with locks held, do the
2757                  * search again for the original key.  It is possible
2758                  * this will race with a balance and return a path that
2759                  * we could modify, but this drop is just an optimization
2760                  * and is allowed to miss some leaves.
2761                  */
2762                 btrfs_release_path(root, path);
2763                 found_key.offset++;
2764
2765                 /* setup a max key for search_forward */
2766                 other_key.offset = (u64)-1;
2767                 other_key.type = key.type;
2768                 other_key.objectid = key.objectid;
2769
2770                 path->keep_locks = 1;
2771                 ret = btrfs_search_forward(root, &found_key, &other_key,
2772                                            path, 0, 0);
2773                 path->keep_locks = 0;
2774                 if (ret || found_key.objectid != key.objectid ||
2775                     found_key.type != key.type) {
2776                         ret = 0;
2777                         goto out;
2778                 }
2779
2780                 key.offset = found_key.offset;
2781                 btrfs_release_path(root, path);
2782                 cond_resched();
2783                 goto again;
2784         }
2785
2786         /* we know there's one more slot after us in the tree,
2787          * read that key so we can verify it is also a checksum item
2788          */
2789         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2790
2791         if (found_key.objectid < inode->i_ino)
2792                 goto next_key;
2793
2794         if (found_key.type != key.type || found_key.offset < new_size)
2795                 goto next_key;
2796
2797         /*
2798          * if the key for the next leaf isn't a csum key from this objectid,
2799          * we can't be sure there aren't good items inside this leaf.
2800          * Bail out
2801          */
2802         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2803                 goto out;
2804
2805         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2806         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2807         /*
2808          * it is safe to delete this leaf, it contains only
2809          * csum items from this inode at an offset >= new_size
2810          */
2811         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2812         BUG_ON(ret);
2813
2814         if (root->ref_cows && leaf_gen < trans->transid) {
2815                 ref = btrfs_alloc_leaf_ref(root, 0);
2816                 if (ref) {
2817                         ref->root_gen = root->root_key.offset;
2818                         ref->bytenr = leaf_start;
2819                         ref->owner = 0;
2820                         ref->generation = leaf_gen;
2821                         ref->nritems = 0;
2822
2823                         btrfs_sort_leaf_ref(ref);
2824
2825                         ret = btrfs_add_leaf_ref(root, ref, 0);
2826                         WARN_ON(ret);
2827                         btrfs_free_leaf_ref(root, ref);
2828                 } else {
2829                         WARN_ON(1);
2830                 }
2831         }
2832 next_key:
2833         btrfs_release_path(root, path);
2834
2835         if (other_key.objectid == inode->i_ino &&
2836             other_key.type == key.type && other_key.offset > key.offset) {
2837                 key.offset = other_key.offset;
2838                 cond_resched();
2839                 goto again;
2840         }
2841         ret = 0;
2842 out:
2843         /* fixup any changes we've made to the path */
2844         path->lowest_level = 0;
2845         path->keep_locks = 0;
2846         btrfs_release_path(root, path);
2847         return ret;
2848 }
2849
2850 #endif
2851
2852 /*
2853  * this can truncate away extent items, csum items and directory items.
2854  * It starts at a high offset and removes keys until it can't find
2855  * any higher than new_size
2856  *
2857  * csum items that cross the new i_size are truncated to the new size
2858  * as well.
2859  *
2860  * min_type is the minimum key type to truncate down to.  If set to 0, this
2861  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2862  */
2863 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2864                                struct btrfs_root *root,
2865                                struct inode *inode,
2866                                u64 new_size, u32 min_type)
2867 {
2868         struct btrfs_path *path;
2869         struct extent_buffer *leaf;
2870         struct btrfs_file_extent_item *fi;
2871         struct btrfs_key key;
2872         struct btrfs_key found_key;
2873         u64 extent_start = 0;
2874         u64 extent_num_bytes = 0;
2875         u64 extent_offset = 0;
2876         u64 item_end = 0;
2877         u64 mask = root->sectorsize - 1;
2878         u32 found_type = (u8)-1;
2879         int found_extent;
2880         int del_item;
2881         int pending_del_nr = 0;
2882         int pending_del_slot = 0;
2883         int extent_type = -1;
2884         int encoding;
2885         int ret;
2886         int err = 0;
2887
2888         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2889
2890         if (root->ref_cows)
2891                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2892
2893         path = btrfs_alloc_path();
2894         BUG_ON(!path);
2895         path->reada = -1;
2896
2897         key.objectid = inode->i_ino;
2898         key.offset = (u64)-1;
2899         key.type = (u8)-1;
2900
2901 search_again:
2902         path->leave_spinning = 1;
2903         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2904         if (ret < 0) {
2905                 err = ret;
2906                 goto out;
2907         }
2908
2909         if (ret > 0) {
2910                 /* there are no items in the tree for us to truncate, we're
2911                  * done
2912                  */
2913                 if (path->slots[0] == 0)
2914                         goto out;
2915                 path->slots[0]--;
2916         }
2917
2918         while (1) {
2919                 fi = NULL;
2920                 leaf = path->nodes[0];
2921                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2922                 found_type = btrfs_key_type(&found_key);
2923                 encoding = 0;
2924
2925                 if (found_key.objectid != inode->i_ino)
2926                         break;
2927
2928                 if (found_type < min_type)
2929                         break;
2930
2931                 item_end = found_key.offset;
2932                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2933                         fi = btrfs_item_ptr(leaf, path->slots[0],
2934                                             struct btrfs_file_extent_item);
2935                         extent_type = btrfs_file_extent_type(leaf, fi);
2936                         encoding = btrfs_file_extent_compression(leaf, fi);
2937                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2938                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2939
2940                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2941                                 item_end +=
2942                                     btrfs_file_extent_num_bytes(leaf, fi);
2943                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2944                                 item_end += btrfs_file_extent_inline_len(leaf,
2945                                                                          fi);
2946                         }
2947                         item_end--;
2948                 }
2949                 if (found_type > min_type) {
2950                         del_item = 1;
2951                 } else {
2952                         if (item_end < new_size)
2953                                 break;
2954                         if (found_key.offset >= new_size)
2955                                 del_item = 1;
2956                         else
2957                                 del_item = 0;
2958                 }
2959                 found_extent = 0;
2960                 /* FIXME, shrink the extent if the ref count is only 1 */
2961                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2962                         goto delete;
2963
2964                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2965                         u64 num_dec;
2966                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2967                         if (!del_item && !encoding) {
2968                                 u64 orig_num_bytes =
2969                                         btrfs_file_extent_num_bytes(leaf, fi);
2970                                 extent_num_bytes = new_size -
2971                                         found_key.offset + root->sectorsize - 1;
2972                                 extent_num_bytes = extent_num_bytes &
2973                                         ~((u64)root->sectorsize - 1);
2974                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2975                                                          extent_num_bytes);
2976                                 num_dec = (orig_num_bytes -
2977                                            extent_num_bytes);
2978                                 if (root->ref_cows && extent_start != 0)
2979                                         inode_sub_bytes(inode, num_dec);
2980                                 btrfs_mark_buffer_dirty(leaf);
2981                         } else {
2982                                 extent_num_bytes =
2983                                         btrfs_file_extent_disk_num_bytes(leaf,
2984                                                                          fi);
2985                                 extent_offset = found_key.offset -
2986                                         btrfs_file_extent_offset(leaf, fi);
2987
2988                                 /* FIXME blocksize != 4096 */
2989                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2990                                 if (extent_start != 0) {
2991                                         found_extent = 1;
2992                                         if (root->ref_cows)
2993                                                 inode_sub_bytes(inode, num_dec);
2994                                 }
2995                         }
2996                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2997                         /*
2998                          * we can't truncate inline items that have had
2999                          * special encodings
3000                          */
3001                         if (!del_item &&
3002                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3003                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3004                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3005                                 u32 size = new_size - found_key.offset;
3006
3007                                 if (root->ref_cows) {
3008                                         inode_sub_bytes(inode, item_end + 1 -
3009                                                         new_size);
3010                                 }
3011                                 size =
3012                                     btrfs_file_extent_calc_inline_size(size);
3013                                 ret = btrfs_truncate_item(trans, root, path,
3014                                                           size, 1);
3015                                 BUG_ON(ret);
3016                         } else if (root->ref_cows) {
3017                                 inode_sub_bytes(inode, item_end + 1 -
3018                                                 found_key.offset);
3019                         }
3020                 }
3021 delete:
3022                 if (del_item) {
3023                         if (!pending_del_nr) {
3024                                 /* no pending yet, add ourselves */
3025                                 pending_del_slot = path->slots[0];
3026                                 pending_del_nr = 1;
3027                         } else if (pending_del_nr &&
3028                                    path->slots[0] + 1 == pending_del_slot) {
3029                                 /* hop on the pending chunk */
3030                                 pending_del_nr++;
3031                                 pending_del_slot = path->slots[0];
3032                         } else {
3033                                 BUG();
3034                         }
3035                 } else {
3036                         break;
3037                 }
3038                 if (found_extent && root->ref_cows) {
3039                         btrfs_set_path_blocking(path);
3040                         ret = btrfs_free_extent(trans, root, extent_start,
3041                                                 extent_num_bytes, 0,
3042                                                 btrfs_header_owner(leaf),
3043                                                 inode->i_ino, extent_offset);
3044                         BUG_ON(ret);
3045                 }
3046
3047                 if (found_type == BTRFS_INODE_ITEM_KEY)
3048                         break;
3049
3050                 if (path->slots[0] == 0 ||
3051                     path->slots[0] != pending_del_slot) {
3052                         if (root->ref_cows) {
3053                                 err = -EAGAIN;
3054                                 goto out;
3055                         }
3056                         if (pending_del_nr) {
3057                                 ret = btrfs_del_items(trans, root, path,
3058                                                 pending_del_slot,
3059                                                 pending_del_nr);
3060                                 BUG_ON(ret);
3061                                 pending_del_nr = 0;
3062                         }
3063                         btrfs_release_path(root, path);
3064                         goto search_again;
3065                 } else {
3066                         path->slots[0]--;
3067                 }
3068         }
3069 out:
3070         if (pending_del_nr) {
3071                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3072                                       pending_del_nr);
3073         }
3074         btrfs_free_path(path);
3075         return err;
3076 }
3077
3078 /*
3079  * taken from block_truncate_page, but does cow as it zeros out
3080  * any bytes left in the last page in the file.
