Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-drm-fsl-dcu.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 <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include "compat.h"
43 #include "ctree.h"
44 #include "disk-io.h"
45 #include "transaction.h"
46 #include "btrfs_inode.h"
47 #include "ioctl.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
50 #include "xattr.h"
51 #include "tree-log.h"
52 #include "volumes.h"
53 #include "compression.h"
54 #include "locking.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
57
58 struct btrfs_iget_args {
59         u64 ino;
60         struct btrfs_root *root;
61 };
62
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
72
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78
79 #define S_SHIFT 12
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
88 };
89
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 static noinline int cow_file_range(struct inode *inode,
94                                    struct page *locked_page,
95                                    u64 start, u64 end, int *page_started,
96                                    unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98                                 struct btrfs_root *root, struct inode *inode);
99
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101                                      struct inode *inode,  struct inode *dir,
102                                      const struct qstr *qstr)
103 {
104         int err;
105
106         err = btrfs_init_acl(trans, inode, dir);
107         if (!err)
108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
109         return err;
110 }
111
112 /*
113  * this does all the hard work for inserting an inline extent into
114  * the btree.  The caller should have done a btrfs_drop_extents so that
115  * no overlapping inline items exist in the btree
116  */
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118                                 struct btrfs_root *root, struct inode *inode,
119                                 u64 start, size_t size, size_t compressed_size,
120                                 int compress_type,
121                                 struct page **compressed_pages)
122 {
123         struct btrfs_key key;
124         struct btrfs_path *path;
125         struct extent_buffer *leaf;
126         struct page *page = NULL;
127         char *kaddr;
128         unsigned long ptr;
129         struct btrfs_file_extent_item *ei;
130         int err = 0;
131         int ret;
132         size_t cur_size = size;
133         size_t datasize;
134         unsigned long offset;
135
136         if (compressed_size && compressed_pages)
137                 cur_size = compressed_size;
138
139         path = btrfs_alloc_path();
140         if (!path)
141                 return -ENOMEM;
142
143         path->leave_spinning = 1;
144
145         key.objectid = btrfs_ino(inode);
146         key.offset = start;
147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
149
150         inode_add_bytes(inode, size);
151         ret = btrfs_insert_empty_item(trans, root, path, &key,
152                                       datasize);
153         if (ret) {
154                 err = ret;
155                 goto fail;
156         }
157         leaf = path->nodes[0];
158         ei = btrfs_item_ptr(leaf, path->slots[0],
159                             struct btrfs_file_extent_item);
160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162         btrfs_set_file_extent_encryption(leaf, ei, 0);
163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165         ptr = btrfs_file_extent_inline_start(ei);
166
167         if (compress_type != BTRFS_COMPRESS_NONE) {
168                 struct page *cpage;
169                 int i = 0;
170                 while (compressed_size > 0) {
171                         cpage = compressed_pages[i];
172                         cur_size = min_t(unsigned long, compressed_size,
173                                        PAGE_CACHE_SIZE);
174
175                         kaddr = kmap_atomic(cpage);
176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
177                         kunmap_atomic(kaddr);
178
179                         i++;
180                         ptr += cur_size;
181                         compressed_size -= cur_size;
182                 }
183                 btrfs_set_file_extent_compression(leaf, ei,
184                                                   compress_type);
185         } else {
186                 page = find_get_page(inode->i_mapping,
187                                      start >> PAGE_CACHE_SHIFT);
188                 btrfs_set_file_extent_compression(leaf, ei, 0);
189                 kaddr = kmap_atomic(page);
190                 offset = start & (PAGE_CACHE_SIZE - 1);
191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192                 kunmap_atomic(kaddr);
193                 page_cache_release(page);
194         }
195         btrfs_mark_buffer_dirty(leaf);
196         btrfs_free_path(path);
197
198         /*
199          * we're an inline extent, so nobody can
200          * extend the file past i_size without locking
201          * a page we already have locked.
202          *
203          * We must do any isize and inode updates
204          * before we unlock the pages.  Otherwise we
205          * could end up racing with unlink.
206          */
207         BTRFS_I(inode)->disk_i_size = inode->i_size;
208         ret = btrfs_update_inode(trans, root, inode);
209
210         return ret;
211 fail:
212         btrfs_free_path(path);
213         return err;
214 }
215
216
217 /*
218  * conditionally insert an inline extent into the file.  This
219  * does the checks required to make sure the data is small enough
220  * to fit as an inline extent.
