new helper: ihold()
[linux-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93                                      struct inode *inode,  struct inode *dir)
94 {
95         int err;
96
97         err = btrfs_init_acl(trans, inode, dir);
98         if (!err)
99                 err = btrfs_xattr_security_init(trans, inode, dir);
100         return err;
101 }
102
103 /*
104  * this does all the hard work for inserting an inline extent into
105  * the btree.  The caller should have done a btrfs_drop_extents so that
106  * no overlapping inline items exist in the btree
107  */
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109                                 struct btrfs_root *root, struct inode *inode,
110                                 u64 start, size_t size, size_t compressed_size,
111                                 struct page **compressed_pages)
112 {
113         struct btrfs_key key;
114         struct btrfs_path *path;
115         struct extent_buffer *leaf;
116         struct page *page = NULL;
117         char *kaddr;
118         unsigned long ptr;
119         struct btrfs_file_extent_item *ei;
120         int err = 0;
121         int ret;
122         size_t cur_size = size;
123         size_t datasize;
124         unsigned long offset;
125         int use_compress = 0;
126
127         if (compressed_size && compressed_pages) {
128                 use_compress = 1;
129                 cur_size = compressed_size;
130         }
131
132         path = btrfs_alloc_path();
133         if (!path)
134                 return -ENOMEM;
135
136         path->leave_spinning = 1;
137         btrfs_set_trans_block_group(trans, inode);
138
139         key.objectid = inode->i_ino;
140         key.offset = start;
141         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142         datasize = btrfs_file_extent_calc_inline_size(cur_size);
143
144         inode_add_bytes(inode, size);
145         ret = btrfs_insert_empty_item(trans, root, path, &key,
146                                       datasize);
147         BUG_ON(ret);
148         if (ret) {
149                 err = ret;
150                 goto fail;
151         }
152         leaf = path->nodes[0];
153         ei = btrfs_item_ptr(leaf, path->slots[0],
154                             struct btrfs_file_extent_item);
155         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157         btrfs_set_file_extent_encryption(leaf, ei, 0);
158         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160         ptr = btrfs_file_extent_inline_start(ei);
161
162         if (use_compress) {
163                 struct page *cpage;
164                 int i = 0;
165                 while (compressed_size > 0) {
166                         cpage = compressed_pages[i];
167                         cur_size = min_t(unsigned long, compressed_size,
168                                        PAGE_CACHE_SIZE);
169
170                         kaddr = kmap_atomic(cpage, KM_USER0);
171                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
172                         kunmap_atomic(kaddr, KM_USER0);
173
174                         i++;
175                         ptr += cur_size;
176                         compressed_size -= cur_size;
177                 }
178                 btrfs_set_file_extent_compression(leaf, ei,
179                                                   BTRFS_COMPRESS_ZLIB);
180         } else {
181                 page = find_get_page(inode->i_mapping,
182                                      start >> PAGE_CACHE_SHIFT);
183                 btrfs_set_file_extent_compression(leaf, ei, 0);
184                 kaddr = kmap_atomic(page, KM_USER0);
185                 offset = start & (PAGE_CACHE_SIZE - 1);
186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187                 kunmap_atomic(kaddr, KM_USER0);
188                 page_cache_release(page);
189         }
190         btrfs_mark_buffer_dirty(leaf);
191         btrfs_free_path(path);
192
193         /*
194          * we're an inline extent, so nobody can
195          * extend the file past i_size without locking
196          * a page we already have locked.
197          *
198          * We must do any isize and inode updates
199          * before we unlock the pages.  Otherwise we
200          * could end up racing with unlink.
201          */
202         BTRFS_I(inode)->disk_i_size = inode->i_size;
203         btrfs_update_inode(trans, root, inode);
204
205         return 0;
206 fail:
207         btrfs_free_path(path);
208         return err;
209 }
210
211
212 /*
213  * conditionally insert an inline extent into the file.  This
214  * does the checks required to make sure the data is small enough
215  * to fit as an inline extent.
216  */
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218                                  struct btrfs_root *root,
219                                  struct inode *inode, u64 start, u64 end,
220                                  size_t compressed_size,
221                                  struct page **compressed_pages)
222 {
223         u64 isize = i_size_read(inode);
224         u64 actual_end = min(end + 1, isize);
225         u64 inline_len = actual_end - start;
226         u64 aligned_end = (end + root->sectorsize - 1) &
227                         ~((u64)root->sectorsize - 1);
228         u64 hint_byte;
229         u64 data_len = inline_len;
230         int ret;
231
232         if (compressed_size)
233                 data_len = compressed_size;
234
235         if (start > 0 ||
236             actual_end >= PAGE_CACHE_SIZE ||
237             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238             (!compressed_size &&
239             (actual_end & (root->sectorsize - 1)) == 0) ||
240             end + 1 < isize ||
241             data_len > root->fs_info->max_inline) {
242                 return 1;
243         }
244
245         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
246                                  &hint_byte, 1);
247         BUG_ON(ret);
248
249         if (isize > actual_end)
250                 inline_len = min_t(u64, isize, actual_end);
251         ret = insert_inline_extent(trans, root, inode, start,
252                                    inline_len, compressed_size,
253                                    compressed_pages);
254         BUG_ON(ret);
255         btrfs_delalloc_release_metadata(inode, end + 1 - start);
256         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
257         return 0;
258 }
259
260 struct async_extent {
261         u64 start;
262         u64 ram_size;
263         u64 compressed_size;
264         struct page **pages;
265         unsigned long nr_pages;
266         struct list_head list;
267 };
268
269 struct async_cow {
270         struct inode *inode;
271         struct btrfs_root *root;
272         struct page *locked_page;
273         u64 start;
274         u64 end;
275         struct list_head extents;
276         struct btrfs_work work;
277 };
278
279 static noinline int add_async_extent(struct async_cow *cow,
280                                      u64 start, u64 ram_size,
281                                      u64 compressed_size,
282                                      struct page **pages,
283                                      unsigned long nr_pages)
284 {
285         struct async_extent *async_extent;
286
287         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
288         async_extent->start = start;
289         async_extent->ram_size = ram_size;
290         async_extent->compressed_size = compressed_size;
291         async_extent->pages = pages;
292         async_extent->nr_pages = nr_pages;
293         list_add_tail(&async_extent->list, &cow->extents);
294         return 0;
295 }
296
297 /*
298  * we create compressed extents in two phases.  The first
299  * phase compresses a range of pages that have already been
300  * locked (both pages and state bits are locked).
301  *
302  * This is done inside an ordered work queue, and the compression
303  * is spread across many cpus.  The actual IO submission is step
304  * two, and the ordered work queue takes care of making sure that
305  * happens in the same order things were put onto the queue by
306  * writepages and friends.
307  *
308  * If this code finds it can't get good compression, it puts an
309  * entry onto the work queue to write the uncompressed bytes.  This
310  * makes sure that both compressed inodes and uncompressed inodes
311  * are written in the same order that pdflush sent them down.
312  */
313 static noinline int compress_file_range(struct inode *inode,
314                                         struct page *locked_page,
315                                         u64 start, u64 end,
316                                         struct async_cow *async_cow,
317                                         int *num_added)
318 {
319         struct btrfs_root *root = BTRFS_I(inode)->root;
320         struct btrfs_trans_handle *trans;
321         u64 num_bytes;
322         u64 orig_start;
323         u64 disk_num_bytes;
324         u64 blocksize = root->sectorsize;
325         u64 actual_end;
326         u64 isize = i_size_read(inode);
327         int ret = 0;
328         struct page **pages = NULL;
329         unsigned long nr_pages;
330         unsigned long nr_pages_ret = 0;
331         unsigned long total_compressed = 0;
332         unsigned long total_in = 0;
333         unsigned long max_compressed = 128 * 1024;
334         unsigned long max_uncompressed = 128 * 1024;
335         int i;
336         int will_compress;
337
338         orig_start = start;
339
340         actual_end = min_t(u64, isize, end + 1);
341 again:
342         will_compress = 0;
343         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
345
346         /*
347          * we don't want to send crud past the end of i_size through
348          * compression, that's just a waste of CPU time.  So, if the
349          * end of the file is before the start of our current
350          * requested range of bytes, we bail out to the uncompressed
351          * cleanup code that can deal with all of this.
352          *
353          * It isn't really the fastest way to fix things, but this is a
354          * very uncommon corner.
355          */
356         if (actual_end <= start)
357                 goto cleanup_and_bail_uncompressed;
358
359         total_compressed = actual_end - start;
360
361         /* we want to make sure that amount of ram required to uncompress
362          * an extent is reasonable, so we limit the total size in ram
363          * of a compressed extent to 128k.  This is a crucial number
364          * because it also controls how easily we can spread reads across
365          * cpus for decompression.
366          *
367          * We also want to make sure the amount of IO required to do
368          * a random read is reasonably small, so we limit the size of
369          * a compressed extent to 128k.
370          */
371         total_compressed = min(total_compressed, max_uncompressed);
372         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373         num_bytes = max(blocksize,  num_bytes);
374         disk_num_bytes = num_bytes;
375         total_in = 0;
376         ret = 0;
377
378         /*
379          * we do compression for mount -o compress and when the
380          * inode has not been flagged as nocompress.  This flag can
381          * change at any time if we discover bad compression ratios.
382          */
383         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384             (btrfs_test_opt(root, COMPRESS) ||
385              (BTRFS_I(inode)->force_compress))) {
386                 WARN_ON(pages);
387                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388
389                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
390                                                 total_compressed, pages,
391                                                 nr_pages, &nr_pages_ret,
392                                                 &total_in,
393                                                 &total_compressed,
394                                                 max_compressed);
395
396                 if (!ret) {
397                         unsigned long offset = total_compressed &
398                                 (PAGE_CACHE_SIZE - 1);
399                         struct page *page = pages[nr_pages_ret - 1];
400                         char *kaddr;
401
402                         /* zero the tail end of the last page, we might be
403                          * sending it down to disk
404                          */
405                         if (offset) {
406                                 kaddr = kmap_atomic(page, KM_USER0);
407                                 memset(kaddr + offset, 0,
408                                        PAGE_CACHE_SIZE - offset);
409                                 kunmap_atomic(kaddr, KM_USER0);
410                         }
411                         will_compress = 1;
412                 }
413         }
414         if (start == 0) {
415                 trans = btrfs_join_transaction(root, 1);
416                 BUG_ON(!trans);
417                 btrfs_set_trans_block_group(trans, inode);
418                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
419
420                 /* lets try to make an inline extent */
421                 if (ret || total_in < (actual_end - start)) {
422                         /* we didn't compress the entire range, try
423                          * to make an uncompressed inline extent.
424                          */
425                         ret = cow_file_range_inline(trans, root, inode,
426                                                     start, end, 0, NULL);
427                 } else {
428                         /* try making a compressed inline extent */
429                         ret = cow_file_range_inline(trans, root, inode,
430                                                     start, end,
431                                                     total_compressed, pages);
432                 }
433                 if (ret == 0) {
434                         /*
435                          * inline extent creation worked, we don't need
436                          * to create any more async work items.  Unlock
437                          * and free up our temp pages.
438                          */
439                         extent_clear_unlock_delalloc(inode,
440                              &BTRFS_I(inode)->io_tree,
441                              start, end, NULL,
442                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
443                              EXTENT_CLEAR_DELALLOC |
444                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
445
446                         btrfs_end_transaction(trans, root);
447                         goto free_pages_out;
448                 }
449                 btrfs_end_transaction(trans, root);
450         }
451
452         if (will_compress) {
453                 /*
454                  * we aren't doing an inline extent round the compressed size
455                  * up to a block size boundary so the allocator does sane
456                  * things
457                  */
458                 total_compressed = (total_compressed + blocksize - 1) &
459                         ~(blocksize - 1);
460
461                 /*
462                  * one last check to make sure the compression is really a
463                  * win, compare the page count read with the blocks on disk
464                  */
465                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
466                         ~(PAGE_CACHE_SIZE - 1);
467                 if (total_compressed >= total_in) {
468                         will_compress = 0;
469                 } else {
470                         disk_num_bytes = total_compressed;
471                         num_bytes = total_in;
472                 }
473         }
474         if (!will_compress && pages) {
475                 /*
476                  * the compression code ran but failed to make things smaller,
477                  * free any pages it allocated and our page pointer array
478                  */
479                 for (i = 0; i < nr_pages_ret; i++) {
480                         WARN_ON(pages[i]->mapping);
481                         page_cache_release(pages[i]);
482                 }
483                 kfree(pages);
484                 pages = NULL;
485                 total_compressed = 0;
486                 nr_pages_ret = 0;
487
488                 /* flag the file so we don't compress in the future */
489                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
490                     !(BTRFS_I(inode)->force_compress)) {
491                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
492                 }
493         }
494         if (will_compress) {
495                 *num_added += 1;
496
497                 /* the async work queues will take care of doing actual
498                  * allocation on disk for these compressed pages,
499                  * and will submit them to the elevator.
500                  */
501                 add_async_extent(async_cow, start, num_bytes,
502                                  total_compressed, pages, nr_pages_ret);
503
504                 if (start + num_bytes < end && start + num_bytes < actual_end) {
505                         start += num_bytes;
506                         pages = NULL;
507                         cond_resched();
508                         goto again;
509                 }
510         } else {
511 cleanup_and_bail_uncompressed:
512                 /*
513                  * No compression, but we still need to write the pages in
514                  * the file we've been given so far.  redirty the locked
515                  * page if it corresponds to our extent and set things up
516                  * for the async work queue to run cow_file_range to do
517                  * the normal delalloc dance
518                  */
519                 if (page_offset(locked_page) >= start &&
520                     page_offset(locked_page) <= end) {
521                         __set_page_dirty_nobuffers(locked_page);
522                         /* unlocked later on in the async handlers */
523                 }
524                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
525                 *num_added += 1;
526         }
527
528 out:
529         return 0;
530
531 free_pages_out:
532         for (i = 0; i < nr_pages_ret; i++) {
533                 WARN_ON(pages[i]->mapping);
534                 page_cache_release(pages[i]);
535         }
536         kfree(pages);
537
538         goto out;
539 }
540
541 /*
542  * phase two of compressed writeback.  This is the ordered portion
543  * of the code, which only gets called in the order the work was
544  * queued.  We walk all the async extents created by compress_file_range
545  * and send them down to the disk.
546  */
547 static noinline int submit_compressed_extents(struct inode *inode,
548                                               struct async_cow *async_cow)
549 {
550         struct async_extent *async_extent;
551         u64 alloc_hint = 0;
552         struct btrfs_trans_handle *trans;
553         struct btrfs_key ins;
554         struct extent_map *em;
555         struct btrfs_root *root = BTRFS_I(inode)->root;
556         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
557         struct extent_io_tree *io_tree;
558         int ret = 0;
559
560         if (list_empty(&async_cow->extents))
561                 return 0;
562
563
564         while (!list_empty(&async_cow->extents)) {
565                 async_extent = list_entry(async_cow->extents.next,
566                                           struct async_extent, list);
567                 list_del(&async_extent->list);
568
569                 io_tree = &BTRFS_I(inode)->io_tree;
570
571 retry:
572                 /* did the compression code fall back to uncompressed IO? */
573                 if (!async_extent->pages) {
574                         int page_started = 0;
575                         unsigned long nr_written = 0;
576
577                         lock_extent(io_tree, async_extent->start,
578                                          async_extent->start +
579                                          async_extent->ram_size - 1, GFP_NOFS);
580
581                         /* allocate blocks */
582                         ret = cow_file_range(inode, async_cow->locked_page,
583                                              async_extent->start,
584                                              async_extent->start +
585                                              async_extent->ram_size - 1,
586                                              &page_started, &nr_written, 0);
587
588                         /*
589                          * if page_started, cow_file_range inserted an
590                          * inline extent and took care of all the unlocking
591                          * and IO for us.  Otherwise, we need to submit
592                          * all those pages down to the drive.
593                          */
594                         if (!page_started && !ret)
595                                 extent_write_locked_range(io_tree,
596                                                   inode, async_extent->start,
597                                                   async_extent->start +
598                                                   async_extent->ram_size - 1,
599                                                   btrfs_get_extent,
600                                                   WB_SYNC_ALL);
601                         kfree(async_extent);
602                         cond_resched();
603                         continue;
604                 }
605
606                 lock_extent(io_tree, async_extent->start,
607                             async_extent->start + async_extent->ram_size - 1,
608                             GFP_NOFS);
609
610                 trans = btrfs_join_transaction(root, 1);
611                 ret = btrfs_reserve_extent(trans, root,
612                                            async_extent->compressed_size,
613                                            async_extent->compressed_size,
614                                            0, alloc_hint,
615                                            (u64)-1, &ins, 1);
616                 btrfs_end_transaction(trans, root);
617
618                 if (ret) {
619                         int i;
620                         for (i = 0; i < async_extent->nr_pages; i++) {
621                                 WARN_ON(async_extent->pages[i]->mapping);
622                                 page_cache_release(async_extent->pages[i]);
623                         }
624                         kfree(async_extent->pages);
625                         async_extent->nr_pages = 0;
626                         async_extent->pages = NULL;
627                         unlock_extent(io_tree, async_extent->start,
628                                       async_extent->start +
629                                       async_extent->ram_size - 1, GFP_NOFS);
630                         goto retry;
631                 }
632
633                 /*
634                  * here we're doing allocation and writeback of the
635                  * compressed pages
636                  */
637                 btrfs_drop_extent_cache(inode, async_extent->start,
638                                         async_extent->start +
639                                         async_extent->ram_size - 1, 0);
640
641                 em = alloc_extent_map(GFP_NOFS);
642                 em->start = async_extent->start;
643                 em->len = async_extent->ram_size;
644                 em->orig_start = em->start;
645
646                 em->block_start = ins.objectid;
647                 em->block_len = ins.offset;
648                 em->bdev = root->fs_info->fs_devices->latest_bdev;
649                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
650                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
651
652                 while (1) {
653                         write_lock(&em_tree->lock);
654                         ret = add_extent_mapping(em_tree, em);
655                         write_unlock(&em_tree->lock);
656                         if (ret != -EEXIST) {
657                                 free_extent_map(em);
658                                 break;
659                         }
660                         btrfs_drop_extent_cache(inode, async_extent->start,
661                                                 async_extent->start +
662                                                 async_extent->ram_size - 1, 0);
663                 }
664
665                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
666                                                ins.objectid,
667                                                async_extent->ram_size,
668                                                ins.offset,
669                                                BTRFS_ORDERED_COMPRESSED);
670                 BUG_ON(ret);
671
672                 /*
673                  * clear dirty, set writeback and unlock the pages.
674                  */
675                 extent_clear_unlock_delalloc(inode,
676                                 &BTRFS_I(inode)->io_tree,
677                                 async_extent->start,
678                                 async_extent->start +
679                                 async_extent->ram_size - 1,
680                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
681                                 EXTENT_CLEAR_UNLOCK |
682                                 EXTENT_CLEAR_DELALLOC |
683                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
684
685                 ret = btrfs_submit_compressed_write(inode,
686                                     async_extent->start,
687                                     async_extent->ram_size,
688                                     ins.objectid,
689                                     ins.offset, async_extent->pages,
690                                     async_extent->nr_pages);
691
692                 BUG_ON(ret);
693                 alloc_hint = ins.objectid + ins.offset;
694                 kfree(async_extent);
695                 cond_resched();
696         }
697
698         return 0;
699 }
700
701 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
702                                       u64 num_bytes)
703 {
704         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
705         struct extent_map *em;
706         u64 alloc_hint = 0;
707
708         read_lock(&em_tree->lock);
709         em = search_extent_mapping(em_tree, start, num_bytes);
710         if (em) {
711                 /*
712                  * if block start isn't an actual block number then find the
713                  * first block in this inode and use that as a hint.  If that
714                  * block is also bogus then just don't worry about it.
715                  */
716                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
717                         free_extent_map(em);
718                         em = search_extent_mapping(em_tree, 0, 0);
719                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
720                                 alloc_hint = em->block_start;
721                         if (em)
722                                 free_extent_map(em);
723                 } else {
724                         alloc_hint = em->block_start;
725                         free_extent_map(em);
726                 }
727         }
728         read_unlock(&em_tree->lock);
729
730         return alloc_hint;
731 }
732
733 /*
734  * when extent_io.c finds a delayed allocation range in the file,
735  * the call backs end up in this code.  The basic idea is to
736  * allocate extents on disk for the range, and create ordered data structs
737  * in ram to track those extents.
738  *
739  * locked_page is the page that writepage had locked already.  We use
740  * it to make sure we don't do extra locks or unlocks.
741  *
742  * *page_started is set to one if we unlock locked_page and do everything
743  * required to start IO on it.  It may be clean and already done with
744  * IO when we return.