3081  */
3082 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3083 {
3084         struct inode *inode = mapping->host;
3085         struct btrfs_root *root = BTRFS_I(inode)->root;
3086         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3087         struct btrfs_ordered_extent *ordered;
3088         char *kaddr;
3089         u32 blocksize = root->sectorsize;
3090         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3091         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3092         struct page *page;
3093         int ret = 0;
3094         u64 page_start;
3095         u64 page_end;
3096
3097         if ((offset & (blocksize - 1)) == 0)
3098                 goto out;
3099         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3100         if (ret)
3101                 goto out;
3102
3103         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3104         if (ret)
3105                 goto out;
3106
3107         ret = -ENOMEM;
3108 again:
3109         page = grab_cache_page(mapping, index);
3110         if (!page) {
3111                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3112                 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3113                 goto out;
3114         }
3115
3116         page_start = page_offset(page);
3117         page_end = page_start + PAGE_CACHE_SIZE - 1;
3118
3119         if (!PageUptodate(page)) {
3120                 ret = btrfs_readpage(NULL, page);
3121                 lock_page(page);
3122                 if (page->mapping != mapping) {
3123                         unlock_page(page);
3124                         page_cache_release(page);
3125                         goto again;
3126                 }
3127                 if (!PageUptodate(page)) {
3128                         ret = -EIO;
3129                         goto out_unlock;
3130                 }
3131         }
3132         wait_on_page_writeback(page);
3133
3134         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3135         set_page_extent_mapped(page);
3136
3137         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3138         if (ordered) {
3139                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3140                 unlock_page(page);
3141                 page_cache_release(page);
3142                 btrfs_start_ordered_extent(inode, ordered, 1);
3143                 btrfs_put_ordered_extent(ordered);
3144                 goto again;
3145         }
3146
3147         clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
3148                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3149                           GFP_NOFS);
3150
3151         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3152         if (ret) {
3153                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3154                 goto out_unlock;
3155         }
3156
3157         ret = 0;
3158         if (offset != PAGE_CACHE_SIZE) {
3159                 kaddr = kmap(page);
3160                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3161                 flush_dcache_page(page);
3162                 kunmap(page);
3163         }
3164         ClearPageChecked(page);
3165         set_page_dirty(page);
3166         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3167
3168 out_unlock:
3169         if (ret)
3170                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3171         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3172         unlock_page(page);
3173         page_cache_release(page);
3174 out:
3175         return ret;
3176 }
3177
3178 int btrfs_cont_expand(struct inode *inode, loff_t size)
3179 {
3180         struct btrfs_trans_handle *trans;
3181         struct btrfs_root *root = BTRFS_I(inode)->root;
3182         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3183         struct extent_map *em;
3184         u64 mask = root->sectorsize - 1;
3185         u64 hole_start = (inode->i_size + mask) & ~mask;
3186         u64 block_end = (size + mask) & ~mask;
3187         u64 last_byte;
3188         u64 cur_offset;
3189         u64 hole_size;
3190         int err = 0;
3191
3192         if (size <= hole_start)
3193                 return 0;
3194
3195         while (1) {
3196                 struct btrfs_ordered_extent *ordered;
3197                 btrfs_wait_ordered_range(inode, hole_start,
3198                                          block_end - hole_start);
3199                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3200                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3201                 if (!ordered)
3202                         break;
3203                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3204                 btrfs_put_ordered_extent(ordered);
3205         }
3206
3207         cur_offset = hole_start;
3208         while (1) {
3209                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3210                                 block_end - cur_offset, 0);
3211                 BUG_ON(IS_ERR(em) || !em);
3212                 last_byte = min(extent_map_end(em), block_end);
3213                 last_byte = (last_byte + mask) & ~mask;
3214                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3215                         u64 hint_byte = 0;
3216                         hole_size = last_byte - cur_offset;
3217
3218                         err = btrfs_reserve_metadata_space(root, 2);
3219                         if (err)
3220                                 break;
3221
3222                         trans = btrfs_start_transaction(root, 1);
3223                         btrfs_set_trans_block_group(trans, inode);
3224
3225                         err = btrfs_drop_extents(trans, inode, cur_offset,
3226                                                  cur_offset + hole_size,
3227                                                  &hint_byte, 1);
3228                         BUG_ON(err);
3229
3230                         err = btrfs_insert_file_extent(trans, root,
3231                                         inode->i_ino, cur_offset, 0,
3232                                         0, hole_size, 0, hole_size,
3233                                         0, 0, 0);
3234                         BUG_ON(err);
3235
3236                         btrfs_drop_extent_cache(inode, hole_start,
3237                                         last_byte - 1, 0);
3238
3239                         btrfs_end_transaction(trans, root);
3240                         btrfs_unreserve_metadata_space(root, 2);
3241                 }
3242                 free_extent_map(em);
3243                 cur_offset = last_byte;
3244                 if (cur_offset >= block_end)
3245                         break;
3246         }
3247
3248         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3249         return err;
3250 }
3251
3252 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3253 {
3254         struct btrfs_root *root = BTRFS_I(inode)->root;
3255         struct btrfs_trans_handle *trans;
3256         unsigned long nr;
3257         int ret;
3258
3259         if (attr->ia_size == inode->i_size)
3260                 return 0;
3261
3262         if (attr->ia_size > inode->i_size) {
3263                 unsigned long limit;
3264                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3265                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3266                         return -EFBIG;
3267                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3268                         send_sig(SIGXFSZ, current, 0);
3269                         return -EFBIG;
3270                 }
3271         }
3272
3273         ret = btrfs_reserve_metadata_space(root, 1);
3274         if (ret)
3275                 return ret;
3276
3277         trans = btrfs_start_transaction(root, 1);
3278         btrfs_set_trans_block_group(trans, inode);
3279
3280         ret = btrfs_orphan_add(trans, inode);
3281         BUG_ON(ret);
3282
3283         nr = trans->blocks_used;
3284         btrfs_end_transaction(trans, root);
3285         btrfs_unreserve_metadata_space(root, 1);
3286         btrfs_btree_balance_dirty(root, nr);
3287
3288         if (attr->ia_size > inode->i_size) {
3289                 ret = btrfs_cont_expand(inode, attr->ia_size);
3290                 if (ret) {
3291                         btrfs_truncate(inode);
3292                         return ret;
3293                 }
3294
3295                 i_size_write(inode, attr->ia_size);
3296                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3297
3298                 trans = btrfs_start_transaction(root, 1);
3299                 btrfs_set_trans_block_group(trans, inode);
3300
3301                 ret = btrfs_update_inode(trans, root, inode);
3302                 BUG_ON(ret);
3303                 if (inode->i_nlink > 0) {
3304                         ret = btrfs_orphan_del(trans, inode);
3305                         BUG_ON(ret);
3306                 }
3307                 nr = trans->blocks_used;
3308                 btrfs_end_transaction(trans, root);
3309                 btrfs_btree_balance_dirty(root, nr);
3310                 return 0;
3311         }
3312
3313         /*
3314          * We're truncating a file that used to have good data down to
3315          * zero. Make sure it gets into the ordered flush list so that
3316          * any new writes get down to disk quickly.
3317          */
3318         if (attr->ia_size == 0)
3319                 BTRFS_I(inode)->ordered_data_close = 1;
3320
3321         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3322         ret = vmtruncate(inode, attr->ia_size);
3323         BUG_ON(ret);
3324
3325         return 0;
3326 }
3327
3328 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3329 {
3330         struct inode *inode = dentry->d_inode;
3331         int err;
3332
3333         err = inode_change_ok(inode, attr);
3334         if (err)
3335                 return err;
3336
3337         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3338                 err = btrfs_setattr_size(inode, attr);
3339                 if (err)
3340                         return err;
3341         }
3342         attr->ia_valid &= ~ATTR_SIZE;
3343
3344         if (attr->ia_valid)
3345                 err = inode_setattr(inode, attr);
3346
3347         if (!err && ((attr->ia_valid & ATTR_MODE)))
3348                 err = btrfs_acl_chmod(inode);
3349         return err;
3350 }
3351
3352 void btrfs_delete_inode(struct inode *inode)
3353 {
3354         struct btrfs_trans_handle *trans;
3355         struct btrfs_root *root = BTRFS_I(inode)->root;
3356         unsigned long nr;
3357         int ret;
3358
3359         truncate_inode_pages(&inode->i_data, 0);
3360         if (is_bad_inode(inode)) {
3361                 btrfs_orphan_del(NULL, inode);
3362                 goto no_delete;
3363         }
3364         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3365
3366         if (root->fs_info->log_root_recovering) {
3367                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3368                 goto no_delete;
3369         }
3370
3371         if (inode->i_nlink > 0) {
3372                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3373                 goto no_delete;
3374         }
3375
3376         btrfs_i_size_write(inode, 0);
3377
3378         while (1) {
3379                 trans = btrfs_start_transaction(root, 1);
3380                 btrfs_set_trans_block_group(trans, inode);
3381                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3382
3383                 if (ret != -EAGAIN)
3384                         break;
3385
3386                 nr = trans->blocks_used;
3387                 btrfs_end_transaction(trans, root);
3388                 trans = NULL;
3389                 btrfs_btree_balance_dirty(root, nr);
3390         }
3391
3392         if (ret == 0) {
3393                 ret = btrfs_orphan_del(trans, inode);
3394                 BUG_ON(ret);
3395         }
3396
3397         nr = trans->blocks_used;
3398         btrfs_end_transaction(trans, root);
3399         btrfs_btree_balance_dirty(root, nr);
3400 no_delete:
3401         clear_inode(inode);
3402         return;
3403 }
3404
3405 /*
3406  * this returns the key found in the dir entry in the location pointer.
3407  * If no dir entries were found, location->objectid is 0.
3408  */
3409 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3410                                struct btrfs_key *location)
3411 {
3412         const char *name = dentry->d_name.name;
3413         int namelen = dentry->d_name.len;
3414         struct btrfs_dir_item *di;
3415         struct btrfs_path *path;
3416         struct btrfs_root *root = BTRFS_I(dir)->root;
3417         int ret = 0;
3418
3419         path = btrfs_alloc_path();
3420         BUG_ON(!path);
3421
3422         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3423                                     namelen, 0);
3424         if (IS_ERR(di))
3425                 ret = PTR_ERR(di);
3426
3427         if (!di || IS_ERR(di))
3428                 goto out_err;
3429
3430         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3431 out:
3432         btrfs_free_path(path);
3433         return ret;
3434 out_err:
3435         location->objectid = 0;
3436         goto out;
3437 }
3438
3439 /*
3440  * when we hit a tree root in a directory, the btrfs part of the inode
3441  * needs to be changed to reflect the root directory of the tree root.  This
3442  * is kind of like crossing a mount point.