221  */
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223                                  struct btrfs_root *root,
224                                  struct inode *inode, u64 start, u64 end,
225                                  size_t compressed_size, int compress_type,
226                                  struct page **compressed_pages)
227 {
228         u64 isize = i_size_read(inode);
229         u64 actual_end = min(end + 1, isize);
230         u64 inline_len = actual_end - start;
231         u64 aligned_end = (end + root->sectorsize - 1) &
232                         ~((u64)root->sectorsize - 1);
233         u64 hint_byte;
234         u64 data_len = inline_len;
235         int ret;
236
237         if (compressed_size)
238                 data_len = compressed_size;
239
240         if (start > 0 ||
241             actual_end >= PAGE_CACHE_SIZE ||
242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243             (!compressed_size &&
244             (actual_end & (root->sectorsize - 1)) == 0) ||
245             end + 1 < isize ||
246             data_len > root->fs_info->max_inline) {
247                 return 1;
248         }
249
250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
251                                  &hint_byte, 1);
252         if (ret)
253                 return ret;
254
255         if (isize > actual_end)
256                 inline_len = min_t(u64, isize, actual_end);
257         ret = insert_inline_extent(trans, root, inode, start,
258                                    inline_len, compressed_size,
259                                    compress_type, compressed_pages);
260         if (ret && ret != -ENOSPC) {
261                 btrfs_abort_transaction(trans, root, ret);
262                 return ret;
263         } else if (ret == -ENOSPC) {
264                 return 1;
265         }
266
267         btrfs_delalloc_release_metadata(inode, end + 1 - start);
268         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
269         return 0;
270 }
271
272 struct async_extent {
273         u64 start;
274         u64 ram_size;
275         u64 compressed_size;
276         struct page **pages;
277         unsigned long nr_pages;
278         int compress_type;
279         struct list_head list;
280 };
281
282 struct async_cow {
283         struct inode *inode;
284         struct btrfs_root *root;
285         struct page *locked_page;
286         u64 start;
287         u64 end;
288         struct list_head extents;
289         struct btrfs_work work;
290 };
291
292 static noinline int add_async_extent(struct async_cow *cow,
293                                      u64 start, u64 ram_size,
294                                      u64 compressed_size,
295                                      struct page **pages,
296                                      unsigned long nr_pages,
297                                      int compress_type)
298 {
299         struct async_extent *async_extent;
300
301         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
302         BUG_ON(!async_extent); /* -ENOMEM */
303         async_extent->start = start;
304         async_extent->ram_size = ram_size;
305         async_extent->compressed_size = compressed_size;
306         async_extent->pages = pages;
307         async_extent->nr_pages = nr_pages;
308         async_extent->compress_type = compress_type;
309         list_add_tail(&async_extent->list, &cow->extents);
310         return 0;
311 }
312
313 /*
314  * we create compressed extents in two phases.  The first
315  * phase compresses a range of pages that have already been
316  * locked (both pages and state bits are locked).
317  *
318  * This is done inside an ordered work queue, and the compression
319  * is spread across many cpus.  The actual IO submission is step
320  * two, and the ordered work queue takes care of making sure that
321  * happens in the same order things were put onto the queue by
322  * writepages and friends.
323  *
324  * If this code finds it can't get good compression, it puts an
325  * entry onto the work queue to write the uncompressed bytes.  This
326  * makes sure that both compressed inodes and uncompressed inodes
327  * are written in the same order that pdflush sent them down.
328  */
329 static noinline int compress_file_range(struct inode *inode,
330                                         struct page *locked_page,
331                                         u64 start, u64 end,
332                                         struct async_cow *async_cow,
333                                         int *num_added)
334 {
335         struct btrfs_root *root = BTRFS_I(inode)->root;
336         struct btrfs_trans_handle *trans;
337         u64 num_bytes;
338         u64 blocksize = root->sectorsize;
339         u64 actual_end;
340         u64 isize = i_size_read(inode);
341         int ret = 0;
342         struct page **pages = NULL;
343         unsigned long nr_pages;
344         unsigned long nr_pages_ret = 0;
345         unsigned long total_compressed = 0;
346         unsigned long total_in = 0;
347         unsigned long max_compressed = 128 * 1024;
348         unsigned long max_uncompressed = 128 * 1024;
349         int i;
350         int will_compress;
351         int compress_type = root->fs_info->compress_type;
352
353         /* if this is a small write inside eof, kick off a defrag */
354         if ((end - start + 1) < 16 * 1024 &&
355             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
356                 btrfs_add_inode_defrag(NULL, inode);
357
358         actual_end = min_t(u64, isize, end + 1);
359 again:
360         will_compress = 0;
361         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
362         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
363
364         /*
365          * we don't want to send crud past the end of i_size through
366          * compression, that's just a waste of CPU time.  So, if the
367          * end of the file is before the start of our current
368          * requested range of bytes, we bail out to the uncompressed
369          * cleanup code that can deal with all of this.