745  */
746 static noinline int cow_file_range(struct inode *inode,
747                                    struct page *locked_page,
748                                    u64 start, u64 end, int *page_started,
749                                    unsigned long *nr_written,
750                                    int unlock)
751 {
752         struct btrfs_root *root = BTRFS_I(inode)->root;
753         struct btrfs_trans_handle *trans;
754         u64 alloc_hint = 0;
755         u64 num_bytes;
756         unsigned long ram_size;
757         u64 disk_num_bytes;
758         u64 cur_alloc_size;
759         u64 blocksize = root->sectorsize;
760         u64 actual_end;
761         u64 isize = i_size_read(inode);
762         struct btrfs_key ins;
763         struct extent_map *em;
764         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
765         int ret = 0;
766
767         trans = btrfs_join_transaction(root, 1);
768         BUG_ON(!trans);
769         btrfs_set_trans_block_group(trans, inode);
770         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771
772         actual_end = min_t(u64, isize, end + 1);
773
774         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
775         num_bytes = max(blocksize,  num_bytes);
776         disk_num_bytes = num_bytes;
777         ret = 0;
778
779         if (start == 0) {
780                 /* lets try to make an inline extent */
781                 ret = cow_file_range_inline(trans, root, inode,
782                                             start, end, 0, NULL);
783                 if (ret == 0) {
784                         extent_clear_unlock_delalloc(inode,
785                                      &BTRFS_I(inode)->io_tree,
786                                      start, end, NULL,
787                                      EXTENT_CLEAR_UNLOCK_PAGE |
788                                      EXTENT_CLEAR_UNLOCK |
789                                      EXTENT_CLEAR_DELALLOC |
790                                      EXTENT_CLEAR_DIRTY |
791                                      EXTENT_SET_WRITEBACK |
792                                      EXTENT_END_WRITEBACK);
793
794                         *nr_written = *nr_written +
795                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
796                         *page_started = 1;
797                         ret = 0;
798                         goto out;
799                 }
800         }
801
802         BUG_ON(disk_num_bytes >
803                btrfs_super_total_bytes(&root->fs_info->super_copy));
804
805         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
806         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
807
808         while (disk_num_bytes > 0) {
809                 unsigned long op;
810
811                 cur_alloc_size = disk_num_bytes;
812                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
813                                            root->sectorsize, 0, alloc_hint,
814                                            (u64)-1, &ins, 1);
815                 BUG_ON(ret);
816
817                 em = alloc_extent_map(GFP_NOFS);
818                 em->start = start;
819                 em->orig_start = em->start;
820                 ram_size = ins.offset;
821                 em->len = ins.offset;
822
823                 em->block_start = ins.objectid;
824                 em->block_len = ins.offset;
825                 em->bdev = root->fs_info->fs_devices->latest_bdev;
826                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
827
828                 while (1) {
829                         write_lock(&em_tree->lock);
830                         ret = add_extent_mapping(em_tree, em);
831                         write_unlock(&em_tree->lock);
832                         if (ret != -EEXIST) {
833                                 free_extent_map(em);
834                                 break;
835                         }
836                         btrfs_drop_extent_cache(inode, start,
837                                                 start + ram_size - 1, 0);
838                 }
839
840                 cur_alloc_size = ins.offset;
841                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
842                                                ram_size, cur_alloc_size, 0);
843                 BUG_ON(ret);
844
845                 if (root->root_key.objectid ==
846                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
847                         ret = btrfs_reloc_clone_csums(inode, start,
848                                                       cur_alloc_size);
849                         BUG_ON(ret);
850                 }
851
852                 if (disk_num_bytes < cur_alloc_size)
853                         break;
854
855                 /* we're not doing compressed IO, don't unlock the first
856                  * page (which the caller expects to stay locked), don't
857                  * clear any dirty bits and don't set any writeback bits
858                  *
859                  * Do set the Private2 bit so we know this page was properly
860                  * setup for writepage
861                  */
862                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
863                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
864                         EXTENT_SET_PRIVATE2;
865
866                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
867                                              start, start + ram_size - 1,
868                                              locked_page, op);
869                 disk_num_bytes -= cur_alloc_size;
870                 num_bytes -= cur_alloc_size;
871                 alloc_hint = ins.objectid + ins.offset;
872                 start += cur_alloc_size;
873         }
874 out:
875         ret = 0;
876         btrfs_end_transaction(trans, root);
877
878         return ret;
879 }
880
881 /*
882  * work queue call back to started compression on a file and pages
883  */
884 static noinline void async_cow_start(struct btrfs_work *work)
885 {
886         struct async_cow *async_cow;
887         int num_added = 0;
888         async_cow = container_of(work, struct async_cow, work);
889
890         compress_file_range(async_cow->inode, async_cow->locked_page,
891                             async_cow->start, async_cow->end, async_cow,
892                             &num_added);
893         if (num_added == 0)
894                 async_cow->inode = NULL;
895 }
896
897 /*
898  * work queue call back to submit previously compressed pages
899  */
900 static noinline void async_cow_submit(struct btrfs_work *work)
901 {
902         struct async_cow *async_cow;
903         struct btrfs_root *root;
904         unsigned long nr_pages;
905
906         async_cow = container_of(work, struct async_cow, work);
907
908         root = async_cow->root;
909         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
910                 PAGE_CACHE_SHIFT;
911
912         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
913
914         if (atomic_read(&root->fs_info->async_delalloc_pages) <
915             5 * 1042 * 1024 &&
916             waitqueue_active(&root->fs_info->async_submit_wait))
917                 wake_up(&root->fs_info->async_submit_wait);
918
919         if (async_cow->inode)
920                 submit_compressed_extents(async_cow->inode, async_cow);
921 }
922
923 static noinline void async_cow_free(struct btrfs_work *work)
924 {
925         struct async_cow *async_cow;
926         async_cow = container_of(work, struct async_cow, work);
927         kfree(async_cow);
928 }
929
930 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
931                                 u64 start, u64 end, int *page_started,
932                                 unsigned long *nr_written)
933 {
934         struct async_cow *async_cow;
935         struct btrfs_root *root = BTRFS_I(inode)->root;
936         unsigned long nr_pages;
937         u64 cur_end;
938         int limit = 10 * 1024 * 1042;
939
940         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
941                          1, 0, NULL, GFP_NOFS);
942         while (start < end) {
943                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
944                 async_cow->inode = inode;
945                 async_cow->root = root;
946                 async_cow->locked_page = locked_page;
947                 async_cow->start = start;
948
949                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
950                         cur_end = end;
951                 else
952                         cur_end = min(end, start + 512 * 1024 - 1);
953
954                 async_cow->end = cur_end;
955                 INIT_LIST_HEAD(&async_cow->extents);
956
957                 async_cow->work.func = async_cow_start;
958                 async_cow->work.ordered_func = async_cow_submit;
959                 async_cow->work.ordered_free = async_cow_free;
960                 async_cow->work.flags = 0;
961
962                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
963                         PAGE_CACHE_SHIFT;
964                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
965
966                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
967                                    &async_cow->work);
968
969                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
970                         wait_event(root->fs_info->async_submit_wait,
971                            (atomic_read(&root->fs_info->async_delalloc_pages) <
972                             limit));
973                 }
974
975                 while (atomic_read(&root->fs_info->async_submit_draining) &&
976                       atomic_read(&root->fs_info->async_delalloc_pages)) {
977                         wait_event(root->fs_info->async_submit_wait,
978                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
979                            0));
980                 }
981
982                 *nr_written += nr_pages;
983                 start = cur_end + 1;
984         }
985         *page_started = 1;
986         return 0;
987 }
988
989 static noinline int csum_exist_in_range(struct btrfs_root *root,
990                                         u64 bytenr, u64 num_bytes)
991 {
992         int ret;
993         struct btrfs_ordered_sum *sums;
994         LIST_HEAD(list);
995
996         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
997                                        bytenr + num_bytes - 1, &list);
998         if (ret == 0 && list_empty(&list))
999                 return 0;
1000
1001         while (!list_empty(&list)) {
1002                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1003                 list_del(&sums->list);
1004                 kfree(sums);
1005         }
1006         return 1;
1007 }
1008
1009 /*
1010  * when nowcow writeback call back.  This checks for snapshots or COW copies
1011  * of the extents that exist in the file, and COWs the file as required.
1012  *
1013  * If no cow copies or snapshots exist, we write directly to the existing
1014  * blocks on disk
1015  */
1016 static noinline int run_delalloc_nocow(struct inode *inode,
1017                                        struct page *locked_page,
1018                               u64 start, u64 end, int *page_started, int force,
1019                               unsigned long *nr_written)
1020 {
1021         struct btrfs_root *root = BTRFS_I(inode)->root;
1022         struct btrfs_trans_handle *trans;
1023         struct extent_buffer *leaf;
1024         struct btrfs_path *path;
1025         struct btrfs_file_extent_item *fi;
1026         struct btrfs_key found_key;
1027         u64 cow_start;
1028         u64 cur_offset;
1029         u64 extent_end;
1030         u64 extent_offset;
1031         u64 disk_bytenr;
1032         u64 num_bytes;
1033         int extent_type;
1034         int ret;
1035         int type;
1036         int nocow;
1037         int check_prev = 1;
1038
1039         path = btrfs_alloc_path();
1040         BUG_ON(!path);
1041         trans = btrfs_join_transaction(root, 1);
1042         BUG_ON(!trans);
1043
1044         cow_start = (u64)-1;
1045         cur_offset = start;
1046         while (1) {
1047                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1048                                                cur_offset, 0);
1049                 BUG_ON(ret < 0);
1050                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1051                         leaf = path->nodes[0];
1052                         btrfs_item_key_to_cpu(leaf, &found_key,
1053                                               path->slots[0] - 1);
1054                         if (found_key.objectid == inode->i_ino &&
1055                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1056                                 path->slots[0]--;
1057                 }
1058                 check_prev = 0;
1059 next_slot:
1060                 leaf = path->nodes[0];
1061                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1062                         ret = btrfs_next_leaf(root, path);
1063                         if (ret < 0)
1064                                 BUG_ON(1);
1065                         if (ret > 0)
1066                                 break;
1067                         leaf = path->nodes[0];
1068                 }
1069
1070                 nocow = 0;
1071                 disk_bytenr = 0;
1072                 num_bytes = 0;
1073                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1074
1075                 if (found_key.objectid > inode->i_ino ||
1076                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1077                     found_key.offset > end)
1078                         break;
1079
1080                 if (found_key.offset > cur_offset) {
1081                         extent_end = found_key.offset;
1082                         extent_type = 0;
1083                         goto out_check;
1084                 }
1085
1086                 fi = btrfs_item_ptr(leaf, path->slots[0],
1087                                     struct btrfs_file_extent_item);
1088                 extent_type = btrfs_file_extent_type(leaf, fi);
1089
1090                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1091                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1092                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1093                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1094                         extent_end = found_key.offset +
1095                                 btrfs_file_extent_num_bytes(leaf, fi);
1096                         if (extent_end <= start) {
1097                                 path->slots[0]++;
1098                                 goto next_slot;
1099                         }
1100                         if (disk_bytenr == 0)
1101                                 goto out_check;
1102                         if (btrfs_file_extent_compression(leaf, fi) ||
1103                             btrfs_file_extent_encryption(leaf, fi) ||
1104                             btrfs_file_extent_other_encoding(leaf, fi))
1105                                 goto out_check;
1106                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1107                                 goto out_check;
1108                         if (btrfs_extent_readonly(root, disk_bytenr))
1109                                 goto out_check;
1110                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1111                                                   found_key.offset -
1112                                                   extent_offset, disk_bytenr))
1113                                 goto out_check;
1114                         disk_bytenr += extent_offset;
1115                         disk_bytenr += cur_offset - found_key.offset;
1116                         num_bytes = min(end + 1, extent_end) - cur_offset;
1117                         /*
1118                          * force cow if csum exists in the range.
1119                          * this ensure that csum for a given extent are
1120                          * either valid or do not exist.
1121                          */
1122                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1123                                 goto out_check;
1124                         nocow = 1;
1125                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1126                         extent_end = found_key.offset +
1127                                 btrfs_file_extent_inline_len(leaf, fi);
1128                         extent_end = ALIGN(extent_end, root->sectorsize);
1129                 } else {
1130                         BUG_ON(1);
1131                 }
1132 out_check:
1133                 if (extent_end <= start) {
1134                         path->slots[0]++;
1135                         goto next_slot;
1136                 }
1137                 if (!nocow) {
1138                         if (cow_start == (u64)-1)
1139                                 cow_start = cur_offset;
1140                         cur_offset = extent_end;
1141                         if (cur_offset > end)
1142                                 break;
1143                         path->slots[0]++;
1144                         goto next_slot;
1145                 }
1146
1147                 btrfs_release_path(root, path);
1148                 if (cow_start != (u64)-1) {
1149                         ret = cow_file_range(inode, locked_page, cow_start,
1150                                         found_key.offset - 1, page_started,
1151                                         nr_written, 1);
1152                         BUG_ON(ret);
1153                         cow_start = (u64)-1;
1154                 }
1155
1156                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1157                         struct extent_map *em;
1158                         struct extent_map_tree *em_tree;
1159                         em_tree = &BTRFS_I(inode)->extent_tree;
1160                         em = alloc_extent_map(GFP_NOFS);
1161                         em->start = cur_offset;
1162                         em->orig_start = em->start;
1163                         em->len = num_bytes;
1164                         em->block_len = num_bytes;
1165                         em->block_start = disk_bytenr;
1166                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1167                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1168                         while (1) {
1169                                 write_lock(&em_tree->lock);
1170                                 ret = add_extent_mapping(em_tree, em);
1171                                 write_unlock(&em_tree->lock);
1172                                 if (ret != -EEXIST) {
1173                                         free_extent_map(em);
1174                                         break;
1175                                 }
1176                                 btrfs_drop_extent_cache(inode, em->start,
1177                                                 em->start + em->len - 1, 0);
1178                         }
1179                         type = BTRFS_ORDERED_PREALLOC;
1180                 } else {
1181                         type = BTRFS_ORDERED_NOCOW;
1182                 }
1183
1184                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1185                                                num_bytes, num_bytes, type);
1186                 BUG_ON(ret);
1187
1188                 if (root->root_key.objectid ==
1189                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1190                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1191                                                       num_bytes);
1192                         BUG_ON(ret);
1193                 }
1194
1195                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1196                                 cur_offset, cur_offset + num_bytes - 1,
1197                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1198                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1199                                 EXTENT_SET_PRIVATE2);
1200                 cur_offset = extent_end;
1201                 if (cur_offset > end)
1202                         break;
1203         }
1204         btrfs_release_path(root, path);
1205
1206         if (cur_offset <= end && cow_start == (u64)-1)
1207                 cow_start = cur_offset;
1208         if (cow_start != (u64)-1) {
1209                 ret = cow_file_range(inode, locked_page, cow_start, end,
1210                                      page_started, nr_written, 1);
1211                 BUG_ON(ret);
1212         }
1213
1214         ret = btrfs_end_transaction(trans, root);
1215         BUG_ON(ret);
1216         btrfs_free_path(path);
1217         return 0;
1218 }
1219
1220 /*
1221  * extent_io.c call back to do delayed allocation processing
1222  */
1223 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1224                               u64 start, u64 end, int *page_started,
1225                               unsigned long *nr_written)
1226 {
1227         int ret;
1228         struct btrfs_root *root = BTRFS_I(inode)->root;
1229
1230         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1231                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1232                                          page_started, 1, nr_written);
1233         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1234                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1235                                          page_started, 0, nr_written);
1236         else if (!btrfs_test_opt(root, COMPRESS) &&
1237                  !(BTRFS_I(inode)->force_compress))
1238                 ret = cow_file_range(inode, locked_page, start, end,
1239                                       page_started, nr_written, 1);
1240         else
1241                 ret = cow_file_range_async(inode, locked_page, start, end,
1242                                            page_started, nr_written);
1243         return ret;
1244 }
1245
1246 static int btrfs_split_extent_hook(struct inode *inode,
1247                                    struct extent_state *orig, u64 split)
1248 {
1249         /* not delalloc, ignore it */
1250         if (!(orig->state & EXTENT_DELALLOC))
1251                 return 0;
1252
1253         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1254         return 0;
1255 }
1256
1257 /*
1258  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1259  * extents so we can keep track of new extents that are just merged onto old
1260  * extents, such as when we are doing sequential writes, so we can properly
1261  * account for the metadata space we'll need.
1262  */
1263 static int btrfs_merge_extent_hook(struct inode *inode,
1264                                    struct extent_state *new,
1265                                    struct extent_state *other)
1266 {
1267         /* not delalloc, ignore it */
1268         if (!(other->state & EXTENT_DELALLOC))
1269                 return 0;
1270
1271         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1272         return 0;
1273 }
1274
1275 /*
1276  * extent_io.c set_bit_hook, used to track delayed allocation
1277  * bytes in this file, and to maintain the list of inodes that
1278  * have pending delalloc work to be done.
1279  */
1280 static int btrfs_set_bit_hook(struct inode *inode,
1281                               struct extent_state *state, int *bits)
1282 {
1283
1284         /*
1285          * set_bit and clear bit hooks normally require _irqsave/restore
1286          * but in this case, we are only testeing for the DELALLOC
1287          * bit, which is only set or cleared with irqs on
1288          */
1289         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1290                 struct btrfs_root *root = BTRFS_I(inode)->root;
1291                 u64 len = state->end + 1 - state->start;
1292
1293                 if (*bits & EXTENT_FIRST_DELALLOC)
1294                         *bits &= ~EXTENT_FIRST_DELALLOC;
1295                 else
1296                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1297
1298                 spin_lock(&root->fs_info->delalloc_lock);
1299                 BTRFS_I(inode)->delalloc_bytes += len;
1300                 root->fs_info->delalloc_bytes += len;
1301                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1302                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1303                                       &root->fs_info->delalloc_inodes);
1304                 }
1305                 spin_unlock(&root->fs_info->delalloc_lock);
1306         }
1307         return 0;
1308 }
1309
1310 /*
1311  * extent_io.c clear_bit_hook, see set_bit_hook for why
1312  */
1313 static int btrfs_clear_bit_hook(struct inode *inode,
1314                                 struct extent_state *state, int *bits)
1315 {
1316         /*
1317          * set_bit and clear bit hooks normally require _irqsave/restore
1318          * but in this case, we are only testeing for the DELALLOC
1319          * bit, which is only set or cleared with irqs on
1320          */
1321         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1322                 struct btrfs_root *root = BTRFS_I(inode)->root;
1323                 u64 len = state->end + 1 - state->start;
1324
1325                 if (*bits & EXTENT_FIRST_DELALLOC)
1326                         *bits &= ~EXTENT_FIRST_DELALLOC;
1327                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1328                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1329
1330                 if (*bits & EXTENT_DO_ACCOUNTING)
1331                         btrfs_delalloc_release_metadata(inode, len);
1332
1333                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1334                         btrfs_free_reserved_data_space(inode, len);
1335
1336                 spin_lock(&root->fs_info->delalloc_lock);
1337                 root->fs_info->delalloc_bytes -= len;
1338                 BTRFS_I(inode)->delalloc_bytes -= len;
1339
1340                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1341                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1342                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1343                 }
1344                 spin_unlock(&root->fs_info->delalloc_lock);
1345         }
1346         return 0;
1347 }
1348
1349 /*
1350  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1351  * we don't create bios that span stripes or chunks
1352  */
1353 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1354                          size_t size, struct bio *bio,
1355                          unsigned long bio_flags)
1356 {
1357         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1358         struct btrfs_mapping_tree *map_tree;
1359         u64 logical = (u64)bio->bi_sector << 9;
1360         u64 length = 0;
1361         u64 map_length;
1362         int ret;
1363
1364         if (bio_flags & EXTENT_BIO_COMPRESSED)
1365                 return 0;
1366
1367         length = bio->bi_size;
1368         map_tree = &root->fs_info->mapping_tree;
1369         map_length = length;
1370         ret = btrfs_map_block(map_tree, READ, logical,
1371                               &map_length, NULL, 0);
1372
1373         if (map_length < length + size)
1374                 return 1;
1375         return 0;
1376 }
1377
1378 /*
1379  * in order to insert checksums into the metadata in large chunks,
1380  * we wait until bio submission time.   All the pages in the bio are
1381  * checksummed and sums are attached onto the ordered extent record.
1382  *
1383  * At IO completion time the cums attached on the ordered extent record
1384  * are inserted into the btree
1385  */
1386 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1387                                     struct bio *bio, int mirror_num,
1388                                     unsigned long bio_flags,
1389                                     u64 bio_offset)
1390 {
1391         struct btrfs_root *root = BTRFS_I(inode)->root;
1392         int ret = 0;
1393
1394         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1395         BUG_ON(ret);
1396         return 0;
1397 }
1398
1399 /*
1400  * in order to insert checksums into the metadata in large chunks,
1401  * we wait until bio submission time.   All the pages in the bio are
1402  * checksummed and sums are attached onto the ordered extent record.
1403  *
1404  * At IO completion time the cums attached on the ordered extent record
1405  * are inserted into the btree
1406  */
1407 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1408                           int mirror_num, unsigned long bio_flags,
1409                           u64 bio_offset)
1410 {
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1413 }
1414
1415 /*
1416  * extent_io.c submission hook. This does the right thing for csum calculation
1417  * on write, or reading the csums from the tree before a read
1418  */
1419 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1420                           int mirror_num, unsigned long bio_flags,
1421                           u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425         int skip_sum;
1426
1427         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1428
1429         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1430         BUG_ON(ret);
1431
1432         if (!(rw & REQ_WRITE)) {
1433                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1434                         return btrfs_submit_compressed_read(inode, bio,
1435                                                     mirror_num, bio_flags);
1436                 } else if (!skip_sum)
1437                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1438                 goto mapit;
1439         } else if (!skip_sum) {
1440                 /* csum items have already been cloned */
1441                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1442                         goto mapit;
1443                 /* we're doing a write, do the async checksumming */
1444                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1445                                    inode, rw, bio, mirror_num,
1446                                    bio_flags, bio_offset,
1447                                    __btrfs_submit_bio_start,
1448                                    __btrfs_submit_bio_done);
1449         }
1450
1451 mapit:
1452         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1453 }
1454
1455 /*
1456  * given a list of ordered sums record them in the inode.  This happens
1457  * at IO completion time based on sums calculated at bio submission time.