3443  */
3444 static int fixup_tree_root_location(struct btrfs_root *root,
3445                                     struct inode *dir,
3446                                     struct dentry *dentry,
3447                                     struct btrfs_key *location,
3448                                     struct btrfs_root **sub_root)
3449 {
3450         struct btrfs_path *path;
3451         struct btrfs_root *new_root;
3452         struct btrfs_root_ref *ref;
3453         struct extent_buffer *leaf;
3454         int ret;
3455         int err = 0;
3456
3457         path = btrfs_alloc_path();
3458         if (!path) {
3459                 err = -ENOMEM;
3460                 goto out;
3461         }
3462
3463         err = -ENOENT;
3464         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3465                                   BTRFS_I(dir)->root->root_key.objectid,
3466                                   location->objectid);
3467         if (ret) {
3468                 if (ret < 0)
3469                         err = ret;
3470                 goto out;
3471         }
3472
3473         leaf = path->nodes[0];
3474         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3475         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3476             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3477                 goto out;
3478
3479         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3480                                    (unsigned long)(ref + 1),
3481                                    dentry->d_name.len);
3482         if (ret)
3483                 goto out;
3484
3485         btrfs_release_path(root->fs_info->tree_root, path);
3486
3487         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3488         if (IS_ERR(new_root)) {
3489                 err = PTR_ERR(new_root);
3490                 goto out;
3491         }
3492
3493         if (btrfs_root_refs(&new_root->root_item) == 0) {
3494                 err = -ENOENT;
3495                 goto out;
3496         }
3497
3498         *sub_root = new_root;
3499         location->objectid = btrfs_root_dirid(&new_root->root_item);
3500         location->type = BTRFS_INODE_ITEM_KEY;
3501         location->offset = 0;
3502         err = 0;
3503 out:
3504         btrfs_free_path(path);
3505         return err;
3506 }
3507
3508 static void inode_tree_add(struct inode *inode)
3509 {
3510         struct btrfs_root *root = BTRFS_I(inode)->root;
3511         struct btrfs_inode *entry;
3512         struct rb_node **p;
3513         struct rb_node *parent;
3514 again:
3515         p = &root->inode_tree.rb_node;
3516         parent = NULL;
3517
3518         if (hlist_unhashed(&inode->i_hash))
3519                 return;
3520
3521         spin_lock(&root->inode_lock);
3522         while (*p) {
3523                 parent = *p;
3524                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3525
3526                 if (inode->i_ino < entry->vfs_inode.i_ino)
3527                         p = &parent->rb_left;
3528                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3529                         p = &parent->rb_right;
3530                 else {
3531                         WARN_ON(!(entry->vfs_inode.i_state &
3532                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3533                         rb_erase(parent, &root->inode_tree);
3534                         RB_CLEAR_NODE(parent);
3535                         spin_unlock(&root->inode_lock);
3536                         goto again;
3537                 }
3538         }
3539         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3540         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3541         spin_unlock(&root->inode_lock);
3542 }
3543
3544 static void inode_tree_del(struct inode *inode)
3545 {
3546         struct btrfs_root *root = BTRFS_I(inode)->root;
3547         int empty = 0;
3548
3549         spin_lock(&root->inode_lock);
3550         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3551                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3552                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3553                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3554         }
3555         spin_unlock(&root->inode_lock);
3556
3557         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3558                 synchronize_srcu(&root->fs_info->subvol_srcu);
3559                 spin_lock(&root->inode_lock);
3560                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3561                 spin_unlock(&root->inode_lock);
3562                 if (empty)
3563                         btrfs_add_dead_root(root);
3564         }
3565 }
3566
3567 int btrfs_invalidate_inodes(struct btrfs_root *root)
3568 {
3569         struct rb_node *node;
3570         struct rb_node *prev;
3571         struct btrfs_inode *entry;
3572         struct inode *inode;
3573         u64 objectid = 0;
3574
3575         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3576
3577         spin_lock(&root->inode_lock);
3578 again:
3579         node = root->inode_tree.rb_node;
3580         prev = NULL;
3581         while (node) {
3582                 prev = node;
3583                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3584
3585                 if (objectid < entry->vfs_inode.i_ino)
3586                         node = node->rb_left;
3587                 else if (objectid > entry->vfs_inode.i_ino)
3588                         node = node->rb_right;
3589                 else
3590                         break;
3591         }
3592         if (!node) {
3593                 while (prev) {
3594                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3595                         if (objectid <= entry->vfs_inode.i_ino) {
3596                                 node = prev;
3597                                 break;
3598                         }
3599                         prev = rb_next(prev);
3600                 }
3601         }
3602         while (node) {
3603                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3604                 objectid = entry->vfs_inode.i_ino + 1;
3605                 inode = igrab(&entry->vfs_inode);
3606                 if (inode) {
3607                         spin_unlock(&root->inode_lock);
3608                         if (atomic_read(&inode->i_count) > 1)
3609                                 d_prune_aliases(inode);
3610                         /*
3611                          * btrfs_drop_inode will remove it from
3612                          * the inode cache when its usage count
3613                          * hits zero.
3614                          */
3615                         iput(inode);
3616                         cond_resched();
3617                         spin_lock(&root->inode_lock);
3618                         goto again;
3619                 }
3620
3621                 if (cond_resched_lock(&root->inode_lock))
3622                         goto again;
3623
3624                 node = rb_next(node);
3625         }
3626         spin_unlock(&root->inode_lock);
3627         return 0;
3628 }
3629
3630 static noinline void init_btrfs_i(struct inode *inode)
3631 {
3632         struct btrfs_inode *bi = BTRFS_I(inode);
3633
3634         bi->generation = 0;
3635         bi->sequence = 0;
3636         bi->last_trans = 0;
3637         bi->last_sub_trans = 0;
3638         bi->logged_trans = 0;
3639         bi->delalloc_bytes = 0;
3640         bi->reserved_bytes = 0;
3641         bi->disk_i_size = 0;
3642         bi->flags = 0;
3643         bi->index_cnt = (u64)-1;
3644         bi->last_unlink_trans = 0;
3645         bi->ordered_data_close = 0;
3646         bi->force_compress = 0;
3647         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3648         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3649                              inode->i_mapping, GFP_NOFS);
3650         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3651                              inode->i_mapping, GFP_NOFS);
3652         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3653         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3654         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3655         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3656         mutex_init(&BTRFS_I(inode)->log_mutex);
3657 }
3658
3659 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3660 {
3661         struct btrfs_iget_args *args = p;
3662         inode->i_ino = args->ino;
3663         init_btrfs_i(inode);
3664         BTRFS_I(inode)->root = args->root;
3665         btrfs_set_inode_space_info(args->root, inode);
3666         return 0;
3667 }
3668
3669 static int btrfs_find_actor(struct inode *inode, void *opaque)
3670 {
3671         struct btrfs_iget_args *args = opaque;
3672         return args->ino == inode->i_ino &&
3673                 args->root == BTRFS_I(inode)->root;
3674 }
3675
3676 static struct inode *btrfs_iget_locked(struct super_block *s,
3677                                        u64 objectid,
3678                                        struct btrfs_root *root)
3679 {
3680         struct inode *inode;
3681         struct btrfs_iget_args args;
3682         args.ino = objectid;
3683         args.root = root;
3684
3685         inode = iget5_locked(s, objectid, btrfs_find_actor,
3686                              btrfs_init_locked_inode,
3687                              (void *)&args);
3688         return inode;
3689 }
3690
3691 /* Get an inode object given its location and corresponding root.
3692  * Returns in *is_new if the inode was read from disk
3693  */
3694 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3695                          struct btrfs_root *root, int *new)
3696 {
3697         struct inode *inode;
3698
3699         inode = btrfs_iget_locked(s, location->objectid, root);
3700         if (!inode)
3701                 return ERR_PTR(-ENOMEM);
3702
3703         if (inode->i_state & I_NEW) {
3704                 BTRFS_I(inode)->root = root;
3705                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3706                 btrfs_read_locked_inode(inode);
3707
3708                 inode_tree_add(inode);
3709                 unlock_new_inode(inode);
3710                 if (new)
3711                         *new = 1;
3712         }
3713
3714         return inode;
3715 }
3716
3717 static struct inode *new_simple_dir(struct super_block *s,
3718                                     struct btrfs_key *key,
3719                                     struct btrfs_root *root)
3720 {
3721         struct inode *inode = new_inode(s);
3722
3723         if (!inode)
3724                 return ERR_PTR(-ENOMEM);
3725
3726         init_btrfs_i(inode);
3727
3728         BTRFS_I(inode)->root = root;
3729         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3730         BTRFS_I(inode)->dummy_inode = 1;
3731
3732         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3733         inode->i_op = &simple_dir_inode_operations;
3734         inode->i_fop = &simple_dir_operations;
3735         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3736         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3737
3738         return inode;
3739 }
3740
3741 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3742 {
3743         struct inode *inode;
3744         struct btrfs_root *root = BTRFS_I(dir)->root;
3745         struct btrfs_root *sub_root = root;
3746         struct btrfs_key location;
3747         int index;
3748         int ret;
3749
3750         dentry->d_op = &btrfs_dentry_operations;
3751
3752         if (dentry->d_name.len > BTRFS_NAME_LEN)
3753                 return ERR_PTR(-ENAMETOOLONG);
3754
3755         ret = btrfs_inode_by_name(dir, dentry, &location);
3756
3757         if (ret < 0)
3758                 return ERR_PTR(ret);
3759
3760         if (location.objectid == 0)
3761                 return NULL;
3762
3763         if (location.type == BTRFS_INODE_ITEM_KEY) {
3764                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3765                 return inode;
3766         }
3767
3768         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3769
3770         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3771         ret = fixup_tree_root_location(root, dir, dentry,
3772                                        &location, &sub_root);
3773         if (ret < 0) {
3774                 if (ret != -ENOENT)
3775                         inode = ERR_PTR(ret);
3776                 else
3777                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3778         } else {
3779                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3780         }
3781         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3782
3783         if (root != sub_root) {
3784                 down_read(&root->fs_info->cleanup_work_sem);
3785                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3786                         btrfs_orphan_cleanup(sub_root);
3787                 up_read(&root->fs_info->cleanup_work_sem);
3788         }
3789
3790         return inode;
3791 }
3792
3793 static int btrfs_dentry_delete(struct dentry *dentry)
3794 {
3795         struct btrfs_root *root;
3796
3797         if (!dentry->d_inode && !IS_ROOT(dentry))
3798                 dentry = dentry->d_parent;
3799
3800         if (dentry->d_inode) {
3801                 root = BTRFS_I(dentry->d_inode)->root;
3802                 if (btrfs_root_refs(&root->root_item) == 0)
3803                         return 1;
3804         }
3805         return 0;
3806 }
3807
3808 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3809                                    struct nameidata *nd)
3810 {
3811         struct inode *inode;
3812
3813         inode = btrfs_lookup_dentry(dir, dentry);
3814         if (IS_ERR(inode))
3815                 return ERR_CAST(inode);
3816
3817         return d_splice_alias(inode, dentry);
3818 }
3819
3820 static unsigned char btrfs_filetype_table[] = {
3821         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3822 };
3823
3824 static int btrfs_real_readdir(struct file *filp, void *dirent,
3825                               filldir_t filldir)
3826 {
3827         struct inode *inode = filp->f_dentry->d_inode;
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         struct btrfs_item *item;
3830         struct btrfs_dir_item *di;
3831         struct btrfs_key key;
3832         struct btrfs_key found_key;
3833         struct btrfs_path *path;
3834         int ret;
3835         u32 nritems;
3836         struct extent_buffer *leaf;
3837         int slot;
3838         int advance;
3839         unsigned char d_type;
3840         int over = 0;
3841         u32 di_cur;
3842         u32 di_total;
3843         u32 di_len;
3844         int key_type = BTRFS_DIR_INDEX_KEY;
3845         char tmp_name[32];
3846         char *name_ptr;
3847         int name_len;
3848
3849         /* FIXME, use a real flag for deciding about the key type */
3850         if (root->fs_info->tree_root == root)
3851                 key_type = BTRFS_DIR_ITEM_KEY;
3852
3853         /* special case for "." */
3854         if (filp->f_pos == 0) {
3855                 over = filldir(dirent, ".", 1,
3856                                1, inode->i_ino,
3857                                DT_DIR);
3858                 if (over)
3859                         return 0;
3860                 filp->f_pos = 1;
3861         }
3862         /* special case for .., just use the back ref */
3863         if (filp->f_pos == 1) {
3864                 u64 pino = parent_ino(filp->f_path.dentry);
3865                 over = filldir(dirent, "..", 2,
3866                                2, pino, DT_DIR);
3867                 if (over)
3868                         return 0;
3869                 filp->f_pos = 2;
3870         }
3871         path = btrfs_alloc_path();
3872         path->reada = 2;
3873
3874         btrfs_set_key_type(&key, key_type);
3875         key.offset = filp->f_pos;
3876         key.objectid = inode->i_ino;
3877
3878         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3879         if (ret < 0)
3880                 goto err;
3881         advance = 0;
3882
3883         while (1) {
3884                 leaf = path->nodes[0];
3885                 nritems = btrfs_header_nritems(leaf);
3886                 slot = path->slots[0];
3887                 if (advance || slot >= nritems) {
3888                         if (slot >= nritems - 1) {
3889                                 ret = btrfs_next_leaf(root, path);
3890                                 if (ret)
3891                                         break;
3892                                 leaf = path->nodes[0];
3893                                 nritems = btrfs_header_nritems(leaf);
3894                                 slot = path->slots[0];
3895                         } else {
3896                                 slot++;
3897                                 path->slots[0]++;
3898                         }
3899                 }
3900
3901                 advance = 1;
3902                 item = btrfs_item_nr(leaf, slot);
3903                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3904
3905                 if (found_key.objectid != key.objectid)
3906                         break;
3907                 if (btrfs_key_type(&found_key) != key_type)
3908                         break;
3909                 if (found_key.offset < filp->f_pos)
3910                         continue;
3911
3912                 filp->f_pos = found_key.offset;
3913
3914                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3915                 di_cur = 0;
3916                 di_total = btrfs_item_size(leaf, item);
3917
3918                 while (di_cur < di_total) {
3919                         struct btrfs_key location;
3920
3921                         name_len = btrfs_dir_name_len(leaf, di);
3922                         if (name_len <= sizeof(tmp_name)) {
3923                                 name_ptr = tmp_name;
3924                         } else {
3925                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3926                                 if (!name_ptr) {
3927                                         ret = -ENOMEM;
3928                                         goto err;
3929                                 }
3930                         }
3931                         read_extent_buffer(leaf, name_ptr,
3932                                            (unsigned long)(di + 1), name_len);
3933
3934                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3935                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3936
3937                         /* is this a reference to our own snapshot? If so
3938                          * skip it
3939                          */
3940                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3941                             location.objectid == root->root_key.objectid) {
3942                                 over = 0;
3943                                 goto skip;
3944                         }
3945                         over = filldir(dirent, name_ptr, name_len,
3946                                        found_key.offset, location.objectid,
3947                                        d_type);
3948
3949 skip:
3950                         if (name_ptr != tmp_name)
3951                                 kfree(name_ptr);
3952
3953                         if (over)
3954                                 goto nopos;
3955                         di_len = btrfs_dir_name_len(leaf, di) +
3956                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3957                         di_cur += di_len;
3958                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3959                 }
3960         }
3961
3962         /* Reached end of directory/root. Bump pos past the last item. */
3963         if (key_type == BTRFS_DIR_INDEX_KEY)
3964                 /*
3965                  * 32-bit glibc will use getdents64, but then strtol -
3966                  * so the last number we can serve is this.
