Btrfs: cache ordered extent when completing io
[linux-3.10.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include "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, &ordered_extent, start,
1702                                              end - start + 1);
1703         if (!ret)
1704                 return 0;
1705         BUG_ON(!ordered_extent);
1706
1707         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1708                 BUG_ON(!list_empty(&ordered_extent->list));
1709                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1710                 if (!ret) {
1711                         trans = btrfs_join_transaction(root, 1);
1712                         ret = btrfs_update_inode(trans, root, inode);
1713                         BUG_ON(ret);
1714                         btrfs_end_transaction(trans, root);
1715                 }
1716                 goto out;
1717         }
1718
1719         lock_extent(io_tree, ordered_extent->file_offset,
1720                     ordered_extent->file_offset + ordered_extent->len - 1,
1721                     GFP_NOFS);
1722
1723         trans = btrfs_join_transaction(root, 1);
1724
1725         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1726                 compressed = 1;
1727         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1728                 BUG_ON(compressed);
1729                 ret = btrfs_mark_extent_written(trans, inode,
1730                                                 ordered_extent->file_offset,
1731                                                 ordered_extent->file_offset +
1732                                                 ordered_extent->len);
1733                 BUG_ON(ret);
1734         } else {
1735                 ret = insert_reserved_file_extent(trans, inode,
1736                                                 ordered_extent->file_offset,
1737                                                 ordered_extent->start,
1738                                                 ordered_extent->disk_len,
1739                                                 ordered_extent->len,
1740                                                 ordered_extent->len,
1741                                                 compressed, 0, 0,
1742                                                 BTRFS_FILE_EXTENT_REG);
1743                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1744                                    ordered_extent->file_offset,
1745                                    ordered_extent->len);
1746                 BUG_ON(ret);
1747         }
1748         unlock_extent(io_tree, ordered_extent->file_offset,
1749                     ordered_extent->file_offset + ordered_extent->len - 1,
1750                     GFP_NOFS);
1751         add_pending_csums(trans, inode, ordered_extent->file_offset,
1752                           &ordered_extent->list);
1753
1754         /* this also removes the ordered extent from the tree */
1755         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1756         ret = btrfs_update_inode(trans, root, inode);
1757         BUG_ON(ret);
1758         btrfs_end_transaction(trans, root);
1759 out:
1760         /* once for us */
1761         btrfs_put_ordered_extent(ordered_extent);
1762         /* once for the tree */
1763         btrfs_put_ordered_extent(ordered_extent);
1764
1765         return 0;
1766 }
1767
1768 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1769                                 struct extent_state *state, int uptodate)
1770 {
1771         ClearPagePrivate2(page);
1772         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1773 }
1774
1775 /*
1776  * When IO fails, either with EIO or csum verification fails, we
1777  * try other mirrors that might have a good copy of the data.  This
1778  * io_failure_record is used to record state as we go through all the
1779  * mirrors.  If another mirror has good data, the page is set up to date
1780  * and things continue.  If a good mirror can't be found, the original
1781  * bio end_io callback is called to indicate things have failed.
1782  */
1783 struct io_failure_record {
1784         struct page *page;
1785         u64 start;
1786         u64 len;
1787         u64 logical;
1788         unsigned long bio_flags;
1789         int last_mirror;
1790 };
1791
1792 static int btrfs_io_failed_hook(struct bio *failed_bio,
1793                          struct page *page, u64 start, u64 end,
1794                          struct extent_state *state)
1795 {
1796         struct io_failure_record *failrec = NULL;
1797         u64 private;
1798         struct extent_map *em;
1799         struct inode *inode = page->mapping->host;
1800         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1801         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1802         struct bio *bio;
1803         int num_copies;
1804         int ret;
1805         int rw;
1806         u64 logical;
1807
1808         ret = get_state_private(failure_tree, start, &private);
1809         if (ret) {
1810                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1811                 if (!failrec)
1812                         return -ENOMEM;
1813                 failrec->start = start;
1814                 failrec->len = end - start + 1;
1815                 failrec->last_mirror = 0;
1816                 failrec->bio_flags = 0;
1817
1818                 read_lock(&em_tree->lock);
1819                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1820                 if (em->start > start || em->start + em->len < start) {
1821                         free_extent_map(em);
1822                         em = NULL;
1823                 }
1824                 read_unlock(&em_tree->lock);
1825
1826                 if (!em || IS_ERR(em)) {
1827                         kfree(failrec);
1828                         return -EIO;
1829                 }
1830                 logical = start - em->start;
1831                 logical = em->block_start + logical;
1832                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1833                         logical = em->block_start;
1834                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1835                 }
1836                 failrec->logical = logical;
1837                 free_extent_map(em);
1838                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1839                                 EXTENT_DIRTY, GFP_NOFS);
1840                 set_state_private(failure_tree, start,
1841                                  (u64)(unsigned long)failrec);
1842         } else {
1843                 failrec = (struct io_failure_record *)(unsigned long)private;
1844         }
1845         num_copies = btrfs_num_copies(
1846                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1847                               failrec->logical, failrec->len);
1848         failrec->last_mirror++;
1849         if (!state) {
1850                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1851                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1852                                                     failrec->start,
1853                                                     EXTENT_LOCKED);
1854                 if (state && state->start != failrec->start)
1855                         state = NULL;
1856                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1857         }
1858         if (!state || failrec->last_mirror > num_copies) {
1859                 set_state_private(failure_tree, failrec->start, 0);
1860                 clear_extent_bits(failure_tree, failrec->start,
1861                                   failrec->start + failrec->len - 1,
1862                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1863                 kfree(failrec);
1864                 return -EIO;
1865         }
1866         bio = bio_alloc(GFP_NOFS, 1);
1867         bio->bi_private = state;
1868         bio->bi_end_io = failed_bio->bi_end_io;
1869         bio->bi_sector = failrec->logical >> 9;
1870         bio->bi_bdev = failed_bio->bi_bdev;
1871         bio->bi_size = 0;
1872
1873         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1874         if (failed_bio->bi_rw & (1 << BIO_RW))
1875                 rw = WRITE;
1876         else
1877                 rw = READ;
1878
1879         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1880                                                       failrec->last_mirror,
1881                                                       failrec->bio_flags);
1882         return 0;
1883 }
1884
1885 /*
1886  * each time an IO finishes, we do a fast check in the IO failure tree
1887  * to see if we need to process or clean up an io_failure_record
1888  */
1889 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1890 {
1891         u64 private;
1892         u64 private_failure;
1893         struct io_failure_record *failure;
1894         int ret;
1895
1896         private = 0;
1897         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1898                              (u64)-1, 1, EXTENT_DIRTY)) {
1899                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1900                                         start, &private_failure);
1901                 if (ret == 0) {
1902                         failure = (struct io_failure_record *)(unsigned long)
1903                                    private_failure;
1904                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1905                                           failure->start, 0);
1906                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1907                                           failure->start,
1908                                           failure->start + failure->len - 1,
1909                                           EXTENT_DIRTY | EXTENT_LOCKED,
1910                                           GFP_NOFS);
1911                         kfree(failure);
1912                 }
1913         }
1914         return 0;
1915 }
1916
1917 /*
1918  * when reads are done, we need to check csums to verify the data is correct
1919  * if there's a match, we allow the bio to finish.  If not, we go through
1920  * the io_failure_record routines to find good copies
1921  */
1922 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1923                                struct extent_state *state)
1924 {
1925         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1926         struct inode *inode = page->mapping->host;
1927         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1928         char *kaddr;
1929         u64 private = ~(u32)0;
1930         int ret;
1931         struct btrfs_root *root = BTRFS_I(inode)->root;
1932         u32 csum = ~(u32)0;
1933
1934         if (PageChecked(page)) {
1935                 ClearPageChecked(page);
1936                 goto good;
1937         }
1938
1939         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1940                 return 0;
1941
1942         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1943             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1944                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1945                                   GFP_NOFS);
1946                 return 0;
1947         }
1948
1949         if (state && state->start == start) {
1950                 private = state->private;
1951                 ret = 0;
1952         } else {
1953                 ret = get_state_private(io_tree, start, &private);
1954         }
1955         kaddr = kmap_atomic(page, KM_USER0);
1956         if (ret)
1957                 goto zeroit;
1958
1959         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1960         btrfs_csum_final(csum, (char *)&csum);
1961         if (csum != private)
1962                 goto zeroit;
1963
1964         kunmap_atomic(kaddr, KM_USER0);
1965 good:
1966         /* if the io failure tree for this inode is non-empty,
1967          * check to see if we've recovered from a failed IO
1968          */
1969         btrfs_clean_io_failures(inode, start);
1970         return 0;
1971
1972 zeroit:
1973         if (printk_ratelimit()) {
1974                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1975                        "private %llu\n", page->mapping->host->i_ino,
1976                        (unsigned long long)start, csum,
1977                        (unsigned long long)private);
1978         }
1979         memset(kaddr + offset, 1, end - start + 1);
1980         flush_dcache_page(page);
1981         kunmap_atomic(kaddr, KM_USER0);
1982         if (private == 0)
1983                 return 0;
1984         return -EIO;
1985 }
1986
1987 struct delayed_iput {
1988         struct list_head list;
1989         struct inode *inode;
1990 };
1991
1992 void btrfs_add_delayed_iput(struct inode *inode)
1993 {
1994         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1995         struct delayed_iput *delayed;
1996
1997         if (atomic_add_unless(&inode->i_count, -1, 1))
1998                 return;
1999
2000         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2001         delayed->inode = inode;
2002
2003         spin_lock(&fs_info->delayed_iput_lock);
2004         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2005         spin_unlock(&fs_info->delayed_iput_lock);
2006 }
2007
2008 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2009 {
2010         LIST_HEAD(list);
2011         struct btrfs_fs_info *fs_info = root->fs_info;
2012         struct delayed_iput *delayed;
2013         int empty;
2014
2015         spin_lock(&fs_info->delayed_iput_lock);
2016         empty = list_empty(&fs_info->delayed_iputs);
2017         spin_unlock(&fs_info->delayed_iput_lock);
2018         if (empty)
2019                 return;
2020
2021         down_read(&root->fs_info->cleanup_work_sem);
2022         spin_lock(&fs_info->delayed_iput_lock);
2023         list_splice_init(&fs_info->delayed_iputs, &list);
2024         spin_unlock(&fs_info->delayed_iput_lock);
2025
2026         while (!list_empty(&list)) {
2027                 delayed = list_entry(list.next, struct delayed_iput, list);
2028                 list_del(&delayed->list);
2029                 iput(delayed->inode);
2030                 kfree(delayed);
2031         }
2032         up_read(&root->fs_info->cleanup_work_sem);
2033 }
2034
2035 /*
2036  * This creates an orphan entry for the given inode in case something goes
2037  * wrong in the middle of an unlink/truncate.
2038  */
2039 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2040 {
2041         struct btrfs_root *root = BTRFS_I(inode)->root;
2042         int ret = 0;
2043
2044         spin_lock(&root->list_lock);
2045
2046         /* already on the orphan list, we're good */
2047         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2048                 spin_unlock(&root->list_lock);
2049                 return 0;
2050         }
2051
2052         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2053
2054         spin_unlock(&root->list_lock);
2055
2056         /*
2057          * insert an orphan item to track this unlinked/truncated file
2058          */
2059         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2060
2061         return ret;
2062 }
2063
2064 /*
2065  * We have done the truncate/delete so we can go ahead and remove the orphan
2066  * item for this particular inode.
2067  */
2068 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2069 {
2070         struct btrfs_root *root = BTRFS_I(inode)->root;
2071         int ret = 0;
2072
2073         spin_lock(&root->list_lock);
2074
2075         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2076                 spin_unlock(&root->list_lock);
2077                 return 0;
2078         }
2079
2080         list_del_init(&BTRFS_I(inode)->i_orphan);
2081         if (!trans) {
2082                 spin_unlock(&root->list_lock);
2083                 return 0;
2084         }
2085
2086         spin_unlock(&root->list_lock);
2087
2088         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2089
2090         return ret;
2091 }
2092
2093 /*
2094  * this cleans up any orphans that may be left on the list from the last use
2095  * of this root.
2096  */
2097 void btrfs_orphan_cleanup(struct btrfs_root *root)
2098 {
2099         struct btrfs_path *path;
2100         struct extent_buffer *leaf;
2101         struct btrfs_item *item;
2102         struct btrfs_key key, found_key;
2103         struct btrfs_trans_handle *trans;
2104         struct inode *inode;
2105         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2106
2107         if (!xchg(&root->clean_orphans, 0))
2108                 return;
2109
2110         path = btrfs_alloc_path();
2111         BUG_ON(!path);
2112         path->reada = -1;
2113
2114         key.objectid = BTRFS_ORPHAN_OBJECTID;
2115         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2116         key.offset = (u64)-1;
2117
2118         while (1) {
2119                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2120                 if (ret < 0) {
2121                         printk(KERN_ERR "Error searching slot for orphan: %d"
2122                                "\n", ret);
2123                         break;
2124                 }
2125
2126                 /*
2127                  * if ret == 0 means we found what we were searching for, which
2128                  * is weird, but possible, so only screw with path if we didnt
2129                  * find the key and see if we have stuff that matches
2130                  */
2131                 if (ret > 0) {
2132                         if (path->slots[0] == 0)
2133                                 break;
2134                         path->slots[0]--;
2135                 }
2136
2137                 /* pull out the item */
2138                 leaf = path->nodes[0];
2139                 item = btrfs_item_nr(leaf, path->slots[0]);
2140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2141
2142                 /* make sure the item matches what we want */
2143                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2144                         break;
2145                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2146                         break;
2147
2148                 /* release the path since we're done with it */
2149                 btrfs_release_path(root, path);
2150
2151                 /*
2152                  * this is where we are basically btrfs_lookup, without the
2153                  * crossing root thing.  we store the inode number in the
2154                  * offset of the orphan item.
2155                  */
2156                 found_key.objectid = found_key.offset;
2157                 found_key.type = BTRFS_INODE_ITEM_KEY;
2158                 found_key.offset = 0;
2159                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2160                 if (IS_ERR(inode))
2161                         break;
2162
2163                 /*
2164                  * add this inode to the orphan list so btrfs_orphan_del does
2165                  * the proper thing when we hit it
2166                  */
2167                 spin_lock(&root->list_lock);
2168                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2169                 spin_unlock(&root->list_lock);
2170
2171                 /*
2172                  * if this is a bad inode, means we actually succeeded in
2173                  * removing the inode, but not the orphan record, which means
2174                  * we need to manually delete the orphan since iput will just
2175                  * do a destroy_inode
2176                  */
2177                 if (is_bad_inode(inode)) {
2178                         trans = btrfs_start_transaction(root, 1);
2179                         btrfs_orphan_del(trans, inode);
2180                         btrfs_end_transaction(trans, root);
2181                         iput(inode);
2182                         continue;
2183                 }
2184
2185                 /* if we have links, this was a truncate, lets do that */
2186                 if (inode->i_nlink) {
2187                         nr_truncate++;
2188                         btrfs_truncate(inode);
2189                 } else {
2190                         nr_unlink++;
2191                 }
2192
2193                 /* this will do delete_inode and everything for us */
2194                 iput(inode);
2195         }
2196
2197         if (nr_unlink)
2198                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2199         if (nr_truncate)
2200                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2201
2202         btrfs_free_path(path);
2203 }
2204
2205 /*
2206  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2207  * don't find any xattrs, we know there can't be any acls.