1458  */
1459 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1460                              struct inode *inode, u64 file_offset,
1461                              struct list_head *list)
1462 {
1463         struct btrfs_ordered_sum *sum;
1464
1465         btrfs_set_trans_block_group(trans, inode);
1466
1467         list_for_each_entry(sum, list, list) {
1468                 btrfs_csum_file_blocks(trans,
1469                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1470         }
1471         return 0;
1472 }
1473
1474 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1475                               struct extent_state **cached_state)
1476 {
1477         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1478                 WARN_ON(1);
1479         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1480                                    cached_state, GFP_NOFS);
1481 }
1482
1483 /* see btrfs_writepage_start_hook for details on why this is required */
1484 struct btrfs_writepage_fixup {
1485         struct page *page;
1486         struct btrfs_work work;
1487 };
1488
1489 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1490 {
1491         struct btrfs_writepage_fixup *fixup;
1492         struct btrfs_ordered_extent *ordered;
1493         struct extent_state *cached_state = NULL;
1494         struct page *page;
1495         struct inode *inode;
1496         u64 page_start;
1497         u64 page_end;
1498
1499         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1500         page = fixup->page;
1501 again:
1502         lock_page(page);
1503         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1504                 ClearPageChecked(page);
1505                 goto out_page;
1506         }
1507
1508         inode = page->mapping->host;
1509         page_start = page_offset(page);
1510         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1511
1512         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1513                          &cached_state, GFP_NOFS);
1514
1515         /* already ordered? We're done */
1516         if (PagePrivate2(page))
1517                 goto out;
1518
1519         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1520         if (ordered) {
1521                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1522                                      page_end, &cached_state, GFP_NOFS);
1523                 unlock_page(page);
1524                 btrfs_start_ordered_extent(inode, ordered, 1);
1525                 goto again;
1526         }
1527
1528         BUG();
1529         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1530         ClearPageChecked(page);
1531 out:
1532         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1533                              &cached_state, GFP_NOFS);
1534 out_page:
1535         unlock_page(page);
1536         page_cache_release(page);
1537 }
1538
1539 /*
1540  * There are a few paths in the higher layers of the kernel that directly
1541  * set the page dirty bit without asking the filesystem if it is a
1542  * good idea.  This causes problems because we want to make sure COW
1543  * properly happens and the data=ordered rules are followed.
1544  *
1545  * In our case any range that doesn't have the ORDERED bit set
1546  * hasn't been properly setup for IO.  We kick off an async process
1547  * to fix it up.  The async helper will wait for ordered extents, set
1548  * the delalloc bit and make it safe to write the page.
1549  */
1550 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1551 {
1552         struct inode *inode = page->mapping->host;
1553         struct btrfs_writepage_fixup *fixup;
1554         struct btrfs_root *root = BTRFS_I(inode)->root;
1555
1556         /* this page is properly in the ordered list */
1557         if (TestClearPagePrivate2(page))
1558                 return 0;
1559
1560         if (PageChecked(page))
1561                 return -EAGAIN;
1562
1563         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1564         if (!fixup)
1565                 return -EAGAIN;
1566
1567         SetPageChecked(page);
1568         page_cache_get(page);
1569         fixup->work.func = btrfs_writepage_fixup_worker;
1570         fixup->page = page;
1571         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1572         return -EAGAIN;
1573 }
1574
1575 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1576                                        struct inode *inode, u64 file_pos,
1577                                        u64 disk_bytenr, u64 disk_num_bytes,
1578                                        u64 num_bytes, u64 ram_bytes,
1579                                        u8 compression, u8 encryption,
1580                                        u16 other_encoding, int extent_type)
1581 {
1582         struct btrfs_root *root = BTRFS_I(inode)->root;
1583         struct btrfs_file_extent_item *fi;
1584         struct btrfs_path *path;
1585         struct extent_buffer *leaf;
1586         struct btrfs_key ins;
1587         u64 hint;
1588         int ret;
1589
1590         path = btrfs_alloc_path();
1591         BUG_ON(!path);
1592
1593         path->leave_spinning = 1;
1594
1595         /*
1596          * we may be replacing one extent in the tree with another.
1597          * The new extent is pinned in the extent map, and we don't want
1598          * to drop it from the cache until it is completely in the btree.
1599          *
1600          * So, tell btrfs_drop_extents to leave this extent in the cache.
1601          * the caller is expected to unpin it and allow it to be merged
1602          * with the others.
1603          */
1604         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1605                                  &hint, 0);
1606         BUG_ON(ret);
1607
1608         ins.objectid = inode->i_ino;
1609         ins.offset = file_pos;
1610         ins.type = BTRFS_EXTENT_DATA_KEY;
1611         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1612         BUG_ON(ret);
1613         leaf = path->nodes[0];
1614         fi = btrfs_item_ptr(leaf, path->slots[0],
1615                             struct btrfs_file_extent_item);
1616         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1617         btrfs_set_file_extent_type(leaf, fi, extent_type);
1618         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1619         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1620         btrfs_set_file_extent_offset(leaf, fi, 0);
1621         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1622         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1623         btrfs_set_file_extent_compression(leaf, fi, compression);
1624         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1625         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1626
1627         btrfs_unlock_up_safe(path, 1);
1628         btrfs_set_lock_blocking(leaf);
1629
1630         btrfs_mark_buffer_dirty(leaf);
1631
1632         inode_add_bytes(inode, num_bytes);
1633
1634         ins.objectid = disk_bytenr;
1635         ins.offset = disk_num_bytes;
1636         ins.type = BTRFS_EXTENT_ITEM_KEY;
1637         ret = btrfs_alloc_reserved_file_extent(trans, root,
1638                                         root->root_key.objectid,
1639                                         inode->i_ino, file_pos, &ins);
1640         BUG_ON(ret);
1641         btrfs_free_path(path);
1642
1643         return 0;
1644 }
1645
1646 /*
1647  * helper function for btrfs_finish_ordered_io, this
1648  * just reads in some of the csum leaves to prime them into ram
1649  * before we start the transaction.  It limits the amount of btree
1650  * reads required while inside the transaction.
1651  */
1652 /* as ordered data IO finishes, this gets called so we can finish
1653  * an ordered extent if the range of bytes in the file it covers are
1654  * fully written.
1655  */
1656 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1657 {
1658         struct btrfs_root *root = BTRFS_I(inode)->root;
1659         struct btrfs_trans_handle *trans = NULL;
1660         struct btrfs_ordered_extent *ordered_extent = NULL;
1661         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1662         struct extent_state *cached_state = NULL;
1663         int compressed = 0;
1664         int ret;
1665
1666         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1667                                              end - start + 1);
1668         if (!ret)
1669                 return 0;
1670         BUG_ON(!ordered_extent);
1671
1672         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1673                 BUG_ON(!list_empty(&ordered_extent->list));
1674                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1675                 if (!ret) {
1676                         trans = btrfs_join_transaction(root, 1);
1677                         btrfs_set_trans_block_group(trans, inode);
1678                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1679                         ret = btrfs_update_inode(trans, root, inode);
1680                         BUG_ON(ret);
1681                 }
1682                 goto out;
1683         }
1684
1685         lock_extent_bits(io_tree, ordered_extent->file_offset,
1686                          ordered_extent->file_offset + ordered_extent->len - 1,
1687                          0, &cached_state, GFP_NOFS);
1688
1689         trans = btrfs_join_transaction(root, 1);
1690         btrfs_set_trans_block_group(trans, inode);
1691         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1692
1693         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1694                 compressed = 1;
1695         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1696                 BUG_ON(compressed);
1697                 ret = btrfs_mark_extent_written(trans, inode,
1698                                                 ordered_extent->file_offset,
1699                                                 ordered_extent->file_offset +
1700                                                 ordered_extent->len);
1701                 BUG_ON(ret);
1702         } else {
1703                 ret = insert_reserved_file_extent(trans, inode,
1704                                                 ordered_extent->file_offset,
1705                                                 ordered_extent->start,
1706                                                 ordered_extent->disk_len,
1707                                                 ordered_extent->len,
1708                                                 ordered_extent->len,
1709                                                 compressed, 0, 0,
1710                                                 BTRFS_FILE_EXTENT_REG);
1711                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1712                                    ordered_extent->file_offset,
1713                                    ordered_extent->len);
1714                 BUG_ON(ret);
1715         }
1716         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1717                              ordered_extent->file_offset +
1718                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1719
1720         add_pending_csums(trans, inode, ordered_extent->file_offset,
1721                           &ordered_extent->list);
1722
1723         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1724         ret = btrfs_update_inode(trans, root, inode);
1725         BUG_ON(ret);
1726 out:
1727         btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1728         if (trans)
1729                 btrfs_end_transaction(trans, root);
1730         /* once for us */
1731         btrfs_put_ordered_extent(ordered_extent);
1732         /* once for the tree */
1733         btrfs_put_ordered_extent(ordered_extent);
1734
1735         return 0;
1736 }
1737
1738 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1739                                 struct extent_state *state, int uptodate)
1740 {
1741         ClearPagePrivate2(page);
1742         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1743 }
1744
1745 /*
1746  * When IO fails, either with EIO or csum verification fails, we
1747  * try other mirrors that might have a good copy of the data.  This
1748  * io_failure_record is used to record state as we go through all the
1749  * mirrors.  If another mirror has good data, the page is set up to date
1750  * and things continue.  If a good mirror can't be found, the original
1751  * bio end_io callback is called to indicate things have failed.
1752  */
1753 struct io_failure_record {
1754         struct page *page;
1755         u64 start;
1756         u64 len;
1757         u64 logical;
1758         unsigned long bio_flags;
1759         int last_mirror;
1760 };
1761
1762 static int btrfs_io_failed_hook(struct bio *failed_bio,
1763                          struct page *page, u64 start, u64 end,
1764                          struct extent_state *state)
1765 {
1766         struct io_failure_record *failrec = NULL;
1767         u64 private;
1768         struct extent_map *em;
1769         struct inode *inode = page->mapping->host;
1770         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1771         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1772         struct bio *bio;
1773         int num_copies;
1774         int ret;
1775         int rw;
1776         u64 logical;
1777
1778         ret = get_state_private(failure_tree, start, &private);
1779         if (ret) {
1780                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1781                 if (!failrec)
1782                         return -ENOMEM;
1783                 failrec->start = start;
1784                 failrec->len = end - start + 1;
1785                 failrec->last_mirror = 0;
1786                 failrec->bio_flags = 0;
1787
1788                 read_lock(&em_tree->lock);
1789                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1790                 if (em->start > start || em->start + em->len < start) {
1791                         free_extent_map(em);
1792                         em = NULL;
1793                 }
1794                 read_unlock(&em_tree->lock);
1795
1796                 if (!em || IS_ERR(em)) {
1797                         kfree(failrec);
1798                         return -EIO;
1799                 }
1800                 logical = start - em->start;
1801                 logical = em->block_start + logical;
1802                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1803                         logical = em->block_start;
1804                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1805                 }
1806                 failrec->logical = logical;
1807                 free_extent_map(em);
1808                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1809                                 EXTENT_DIRTY, GFP_NOFS);
1810                 set_state_private(failure_tree, start,
1811                                  (u64)(unsigned long)failrec);
1812         } else {
1813                 failrec = (struct io_failure_record *)(unsigned long)private;
1814         }
1815         num_copies = btrfs_num_copies(
1816                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1817                               failrec->logical, failrec->len);
1818         failrec->last_mirror++;
1819         if (!state) {
1820                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1821                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1822                                                     failrec->start,
1823                                                     EXTENT_LOCKED);
1824                 if (state && state->start != failrec->start)
1825                         state = NULL;
1826                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1827         }
1828         if (!state || failrec->last_mirror > num_copies) {
1829                 set_state_private(failure_tree, failrec->start, 0);
1830                 clear_extent_bits(failure_tree, failrec->start,
1831                                   failrec->start + failrec->len - 1,
1832                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1833                 kfree(failrec);
1834                 return -EIO;
1835         }
1836         bio = bio_alloc(GFP_NOFS, 1);
1837         bio->bi_private = state;
1838         bio->bi_end_io = failed_bio->bi_end_io;
1839         bio->bi_sector = failrec->logical >> 9;
1840         bio->bi_bdev = failed_bio->bi_bdev;
1841         bio->bi_size = 0;
1842
1843         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1844         if (failed_bio->bi_rw & REQ_WRITE)
1845                 rw = WRITE;
1846         else
1847                 rw = READ;
1848
1849         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1850                                                       failrec->last_mirror,
1851                                                       failrec->bio_flags, 0);
1852         return 0;
1853 }
1854
1855 /*
1856  * each time an IO finishes, we do a fast check in the IO failure tree
1857  * to see if we need to process or clean up an io_failure_record
1858  */
1859 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1860 {
1861         u64 private;
1862         u64 private_failure;
1863         struct io_failure_record *failure;
1864         int ret;
1865
1866         private = 0;
1867         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1868                              (u64)-1, 1, EXTENT_DIRTY)) {
1869                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1870                                         start, &private_failure);
1871                 if (ret == 0) {
1872                         failure = (struct io_failure_record *)(unsigned long)
1873                                    private_failure;
1874                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1875                                           failure->start, 0);
1876                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1877                                           failure->start,
1878                                           failure->start + failure->len - 1,
1879                                           EXTENT_DIRTY | EXTENT_LOCKED,
1880                                           GFP_NOFS);
1881                         kfree(failure);
1882                 }
1883         }
1884         return 0;
1885 }
1886
1887 /*
1888  * when reads are done, we need to check csums to verify the data is correct
1889  * if there's a match, we allow the bio to finish.  If not, we go through
1890  * the io_failure_record routines to find good copies
1891  */
1892 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1893                                struct extent_state *state)
1894 {
1895         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1896         struct inode *inode = page->mapping->host;
1897         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1898         char *kaddr;
1899         u64 private = ~(u32)0;
1900         int ret;
1901         struct btrfs_root *root = BTRFS_I(inode)->root;
1902         u32 csum = ~(u32)0;
1903
1904         if (PageChecked(page)) {
1905                 ClearPageChecked(page);
1906                 goto good;
1907         }
1908
1909         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1910                 return 0;
1911
1912         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1913             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1914                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1915                                   GFP_NOFS);
1916                 return 0;
1917         }
1918
1919         if (state && state->start == start) {
1920                 private = state->private;
1921                 ret = 0;
1922         } else {
1923                 ret = get_state_private(io_tree, start, &private);
1924         }
1925         kaddr = kmap_atomic(page, KM_USER0);
1926         if (ret)
1927                 goto zeroit;
1928
1929         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1930         btrfs_csum_final(csum, (char *)&csum);
1931         if (csum != private)
1932                 goto zeroit;
1933
1934         kunmap_atomic(kaddr, KM_USER0);
1935 good:
1936         /* if the io failure tree for this inode is non-empty,
1937          * check to see if we've recovered from a failed IO
1938          */
1939         btrfs_clean_io_failures(inode, start);
1940         return 0;
1941
1942 zeroit:
1943         if (printk_ratelimit()) {
1944                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1945                        "private %llu\n", page->mapping->host->i_ino,
1946                        (unsigned long long)start, csum,
1947                        (unsigned long long)private);
1948         }
1949         memset(kaddr + offset, 1, end - start + 1);
1950         flush_dcache_page(page);
1951         kunmap_atomic(kaddr, KM_USER0);
1952         if (private == 0)
1953                 return 0;
1954         return -EIO;
1955 }
1956
1957 struct delayed_iput {
1958         struct list_head list;
1959         struct inode *inode;
1960 };
1961
1962 void btrfs_add_delayed_iput(struct inode *inode)
1963 {
1964         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1965         struct delayed_iput *delayed;
1966
1967         if (atomic_add_unless(&inode->i_count, -1, 1))
1968                 return;
1969
1970         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1971         delayed->inode = inode;
1972
1973         spin_lock(&fs_info->delayed_iput_lock);
1974         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1975         spin_unlock(&fs_info->delayed_iput_lock);
1976 }
1977
1978 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1979 {
1980         LIST_HEAD(list);
1981         struct btrfs_fs_info *fs_info = root->fs_info;
1982         struct delayed_iput *delayed;
1983         int empty;
1984
1985         spin_lock(&fs_info->delayed_iput_lock);
1986         empty = list_empty(&fs_info->delayed_iputs);
1987         spin_unlock(&fs_info->delayed_iput_lock);
1988         if (empty)
1989                 return;
1990
1991         down_read(&root->fs_info->cleanup_work_sem);
1992         spin_lock(&fs_info->delayed_iput_lock);
1993         list_splice_init(&fs_info->delayed_iputs, &list);
1994         spin_unlock(&fs_info->delayed_iput_lock);
1995
1996         while (!list_empty(&list)) {
1997                 delayed = list_entry(list.next, struct delayed_iput, list);
1998                 list_del(&delayed->list);
1999                 iput(delayed->inode);
2000                 kfree(delayed);
2001         }
2002         up_read(&root->fs_info->cleanup_work_sem);
2003 }
2004
2005 /*
2006  * calculate extra metadata reservation when snapshotting a subvolume
2007  * contains orphan files.
2008  */
2009 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2010                                 struct btrfs_pending_snapshot *pending,
2011                                 u64 *bytes_to_reserve)
2012 {
2013         struct btrfs_root *root;
2014         struct btrfs_block_rsv *block_rsv;
2015         u64 num_bytes;
2016         int index;
2017
2018         root = pending->root;
2019         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2020                 return;
2021
2022         block_rsv = root->orphan_block_rsv;
2023
2024         /* orphan block reservation for the snapshot */
2025         num_bytes = block_rsv->size;
2026
2027         /*
2028          * after the snapshot is created, COWing tree blocks may use more
2029          * space than it frees. So we should make sure there is enough
2030          * reserved space.
2031          */
2032         index = trans->transid & 0x1;
2033         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2034                 num_bytes += block_rsv->size -
2035                              (block_rsv->reserved + block_rsv->freed[index]);
2036         }
2037
2038         *bytes_to_reserve += num_bytes;
2039 }
2040
2041 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2042                                 struct btrfs_pending_snapshot *pending)
2043 {
2044         struct btrfs_root *root = pending->root;
2045         struct btrfs_root *snap = pending->snap;
2046         struct btrfs_block_rsv *block_rsv;
2047         u64 num_bytes;
2048         int index;
2049         int ret;
2050
2051         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2052                 return;
2053
2054         /* refill source subvolume's orphan block reservation */
2055         block_rsv = root->orphan_block_rsv;
2056         index = trans->transid & 0x1;
2057         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2058                 num_bytes = block_rsv->size -
2059                             (block_rsv->reserved + block_rsv->freed[index]);
2060                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2061                                               root->orphan_block_rsv,
2062                                               num_bytes);
2063                 BUG_ON(ret);
2064         }
2065
2066         /* setup orphan block reservation for the snapshot */
2067         block_rsv = btrfs_alloc_block_rsv(snap);
2068         BUG_ON(!block_rsv);
2069
2070         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2071         snap->orphan_block_rsv = block_rsv;
2072
2073         num_bytes = root->orphan_block_rsv->size;
2074         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2075                                       block_rsv, num_bytes);
2076         BUG_ON(ret);
2077
2078 #if 0
2079         /* insert orphan item for the snapshot */
2080         WARN_ON(!root->orphan_item_inserted);
2081         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2082                                        snap->root_key.objectid);
2083         BUG_ON(ret);
2084         snap->orphan_item_inserted = 1;
2085 #endif
2086 }
2087
2088 enum btrfs_orphan_cleanup_state {
2089         ORPHAN_CLEANUP_STARTED  = 1,
2090         ORPHAN_CLEANUP_DONE     = 2,
2091 };
2092
2093 /*
2094  * This is called in transaction commmit time. If there are no orphan
2095  * files in the subvolume, it removes orphan item and frees block_rsv
2096  * structure.
2097  */
2098 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2099                               struct btrfs_root *root)
2100 {
2101         int ret;
2102
2103         if (!list_empty(&root->orphan_list) ||
2104             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2105                 return;
2106
2107         if (root->orphan_item_inserted &&
2108             btrfs_root_refs(&root->root_item) > 0) {
2109                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2110                                             root->root_key.objectid);
2111                 BUG_ON(ret);
2112                 root->orphan_item_inserted = 0;
2113         }
2114
2115         if (root->orphan_block_rsv) {
2116                 WARN_ON(root->orphan_block_rsv->size > 0);
2117                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2118                 root->orphan_block_rsv = NULL;
2119         }
2120 }
2121
2122 /*
2123  * This creates an orphan entry for the given inode in case something goes
2124  * wrong in the middle of an unlink/truncate.
2125  *
2126  * NOTE: caller of this function should reserve 5 units of metadata for
2127  *       this function.