3967                  */
3968                 filp->f_pos = 0x7fffffff;
3969         else
3970                 filp->f_pos++;
3971 nopos:
3972         ret = 0;
3973 err:
3974         btrfs_free_path(path);
3975         return ret;
3976 }
3977
3978 int btrfs_write_inode(struct inode *inode, int wait)
3979 {
3980         struct btrfs_root *root = BTRFS_I(inode)->root;
3981         struct btrfs_trans_handle *trans;
3982         int ret = 0;
3983
3984         if (root->fs_info->btree_inode == inode)
3985                 return 0;
3986
3987         if (wait) {
3988                 trans = btrfs_join_transaction(root, 1);
3989                 btrfs_set_trans_block_group(trans, inode);
3990                 ret = btrfs_commit_transaction(trans, root);
3991         }
3992         return ret;
3993 }
3994
3995 /*
3996  * This is somewhat expensive, updating the tree every time the
3997  * inode changes.  But, it is most likely to find the inode in cache.
3998  * FIXME, needs more benchmarking...there are no reasons other than performance
3999  * to keep or drop this code.
4000  */
4001 void btrfs_dirty_inode(struct inode *inode)
4002 {
4003         struct btrfs_root *root = BTRFS_I(inode)->root;
4004         struct btrfs_trans_handle *trans;
4005
4006         trans = btrfs_join_transaction(root, 1);
4007         btrfs_set_trans_block_group(trans, inode);
4008         btrfs_update_inode(trans, root, inode);
4009         btrfs_end_transaction(trans, root);
4010 }
4011
4012 /*
4013  * find the highest existing sequence number in a directory
4014  * and then set the in-memory index_cnt variable to reflect
4015  * free sequence numbers
4016  */
4017 static int btrfs_set_inode_index_count(struct inode *inode)
4018 {
4019         struct btrfs_root *root = BTRFS_I(inode)->root;
4020         struct btrfs_key key, found_key;
4021         struct btrfs_path *path;
4022         struct extent_buffer *leaf;
4023         int ret;
4024
4025         key.objectid = inode->i_ino;
4026         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4027         key.offset = (u64)-1;
4028
4029         path = btrfs_alloc_path();
4030         if (!path)
4031                 return -ENOMEM;
4032
4033         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4034         if (ret < 0)
4035                 goto out;
4036         /* FIXME: we should be able to handle this */
4037         if (ret == 0)
4038                 goto out;
4039         ret = 0;
4040
4041         /*
4042          * MAGIC NUMBER EXPLANATION:
4043          * since we search a directory based on f_pos we have to start at 2
4044          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4045          * else has to start at 2
4046          */
4047         if (path->slots[0] == 0) {
4048                 BTRFS_I(inode)->index_cnt = 2;
4049                 goto out;
4050         }
4051
4052         path->slots[0]--;
4053
4054         leaf = path->nodes[0];
4055         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4056
4057         if (found_key.objectid != inode->i_ino ||
4058             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4059                 BTRFS_I(inode)->index_cnt = 2;
4060                 goto out;
4061         }
4062
4063         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4064 out:
4065         btrfs_free_path(path);
4066         return ret;
4067 }
4068
4069 /*
4070  * helper to find a free sequence number in a given directory.  This current
4071  * code is very simple, later versions will do smarter things in the btree
4072  */
4073 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4074 {
4075         int ret = 0;
4076
4077         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4078                 ret = btrfs_set_inode_index_count(dir);
4079                 if (ret)
4080                         return ret;
4081         }
4082
4083         *index = BTRFS_I(dir)->index_cnt;
4084         BTRFS_I(dir)->index_cnt++;
4085
4086         return ret;
4087 }
4088
4089 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4090                                      struct btrfs_root *root,
4091                                      struct inode *dir,
4092                                      const char *name, int name_len,
4093                                      u64 ref_objectid, u64 objectid,
4094                                      u64 alloc_hint, int mode, u64 *index)
4095 {
4096         struct inode *inode;
4097         struct btrfs_inode_item *inode_item;
4098         struct btrfs_key *location;
4099         struct btrfs_path *path;
4100         struct btrfs_inode_ref *ref;
4101         struct btrfs_key key[2];
4102         u32 sizes[2];
4103         unsigned long ptr;
4104         int ret;
4105         int owner;
4106
4107         path = btrfs_alloc_path();
4108         BUG_ON(!path);
4109
4110         inode = new_inode(root->fs_info->sb);
4111         if (!inode)
4112                 return ERR_PTR(-ENOMEM);
4113
4114         if (dir) {
4115                 ret = btrfs_set_inode_index(dir, index);
4116                 if (ret) {
4117                         iput(inode);
4118                         return ERR_PTR(ret);
4119                 }
4120         }
4121         /*
4122          * index_cnt is ignored for everything but a dir,
4123          * btrfs_get_inode_index_count has an explanation for the magic
4124          * number
4125          */
4126         init_btrfs_i(inode);
4127         BTRFS_I(inode)->index_cnt = 2;
4128         BTRFS_I(inode)->root = root;
4129         BTRFS_I(inode)->generation = trans->transid;
4130         btrfs_set_inode_space_info(root, inode);
4131
4132         if (mode & S_IFDIR)
4133                 owner = 0;
4134         else
4135                 owner = 1;
4136         BTRFS_I(inode)->block_group =
4137                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4138
4139         key[0].objectid = objectid;
4140         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4141         key[0].offset = 0;
4142
4143         key[1].objectid = objectid;
4144         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4145         key[1].offset = ref_objectid;
4146
4147         sizes[0] = sizeof(struct btrfs_inode_item);
4148         sizes[1] = name_len + sizeof(*ref);
4149
4150         path->leave_spinning = 1;
4151         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4152         if (ret != 0)
4153                 goto fail;
4154
4155         inode->i_uid = current_fsuid();
4156
4157         if (dir && (dir->i_mode & S_ISGID)) {
4158                 inode->i_gid = dir->i_gid;
4159                 if (S_ISDIR(mode))
4160                         mode |= S_ISGID;
4161         } else
4162                 inode->i_gid = current_fsgid();
4163
4164         inode->i_mode = mode;
4165         inode->i_ino = objectid;
4166         inode_set_bytes(inode, 0);
4167         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4168         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4169                                   struct btrfs_inode_item);
4170         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4171
4172         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4173                              struct btrfs_inode_ref);
4174         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4175         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4176         ptr = (unsigned long)(ref + 1);
4177         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4178
4179         btrfs_mark_buffer_dirty(path->nodes[0]);
4180         btrfs_free_path(path);
4181
4182         location = &BTRFS_I(inode)->location;
4183         location->objectid = objectid;
4184         location->offset = 0;
4185         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4186
4187         btrfs_inherit_iflags(inode, dir);
4188
4189         if ((mode & S_IFREG)) {
4190                 if (btrfs_test_opt(root, NODATASUM))
4191                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4192                 if (btrfs_test_opt(root, NODATACOW))
4193                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4194         }
4195
4196         insert_inode_hash(inode);
4197         inode_tree_add(inode);
4198         return inode;
4199 fail:
4200         if (dir)
4201                 BTRFS_I(dir)->index_cnt--;
4202         btrfs_free_path(path);
4203         iput(inode);
4204         return ERR_PTR(ret);
4205 }
4206
4207 static inline u8 btrfs_inode_type(struct inode *inode)
4208 {
4209         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4210 }
4211
4212 /*
4213  * utility function to add 'inode' into 'parent_inode' with
4214  * a give name and a given sequence number.
4215  * if 'add_backref' is true, also insert a backref from the
4216  * inode to the parent directory.