2208  *
2209  * slot is the slot the inode is in, objectid is the objectid of the inode
2210  */
2211 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2212                                           int slot, u64 objectid)
2213 {
2214         u32 nritems = btrfs_header_nritems(leaf);
2215         struct btrfs_key found_key;
2216         int scanned = 0;
2217
2218         slot++;
2219         while (slot < nritems) {
2220                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2221
2222                 /* we found a different objectid, there must not be acls */
2223                 if (found_key.objectid != objectid)
2224                         return 0;
2225
2226                 /* we found an xattr, assume we've got an acl */
2227                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2228                         return 1;
2229
2230                 /*
2231                  * we found a key greater than an xattr key, there can't
2232                  * be any acls later on
2233                  */
2234                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2235                         return 0;
2236
2237                 slot++;
2238                 scanned++;
2239
2240                 /*
2241                  * it goes inode, inode backrefs, xattrs, extents,
2242                  * so if there are a ton of hard links to an inode there can
2243                  * be a lot of backrefs.  Don't waste time searching too hard,
2244                  * this is just an optimization
2245                  */
2246                 if (scanned >= 8)
2247                         break;
2248         }
2249         /* we hit the end of the leaf before we found an xattr or
2250          * something larger than an xattr.  We have to assume the inode
2251          * has acls
2252          */
2253         return 1;
2254 }
2255
2256 /*
2257  * read an inode from the btree into the in-memory inode
2258  */
2259 static void btrfs_read_locked_inode(struct inode *inode)
2260 {
2261         struct btrfs_path *path;
2262         struct extent_buffer *leaf;
2263         struct btrfs_inode_item *inode_item;
2264         struct btrfs_timespec *tspec;
2265         struct btrfs_root *root = BTRFS_I(inode)->root;
2266         struct btrfs_key location;
2267         int maybe_acls;
2268         u64 alloc_group_block;
2269         u32 rdev;
2270         int ret;
2271
2272         path = btrfs_alloc_path();
2273         BUG_ON(!path);
2274         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2275
2276         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2277         if (ret)
2278                 goto make_bad;
2279
2280         leaf = path->nodes[0];
2281         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2282                                     struct btrfs_inode_item);
2283
2284         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2285         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2286         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2287         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2288         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2289
2290         tspec = btrfs_inode_atime(inode_item);
2291         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2292         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2293
2294         tspec = btrfs_inode_mtime(inode_item);
2295         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2296         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2297
2298         tspec = btrfs_inode_ctime(inode_item);
2299         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2300         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2301
2302         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2303         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2304         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2305         inode->i_generation = BTRFS_I(inode)->generation;
2306         inode->i_rdev = 0;
2307         rdev = btrfs_inode_rdev(leaf, inode_item);
2308
2309         BTRFS_I(inode)->index_cnt = (u64)-1;
2310         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2311
2312         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2313
2314         /*
2315          * try to precache a NULL acl entry for files that don't have
2316          * any xattrs or acls
2317          */
2318         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2319         if (!maybe_acls)
2320                 cache_no_acl(inode);
2321
2322         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2323                                                 alloc_group_block, 0);
2324         btrfs_free_path(path);
2325         inode_item = NULL;
2326
2327         switch (inode->i_mode & S_IFMT) {
2328         case S_IFREG:
2329                 inode->i_mapping->a_ops = &btrfs_aops;
2330                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2331                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2332                 inode->i_fop = &btrfs_file_operations;
2333                 inode->i_op = &btrfs_file_inode_operations;
2334                 break;
2335         case S_IFDIR:
2336                 inode->i_fop = &btrfs_dir_file_operations;
2337                 if (root == root->fs_info->tree_root)
2338                         inode->i_op = &btrfs_dir_ro_inode_operations;
2339                 else
2340                         inode->i_op = &btrfs_dir_inode_operations;
2341                 break;
2342         case S_IFLNK:
2343                 inode->i_op = &btrfs_symlink_inode_operations;
2344                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2345                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2346                 break;
2347         default:
2348                 inode->i_op = &btrfs_special_inode_operations;
2349                 init_special_inode(inode, inode->i_mode, rdev);
2350                 break;
2351         }
2352
2353         btrfs_update_iflags(inode);
2354         return;
2355
2356 make_bad:
2357         btrfs_free_path(path);
2358         make_bad_inode(inode);
2359 }
2360
2361 /*
2362  * given a leaf and an inode, copy the inode fields into the leaf
2363  */
2364 static void fill_inode_item(struct btrfs_trans_handle *trans,
2365                             struct extent_buffer *leaf,
2366                             struct btrfs_inode_item *item,
2367                             struct inode *inode)
2368 {
2369         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2370         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2371         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2372         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2373         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2374
2375         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2376                                inode->i_atime.tv_sec);
2377         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2378                                 inode->i_atime.tv_nsec);
2379
2380         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2381                                inode->i_mtime.tv_sec);
2382         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2383                                 inode->i_mtime.tv_nsec);
2384
2385         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2386                                inode->i_ctime.tv_sec);
2387         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2388                                 inode->i_ctime.tv_nsec);
2389
2390         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2391         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2392         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2393         btrfs_set_inode_transid(leaf, item, trans->transid);
2394         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2395         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2396         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2397 }
2398
2399 /*
2400  * copy everything in the in-memory inode into the btree.
2401  */
2402 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2403                                 struct btrfs_root *root, struct inode *inode)
2404 {
2405         struct btrfs_inode_item *inode_item;
2406         struct btrfs_path *path;
2407         struct extent_buffer *leaf;
2408         int ret;
2409
2410         path = btrfs_alloc_path();
2411         BUG_ON(!path);
2412         path->leave_spinning = 1;
2413         ret = btrfs_lookup_inode(trans, root, path,
2414                                  &BTRFS_I(inode)->location, 1);
2415         if (ret) {
2416                 if (ret > 0)
2417                         ret = -ENOENT;
2418                 goto failed;
2419         }
2420
2421         btrfs_unlock_up_safe(path, 1);
2422         leaf = path->nodes[0];
2423         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2424                                   struct btrfs_inode_item);
2425
2426         fill_inode_item(trans, leaf, inode_item, inode);
2427         btrfs_mark_buffer_dirty(leaf);
2428         btrfs_set_inode_last_trans(trans, inode);
2429         ret = 0;
2430 failed:
2431         btrfs_free_path(path);
2432         return ret;
2433 }
2434
2435
2436 /*
2437  * unlink helper that gets used here in inode.c and in the tree logging
2438  * recovery code.  It remove a link in a directory with a given name, and
2439  * also drops the back refs in the inode to the directory
2440  */
2441 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2442                        struct btrfs_root *root,
2443                        struct inode *dir, struct inode *inode,
2444                        const char *name, int name_len)
2445 {
2446         struct btrfs_path *path;
2447         int ret = 0;
2448         struct extent_buffer *leaf;
2449         struct btrfs_dir_item *di;
2450         struct btrfs_key key;
2451         u64 index;
2452
2453         path = btrfs_alloc_path();
2454         if (!path) {
2455                 ret = -ENOMEM;
2456                 goto err;
2457         }
2458
2459         path->leave_spinning = 1;
2460         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2461                                     name, name_len, -1);
2462         if (IS_ERR(di)) {
2463                 ret = PTR_ERR(di);
2464                 goto err;
2465         }
2466         if (!di) {
2467                 ret = -ENOENT;
2468                 goto err;
2469         }
2470         leaf = path->nodes[0];
2471         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2472         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2473         if (ret)
2474                 goto err;
2475         btrfs_release_path(root, path);
2476
2477         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2478                                   inode->i_ino,
2479                                   dir->i_ino, &index);
2480         if (ret) {
2481                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2482                        "inode %lu parent %lu\n", name_len, name,
2483                        inode->i_ino, dir->i_ino);
2484                 goto err;
2485         }
2486
2487         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2488                                          index, name, name_len, -1);
2489         if (IS_ERR(di)) {
2490                 ret = PTR_ERR(di);
2491                 goto err;
2492         }
2493         if (!di) {
2494                 ret = -ENOENT;
2495                 goto err;
2496         }
2497         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2498         btrfs_release_path(root, path);
2499
2500         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2501                                          inode, dir->i_ino);
2502         BUG_ON(ret != 0 && ret != -ENOENT);
2503
2504         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2505                                            dir, index);
2506         BUG_ON(ret);
2507 err:
2508         btrfs_free_path(path);
2509         if (ret)
2510                 goto out;
2511
2512         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2513         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2514         btrfs_update_inode(trans, root, dir);
2515         btrfs_drop_nlink(inode);
2516         ret = btrfs_update_inode(trans, root, inode);
2517 out:
2518         return ret;
2519 }
2520
2521 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2522 {
2523         struct btrfs_root *root;
2524         struct btrfs_trans_handle *trans;
2525         struct inode *inode = dentry->d_inode;
2526         int ret;
2527         unsigned long nr = 0;
2528
2529         root = BTRFS_I(dir)->root;
2530
2531         /*
2532          * 5 items for unlink inode
2533          * 1 for orphan
2534          */
2535         ret = btrfs_reserve_metadata_space(root, 6);
2536         if (ret)
2537                 return ret;
2538
2539         trans = btrfs_start_transaction(root, 1);
2540         if (IS_ERR(trans)) {
2541                 btrfs_unreserve_metadata_space(root, 6);
2542                 return PTR_ERR(trans);
2543         }
2544
2545         btrfs_set_trans_block_group(trans, dir);
2546
2547         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2548
2549         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2550                                  dentry->d_name.name, dentry->d_name.len);
2551
2552         if (inode->i_nlink == 0)
2553                 ret = btrfs_orphan_add(trans, inode);
2554
2555         nr = trans->blocks_used;
2556
2557         btrfs_end_transaction_throttle(trans, root);
2558         btrfs_unreserve_metadata_space(root, 6);
2559         btrfs_btree_balance_dirty(root, nr);
2560         return ret;
2561 }
2562
2563 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2564                         struct btrfs_root *root,
2565                         struct inode *dir, u64 objectid,
2566                         const char *name, int name_len)
2567 {
2568         struct btrfs_path *path;
2569         struct extent_buffer *leaf;
2570         struct btrfs_dir_item *di;
2571         struct btrfs_key key;
2572         u64 index;
2573         int ret;
2574
2575         path = btrfs_alloc_path();
2576         if (!path)
2577                 return -ENOMEM;
2578
2579         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2580                                    name, name_len, -1);
2581         BUG_ON(!di || IS_ERR(di));
2582
2583         leaf = path->nodes[0];
2584         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2585         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2586         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2587         BUG_ON(ret);
2588         btrfs_release_path(root, path);
2589
2590         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2591                                  objectid, root->root_key.objectid,
2592                                  dir->i_ino, &index, name, name_len);
2593         if (ret < 0) {
2594                 BUG_ON(ret != -ENOENT);
2595                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2596                                                  name, name_len);
2597                 BUG_ON(!di || IS_ERR(di));
2598
2599                 leaf = path->nodes[0];
2600                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2601                 btrfs_release_path(root, path);
2602                 index = key.offset;
2603         }
2604
2605         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2606                                          index, name, name_len, -1);
2607         BUG_ON(!di || IS_ERR(di));
2608
2609         leaf = path->nodes[0];
2610         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2611         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2612         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2613         BUG_ON(ret);
2614         btrfs_release_path(root, path);
2615
2616         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2617         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2618         ret = btrfs_update_inode(trans, root, dir);
2619         BUG_ON(ret);
2620         dir->i_sb->s_dirt = 1;
2621
2622         btrfs_free_path(path);
2623         return 0;
2624 }
2625
2626 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2627 {
2628         struct inode *inode = dentry->d_inode;
2629         int err = 0;
2630         int ret;
2631         struct btrfs_root *root = BTRFS_I(dir)->root;
2632         struct btrfs_trans_handle *trans;
2633         unsigned long nr = 0;
2634
2635         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2636             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2637                 return -ENOTEMPTY;
2638
2639         ret = btrfs_reserve_metadata_space(root, 5);
2640         if (ret)
2641                 return ret;
2642
2643         trans = btrfs_start_transaction(root, 1);
2644         if (IS_ERR(trans)) {
2645                 btrfs_unreserve_metadata_space(root, 5);
2646                 return PTR_ERR(trans);
2647         }
2648
2649         btrfs_set_trans_block_group(trans, dir);
2650
2651         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2652                 err = btrfs_unlink_subvol(trans, root, dir,
2653                                           BTRFS_I(inode)->location.objectid,
2654                                           dentry->d_name.name,
2655                                           dentry->d_name.len);
2656                 goto out;
2657         }
2658
2659         err = btrfs_orphan_add(trans, inode);
2660         if (err)
2661                 goto out;
2662
2663         /* now the directory is empty */
2664         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2665                                  dentry->d_name.name, dentry->d_name.len);
2666         if (!err)
2667                 btrfs_i_size_write(inode, 0);
2668 out:
2669         nr = trans->blocks_used;
2670         ret = btrfs_end_transaction_throttle(trans, root);
2671         btrfs_unreserve_metadata_space(root, 5);
2672         btrfs_btree_balance_dirty(root, nr);
2673
2674         if (ret && !err)
2675                 err = ret;
2676         return err;
2677 }
2678
2679 #if 0
2680 /*
2681  * when truncating bytes in a file, it is possible to avoid reading
2682  * the leaves that contain only checksum items.  This can be the
2683  * majority of the IO required to delete a large file, but it must
2684  * be done carefully.
2685  *
2686  * The keys in the level just above the leaves are checked to make sure
2687  * the lowest key in a given leaf is a csum key, and starts at an offset
2688  * after the new  size.
2689  *
2690  * Then the key for the next leaf is checked to make sure it also has
2691  * a checksum item for the same file.  If it does, we know our target leaf
2692  * contains only checksum items, and it can be safely freed without reading
2693  * it.
2694  *
2695  * This is just an optimization targeted at large files.  It may do
2696  * nothing.  It will return 0 unless things went badly.
2697  */
2698 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2699                                      struct btrfs_root *root,
2700                                      struct btrfs_path *path,
2701                                      struct inode *inode, u64 new_size)
2702 {
2703         struct btrfs_key key;
2704         int ret;
2705         int nritems;
2706         struct btrfs_key found_key;
2707         struct btrfs_key other_key;
2708         struct btrfs_leaf_ref *ref;
2709         u64 leaf_gen;
2710         u64 leaf_start;
2711
2712         path->lowest_level = 1;
2713         key.objectid = inode->i_ino;
2714         key.type = BTRFS_CSUM_ITEM_KEY;
2715         key.offset = new_size;
2716 again:
2717         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2718         if (ret < 0)
2719                 goto out;
2720
2721         if (path->nodes[1] == NULL) {
2722                 ret = 0;
2723                 goto out;
2724         }
2725         ret = 0;
2726         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2727         nritems = btrfs_header_nritems(path->nodes[1]);
2728
2729         if (!nritems)
2730                 goto out;
2731
2732         if (path->slots[1] >= nritems)
2733                 goto next_node;
2734
2735         /* did we find a key greater than anything we want to delete? */
2736         if (found_key.objectid > inode->i_ino ||
2737            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2738                 goto out;
2739
2740         /* we check the next key in the node to make sure the leave contains
2741          * only checksum items.  This comparison doesn't work if our
2742          * leaf is the last one in the node
2743          */
2744         if (path->slots[1] + 1 >= nritems) {
2745 next_node:
2746                 /* search forward from the last key in the node, this
2747                  * will bring us into the next node in the tree
2748                  */
2749                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2750
2751                 /* unlikely, but we inc below, so check to be safe */
2752                 if (found_key.offset == (u64)-1)
2753                         goto out;
2754
2755                 /* search_forward needs a path with locks held, do the
2756                  * search again for the original key.  It is possible
2757                  * this will race with a balance and return a path that
2758                  * we could modify, but this drop is just an optimization
2759                  * and is allowed to miss some leaves.