2128  */
2129 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2130 {
2131         struct btrfs_root *root = BTRFS_I(inode)->root;
2132         struct btrfs_block_rsv *block_rsv = NULL;
2133         int reserve = 0;
2134         int insert = 0;
2135         int ret;
2136
2137         if (!root->orphan_block_rsv) {
2138                 block_rsv = btrfs_alloc_block_rsv(root);
2139                 BUG_ON(!block_rsv);
2140         }
2141
2142         spin_lock(&root->orphan_lock);
2143         if (!root->orphan_block_rsv) {
2144                 root->orphan_block_rsv = block_rsv;
2145         } else if (block_rsv) {
2146                 btrfs_free_block_rsv(root, block_rsv);
2147                 block_rsv = NULL;
2148         }
2149
2150         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2151                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2152 #if 0
2153                 /*
2154                  * For proper ENOSPC handling, we should do orphan
2155                  * cleanup when mounting. But this introduces backward
2156                  * compatibility issue.
2157                  */
2158                 if (!xchg(&root->orphan_item_inserted, 1))
2159                         insert = 2;
2160                 else
2161                         insert = 1;
2162 #endif
2163                 insert = 1;
2164         } else {
2165                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2166         }
2167
2168         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2169                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2170                 reserve = 1;
2171         }
2172         spin_unlock(&root->orphan_lock);
2173
2174         if (block_rsv)
2175                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2176
2177         /* grab metadata reservation from transaction handle */
2178         if (reserve) {
2179                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2180                 BUG_ON(ret);
2181         }
2182
2183         /* insert an orphan item to track this unlinked/truncated file */
2184         if (insert >= 1) {
2185                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2186                 BUG_ON(ret);
2187         }
2188
2189         /* insert an orphan item to track subvolume contains orphan files */
2190         if (insert >= 2) {
2191                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2192                                                root->root_key.objectid);
2193                 BUG_ON(ret);
2194         }
2195         return 0;
2196 }
2197
2198 /*
2199  * We have done the truncate/delete so we can go ahead and remove the orphan
2200  * item for this particular inode.
2201  */
2202 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2203 {
2204         struct btrfs_root *root = BTRFS_I(inode)->root;
2205         int delete_item = 0;
2206         int release_rsv = 0;
2207         int ret = 0;
2208
2209         spin_lock(&root->orphan_lock);
2210         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2211                 list_del_init(&BTRFS_I(inode)->i_orphan);
2212                 delete_item = 1;
2213         }
2214
2215         if (BTRFS_I(inode)->orphan_meta_reserved) {
2216                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2217                 release_rsv = 1;
2218         }
2219         spin_unlock(&root->orphan_lock);
2220
2221         if (trans && delete_item) {
2222                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2223                 BUG_ON(ret);
2224         }
2225
2226         if (release_rsv)
2227                 btrfs_orphan_release_metadata(inode);
2228
2229         return 0;
2230 }
2231
2232 /*
2233  * this cleans up any orphans that may be left on the list from the last use
2234  * of this root.
2235  */
2236 void btrfs_orphan_cleanup(struct btrfs_root *root)
2237 {
2238         struct btrfs_path *path;
2239         struct extent_buffer *leaf;
2240         struct btrfs_item *item;
2241         struct btrfs_key key, found_key;
2242         struct btrfs_trans_handle *trans;
2243         struct inode *inode;
2244         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2245
2246         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2247                 return;
2248
2249         path = btrfs_alloc_path();
2250         BUG_ON(!path);
2251         path->reada = -1;
2252
2253         key.objectid = BTRFS_ORPHAN_OBJECTID;
2254         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2255         key.offset = (u64)-1;
2256
2257         while (1) {
2258                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2259                 if (ret < 0) {
2260                         printk(KERN_ERR "Error searching slot for orphan: %d"
2261                                "\n", ret);
2262                         break;
2263                 }
2264
2265                 /*
2266                  * if ret == 0 means we found what we were searching for, which
2267                  * is weird, but possible, so only screw with path if we didnt
2268                  * find the key and see if we have stuff that matches
2269                  */
2270                 if (ret > 0) {
2271                         if (path->slots[0] == 0)
2272                                 break;
2273                         path->slots[0]--;
2274                 }
2275
2276                 /* pull out the item */
2277                 leaf = path->nodes[0];
2278                 item = btrfs_item_nr(leaf, path->slots[0]);
2279                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2280
2281                 /* make sure the item matches what we want */
2282                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2283                         break;
2284                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2285                         break;
2286
2287                 /* release the path since we're done with it */
2288                 btrfs_release_path(root, path);
2289
2290                 /*
2291                  * this is where we are basically btrfs_lookup, without the
2292                  * crossing root thing.  we store the inode number in the
2293                  * offset of the orphan item.
2294                  */
2295                 found_key.objectid = found_key.offset;
2296                 found_key.type = BTRFS_INODE_ITEM_KEY;
2297                 found_key.offset = 0;
2298                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2299                 BUG_ON(IS_ERR(inode));
2300
2301                 /*
2302                  * add this inode to the orphan list so btrfs_orphan_del does
2303                  * the proper thing when we hit it
2304                  */
2305                 spin_lock(&root->orphan_lock);
2306                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2307                 spin_unlock(&root->orphan_lock);
2308
2309                 /*
2310                  * if this is a bad inode, means we actually succeeded in
2311                  * removing the inode, but not the orphan record, which means
2312                  * we need to manually delete the orphan since iput will just
2313                  * do a destroy_inode
2314                  */
2315                 if (is_bad_inode(inode)) {
2316                         trans = btrfs_start_transaction(root, 0);
2317                         btrfs_orphan_del(trans, inode);
2318                         btrfs_end_transaction(trans, root);
2319                         iput(inode);
2320                         continue;
2321                 }
2322
2323                 /* if we have links, this was a truncate, lets do that */
2324                 if (inode->i_nlink) {
2325                         nr_truncate++;
2326                         btrfs_truncate(inode);
2327                 } else {
2328                         nr_unlink++;
2329                 }
2330
2331                 /* this will do delete_inode and everything for us */
2332                 iput(inode);
2333         }
2334         btrfs_free_path(path);
2335
2336         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2337
2338         if (root->orphan_block_rsv)
2339                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2340                                         (u64)-1);
2341
2342         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2343                 trans = btrfs_join_transaction(root, 1);
2344                 btrfs_end_transaction(trans, root);
2345         }
2346
2347         if (nr_unlink)
2348                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2349         if (nr_truncate)
2350                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2351 }
2352
2353 /*
2354  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2355  * don't find any xattrs, we know there can't be any acls.
2356  *
2357  * slot is the slot the inode is in, objectid is the objectid of the inode
2358  */
2359 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2360                                           int slot, u64 objectid)
2361 {
2362         u32 nritems = btrfs_header_nritems(leaf);
2363         struct btrfs_key found_key;
2364         int scanned = 0;
2365
2366         slot++;
2367         while (slot < nritems) {
2368                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2369
2370                 /* we found a different objectid, there must not be acls */
2371                 if (found_key.objectid != objectid)
2372                         return 0;
2373
2374                 /* we found an xattr, assume we've got an acl */
2375                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2376                         return 1;
2377
2378                 /*
2379                  * we found a key greater than an xattr key, there can't
2380                  * be any acls later on
2381                  */
2382                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2383                         return 0;
2384
2385                 slot++;
2386                 scanned++;
2387
2388                 /*
2389                  * it goes inode, inode backrefs, xattrs, extents,
2390                  * so if there are a ton of hard links to an inode there can
2391                  * be a lot of backrefs.  Don't waste time searching too hard,
2392                  * this is just an optimization
2393                  */
2394                 if (scanned >= 8)
2395                         break;
2396         }
2397         /* we hit the end of the leaf before we found an xattr or
2398          * something larger than an xattr.  We have to assume the inode
2399          * has acls
2400          */
2401         return 1;
2402 }
2403
2404 /*
2405  * read an inode from the btree into the in-memory inode
2406  */
2407 static void btrfs_read_locked_inode(struct inode *inode)
2408 {
2409         struct btrfs_path *path;
2410         struct extent_buffer *leaf;
2411         struct btrfs_inode_item *inode_item;
2412         struct btrfs_timespec *tspec;
2413         struct btrfs_root *root = BTRFS_I(inode)->root;
2414         struct btrfs_key location;
2415         int maybe_acls;
2416         u64 alloc_group_block;
2417         u32 rdev;
2418         int ret;
2419
2420         path = btrfs_alloc_path();
2421         BUG_ON(!path);
2422         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2423
2424         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2425         if (ret)
2426                 goto make_bad;
2427
2428         leaf = path->nodes[0];
2429         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2430                                     struct btrfs_inode_item);
2431
2432         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2433         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2434         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2435         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2436         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2437
2438         tspec = btrfs_inode_atime(inode_item);
2439         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2440         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2441
2442         tspec = btrfs_inode_mtime(inode_item);
2443         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2444         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2445
2446         tspec = btrfs_inode_ctime(inode_item);
2447         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2448         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2449
2450         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2451         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2452         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2453         inode->i_generation = BTRFS_I(inode)->generation;
2454         inode->i_rdev = 0;
2455         rdev = btrfs_inode_rdev(leaf, inode_item);
2456
2457         BTRFS_I(inode)->index_cnt = (u64)-1;
2458         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2459
2460         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2461
2462         /*
2463          * try to precache a NULL acl entry for files that don't have
2464          * any xattrs or acls
2465          */
2466         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2467         if (!maybe_acls)
2468                 cache_no_acl(inode);
2469
2470         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2471                                                 alloc_group_block, 0);
2472         btrfs_free_path(path);
2473         inode_item = NULL;
2474
2475         switch (inode->i_mode & S_IFMT) {
2476         case S_IFREG:
2477                 inode->i_mapping->a_ops = &btrfs_aops;
2478                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2479                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2480                 inode->i_fop = &btrfs_file_operations;
2481                 inode->i_op = &btrfs_file_inode_operations;
2482                 break;
2483         case S_IFDIR:
2484                 inode->i_fop = &btrfs_dir_file_operations;
2485                 if (root == root->fs_info->tree_root)
2486                         inode->i_op = &btrfs_dir_ro_inode_operations;
2487                 else
2488                         inode->i_op = &btrfs_dir_inode_operations;
2489                 break;
2490         case S_IFLNK:
2491                 inode->i_op = &btrfs_symlink_inode_operations;
2492                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2493                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2494                 break;
2495         default:
2496                 inode->i_op = &btrfs_special_inode_operations;
2497                 init_special_inode(inode, inode->i_mode, rdev);
2498                 break;
2499         }
2500
2501         btrfs_update_iflags(inode);
2502         return;
2503
2504 make_bad:
2505         btrfs_free_path(path);
2506         make_bad_inode(inode);
2507 }
2508
2509 /*
2510  * given a leaf and an inode, copy the inode fields into the leaf
2511  */
2512 static void fill_inode_item(struct btrfs_trans_handle *trans,
2513                             struct extent_buffer *leaf,
2514                             struct btrfs_inode_item *item,
2515                             struct inode *inode)
2516 {
2517         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2518         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2519         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2520         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2521         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2522
2523         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2524                                inode->i_atime.tv_sec);
2525         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2526                                 inode->i_atime.tv_nsec);
2527
2528         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2529                                inode->i_mtime.tv_sec);
2530         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2531                                 inode->i_mtime.tv_nsec);
2532
2533         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2534                                inode->i_ctime.tv_sec);
2535         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2536                                 inode->i_ctime.tv_nsec);
2537
2538         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2539         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2540         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2541         btrfs_set_inode_transid(leaf, item, trans->transid);
2542         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2543         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2544         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2545 }
2546
2547 /*
2548  * copy everything in the in-memory inode into the btree.
2549  */
2550 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2551                                 struct btrfs_root *root, struct inode *inode)
2552 {
2553         struct btrfs_inode_item *inode_item;
2554         struct btrfs_path *path;
2555         struct extent_buffer *leaf;
2556         int ret;
2557
2558         path = btrfs_alloc_path();
2559         BUG_ON(!path);
2560         path->leave_spinning = 1;
2561         ret = btrfs_lookup_inode(trans, root, path,
2562                                  &BTRFS_I(inode)->location, 1);
2563         if (ret) {
2564                 if (ret > 0)
2565                         ret = -ENOENT;
2566                 goto failed;
2567         }
2568
2569         btrfs_unlock_up_safe(path, 1);
2570         leaf = path->nodes[0];
2571         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2572                                   struct btrfs_inode_item);
2573
2574         fill_inode_item(trans, leaf, inode_item, inode);
2575         btrfs_mark_buffer_dirty(leaf);
2576         btrfs_set_inode_last_trans(trans, inode);
2577         ret = 0;
2578 failed:
2579         btrfs_free_path(path);
2580         return ret;
2581 }
2582
2583
2584 /*
2585  * unlink helper that gets used here in inode.c and in the tree logging
2586  * recovery code.  It remove a link in a directory with a given name, and
2587  * also drops the back refs in the inode to the directory
2588  */
2589 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2590                        struct btrfs_root *root,
2591                        struct inode *dir, struct inode *inode,
2592                        const char *name, int name_len)
2593 {
2594         struct btrfs_path *path;
2595         int ret = 0;
2596         struct extent_buffer *leaf;
2597         struct btrfs_dir_item *di;
2598         struct btrfs_key key;
2599         u64 index;
2600
2601         path = btrfs_alloc_path();
2602         if (!path) {
2603                 ret = -ENOMEM;
2604                 goto err;
2605         }
2606
2607         path->leave_spinning = 1;
2608         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2609                                     name, name_len, -1);
2610         if (IS_ERR(di)) {
2611                 ret = PTR_ERR(di);
2612                 goto err;
2613         }
2614         if (!di) {
2615                 ret = -ENOENT;
2616                 goto err;
2617         }
2618         leaf = path->nodes[0];
2619         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2620         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2621         if (ret)
2622                 goto err;
2623         btrfs_release_path(root, path);
2624
2625         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2626                                   inode->i_ino,
2627                                   dir->i_ino, &index);
2628         if (ret) {
2629                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2630                        "inode %lu parent %lu\n", name_len, name,
2631                        inode->i_ino, dir->i_ino);
2632                 goto err;
2633         }
2634
2635         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2636                                          index, name, name_len, -1);
2637         if (IS_ERR(di)) {
2638                 ret = PTR_ERR(di);
2639                 goto err;
2640         }
2641         if (!di) {
2642                 ret = -ENOENT;
2643                 goto err;
2644         }
2645         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2646         btrfs_release_path(root, path);
2647
2648         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2649                                          inode, dir->i_ino);
2650         BUG_ON(ret != 0 && ret != -ENOENT);
2651
2652         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2653                                            dir, index);
2654         BUG_ON(ret);
2655 err:
2656         btrfs_free_path(path);
2657         if (ret)
2658                 goto out;
2659
2660         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2661         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2662         btrfs_update_inode(trans, root, dir);
2663         btrfs_drop_nlink(inode);
2664         ret = btrfs_update_inode(trans, root, inode);
2665 out:
2666         return ret;
2667 }
2668
2669 /* helper to check if there is any shared block in the path */
2670 static int check_path_shared(struct btrfs_root *root,
2671                              struct btrfs_path *path)
2672 {
2673         struct extent_buffer *eb;
2674         int level;
2675         int ret;
2676         u64 refs = 1;
2677
2678         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2679                 if (!path->nodes[level])
2680                         break;
2681                 eb = path->nodes[level];
2682                 if (!btrfs_block_can_be_shared(root, eb))
2683                         continue;
2684                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2685                                                &refs, NULL);
2686                 if (refs > 1)
2687                         return 1;
2688         }
2689         return 0;
2690 }
2691
2692 /*
2693  * helper to start transaction for unlink and rmdir.
2694  *
2695  * unlink and rmdir are special in btrfs, they do not always free space.
2696  * so in enospc case, we should make sure they will free space before
2697  * allowing them to use the global metadata reservation.
2698  */
2699 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2700                                                        struct dentry *dentry)
2701 {
2702         struct btrfs_trans_handle *trans;
2703         struct btrfs_root *root = BTRFS_I(dir)->root;
2704         struct btrfs_path *path;
2705         struct btrfs_inode_ref *ref;
2706         struct btrfs_dir_item *di;
2707         struct inode *inode = dentry->d_inode;
2708         u64 index;
2709         int check_link = 1;
2710         int err = -ENOSPC;
2711         int ret;
2712
2713         trans = btrfs_start_transaction(root, 10);
2714         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2715                 return trans;
2716
2717         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2718                 return ERR_PTR(-ENOSPC);
2719
2720         /* check if there is someone else holds reference */
2721         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2722                 return ERR_PTR(-ENOSPC);
2723
2724         if (atomic_read(&inode->i_count) > 2)
2725                 return ERR_PTR(-ENOSPC);
2726
2727         if (xchg(&root->fs_info->enospc_unlink, 1))
2728                 return ERR_PTR(-ENOSPC);
2729
2730         path = btrfs_alloc_path();
2731         if (!path) {
2732                 root->fs_info->enospc_unlink = 0;
2733                 return ERR_PTR(-ENOMEM);
2734         }
2735
2736         trans = btrfs_start_transaction(root, 0);
2737         if (IS_ERR(trans)) {
2738                 btrfs_free_path(path);
2739                 root->fs_info->enospc_unlink = 0;
2740                 return trans;
2741         }
2742
2743         path->skip_locking = 1;
2744         path->search_commit_root = 1;
2745
2746         ret = btrfs_lookup_inode(trans, root, path,
2747                                 &BTRFS_I(dir)->location, 0);
2748         if (ret < 0) {
2749                 err = ret;
2750                 goto out;
2751         }
2752         if (ret == 0) {
2753                 if (check_path_shared(root, path))
2754                         goto out;
2755         } else {
2756                 check_link = 0;
2757         }
2758         btrfs_release_path(root, path);
2759
2760         ret = btrfs_lookup_inode(trans, root, path,
2761                                 &BTRFS_I(inode)->location, 0);
2762         if (ret < 0) {
2763                 err = ret;
2764                 goto out;
2765         }
2766         if (ret == 0) {
2767                 if (check_path_shared(root, path))
2768                         goto out;
2769         } else {
2770                 check_link = 0;
2771         }
2772         btrfs_release_path(root, path);
2773
2774         if (ret == 0 && S_ISREG(inode->i_mode)) {
2775                 ret = btrfs_lookup_file_extent(trans, root, path,
2776                                                inode->i_ino, (u64)-1, 0);
2777                 if (ret < 0) {
2778                         err = ret;
2779                         goto out;
2780                 }
2781                 BUG_ON(ret == 0);
2782                 if (check_path_shared(root, path))
2783                         goto out;
2784                 btrfs_release_path(root, path);
2785         }
2786
2787         if (!check_link) {
2788                 err = 0;
2789                 goto out;
2790         }
2791
2792         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2793                                 dentry->d_name.name, dentry->d_name.len, 0);
2794         if (IS_ERR(di)) {
2795                 err = PTR_ERR(di);
2796                 goto out;
2797         }
2798         if (di) {
2799                 if (check_path_shared(root, path))
2800                         goto out;
2801         } else {
2802                 err = 0;
2803                 goto out;
2804         }
2805         btrfs_release_path(root, path);
2806
2807         ref = btrfs_lookup_inode_ref(trans, root, path,
2808                                 dentry->d_name.name, dentry->d_name.len,
2809                                 inode->i_ino, dir->i_ino, 0);
2810         if (IS_ERR(ref)) {
2811                 err = PTR_ERR(ref);
2812                 goto out;
2813         }
2814         BUG_ON(!ref);
2815         if (check_path_shared(root, path))
2816                 goto out;
2817         index = btrfs_inode_ref_index(path->nodes[0], ref);
2818         btrfs_release_path(root, path);
2819
2820         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2821                                 dentry->d_name.name, dentry->d_name.len, 0);
2822         if (IS_ERR(di)) {
2823                 err = PTR_ERR(di);
2824                 goto out;
2825         }
2826         BUG_ON(ret == -ENOENT);
2827         if (check_path_shared(root, path))
2828                 goto out;
2829
2830         err = 0;
2831 out:
2832         btrfs_free_path(path);
2833         if (err) {
2834                 btrfs_end_transaction(trans, root);
2835                 root->fs_info->enospc_unlink = 0;
2836                 return ERR_PTR(err);
2837         }
2838
2839         trans->block_rsv = &root->fs_info->global_block_rsv;
2840         return trans;
2841 }
2842
2843 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2844                                struct btrfs_root *root)
2845 {
2846         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2847                 BUG_ON(!root->fs_info->enospc_unlink);
2848                 root->fs_info->enospc_unlink = 0;
2849         }
2850         btrfs_end_transaction_throttle(trans, root);
2851 }
2852
2853 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2854 {
2855         struct btrfs_root *root = BTRFS_I(dir)->root;
2856         struct btrfs_trans_handle *trans;
2857         struct inode *inode = dentry->d_inode;
2858         int ret;
2859         unsigned long nr = 0;
2860
2861         trans = __unlink_start_trans(dir, dentry);
2862         if (IS_ERR(trans))
2863                 return PTR_ERR(trans);
2864
2865         btrfs_set_trans_block_group(trans, dir);
2866
2867         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2868
2869         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2870                                  dentry->d_name.name, dentry->d_name.len);
2871         BUG_ON(ret);
2872
2873         if (inode->i_nlink == 0) {
2874                 ret = btrfs_orphan_add(trans, inode);
2875                 BUG_ON(ret);
2876         }
2877
2878         nr = trans->blocks_used;
2879         __unlink_end_trans(trans, root);
2880         btrfs_btree_balance_dirty(root, nr);
2881         return ret;
2882 }
2883
2884 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2885                         struct btrfs_root *root,
2886                         struct inode *dir, u64 objectid,
2887                         const char *name, int name_len)
2888 {
2889         struct btrfs_path *path;
2890         struct extent_buffer *leaf;
2891         struct btrfs_dir_item *di;
2892         struct btrfs_key key;
2893         u64 index;
2894         int ret;
2895
2896         path = btrfs_alloc_path();
2897         if (!path)
2898                 return -ENOMEM;
2899
2900         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2901                                    name, name_len, -1);
2902         BUG_ON(!di || IS_ERR(di));
2903
2904         leaf = path->nodes[0];
2905         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2906         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2907         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2908         BUG_ON(ret);
2909         btrfs_release_path(root, path);
2910
2911         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2912                                  objectid, root->root_key.objectid,
2913                                  dir->i_ino, &index, name, name_len);
2914         if (ret < 0) {
2915                 BUG_ON(ret != -ENOENT);
2916                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2917                                                  name, name_len);
2918                 BUG_ON(!di || IS_ERR(di));
2919
2920                 leaf = path->nodes[0];
2921                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2922                 btrfs_release_path(root, path);
2923                 index = key.offset;
2924         }
2925
2926         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2927                                          index, name, name_len, -1);
2928         BUG_ON(!di || IS_ERR(di));
2929
2930         leaf = path->nodes[0];
2931         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2932         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2933         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2934         BUG_ON(ret);
2935         btrfs_release_path(root, path);
2936
2937         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2938         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2939         ret = btrfs_update_inode(trans, root, dir);
2940         BUG_ON(ret);
2941
2942         btrfs_free_path(path);
2943         return 0;
2944 }
2945
2946 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2947 {
2948         struct inode *inode = dentry->d_inode;
2949         int err = 0;
2950         struct btrfs_root *root = BTRFS_I(dir)->root;
2951         struct btrfs_trans_handle *trans;
2952         unsigned long nr = 0;
2953
2954         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2955             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2956                 return -ENOTEMPTY;
2957
2958         trans = __unlink_start_trans(dir, dentry);
2959         if (IS_ERR(trans))
2960                 return PTR_ERR(trans);
2961
2962         btrfs_set_trans_block_group(trans, dir);
2963
2964         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2965                 err = btrfs_unlink_subvol(trans, root, dir,
2966                                           BTRFS_I(inode)->location.objectid,
2967                                           dentry->d_name.name,
2968                                           dentry->d_name.len);
2969                 goto out;
2970         }
2971
2972         err = btrfs_orphan_add(trans, inode);
2973         if (err)
2974                 goto out;
2975
2976         /* now the directory is empty */
2977         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2978                                  dentry->d_name.name, dentry->d_name.len);
2979         if (!err)
2980                 btrfs_i_size_write(inode, 0);
2981 out:
2982         nr = trans->blocks_used;
2983         __unlink_end_trans(trans, root);
2984         btrfs_btree_balance_dirty(root, nr);
2985
2986         return err;
2987 }
2988
2989 #if 0
2990 /*
2991  * when truncating bytes in a file, it is possible to avoid reading
2992  * the leaves that contain only checksum items.  This can be the
2993  * majority of the IO required to delete a large file, but it must
2994  * be done carefully.