4217  */
4218 int btrfs_add_link(struct btrfs_trans_handle *trans,
4219                    struct inode *parent_inode, struct inode *inode,
4220                    const char *name, int name_len, int add_backref, u64 index)
4221 {
4222         int ret = 0;
4223         struct btrfs_key key;
4224         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4225
4226         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4227                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4228         } else {
4229                 key.objectid = inode->i_ino;
4230                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4231                 key.offset = 0;
4232         }
4233
4234         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4235                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4236                                          key.objectid, root->root_key.objectid,
4237                                          parent_inode->i_ino,
4238                                          index, name, name_len);
4239         } else if (add_backref) {
4240                 ret = btrfs_insert_inode_ref(trans, root,
4241                                              name, name_len, inode->i_ino,
4242                                              parent_inode->i_ino, index);
4243         }
4244
4245         if (ret == 0) {
4246                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4247                                             parent_inode->i_ino, &key,
4248                                             btrfs_inode_type(inode), index);
4249                 BUG_ON(ret);
4250
4251                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4252                                    name_len * 2);
4253                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4254                 ret = btrfs_update_inode(trans, root, parent_inode);
4255         }
4256         return ret;
4257 }
4258
4259 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4260                             struct dentry *dentry, struct inode *inode,
4261                             int backref, u64 index)
4262 {
4263         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4264                                  inode, dentry->d_name.name,
4265                                  dentry->d_name.len, backref, index);
4266         if (!err) {
4267                 d_instantiate(dentry, inode);
4268                 return 0;
4269         }
4270         if (err > 0)
4271                 err = -EEXIST;
4272         return err;
4273 }
4274
4275 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4276                         int mode, dev_t rdev)
4277 {
4278         struct btrfs_trans_handle *trans;
4279         struct btrfs_root *root = BTRFS_I(dir)->root;
4280         struct inode *inode = NULL;
4281         int err;
4282         int drop_inode = 0;
4283         u64 objectid;
4284         unsigned long nr = 0;
4285         u64 index = 0;
4286
4287         if (!new_valid_dev(rdev))
4288                 return -EINVAL;
4289
4290         /*
4291          * 2 for inode item and ref
4292          * 2 for dir items
4293          * 1 for xattr if selinux is on
4294          */
4295         err = btrfs_reserve_metadata_space(root, 5);
4296         if (err)
4297                 return err;
4298
4299         trans = btrfs_start_transaction(root, 1);
4300         if (!trans)
4301                 goto fail;
4302         btrfs_set_trans_block_group(trans, dir);
4303
4304         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4305         if (err) {
4306                 err = -ENOSPC;
4307                 goto out_unlock;
4308         }
4309
4310         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4311                                 dentry->d_name.len,
4312                                 dentry->d_parent->d_inode->i_ino, objectid,
4313                                 BTRFS_I(dir)->block_group, mode, &index);
4314         err = PTR_ERR(inode);
4315         if (IS_ERR(inode))
4316                 goto out_unlock;
4317
4318         err = btrfs_init_inode_security(trans, inode, dir);
4319         if (err) {
4320                 drop_inode = 1;
4321                 goto out_unlock;
4322         }
4323
4324         btrfs_set_trans_block_group(trans, inode);
4325         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4326         if (err)
4327                 drop_inode = 1;
4328         else {
4329                 inode->i_op = &btrfs_special_inode_operations;
4330                 init_special_inode(inode, inode->i_mode, rdev);
4331                 btrfs_update_inode(trans, root, inode);
4332         }
4333         btrfs_update_inode_block_group(trans, inode);
4334         btrfs_update_inode_block_group(trans, dir);
4335 out_unlock:
4336         nr = trans->blocks_used;
4337         btrfs_end_transaction_throttle(trans, root);
4338 fail:
4339         btrfs_unreserve_metadata_space(root, 5);
4340         if (drop_inode) {
4341                 inode_dec_link_count(inode);
4342                 iput(inode);
4343         }
4344         btrfs_btree_balance_dirty(root, nr);
4345         return err;
4346 }
4347
4348 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4349                         int mode, struct nameidata *nd)
4350 {
4351         struct btrfs_trans_handle *trans;
4352         struct btrfs_root *root = BTRFS_I(dir)->root;
4353         struct inode *inode = NULL;
4354         int err;
4355         int drop_inode = 0;
4356         unsigned long nr = 0;
4357         u64 objectid;
4358         u64 index = 0;
4359
4360         /*
4361          * 2 for inode item and ref
4362          * 2 for dir items
4363          * 1 for xattr if selinux is on
4364          */
4365         err = btrfs_reserve_metadata_space(root, 5);
4366         if (err)
4367                 return err;
4368
4369         trans = btrfs_start_transaction(root, 1);
4370         if (!trans)
4371                 goto fail;
4372         btrfs_set_trans_block_group(trans, dir);
4373
4374         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4375         if (err) {
4376                 err = -ENOSPC;
4377                 goto out_unlock;
4378         }
4379
4380         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4381                                 dentry->d_name.len,
4382                                 dentry->d_parent->d_inode->i_ino,
4383                                 objectid, BTRFS_I(dir)->block_group, mode,
4384                                 &index);
4385         err = PTR_ERR(inode);
4386         if (IS_ERR(inode))
4387                 goto out_unlock;
4388
4389         err = btrfs_init_inode_security(trans, inode, dir);
4390         if (err) {
4391                 drop_inode = 1;
4392                 goto out_unlock;
4393         }
4394
4395         btrfs_set_trans_block_group(trans, inode);
4396         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4397         if (err)
4398                 drop_inode = 1;
4399         else {
4400                 inode->i_mapping->a_ops = &btrfs_aops;
4401                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4402                 inode->i_fop = &btrfs_file_operations;
4403                 inode->i_op = &btrfs_file_inode_operations;
4404                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4405         }
4406         btrfs_update_inode_block_group(trans, inode);
4407         btrfs_update_inode_block_group(trans, dir);
4408 out_unlock:
4409         nr = trans->blocks_used;
4410         btrfs_end_transaction_throttle(trans, root);
4411 fail:
4412         btrfs_unreserve_metadata_space(root, 5);
4413         if (drop_inode) {
4414                 inode_dec_link_count(inode);
4415                 iput(inode);
4416         }
4417         btrfs_btree_balance_dirty(root, nr);
4418         return err;
4419 }
4420
4421 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4422                       struct dentry *dentry)
4423 {
4424         struct btrfs_trans_handle *trans;
4425         struct btrfs_root *root = BTRFS_I(dir)->root;
4426         struct inode *inode = old_dentry->d_inode;
4427         u64 index;
4428         unsigned long nr = 0;
4429         int err;
4430         int drop_inode = 0;
4431
4432         if (inode->i_nlink == 0)
4433                 return -ENOENT;
4434
4435         /* do not allow sys_link's with other subvols of the same device */
4436         if (root->objectid != BTRFS_I(inode)->root->objectid)
4437                 return -EPERM;
4438
4439         /*
4440          * 1 item for inode ref
4441          * 2 items for dir items
4442          */
4443         err = btrfs_reserve_metadata_space(root, 3);
4444         if (err)
4445                 return err;
4446
4447         btrfs_inc_nlink(inode);
4448
4449         err = btrfs_set_inode_index(dir, &index);
4450         if (err)
4451                 goto fail;
4452
4453         trans = btrfs_start_transaction(root, 1);
4454
4455         btrfs_set_trans_block_group(trans, dir);
4456         atomic_inc(&inode->i_count);
4457
4458         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4459
4460         if (err) {
4461                 drop_inode = 1;
4462         } else {
4463                 btrfs_update_inode_block_group(trans, dir);
4464                 err = btrfs_update_inode(trans, root, inode);
4465                 BUG_ON(err);
4466                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4467         }
4468
4469         nr = trans->blocks_used;
4470         btrfs_end_transaction_throttle(trans, root);
4471 fail:
4472         btrfs_unreserve_metadata_space(root, 3);
4473         if (drop_inode) {
4474                 inode_dec_link_count(inode);
4475                 iput(inode);
4476         }
4477         btrfs_btree_balance_dirty(root, nr);
4478         return err;
4479 }
4480
4481 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4482 {
4483         struct inode *inode = NULL;
4484         struct btrfs_trans_handle *trans;
4485         struct btrfs_root *root = BTRFS_I(dir)->root;
4486         int err = 0;
4487         int drop_on_err = 0;
4488         u64 objectid = 0;
4489         u64 index = 0;
4490         unsigned long nr = 1;
4491
4492         /*
4493          * 2 items for inode and ref
4494          * 2 items for dir items
4495          * 1 for xattr if selinux is on
4496          */
4497         err = btrfs_reserve_metadata_space(root, 5);
4498         if (err)
4499                 return err;
4500
4501         trans = btrfs_start_transaction(root, 1);
4502         if (!trans) {
4503                 err = -ENOMEM;
4504                 goto out_unlock;
4505         }
4506         btrfs_set_trans_block_group(trans, dir);
4507
4508         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4509         if (err) {
4510                 err = -ENOSPC;
4511                 goto out_unlock;
4512         }
4513
4514         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4515                                 dentry->d_name.len,
4516                                 dentry->d_parent->d_inode->i_ino, objectid,
4517                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4518                                 &index);
4519         if (IS_ERR(inode)) {
4520                 err = PTR_ERR(inode);
4521                 goto out_fail;
4522         }
4523
4524         drop_on_err = 1;
4525
4526         err = btrfs_init_inode_security(trans, inode, dir);
4527         if (err)
4528                 goto out_fail;
4529
4530         inode->i_op = &btrfs_dir_inode_operations;
4531         inode->i_fop = &btrfs_dir_file_operations;
4532         btrfs_set_trans_block_group(trans, inode);
4533
4534         btrfs_i_size_write(inode, 0);
4535         err = btrfs_update_inode(trans, root, inode);
4536         if (err)
4537                 goto out_fail;
4538
4539         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4540                                  inode, dentry->d_name.name,
4541                                  dentry->d_name.len, 0, index);
4542         if (err)
4543                 goto out_fail;
4544
4545         d_instantiate(dentry, inode);
4546         drop_on_err = 0;
4547         btrfs_update_inode_block_group(trans, inode);
4548         btrfs_update_inode_block_group(trans, dir);
4549
4550 out_fail:
4551         nr = trans->blocks_used;
4552         btrfs_end_transaction_throttle(trans, root);
4553
4554 out_unlock:
4555         btrfs_unreserve_metadata_space(root, 5);
4556         if (drop_on_err)
4557                 iput(inode);
4558         btrfs_btree_balance_dirty(root, nr);
4559         return err;
4560 }
4561
4562 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4563  * and an extent that you want to insert, deal with overlap and insert
4564  * the new extent into the tree.
4565  */
4566 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4567                                 struct extent_map *existing,
4568                                 struct extent_map *em,
4569                                 u64 map_start, u64 map_len)
4570 {
4571         u64 start_diff;
4572
4573         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4574         start_diff = map_start - em->start;
4575         em->start = map_start;
4576         em->len = map_len;
4577         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4578             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4579                 em->block_start += start_diff;
4580                 em->block_len -= start_diff;
4581         }
4582         return add_extent_mapping(em_tree, em);
4583 }
4584
4585 static noinline int uncompress_inline(struct btrfs_path *path,
4586                                       struct inode *inode, struct page *page,
4587                                       size_t pg_offset, u64 extent_offset,
4588                                       struct btrfs_file_extent_item *item)
4589 {
4590         int ret;
4591         struct extent_buffer *leaf = path->nodes[0];
4592         char *tmp;
4593         size_t max_size;
4594         unsigned long inline_size;
4595         unsigned long ptr;
4596
4597         WARN_ON(pg_offset != 0);
4598         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4599         inline_size = btrfs_file_extent_inline_item_len(leaf,
4600                                         btrfs_item_nr(leaf, path->slots[0]));
4601         tmp = kmalloc(inline_size, GFP_NOFS);
4602         ptr = btrfs_file_extent_inline_start(item);
4603
4604         read_extent_buffer(leaf, tmp, ptr, inline_size);
4605
4606         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4607         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4608                                     inline_size, max_size);
4609         if (ret) {
4610                 char *kaddr = kmap_atomic(page, KM_USER0);
4611                 unsigned long copy_size = min_t(u64,
4612                                   PAGE_CACHE_SIZE - pg_offset,
4613                                   max_size - extent_offset);
4614                 memset(kaddr + pg_offset, 0, copy_size);
4615                 kunmap_atomic(kaddr, KM_USER0);
4616         }
4617         kfree(tmp);
4618         return 0;
4619 }
4620
4621 /*
4622  * a bit scary, this does extent mapping from logical file offset to the disk.
4623  * the ugly parts come from merging extents from the disk with the in-ram
4624  * representation.  This gets more complex because of the data=ordered code,
4625  * where the in-ram extents might be locked pending data=ordered completion.
4626  *
4627  * This also copies inline extents directly into the page.