370          *
371          * It isn't really the fastest way to fix things, but this is a
372          * very uncommon corner.
373          */
374         if (actual_end <= start)
375                 goto cleanup_and_bail_uncompressed;
376
377         total_compressed = actual_end - start;
378
379         /* we want to make sure that amount of ram required to uncompress
380          * an extent is reasonable, so we limit the total size in ram
381          * of a compressed extent to 128k.  This is a crucial number
382          * because it also controls how easily we can spread reads across
383          * cpus for decompression.
384          *
385          * We also want to make sure the amount of IO required to do
386          * a random read is reasonably small, so we limit the size of
387          * a compressed extent to 128k.
388          */
389         total_compressed = min(total_compressed, max_uncompressed);
390         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
391         num_bytes = max(blocksize,  num_bytes);
392         total_in = 0;
393         ret = 0;
394
395         /*
396          * we do compression for mount -o compress and when the
397          * inode has not been flagged as nocompress.  This flag can
398          * change at any time if we discover bad compression ratios.
399          */
400         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
401             (btrfs_test_opt(root, COMPRESS) ||
402              (BTRFS_I(inode)->force_compress) ||
403              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
404                 WARN_ON(pages);
405                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
406                 if (!pages) {
407                         /* just bail out to the uncompressed code */
408                         goto cont;
409                 }
410
411                 if (BTRFS_I(inode)->force_compress)
412                         compress_type = BTRFS_I(inode)->force_compress;
413
414                 ret = btrfs_compress_pages(compress_type,
415                                            inode->i_mapping, start,
416                                            total_compressed, pages,
417                                            nr_pages, &nr_pages_ret,
418                                            &total_in,
419                                            &total_compressed,
420                                            max_compressed);
421
422                 if (!ret) {
423                         unsigned long offset = total_compressed &
424                                 (PAGE_CACHE_SIZE - 1);
425                         struct page *page = pages[nr_pages_ret - 1];
426                         char *kaddr;
427
428                         /* zero the tail end of the last page, we might be
429                          * sending it down to disk
430                          */
431                         if (offset) {
432                                 kaddr = kmap_atomic(page);
433                                 memset(kaddr + offset, 0,
434                                        PAGE_CACHE_SIZE - offset);
435                                 kunmap_atomic(kaddr);
436                         }
437                         will_compress = 1;
438                 }
439         }
440 cont:
441         if (start == 0) {
442                 trans = btrfs_join_transaction(root);
443                 if (IS_ERR(trans)) {
444                         ret = PTR_ERR(trans);
445                         trans = NULL;
446                         goto cleanup_and_out;
447                 }
448                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
449
450                 /* lets try to make an inline extent */
451                 if (ret || total_in < (actual_end - start)) {
452                         /* we didn't compress the entire range, try
453                          * to make an uncompressed inline extent.
454                          */
455                         ret = cow_file_range_inline(trans, root, inode,
456                                                     start, end, 0, 0, NULL);
457                 } else {
458                         /* try making a compressed inline extent */
459                         ret = cow_file_range_inline(trans, root, inode,
460                                                     start, end,
461                                                     total_compressed,
462                                                     compress_type, pages);
463                 }
464                 if (ret <= 0) {
465                         /*
466                          * inline extent creation worked or returned error,
467                          * we don't need to create any more async work items.
468                          * Unlock and free up our temp pages.