2995  *
2996  * The keys in the level just above the leaves are checked to make sure
2997  * the lowest key in a given leaf is a csum key, and starts at an offset
2998  * after the new  size.
2999  *
3000  * Then the key for the next leaf is checked to make sure it also has
3001  * a checksum item for the same file.  If it does, we know our target leaf
3002  * contains only checksum items, and it can be safely freed without reading
3003  * it.
3004  *
3005  * This is just an optimization targeted at large files.  It may do
3006  * nothing.  It will return 0 unless things went badly.
3007  */
3008 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3009                                      struct btrfs_root *root,
3010                                      struct btrfs_path *path,
3011                                      struct inode *inode, u64 new_size)
3012 {
3013         struct btrfs_key key;
3014         int ret;
3015         int nritems;
3016         struct btrfs_key found_key;
3017         struct btrfs_key other_key;
3018         struct btrfs_leaf_ref *ref;
3019         u64 leaf_gen;
3020         u64 leaf_start;
3021
3022         path->lowest_level = 1;
3023         key.objectid = inode->i_ino;
3024         key.type = BTRFS_CSUM_ITEM_KEY;
3025         key.offset = new_size;
3026 again:
3027         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3028         if (ret < 0)
3029                 goto out;
3030
3031         if (path->nodes[1] == NULL) {
3032                 ret = 0;
3033                 goto out;
3034         }
3035         ret = 0;
3036         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3037         nritems = btrfs_header_nritems(path->nodes[1]);
3038
3039         if (!nritems)
3040                 goto out;
3041
3042         if (path->slots[1] >= nritems)
3043                 goto next_node;
3044
3045         /* did we find a key greater than anything we want to delete? */
3046         if (found_key.objectid > inode->i_ino ||
3047            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3048                 goto out;
3049
3050         /* we check the next key in the node to make sure the leave contains
3051          * only checksum items.  This comparison doesn't work if our
3052          * leaf is the last one in the node
3053          */
3054         if (path->slots[1] + 1 >= nritems) {
3055 next_node:
3056                 /* search forward from the last key in the node, this
3057                  * will bring us into the next node in the tree
3058                  */
3059                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3060
3061                 /* unlikely, but we inc below, so check to be safe */
3062                 if (found_key.offset == (u64)-1)
3063                         goto out;
3064
3065                 /* search_forward needs a path with locks held, do the
3066                  * search again for the original key.  It is possible
3067                  * this will race with a balance and return a path that
3068                  * we could modify, but this drop is just an optimization
3069                  * and is allowed to miss some leaves.
3070                  */
3071                 btrfs_release_path(root, path);
3072                 found_key.offset++;
3073
3074                 /* setup a max key for search_forward */
3075                 other_key.offset = (u64)-1;
3076                 other_key.type = key.type;
3077                 other_key.objectid = key.objectid;
3078
3079                 path->keep_locks = 1;
3080                 ret = btrfs_search_forward(root, &found_key, &other_key,
3081                                            path, 0, 0);
3082                 path->keep_locks = 0;
3083                 if (ret || found_key.objectid != key.objectid ||
3084                     found_key.type != key.type) {
3085                         ret = 0;
3086                         goto out;
3087                 }
3088
3089                 key.offset = found_key.offset;
3090                 btrfs_release_path(root, path);
3091                 cond_resched();
3092                 goto again;
3093         }
3094
3095         /* we know there's one more slot after us in the tree,
3096          * read that key so we can verify it is also a checksum item
3097          */
3098         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3099
3100         if (found_key.objectid < inode->i_ino)
3101                 goto next_key;
3102
3103         if (found_key.type != key.type || found_key.offset < new_size)
3104                 goto next_key;
3105
3106         /*
3107          * if the key for the next leaf isn't a csum key from this objectid,
3108          * we can't be sure there aren't good items inside this leaf.
3109          * Bail out
3110          */
3111         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3112                 goto out;
3113
3114         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3115         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3116         /*
3117          * it is safe to delete this leaf, it contains only
3118          * csum items from this inode at an offset >= new_size
3119          */
3120         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3121         BUG_ON(ret);
3122
3123         if (root->ref_cows && leaf_gen < trans->transid) {
3124                 ref = btrfs_alloc_leaf_ref(root, 0);
3125                 if (ref) {
3126                         ref->root_gen = root->root_key.offset;
3127                         ref->bytenr = leaf_start;
3128                         ref->owner = 0;
3129                         ref->generation = leaf_gen;
3130                         ref->nritems = 0;
3131
3132                         btrfs_sort_leaf_ref(ref);
3133
3134                         ret = btrfs_add_leaf_ref(root, ref, 0);
3135                         WARN_ON(ret);
3136                         btrfs_free_leaf_ref(root, ref);
3137                 } else {
3138                         WARN_ON(1);
3139                 }
3140         }
3141 next_key:
3142         btrfs_release_path(root, path);
3143
3144         if (other_key.objectid == inode->i_ino &&
3145             other_key.type == key.type && other_key.offset > key.offset) {
3146                 key.offset = other_key.offset;
3147                 cond_resched();
3148                 goto again;
3149         }
3150         ret = 0;
3151 out:
3152         /* fixup any changes we've made to the path */
3153         path->lowest_level = 0;
3154         path->keep_locks = 0;
3155         btrfs_release_path(root, path);
3156         return ret;
3157 }
3158
3159 #endif
3160
3161 /*
3162  * this can truncate away extent items, csum items and directory items.
3163  * It starts at a high offset and removes keys until it can't find
3164  * any higher than new_size
3165  *
3166  * csum items that cross the new i_size are truncated to the new size
3167  * as well.
3168  *
3169  * min_type is the minimum key type to truncate down to.  If set to 0, this
3170  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3171  */
3172 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3173                                struct btrfs_root *root,
3174                                struct inode *inode,
3175                                u64 new_size, u32 min_type)
3176 {
3177         struct btrfs_path *path;
3178         struct extent_buffer *leaf;
3179         struct btrfs_file_extent_item *fi;
3180         struct btrfs_key key;
3181         struct btrfs_key found_key;
3182         u64 extent_start = 0;
3183         u64 extent_num_bytes = 0;
3184         u64 extent_offset = 0;
3185         u64 item_end = 0;
3186         u64 mask = root->sectorsize - 1;
3187         u32 found_type = (u8)-1;
3188         int found_extent;
3189         int del_item;
3190         int pending_del_nr = 0;
3191         int pending_del_slot = 0;
3192         int extent_type = -1;
3193         int encoding;
3194         int ret;
3195         int err = 0;
3196
3197         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3198
3199         if (root->ref_cows)
3200                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3201
3202         path = btrfs_alloc_path();
3203         BUG_ON(!path);
3204         path->reada = -1;
3205
3206         key.objectid = inode->i_ino;
3207         key.offset = (u64)-1;
3208         key.type = (u8)-1;
3209
3210 search_again:
3211         path->leave_spinning = 1;
3212         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3213         if (ret < 0) {
3214                 err = ret;
3215                 goto out;
3216         }
3217
3218         if (ret > 0) {
3219                 /* there are no items in the tree for us to truncate, we're
3220                  * done
3221                  */
3222                 if (path->slots[0] == 0)
3223                         goto out;
3224                 path->slots[0]--;
3225         }
3226
3227         while (1) {
3228                 fi = NULL;
3229                 leaf = path->nodes[0];
3230                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3231                 found_type = btrfs_key_type(&found_key);
3232                 encoding = 0;
3233
3234                 if (found_key.objectid != inode->i_ino)
3235                         break;
3236
3237                 if (found_type < min_type)
3238                         break;
3239
3240                 item_end = found_key.offset;
3241                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3242                         fi = btrfs_item_ptr(leaf, path->slots[0],
3243                                             struct btrfs_file_extent_item);
3244                         extent_type = btrfs_file_extent_type(leaf, fi);
3245                         encoding = btrfs_file_extent_compression(leaf, fi);
3246                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3247                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3248
3249                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3250                                 item_end +=
3251                                     btrfs_file_extent_num_bytes(leaf, fi);
3252                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3253                                 item_end += btrfs_file_extent_inline_len(leaf,
3254                                                                          fi);
3255                         }
3256                         item_end--;
3257                 }
3258                 if (found_type > min_type) {
3259                         del_item = 1;
3260                 } else {
3261                         if (item_end < new_size)
3262                                 break;
3263                         if (found_key.offset >= new_size)
3264                                 del_item = 1;
3265                         else
3266                                 del_item = 0;
3267                 }
3268                 found_extent = 0;
3269                 /* FIXME, shrink the extent if the ref count is only 1 */
3270                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3271                         goto delete;
3272
3273                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3274                         u64 num_dec;
3275                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3276                         if (!del_item && !encoding) {
3277                                 u64 orig_num_bytes =
3278                                         btrfs_file_extent_num_bytes(leaf, fi);
3279                                 extent_num_bytes = new_size -
3280                                         found_key.offset + root->sectorsize - 1;
3281                                 extent_num_bytes = extent_num_bytes &
3282                                         ~((u64)root->sectorsize - 1);
3283                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3284                                                          extent_num_bytes);
3285                                 num_dec = (orig_num_bytes -
3286                                            extent_num_bytes);
3287                                 if (root->ref_cows && extent_start != 0)
3288                                         inode_sub_bytes(inode, num_dec);
3289                                 btrfs_mark_buffer_dirty(leaf);
3290                         } else {
3291                                 extent_num_bytes =
3292                                         btrfs_file_extent_disk_num_bytes(leaf,
3293                                                                          fi);
3294                                 extent_offset = found_key.offset -
3295                                         btrfs_file_extent_offset(leaf, fi);
3296
3297                                 /* FIXME blocksize != 4096 */
3298                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3299                                 if (extent_start != 0) {
3300                                         found_extent = 1;
3301                                         if (root->ref_cows)
3302                                                 inode_sub_bytes(inode, num_dec);
3303                                 }
3304                         }
3305                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3306                         /*
3307                          * we can't truncate inline items that have had
3308                          * special encodings
3309                          */
3310                         if (!del_item &&
3311                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3312                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3313                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3314                                 u32 size = new_size - found_key.offset;
3315
3316                                 if (root->ref_cows) {
3317                                         inode_sub_bytes(inode, item_end + 1 -
3318                                                         new_size);
3319                                 }
3320                                 size =
3321                                     btrfs_file_extent_calc_inline_size(size);
3322                                 ret = btrfs_truncate_item(trans, root, path,
3323                                                           size, 1);
3324                                 BUG_ON(ret);
3325                         } else if (root->ref_cows) {
3326                                 inode_sub_bytes(inode, item_end + 1 -
3327                                                 found_key.offset);
3328                         }
3329                 }
3330 delete:
3331                 if (del_item) {
3332                         if (!pending_del_nr) {
3333                                 /* no pending yet, add ourselves */
3334                                 pending_del_slot = path->slots[0];
3335                                 pending_del_nr = 1;
3336                         } else if (pending_del_nr &&
3337                                    path->slots[0] + 1 == pending_del_slot) {
3338                                 /* hop on the pending chunk */
3339                                 pending_del_nr++;
3340                                 pending_del_slot = path->slots[0];
3341                         } else {
3342                                 BUG();
3343                         }
3344                 } else {
3345                         break;
3346                 }
3347                 if (found_extent && root->ref_cows) {
3348                         btrfs_set_path_blocking(path);
3349                         ret = btrfs_free_extent(trans, root, extent_start,
3350                                                 extent_num_bytes, 0,
3351                                                 btrfs_header_owner(leaf),
3352                                                 inode->i_ino, extent_offset);
3353                         BUG_ON(ret);
3354                 }
3355
3356                 if (found_type == BTRFS_INODE_ITEM_KEY)
3357                         break;
3358
3359                 if (path->slots[0] == 0 ||
3360                     path->slots[0] != pending_del_slot) {
3361                         if (root->ref_cows) {
3362                                 err = -EAGAIN;
3363                                 goto out;
3364                         }
3365                         if (pending_del_nr) {
3366                                 ret = btrfs_del_items(trans, root, path,
3367                                                 pending_del_slot,
3368                                                 pending_del_nr);
3369                                 BUG_ON(ret);
3370                                 pending_del_nr = 0;
3371                         }
3372                         btrfs_release_path(root, path);
3373                         goto search_again;
3374                 } else {
3375                         path->slots[0]--;
3376                 }
3377         }
3378 out:
3379         if (pending_del_nr) {
3380                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3381                                       pending_del_nr);
3382                 BUG_ON(ret);
3383         }
3384         btrfs_free_path(path);
3385         return err;
3386 }
3387
3388 /*
3389  * taken from block_truncate_page, but does cow as it zeros out
3390  * any bytes left in the last page in the file.
3391  */
3392 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3393 {
3394         struct inode *inode = mapping->host;
3395         struct btrfs_root *root = BTRFS_I(inode)->root;
3396         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3397         struct btrfs_ordered_extent *ordered;
3398         struct extent_state *cached_state = NULL;
3399         char *kaddr;
3400         u32 blocksize = root->sectorsize;
3401         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3402         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3403         struct page *page;
3404         int ret = 0;
3405         u64 page_start;
3406         u64 page_end;
3407
3408         if ((offset & (blocksize - 1)) == 0)
3409                 goto out;
3410         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3411         if (ret)
3412                 goto out;
3413
3414         ret = -ENOMEM;
3415 again:
3416         page = grab_cache_page(mapping, index);
3417         if (!page) {
3418                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3419                 goto out;
3420         }
3421
3422         page_start = page_offset(page);
3423         page_end = page_start + PAGE_CACHE_SIZE - 1;
3424
3425         if (!PageUptodate(page)) {
3426                 ret = btrfs_readpage(NULL, page);
3427                 lock_page(page);
3428                 if (page->mapping != mapping) {
3429                         unlock_page(page);
3430                         page_cache_release(page);
3431                         goto again;
3432                 }
3433                 if (!PageUptodate(page)) {
3434                         ret = -EIO;
3435                         goto out_unlock;
3436                 }
3437         }
3438         wait_on_page_writeback(page);
3439
3440         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3441                          GFP_NOFS);
3442         set_page_extent_mapped(page);
3443
3444         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3445         if (ordered) {
3446                 unlock_extent_cached(io_tree, page_start, page_end,
3447                                      &cached_state, GFP_NOFS);
3448                 unlock_page(page);
3449                 page_cache_release(page);
3450                 btrfs_start_ordered_extent(inode, ordered, 1);
3451                 btrfs_put_ordered_extent(ordered);
3452                 goto again;
3453         }
3454
3455         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3456                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3457                           0, 0, &cached_state, GFP_NOFS);
3458
3459         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3460                                         &cached_state);
3461         if (ret) {
3462                 unlock_extent_cached(io_tree, page_start, page_end,
3463                                      &cached_state, GFP_NOFS);
3464                 goto out_unlock;
3465         }
3466
3467         ret = 0;
3468         if (offset != PAGE_CACHE_SIZE) {
3469                 kaddr = kmap(page);
3470                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3471                 flush_dcache_page(page);
3472                 kunmap(page);
3473         }
3474         ClearPageChecked(page);
3475         set_page_dirty(page);
3476         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3477                              GFP_NOFS);
3478
3479 out_unlock:
3480         if (ret)
3481                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3482         unlock_page(page);
3483         page_cache_release(page);
3484 out:
3485         return ret;
3486 }
3487
3488 int btrfs_cont_expand(struct inode *inode, loff_t size)
3489 {
3490         struct btrfs_trans_handle *trans;
3491         struct btrfs_root *root = BTRFS_I(inode)->root;
3492         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3493         struct extent_map *em = NULL;
3494         struct extent_state *cached_state = NULL;
3495         u64 mask = root->sectorsize - 1;
3496         u64 hole_start = (inode->i_size + mask) & ~mask;
3497         u64 block_end = (size + mask) & ~mask;
3498         u64 last_byte;
3499         u64 cur_offset;
3500         u64 hole_size;
3501         int err = 0;
3502
3503         if (size <= hole_start)
3504                 return 0;
3505
3506         while (1) {
3507                 struct btrfs_ordered_extent *ordered;
3508                 btrfs_wait_ordered_range(inode, hole_start,
3509                                          block_end - hole_start);
3510                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3511                                  &cached_state, GFP_NOFS);
3512                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3513                 if (!ordered)
3514                         break;
3515                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3516                                      &cached_state, GFP_NOFS);
3517                 btrfs_put_ordered_extent(ordered);
3518         }
3519
3520         cur_offset = hole_start;
3521         while (1) {
3522                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3523                                 block_end - cur_offset, 0);
3524                 BUG_ON(IS_ERR(em) || !em);
3525                 last_byte = min(extent_map_end(em), block_end);
3526                 last_byte = (last_byte + mask) & ~mask;
3527                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3528                         u64 hint_byte = 0;
3529                         hole_size = last_byte - cur_offset;
3530
3531                         trans = btrfs_start_transaction(root, 2);
3532                         if (IS_ERR(trans)) {
3533                                 err = PTR_ERR(trans);
3534                                 break;
3535                         }
3536                         btrfs_set_trans_block_group(trans, inode);
3537
3538                         err = btrfs_drop_extents(trans, inode, cur_offset,
3539                                                  cur_offset + hole_size,
3540                                                  &hint_byte, 1);
3541                         BUG_ON(err);
3542
3543                         err = btrfs_insert_file_extent(trans, root,
3544                                         inode->i_ino, cur_offset, 0,
3545                                         0, hole_size, 0, hole_size,
3546                                         0, 0, 0);
3547                         BUG_ON(err);
3548
3549                         btrfs_drop_extent_cache(inode, hole_start,
3550                                         last_byte - 1, 0);
3551
3552                         btrfs_end_transaction(trans, root);
3553                 }
3554                 free_extent_map(em);
3555                 em = NULL;
3556                 cur_offset = last_byte;
3557                 if (cur_offset >= block_end)
3558                         break;
3559         }
3560
3561         free_extent_map(em);
3562         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3563                              GFP_NOFS);
3564         return err;
3565 }
3566
3567 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3568 {
3569         struct btrfs_root *root = BTRFS_I(inode)->root;
3570         struct btrfs_trans_handle *trans;
3571         unsigned long nr;
3572         int ret;
3573
3574         if (attr->ia_size == inode->i_size)
3575                 return 0;
3576
3577         if (attr->ia_size > inode->i_size) {
3578                 unsigned long limit;
3579                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3580                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3581                         return -EFBIG;
3582                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3583                         send_sig(SIGXFSZ, current, 0);
3584                         return -EFBIG;
3585                 }
3586         }
3587
3588         trans = btrfs_start_transaction(root, 5);
3589         if (IS_ERR(trans))
3590                 return PTR_ERR(trans);
3591
3592         btrfs_set_trans_block_group(trans, inode);
3593
3594         ret = btrfs_orphan_add(trans, inode);
3595         BUG_ON(ret);
3596
3597         nr = trans->blocks_used;
3598         btrfs_end_transaction(trans, root);
3599         btrfs_btree_balance_dirty(root, nr);
3600
3601         if (attr->ia_size > inode->i_size) {
3602                 ret = btrfs_cont_expand(inode, attr->ia_size);
3603                 if (ret) {
3604                         btrfs_truncate(inode);
3605                         return ret;
3606                 }
3607
3608                 i_size_write(inode, attr->ia_size);
3609                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3610
3611                 trans = btrfs_start_transaction(root, 0);
3612                 BUG_ON(IS_ERR(trans));
3613                 btrfs_set_trans_block_group(trans, inode);
3614                 trans->block_rsv = root->orphan_block_rsv;
3615                 BUG_ON(!trans->block_rsv);
3616
3617                 ret = btrfs_update_inode(trans, root, inode);
3618                 BUG_ON(ret);
3619                 if (inode->i_nlink > 0) {
3620                         ret = btrfs_orphan_del(trans, inode);
3621                         BUG_ON(ret);
3622                 }
3623                 nr = trans->blocks_used;
3624                 btrfs_end_transaction(trans, root);
3625                 btrfs_btree_balance_dirty(root, nr);
3626                 return 0;
3627         }
3628
3629         /*
3630          * We're truncating a file that used to have good data down to
3631          * zero. Make sure it gets into the ordered flush list so that
3632          * any new writes get down to disk quickly.