4628  */
4629
4630 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4631                                     size_t pg_offset, u64 start, u64 len,
4632                                     int create)
4633 {
4634         int ret;
4635         int err = 0;
4636         u64 bytenr;
4637         u64 extent_start = 0;
4638         u64 extent_end = 0;
4639         u64 objectid = inode->i_ino;
4640         u32 found_type;
4641         struct btrfs_path *path = NULL;
4642         struct btrfs_root *root = BTRFS_I(inode)->root;
4643         struct btrfs_file_extent_item *item;
4644         struct extent_buffer *leaf;
4645         struct btrfs_key found_key;
4646         struct extent_map *em = NULL;
4647         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4648         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4649         struct btrfs_trans_handle *trans = NULL;
4650         int compressed;
4651
4652 again:
4653         read_lock(&em_tree->lock);
4654         em = lookup_extent_mapping(em_tree, start, len);
4655         if (em)
4656                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4657         read_unlock(&em_tree->lock);
4658
4659         if (em) {
4660                 if (em->start > start || em->start + em->len <= start)
4661                         free_extent_map(em);
4662                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4663                         free_extent_map(em);
4664                 else
4665                         goto out;
4666         }
4667         em = alloc_extent_map(GFP_NOFS);
4668         if (!em) {
4669                 err = -ENOMEM;
4670                 goto out;
4671         }
4672         em->bdev = root->fs_info->fs_devices->latest_bdev;
4673         em->start = EXTENT_MAP_HOLE;
4674         em->orig_start = EXTENT_MAP_HOLE;
4675         em->len = (u64)-1;
4676         em->block_len = (u64)-1;
4677
4678         if (!path) {
4679                 path = btrfs_alloc_path();
4680                 BUG_ON(!path);
4681         }
4682
4683         ret = btrfs_lookup_file_extent(trans, root, path,
4684                                        objectid, start, trans != NULL);
4685         if (ret < 0) {
4686                 err = ret;
4687                 goto out;
4688         }
4689
4690         if (ret != 0) {
4691                 if (path->slots[0] == 0)
4692                         goto not_found;
4693                 path->slots[0]--;
4694         }
4695
4696         leaf = path->nodes[0];
4697         item = btrfs_item_ptr(leaf, path->slots[0],
4698                               struct btrfs_file_extent_item);
4699         /* are we inside the extent that was found? */
4700         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4701         found_type = btrfs_key_type(&found_key);
4702         if (found_key.objectid != objectid ||
4703             found_type != BTRFS_EXTENT_DATA_KEY) {
4704                 goto not_found;
4705         }
4706
4707         found_type = btrfs_file_extent_type(leaf, item);
4708         extent_start = found_key.offset;
4709         compressed = btrfs_file_extent_compression(leaf, item);
4710         if (found_type == BTRFS_FILE_EXTENT_REG ||
4711             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4712                 extent_end = extent_start +
4713                        btrfs_file_extent_num_bytes(leaf, item);
4714         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4715                 size_t size;
4716                 size = btrfs_file_extent_inline_len(leaf, item);
4717                 extent_end = (extent_start + size + root->sectorsize - 1) &
4718                         ~((u64)root->sectorsize - 1);
4719         }
4720
4721         if (start >= extent_end) {
4722                 path->slots[0]++;
4723                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4724                         ret = btrfs_next_leaf(root, path);
4725                         if (ret < 0) {
4726                                 err = ret;
4727                                 goto out;
4728                         }
4729                         if (ret > 0)
4730                                 goto not_found;
4731                         leaf = path->nodes[0];
4732                 }
4733                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4734                 if (found_key.objectid != objectid ||
4735                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4736                         goto not_found;
4737                 if (start + len <= found_key.offset)
4738                         goto not_found;
4739                 em->start = start;
4740                 em->len = found_key.offset - start;
4741                 goto not_found_em;
4742         }
4743
4744         if (found_type == BTRFS_FILE_EXTENT_REG ||
4745             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4746                 em->start = extent_start;
4747                 em->len = extent_end - extent_start;
4748                 em->orig_start = extent_start -
4749                                  btrfs_file_extent_offset(leaf, item);
4750                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4751                 if (bytenr == 0) {
4752                         em->block_start = EXTENT_MAP_HOLE;
4753                         goto insert;
4754                 }
4755                 if (compressed) {
4756                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4757                         em->block_start = bytenr;
4758                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4759                                                                          item);
4760                 } else {
4761                         bytenr += btrfs_file_extent_offset(leaf, item);
4762                         em->block_start = bytenr;
4763                         em->block_len = em->len;
4764                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4765                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4766                 }
4767                 goto insert;
4768         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4769                 unsigned long ptr;
4770                 char *map;
4771                 size_t size;
4772                 size_t extent_offset;
4773                 size_t copy_size;
4774
4775                 em->block_start = EXTENT_MAP_INLINE;
4776                 if (!page || create) {
4777                         em->start = extent_start;
4778                         em->len = extent_end - extent_start;
4779                         goto out;
4780                 }
4781
4782                 size = btrfs_file_extent_inline_len(leaf, item);
4783                 extent_offset = page_offset(page) + pg_offset - extent_start;
4784                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4785                                 size - extent_offset);
4786                 em->start = extent_start + extent_offset;
4787                 em->len = (copy_size + root->sectorsize - 1) &
4788                         ~((u64)root->sectorsize - 1);
4789                 em->orig_start = EXTENT_MAP_INLINE;
4790                 if (compressed)
4791                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4792                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4793                 if (create == 0 && !PageUptodate(page)) {
4794                         if (btrfs_file_extent_compression(leaf, item) ==
4795                             BTRFS_COMPRESS_ZLIB) {
4796                                 ret = uncompress_inline(path, inode, page,
4797                                                         pg_offset,
4798                                                         extent_offset, item);
4799                                 BUG_ON(ret);
4800                         } else {
4801                                 map = kmap(page);
4802                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4803                                                    copy_size);
4804                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4805                                         memset(map + pg_offset + copy_size, 0,
4806                                                PAGE_CACHE_SIZE - pg_offset -
4807                                                copy_size);
4808                                 }
4809                                 kunmap(page);
4810                         }
4811                         flush_dcache_page(page);
4812                 } else if (create && PageUptodate(page)) {
4813                         if (!trans) {
4814                                 kunmap(page);
4815                                 free_extent_map(em);
4816                                 em = NULL;
4817                                 btrfs_release_path(root, path);
4818                                 trans = btrfs_join_transaction(root, 1);
4819                                 goto again;
4820                         }
4821                         map = kmap(page);
4822                         write_extent_buffer(leaf, map + pg_offset, ptr,
4823                                             copy_size);
4824                         kunmap(page);
4825                         btrfs_mark_buffer_dirty(leaf);
4826                 }
4827                 set_extent_uptodate(io_tree, em->start,
4828                                     extent_map_end(em) - 1, GFP_NOFS);
4829                 goto insert;
4830         } else {
4831                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4832                 WARN_ON(1);
4833         }
4834 not_found:
4835         em->start = start;
4836         em->len = len;
4837 not_found_em:
4838         em->block_start = EXTENT_MAP_HOLE;
4839         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4840 insert:
4841         btrfs_release_path(root, path);
4842         if (em->start > start || extent_map_end(em) <= start) {
4843                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4844                        "[%llu %llu]\n", (unsigned long long)em->start,
4845                        (unsigned long long)em->len,
4846                        (unsigned long long)start,
4847                        (unsigned long long)len);
4848                 err = -EIO;
4849                 goto out;
4850         }
4851
4852         err = 0;
4853         write_lock(&em_tree->lock);
4854         ret = add_extent_mapping(em_tree, em);
4855         /* it is possible that someone inserted the extent into the tree
4856          * while we had the lock dropped.  It is also possible that
4857          * an overlapping map exists in the tree
4858          */
4859         if (ret == -EEXIST) {
4860                 struct extent_map *existing;
4861
4862                 ret = 0;
4863
4864                 existing = lookup_extent_mapping(em_tree, start, len);
4865                 if (existing && (existing->start > start ||
4866                     existing->start + existing->len <= start)) {
4867                         free_extent_map(existing);
4868                         existing = NULL;
4869                 }
4870                 if (!existing) {
4871                         existing = lookup_extent_mapping(em_tree, em->start,
4872                                                          em->len);
4873                         if (existing) {
4874                                 err = merge_extent_mapping(em_tree, existing,
4875                                                            em, start,
4876                                                            root->sectorsize);
4877                                 free_extent_map(existing);
4878                                 if (err) {
4879                                         free_extent_map(em);
4880                                         em = NULL;
4881                                 }
4882                         } else {
4883                                 err = -EIO;
4884                                 free_extent_map(em);
4885                                 em = NULL;
4886                         }
4887                 } else {
4888                         free_extent_map(em);
4889                         em = existing;
4890                         err = 0;
4891                 }
4892         }
4893         write_unlock(&em_tree->lock);
4894 out:
4895         if (path)
4896                 btrfs_free_path(path);
4897         if (trans) {
4898                 ret = btrfs_end_transaction(trans, root);
4899                 if (!err)
4900                         err = ret;
4901         }
4902         if (err) {
4903                 free_extent_map(em);
4904                 return ERR_PTR(err);
4905         }
4906         return em;
4907 }
4908
4909 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4910                         const struct iovec *iov, loff_t offset,
4911                         unsigned long nr_segs)
4912 {
4913         return -EINVAL;
4914 }
4915
4916 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4917                 __u64 start, __u64 len)
4918 {
4919         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4920 }
4921
4922 int btrfs_readpage(struct file *file, struct page *page)
4923 {
4924         struct extent_io_tree *tree;
4925         tree = &BTRFS_I(page->mapping->host)->io_tree;
4926         return extent_read_full_page(tree, page, btrfs_get_extent);
4927 }
4928
4929 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4930 {
4931         struct extent_io_tree *tree;
4932
4933
4934         if (current->flags & PF_MEMALLOC) {
4935                 redirty_page_for_writepage(wbc, page);
4936                 unlock_page(page);
4937                 return 0;
4938         }
4939         tree = &BTRFS_I(page->mapping->host)->io_tree;
4940         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4941 }
4942
4943 int btrfs_writepages(struct address_space *mapping,
4944                      struct writeback_control *wbc)
4945 {
4946         struct extent_io_tree *tree;
4947
4948         tree = &BTRFS_I(mapping->host)->io_tree;
4949         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4950 }
4951
4952 static int
4953 btrfs_readpages(struct file *file, struct address_space *mapping,
4954                 struct list_head *pages, unsigned nr_pages)
4955 {
4956         struct extent_io_tree *tree;
4957         tree = &BTRFS_I(mapping->host)->io_tree;
4958         return extent_readpages(tree, mapping, pages, nr_pages,
4959                                 btrfs_get_extent);
4960 }
4961 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4962 {
4963         struct extent_io_tree *tree;
4964         struct extent_map_tree *map;
4965         int ret;
4966
4967         tree = &BTRFS_I(page->mapping->host)->io_tree;
4968         map = &BTRFS_I(page->mapping->host)->extent_tree;
4969         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4970         if (ret == 1) {
4971                 ClearPagePrivate(page);
4972                 set_page_private(page, 0);
4973                 page_cache_release(page);
4974         }
4975         return ret;
4976 }
4977
4978 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4979 {
4980         if (PageWriteback(page) || PageDirty(page))
4981                 return 0;
4982         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4983 }
4984
4985 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4986 {
4987         struct extent_io_tree *tree;
4988         struct btrfs_ordered_extent *ordered;
4989         u64 page_start = page_offset(page);
4990         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4991
4992
4993         /*
4994          * we have the page locked, so new writeback can't start,
4995          * and the dirty bit won't be cleared while we are here.