2760                  */
2761                 btrfs_release_path(root, path);
2762                 found_key.offset++;
2763
2764                 /* setup a max key for search_forward */
2765                 other_key.offset = (u64)-1;
2766                 other_key.type = key.type;
2767                 other_key.objectid = key.objectid;
2768
2769                 path->keep_locks = 1;
2770                 ret = btrfs_search_forward(root, &found_key, &other_key,
2771                                            path, 0, 0);
2772                 path->keep_locks = 0;
2773                 if (ret || found_key.objectid != key.objectid ||
2774                     found_key.type != key.type) {
2775                         ret = 0;
2776                         goto out;
2777                 }
2778
2779                 key.offset = found_key.offset;
2780                 btrfs_release_path(root, path);
2781                 cond_resched();
2782                 goto again;
2783         }
2784
2785         /* we know there's one more slot after us in the tree,
2786          * read that key so we can verify it is also a checksum item
2787          */
2788         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2789
2790         if (found_key.objectid < inode->i_ino)
2791                 goto next_key;
2792
2793         if (found_key.type != key.type || found_key.offset < new_size)
2794                 goto next_key;
2795
2796         /*
2797          * if the key for the next leaf isn't a csum key from this objectid,
2798          * we can't be sure there aren't good items inside this leaf.
2799          * Bail out
2800          */
2801         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2802                 goto out;
2803
2804         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2805         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2806         /*
2807          * it is safe to delete this leaf, it contains only
2808          * csum items from this inode at an offset >= new_size
2809          */
2810         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2811         BUG_ON(ret);
2812
2813         if (root->ref_cows && leaf_gen < trans->transid) {
2814                 ref = btrfs_alloc_leaf_ref(root, 0);
2815                 if (ref) {
2816                         ref->root_gen = root->root_key.offset;
2817                         ref->bytenr = leaf_start;
2818                         ref->owner = 0;
2819                         ref->generation = leaf_gen;
2820                         ref->nritems = 0;
2821
2822                         btrfs_sort_leaf_ref(ref);
2823
2824                         ret = btrfs_add_leaf_ref(root, ref, 0);
2825                         WARN_ON(ret);
2826                         btrfs_free_leaf_ref(root, ref);
2827                 } else {
2828                         WARN_ON(1);
2829                 }
2830         }
2831 next_key:
2832         btrfs_release_path(root, path);
2833
2834         if (other_key.objectid == inode->i_ino &&
2835             other_key.type == key.type && other_key.offset > key.offset) {
2836                 key.offset = other_key.offset;
2837                 cond_resched();
2838                 goto again;
2839         }
2840         ret = 0;
2841 out:
2842         /* fixup any changes we've made to the path */
2843         path->lowest_level = 0;
2844         path->keep_locks = 0;
2845         btrfs_release_path(root, path);
2846         return ret;
2847 }
2848
2849 #endif
2850
2851 /*
2852  * this can truncate away extent items, csum items and directory items.
2853  * It starts at a high offset and removes keys until it can't find
2854  * any higher than new_size
2855  *
2856  * csum items that cross the new i_size are truncated to the new size
2857  * as well.
2858  *
2859  * min_type is the minimum key type to truncate down to.  If set to 0, this
2860  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2861  */
2862 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2863                                struct btrfs_root *root,
2864                                struct inode *inode,
2865                                u64 new_size, u32 min_type)
2866 {
2867         struct btrfs_path *path;
2868         struct extent_buffer *leaf;
2869         struct btrfs_file_extent_item *fi;
2870         struct btrfs_key key;
2871         struct btrfs_key found_key;
2872         u64 extent_start = 0;
2873         u64 extent_num_bytes = 0;
2874         u64 extent_offset = 0;
2875         u64 item_end = 0;
2876         u64 mask = root->sectorsize - 1;
2877         u32 found_type = (u8)-1;
2878         int found_extent;
2879         int del_item;
2880         int pending_del_nr = 0;
2881         int pending_del_slot = 0;
2882         int extent_type = -1;
2883         int encoding;
2884         int ret;
2885         int err = 0;
2886
2887         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2888
2889         if (root->ref_cows)
2890                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2891
2892         path = btrfs_alloc_path();
2893         BUG_ON(!path);
2894         path->reada = -1;
2895
2896         key.objectid = inode->i_ino;
2897         key.offset = (u64)-1;
2898         key.type = (u8)-1;
2899
2900 search_again:
2901         path->leave_spinning = 1;
2902         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2903         if (ret < 0) {
2904                 err = ret;
2905                 goto out;
2906         }
2907
2908         if (ret > 0) {
2909                 /* there are no items in the tree for us to truncate, we're
2910                  * done
2911                  */
2912                 if (path->slots[0] == 0)
2913                         goto out;
2914                 path->slots[0]--;
2915         }
2916
2917         while (1) {
2918                 fi = NULL;
2919                 leaf = path->nodes[0];
2920                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2921                 found_type = btrfs_key_type(&found_key);
2922                 encoding = 0;
2923
2924                 if (found_key.objectid != inode->i_ino)
2925                         break;
2926
2927                 if (found_type < min_type)
2928                         break;
2929
2930                 item_end = found_key.offset;
2931                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2932                         fi = btrfs_item_ptr(leaf, path->slots[0],
2933                                             struct btrfs_file_extent_item);
2934                         extent_type = btrfs_file_extent_type(leaf, fi);
2935                         encoding = btrfs_file_extent_compression(leaf, fi);
2936                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2937                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2938
2939                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2940                                 item_end +=
2941                                     btrfs_file_extent_num_bytes(leaf, fi);
2942                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2943                                 item_end += btrfs_file_extent_inline_len(leaf,
2944                                                                          fi);
2945                         }
2946                         item_end--;
2947                 }
2948                 if (found_type > min_type) {
2949                         del_item = 1;
2950                 } else {
2951                         if (item_end < new_size)
2952                                 break;
2953                         if (found_key.offset >= new_size)
2954                                 del_item = 1;
2955                         else
2956                                 del_item = 0;
2957                 }
2958                 found_extent = 0;
2959                 /* FIXME, shrink the extent if the ref count is only 1 */
2960                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2961                         goto delete;
2962
2963                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2964                         u64 num_dec;
2965                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2966                         if (!del_item && !encoding) {
2967                                 u64 orig_num_bytes =
2968                                         btrfs_file_extent_num_bytes(leaf, fi);
2969                                 extent_num_bytes = new_size -
2970                                         found_key.offset + root->sectorsize - 1;
2971                                 extent_num_bytes = extent_num_bytes &
2972                                         ~((u64)root->sectorsize - 1);
2973                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2974                                                          extent_num_bytes);
2975                                 num_dec = (orig_num_bytes -
2976                                            extent_num_bytes);
2977                                 if (root->ref_cows && extent_start != 0)
2978                                         inode_sub_bytes(inode, num_dec);
2979                                 btrfs_mark_buffer_dirty(leaf);
2980                         } else {
2981                                 extent_num_bytes =
2982                                         btrfs_file_extent_disk_num_bytes(leaf,
2983                                                                          fi);
2984                                 extent_offset = found_key.offset -
2985                                         btrfs_file_extent_offset(leaf, fi);
2986
2987                                 /* FIXME blocksize != 4096 */
2988                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2989                                 if (extent_start != 0) {
2990                                         found_extent = 1;
2991                                         if (root->ref_cows)
2992                                                 inode_sub_bytes(inode, num_dec);
2993                                 }
2994                         }
2995                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2996                         /*
2997                          * we can't truncate inline items that have had
2998                          * special encodings
2999                          */
3000                         if (!del_item &&
3001                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3002                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3003                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3004                                 u32 size = new_size - found_key.offset;
3005
3006                                 if (root->ref_cows) {
3007                                         inode_sub_bytes(inode, item_end + 1 -
3008                                                         new_size);
3009                                 }
3010                                 size =
3011                                     btrfs_file_extent_calc_inline_size(size);
3012                                 ret = btrfs_truncate_item(trans, root, path,
3013                                                           size, 1);
3014                                 BUG_ON(ret);
3015                         } else if (root->ref_cows) {
3016                                 inode_sub_bytes(inode, item_end + 1 -
3017                                                 found_key.offset);
3018                         }
3019                 }
3020 delete:
3021                 if (del_item) {
3022                         if (!pending_del_nr) {
3023                                 /* no pending yet, add ourselves */
3024                                 pending_del_slot = path->slots[0];
3025                                 pending_del_nr = 1;
3026                         } else if (pending_del_nr &&
3027                                    path->slots[0] + 1 == pending_del_slot) {
3028                                 /* hop on the pending chunk */
3029                                 pending_del_nr++;
3030                                 pending_del_slot = path->slots[0];
3031                         } else {
3032                                 BUG();
3033                         }
3034                 } else {
3035                         break;
3036                 }
3037                 if (found_extent && root->ref_cows) {
3038                         btrfs_set_path_blocking(path);
3039                         ret = btrfs_free_extent(trans, root, extent_start,
3040                                                 extent_num_bytes, 0,
3041                                                 btrfs_header_owner(leaf),
3042                                                 inode->i_ino, extent_offset);
3043                         BUG_ON(ret);
3044                 }
3045
3046                 if (found_type == BTRFS_INODE_ITEM_KEY)
3047                         break;
3048
3049                 if (path->slots[0] == 0 ||
3050                     path->slots[0] != pending_del_slot) {
3051                         if (root->ref_cows) {
3052                                 err = -EAGAIN;
3053                                 goto out;
3054                         }
3055                         if (pending_del_nr) {
3056                                 ret = btrfs_del_items(trans, root, path,
3057                                                 pending_del_slot,
3058                                                 pending_del_nr);
3059                                 BUG_ON(ret);
3060                                 pending_del_nr = 0;
3061                         }
3062                         btrfs_release_path(root, path);
3063                         goto search_again;
3064                 } else {
3065                         path->slots[0]--;
3066                 }
3067         }
3068 out:
3069         if (pending_del_nr) {
3070                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3071                                       pending_del_nr);
3072         }
3073         btrfs_free_path(path);
3074         return err;
3075 }
3076
3077 /*
3078  * taken from block_truncate_page, but does cow as it zeros out
3079  * any bytes left in the last page in the file.
3080  */
3081 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3082 {
3083         struct inode *inode = mapping->host;
3084         struct btrfs_root *root = BTRFS_I(inode)->root;
3085         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3086         struct btrfs_ordered_extent *ordered;
3087         char *kaddr;
3088         u32 blocksize = root->sectorsize;
3089         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3090         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3091         struct page *page;
3092         int ret = 0;
3093         u64 page_start;
3094         u64 page_end;
3095
3096         if ((offset & (blocksize - 1)) == 0)
3097                 goto out;
3098         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3099         if (ret)
3100                 goto out;
3101
3102         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3103         if (ret)
3104                 goto out;
3105
3106         ret = -ENOMEM;
3107 again:
3108         page = grab_cache_page(mapping, index);
3109         if (!page) {
3110                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3111                 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3112                 goto out;
3113         }
3114
3115         page_start = page_offset(page);
3116         page_end = page_start + PAGE_CACHE_SIZE - 1;
3117
3118         if (!PageUptodate(page)) {
3119                 ret = btrfs_readpage(NULL, page);
3120                 lock_page(page);
3121                 if (page->mapping != mapping) {
3122                         unlock_page(page);
3123                         page_cache_release(page);
3124                         goto again;
3125                 }
3126                 if (!PageUptodate(page)) {
3127                         ret = -EIO;
3128                         goto out_unlock;
3129                 }
3130         }
3131         wait_on_page_writeback(page);
3132
3133         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3134         set_page_extent_mapped(page);
3135
3136         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3137         if (ordered) {
3138                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3139                 unlock_page(page);
3140                 page_cache_release(page);
3141                 btrfs_start_ordered_extent(inode, ordered, 1);
3142                 btrfs_put_ordered_extent(ordered);
3143                 goto again;
3144         }
3145
3146         clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
3147                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3148                           GFP_NOFS);
3149
3150         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3151         if (ret) {
3152                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3153                 goto out_unlock;
3154         }
3155
3156         ret = 0;
3157         if (offset != PAGE_CACHE_SIZE) {
3158                 kaddr = kmap(page);
3159                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3160                 flush_dcache_page(page);
3161                 kunmap(page);
3162         }
3163         ClearPageChecked(page);
3164         set_page_dirty(page);
3165         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3166
3167 out_unlock:
3168         if (ret)
3169                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3170         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3171         unlock_page(page);
3172         page_cache_release(page);
3173 out:
3174         return ret;
3175 }
3176
3177 int btrfs_cont_expand(struct inode *inode, loff_t size)
3178 {
3179         struct btrfs_trans_handle *trans;
3180         struct btrfs_root *root = BTRFS_I(inode)->root;
3181         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3182         struct extent_map *em;
3183         u64 mask = root->sectorsize - 1;
3184         u64 hole_start = (inode->i_size + mask) & ~mask;
3185         u64 block_end = (size + mask) & ~mask;
3186         u64 last_byte;
3187         u64 cur_offset;
3188         u64 hole_size;
3189         int err = 0;
3190
3191         if (size <= hole_start)
3192                 return 0;
3193
3194         while (1) {
3195                 struct btrfs_ordered_extent *ordered;
3196                 btrfs_wait_ordered_range(inode, hole_start,
3197                                          block_end - hole_start);
3198                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3199                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3200                 if (!ordered)
3201                         break;
3202                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3203                 btrfs_put_ordered_extent(ordered);
3204         }
3205
3206         cur_offset = hole_start;
3207         while (1) {
3208                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3209                                 block_end - cur_offset, 0);
3210                 BUG_ON(IS_ERR(em) || !em);
3211                 last_byte = min(extent_map_end(em), block_end);
3212                 last_byte = (last_byte + mask) & ~mask;
3213                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3214                         u64 hint_byte = 0;
3215                         hole_size = last_byte - cur_offset;
3216
3217                         err = btrfs_reserve_metadata_space(root, 2);
3218                         if (err)
3219                                 break;
3220
3221                         trans = btrfs_start_transaction(root, 1);
3222                         btrfs_set_trans_block_group(trans, inode);
3223
3224                         err = btrfs_drop_extents(trans, inode, cur_offset,
3225                                                  cur_offset + hole_size,
3226                                                  &hint_byte, 1);
3227                         BUG_ON(err);
3228
3229                         err = btrfs_insert_file_extent(trans, root,
3230                                         inode->i_ino, cur_offset, 0,
3231                                         0, hole_size, 0, hole_size,
3232                                         0, 0, 0);
3233                         BUG_ON(err);
3234
3235                         btrfs_drop_extent_cache(inode, hole_start,
3236                                         last_byte - 1, 0);
3237
3238                         btrfs_end_transaction(trans, root);
3239                         btrfs_unreserve_metadata_space(root, 2);
3240                 }
3241                 free_extent_map(em);
3242                 cur_offset = last_byte;
3243                 if (cur_offset >= block_end)
3244                         break;
3245         }
3246
3247         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3248         return err;
3249 }
3250
3251 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3252 {
3253         struct btrfs_root *root = BTRFS_I(inode)->root;
3254         struct btrfs_trans_handle *trans;
3255         unsigned long nr;
3256         int ret;
3257
3258         if (attr->ia_size == inode->i_size)
3259                 return 0;
3260
3261         if (attr->ia_size > inode->i_size) {
3262                 unsigned long limit;
3263                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3264                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3265                         return -EFBIG;
3266                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3267                         send_sig(SIGXFSZ, current, 0);
3268                         return -EFBIG;
3269                 }
3270         }
3271
3272         ret = btrfs_reserve_metadata_space(root, 1);
3273         if (ret)
3274                 return ret;
3275
3276         trans = btrfs_start_transaction(root, 1);
3277         btrfs_set_trans_block_group(trans, inode);
3278
3279         ret = btrfs_orphan_add(trans, inode);
3280         BUG_ON(ret);
3281
3282         nr = trans->blocks_used;
3283         btrfs_end_transaction(trans, root);
3284         btrfs_unreserve_metadata_space(root, 1);
3285         btrfs_btree_balance_dirty(root, nr);
3286
3287         if (attr->ia_size > inode->i_size) {
3288                 ret = btrfs_cont_expand(inode, attr->ia_size);
3289                 if (ret) {
3290                         btrfs_truncate(inode);
3291                         return ret;
3292                 }
3293
3294                 i_size_write(inode, attr->ia_size);
3295                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3296
3297                 trans = btrfs_start_transaction(root, 1);
3298                 btrfs_set_trans_block_group(trans, inode);
3299
3300                 ret = btrfs_update_inode(trans, root, inode);
3301                 BUG_ON(ret);
3302                 if (inode->i_nlink > 0) {
3303                         ret = btrfs_orphan_del(trans, inode);
3304                         BUG_ON(ret);
3305                 }
3306                 nr = trans->blocks_used;
3307                 btrfs_end_transaction(trans, root);
3308                 btrfs_btree_balance_dirty(root, nr);
3309                 return 0;
3310         }
3311
3312         /*
3313          * We're truncating a file that used to have good data down to
3314          * zero. Make sure it gets into the ordered flush list so that
3315          * any new writes get down to disk quickly.