469                          */
470                         extent_clear_unlock_delalloc(inode,
471                              &BTRFS_I(inode)->io_tree,
472                              start, end, NULL,
473                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
474                              EXTENT_CLEAR_DELALLOC |
475                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
476
477                         btrfs_end_transaction(trans, root);
478                         goto free_pages_out;
479                 }
480                 btrfs_end_transaction(trans, root);
481         }
482
483         if (will_compress) {
484                 /*
485                  * we aren't doing an inline extent round the compressed size
486                  * up to a block size boundary so the allocator does sane
487                  * things
488                  */
489                 total_compressed = (total_compressed + blocksize - 1) &
490                         ~(blocksize - 1);
491
492                 /*
493                  * one last check to make sure the compression is really a
494                  * win, compare the page count read with the blocks on disk
495                  */
496                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
497                         ~(PAGE_CACHE_SIZE - 1);
498                 if (total_compressed >= total_in) {
499                         will_compress = 0;
500                 } else {
501                         num_bytes = total_in;
502                 }
503         }
504         if (!will_compress && pages) {
505                 /*
506                  * the compression code ran but failed to make things smaller,
507                  * free any pages it allocated and our page pointer array
508                  */
509                 for (i = 0; i < nr_pages_ret; i++) {
510                         WARN_ON(pages[i]->mapping);
511                         page_cache_release(pages[i]);
512                 }
513                 kfree(pages);
514                 pages = NULL;
515                 total_compressed = 0;
516                 nr_pages_ret = 0;
517
518                 /* flag the file so we don't compress in the future */
519                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
520                     !(BTRFS_I(inode)->force_compress)) {
521                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
522                 }
523         }
524         if (will_compress) {
525                 *num_added += 1;
526
527                 /* the async work queues will take care of doing actual
528                  * allocation on disk for these compressed pages,
529                  * and will submit them to the elevator.
530                  */
531                 add_async_extent(async_cow, start, num_bytes,
532                                  total_compressed, pages, nr_pages_ret,
533                                  compress_type);
534
535                 if (start + num_bytes < end) {
536                         start += num_bytes;
537                         pages = NULL;
538                         cond_resched();
539                         goto again;
540                 }
541         } else {
542 cleanup_and_bail_uncompressed:
543                 /*
544                  * No compression, but we still need to write the pages in
545                  * the file we've been given so far.  redirty the locked
546                  * page if it corresponds to our extent and set things up
547                  * for the async work queue to run cow_file_range to do
548                  * the normal delalloc dance
549                  */
550                 if (page_offset(locked_page) >= start &&
551                     page_offset(locked_page) <= end) {
552                         __set_page_dirty_nobuffers(locked_page);
553                         /* unlocked later on in the async handlers */
554                 }
555                 add_async_extent(async_cow, start, end - start + 1,
556                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
557                 *num_added += 1;
558         }
559
560 out:
561         return ret;
562
563 free_pages_out:
564         for (i = 0; i < nr_pages_ret; i++) {
565                 WARN_ON(pages[i]->mapping);
566                 page_cache_release(pages[i]);
567         }
568         kfree(pages);
569
570         goto out;
571
572 cleanup_and_out:
573         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
574                                      start, end, NULL,
575                                      EXTENT_CLEAR_UNLOCK_PAGE |
576                                      EXTENT_CLEAR_DIRTY |
577                                      EXTENT_CLEAR_DELALLOC |
578                                      EXTENT_SET_WRITEBACK |
579                                      EXTENT_END_WRITEBACK);
580         if (!trans || IS_ERR(trans))
581                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
582         else
583                 btrfs_abort_transaction(trans, root, ret);
584         goto free_pages_out;
585 }
586
587 /*
588  * phase two of compressed writeback.  This is the ordered portion
589  * of the code, which only gets called in the order the work was
590  * queued.  We walk all the async extents created by compress_file_range
591  * and send them down to the disk.
592  */
593 static noinline int submit_compressed_extents(struct inode *inode,
594                                               struct async_cow *async_cow)
595 {
596         struct async_extent *async_extent;
597         u64 alloc_hint = 0;
598         struct btrfs_trans_handle *trans;
599         struct btrfs_key ins;
600         struct extent_map *em;
601         struct btrfs_root *root = BTRFS_I(inode)->root;
602         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
603         struct extent_io_tree *io_tree;
604         int ret = 0;
605
606         if (list_empty(&async_cow->extents))
607                 return 0;
608
609
610         while (!list_empty(&async_cow->extents)) {
611                 async_extent = list_entry(async_cow->extents.next,
612                                           struct async_extent, list);
613                 list_del(&async_extent->list);
614
615                 io_tree = &BTRFS_I(inode)->io_tree;
616
617 retry:
618                 /* did the compression code fall back to uncompressed IO? */
619                 if (!async_extent->pages) {
620                         int page_started = 0;
621                         unsigned long nr_written = 0;
622
623                         lock_extent(io_tree, async_extent->start,
624                                          async_extent->start +
625                                          async_extent->ram_size - 1);
626
627                         /* allocate blocks */
628                         ret = cow_file_range(inode, async_cow->locked_page,
629                                              async_extent->start,
630                                              async_extent->start +
631                                              async_extent->ram_size - 1,
632                                              &page_started, &nr_written, 0);
633
634                         /* JDM XXX */
635
636                         /*
637                          * if page_started, cow_file_range inserted an
638                          * inline extent and took care of all the unlocking
639                          * and IO for us.  Otherwise, we need to submit
640                          * all those pages down to the drive.