3633          */
3634         if (attr->ia_size == 0)
3635                 BTRFS_I(inode)->ordered_data_close = 1;
3636
3637         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3638         ret = vmtruncate(inode, attr->ia_size);
3639         BUG_ON(ret);
3640
3641         return 0;
3642 }
3643
3644 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3645 {
3646         struct inode *inode = dentry->d_inode;
3647         int err;
3648
3649         err = inode_change_ok(inode, attr);
3650         if (err)
3651                 return err;
3652
3653         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3654                 err = btrfs_setattr_size(inode, attr);
3655                 if (err)
3656                         return err;
3657         }
3658
3659         if (attr->ia_valid) {
3660                 setattr_copy(inode, attr);
3661                 mark_inode_dirty(inode);
3662
3663                 if (attr->ia_valid & ATTR_MODE)
3664                         err = btrfs_acl_chmod(inode);
3665         }
3666
3667         return err;
3668 }
3669
3670 void btrfs_evict_inode(struct inode *inode)
3671 {
3672         struct btrfs_trans_handle *trans;
3673         struct btrfs_root *root = BTRFS_I(inode)->root;
3674         unsigned long nr;
3675         int ret;
3676
3677         truncate_inode_pages(&inode->i_data, 0);
3678         if (inode->i_nlink && btrfs_root_refs(&root->root_item) != 0)
3679                 goto no_delete;
3680
3681         if (is_bad_inode(inode)) {
3682                 btrfs_orphan_del(NULL, inode);
3683                 goto no_delete;
3684         }
3685         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3686         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3687
3688         if (root->fs_info->log_root_recovering) {
3689                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3690                 goto no_delete;
3691         }
3692
3693         if (inode->i_nlink > 0) {
3694                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3695                 goto no_delete;
3696         }
3697
3698         btrfs_i_size_write(inode, 0);
3699
3700         while (1) {
3701                 trans = btrfs_start_transaction(root, 0);
3702                 BUG_ON(IS_ERR(trans));
3703                 btrfs_set_trans_block_group(trans, inode);
3704                 trans->block_rsv = root->orphan_block_rsv;
3705
3706                 ret = btrfs_block_rsv_check(trans, root,
3707                                             root->orphan_block_rsv, 0, 5);
3708                 if (ret) {
3709                         BUG_ON(ret != -EAGAIN);
3710                         ret = btrfs_commit_transaction(trans, root);
3711                         BUG_ON(ret);
3712                         continue;
3713                 }
3714
3715                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3716                 if (ret != -EAGAIN)
3717                         break;
3718
3719                 nr = trans->blocks_used;
3720                 btrfs_end_transaction(trans, root);
3721                 trans = NULL;
3722                 btrfs_btree_balance_dirty(root, nr);
3723
3724         }
3725
3726         if (ret == 0) {
3727                 ret = btrfs_orphan_del(trans, inode);
3728                 BUG_ON(ret);
3729         }
3730
3731         nr = trans->blocks_used;
3732         btrfs_end_transaction(trans, root);
3733         btrfs_btree_balance_dirty(root, nr);
3734 no_delete:
3735         end_writeback(inode);
3736         return;
3737 }
3738
3739 /*
3740  * this returns the key found in the dir entry in the location pointer.
3741  * If no dir entries were found, location->objectid is 0.
3742  */
3743 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3744                                struct btrfs_key *location)
3745 {
3746         const char *name = dentry->d_name.name;
3747         int namelen = dentry->d_name.len;
3748         struct btrfs_dir_item *di;
3749         struct btrfs_path *path;
3750         struct btrfs_root *root = BTRFS_I(dir)->root;
3751         int ret = 0;
3752
3753         path = btrfs_alloc_path();
3754         BUG_ON(!path);
3755
3756         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3757                                     namelen, 0);
3758         if (IS_ERR(di))
3759                 ret = PTR_ERR(di);
3760
3761         if (!di || IS_ERR(di))
3762                 goto out_err;
3763
3764         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3765 out:
3766         btrfs_free_path(path);
3767         return ret;
3768 out_err:
3769         location->objectid = 0;
3770         goto out;
3771 }
3772
3773 /*
3774  * when we hit a tree root in a directory, the btrfs part of the inode
3775  * needs to be changed to reflect the root directory of the tree root.  This
3776  * is kind of like crossing a mount point.
3777  */
3778 static int fixup_tree_root_location(struct btrfs_root *root,
3779                                     struct inode *dir,
3780                                     struct dentry *dentry,
3781                                     struct btrfs_key *location,
3782                                     struct btrfs_root **sub_root)
3783 {
3784         struct btrfs_path *path;
3785         struct btrfs_root *new_root;
3786         struct btrfs_root_ref *ref;
3787         struct extent_buffer *leaf;
3788         int ret;
3789         int err = 0;
3790
3791         path = btrfs_alloc_path();
3792         if (!path) {
3793                 err = -ENOMEM;
3794                 goto out;
3795         }
3796
3797         err = -ENOENT;
3798         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3799                                   BTRFS_I(dir)->root->root_key.objectid,
3800                                   location->objectid);
3801         if (ret) {
3802                 if (ret < 0)
3803                         err = ret;
3804                 goto out;
3805         }
3806
3807         leaf = path->nodes[0];
3808         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3809         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3810             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3811                 goto out;
3812
3813         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3814                                    (unsigned long)(ref + 1),
3815                                    dentry->d_name.len);
3816         if (ret)
3817                 goto out;
3818
3819         btrfs_release_path(root->fs_info->tree_root, path);
3820
3821         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3822         if (IS_ERR(new_root)) {
3823                 err = PTR_ERR(new_root);
3824                 goto out;
3825         }
3826
3827         if (btrfs_root_refs(&new_root->root_item) == 0) {
3828                 err = -ENOENT;
3829                 goto out;
3830         }
3831
3832         *sub_root = new_root;
3833         location->objectid = btrfs_root_dirid(&new_root->root_item);
3834         location->type = BTRFS_INODE_ITEM_KEY;
3835         location->offset = 0;
3836         err = 0;
3837 out:
3838         btrfs_free_path(path);
3839         return err;
3840 }
3841
3842 static void inode_tree_add(struct inode *inode)
3843 {
3844         struct btrfs_root *root = BTRFS_I(inode)->root;
3845         struct btrfs_inode *entry;
3846         struct rb_node **p;
3847         struct rb_node *parent;
3848 again:
3849         p = &root->inode_tree.rb_node;
3850         parent = NULL;
3851
3852         if (inode_unhashed(inode))
3853                 return;
3854
3855         spin_lock(&root->inode_lock);
3856         while (*p) {
3857                 parent = *p;
3858                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3859
3860                 if (inode->i_ino < entry->vfs_inode.i_ino)
3861                         p = &parent->rb_left;
3862                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3863                         p = &parent->rb_right;
3864                 else {
3865                         WARN_ON(!(entry->vfs_inode.i_state &
3866                                   (I_WILL_FREE | I_FREEING)));
3867                         rb_erase(parent, &root->inode_tree);
3868                         RB_CLEAR_NODE(parent);
3869                         spin_unlock(&root->inode_lock);
3870                         goto again;
3871                 }
3872         }
3873         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3874         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3875         spin_unlock(&root->inode_lock);
3876 }
3877
3878 static void inode_tree_del(struct inode *inode)
3879 {
3880         struct btrfs_root *root = BTRFS_I(inode)->root;
3881         int empty = 0;
3882
3883         spin_lock(&root->inode_lock);
3884         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3885                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3886                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3887                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3888         }
3889         spin_unlock(&root->inode_lock);
3890
3891         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3892                 synchronize_srcu(&root->fs_info->subvol_srcu);
3893                 spin_lock(&root->inode_lock);
3894                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3895                 spin_unlock(&root->inode_lock);
3896                 if (empty)
3897                         btrfs_add_dead_root(root);
3898         }
3899 }
3900
3901 int btrfs_invalidate_inodes(struct btrfs_root *root)
3902 {
3903         struct rb_node *node;
3904         struct rb_node *prev;
3905         struct btrfs_inode *entry;
3906         struct inode *inode;
3907         u64 objectid = 0;
3908
3909         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3910
3911         spin_lock(&root->inode_lock);
3912 again:
3913         node = root->inode_tree.rb_node;
3914         prev = NULL;
3915         while (node) {
3916                 prev = node;
3917                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3918
3919                 if (objectid < entry->vfs_inode.i_ino)
3920                         node = node->rb_left;
3921                 else if (objectid > entry->vfs_inode.i_ino)
3922                         node = node->rb_right;
3923                 else
3924                         break;
3925         }
3926         if (!node) {
3927                 while (prev) {
3928                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3929                         if (objectid <= entry->vfs_inode.i_ino) {
3930                                 node = prev;
3931                                 break;
3932                         }
3933                         prev = rb_next(prev);
3934                 }
3935         }
3936         while (node) {
3937                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3938                 objectid = entry->vfs_inode.i_ino + 1;
3939                 inode = igrab(&entry->vfs_inode);
3940                 if (inode) {
3941                         spin_unlock(&root->inode_lock);
3942                         if (atomic_read(&inode->i_count) > 1)
3943                                 d_prune_aliases(inode);
3944                         /*
3945                          * btrfs_drop_inode will have it removed from
3946                          * the inode cache when its usage count
3947                          * hits zero.
3948                          */
3949                         iput(inode);
3950                         cond_resched();
3951                         spin_lock(&root->inode_lock);
3952                         goto again;
3953                 }
3954
3955                 if (cond_resched_lock(&root->inode_lock))
3956                         goto again;
3957
3958                 node = rb_next(node);
3959         }
3960         spin_unlock(&root->inode_lock);
3961         return 0;
3962 }
3963
3964 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3965 {
3966         struct btrfs_iget_args *args = p;
3967         inode->i_ino = args->ino;
3968         BTRFS_I(inode)->root = args->root;
3969         btrfs_set_inode_space_info(args->root, inode);
3970         return 0;
3971 }
3972
3973 static int btrfs_find_actor(struct inode *inode, void *opaque)
3974 {
3975         struct btrfs_iget_args *args = opaque;
3976         return args->ino == inode->i_ino &&
3977                 args->root == BTRFS_I(inode)->root;
3978 }
3979
3980 static struct inode *btrfs_iget_locked(struct super_block *s,
3981                                        u64 objectid,
3982                                        struct btrfs_root *root)
3983 {
3984         struct inode *inode;
3985         struct btrfs_iget_args args;
3986         args.ino = objectid;
3987         args.root = root;
3988
3989         inode = iget5_locked(s, objectid, btrfs_find_actor,
3990                              btrfs_init_locked_inode,
3991                              (void *)&args);
3992         return inode;
3993 }
3994
3995 /* Get an inode object given its location and corresponding root.
3996  * Returns in *is_new if the inode was read from disk
3997  */
3998 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3999                          struct btrfs_root *root, int *new)
4000 {
4001         struct inode *inode;
4002
4003         inode = btrfs_iget_locked(s, location->objectid, root);
4004         if (!inode)
4005                 return ERR_PTR(-ENOMEM);
4006
4007         if (inode->i_state & I_NEW) {
4008                 BTRFS_I(inode)->root = root;
4009                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4010                 btrfs_read_locked_inode(inode);
4011
4012                 inode_tree_add(inode);
4013                 unlock_new_inode(inode);
4014                 if (new)
4015                         *new = 1;
4016         }
4017
4018         return inode;
4019 }
4020
4021 static struct inode *new_simple_dir(struct super_block *s,
4022                                     struct btrfs_key *key,
4023                                     struct btrfs_root *root)
4024 {
4025         struct inode *inode = new_inode(s);
4026
4027         if (!inode)
4028                 return ERR_PTR(-ENOMEM);
4029
4030         BTRFS_I(inode)->root = root;
4031         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4032         BTRFS_I(inode)->dummy_inode = 1;
4033
4034         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4035         inode->i_op = &simple_dir_inode_operations;
4036         inode->i_fop = &simple_dir_operations;
4037         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4038         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4039
4040         return inode;
4041 }
4042
4043 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4044 {
4045         struct inode *inode;
4046         struct btrfs_root *root = BTRFS_I(dir)->root;
4047         struct btrfs_root *sub_root = root;
4048         struct btrfs_key location;
4049         int index;
4050         int ret;
4051
4052         dentry->d_op = &btrfs_dentry_operations;
4053
4054         if (dentry->d_name.len > BTRFS_NAME_LEN)
4055                 return ERR_PTR(-ENAMETOOLONG);
4056
4057         ret = btrfs_inode_by_name(dir, dentry, &location);
4058
4059         if (ret < 0)
4060                 return ERR_PTR(ret);
4061
4062         if (location.objectid == 0)
4063                 return NULL;
4064
4065         if (location.type == BTRFS_INODE_ITEM_KEY) {
4066                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4067                 return inode;
4068         }
4069
4070         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4071
4072         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4073         ret = fixup_tree_root_location(root, dir, dentry,
4074                                        &location, &sub_root);
4075         if (ret < 0) {
4076                 if (ret != -ENOENT)
4077                         inode = ERR_PTR(ret);
4078                 else
4079                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4080         } else {
4081                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4082         }
4083         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4084
4085         if (root != sub_root) {
4086                 down_read(&root->fs_info->cleanup_work_sem);
4087                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4088                         btrfs_orphan_cleanup(sub_root);
4089                 up_read(&root->fs_info->cleanup_work_sem);
4090         }
4091
4092         return inode;
4093 }
4094
4095 static int btrfs_dentry_delete(struct dentry *dentry)
4096 {
4097         struct btrfs_root *root;
4098
4099         if (!dentry->d_inode && !IS_ROOT(dentry))
4100                 dentry = dentry->d_parent;
4101
4102         if (dentry->d_inode) {
4103                 root = BTRFS_I(dentry->d_inode)->root;
4104                 if (btrfs_root_refs(&root->root_item) == 0)
4105                         return 1;
4106         }
4107         return 0;
4108 }
4109
4110 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4111                                    struct nameidata *nd)
4112 {
4113         struct inode *inode;
4114
4115         inode = btrfs_lookup_dentry(dir, dentry);
4116         if (IS_ERR(inode))
4117                 return ERR_CAST(inode);
4118
4119         return d_splice_alias(inode, dentry);
4120 }
4121
4122 static unsigned char btrfs_filetype_table[] = {
4123         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4124 };
4125
4126 static int btrfs_real_readdir(struct file *filp, void *dirent,
4127                               filldir_t filldir)
4128 {
4129         struct inode *inode = filp->f_dentry->d_inode;
4130         struct btrfs_root *root = BTRFS_I(inode)->root;
4131         struct btrfs_item *item;
4132         struct btrfs_dir_item *di;
4133         struct btrfs_key key;
4134         struct btrfs_key found_key;
4135         struct btrfs_path *path;
4136         int ret;
4137         u32 nritems;
4138         struct extent_buffer *leaf;
4139         int slot;
4140         int advance;
4141         unsigned char d_type;
4142         int over = 0;
4143         u32 di_cur;
4144         u32 di_total;
4145         u32 di_len;
4146         int key_type = BTRFS_DIR_INDEX_KEY;
4147         char tmp_name[32];
4148         char *name_ptr;
4149         int name_len;
4150
4151         /* FIXME, use a real flag for deciding about the key type */
4152         if (root->fs_info->tree_root == root)
4153                 key_type = BTRFS_DIR_ITEM_KEY;
4154
4155         /* special case for "." */
4156         if (filp->f_pos == 0) {
4157                 over = filldir(dirent, ".", 1,
4158                                1, inode->i_ino,
4159                                DT_DIR);
4160                 if (over)
4161                         return 0;
4162                 filp->f_pos = 1;
4163         }
4164         /* special case for .., just use the back ref */
4165         if (filp->f_pos == 1) {
4166                 u64 pino = parent_ino(filp->f_path.dentry);
4167                 over = filldir(dirent, "..", 2,
4168                                2, pino, DT_DIR);
4169                 if (over)
4170                         return 0;
4171                 filp->f_pos = 2;
4172         }
4173         path = btrfs_alloc_path();
4174         path->reada = 2;
4175
4176         btrfs_set_key_type(&key, key_type);
4177         key.offset = filp->f_pos;
4178         key.objectid = inode->i_ino;
4179
4180         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4181         if (ret < 0)
4182                 goto err;
4183         advance = 0;
4184
4185         while (1) {
4186                 leaf = path->nodes[0];
4187                 nritems = btrfs_header_nritems(leaf);
4188                 slot = path->slots[0];
4189                 if (advance || slot >= nritems) {
4190                         if (slot >= nritems - 1) {
4191                                 ret = btrfs_next_leaf(root, path);
4192                                 if (ret)
4193                                         break;
4194                                 leaf = path->nodes[0];
4195                                 nritems = btrfs_header_nritems(leaf);
4196                                 slot = path->slots[0];
4197                         } else {
4198                                 slot++;
4199                                 path->slots[0]++;
4200                         }
4201                 }
4202
4203                 advance = 1;
4204                 item = btrfs_item_nr(leaf, slot);
4205                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4206
4207                 if (found_key.objectid != key.objectid)
4208                         break;
4209                 if (btrfs_key_type(&found_key) != key_type)
4210                         break;
4211                 if (found_key.offset < filp->f_pos)
4212                         continue;
4213
4214                 filp->f_pos = found_key.offset;
4215
4216                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4217                 di_cur = 0;
4218                 di_total = btrfs_item_size(leaf, item);
4219
4220                 while (di_cur < di_total) {
4221                         struct btrfs_key location;
4222
4223                         name_len = btrfs_dir_name_len(leaf, di);
4224                         if (name_len <= sizeof(tmp_name)) {
4225                                 name_ptr = tmp_name;
4226                         } else {
4227                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4228                                 if (!name_ptr) {
4229                                         ret = -ENOMEM;
4230                                         goto err;
4231                                 }
4232                         }
4233                         read_extent_buffer(leaf, name_ptr,
4234                                            (unsigned long)(di + 1), name_len);
4235
4236                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4237                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4238
4239                         /* is this a reference to our own snapshot? If so
4240                          * skip it
4241                          */
4242                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4243                             location.objectid == root->root_key.objectid) {
4244                                 over = 0;
4245                                 goto skip;
4246                         }
4247                         over = filldir(dirent, name_ptr, name_len,
4248                                        found_key.offset, location.objectid,
4249                                        d_type);
4250
4251 skip:
4252                         if (name_ptr != tmp_name)
4253                                 kfree(name_ptr);
4254
4255                         if (over)
4256                                 goto nopos;
4257                         di_len = btrfs_dir_name_len(leaf, di) +
4258                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4259                         di_cur += di_len;
4260                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4261                 }
4262         }
4263
4264         /* Reached end of directory/root. Bump pos past the last item. */
4265         if (key_type == BTRFS_DIR_INDEX_KEY)
4266                 /*
4267                  * 32-bit glibc will use getdents64, but then strtol -
4268                  * so the last number we can serve is this.
4269                  */
4270                 filp->f_pos = 0x7fffffff;
4271         else
4272                 filp->f_pos++;
4273 nopos:
4274         ret = 0;
4275 err:
4276         btrfs_free_path(path);
4277         return ret;
4278 }
4279
4280 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4281 {
4282         struct btrfs_root *root = BTRFS_I(inode)->root;
4283         struct btrfs_trans_handle *trans;
4284         int ret = 0;
4285
4286         if (BTRFS_I(inode)->dummy_inode)
4287                 return 0;
4288
4289         if (wbc->sync_mode == WB_SYNC_ALL) {
4290                 trans = btrfs_join_transaction(root, 1);
4291                 btrfs_set_trans_block_group(trans, inode);
4292                 ret = btrfs_commit_transaction(trans, root);
4293         }
4294         return ret;
4295 }
4296
4297 /*
4298  * This is somewhat expensive, updating the tree every time the
4299  * inode changes.  But, it is most likely to find the inode in cache.