3316          */
3317         if (attr->ia_size == 0)
3318                 BTRFS_I(inode)->ordered_data_close = 1;
3319
3320         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3321         ret = vmtruncate(inode, attr->ia_size);
3322         BUG_ON(ret);
3323
3324         return 0;
3325 }
3326
3327 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3328 {
3329         struct inode *inode = dentry->d_inode;
3330         int err;
3331
3332         err = inode_change_ok(inode, attr);
3333         if (err)
3334                 return err;
3335
3336         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3337                 err = btrfs_setattr_size(inode, attr);
3338                 if (err)
3339                         return err;
3340         }
3341         attr->ia_valid &= ~ATTR_SIZE;
3342
3343         if (attr->ia_valid)
3344                 err = inode_setattr(inode, attr);
3345
3346         if (!err && ((attr->ia_valid & ATTR_MODE)))
3347                 err = btrfs_acl_chmod(inode);
3348         return err;
3349 }
3350
3351 void btrfs_delete_inode(struct inode *inode)
3352 {
3353         struct btrfs_trans_handle *trans;
3354         struct btrfs_root *root = BTRFS_I(inode)->root;
3355         unsigned long nr;
3356         int ret;
3357
3358         truncate_inode_pages(&inode->i_data, 0);
3359         if (is_bad_inode(inode)) {
3360                 btrfs_orphan_del(NULL, inode);
3361                 goto no_delete;
3362         }
3363         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3364
3365         if (root->fs_info->log_root_recovering) {
3366                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3367                 goto no_delete;
3368         }
3369
3370         if (inode->i_nlink > 0) {
3371                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3372                 goto no_delete;
3373         }
3374
3375         btrfs_i_size_write(inode, 0);
3376
3377         while (1) {
3378                 trans = btrfs_start_transaction(root, 1);
3379                 btrfs_set_trans_block_group(trans, inode);
3380                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3381
3382                 if (ret != -EAGAIN)
3383                         break;
3384
3385                 nr = trans->blocks_used;
3386                 btrfs_end_transaction(trans, root);
3387                 trans = NULL;
3388                 btrfs_btree_balance_dirty(root, nr);
3389         }
3390
3391         if (ret == 0) {
3392                 ret = btrfs_orphan_del(trans, inode);
3393                 BUG_ON(ret);
3394         }
3395
3396         nr = trans->blocks_used;
3397         btrfs_end_transaction(trans, root);
3398         btrfs_btree_balance_dirty(root, nr);
3399 no_delete:
3400         clear_inode(inode);
3401         return;
3402 }
3403
3404 /*
3405  * this returns the key found in the dir entry in the location pointer.
3406  * If no dir entries were found, location->objectid is 0.
3407  */
3408 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3409                                struct btrfs_key *location)
3410 {
3411         const char *name = dentry->d_name.name;
3412         int namelen = dentry->d_name.len;
3413         struct btrfs_dir_item *di;
3414         struct btrfs_path *path;
3415         struct btrfs_root *root = BTRFS_I(dir)->root;
3416         int ret = 0;
3417
3418         path = btrfs_alloc_path();
3419         BUG_ON(!path);
3420
3421         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3422                                     namelen, 0);
3423         if (IS_ERR(di))
3424                 ret = PTR_ERR(di);
3425
3426         if (!di || IS_ERR(di))
3427                 goto out_err;
3428
3429         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3430 out:
3431         btrfs_free_path(path);
3432         return ret;
3433 out_err:
3434         location->objectid = 0;
3435         goto out;
3436 }
3437
3438 /*
3439  * when we hit a tree root in a directory, the btrfs part of the inode
3440  * needs to be changed to reflect the root directory of the tree root.  This
3441  * is kind of like crossing a mount point.
3442  */
3443 static int fixup_tree_root_location(struct btrfs_root *root,
3444                                     struct inode *dir,
3445                                     struct dentry *dentry,
3446                                     struct btrfs_key *location,
3447                                     struct btrfs_root **sub_root)
3448 {
3449         struct btrfs_path *path;
3450         struct btrfs_root *new_root;
3451         struct btrfs_root_ref *ref;
3452         struct extent_buffer *leaf;
3453         int ret;
3454         int err = 0;
3455
3456         path = btrfs_alloc_path();
3457         if (!path) {
3458                 err = -ENOMEM;
3459                 goto out;
3460         }
3461
3462         err = -ENOENT;
3463         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3464                                   BTRFS_I(dir)->root->root_key.objectid,
3465                                   location->objectid);
3466         if (ret) {
3467                 if (ret < 0)
3468                         err = ret;
3469                 goto out;
3470         }
3471
3472         leaf = path->nodes[0];
3473         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3474         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3475             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3476                 goto out;
3477
3478         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3479                                    (unsigned long)(ref + 1),
3480                                    dentry->d_name.len);
3481         if (ret)
3482                 goto out;
3483
3484         btrfs_release_path(root->fs_info->tree_root, path);
3485
3486         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3487         if (IS_ERR(new_root)) {
3488                 err = PTR_ERR(new_root);
3489                 goto out;
3490         }
3491
3492         if (btrfs_root_refs(&new_root->root_item) == 0) {
3493                 err = -ENOENT;
3494                 goto out;
3495         }
3496
3497         *sub_root = new_root;
3498         location->objectid = btrfs_root_dirid(&new_root->root_item);
3499         location->type = BTRFS_INODE_ITEM_KEY;
3500         location->offset = 0;
3501         err = 0;
3502 out:
3503         btrfs_free_path(path);
3504         return err;
3505 }
3506
3507 static void inode_tree_add(struct inode *inode)
3508 {
3509         struct btrfs_root *root = BTRFS_I(inode)->root;
3510         struct btrfs_inode *entry;
3511         struct rb_node **p;
3512         struct rb_node *parent;
3513 again:
3514         p = &root->inode_tree.rb_node;
3515         parent = NULL;
3516
3517         if (hlist_unhashed(&inode->i_hash))
3518                 return;
3519
3520         spin_lock(&root->inode_lock);
3521         while (*p) {
3522                 parent = *p;
3523                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3524
3525                 if (inode->i_ino < entry->vfs_inode.i_ino)
3526                         p = &parent->rb_left;
3527                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3528                         p = &parent->rb_right;
3529                 else {
3530                         WARN_ON(!(entry->vfs_inode.i_state &
3531                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3532                         rb_erase(parent, &root->inode_tree);
3533                         RB_CLEAR_NODE(parent);
3534                         spin_unlock(&root->inode_lock);
3535                         goto again;
3536                 }
3537         }
3538         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3539         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3540         spin_unlock(&root->inode_lock);
3541 }
3542
3543 static void inode_tree_del(struct inode *inode)
3544 {
3545         struct btrfs_root *root = BTRFS_I(inode)->root;
3546         int empty = 0;
3547
3548         spin_lock(&root->inode_lock);
3549         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3550                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3551                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3552                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3553         }
3554         spin_unlock(&root->inode_lock);
3555
3556         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3557                 synchronize_srcu(&root->fs_info->subvol_srcu);
3558                 spin_lock(&root->inode_lock);
3559                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3560                 spin_unlock(&root->inode_lock);
3561                 if (empty)
3562                         btrfs_add_dead_root(root);
3563         }
3564 }
3565
3566 int btrfs_invalidate_inodes(struct btrfs_root *root)
3567 {
3568         struct rb_node *node;
3569         struct rb_node *prev;
3570         struct btrfs_inode *entry;
3571         struct inode *inode;
3572         u64 objectid = 0;
3573
3574         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3575
3576         spin_lock(&root->inode_lock);
3577 again:
3578         node = root->inode_tree.rb_node;
3579         prev = NULL;
3580         while (node) {
3581                 prev = node;
3582                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3583
3584                 if (objectid < entry->vfs_inode.i_ino)
3585                         node = node->rb_left;
3586                 else if (objectid > entry->vfs_inode.i_ino)
3587                         node = node->rb_right;
3588                 else
3589                         break;
3590         }
3591         if (!node) {
3592                 while (prev) {
3593                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3594                         if (objectid <= entry->vfs_inode.i_ino) {
3595                                 node = prev;
3596                                 break;
3597                         }
3598                         prev = rb_next(prev);
3599                 }
3600         }
3601         while (node) {
3602                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3603                 objectid = entry->vfs_inode.i_ino + 1;
3604                 inode = igrab(&entry->vfs_inode);
3605                 if (inode) {
3606                         spin_unlock(&root->inode_lock);
3607                         if (atomic_read(&inode->i_count) > 1)
3608                                 d_prune_aliases(inode);
3609                         /*
3610                          * btrfs_drop_inode will remove it from
3611                          * the inode cache when its usage count
3612                          * hits zero.
3613                          */
3614                         iput(inode);
3615                         cond_resched();
3616                         spin_lock(&root->inode_lock);
3617                         goto again;
3618                 }
3619
3620                 if (cond_resched_lock(&root->inode_lock))
3621                         goto again;
3622
3623                 node = rb_next(node);
3624         }
3625         spin_unlock(&root->inode_lock);
3626         return 0;
3627 }
3628
3629 static noinline void init_btrfs_i(struct inode *inode)
3630 {
3631         struct btrfs_inode *bi = BTRFS_I(inode);
3632
3633         bi->generation = 0;
3634         bi->sequence = 0;
3635         bi->last_trans = 0;
3636         bi->last_sub_trans = 0;
3637         bi->logged_trans = 0;
3638         bi->delalloc_bytes = 0;
3639         bi->reserved_bytes = 0;
3640         bi->disk_i_size = 0;
3641         bi->flags = 0;
3642         bi->index_cnt = (u64)-1;
3643         bi->last_unlink_trans = 0;
3644         bi->ordered_data_close = 0;
3645         bi->force_compress = 0;
3646         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3647         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3648                              inode->i_mapping, GFP_NOFS);
3649         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3650                              inode->i_mapping, GFP_NOFS);
3651         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3652         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3653         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3654         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3655         mutex_init(&BTRFS_I(inode)->log_mutex);
3656 }
3657
3658 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3659 {
3660         struct btrfs_iget_args *args = p;
3661         inode->i_ino = args->ino;
3662         init_btrfs_i(inode);
3663         BTRFS_I(inode)->root = args->root;
3664         btrfs_set_inode_space_info(args->root, inode);
3665         return 0;
3666 }
3667
3668 static int btrfs_find_actor(struct inode *inode, void *opaque)
3669 {
3670         struct btrfs_iget_args *args = opaque;
3671         return args->ino == inode->i_ino &&
3672                 args->root == BTRFS_I(inode)->root;
3673 }
3674
3675 static struct inode *btrfs_iget_locked(struct super_block *s,
3676                                        u64 objectid,
3677                                        struct btrfs_root *root)
3678 {
3679         struct inode *inode;
3680         struct btrfs_iget_args args;
3681         args.ino = objectid;
3682         args.root = root;
3683
3684         inode = iget5_locked(s, objectid, btrfs_find_actor,
3685                              btrfs_init_locked_inode,
3686                              (void *)&args);
3687         return inode;
3688 }
3689
3690 /* Get an inode object given its location and corresponding root.