641                          */
642                         if (!page_started && !ret)
643                                 extent_write_locked_range(io_tree,
644                                                   inode, async_extent->start,
645                                                   async_extent->start +
646                                                   async_extent->ram_size - 1,
647                                                   btrfs_get_extent,
648                                                   WB_SYNC_ALL);
649                         kfree(async_extent);
650                         cond_resched();
651                         continue;
652                 }
653
654                 lock_extent(io_tree, async_extent->start,
655                             async_extent->start + async_extent->ram_size - 1);
656
657                 trans = btrfs_join_transaction(root);
658                 if (IS_ERR(trans)) {
659                         ret = PTR_ERR(trans);
660                 } else {
661                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
662                         ret = btrfs_reserve_extent(trans, root,
663                                            async_extent->compressed_size,
664                                            async_extent->compressed_size,
665                                            0, alloc_hint, &ins, 1);
666                         if (ret)
667                                 btrfs_abort_transaction(trans, root, ret);
668                         btrfs_end_transaction(trans, root);
669                 }
670
671                 if (ret) {
672                         int i;
673                         for (i = 0; i < async_extent->nr_pages; i++) {
674                                 WARN_ON(async_extent->pages[i]->mapping);
675                                 page_cache_release(async_extent->pages[i]);
676                         }
677                         kfree(async_extent->pages);
678                         async_extent->nr_pages = 0;
679                         async_extent->pages = NULL;
680                         unlock_extent(io_tree, async_extent->start,
681                                       async_extent->start +
682                                       async_extent->ram_size - 1);
683                         if (ret == -ENOSPC)
684                                 goto retry;
685                         goto out_free; /* JDM: Requeue? */
686                 }
687
688                 /*
689                  * here we're doing allocation and writeback of the
690                  * compressed pages
691                  */
692                 btrfs_drop_extent_cache(inode, async_extent->start,
693                                         async_extent->start +
694                                         async_extent->ram_size - 1, 0);
695
696                 em = alloc_extent_map();
697                 BUG_ON(!em); /* -ENOMEM */
698                 em->start = async_extent->start;
699                 em->len = async_extent->ram_size;
700                 em->orig_start = em->start;
701
702                 em->block_start = ins.objectid;
703                 em->block_len = ins.offset;
704                 em->bdev = root->fs_info->fs_devices->latest_bdev;
705                 em->compress_type = async_extent->compress_type;
706                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
707                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
708
709                 while (1) {
710                         write_lock(&em_tree->lock);
711                         ret = add_extent_mapping(em_tree, em);
712                         write_unlock(&em_tree->lock);
713                         if (ret != -EEXIST) {
714                                 free_extent_map(em);
715                                 break;
716                         }
717                         btrfs_drop_extent_cache(inode, async_extent->start,
718                                                 async_extent->start +
719                                                 async_extent->ram_size - 1, 0);
720                 }
721
722                 ret = btrfs_add_ordered_extent_compress(inode,
723                                                 async_extent->start,
724                                                 ins.objectid,
725                                                 async_extent->ram_size,
726                                                 ins.offset,
727                                                 BTRFS_ORDERED_COMPRESSED,
728                                                 async_extent->compress_type);
729                 BUG_ON(ret); /* -ENOMEM */
730
731                 /*
732                  * clear dirty, set writeback and unlock the pages.