4300  * FIXME, needs more benchmarking...there are no reasons other than performance
4301  * to keep or drop this code.
4302  */
4303 void btrfs_dirty_inode(struct inode *inode)
4304 {
4305         struct btrfs_root *root = BTRFS_I(inode)->root;
4306         struct btrfs_trans_handle *trans;
4307         int ret;
4308
4309         if (BTRFS_I(inode)->dummy_inode)
4310                 return;
4311
4312         trans = btrfs_join_transaction(root, 1);
4313         btrfs_set_trans_block_group(trans, inode);
4314
4315         ret = btrfs_update_inode(trans, root, inode);
4316         if (ret && ret == -ENOSPC) {
4317                 /* whoops, lets try again with the full transaction */
4318                 btrfs_end_transaction(trans, root);
4319                 trans = btrfs_start_transaction(root, 1);
4320                 if (IS_ERR(trans)) {
4321                         if (printk_ratelimit()) {
4322                                 printk(KERN_ERR "btrfs: fail to "
4323                                        "dirty  inode %lu error %ld\n",
4324                                        inode->i_ino, PTR_ERR(trans));
4325                         }
4326                         return;
4327                 }
4328                 btrfs_set_trans_block_group(trans, inode);
4329
4330                 ret = btrfs_update_inode(trans, root, inode);
4331                 if (ret) {
4332                         if (printk_ratelimit()) {
4333                                 printk(KERN_ERR "btrfs: fail to "
4334                                        "dirty  inode %lu error %d\n",
4335                                        inode->i_ino, ret);
4336                         }
4337                 }
4338         }
4339         btrfs_end_transaction(trans, root);
4340 }
4341
4342 /*
4343  * find the highest existing sequence number in a directory
4344  * and then set the in-memory index_cnt variable to reflect
4345  * free sequence numbers
4346  */
4347 static int btrfs_set_inode_index_count(struct inode *inode)
4348 {
4349         struct btrfs_root *root = BTRFS_I(inode)->root;
4350         struct btrfs_key key, found_key;
4351         struct btrfs_path *path;
4352         struct extent_buffer *leaf;
4353         int ret;
4354
4355         key.objectid = inode->i_ino;
4356         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4357         key.offset = (u64)-1;
4358
4359         path = btrfs_alloc_path();
4360         if (!path)
4361                 return -ENOMEM;
4362
4363         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4364         if (ret < 0)
4365                 goto out;
4366         /* FIXME: we should be able to handle this */
4367         if (ret == 0)
4368                 goto out;
4369         ret = 0;
4370
4371         /*
4372          * MAGIC NUMBER EXPLANATION:
4373          * since we search a directory based on f_pos we have to start at 2
4374          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4375          * else has to start at 2
4376          */
4377         if (path->slots[0] == 0) {
4378                 BTRFS_I(inode)->index_cnt = 2;
4379                 goto out;
4380         }
4381
4382         path->slots[0]--;
4383
4384         leaf = path->nodes[0];
4385         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4386
4387         if (found_key.objectid != inode->i_ino ||
4388             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4389                 BTRFS_I(inode)->index_cnt = 2;
4390                 goto out;
4391         }
4392
4393         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4394 out:
4395         btrfs_free_path(path);
4396         return ret;
4397 }
4398
4399 /*
4400  * helper to find a free sequence number in a given directory.  This current
4401  * code is very simple, later versions will do smarter things in the btree
4402  */
4403 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4404 {
4405         int ret = 0;
4406
4407         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4408                 ret = btrfs_set_inode_index_count(dir);
4409                 if (ret)
4410                         return ret;
4411         }
4412
4413         *index = BTRFS_I(dir)->index_cnt;
4414         BTRFS_I(dir)->index_cnt++;
4415
4416         return ret;
4417 }
4418
4419 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4420                                      struct btrfs_root *root,
4421                                      struct inode *dir,
4422                                      const char *name, int name_len,
4423                                      u64 ref_objectid, u64 objectid,
4424                                      u64 alloc_hint, int mode, u64 *index)
4425 {
4426         struct inode *inode;
4427         struct btrfs_inode_item *inode_item;
4428         struct btrfs_key *location;
4429         struct btrfs_path *path;
4430         struct btrfs_inode_ref *ref;
4431         struct btrfs_key key[2];
4432         u32 sizes[2];
4433         unsigned long ptr;
4434         int ret;
4435         int owner;
4436
4437         path = btrfs_alloc_path();
4438         BUG_ON(!path);
4439
4440         inode = new_inode(root->fs_info->sb);
4441         if (!inode)
4442                 return ERR_PTR(-ENOMEM);
4443
4444         if (dir) {
4445                 ret = btrfs_set_inode_index(dir, index);
4446                 if (ret) {
4447                         iput(inode);
4448                         return ERR_PTR(ret);
4449                 }
4450         }
4451         /*
4452          * index_cnt is ignored for everything but a dir,
4453          * btrfs_get_inode_index_count has an explanation for the magic
4454          * number
4455          */
4456         BTRFS_I(inode)->index_cnt = 2;
4457         BTRFS_I(inode)->root = root;
4458         BTRFS_I(inode)->generation = trans->transid;
4459         btrfs_set_inode_space_info(root, inode);
4460
4461         if (mode & S_IFDIR)
4462                 owner = 0;
4463         else
4464                 owner = 1;
4465         BTRFS_I(inode)->block_group =
4466                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4467
4468         key[0].objectid = objectid;
4469         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4470         key[0].offset = 0;
4471
4472         key[1].objectid = objectid;
4473         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4474         key[1].offset = ref_objectid;
4475
4476         sizes[0] = sizeof(struct btrfs_inode_item);
4477         sizes[1] = name_len + sizeof(*ref);
4478
4479         path->leave_spinning = 1;
4480         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4481         if (ret != 0)
4482                 goto fail;
4483
4484         inode_init_owner(inode, dir, mode);
4485         inode->i_ino = objectid;
4486         inode_set_bytes(inode, 0);
4487         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4488         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4489                                   struct btrfs_inode_item);
4490         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4491
4492         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4493                              struct btrfs_inode_ref);
4494         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4495         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4496         ptr = (unsigned long)(ref + 1);
4497         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4498
4499         btrfs_mark_buffer_dirty(path->nodes[0]);
4500         btrfs_free_path(path);
4501
4502         location = &BTRFS_I(inode)->location;
4503         location->objectid = objectid;
4504         location->offset = 0;
4505         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4506
4507         btrfs_inherit_iflags(inode, dir);
4508
4509         if ((mode & S_IFREG)) {
4510                 if (btrfs_test_opt(root, NODATASUM))
4511                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4512                 if (btrfs_test_opt(root, NODATACOW))
4513                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4514         }
4515
4516         insert_inode_hash(inode);
4517         inode_tree_add(inode);
4518         return inode;
4519 fail:
4520         if (dir)
4521                 BTRFS_I(dir)->index_cnt--;
4522         btrfs_free_path(path);
4523         iput(inode);
4524         return ERR_PTR(ret);
4525 }
4526
4527 static inline u8 btrfs_inode_type(struct inode *inode)
4528 {
4529         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4530 }
4531
4532 /*
4533  * utility function to add 'inode' into 'parent_inode' with
4534  * a give name and a given sequence number.
4535  * if 'add_backref' is true, also insert a backref from the
4536  * inode to the parent directory.
4537  */
4538 int btrfs_add_link(struct btrfs_trans_handle *trans,
4539                    struct inode *parent_inode, struct inode *inode,
4540                    const char *name, int name_len, int add_backref, u64 index)
4541 {
4542         int ret = 0;
4543         struct btrfs_key key;
4544         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4545
4546         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4547                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4548         } else {
4549                 key.objectid = inode->i_ino;
4550                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4551                 key.offset = 0;
4552         }
4553
4554         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4555                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4556                                          key.objectid, root->root_key.objectid,
4557                                          parent_inode->i_ino,
4558                                          index, name, name_len);
4559         } else if (add_backref) {
4560                 ret = btrfs_insert_inode_ref(trans, root,
4561                                              name, name_len, inode->i_ino,
4562                                              parent_inode->i_ino, index);
4563         }
4564
4565         if (ret == 0) {
4566                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4567                                             parent_inode->i_ino, &key,
4568                                             btrfs_inode_type(inode), index);
4569                 BUG_ON(ret);
4570
4571                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4572                                    name_len * 2);
4573                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4574                 ret = btrfs_update_inode(trans, root, parent_inode);
4575         }
4576         return ret;
4577 }
4578
4579 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4580                             struct dentry *dentry, struct inode *inode,
4581                             int backref, u64 index)
4582 {
4583         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4584                                  inode, dentry->d_name.name,
4585                                  dentry->d_name.len, backref, index);
4586         if (!err) {
4587                 d_instantiate(dentry, inode);
4588                 return 0;
4589         }
4590         if (err > 0)
4591                 err = -EEXIST;
4592         return err;
4593 }
4594
4595 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4596                         int mode, dev_t rdev)
4597 {
4598         struct btrfs_trans_handle *trans;
4599         struct btrfs_root *root = BTRFS_I(dir)->root;
4600         struct inode *inode = NULL;
4601         int err;
4602         int drop_inode = 0;
4603         u64 objectid;
4604         unsigned long nr = 0;
4605         u64 index = 0;
4606
4607         if (!new_valid_dev(rdev))
4608                 return -EINVAL;
4609
4610         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4611         if (err)
4612                 return err;
4613
4614         /*
4615          * 2 for inode item and ref
4616          * 2 for dir items
4617          * 1 for xattr if selinux is on
4618          */
4619         trans = btrfs_start_transaction(root, 5);
4620         if (IS_ERR(trans))
4621                 return PTR_ERR(trans);
4622
4623         btrfs_set_trans_block_group(trans, dir);
4624
4625         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4626                                 dentry->d_name.len,
4627                                 dentry->d_parent->d_inode->i_ino, objectid,
4628                                 BTRFS_I(dir)->block_group, mode, &index);
4629         err = PTR_ERR(inode);
4630         if (IS_ERR(inode))
4631                 goto out_unlock;
4632
4633         err = btrfs_init_inode_security(trans, inode, dir);
4634         if (err) {
4635                 drop_inode = 1;
4636                 goto out_unlock;
4637         }
4638
4639         btrfs_set_trans_block_group(trans, inode);
4640         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4641         if (err)
4642                 drop_inode = 1;
4643         else {
4644                 inode->i_op = &btrfs_special_inode_operations;
4645                 init_special_inode(inode, inode->i_mode, rdev);
4646                 btrfs_update_inode(trans, root, inode);
4647         }
4648         btrfs_update_inode_block_group(trans, inode);
4649         btrfs_update_inode_block_group(trans, dir);
4650 out_unlock:
4651         nr = trans->blocks_used;
4652         btrfs_end_transaction_throttle(trans, root);
4653         btrfs_btree_balance_dirty(root, nr);
4654         if (drop_inode) {
4655                 inode_dec_link_count(inode);
4656                 iput(inode);
4657         }
4658         return err;
4659 }
4660
4661 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4662                         int mode, struct nameidata *nd)
4663 {
4664         struct btrfs_trans_handle *trans;
4665         struct btrfs_root *root = BTRFS_I(dir)->root;
4666         struct inode *inode = NULL;
4667         int drop_inode = 0;
4668         int err;
4669         unsigned long nr = 0;
4670         u64 objectid;
4671         u64 index = 0;
4672
4673         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4674         if (err)
4675                 return err;
4676         /*
4677          * 2 for inode item and ref
4678          * 2 for dir items
4679          * 1 for xattr if selinux is on
4680          */
4681         trans = btrfs_start_transaction(root, 5);
4682         if (IS_ERR(trans))
4683                 return PTR_ERR(trans);
4684
4685         btrfs_set_trans_block_group(trans, dir);
4686
4687         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4688                                 dentry->d_name.len,
4689                                 dentry->d_parent->d_inode->i_ino,
4690                                 objectid, BTRFS_I(dir)->block_group, mode,
4691                                 &index);
4692         err = PTR_ERR(inode);
4693         if (IS_ERR(inode))
4694                 goto out_unlock;
4695
4696         err = btrfs_init_inode_security(trans, inode, dir);
4697         if (err) {
4698                 drop_inode = 1;
4699                 goto out_unlock;
4700         }
4701
4702         btrfs_set_trans_block_group(trans, inode);
4703         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4704         if (err)
4705                 drop_inode = 1;
4706         else {
4707                 inode->i_mapping->a_ops = &btrfs_aops;
4708                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4709                 inode->i_fop = &btrfs_file_operations;
4710                 inode->i_op = &btrfs_file_inode_operations;
4711                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4712         }
4713         btrfs_update_inode_block_group(trans, inode);
4714         btrfs_update_inode_block_group(trans, dir);
4715 out_unlock:
4716         nr = trans->blocks_used;
4717         btrfs_end_transaction_throttle(trans, root);
4718         if (drop_inode) {
4719                 inode_dec_link_count(inode);
4720                 iput(inode);
4721         }
4722         btrfs_btree_balance_dirty(root, nr);
4723         return err;
4724 }
4725
4726 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4727                       struct dentry *dentry)
4728 {
4729         struct btrfs_trans_handle *trans;
4730         struct btrfs_root *root = BTRFS_I(dir)->root;
4731         struct inode *inode = old_dentry->d_inode;
4732         u64 index;
4733         unsigned long nr = 0;
4734         int err;
4735         int drop_inode = 0;
4736
4737         if (inode->i_nlink == 0)
4738                 return -ENOENT;
4739
4740         /* do not allow sys_link's with other subvols of the same device */
4741         if (root->objectid != BTRFS_I(inode)->root->objectid)
4742                 return -EPERM;
4743
4744         btrfs_inc_nlink(inode);
4745
4746         err = btrfs_set_inode_index(dir, &index);
4747         if (err)
4748                 goto fail;
4749
4750         /*
4751          * 1 item for inode ref
4752          * 2 items for dir items
4753          */
4754         trans = btrfs_start_transaction(root, 3);
4755         if (IS_ERR(trans)) {
4756                 err = PTR_ERR(trans);
4757                 goto fail;
4758         }
4759
4760         btrfs_set_trans_block_group(trans, dir);
4761         ihold(inode);
4762
4763         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4764
4765         if (err) {
4766                 drop_inode = 1;
4767         } else {
4768                 btrfs_update_inode_block_group(trans, dir);
4769                 err = btrfs_update_inode(trans, root, inode);
4770                 BUG_ON(err);
4771                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4772         }
4773
4774         nr = trans->blocks_used;
4775         btrfs_end_transaction_throttle(trans, root);
4776 fail:
4777         if (drop_inode) {
4778                 inode_dec_link_count(inode);
4779                 iput(inode);
4780         }
4781         btrfs_btree_balance_dirty(root, nr);
4782         return err;
4783 }
4784
4785 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4786 {
4787         struct inode *inode = NULL;
4788         struct btrfs_trans_handle *trans;
4789         struct btrfs_root *root = BTRFS_I(dir)->root;
4790         int err = 0;
4791         int drop_on_err = 0;
4792         u64 objectid = 0;
4793         u64 index = 0;
4794         unsigned long nr = 1;
4795
4796         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4797         if (err)
4798                 return err;
4799
4800         /*
4801          * 2 items for inode and ref
4802          * 2 items for dir items
4803          * 1 for xattr if selinux is on
4804          */
4805         trans = btrfs_start_transaction(root, 5);
4806         if (IS_ERR(trans))
4807                 return PTR_ERR(trans);
4808         btrfs_set_trans_block_group(trans, dir);
4809
4810         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4811                                 dentry->d_name.len,
4812                                 dentry->d_parent->d_inode->i_ino, objectid,
4813                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4814                                 &index);
4815         if (IS_ERR(inode)) {
4816                 err = PTR_ERR(inode);
4817                 goto out_fail;
4818         }
4819
4820         drop_on_err = 1;
4821
4822         err = btrfs_init_inode_security(trans, inode, dir);
4823         if (err)
4824                 goto out_fail;
4825
4826         inode->i_op = &btrfs_dir_inode_operations;
4827         inode->i_fop = &btrfs_dir_file_operations;
4828         btrfs_set_trans_block_group(trans, inode);
4829
4830         btrfs_i_size_write(inode, 0);
4831         err = btrfs_update_inode(trans, root, inode);
4832         if (err)
4833                 goto out_fail;
4834
4835         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4836                                  inode, dentry->d_name.name,
4837                                  dentry->d_name.len, 0, index);
4838         if (err)
4839                 goto out_fail;
4840
4841         d_instantiate(dentry, inode);
4842         drop_on_err = 0;
4843         btrfs_update_inode_block_group(trans, inode);
4844         btrfs_update_inode_block_group(trans, dir);
4845
4846 out_fail:
4847         nr = trans->blocks_used;
4848         btrfs_end_transaction_throttle(trans, root);
4849         if (drop_on_err)
4850                 iput(inode);
4851         btrfs_btree_balance_dirty(root, nr);
4852         return err;
4853 }
4854
4855 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4856  * and an extent that you want to insert, deal with overlap and insert
4857  * the new extent into the tree.
4858  */
4859 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4860                                 struct extent_map *existing,
4861                                 struct extent_map *em,
4862                                 u64 map_start, u64 map_len)
4863 {
4864         u64 start_diff;
4865
4866         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4867         start_diff = map_start - em->start;
4868         em->start = map_start;
4869         em->len = map_len;
4870         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4871             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4872                 em->block_start += start_diff;
4873                 em->block_len -= start_diff;
4874         }
4875         return add_extent_mapping(em_tree, em);
4876 }
4877
4878 static noinline int uncompress_inline(struct btrfs_path *path,
4879                                       struct inode *inode, struct page *page,
4880                                       size_t pg_offset, u64 extent_offset,
4881                                       struct btrfs_file_extent_item *item)
4882 {
4883         int ret;
4884         struct extent_buffer *leaf = path->nodes[0];
4885         char *tmp;
4886         size_t max_size;
4887         unsigned long inline_size;
4888         unsigned long ptr;
4889
4890         WARN_ON(pg_offset != 0);
4891         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4892         inline_size = btrfs_file_extent_inline_item_len(leaf,
4893                                         btrfs_item_nr(leaf, path->slots[0]));
4894         tmp = kmalloc(inline_size, GFP_NOFS);
4895         ptr = btrfs_file_extent_inline_start(item);
4896
4897         read_extent_buffer(leaf, tmp, ptr, inline_size);
4898
4899         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4900         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4901                                     inline_size, max_size);
4902         if (ret) {
4903                 char *kaddr = kmap_atomic(page, KM_USER0);
4904                 unsigned long copy_size = min_t(u64,
4905                                   PAGE_CACHE_SIZE - pg_offset,
4906                                   max_size - extent_offset);
4907                 memset(kaddr + pg_offset, 0, copy_size);
4908                 kunmap_atomic(kaddr, KM_USER0);
4909         }
4910         kfree(tmp);
4911         return 0;
4912 }
4913
4914 /*
4915  * a bit scary, this does extent mapping from logical file offset to the disk.
4916  * the ugly parts come from merging extents from the disk with the in-ram
4917  * representation.  This gets more complex because of the data=ordered code,
4918  * where the in-ram extents might be locked pending data=ordered completion.
4919  *
4920  * This also copies inline extents directly into the page.