3691  * Returns in *is_new if the inode was read from disk
3692  */
3693 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3694                          struct btrfs_root *root, int *new)
3695 {
3696         struct inode *inode;
3697
3698         inode = btrfs_iget_locked(s, location->objectid, root);
3699         if (!inode)
3700                 return ERR_PTR(-ENOMEM);
3701
3702         if (inode->i_state & I_NEW) {
3703                 BTRFS_I(inode)->root = root;
3704                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3705                 btrfs_read_locked_inode(inode);
3706
3707                 inode_tree_add(inode);
3708                 unlock_new_inode(inode);
3709                 if (new)
3710                         *new = 1;
3711         }
3712
3713         return inode;
3714 }
3715
3716 static struct inode *new_simple_dir(struct super_block *s,
3717                                     struct btrfs_key *key,
3718                                     struct btrfs_root *root)
3719 {
3720         struct inode *inode = new_inode(s);
3721
3722         if (!inode)
3723                 return ERR_PTR(-ENOMEM);
3724
3725         init_btrfs_i(inode);
3726
3727         BTRFS_I(inode)->root = root;
3728         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3729         BTRFS_I(inode)->dummy_inode = 1;
3730
3731         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3732         inode->i_op = &simple_dir_inode_operations;
3733         inode->i_fop = &simple_dir_operations;
3734         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3735         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3736
3737         return inode;
3738 }
3739
3740 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3741 {
3742         struct inode *inode;
3743         struct btrfs_root *root = BTRFS_I(dir)->root;
3744         struct btrfs_root *sub_root = root;
3745         struct btrfs_key location;
3746         int index;
3747         int ret;
3748
3749         dentry->d_op = &btrfs_dentry_operations;
3750
3751         if (dentry->d_name.len > BTRFS_NAME_LEN)
3752                 return ERR_PTR(-ENAMETOOLONG);
3753
3754         ret = btrfs_inode_by_name(dir, dentry, &location);
3755
3756         if (ret < 0)
3757                 return ERR_PTR(ret);
3758
3759         if (location.objectid == 0)
3760                 return NULL;
3761
3762         if (location.type == BTRFS_INODE_ITEM_KEY) {
3763                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3764                 return inode;
3765         }
3766
3767         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3768
3769         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3770         ret = fixup_tree_root_location(root, dir, dentry,
3771                                        &location, &sub_root);
3772         if (ret < 0) {
3773                 if (ret != -ENOENT)
3774                         inode = ERR_PTR(ret);
3775                 else
3776                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3777         } else {
3778                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3779         }
3780         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3781
3782         if (root != sub_root) {
3783                 down_read(&root->fs_info->cleanup_work_sem);
3784                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3785                         btrfs_orphan_cleanup(sub_root);
3786                 up_read(&root->fs_info->cleanup_work_sem);
3787         }
3788
3789         return inode;
3790 }
3791
3792 static int btrfs_dentry_delete(struct dentry *dentry)
3793 {
3794         struct btrfs_root *root;
3795
3796         if (!dentry->d_inode && !IS_ROOT(dentry))
3797                 dentry = dentry->d_parent;
3798
3799         if (dentry->d_inode) {
3800                 root = BTRFS_I(dentry->d_inode)->root;
3801                 if (btrfs_root_refs(&root->root_item) == 0)
3802                         return 1;
3803         }
3804         return 0;
3805 }
3806
3807 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3808                                    struct nameidata *nd)
3809 {
3810         struct inode *inode;
3811
3812         inode = btrfs_lookup_dentry(dir, dentry);
3813         if (IS_ERR(inode))
3814                 return ERR_CAST(inode);
3815
3816         return d_splice_alias(inode, dentry);
3817 }
3818
3819 static unsigned char btrfs_filetype_table[] = {
3820         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3821 };
3822
3823 static int btrfs_real_readdir(struct file *filp, void *dirent,
3824                               filldir_t filldir)
3825 {
3826         struct inode *inode = filp->f_dentry->d_inode;
3827         struct btrfs_root *root = BTRFS_I(inode)->root;
3828         struct btrfs_item *item;
3829         struct btrfs_dir_item *di;
3830         struct btrfs_key key;
3831         struct btrfs_key found_key;
3832         struct btrfs_path *path;
3833         int ret;
3834         u32 nritems;
3835         struct extent_buffer *leaf;
3836         int slot;
3837         int advance;
3838         unsigned char d_type;
3839         int over = 0;
3840         u32 di_cur;
3841         u32 di_total;
3842         u32 di_len;
3843         int key_type = BTRFS_DIR_INDEX_KEY;
3844         char tmp_name[32];
3845         char *name_ptr;
3846         int name_len;
3847
3848         /* FIXME, use a real flag for deciding about the key type */
3849         if (root->fs_info->tree_root == root)
3850                 key_type = BTRFS_DIR_ITEM_KEY;
3851
3852         /* special case for "." */
3853         if (filp->f_pos == 0) {
3854                 over = filldir(dirent, ".", 1,
3855                                1, inode->i_ino,
3856                                DT_DIR);
3857                 if (over)
3858                         return 0;
3859                 filp->f_pos = 1;
3860         }
3861         /* special case for .., just use the back ref */
3862         if (filp->f_pos == 1) {
3863                 u64 pino = parent_ino(filp->f_path.dentry);
3864                 over = filldir(dirent, "..", 2,
3865                                2, pino, DT_DIR);
3866                 if (over)
3867                         return 0;
3868                 filp->f_pos = 2;
3869         }
3870         path = btrfs_alloc_path();
3871         path->reada = 2;
3872
3873         btrfs_set_key_type(&key, key_type);
3874         key.offset = filp->f_pos;
3875         key.objectid = inode->i_ino;
3876
3877         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3878         if (ret < 0)
3879                 goto err;
3880         advance = 0;
3881
3882         while (1) {
3883                 leaf = path->nodes[0];
3884                 nritems = btrfs_header_nritems(leaf);
3885                 slot = path->slots[0];
3886                 if (advance || slot >= nritems) {
3887                         if (slot >= nritems - 1) {
3888                                 ret = btrfs_next_leaf(root, path);
3889                                 if (ret)
3890                                         break;
3891                                 leaf = path->nodes[0];
3892                                 nritems = btrfs_header_nritems(leaf);
3893                                 slot = path->slots[0];
3894                         } else {
3895                                 slot++;
3896                                 path->slots[0]++;
3897                         }
3898                 }
3899
3900                 advance = 1;
3901                 item = btrfs_item_nr(leaf, slot);
3902                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3903
3904                 if (found_key.objectid != key.objectid)
3905                         break;
3906                 if (btrfs_key_type(&found_key) != key_type)
3907                         break;
3908                 if (found_key.offset < filp->f_pos)
3909                         continue;
3910
3911                 filp->f_pos = found_key.offset;
3912
3913                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3914                 di_cur = 0;
3915                 di_total = btrfs_item_size(leaf, item);
3916
3917                 while (di_cur < di_total) {
3918                         struct btrfs_key location;
3919
3920                         name_len = btrfs_dir_name_len(leaf, di);
3921                         if (name_len <= sizeof(tmp_name)) {
3922                                 name_ptr = tmp_name;
3923                         } else {
3924                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3925                                 if (!name_ptr) {
3926                                         ret = -ENOMEM;
3927                                         goto err;
3928                                 }
3929                         }
3930                         read_extent_buffer(leaf, name_ptr,
3931                                            (unsigned long)(di + 1), name_len);
3932
3933                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3934                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3935
3936                         /* is this a reference to our own snapshot? If so
3937                          * skip it
3938                          */
3939                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3940                             location.objectid == root->root_key.objectid) {
3941                                 over = 0;
3942                                 goto skip;
3943                         }
3944                         over = filldir(dirent, name_ptr, name_len,
3945                                        found_key.offset, location.objectid,
3946                                        d_type);
3947
3948 skip:
3949                         if (name_ptr != tmp_name)
3950                                 kfree(name_ptr);
3951
3952                         if (over)
3953                                 goto nopos;
3954                         di_len = btrfs_dir_name_len(leaf, di) +
3955                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3956                         di_cur += di_len;
3957                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3958                 }
3959         }
3960
3961         /* Reached end of directory/root. Bump pos past the last item. */
3962         if (key_type == BTRFS_DIR_INDEX_KEY)
3963                 /*
3964                  * 32-bit glibc will use getdents64, but then strtol -
3965                  * so the last number we can serve is this.
3966                  */
3967                 filp->f_pos = 0x7fffffff;
3968         else
3969                 filp->f_pos++;
3970 nopos:
3971         ret = 0;
3972 err:
3973         btrfs_free_path(path);
3974         return ret;
3975 }
3976
3977 int btrfs_write_inode(struct inode *inode, int wait)
3978 {
3979         struct btrfs_root *root = BTRFS_I(inode)->root;
3980         struct btrfs_trans_handle *trans;
3981         int ret = 0;
3982
3983         if (root->fs_info->btree_inode == inode)
3984                 return 0;
3985
3986         if (wait) {
3987                 trans = btrfs_join_transaction(root, 1);
3988                 btrfs_set_trans_block_group(trans, inode);
3989                 ret = btrfs_commit_transaction(trans, root);
3990         }
3991         return ret;
3992 }
3993
3994 /*
3995  * This is somewhat expensive, updating the tree every time the
3996  * inode changes.  But, it is most likely to find the inode in cache.
3997  * FIXME, needs more benchmarking...there are no reasons other than performance
3998  * to keep or drop this code.
3999  */
4000 void btrfs_dirty_inode(struct inode *inode)
4001 {
4002         struct btrfs_root *root = BTRFS_I(inode)->root;
4003         struct btrfs_trans_handle *trans;
4004
4005         trans = btrfs_join_transaction(root, 1);
4006         btrfs_set_trans_block_group(trans, inode);
4007         btrfs_update_inode(trans, root, inode);
4008         btrfs_end_transaction(trans, root);
4009 }
4010
4011 /*
4012  * find the highest existing sequence number in a directory
4013  * and then set the in-memory index_cnt variable to reflect
4014  * free sequence numbers
4015  */
4016 static int btrfs_set_inode_index_count(struct inode *inode)
4017 {
4018         struct btrfs_root *root = BTRFS_I(inode)->root;
4019         struct btrfs_key key, found_key;
4020         struct btrfs_path *path;
4021         struct extent_buffer *leaf;
4022         int ret;
4023
4024         key.objectid = inode->i_ino;
4025         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4026         key.offset = (u64)-1;
4027
4028         path = btrfs_alloc_path();
4029         if (!path)
4030                 return -ENOMEM;
4031
4032         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4033         if (ret < 0)
4034                 goto out;
4035         /* FIXME: we should be able to handle this */
4036         if (ret == 0)
4037                 goto out;
4038         ret = 0;
4039
4040         /*
4041          * MAGIC NUMBER EXPLANATION:
4042          * since we search a directory based on f_pos we have to start at 2
4043          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4044          * else has to start at 2
4045          */
4046         if (path->slots[0] == 0) {
4047                 BTRFS_I(inode)->index_cnt = 2;
4048                 goto out;
4049         }
4050
4051         path->slots[0]--;
4052
4053         leaf = path->nodes[0];
4054         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4055
4056         if (found_key.objectid != inode->i_ino ||
4057             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4058                 BTRFS_I(inode)->index_cnt = 2;
4059                 goto out;
4060         }
4061
4062         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4063 out:
4064         btrfs_free_path(path);
4065         return ret;
4066 }
4067
4068 /*
4069  * helper to find a free sequence number in a given directory.  This current
4070  * code is very simple, later versions will do smarter things in the btree
4071  */
4072 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4073 {
4074         int ret = 0;
4075
4076         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4077                 ret = btrfs_set_inode_index_count(dir);
4078                 if (ret)
4079                         return ret;
4080         }
4081
4082         *index = BTRFS_I(dir)->index_cnt;
4083         BTRFS_I(dir)->index_cnt++;
4084
4085         return ret;
4086 }
4087
4088 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4089                                      struct btrfs_root *root,
4090                                      struct inode *dir,
4091                                      const char *name, int name_len,
4092                                      u64 ref_objectid, u64 objectid,
4093                                      u64 alloc_hint, int mode, u64 *index)
4094 {
4095         struct inode *inode;
4096         struct btrfs_inode_item *inode_item;
4097         struct btrfs_key *location;
4098         struct btrfs_path *path;
4099         struct btrfs_inode_ref *ref;
4100         struct btrfs_key key[2];
4101         u32 sizes[2];
4102         unsigned long ptr;
4103         int ret;
4104         int owner;
4105
4106         path = btrfs_alloc_path();
4107         BUG_ON(!path);
4108
4109         inode = new_inode(root->fs_info->sb);
4110         if (!inode)
4111                 return ERR_PTR(-ENOMEM);
4112
4113         if (dir) {
4114                 ret = btrfs_set_inode_index(dir, index);
4115                 if (ret) {
4116                         iput(inode);
4117                         return ERR_PTR(ret);
4118                 }
4119         }
4120         /*
4121          * index_cnt is ignored for everything but a dir,
4122          * btrfs_get_inode_index_count has an explanation for the magic
4123          * number
4124          */
4125         init_btrfs_i(inode);
4126         BTRFS_I(inode)->index_cnt = 2;
4127         BTRFS_I(inode)->root = root;
4128         BTRFS_I(inode)->generation = trans->transid;
4129         btrfs_set_inode_space_info(root, inode);
4130
4131         if (mode & S_IFDIR)
4132                 owner = 0;
4133         else
4134                 owner = 1;
4135         BTRFS_I(inode)->block_group =
4136                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4137
4138         key[0].objectid = objectid;
4139         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4140         key[0].offset = 0;
4141
4142         key[1].objectid = objectid;
4143         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4144         key[1].offset = ref_objectid;
4145
4146         sizes[0] = sizeof(struct btrfs_inode_item);
4147         sizes[1] = name_len + sizeof(*ref);
4148
4149         path->leave_spinning = 1;
4150         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4151         if (ret != 0)
4152                 goto fail;
4153
4154         inode->i_uid = current_fsuid();
4155
4156         if (dir && (dir->i_mode & S_ISGID)) {
4157                 inode->i_gid = dir->i_gid;
4158                 if (S_ISDIR(mode))
4159                         mode |= S_ISGID;
4160         } else
4161                 inode->i_gid = current_fsgid();
4162
4163         inode->i_mode = mode;
4164         inode->i_ino = objectid;
4165         inode_set_bytes(inode, 0);
4166         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4167         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4168                                   struct btrfs_inode_item);
4169         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4170
4171         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4172                              struct btrfs_inode_ref);
4173         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4174         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4175         ptr = (unsigned long)(ref + 1);
4176         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4177
4178         btrfs_mark_buffer_dirty(path->nodes[0]);
4179         btrfs_free_path(path);
4180
4181         location = &BTRFS_I(inode)->location;
4182         location->objectid = objectid;
4183         location->offset = 0;
4184         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4185
4186         btrfs_inherit_iflags(inode, dir);
4187
4188         if ((mode & S_IFREG)) {
4189                 if (btrfs_test_opt(root, NODATASUM))
4190                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4191                 if (btrfs_test_opt(root, NODATACOW))
4192                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4193         }
4194
4195         insert_inode_hash(inode);
4196         inode_tree_add(inode);
4197         return inode;
4198 fail:
4199         if (dir)
4200                 BTRFS_I(dir)->index_cnt--;
4201         btrfs_free_path(path);
4202         iput(inode);
4203         return ERR_PTR(ret);
4204 }
4205
4206 static inline u8 btrfs_inode_type(struct inode *inode)
4207 {
4208         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4209 }
4210
4211 /*
4212  * utility function to add 'inode' into 'parent_inode' with
4213  * a give name and a given sequence number.
4214  * if 'add_backref' is true, also insert a backref from the
4215  * inode to the parent directory.