733                  */
734                 extent_clear_unlock_delalloc(inode,
735                                 &BTRFS_I(inode)->io_tree,
736                                 async_extent->start,
737                                 async_extent->start +
738                                 async_extent->ram_size - 1,
739                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
740                                 EXTENT_CLEAR_UNLOCK |
741                                 EXTENT_CLEAR_DELALLOC |
742                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
743
744                 ret = btrfs_submit_compressed_write(inode,
745                                     async_extent->start,
746                                     async_extent->ram_size,
747                                     ins.objectid,
748                                     ins.offset, async_extent->pages,
749                                     async_extent->nr_pages);
750
751                 BUG_ON(ret); /* -ENOMEM */
752                 alloc_hint = ins.objectid + ins.offset;
753                 kfree(async_extent);
754                 cond_resched();
755         }
756         ret = 0;
757 out:
758         return ret;
759 out_free:
760         kfree(async_extent);
761         goto out;
762 }
763
764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
765                                       u64 num_bytes)
766 {
767         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
768         struct extent_map *em;
769         u64 alloc_hint = 0;
770
771         read_lock(&em_tree->lock);
772         em = search_extent_mapping(em_tree, start, num_bytes);
773         if (em) {
774                 /*
775                  * if block start isn't an actual block number then find the
776                  * first block in this inode and use that as a hint.  If that
777                  * block is also bogus then just don't worry about it.
778                  */
779                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
780                         free_extent_map(em);
781                         em = search_extent_mapping(em_tree, 0, 0);
782                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
783                                 alloc_hint = em->block_start;
784                         if (em)
785                                 free_extent_map(em);
786                 } else {
787                         alloc_hint = em->block_start;
788                         free_extent_map(em);
789                 }
790         }
791         read_unlock(&em_tree->lock);
792
793         return alloc_hint;
794 }
795
796 /*
797  * when extent_io.c finds a delayed allocation range in the file,
798  * the call backs end up in this code.  The basic idea is to
799  * allocate extents on disk for the range, and create ordered data structs
800  * in ram to track those extents.
801  *
802  * locked_page is the page that writepage had locked already.  We use
803  * it to make sure we don't do extra locks or unlocks.
804  *
805  * *page_started is set to one if we unlock locked_page and do everything
806  * required to start IO on it.  It may be clean and already done with
807  * IO when we return.
808  */
809 static noinline int cow_file_range(struct inode *inode,
810                                    struct page *locked_page,
811                                    u64 start, u64 end, int *page_started,
812                                    unsigned long *nr_written,
813                                    int unlock)
814 {
815         struct btrfs_root *root = BTRFS_I(inode)->root;
816         struct btrfs_trans_handle *trans;
817         u64 alloc_hint = 0;
818         u64 num_bytes;
819         unsigned long ram_size;
820         u64 disk_num_bytes;
821         u64 cur_alloc_size;
822         u64 blocksize = root->sectorsize;
823         struct btrfs_key ins;
824         struct extent_map *em;
825         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826         int ret = 0;
827
828         BUG_ON(btrfs_is_free_space_inode(inode));
829         trans = btrfs_join_transaction(root);
830         if (IS_ERR(trans)) {
831                 extent_clear_unlock_delalloc(inode,
832                              &BTRFS_I(inode)->io_tree,
833                              start, end, locked_page,
834                              EXTENT_CLEAR_UNLOCK_PAGE |
835                              EXTENT_CLEAR_UNLOCK |
836                              EXTENT_CLEAR_DELALLOC |
837                              EXTENT_CLEAR_DIRTY |
838                              EXTENT_SET_WRITEBACK |
839                              EXTENT_END_WRITEBACK);
840                 return PTR_ERR(trans);
841         }
842         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
843
844         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
845         num_bytes = max(blocksize,  num_bytes);
846         disk_num_bytes = num_bytes;
847         ret = 0;
848
849         /* if this is a small write inside eof, kick off defrag */
850         if (num_bytes < 64 * 1024 &&
851             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
852                 btrfs_add_inode_defrag(trans, inode);
853
854         if (start == 0) {
855                 /* lets try to make an inline extent */
856                 ret = cow_file_range_inline(trans, root, inode,
857                                             start, end, 0, 0, NULL);
858                 if (ret == 0) {
859                         extent_clear_unlock_delalloc(inode,
860                                      &BTRFS_I(inode)->io_tree,
861                                      start, end, NULL,
862                                      EXTENT_CLEAR_UNLOCK_PAGE |
863                                      EXTENT_CLEAR_UNLOCK |
864                                      EXTENT_CLEAR_DELALLOC |