4996          *
4997          * Wait for IO on this page so that we can safely clear
4998          * the PagePrivate2 bit and do ordered accounting
4999          */
5000         wait_on_page_writeback(page);
5001
5002         tree = &BTRFS_I(page->mapping->host)->io_tree;
5003         if (offset) {
5004                 btrfs_releasepage(page, GFP_NOFS);
5005                 return;
5006         }
5007         lock_extent(tree, page_start, page_end, GFP_NOFS);
5008         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5009                                            page_offset(page));
5010         if (ordered) {
5011                 /*
5012                  * IO on this page will never be started, so we need
5013                  * to account for any ordered extents now
5014                  */
5015                 clear_extent_bit(tree, page_start, page_end,
5016                                  EXTENT_DIRTY | EXTENT_DELALLOC |
5017                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5018                                  NULL, GFP_NOFS);
5019                 /*
5020                  * whoever cleared the private bit is responsible
5021                  * for the finish_ordered_io
5022                  */
5023                 if (TestClearPagePrivate2(page)) {
5024                         btrfs_finish_ordered_io(page->mapping->host,
5025                                                 page_start, page_end);
5026                 }
5027                 btrfs_put_ordered_extent(ordered);
5028                 lock_extent(tree, page_start, page_end, GFP_NOFS);
5029         }
5030         clear_extent_bit(tree, page_start, page_end,
5031                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5032                  EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
5033         __btrfs_releasepage(page, GFP_NOFS);
5034
5035         ClearPageChecked(page);
5036         if (PagePrivate(page)) {
5037                 ClearPagePrivate(page);
5038                 set_page_private(page, 0);
5039                 page_cache_release(page);
5040         }
5041 }
5042
5043 /*
5044  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5045  * called from a page fault handler when a page is first dirtied. Hence we must
5046  * be careful to check for EOF conditions here. We set the page up correctly
5047  * for a written page which means we get ENOSPC checking when writing into
5048  * holes and correct delalloc and unwritten extent mapping on filesystems that
5049  * support these features.
5050  *
5051  * We are not allowed to take the i_mutex here so we have to play games to
5052  * protect against truncate races as the page could now be beyond EOF.  Because
5053  * vmtruncate() writes the inode size before removing pages, once we have the
5054  * page lock we can determine safely if the page is beyond EOF. If it is not
5055  * beyond EOF, then the page is guaranteed safe against truncation until we
5056  * unlock the page.
5057  */
5058 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5059 {
5060         struct page *page = vmf->page;
5061         struct inode *inode = fdentry(vma->vm_file)->d_inode;
5062         struct btrfs_root *root = BTRFS_I(inode)->root;
5063         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5064         struct btrfs_ordered_extent *ordered;
5065         char *kaddr;
5066         unsigned long zero_start;
5067         loff_t size;
5068         int ret;
5069         u64 page_start;
5070         u64 page_end;
5071
5072         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5073         if (ret) {
5074                 if (ret == -ENOMEM)
5075                         ret = VM_FAULT_OOM;
5076                 else /* -ENOSPC, -EIO, etc */
5077                         ret = VM_FAULT_SIGBUS;
5078                 goto out;
5079         }
5080
5081         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5082         if (ret) {
5083                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5084                 ret = VM_FAULT_SIGBUS;
5085                 goto out;
5086         }
5087
5088         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5089 again:
5090         lock_page(page);
5091         size = i_size_read(inode);
5092         page_start = page_offset(page);
5093         page_end = page_start + PAGE_CACHE_SIZE - 1;
5094
5095         if ((page->mapping != inode->i_mapping) ||
5096             (page_start >= size)) {
5097                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5098                 /* page got truncated out from underneath us */
5099                 goto out_unlock;
5100         }
5101         wait_on_page_writeback(page);
5102
5103         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5104         set_page_extent_mapped(page);
5105
5106         /*
5107          * we can't set the delalloc bits if there are pending ordered
5108          * extents.  Drop our locks and wait for them to finish
5109          */
5110         ordered = btrfs_lookup_ordered_extent(inode, page_start);
5111         if (ordered) {
5112                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5113                 unlock_page(page);
5114                 btrfs_start_ordered_extent(inode, ordered, 1);
5115                 btrfs_put_ordered_extent(ordered);
5116                 goto again;
5117         }
5118
5119         /*
5120          * XXX - page_mkwrite gets called every time the page is dirtied, even
5121          * if it was already dirty, so for space accounting reasons we need to
5122          * clear any delalloc bits for the range we are fixing to save.  There
5123          * is probably a better way to do this, but for now keep consistent with
5124          * prepare_pages in the normal write path.
5125          */
5126         clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
5127                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5128                           GFP_NOFS);
5129
5130         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
5131         if (ret) {
5132                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5133                 ret = VM_FAULT_SIGBUS;
5134                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5135                 goto out_unlock;
5136         }
5137         ret = 0;
5138
5139         /* page is wholly or partially inside EOF */
5140         if (page_start + PAGE_CACHE_SIZE > size)
5141                 zero_start = size & ~PAGE_CACHE_MASK;
5142         else
5143                 zero_start = PAGE_CACHE_SIZE;
5144
5145         if (zero_start != PAGE_CACHE_SIZE) {
5146                 kaddr = kmap(page);
5147                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5148                 flush_dcache_page(page);
5149                 kunmap(page);
5150         }
5151         ClearPageChecked(page);
5152         set_page_dirty(page);
5153         SetPageUptodate(page);
5154
5155         BTRFS_I(inode)->last_trans = root->fs_info->generation;
5156         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5157
5158         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5159
5160 out_unlock:
5161         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5162         if (!ret)
5163                 return VM_FAULT_LOCKED;
5164         unlock_page(page);
5165 out:
5166         return ret;
5167 }
5168
5169 static void btrfs_truncate(struct inode *inode)
5170 {
5171         struct btrfs_root *root = BTRFS_I(inode)->root;
5172         int ret;
5173         struct btrfs_trans_handle *trans;
5174         unsigned long nr;
5175         u64 mask = root->sectorsize - 1;
5176
5177         if (!S_ISREG(inode->i_mode)) {
5178                 WARN_ON(1);
5179                 return;
5180         }
5181
5182         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5183         if (ret)
5184                 return;
5185
5186         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5187         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5188
5189         trans = btrfs_start_transaction(root, 1);
5190         btrfs_set_trans_block_group(trans, inode);
5191
5192         /*
5193          * setattr is responsible for setting the ordered_data_close flag,
5194          * but that is only tested during the last file release.  That
5195          * could happen well after the next commit, leaving a great big
5196          * window where new writes may get lost if someone chooses to write
5197          * to this file after truncating to zero
5198          *
5199          * The inode doesn't have any dirty data here, and so if we commit
5200          * this is a noop.  If someone immediately starts writing to the inode
5201          * it is very likely we'll catch some of their writes in this
5202          * transaction, and the commit will find this file on the ordered
5203          * data list with good things to send down.
5204          *
5205          * This is a best effort solution, there is still a window where
5206          * using truncate to replace the contents of the file will
5207          * end up with a zero length file after a crash.
5208          */
5209         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5210                 btrfs_add_ordered_operation(trans, root, inode);
5211
5212         while (1) {
5213                 ret = btrfs_truncate_inode_items(trans, root, inode,
5214                                                  inode->i_size,
5215                                                  BTRFS_EXTENT_DATA_KEY);
5216                 if (ret != -EAGAIN)
5217                         break;
5218
5219                 ret = btrfs_update_inode(trans, root, inode);
5220                 BUG_ON(ret);
5221
5222                 nr = trans->blocks_used;
5223                 btrfs_end_transaction(trans, root);
5224                 btrfs_btree_balance_dirty(root, nr);
5225
5226                 trans = btrfs_start_transaction(root, 1);
5227                 btrfs_set_trans_block_group(trans, inode);
5228         }
5229
5230         if (ret == 0 && inode->i_nlink > 0) {
5231                 ret = btrfs_orphan_del(trans, inode);
5232                 BUG_ON(ret);
5233         }
5234
5235         ret = btrfs_update_inode(trans, root, inode);
5236         BUG_ON(ret);
5237
5238         nr = trans->blocks_used;
5239         ret = btrfs_end_transaction_throttle(trans, root);
5240         BUG_ON(ret);
5241         btrfs_btree_balance_dirty(root, nr);
5242 }
5243
5244 /*
5245  * create a new subvolume directory/inode (helper for the ioctl).
5246  */
5247 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5248                              struct btrfs_root *new_root,
5249                              u64 new_dirid, u64 alloc_hint)
5250 {
5251         struct inode *inode;
5252         int err;
5253         u64 index = 0;
5254
5255         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5256                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5257         if (IS_ERR(inode))
5258                 return PTR_ERR(inode);
5259         inode->i_op = &btrfs_dir_inode_operations;
5260         inode->i_fop = &btrfs_dir_file_operations;
5261
5262         inode->i_nlink = 1;
5263         btrfs_i_size_write(inode, 0);
5264
5265         err = btrfs_update_inode(trans, new_root, inode);
5266         BUG_ON(err);
5267
5268         iput(inode);
5269         return 0;
5270 }
5271
5272 /* helper function for file defrag and space balancing.  This
5273  * forces readahead on a given range of bytes in an inode
5274  */
5275 unsigned long btrfs_force_ra(struct address_space *mapping,
5276                               struct file_ra_state *ra, struct file *file,
5277                               pgoff_t offset, pgoff_t last_index)
5278 {
5279         pgoff_t req_size = last_index - offset + 1;
5280
5281         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5282         return offset + req_size;
5283 }
5284
5285 struct inode *btrfs_alloc_inode(struct super_block *sb)
5286 {
5287         struct btrfs_inode *ei;
5288
5289         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5290         if (!ei)
5291                 return NULL;
5292         ei->last_trans = 0;
5293         ei->last_sub_trans = 0;
5294         ei->logged_trans = 0;
5295         ei->outstanding_extents = 0;
5296         ei->reserved_extents = 0;
5297         ei->root = NULL;
5298         spin_lock_init(&ei->accounting_lock);
5299         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5300         INIT_LIST_HEAD(&ei->i_orphan);
5301         INIT_LIST_HEAD(&ei->ordered_operations);
5302         return &ei->vfs_inode;
5303 }
5304
5305 void btrfs_destroy_inode(struct inode *inode)
5306 {
5307         struct btrfs_ordered_extent *ordered;
5308         struct btrfs_root *root = BTRFS_I(inode)->root;
5309
5310         WARN_ON(!list_empty(&inode->i_dentry));
5311         WARN_ON(inode->i_data.nrpages);
5312
5313         /*
5314          * This can happen where we create an inode, but somebody else also
5315          * created the same inode and we need to destroy the one we already
5316          * created.
5317          */
5318         if (!root)
5319                 goto free;
5320
5321         /*
5322          * Make sure we're properly removed from the ordered operation
5323          * lists.