4216  */
4217 int btrfs_add_link(struct btrfs_trans_handle *trans,
4218                    struct inode *parent_inode, struct inode *inode,
4219                    const char *name, int name_len, int add_backref, u64 index)
4220 {
4221         int ret = 0;
4222         struct btrfs_key key;
4223         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4224
4225         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4226                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4227         } else {
4228                 key.objectid = inode->i_ino;
4229                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4230                 key.offset = 0;
4231         }
4232
4233         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4234                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4235                                          key.objectid, root->root_key.objectid,
4236                                          parent_inode->i_ino,
4237                                          index, name, name_len);
4238         } else if (add_backref) {
4239                 ret = btrfs_insert_inode_ref(trans, root,
4240                                              name, name_len, inode->i_ino,
4241                                              parent_inode->i_ino, index);
4242         }
4243
4244         if (ret == 0) {
4245                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4246                                             parent_inode->i_ino, &key,
4247                                             btrfs_inode_type(inode), index);
4248                 BUG_ON(ret);
4249
4250                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4251                                    name_len * 2);
4252                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4253                 ret = btrfs_update_inode(trans, root, parent_inode);
4254         }
4255         return ret;
4256 }
4257
4258 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4259                             struct dentry *dentry, struct inode *inode,
4260                             int backref, u64 index)
4261 {
4262         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4263                                  inode, dentry->d_name.name,
4264                                  dentry->d_name.len, backref, index);
4265         if (!err) {
4266                 d_instantiate(dentry, inode);
4267                 return 0;
4268         }
4269         if (err > 0)
4270                 err = -EEXIST;
4271         return err;
4272 }
4273
4274 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4275                         int mode, dev_t rdev)
4276 {
4277         struct btrfs_trans_handle *trans;
4278         struct btrfs_root *root = BTRFS_I(dir)->root;
4279         struct inode *inode = NULL;
4280         int err;
4281         int drop_inode = 0;
4282         u64 objectid;
4283         unsigned long nr = 0;
4284         u64 index = 0;
4285
4286         if (!new_valid_dev(rdev))
4287                 return -EINVAL;
4288
4289         /*
4290          * 2 for inode item and ref
4291          * 2 for dir items
4292          * 1 for xattr if selinux is on
4293          */
4294         err = btrfs_reserve_metadata_space(root, 5);
4295         if (err)
4296                 return err;
4297
4298         trans = btrfs_start_transaction(root, 1);
4299         if (!trans)
4300                 goto fail;
4301         btrfs_set_trans_block_group(trans, dir);
4302
4303         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4304         if (err) {
4305                 err = -ENOSPC;
4306                 goto out_unlock;
4307         }
4308
4309         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4310                                 dentry->d_name.len,
4311                                 dentry->d_parent->d_inode->i_ino, objectid,
4312                                 BTRFS_I(dir)->block_group, mode, &index);
4313         err = PTR_ERR(inode);
4314         if (IS_ERR(inode))
4315                 goto out_unlock;
4316
4317         err = btrfs_init_inode_security(trans, inode, dir);
4318         if (err) {
4319                 drop_inode = 1;
4320                 goto out_unlock;
4321         }
4322
4323         btrfs_set_trans_block_group(trans, inode);
4324         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4325         if (err)
4326                 drop_inode = 1;
4327         else {
4328                 inode->i_op = &btrfs_special_inode_operations;
4329                 init_special_inode(inode, inode->i_mode, rdev);
4330                 btrfs_update_inode(trans, root, inode);
4331         }
4332         btrfs_update_inode_block_group(trans, inode);
4333         btrfs_update_inode_block_group(trans, dir);
4334 out_unlock:
4335         nr = trans->blocks_used;
4336         btrfs_end_transaction_throttle(trans, root);
4337 fail:
4338         btrfs_unreserve_metadata_space(root, 5);
4339         if (drop_inode) {
4340                 inode_dec_link_count(inode);
4341                 iput(inode);
4342         }
4343         btrfs_btree_balance_dirty(root, nr);
4344         return err;
4345 }
4346
4347 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4348                         int mode, struct nameidata *nd)
4349 {
4350         struct btrfs_trans_handle *trans;
4351         struct btrfs_root *root = BTRFS_I(dir)->root;
4352         struct inode *inode = NULL;
4353         int err;
4354         int drop_inode = 0;
4355         unsigned long nr = 0;
4356         u64 objectid;
4357         u64 index = 0;
4358
4359         /*
4360          * 2 for inode item and ref
4361          * 2 for dir items
4362          * 1 for xattr if selinux is on
4363          */
4364         err = btrfs_reserve_metadata_space(root, 5);
4365         if (err)
4366                 return err;
4367
4368         trans = btrfs_start_transaction(root, 1);
4369         if (!trans)
4370                 goto fail;
4371         btrfs_set_trans_block_group(trans, dir);
4372
4373         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4374         if (err) {
4375                 err = -ENOSPC;
4376                 goto out_unlock;
4377         }
4378
4379         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4380                                 dentry->d_name.len,
4381                                 dentry->d_parent->d_inode->i_ino,
4382                                 objectid, BTRFS_I(dir)->block_group, mode,
4383                                 &index);
4384         err = PTR_ERR(inode);
4385         if (IS_ERR(inode))
4386                 goto out_unlock;
4387
4388         err = btrfs_init_inode_security(trans, inode, dir);
4389         if (err) {
4390                 drop_inode = 1;
4391                 goto out_unlock;
4392         }
4393
4394         btrfs_set_trans_block_group(trans, inode);
4395         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4396         if (err)
4397                 drop_inode = 1;
4398         else {
4399                 inode->i_mapping->a_ops = &btrfs_aops;
4400                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4401                 inode->i_fop = &btrfs_file_operations;
4402                 inode->i_op = &btrfs_file_inode_operations;
4403                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4404         }
4405         btrfs_update_inode_block_group(trans, inode);
4406         btrfs_update_inode_block_group(trans, dir);
4407 out_unlock:
4408         nr = trans->blocks_used;
4409         btrfs_end_transaction_throttle(trans, root);
4410 fail:
4411         btrfs_unreserve_metadata_space(root, 5);
4412         if (drop_inode) {
4413                 inode_dec_link_count(inode);
4414                 iput(inode);
4415         }
4416         btrfs_btree_balance_dirty(root, nr);
4417         return err;
4418 }
4419
4420 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4421                       struct dentry *dentry)
4422 {
4423         struct btrfs_trans_handle *trans;
4424         struct btrfs_root *root = BTRFS_I(dir)->root;
4425         struct inode *inode = old_dentry->d_inode;
4426         u64 index;
4427         unsigned long nr = 0;
4428         int err;
4429         int drop_inode = 0;
4430
4431         if (inode->i_nlink == 0)
4432                 return -ENOENT;
4433
4434         /* do not allow sys_link's with other subvols of the same device */
4435         if (root->objectid != BTRFS_I(inode)->root->objectid)
4436                 return -EPERM;
4437
4438         /*
4439          * 1 item for inode ref
4440          * 2 items for dir items
4441          */
4442         err = btrfs_reserve_metadata_space(root, 3);
4443         if (err)
4444                 return err;
4445
4446         btrfs_inc_nlink(inode);
4447
4448         err = btrfs_set_inode_index(dir, &index);
4449         if (err)
4450                 goto fail;
4451
4452         trans = btrfs_start_transaction(root, 1);
4453
4454         btrfs_set_trans_block_group(trans, dir);
4455         atomic_inc(&inode->i_count);
4456
4457         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4458
4459         if (err) {
4460                 drop_inode = 1;
4461         } else {
4462                 btrfs_update_inode_block_group(trans, dir);
4463                 err = btrfs_update_inode(trans, root, inode);
4464                 BUG_ON(err);
4465                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4466         }
4467
4468         nr = trans->blocks_used;
4469         btrfs_end_transaction_throttle(trans, root);
4470 fail:
4471         btrfs_unreserve_metadata_space(root, 3);
4472         if (drop_inode) {
4473                 inode_dec_link_count(inode);
4474                 iput(inode);
4475         }
4476         btrfs_btree_balance_dirty(root, nr);
4477         return err;
4478 }
4479
4480 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4481 {
4482         struct inode *inode = NULL;
4483         struct btrfs_trans_handle *trans;
4484         struct btrfs_root *root = BTRFS_I(dir)->root;
4485         int err = 0;
4486         int drop_on_err = 0;
4487         u64 objectid = 0;
4488         u64 index = 0;
4489         unsigned long nr = 1;
4490
4491         /*
4492          * 2 items for inode and ref
4493          * 2 items for dir items
4494          * 1 for xattr if selinux is on
4495          */
4496         err = btrfs_reserve_metadata_space(root, 5);
4497         if (err)
4498                 return err;
4499
4500         trans = btrfs_start_transaction(root, 1);
4501         if (!trans) {
4502                 err = -ENOMEM;
4503                 goto out_unlock;
4504         }
4505         btrfs_set_trans_block_group(trans, dir);
4506
4507         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4508         if (err) {
4509                 err = -ENOSPC;
4510                 goto out_fail;
4511         }
4512
4513         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4514                                 dentry->d_name.len,
4515                                 dentry->d_parent->d_inode->i_ino, objectid,
4516                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4517                                 &index);
4518         if (IS_ERR(inode)) {
4519                 err = PTR_ERR(inode);
4520                 goto out_fail;
4521         }
4522
4523         drop_on_err = 1;
4524
4525         err = btrfs_init_inode_security(trans, inode, dir);
4526         if (err)
4527                 goto out_fail;
4528
4529         inode->i_op = &btrfs_dir_inode_operations;
4530         inode->i_fop = &btrfs_dir_file_operations;
4531         btrfs_set_trans_block_group(trans, inode);
4532
4533         btrfs_i_size_write(inode, 0);
4534         err = btrfs_update_inode(trans, root, inode);
4535         if (err)
4536                 goto out_fail;
4537
4538         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4539                                  inode, dentry->d_name.name,
4540                                  dentry->d_name.len, 0, index);
4541         if (err)
4542                 goto out_fail;
4543
4544         d_instantiate(dentry, inode);
4545         drop_on_err = 0;
4546         btrfs_update_inode_block_group(trans, inode);
4547         btrfs_update_inode_block_group(trans, dir);
4548
4549 out_fail:
4550         nr = trans->blocks_used;
4551         btrfs_end_transaction_throttle(trans, root);
4552
4553 out_unlock:
4554         btrfs_unreserve_metadata_space(root, 5);
4555         if (drop_on_err)
4556                 iput(inode);
4557         btrfs_btree_balance_dirty(root, nr);
4558         return err;
4559 }
4560
4561 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4562  * and an extent that you want to insert, deal with overlap and insert
4563  * the new extent into the tree.
4564  */
4565 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4566                                 struct extent_map *existing,
4567                                 struct extent_map *em,
4568                                 u64 map_start, u64 map_len)
4569 {
4570         u64 start_diff;
4571
4572         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4573         start_diff = map_start - em->start;
4574         em->start = map_start;
4575         em->len = map_len;
4576         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4577             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4578                 em->block_start += start_diff;
4579                 em->block_len -= start_diff;
4580         }
4581         return add_extent_mapping(em_tree, em);
4582 }
4583
4584 static noinline int uncompress_inline(struct btrfs_path *path,
4585                                       struct inode *inode, struct page *page,
4586                                       size_t pg_offset, u64 extent_offset,
4587                                       struct btrfs_file_extent_item *item)
4588 {
4589         int ret;
4590         struct extent_buffer *leaf = path->nodes[0];
4591         char *tmp;
4592         size_t max_size;
4593         unsigned long inline_size;
4594         unsigned long ptr;
4595
4596         WARN_ON(pg_offset != 0);
4597         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4598         inline_size = btrfs_file_extent_inline_item_len(leaf,
4599                                         btrfs_item_nr(leaf, path->slots[0]));
4600         tmp = kmalloc(inline_size, GFP_NOFS);
4601         ptr = btrfs_file_extent_inline_start(item);
4602
4603         read_extent_buffer(leaf, tmp, ptr, inline_size);
4604
4605         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4606         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4607                                     inline_size, max_size);
4608         if (ret) {
4609                 char *kaddr = kmap_atomic(page, KM_USER0);
4610                 unsigned long copy_size = min_t(u64,
4611                                   PAGE_CACHE_SIZE - pg_offset,
4612                                   max_size - extent_offset);
4613                 memset(kaddr + pg_offset, 0, copy_size);
4614                 kunmap_atomic(kaddr, KM_USER0);
4615         }
4616         kfree(tmp);
4617         return 0;
4618 }
4619
4620 /*
4621  * a bit scary, this does extent mapping from logical file offset to the disk.
4622  * the ugly parts come from merging extents from the disk with the in-ram
4623  * representation.  This gets more complex because of the data=ordered code,
4624  * where the in-ram extents might be locked pending data=ordered completion.
4625  *
4626  * This also copies inline extents directly into the page.