4921  */
4922
4923 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4924                                     size_t pg_offset, u64 start, u64 len,
4925                                     int create)
4926 {
4927         int ret;
4928         int err = 0;
4929         u64 bytenr;
4930         u64 extent_start = 0;
4931         u64 extent_end = 0;
4932         u64 objectid = inode->i_ino;
4933         u32 found_type;
4934         struct btrfs_path *path = NULL;
4935         struct btrfs_root *root = BTRFS_I(inode)->root;
4936         struct btrfs_file_extent_item *item;
4937         struct extent_buffer *leaf;
4938         struct btrfs_key found_key;
4939         struct extent_map *em = NULL;
4940         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4941         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4942         struct btrfs_trans_handle *trans = NULL;
4943         int compressed;
4944
4945 again:
4946         read_lock(&em_tree->lock);
4947         em = lookup_extent_mapping(em_tree, start, len);
4948         if (em)
4949                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4950         read_unlock(&em_tree->lock);
4951
4952         if (em) {
4953                 if (em->start > start || em->start + em->len <= start)
4954                         free_extent_map(em);
4955                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4956                         free_extent_map(em);
4957                 else
4958                         goto out;
4959         }
4960         em = alloc_extent_map(GFP_NOFS);
4961         if (!em) {
4962                 err = -ENOMEM;
4963                 goto out;
4964         }
4965         em->bdev = root->fs_info->fs_devices->latest_bdev;
4966         em->start = EXTENT_MAP_HOLE;
4967         em->orig_start = EXTENT_MAP_HOLE;
4968         em->len = (u64)-1;
4969         em->block_len = (u64)-1;
4970
4971         if (!path) {
4972                 path = btrfs_alloc_path();
4973                 BUG_ON(!path);
4974         }
4975
4976         ret = btrfs_lookup_file_extent(trans, root, path,
4977                                        objectid, start, trans != NULL);
4978         if (ret < 0) {
4979                 err = ret;
4980                 goto out;
4981         }
4982
4983         if (ret != 0) {
4984                 if (path->slots[0] == 0)
4985                         goto not_found;
4986                 path->slots[0]--;
4987         }
4988
4989         leaf = path->nodes[0];
4990         item = btrfs_item_ptr(leaf, path->slots[0],
4991                               struct btrfs_file_extent_item);
4992         /* are we inside the extent that was found? */
4993         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4994         found_type = btrfs_key_type(&found_key);
4995         if (found_key.objectid != objectid ||
4996             found_type != BTRFS_EXTENT_DATA_KEY) {
4997                 goto not_found;
4998         }
4999
5000         found_type = btrfs_file_extent_type(leaf, item);
5001         extent_start = found_key.offset;
5002         compressed = btrfs_file_extent_compression(leaf, item);
5003         if (found_type == BTRFS_FILE_EXTENT_REG ||
5004             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5005                 extent_end = extent_start +
5006                        btrfs_file_extent_num_bytes(leaf, item);
5007         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5008                 size_t size;
5009                 size = btrfs_file_extent_inline_len(leaf, item);
5010                 extent_end = (extent_start + size + root->sectorsize - 1) &
5011                         ~((u64)root->sectorsize - 1);
5012         }
5013
5014         if (start >= extent_end) {
5015                 path->slots[0]++;
5016                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5017                         ret = btrfs_next_leaf(root, path);
5018                         if (ret < 0) {
5019                                 err = ret;
5020                                 goto out;
5021                         }
5022                         if (ret > 0)
5023                                 goto not_found;
5024                         leaf = path->nodes[0];
5025                 }
5026                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5027                 if (found_key.objectid != objectid ||
5028                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5029                         goto not_found;
5030                 if (start + len <= found_key.offset)
5031                         goto not_found;
5032                 em->start = start;
5033                 em->len = found_key.offset - start;
5034                 goto not_found_em;
5035         }
5036
5037         if (found_type == BTRFS_FILE_EXTENT_REG ||
5038             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5039                 em->start = extent_start;
5040                 em->len = extent_end - extent_start;
5041                 em->orig_start = extent_start -
5042                                  btrfs_file_extent_offset(leaf, item);
5043                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5044                 if (bytenr == 0) {
5045                         em->block_start = EXTENT_MAP_HOLE;
5046                         goto insert;
5047                 }
5048                 if (compressed) {
5049                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5050                         em->block_start = bytenr;
5051                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5052                                                                          item);
5053                 } else {
5054                         bytenr += btrfs_file_extent_offset(leaf, item);
5055                         em->block_start = bytenr;
5056                         em->block_len = em->len;
5057                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5058                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5059                 }
5060                 goto insert;
5061         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5062                 unsigned long ptr;
5063                 char *map;
5064                 size_t size;
5065                 size_t extent_offset;
5066                 size_t copy_size;
5067
5068                 em->block_start = EXTENT_MAP_INLINE;
5069                 if (!page || create) {
5070                         em->start = extent_start;
5071                         em->len = extent_end - extent_start;
5072                         goto out;
5073                 }
5074
5075                 size = btrfs_file_extent_inline_len(leaf, item);
5076                 extent_offset = page_offset(page) + pg_offset - extent_start;
5077                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5078                                 size - extent_offset);
5079                 em->start = extent_start + extent_offset;
5080                 em->len = (copy_size + root->sectorsize - 1) &
5081                         ~((u64)root->sectorsize - 1);
5082                 em->orig_start = EXTENT_MAP_INLINE;
5083                 if (compressed)
5084                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5085                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5086                 if (create == 0 && !PageUptodate(page)) {
5087                         if (btrfs_file_extent_compression(leaf, item) ==
5088                             BTRFS_COMPRESS_ZLIB) {
5089                                 ret = uncompress_inline(path, inode, page,
5090                                                         pg_offset,
5091                                                         extent_offset, item);
5092                                 BUG_ON(ret);
5093                         } else {
5094                                 map = kmap(page);
5095                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5096                                                    copy_size);
5097                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5098                                         memset(map + pg_offset + copy_size, 0,
5099                                                PAGE_CACHE_SIZE - pg_offset -
5100                                                copy_size);
5101                                 }
5102                                 kunmap(page);
5103                         }
5104                         flush_dcache_page(page);
5105                 } else if (create && PageUptodate(page)) {
5106                         WARN_ON(1);
5107                         if (!trans) {
5108                                 kunmap(page);
5109                                 free_extent_map(em);
5110                                 em = NULL;
5111                                 btrfs_release_path(root, path);
5112                                 trans = btrfs_join_transaction(root, 1);
5113                                 goto again;
5114                         }
5115                         map = kmap(page);
5116                         write_extent_buffer(leaf, map + pg_offset, ptr,
5117                                             copy_size);
5118                         kunmap(page);
5119                         btrfs_mark_buffer_dirty(leaf);
5120                 }
5121                 set_extent_uptodate(io_tree, em->start,
5122                                     extent_map_end(em) - 1, GFP_NOFS);
5123                 goto insert;
5124         } else {
5125                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5126                 WARN_ON(1);
5127         }
5128 not_found:
5129         em->start = start;
5130         em->len = len;
5131 not_found_em:
5132         em->block_start = EXTENT_MAP_HOLE;
5133         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5134 insert:
5135         btrfs_release_path(root, path);
5136         if (em->start > start || extent_map_end(em) <= start) {
5137                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5138                        "[%llu %llu]\n", (unsigned long long)em->start,
5139                        (unsigned long long)em->len,
5140                        (unsigned long long)start,
5141                        (unsigned long long)len);
5142                 err = -EIO;
5143                 goto out;
5144         }
5145
5146         err = 0;
5147         write_lock(&em_tree->lock);
5148         ret = add_extent_mapping(em_tree, em);
5149         /* it is possible that someone inserted the extent into the tree
5150          * while we had the lock dropped.  It is also possible that
5151          * an overlapping map exists in the tree
5152          */
5153         if (ret == -EEXIST) {
5154                 struct extent_map *existing;
5155
5156                 ret = 0;
5157
5158                 existing = lookup_extent_mapping(em_tree, start, len);
5159                 if (existing && (existing->start > start ||
5160                     existing->start + existing->len <= start)) {
5161                         free_extent_map(existing);
5162                         existing = NULL;
5163                 }
5164                 if (!existing) {
5165                         existing = lookup_extent_mapping(em_tree, em->start,
5166                                                          em->len);
5167                         if (existing) {
5168                                 err = merge_extent_mapping(em_tree, existing,
5169                                                            em, start,
5170                                                            root->sectorsize);
5171                                 free_extent_map(existing);
5172                                 if (err) {
5173                                         free_extent_map(em);
5174                                         em = NULL;
5175                                 }
5176                         } else {
5177                                 err = -EIO;
5178                                 free_extent_map(em);
5179                                 em = NULL;
5180                         }
5181                 } else {
5182                         free_extent_map(em);
5183                         em = existing;
5184                         err = 0;
5185                 }
5186         }
5187         write_unlock(&em_tree->lock);
5188 out:
5189         if (path)
5190                 btrfs_free_path(path);
5191         if (trans) {
5192                 ret = btrfs_end_transaction(trans, root);
5193                 if (!err)
5194                         err = ret;
5195         }
5196         if (err) {
5197                 free_extent_map(em);
5198                 return ERR_PTR(err);
5199         }
5200         return em;
5201 }
5202
5203 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5204                                                   u64 start, u64 len)
5205 {
5206         struct btrfs_root *root = BTRFS_I(inode)->root;
5207         struct btrfs_trans_handle *trans;
5208         struct extent_map *em;
5209         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5210         struct btrfs_key ins;
5211         u64 alloc_hint;
5212         int ret;
5213
5214         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5215
5216         trans = btrfs_join_transaction(root, 0);
5217         if (!trans)
5218                 return ERR_PTR(-ENOMEM);
5219
5220         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5221
5222         alloc_hint = get_extent_allocation_hint(inode, start, len);
5223         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5224                                    alloc_hint, (u64)-1, &ins, 1);
5225         if (ret) {
5226                 em = ERR_PTR(ret);
5227                 goto out;
5228         }
5229
5230         em = alloc_extent_map(GFP_NOFS);
5231         if (!em) {
5232                 em = ERR_PTR(-ENOMEM);
5233                 goto out;
5234         }
5235
5236         em->start = start;
5237         em->orig_start = em->start;
5238         em->len = ins.offset;
5239
5240         em->block_start = ins.objectid;
5241         em->block_len = ins.offset;
5242         em->bdev = root->fs_info->fs_devices->latest_bdev;
5243         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5244
5245         while (1) {
5246                 write_lock(&em_tree->lock);
5247                 ret = add_extent_mapping(em_tree, em);
5248                 write_unlock(&em_tree->lock);
5249                 if (ret != -EEXIST)
5250                         break;
5251                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5252         }
5253
5254         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5255                                            ins.offset, ins.offset, 0);
5256         if (ret) {
5257                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5258                 em = ERR_PTR(ret);
5259         }
5260 out:
5261         btrfs_end_transaction(trans, root);
5262         return em;
5263 }
5264
5265 /*
5266  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5267  * block must be cow'd
5268  */
5269 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5270                                       struct inode *inode, u64 offset, u64 len)
5271 {
5272         struct btrfs_path *path;
5273         int ret;
5274         struct extent_buffer *leaf;
5275         struct btrfs_root *root = BTRFS_I(inode)->root;
5276         struct btrfs_file_extent_item *fi;
5277         struct btrfs_key key;
5278         u64 disk_bytenr;
5279         u64 backref_offset;
5280         u64 extent_end;
5281         u64 num_bytes;
5282         int slot;
5283         int found_type;
5284
5285         path = btrfs_alloc_path();
5286         if (!path)
5287                 return -ENOMEM;
5288
5289         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5290                                        offset, 0);
5291         if (ret < 0)
5292                 goto out;
5293
5294         slot = path->slots[0];
5295         if (ret == 1) {
5296                 if (slot == 0) {
5297                         /* can't find the item, must cow */
5298                         ret = 0;
5299                         goto out;
5300                 }
5301                 slot--;
5302         }
5303         ret = 0;
5304         leaf = path->nodes[0];
5305         btrfs_item_key_to_cpu(leaf, &key, slot);
5306         if (key.objectid != inode->i_ino ||
5307             key.type != BTRFS_EXTENT_DATA_KEY) {
5308                 /* not our file or wrong item type, must cow */
5309                 goto out;
5310         }
5311
5312         if (key.offset > offset) {
5313                 /* Wrong offset, must cow */
5314                 goto out;
5315         }
5316
5317         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5318         found_type = btrfs_file_extent_type(leaf, fi);
5319         if (found_type != BTRFS_FILE_EXTENT_REG &&
5320             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5321                 /* not a regular extent, must cow */
5322                 goto out;
5323         }
5324         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5325         backref_offset = btrfs_file_extent_offset(leaf, fi);
5326
5327         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5328         if (extent_end < offset + len) {
5329                 /* extent doesn't include our full range, must cow */
5330                 goto out;
5331         }
5332
5333         if (btrfs_extent_readonly(root, disk_bytenr))
5334                 goto out;
5335
5336         /*
5337          * look for other files referencing this extent, if we
5338          * find any we must cow
5339          */
5340         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5341                                   key.offset - backref_offset, disk_bytenr))
5342                 goto out;
5343
5344         /*
5345          * adjust disk_bytenr and num_bytes to cover just the bytes
5346          * in this extent we are about to write.  If there
5347          * are any csums in that range we have to cow in order
5348          * to keep the csums correct
5349          */
5350         disk_bytenr += backref_offset;
5351         disk_bytenr += offset - key.offset;
5352         num_bytes = min(offset + len, extent_end) - offset;
5353         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5354                                 goto out;
5355         /*
5356          * all of the above have passed, it is safe to overwrite this extent
5357          * without cow
5358          */
5359         ret = 1;
5360 out:
5361         btrfs_free_path(path);
5362         return ret;
5363 }
5364
5365 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5366                                    struct buffer_head *bh_result, int create)
5367 {
5368         struct extent_map *em;
5369         struct btrfs_root *root = BTRFS_I(inode)->root;
5370         u64 start = iblock << inode->i_blkbits;
5371         u64 len = bh_result->b_size;
5372         struct btrfs_trans_handle *trans;
5373
5374         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5375         if (IS_ERR(em))
5376                 return PTR_ERR(em);
5377
5378         /*
5379          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5380          * io.  INLINE is special, and we could probably kludge it in here, but
5381          * it's still buffered so for safety lets just fall back to the generic
5382          * buffered path.
5383          *
5384          * For COMPRESSED we _have_ to read the entire extent in so we can
5385          * decompress it, so there will be buffering required no matter what we
5386          * do, so go ahead and fallback to buffered.
5387          *
5388          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5389          * to buffered IO.  Don't blame me, this is the price we pay for using
5390          * the generic code.
5391          */
5392         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5393             em->block_start == EXTENT_MAP_INLINE) {
5394                 free_extent_map(em);
5395                 return -ENOTBLK;
5396         }
5397
5398         /* Just a good old fashioned hole, return */
5399         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5400                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5401                 free_extent_map(em);
5402                 /* DIO will do one hole at a time, so just unlock a sector */
5403                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5404                               start + root->sectorsize - 1, GFP_NOFS);
5405                 return 0;
5406         }
5407
5408         /*
5409          * We don't allocate a new extent in the following cases
5410          *
5411          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5412          * existing extent.
5413          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5414          * just use the extent.
5415          *
5416          */
5417         if (!create) {
5418                 len = em->len - (start - em->start);
5419                 goto map;
5420         }
5421
5422         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5423             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5424              em->block_start != EXTENT_MAP_HOLE)) {
5425                 int type;
5426                 int ret;
5427                 u64 block_start;
5428
5429                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5430                         type = BTRFS_ORDERED_PREALLOC;
5431                 else
5432                         type = BTRFS_ORDERED_NOCOW;
5433                 len = min(len, em->len - (start - em->start));
5434                 block_start = em->block_start + (start - em->start);
5435
5436                 /*
5437                  * we're not going to log anything, but we do need
5438                  * to make sure the current transaction stays open
5439                  * while we look for nocow cross refs
5440                  */
5441                 trans = btrfs_join_transaction(root, 0);
5442                 if (!trans)
5443                         goto must_cow;
5444
5445                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5446                         ret = btrfs_add_ordered_extent_dio(inode, start,
5447                                            block_start, len, len, type);
5448                         btrfs_end_transaction(trans, root);
5449                         if (ret) {
5450                                 free_extent_map(em);
5451                                 return ret;
5452                         }
5453                         goto unlock;
5454                 }
5455                 btrfs_end_transaction(trans, root);
5456         }
5457 must_cow:
5458         /*
5459          * this will cow the extent, reset the len in case we changed
5460          * it above
5461          */
5462         len = bh_result->b_size;
5463         free_extent_map(em);
5464         em = btrfs_new_extent_direct(inode, start, len);
5465         if (IS_ERR(em))
5466                 return PTR_ERR(em);
5467         len = min(len, em->len - (start - em->start));
5468 unlock:
5469         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5470                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5471                           0, NULL, GFP_NOFS);
5472 map:
5473         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5474                 inode->i_blkbits;
5475         bh_result->b_size = len;
5476         bh_result->b_bdev = em->bdev;
5477         set_buffer_mapped(bh_result);
5478         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5479                 set_buffer_new(bh_result);
5480
5481         free_extent_map(em);
5482
5483         return 0;
5484 }
5485
5486 struct btrfs_dio_private {
5487         struct inode *inode;
5488         u64 logical_offset;
5489         u64 disk_bytenr;
5490         u64 bytes;
5491         u32 *csums;
5492         void *private;
5493 };
5494
5495 static void btrfs_endio_direct_read(struct bio *bio, int err)
5496 {
5497         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5498         struct bio_vec *bvec = bio->bi_io_vec;
5499         struct btrfs_dio_private *dip = bio->bi_private;
5500         struct inode *inode = dip->inode;
5501         struct btrfs_root *root = BTRFS_I(inode)->root;
5502         u64 start;
5503         u32 *private = dip->csums;
5504
5505         start = dip->logical_offset;
5506         do {
5507                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5508                         struct page *page = bvec->bv_page;
5509                         char *kaddr;
5510                         u32 csum = ~(u32)0;
5511                         unsigned long flags;
5512
5513                         local_irq_save(flags);
5514                         kaddr = kmap_atomic(page, KM_IRQ0);
5515                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5516                                                csum, bvec->bv_len);
5517                         btrfs_csum_final(csum, (char *)&csum);
5518                         kunmap_atomic(kaddr, KM_IRQ0);
5519                         local_irq_restore(flags);
5520
5521                         flush_dcache_page(bvec->bv_page);
5522                         if (csum != *private) {
5523                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5524                                       " %llu csum %u private %u\n",
5525                                       inode->i_ino, (unsigned long long)start,
5526                                       csum, *private);
5527                                 err = -EIO;
5528                         }
5529                 }
5530
5531                 start += bvec->bv_len;
5532                 private++;
5533                 bvec++;
5534         } while (bvec <= bvec_end);
5535
5536         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5537                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5538         bio->bi_private = dip->private;
5539
5540         kfree(dip->csums);
5541         kfree(dip);
5542         dio_end_io(bio, err);
5543 }
5544
5545 static void btrfs_endio_direct_write(struct bio *bio, int err)
5546 {
5547         struct btrfs_dio_private *dip = bio->bi_private;
5548         struct inode *inode = dip->inode;
5549         struct btrfs_root *root = BTRFS_I(inode)->root;
5550         struct btrfs_trans_handle *trans;
5551         struct btrfs_ordered_extent *ordered = NULL;
5552         struct extent_state *cached_state = NULL;
5553         int ret;
5554
5555         if (err)
5556                 goto out_done;
5557
5558         ret = btrfs_dec_test_ordered_pending(inode, &ordered,
5559                                              dip->logical_offset, dip->bytes);
5560         if (!ret)
5561                 goto out_done;
5562
5563         BUG_ON(!ordered);
5564
5565         trans = btrfs_join_transaction(root, 1);
5566         if (!trans) {
5567                 err = -ENOMEM;
5568                 goto out;
5569         }
5570         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5571
5572         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5573                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5574                 if (!ret)
5575                         ret = btrfs_update_inode(trans, root, inode);
5576                 err = ret;
5577                 goto out;
5578         }
5579
5580         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5581                          ordered->file_offset + ordered->len - 1, 0,
5582                          &cached_state, GFP_NOFS);
5583
5584         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5585                 ret = btrfs_mark_extent_written(trans, inode,
5586                                                 ordered->file_offset,
5587                                                 ordered->file_offset +
5588                                                 ordered->len);
5589                 if (ret) {
5590                         err = ret;
5591                         goto out_unlock;
5592                 }
5593         } else {
5594                 ret = insert_reserved_file_extent(trans, inode,
5595                                                   ordered->file_offset,
5596                                                   ordered->start,
5597                                                   ordered->disk_len,
5598                                                   ordered->len,
5599                                                   ordered->len,
5600                                                   0, 0, 0,
5601                                                   BTRFS_FILE_EXTENT_REG);
5602                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5603                                    ordered->file_offset, ordered->len);
5604                 if (ret) {
5605                         err = ret;
5606                         WARN_ON(1);
5607                         goto out_unlock;
5608                 }
5609         }
5610
5611         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5612         btrfs_ordered_update_i_size(inode, 0, ordered);
5613         btrfs_update_inode(trans, root, inode);
5614 out_unlock:
5615         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5616                              ordered->file_offset + ordered->len - 1,
5617                              &cached_state, GFP_NOFS);
5618 out:
5619         btrfs_delalloc_release_metadata(inode, ordered->len);
5620         btrfs_end_transaction(trans, root);
5621         btrfs_put_ordered_extent(ordered);
5622         btrfs_put_ordered_extent(ordered);
5623 out_done:
5624         bio->bi_private = dip->private;
5625
5626         kfree(dip->csums);
5627         kfree(dip);
5628         dio_end_io(bio, err);
5629 }
5630
5631 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5632                                     struct bio *bio, int mirror_num,
5633                                     unsigned long bio_flags, u64 offset)
5634 {
5635         int ret;
5636         struct btrfs_root *root = BTRFS_I(inode)->root;
5637         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5638         BUG_ON(ret);
5639         return 0;
5640 }
5641
5642 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5643                                 loff_t file_offset)
5644 {
5645         struct btrfs_root *root = BTRFS_I(inode)->root;
5646         struct btrfs_dio_private *dip;
5647         struct bio_vec *bvec = bio->bi_io_vec;
5648         u64 start;
5649         int skip_sum;
5650         int write = rw & REQ_WRITE;
5651         int ret = 0;
5652
5653         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5654
5655         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5656         if (!dip) {
5657                 ret = -ENOMEM;
5658                 goto free_ordered;
5659         }
5660         dip->csums = NULL;
5661
5662         if (!skip_sum) {
5663                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5664                 if (!dip->csums) {
5665                         ret = -ENOMEM;
5666                         goto free_ordered;
5667                 }
5668         }
5669
5670         dip->private = bio->bi_private;
5671         dip->inode = inode;
5672         dip->logical_offset = file_offset;
5673