5324          */
5325         smp_mb();
5326         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5327                 spin_lock(&root->fs_info->ordered_extent_lock);
5328                 list_del_init(&BTRFS_I(inode)->ordered_operations);
5329                 spin_unlock(&root->fs_info->ordered_extent_lock);
5330         }
5331
5332         spin_lock(&root->list_lock);
5333         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5334                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5335                        inode->i_ino);
5336                 list_del_init(&BTRFS_I(inode)->i_orphan);
5337         }
5338         spin_unlock(&root->list_lock);
5339
5340         while (1) {
5341                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5342                 if (!ordered)
5343                         break;
5344                 else {
5345                         printk(KERN_ERR "btrfs found ordered "
5346                                "extent %llu %llu on inode cleanup\n",
5347                                (unsigned long long)ordered->file_offset,
5348                                (unsigned long long)ordered->len);
5349                         btrfs_remove_ordered_extent(inode, ordered);
5350                         btrfs_put_ordered_extent(ordered);
5351                         btrfs_put_ordered_extent(ordered);
5352                 }
5353         }
5354         inode_tree_del(inode);
5355         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5356 free:
5357         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5358 }
5359
5360 void btrfs_drop_inode(struct inode *inode)
5361 {
5362         struct btrfs_root *root = BTRFS_I(inode)->root;
5363
5364         if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5365                 generic_delete_inode(inode);
5366         else
5367                 generic_drop_inode(inode);
5368 }
5369
5370 static void init_once(void *foo)
5371 {
5372         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5373
5374         inode_init_once(&ei->vfs_inode);
5375 }
5376
5377 void btrfs_destroy_cachep(void)
5378 {
5379         if (btrfs_inode_cachep)
5380                 kmem_cache_destroy(btrfs_inode_cachep);
5381         if (btrfs_trans_handle_cachep)
5382                 kmem_cache_destroy(btrfs_trans_handle_cachep);
5383         if (btrfs_transaction_cachep)
5384                 kmem_cache_destroy(btrfs_transaction_cachep);
5385         if (btrfs_path_cachep)
5386                 kmem_cache_destroy(btrfs_path_cachep);
5387 }
5388
5389 int btrfs_init_cachep(void)
5390 {
5391         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5392                         sizeof(struct btrfs_inode), 0,
5393                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5394         if (!btrfs_inode_cachep)
5395                 goto fail;
5396
5397         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5398                         sizeof(struct btrfs_trans_handle), 0,
5399                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5400         if (!btrfs_trans_handle_cachep)
5401                 goto fail;
5402
5403         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5404                         sizeof(struct btrfs_transaction), 0,
5405                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5406         if (!btrfs_transaction_cachep)
5407                 goto fail;
5408
5409         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5410                         sizeof(struct btrfs_path), 0,
5411                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5412         if (!btrfs_path_cachep)
5413                 goto fail;
5414
5415         return 0;
5416 fail:
5417         btrfs_destroy_cachep();
5418         return -ENOMEM;
5419 }
5420
5421 static int btrfs_getattr(struct vfsmount *mnt,
5422                          struct dentry *dentry, struct kstat *stat)
5423 {
5424         struct inode *inode = dentry->d_inode;
5425         generic_fillattr(inode, stat);
5426         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5427         stat->blksize = PAGE_CACHE_SIZE;
5428         stat->blocks = (inode_get_bytes(inode) +
5429                         BTRFS_I(inode)->delalloc_bytes) >> 9;
5430         return 0;
5431 }
5432
5433 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5434                            struct inode *new_dir, struct dentry *new_dentry)
5435 {
5436         struct btrfs_trans_handle *trans;
5437         struct btrfs_root *root = BTRFS_I(old_dir)->root;
5438         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5439         struct inode *new_inode = new_dentry->d_inode;
5440         struct inode *old_inode = old_dentry->d_inode;
5441         struct timespec ctime = CURRENT_TIME;
5442         u64 index = 0;
5443         u64 root_objectid;
5444         int ret;
5445
5446         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5447                 return -EPERM;
5448
5449         /* we only allow rename subvolume link between subvolumes */
5450         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5451                 return -EXDEV;
5452
5453         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5454             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5455                 return -ENOTEMPTY;
5456
5457         if (S_ISDIR(old_inode->i_mode) && new_inode &&
5458             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5459                 return -ENOTEMPTY;
5460
5461         /*
5462          * We want to reserve the absolute worst case amount of items.  So if
5463          * both inodes are subvols and we need to unlink them then that would
5464          * require 4 item modifications, but if they are both normal inodes it
5465          * would require 5 item modifications, so we'll assume their normal
5466          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5467          * should cover the worst case number of items we'll modify.
5468          */
5469         ret = btrfs_reserve_metadata_space(root, 11);
5470         if (ret)
5471                 return ret;
5472
5473         /*
5474          * we're using rename to replace one file with another.
5475          * and the replacement file is large.  Start IO on it now so
5476          * we don't add too much work to the end of the transaction
5477          */
5478         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5479             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5480                 filemap_flush(old_inode->i_mapping);
5481
5482         /* close the racy window with snapshot create/destroy ioctl */
5483         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5484                 down_read(&root->fs_info->subvol_sem);
5485
5486         trans = btrfs_start_transaction(root, 1);
5487         btrfs_set_trans_block_group(trans, new_dir);
5488
5489         if (dest != root)
5490                 btrfs_record_root_in_trans(trans, dest);
5491
5492         ret = btrfs_set_inode_index(new_dir, &index);
5493         if (ret)
5494                 goto out_fail;
5495
5496         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5497                 /* force full log commit if subvolume involved. */
5498                 root->fs_info->last_trans_log_full_commit = trans->transid;
5499         } else {
5500                 ret = btrfs_insert_inode_ref(trans, dest,
5501                                              new_dentry->d_name.name,
5502                                              new_dentry->d_name.len,
5503                                              old_inode->i_ino,
5504                                              new_dir->i_ino, index);
5505                 if (ret)
5506                         goto out_fail;
5507                 /*
5508                  * this is an ugly little race, but the rename is required
5509                  * to make sure that if we crash, the inode is either at the
5510                  * old name or the new one.  pinning the log transaction lets
5511                  * us make sure we don't allow a log commit to come in after
5512                  * we unlink the name but before we add the new name back in.
5513                  */
5514                 btrfs_pin_log_trans(root);
5515         }
5516         /*
5517          * make sure the inode gets flushed if it is replacing
5518          * something.
5519          */
5520         if (new_inode && new_inode->i_size &&
5521             old_inode && S_ISREG(old_inode->i_mode)) {
5522                 btrfs_add_ordered_operation(trans, root, old_inode);
5523         }
5524
5525         old_dir->i_ctime = old_dir->i_mtime = ctime;
5526         new_dir->i_ctime = new_dir->i_mtime = ctime;
5527         old_inode->i_ctime = ctime;
5528
5529         if (old_dentry->d_parent != new_dentry->d_parent)
5530                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5531
5532         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5533                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5534                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5535                                         old_dentry->d_name.name,
5536                                         old_dentry->d_name.len);
5537         } else {
5538                 btrfs_inc_nlink(old_dentry->d_inode);
5539                 ret = btrfs_unlink_inode(trans, root, old_dir,
5540                                          old_dentry->d_inode,
5541                                          old_dentry->d_name.name,
5542                                          old_dentry->d_name.len);
5543         }
5544         BUG_ON(ret);
5545
5546         if (new_inode) {
5547                 new_inode->i_ctime = CURRENT_TIME;
5548                 if (unlikely(new_inode->i_ino ==
5549                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5550                         root_objectid = BTRFS_I(new_inode)->location.objectid;
5551                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
5552                                                 root_objectid,
5553                                                 new_dentry->d_name.name,
5554                                                 new_dentry->d_name.len);
5555                         BUG_ON(new_inode->i_nlink == 0);
5556                 } else {
5557                         ret = btrfs_unlink_inode(trans, dest, new_dir,
5558                                                  new_dentry->d_inode,
5559                                                  new_dentry->d_name.name,
5560                                                  new_dentry->d_name.len);
5561                 }
5562                 BUG_ON(ret);
5563                 if (new_inode->i_nlink == 0) {
5564                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5565                         BUG_ON(ret);
5566                 }
5567         }
5568
5569         ret = btrfs_add_link(trans, new_dir, old_inode,
5570                              new_dentry->d_name.name,
5571                              new_dentry->d_name.len, 0, index);
5572         BUG_ON(ret);
5573
5574         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5575                 btrfs_log_new_name(trans, old_inode, old_dir,
5576                                    new_dentry->d_parent);
5577                 btrfs_end_log_trans(root);
5578         }
5579 out_fail:
5580         btrfs_end_transaction_throttle(trans, root);
5581
5582         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5583                 up_read(&root->fs_info->subvol_sem);
5584
5585         btrfs_unreserve_metadata_space(root, 11);
5586         return ret;
5587 }
5588
5589 /*
5590  * some fairly slow code that needs optimization. This walks the list
5591  * of all the inodes with pending delalloc and forces them to disk.
5592  */
5593 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5594 {
5595         struct list_head *head = &root->fs_info->delalloc_inodes;
5596         struct btrfs_inode *binode;
5597         struct inode *inode;
5598
5599         if (root->fs_info->sb->s_flags & MS_RDONLY)
5600                 return -EROFS;
5601
5602         spin_lock(&root->fs_info->delalloc_lock);
5603         while (!list_empty(head)) {
5604                 binode = list_entry(head->next, struct btrfs_inode,
5605                                     delalloc_inodes);
5606                 inode = igrab(&binode->vfs_inode);
5607                 if (!inode)
5608                         list_del_init(&binode->delalloc_inodes);
5609                 spin_unlock(&root->fs_info->delalloc_lock);
5610                 if (inode) {
5611                         filemap_flush(inode->i_mapping);
5612                         if (delay_iput)
5613                                 btrfs_add_delayed_iput(inode);
5614                         else
5615                                 iput(inode);
5616                 }
5617                 cond_resched();
5618                 spin_lock(&root->fs_info->delalloc_lock);
5619         }
5620         spin_unlock(&root->fs_info->delalloc_lock);
5621
5622         /* the filemap_flush will queue IO into the worker threads, but
5623          * we have to make sure the IO is actually started and that
5624          * ordered extents get created before we return
5625          */
5626         atomic_inc(&root->fs_info->async_submit_draining);
5627         while (atomic_read(&root->fs_info->nr_async_submits) ||
5628               atomic_read(&root->fs_info->async_delalloc_pages)) {
5629                 wait_event(root->fs_info->async_submit_wait,
5630                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5631                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5632         }
5633         atomic_dec(&root->fs_info->async_submit_draining);
5634         return 0;
5635 }
5636
5637 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5638                          const char *symname)
5639 {
5640         struct btrfs_trans_handle *trans;
5641         struct btrfs_root *root = BTRFS_I(dir)->root;
5642         struct btrfs_path *path;
5643         struct btrfs_key key;
5644         struct inode *inode = NULL;
5645         int err;
5646         int drop_inode = 0;
5647         u64 objectid;
5648         u64 index = 0 ;
5649         int name_len;
5650         int datasize;
5651         unsigned long ptr;
5652         struct btrfs_file_extent_item *ei;
5653         struct extent_buffer *leaf;
5654         unsigned long nr = 0;
5655
5656         name_len = strlen(symname) + 1;
5657         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5658                 return -ENAMETOOLONG;
5659
5660         /*
5661          * 2 items for inode item and ref
5662          * 2 items for dir items
5663          * 1 item for xattr if selinux is on
5664          */
5665         err = btrfs_reserve_metadata_space(root, 5);
5666         if (err)
5667                 return err;
5668
5669         trans = btrfs_start_transaction(root, 1);
5670         if (!trans)
5671                 goto out_fail;
5672         btrfs_set_trans_block_group(trans, dir);
5673
5674         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5675         if (err) {
5676                 err = -ENOSPC;
5677                 goto out_unlock;
5678         }
5679
5680         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5681                                 dentry->d_name.len,
5682                                 dentry->d_parent->d_inode->i_ino, objectid,
5683                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5684                                 &index);
5685         err = PTR_ERR(inode);
5686         if (IS_ERR(inode))
5687                 goto out_unlock;
5688
5689         err = btrfs_init_inode_security(trans, inode, dir);
5690         if (err) {
5691                 drop_inode = 1;
5692                 goto out_unlock;
5693         }
5694
5695         btrfs_set_trans_block_group(trans, inode);
5696         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5697         if (err)
5698                 drop_inode = 1;
5699         else {
5700                 inode->i_mapping->a_ops = &btrfs_aops;
5701                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5702                 inode->i_fop = &btrfs_file_operations;
5703                 inode->i_op = &btrfs_file_inode_operations;
5704                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;