4627  */
4628
4629 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4630                                     size_t pg_offset, u64 start, u64 len,
4631                                     int create)
4632 {
4633         int ret;
4634         int err = 0;
4635         u64 bytenr;
4636         u64 extent_start = 0;
4637         u64 extent_end = 0;
4638         u64 objectid = inode->i_ino;
4639         u32 found_type;
4640         struct btrfs_path *path = NULL;
4641         struct btrfs_root *root = BTRFS_I(inode)->root;
4642         struct btrfs_file_extent_item *item;
4643         struct extent_buffer *leaf;
4644         struct btrfs_key found_key;
4645         struct extent_map *em = NULL;
4646         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4647         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4648         struct btrfs_trans_handle *trans = NULL;
4649         int compressed;
4650
4651 again:
4652         read_lock(&em_tree->lock);
4653         em = lookup_extent_mapping(em_tree, start, len);
4654         if (em)
4655                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4656         read_unlock(&em_tree->lock);
4657
4658         if (em) {
4659                 if (em->start > start || em->start + em->len <= start)
4660                         free_extent_map(em);
4661                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4662                         free_extent_map(em);
4663                 else
4664                         goto out;
4665         }
4666         em = alloc_extent_map(GFP_NOFS);
4667         if (!em) {
4668                 err = -ENOMEM;
4669                 goto out;
4670         }
4671         em->bdev = root->fs_info->fs_devices->latest_bdev;
4672         em->start = EXTENT_MAP_HOLE;
4673         em->orig_start = EXTENT_MAP_HOLE;
4674         em->len = (u64)-1;
4675         em->block_len = (u64)-1;
4676
4677         if (!path) {
4678                 path = btrfs_alloc_path();
4679                 BUG_ON(!path);
4680         }
4681
4682         ret = btrfs_lookup_file_extent(trans, root, path,
4683                                        objectid, start, trans != NULL);
4684         if (ret < 0) {
4685                 err = ret;
4686                 goto out;
4687         }
4688
4689         if (ret != 0) {
4690                 if (path->slots[0] == 0)
4691                         goto not_found;
4692                 path->slots[0]--;
4693         }
4694
4695         leaf = path->nodes[0];
4696         item = btrfs_item_ptr(leaf, path->slots[0],
4697                               struct btrfs_file_extent_item);
4698         /* are we inside the extent that was found? */
4699         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4700         found_type = btrfs_key_type(&found_key);
4701         if (found_key.objectid != objectid ||
4702             found_type != BTRFS_EXTENT_DATA_KEY) {
4703                 goto not_found;
4704         }
4705
4706         found_type = btrfs_file_extent_type(leaf, item);
4707         extent_start = found_key.offset;
4708         compressed = btrfs_file_extent_compression(leaf, item);
4709         if (found_type == BTRFS_FILE_EXTENT_REG ||
4710             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4711                 extent_end = extent_start +
4712                        btrfs_file_extent_num_bytes(leaf, item);
4713         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4714                 size_t size;
4715                 size = btrfs_file_extent_inline_len(leaf, item);
4716                 extent_end = (extent_start + size + root->sectorsize - 1) &
4717                         ~((u64)root->sectorsize - 1);
4718         }
4719
4720         if (start >= extent_end) {
4721                 path->slots[0]++;
4722                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4723                         ret = btrfs_next_leaf(root, path);
4724                         if (ret < 0) {
4725                                 err = ret;
4726                                 goto out;
4727                         }
4728                         if (ret > 0)
4729                                 goto not_found;
4730                         leaf = path->nodes[0];
4731                 }
4732                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4733                 if (found_key.objectid != objectid ||
4734                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4735                         goto not_found;
4736                 if (start + len <= found_key.offset)
4737                         goto not_found;
4738                 em->start = start;
4739                 em->len = found_key.offset - start;
4740                 goto not_found_em;
4741         }
4742
4743         if (found_type == BTRFS_FILE_EXTENT_REG ||
4744             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4745                 em->start = extent_start;
4746                 em->len = extent_end - extent_start;
4747                 em->orig_start = extent_start -
4748                                  btrfs_file_extent_offset(leaf, item);
4749                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4750                 if (bytenr == 0) {
4751                         em->block_start = EXTENT_MAP_HOLE;
4752                         goto insert;
4753                 }
4754                 if (compressed) {
4755                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4756                         em->block_start = bytenr;
4757                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4758                                                                          item);
4759                 } else {
4760                         bytenr += btrfs_file_extent_offset(leaf, item);
4761                         em->block_start = bytenr;
4762                         em->block_len = em->len;
4763                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4764                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4765                 }
4766                 goto insert;
4767         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4768                 unsigned long ptr;
4769                 char *map;
4770                 size_t size;
4771                 size_t extent_offset;
4772                 size_t copy_size;
4773
4774                 em->block_start = EXTENT_MAP_INLINE;
4775                 if (!page || create) {
4776                         em->start = extent_start;
4777                         em->len = extent_end - extent_start;
4778                         goto out;
4779                 }
4780
4781                 size = btrfs_file_extent_inline_len(leaf, item);
4782                 extent_offset = page_offset(page) + pg_offset - extent_start;
4783                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4784                                 size - extent_offset);
4785                 em->start = extent_start + extent_offset;
4786                 em->len = (copy_size + root->sectorsize - 1) &
4787                         ~((u64)root->sectorsize - 1);
4788                 em->orig_start = EXTENT_MAP_INLINE;
4789                 if (compressed)
4790                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4791                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4792                 if (create == 0 && !PageUptodate(page)) {
4793                         if (btrfs_file_extent_compression(leaf, item) ==
4794                             BTRFS_COMPRESS_ZLIB) {
4795                                 ret = uncompress_inline(path, inode, page,
4796                                                         pg_offset,
4797                                                         extent_offset, item);
4798                                 BUG_ON(ret);
4799                         } else {
4800                                 map = kmap(page);
4801                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4802                                                    copy_size);
4803                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4804                                         memset(map + pg_offset + copy_size, 0,
4805                                                PAGE_CACHE_SIZE - pg_offset -
4806                                                copy_size);
4807                                 }
4808                                 kunmap(page);
4809                         }
4810                         flush_dcache_page(page);
4811                 } else if (create && PageUptodate(page)) {
4812                         if (!trans) {
4813                                 kunmap(page);
4814                                 free_extent_map(em);
4815                                 em = NULL;
4816                                 btrfs_release_path(root, path);
4817                                 trans = btrfs_join_transaction(root, 1);
4818                                 goto again;
4819                         }
4820                         map = kmap(page);
4821                         write_extent_buffer(leaf, map + pg_offset, ptr,
4822                                             copy_size);
4823                         kunmap(page);
4824                         btrfs_mark_buffer_dirty(leaf);
4825                 }
4826                 set_extent_uptodate(io_tree, em->start,
4827                                     extent_map_end(em) - 1, GFP_NOFS);
4828                 goto insert;
4829         } else {
4830                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4831                 WARN_ON(1);
4832         }
4833 not_found:
4834         em->start = start;
4835         em->len = len;
4836 not_found_em:
4837         em->block_start = EXTENT_MAP_HOLE;
4838         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4839 insert:
4840         btrfs_release_path(root, path);
4841         if (em->start > start || extent_map_end(em) <= start) {
4842                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4843                        "[%llu %llu]\n", (unsigned long long)em->start,
4844                        (unsigned long long)em->len,
4845                        (unsigned long long)start,
4846                        (unsigned long long)len);
4847                 err = -EIO;
4848                 goto out;
4849         }
4850
4851         err = 0;
4852         write_lock(&em_tree->lock);
4853         ret = add_extent_mapping(em_tree, em);
4854         /* it is possible that someone inserted the extent into the tree
4855          * while we had the lock dropped.  It is also possible that
4856          * an overlapping map exists in the tree
4857          */
4858         if (ret == -EEXIST) {
4859                 struct extent_map *existing;
4860
4861                 ret = 0;
4862
4863                 existing = lookup_extent_mapping(em_tree, start, len);
4864                 if (existing && (existing->start > start ||
4865                     existing->start + existing->len <= start)) {
4866                         free_extent_map(existing);
4867                         existing = NULL;
4868                 }
4869                 if (!existing) {
4870                         existing = lookup_extent_mapping(em_tree, em->start,
4871                                                          em->len);
4872                         if (existing) {
4873                                 err = merge_extent_mapping(em_tree, existing,
4874                                                            em, start,
4875                                                            root->sectorsize);
4876                                 free_extent_map(existing);
4877                                 if (err) {
4878                                         free_extent_map(em);
4879                                         em = NULL;
4880                                 }
4881                         } else {
4882                                 err = -EIO;
4883                                 free_extent_map(em);
4884                                 em = NULL;
4885                         }
4886                 } else {
4887                         free_extent_map(em);
4888                         em = existing;
4889                         err = 0;
4890                 }
4891         }
4892         write_unlock(&em_tree->lock);
4893 out:
4894         if (path)
4895                 btrfs_free_path(path);
4896         if (trans) {
4897                 ret = btrfs_end_transaction(trans, root);
4898                 if (!err)
4899                         err = ret;
4900         }
4901         if (err) {
4902                 free_extent_map(em);
4903                 return ERR_PTR(err);
4904         }
4905         return em;
4906 }
4907
4908 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4909                         const struct iovec *iov, loff_t offset,
4910                         unsigned long nr_segs)
4911 {
4912         return -EINVAL;
4913 }
4914
4915 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4916                 __u64 start, __u64 len)
4917 {
4918         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4919 }
4920
4921 int btrfs_readpage(struct file *file, struct page *page)
4922 {
4923         struct extent_io_tree *tree;
4924         tree = &BTRFS_I(page->mapping->host)->io_tree;
4925         return extent_read_full_page(tree, page, btrfs_get_extent);
4926 }
4927
4928 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4929 {
4930         struct extent_io_tree *tree;
4931
4932
4933         if (current->flags & PF_MEMALLOC) {
4934                 redirty_page_for_writepage(wbc, page);
4935                 unlock_page(page);
4936                 return 0;
4937         }
4938         tree = &BTRFS_I(page->mapping->host)->io_tree;
4939         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4940 }
4941
4942 int btrfs_writepages(struct address_space *mapping,
4943                      struct writeback_control *wbc)
4944 {
4945         struct extent_io_tree *tree;
4946
4947         tree = &BTRFS_I(mapping->host)->io_tree;
4948         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4949 }
4950
4951 static int
4952 btrfs_readpages(struct file *file, struct address_space *mapping,
4953                 struct list_head *pages, unsigned nr_pages)
4954 {
4955         struct extent_io_tree *tree;
4956         tree = &BTRFS_I(mapping->host)->io_tree;
4957         return extent_readpages(tree, mapping, pages, nr_pages,
4958                                 btrfs_get_extent);
4959 }
4960 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4961 {
4962         struct extent_io_tree *tree;
4963         struct extent_map_tree *map;
4964         int ret;
4965
4966         tree = &BTRFS_I(page->mapping->host)->io_tree;
4967         map = &BTRFS_I(page->mapping->host)->extent_tree;
4968         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4969         if (ret == 1) {
4970                 ClearPagePrivate(page);
4971                 set_page_private(page, 0);
4972                 page_cache_release(page);
4973         }
4974         return ret;
4975 }
4976
4977 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4978 {
4979         if (PageWriteback(page) || PageDirty(page))
4980                 return 0;
4981         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4982 }
4983
4984 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4985 {
4986         struct extent_io_tree *tree;
4987         struct btrfs_ordered_extent *ordered;
4988         u64 page_start = page_offset(page);
4989         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4990
4991
4992         /*
4993          * we have the page locked, so new writeback can't start,
4994          * and the dirty bit won't be cleared while we are here.
4995          *
4996          * Wait for IO on this page so that we can safely clear
4997          * the PagePrivate2 bit and do ordered accounting
4998          */
4999         wait_on_page_writeback(page);
5000
5001         tree = &BTRFS_I(page->mapping->host)->io_tree;
5002         if (offset) {
5003                 btrfs_releasepage(page, GFP_NOFS);
5004                 return;
5005         }
5006         lock_extent(tree, page_start, page_end, GFP_NOFS);
5007         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5008                                            page_offset(page));
5009         if (ordered) {
5010                 /*
5011                  * IO on this page will never be started, so we need
5012                  * to account for any ordered extents now
5013                  */
5014                 clear_extent_bit(tree, page_start, page_end,
5015                                  EXTENT_DIRTY | EXTENT_DELALLOC |
5016                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5017                                  NULL, GFP_NOFS);
5018                 /*
5019                  * whoever cleared the private bit is responsible
5020                  * for the finish_ordered_io
5021                  */
5022                 if (TestClearPagePrivate2(page)) {
5023                         btrfs_finish_ordered_io(page->mapping->host,
5024                                                 page_start, page_end);
5025                 }
5026                 btrfs_put_ordered_extent(ordered);
5027                 lock_extent(tree, page_start, page_end, GFP_NOFS);
5028         }
5029         clear_extent_bit(tree, page_start, page_end,
5030                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5031                  EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
5032         __btrfs_releasepage(page, GFP_NOFS);
5033
5034         ClearPageChecked(page);
5035         if (PagePrivate(page)) {
5036                 ClearPagePrivate(page);
5037                 set_page_private(page, 0);
5038                 page_cache_release(page);
5039         }
5040 }
5041
5042 /*
5043  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5044  * called from a page fault handler when a page is first dirtied. Hence we must
5045  * be careful to check for EOF conditions here. We set the page up correctly
5046  * for a written page which means we get ENOSPC checking when writing into
5047  * holes and correct delalloc and unwritten extent mapping on filesystems that
5048  * support these features.
5049  *
5050  * We are not allowed to take the i_mutex here so we have to play games to
5051  * protect against truncate races as the page could now be beyond EOF.  Because
5052  * vmtruncate() writes the inode size before removing pages, once we have the
5053  * page lock we can determine safely if the page is beyond EOF. If it is not
5054  * beyond EOF, then the page is guaranteed safe against truncation until we
5055  * unlock the page.
5056  */
5057 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5058 {
5059         struct page *page = vmf->page;
5060         struct inode *inode = fdentry(vma->vm_file)->d_inode;
5061         struct btrfs_root *root = BTRFS_I(inode)->root;
5062         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5063         struct btrfs_ordered_extent *ordered;
5064         char *kaddr;
5065         unsigned long zero_start;
5066         loff_t size;
5067         int ret;
5068         u64 page_start;
5069         u64 page_end;
5070
5071         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5072         if (ret) {
5073                 if (ret == -ENOMEM)
5074                         ret = VM_FAULT_OOM;
5075                 else /* -ENOSPC, -EIO, etc */
5076                         ret = VM_FAULT_SIGBUS;
5077                 goto out;
5078         }
5079
5080         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5081         if (ret) {
5082                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5083                 ret = VM_FAULT_SIGBUS;
5084                 goto out;
5085         }
5086
5087         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5088 again:
5089         lock_page(page);
5090         size = i_size_read(inode);
5091         page_start = page_offset(page);
5092         page_end = page_start + PAGE_CACHE_SIZE - 1;
5093
5094         if ((page->mapping != inode->i_mapping) ||
5095             (page_start >= size)) {
5096                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5097                 /* page got truncated out from underneath us */
5098                 goto out_unlock;
5099         }
5100         wait_on_page_writeback(page);
5101
5102         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5103         set_page_extent_mapped(page);
5104
5105         /*
5106          * we can't set the delalloc bits if there are pending ordered
5107          * extents.  Drop our locks and wait for them to finish
5108          */
5109         ordered = btrfs_lookup_ordered_extent(inode, page_start);
5110         if (ordered) {
5111                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5112                 unlock_page(page);
5113                 btrfs_start_ordered_extent(inode, ordered, 1);
5114                 btrfs_put_ordered_extent(ordered);
5115                 goto again;
5116         }
5117
5118         /*
5119          * XXX - page_mkwrite gets called every time the page is dirtied, even
5120          * if it was already dirty, so for space accounting reasons we need to
5121          * clear any delalloc bits for the range we are fixing to save.  There
5122          * is probably a better way to do this, but for now keep consistent with
5123          * prepare_pages in the normal write path.
5124          */
5125         clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
5126                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5127                           GFP_NOFS);
5128
5129         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
5130         if (ret) {
5131                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5132                 ret = VM_FAULT_SIGBUS;
5133                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5134                 goto out_unlock;
5135         }
5136         ret = 0;
5137
5138         /* page is wholly or partially inside EOF */
5139         if (page_start + PAGE_CACHE_SIZE > size)
5140                 zero_start = size & ~PAGE_CACHE_MASK;
5141         else
5142                 zero_start = PAGE_CACHE_SIZE;
5143
5144         if (zero_start != PAGE_CACHE_SIZE) {
5145                 kaddr = kmap(page);
5146                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5147                 flush_dcache_page(page);
5148                 kunmap(page);
5149         }
5150         ClearPageChecked(page);
5151         set_page_dirty(page);
5152         SetPageUptodate(page);
5153
5154         BTRFS_I(inode)->last_trans = root->fs_info->generation;
5155         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5156
5157         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5158
5159 out_unlock:
5160         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5161         if (!ret)
5162                 return VM_FAULT_LOCKED;
5163         unlock_page(page);
5164 out:
5165         return ret;
5166 }
5167
5168 static void btrfs_truncate(struct inode *inode)
5169 {
5170         struct btrfs_root *root = BTRFS_I(inode)->root;
5171         int ret;
5172         struct btrfs_trans_handle *trans;
5173         unsigned long nr;
5174         u64 mask = root->sectorsize - 1;
5175
5176         if (!S_ISREG(inode->i_mode)) {
5177                 WARN_ON(1);
5178                 return;
5179         }
5180
5181         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5182         if (ret)
5183                 return;
5184
5185         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5186         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5187
5188         trans = btrfs_start_transaction(root, 1);
5189         btrfs_set_trans_block_group(trans, inode);
5190
5191         /*
5192          * setattr is responsible for setting the ordered_data_close flag,
5193          * but that is only tested during the last file release.  That
5194          * could happen well after the next commit, leaving a great big
5195          * window where new writes may get lost if someone chooses to write
5196          * to this file after truncating to zero
5197          *
5198          * The inode doesn't have any dirty data here, and so if we commit
5199          * this is a noop.  If someone immediately starts writing to the inode
5200          * it is very likely we'll catch some of their writes in this
5201          * transaction, and the commit will find this file on the ordered
5202          * data list with good things to send down.
5203          *
5204          * This is a best effort solution, there is still a window where
5205          * using truncate to replace the contents of the file will
5206          * end up with a zero length file after a crash.
5207          */
5208         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5209                 btrfs_add_ordered_operation(trans, root, inode);
5210
5211         while (1) {
5212                 ret = btrfs_truncate_inode_items(trans, root, inode,
5213                                                  inode->i_size,
5214                                                  BTRFS_EXTENT_DATA_KEY);
5215                 if (ret != -EAGAIN)
5216                         break;
5217
5218                 ret = btrfs_update_inode(trans, root, inode);
5219                 BUG_ON(ret);
5220
5221                 nr = trans->blocks_used;
5222                 btrfs_end_transaction(trans, root);
5223                 btrfs_btree_balance_dirty(root, nr);
5224
5225                 trans = btrfs_start_transaction(root, 1);
5226                 btrfs_set_trans_block_group(trans, inode);
5227         }
5228
5229         if (ret == 0 && inode->i_nlink > 0) {
5230                 ret = btrfs_orphan_del(trans, inode);
5231                 BUG_ON(ret);
5232         }
5233
5234         ret = btrfs_update_inode(trans, root, inode);
5235         BUG_ON(ret);
5236
5237         nr = trans->blocks_used;
5238         ret = btrfs_end_transaction_throttle(trans, root);
5239         BUG_ON(ret);
5240         btrfs_btree_balance_dirty(root, nr);
5241 }
5242
5243 /*
5244  * create a new subvolume directory/inode (helper for the ioctl).
5245  */
5246 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5247                              struct btrfs_root *new_root,
5248                              u64 new_dirid, u64 alloc_hint)
5249 {
5250         struct inode *inode;
5251         int err;
5252         u64 index = 0;
5253
5254         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5255                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5256         if (IS_ERR(inode))
5257                 return PTR_ERR(inode);
5258         inode->i_op = &btrfs_dir_inode_operations;
5259         inode->i_fop = &btrfs_dir_file_operations;
5260
5261         inode->i_nlink = 1;
5262         btrfs_i_size_write(inode, 0);
5263
5264         err = btrfs_update_inode(trans, new_root, inode);
5265         BUG_ON(err);
5266
5267         iput(inode);
5268         return 0;
5269 }
5270
5271 /* helper function for file defrag and space balancing.  This
5272  * forces readahead on a given range of bytes in an inode
5273  */
5274 unsigned long btrfs_force_ra(struct address_space *mapping,
5275                               struct file_ra_state *ra, struct file *file,
5276                               pgoff_t offset, pgoff_t last_index)
5277 {
5278         pgoff_t req_size = last_index - offset + 1;
5279
5280         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5281         return offset + req_size;
5282 }
5283
5284 struct inode *btrfs_alloc_inode(struct super_block *sb)
5285 {
5286         struct btrfs_inode *ei;
5287
5288         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5289         if (!ei)
5290                 return NULL;
5291         ei->last_trans = 0;
5292         ei->last_sub_trans = 0;
5293         ei->logged_trans = 0;
5294         ei->outstanding_extents = 0;
5295         ei->reserved_extents = 0;
5296         ei->root = NULL;
5297         spin_lock_init(&ei->accounting_lock);
5298         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5299         INIT_LIST_HEAD(&ei->i_orphan);
5300         INIT_LIST_HEAD(&ei->ordered_operations);
5301         return &ei->vfs_inode;
5302 }
5303
5304 void btrfs_destroy_inode(struct inode *inode)
5305 {
5306         struct btrfs_ordered_extent *ordered;
5307         struct btrfs_root *root = BTRFS_I(inode)->root;
5308
5309         WARN_ON(!list_empty(&inode->i_dentry));
5310         WARN_ON(inode->i_data.nrpages);
5311
5312         /*
5313          * This can happen where we create an inode, but somebody else also
5314          * created the same inode and we need to destroy the one we already
5315          * created.
5316          */
5317         if (!root)
5318                 goto free;
5319
5320         /*
5321          * Make sure we're properly removed from the ordered operation
5322          * lists.
5323          */
5324         smp_mb();
5325         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5326                 spin_lock(&root->fs_info->ordered_extent_lock);
5327                 list_del_init(&BTRFS_I(inode)->ordered_operations);
5328                 spin_unlock(&root->fs_info->ordered_extent_lock);
5329         }
5330
5331         spin_lock(&root->list_lock);
5332         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5333                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5334                        inode->i_ino);
5335                 list_del_init(&BTRFS_I(inode)->i_orphan);
5336         }
5337         spin_unlock(&root->list_lock);
5338
5339         while (1) {
5340                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5341                 if (!ordered)
5342                         break;
5343                 else {
5344                         printk(KERN_ERR "btrfs found ordered "
5345                                "extent %llu %llu on inode cleanup\n",
5346                                (unsigned long long)ordered->file_offset,
5347                                (unsigned long long)ordered->len);
5348                         btrfs_remove_ordered_extent(inode, ordered);
5349                         btrfs_put_ordered_extent(ordered);
5350                         btrfs_put_ordered_extent(ordered);
5351                 }
5352         }
5353         inode_tree_del(inode);
5354         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5355 free:
5356         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5357 }
5358
5359 void btrfs_drop_inode(struct inode *inode)
5360 {
5361         struct btrfs_root *root = BTRFS_I(inode)->root;
5362
5363         if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5364                 generic_delete_inode(inode);
5365         else
5366                 generic_drop_inode(inode);
5367 }
5368
5369 static void init_once(void *foo)
5370 {
5371         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5372
5373         inode_init_once(&ei->vfs_inode);
5374 }
5375
5376 void btrfs_destroy_cachep(void)
5377 {
5378         if (btrfs_inode_cachep)
5379                 kmem_cache_destroy(btrfs_inode_cachep);
5380         if (btrfs_trans_handle_cachep)
5381                 kmem_cache_destroy(btrfs_trans_handle_cachep);
5382         if (btrfs_transaction_cachep)
5383                 kmem_cache_destroy(btrfs_transaction_cachep);
5384         if (btrfs_path_cachep)
5385                 kmem_cache_destroy(btrfs_path_cachep);
5386 }
5387
5388 int btrfs_init_cachep(void)
5389 {
5390         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5391                         sizeof(struct btrfs_inode), 0,
5392                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5393         if (!btrfs_inode_cachep)
5394                 goto fail;
5395
5396         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5397                         sizeof(struct btrfs_trans_handle), 0,
5398                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5399         if (!btrfs_trans_handle_cachep)
5400                 goto fail;
5401
5402         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5403                         sizeof(struct btrfs_transaction), 0,
5404                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5405         if (!btrfs_transaction_cachep)
5406                 goto fail;
5407
5408         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5409                         sizeof(struct btrfs_path), 0,
5410                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5411         if (!btrfs_path_cachep)
5412                 goto fail;
5413
5414         return 0;
5415 fail:
5416         btrfs_destroy_cachep();
5417         return -ENOMEM;
5418 }
5419
5420 static int btrfs_getattr(struct vfsmount *mnt,
5421                          struct dentry *dentry, struct kstat *stat)
5422 {
5423         struct inode *inode = dentry->d_inode;
5424         generic_fillattr(inode, stat);
5425         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5426         stat->blksize = PAGE_CACHE_SIZE;
5427         stat->blocks = (inode_get_bytes(inode) +
5428                         BTRFS_I(inode)->delalloc_bytes) >> 9;
5429         return 0;
5430 }
5431
5432 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5433                            struct inode *new_dir, struct dentry *new_dentry)
5434 {
5435         struct btrfs_trans_handle *trans;
5436         struct btrfs_root *root = BTRFS_I(old_dir)->root;
5437         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5438         struct inode *new_inode = new_dentry->d_inode;
5439         struct inode *old_inode = old_dentry->d_inode;
5440         struct timespec ctime = CURRENT_TIME;
5441         u64 index = 0;
5442         u64 root_objectid;
5443         int ret;
5444
5445         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5446                 return -EPERM;
5447
5448         /* we only allow rename subvolume link between subvolumes */
5449         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5450                 return -EXDEV;
5451
5452         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5453             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5454                 return -ENOTEMPTY;
5455
5456         if (S_ISDIR(old_inode->i_mode) && new_inode &&
5457             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5458                 return -ENOTEMPTY;
5459
5460         /*
5461          * We want to reserve the absolute worst case amount of items.  So if
5462          * both inodes are subvols and we need to unlink them then that would
5463          * require 4 item modifications, but if they are both normal inodes it
5464          * would require 5 item modifications, so we'll assume their normal
5465          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5466          * should cover the worst case number of items we'll modify.
5467          */
5468         ret = btrfs_reserve_metadata_space(root, 11);
5469         if (ret)
5470                 return ret;
5471
5472         /*
5473          * we're using rename to replace one file with another.
5474          * and the replacement file is large.  Start IO on it now so
5475          * we don't add too much work to the end of the transaction
5476          */
5477         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5478             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5479                 filemap_flush(old_inode->i_mapping);
5480
5481         /* close the racy window with snapshot create/destroy ioctl */
5482         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5483                 down_read(&root->fs_info->subvol_sem);
5484
5485         trans = btrfs_start_transaction(root, 1);
5486         btrfs_set_trans_block_group(trans, new_dir);
5487
5488         if (dest != root)
5489                 btrfs_record_root_in_trans(trans, dest);
5490
5491         ret = btrfs_set_inode_index(new_dir, &index);
5492         if (ret)
5493                 goto out_fail;
5494
5495         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5496                 /* force full log commit if subvolume involved. */
5497                 root->fs_info->last_trans_log_full_commit = trans->transid;
5498         } else {
5499                 ret = btrfs_insert_inode_ref(trans, dest,
5500                                              new_dentry->d_name.name,
5501                                              new_dentry->d_name.len,
5502                                              old_inode->i_ino,
5503                                              new_dir->i_ino, index);
5504                 if (ret)
5505                         goto out_fail;
5506                 /*
5507                  * this is an ugly little race, but the rename is required
5508                  * to make sure that if we crash, the inode is either at the
5509                  * old name or the new one.  pinning the log transaction lets
5510                  * us make sure we don't allow a log commit to come in after
5511                  * we unlink the name but before we add the new name back in.
5512                  */
5513                 btrfs_pin_log_trans(root);
5514         }
5515         /*
5516          * make sure the inode gets flushed if it is replacing
5517          * something.
5518          */
5519         if (new_inode && new_inode->i_size &&
5520             old_inode && S_ISREG(old_inode->i_mode)) {
5521                 btrfs_add_ordered_operation(trans, root, old_inode);
5522         }
5523
5524         old_dir->i_ctime = old_dir->i_mtime = ctime;
5525         new_dir->i_ctime = new_dir->i_mtime = ctime;
5526         old_inode->i_ctime = ctime;
5527
5528         if (old_dentry->d_parent != new_dentry->d_parent)
5529                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5530
5531         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5532                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5533                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5534                                         old_dentry->d_name.name,
5535                                         old_dentry->d_name.len);
5536         } else {
5537                 btrfs_inc_nlink(old_dentry->d_inode);
5538                 ret = btrfs_unlink_inode(trans, root, old_dir,
5539                                          old_dentry->d_inode,
5540                                          old_dentry->d_name.name,
5541                                          old_dentry->d_name.len);
5542         }
5543         BUG_ON(ret);
5544
5545         if (new_inode) {
5546                 new_inode->i_ctime = CURRENT_TIME;
5547                 if (unlikely(new_inode->i_ino ==
5548                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5549                         root_objectid = BTRFS_I(new_inode)->location.objectid;
5550                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
5551                                                 root_objectid,
5552                                                 new_dentry->d_name.name,
5553                                                 new_dentry->d_name.len);
5554                         BUG_ON(new_inode->i_nlink == 0);
5555                 } else {
5556                         ret = btrfs_unlink_inode(trans, dest, new_dir,
5557                                                  new_dentry->d_inode,
5558                                                  new_dentry->d_name.name,
5559                                                  new_dentry->d_name.len);
5560                 }
5561                 BUG_ON(ret);
5562                 if (new_inode->i_nlink == 0) {
5563                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5564                         BUG_ON(ret);
5565                 }
5566         }
5567
5568         ret = btrfs_add_link(trans, new_dir, old_inode,
5569                              new_dentry->d_name.name,
5570                              new_dentry->d_name.len, 0, index);
5571         BUG_ON(ret);
5572
5573         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5574                 btrfs_log_new_name(trans, old_inode, old_dir,
5575                                    new_dentry->d_parent);
5576                 btrfs_end_log_trans(root);
5577         }
5578 out_fail:
5579         btrfs_end_transaction_throttle(trans, root);
5580
5581         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5582                 up_read(&root->fs_info->subvol_sem);
5583
5584         btrfs_unreserve_metadata_space(root, 11);
5585         return ret;
5586 }
5587
5588 /*
5589  * some fairly slow code that needs optimization. This walks the list
5590  * of all the inodes with pending delalloc and forces them to disk.
5591  */
5592 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5593 {
5594         struct list_head *head = &root->fs_info->delalloc_inodes;
5595         struct btrfs_inode *binode;
5596         struct inode *inode;
5597
5598         if (root->fs_info->sb->s_flags & MS_RDONLY)
5599                 return -EROFS;
5600
5601         spin_lock(&root->fs_info->delalloc_lock);
5602         while (!list_empty(head)) {
5603                 binode = list_entry(head->next, struct btrfs_inode,
5604                                     delalloc_inodes);
5605                 inode = igrab(&binode->vfs_inode);
5606                 if (!inode)
5607                         list_del_init(&binode->delalloc_inodes);
5608                 spin_unlock(&root->fs_info->delalloc_lock);
5609                 if (inode) {
5610                         filemap_flush(inode->i_mapping);
5611                         if (delay_iput)
5612                                 btrfs_add_delayed_iput(inode);
5613                         else
5614                                 iput(inode);
5615                 }
5616                 cond_resched();
5617                 spin_lock(&root->fs_info->delalloc_lock);
5618         }
5619         spin_unlock(&root->fs_info->delalloc_lock);
5620
5621         /* the filemap_flush will queue IO into the worker threads, but
5622          * we have to make sure the IO is actually started and that
5623          * ordered extents get created before we return
5624          */
5625         atomic_inc(&root->fs_info->async_submit_draining);
5626         while (atomic_read(&root->fs_info->nr_async_submits) ||
5627               atomic_read(&root->fs_info->async_delalloc_pages)) {
5628                 wait_event(root->fs_info->async_submit_wait,
5629                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5630                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5631         }
5632         atomic_dec(&root->fs_info->async_submit_draining);
5633         return 0;
5634 }
5635
5636 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5637                          const char *symname)
5638 {
5639         struct btrfs_trans_handle *trans;
5640         struct btrfs_root *root = BTRFS_I(dir)->root;
5641         struct btrfs_path *path;
5642         struct btrfs_key key;
5643         struct inode *inode = NULL;
5644         int err;
5645         int drop_inode = 0;
5646         u64 objectid;
5647         u64 index = 0 ;
5648         int name_len;
5649         int datasize;
5650         unsigned long ptr;
5651         struct btrfs_file_extent_item *ei;
5652         struct extent_buffer *leaf;
5653         unsigned long nr = 0;
5654
5655         name_len = strlen(symname) + 1;
5656         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5657                 return -ENAMETOOLONG;
5658
5659         /*
5660          * 2 items for inode item and ref
5661          * 2 items for dir items
5662          * 1 item for xattr if selinux is on
5663          */
5664         err = btrfs_reserve_metadata_space(root, 5);
5665         if (err)
5666                 return err;
5667
5668         trans = btrfs_start_transaction(root, 1);
5669         if (!trans)
5670                 goto out_fail;
5671         btrfs_set_trans_block_group(trans, dir);
5672
5673         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5674         if (err) {
5675                 err = -ENOSPC;
5676                 goto out_unlock;
5677         }
5678
5679         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5680                                 dentry->d_name.len,
5681                                 dentry->d_parent->d_inode->i_ino, objectid,
5682                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5683                                 &index);
5684         err = PTR_ERR(inode);
5685         if (IS_ERR(inode))
5686                 goto out_unlock;
5687
5688         err = btrfs_init_inode_security(trans, inode, dir);
5689         if (err) {
5690                 drop_inode = 1;
5691                 goto out_unlock;
5692         }
5693
5694         btrfs_set_trans_block_group(trans, inode);
5695         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5696         if (err)
5697                 drop_inode = 1;
5698         else {
5699                 inode->i_mapping->a_ops = &btrfs_aops;
5700                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5701                 inode->i_fop = &btrfs_file_operations;
5702                 inode->i_op = &btrfs_file_inode_operations;
5703           &nb