865                                      EXTENT_CLEAR_DIRTY |
866                                      EXTENT_SET_WRITEBACK |
867                                      EXTENT_END_WRITEBACK);
868
869                         *nr_written = *nr_written +
870                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
871                         *page_started = 1;
872                         goto out;
873                 } else if (ret < 0) {
874                         btrfs_abort_transaction(trans, root, ret);
875                         goto out_unlock;
876                 }
877         }
878
879         BUG_ON(disk_num_bytes >
880                btrfs_super_total_bytes(root->fs_info->super_copy));
881
882         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
883         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
884
885         while (disk_num_bytes > 0) {
886                 unsigned long op;
887
888                 cur_alloc_size = disk_num_bytes;
889                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
890                                            root->sectorsize, 0, alloc_hint,
891                                            &ins, 1);
892                 if (ret < 0) {
893                         btrfs_abort_transaction(trans, root, ret);
894                         goto out_unlock;
895                 }
896
897                 em = alloc_extent_map();
898                 BUG_ON(!em); /* -ENOMEM */
899                 em->start = start;
900                 em->orig_start = em->start;
901                 ram_size = ins.offset;
902                 em->len = ins.offset;
903
904                 em->block_start = ins.objectid;
905                 em->block_len = ins.offset;
906                 em->bdev = root->fs_info->fs_devices->latest_bdev;
907                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
908
909                 while (1) {
910                         write_lock(&em_tree->lock);
911                         ret = add_extent_mapping(em_tree, em);
912                         write_unlock(&em_tree->lock);
913                         if (ret != -EEXIST) {
914                                 free_extent_map(em);
915                                 break;
916                         }
917                         btrfs_drop_extent_cache(inode, start,
918                                                 start + ram_size - 1, 0);
919                 }
920
921                 cur_alloc_size = ins.offset;
922                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
923                                                ram_size, cur_alloc_size, 0);
924                 BUG_ON(ret); /* -ENOMEM */
925
926                 if (root->root_key.objectid ==
927                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
928                         ret = btrfs_reloc_clone_csums(inode, start,
929                                                       cur_alloc_size);
930                         if (ret) {
931                                 btrfs_abort_transaction(trans, root, ret);
932                                 goto out_unlock;
933                         }
934                 }
935
936                 if (disk_num_bytes < cur_alloc_size)
937                         break;
938
939                 /* we're not doing compressed IO, don't unlock the first
940                  * page (which the caller expects to stay locked), don't
941                  * clear any dirty bits and don't set any writeback bits
942                  *
943                  * Do set the Private2 bit so we know this page was properly
944                  * setup for writepage
945                  */
946                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
947                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
948                         EXTENT_SET_PRIVATE2;
949
950                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
951                                              start, start + ram_size - 1,
952                                              locked_page, op);
953                 disk_num_bytes -= cur_alloc_size;
954                 num_bytes -= cur_alloc_size;
955                 alloc_hint = ins.objectid + ins.offset;
956                 start += cur_alloc_size;
957         }
958         ret = 0;
959 out:
960         btrfs_end_transaction(trans, root);
961
962         return ret;
963 out_unlock:
964         extent_clear_unlock_delalloc(inode,
965                      &BTRFS_I(inode)->io_tree,
966                      start, end, locked_page,
967                      EXTENT_CLEAR_UNLOCK_PAGE |
968                      EXTENT_CLEAR_UNLOCK |
969                      EXTENT_CLEAR_DELALLOC |
970                      EXTENT_CLEAR_DIRTY |
971                      EXTENT_SET_WRITEBACK |
972                      EXTENT_END_WRITEBACK);
973
974         goto out;
975 }
976
977 /*
978  * work queue call back to started compression on a file and pages
979  */
980 static noinline void async_cow_start(struct btrfs_work *work)
981 {
982         struct async_cow *async_cow;
983         int num_added = 0;
984         async_cow = container_of(work, struct async_cow, work);
985
986         compress_file_range(async_cow->inode, async_cow->locked_page,
987                             async_cow->start, async_cow->end, async_cow,
988                             &num_added);
989         if (num_added == 0) {
990                 btrfs_add_delayed_iput(async_cow->inode);
991                 async_cow->inode = NULL;
992         }
993 }
994
995 /*
996  * work queue call back to submit previously compressed pages
997  */
998 static noinline void async_cow_submit(struct btrfs_work *work)
999 {
1000         struct async_cow *async_cow;
1001         struct btrfs_root *root;
1002         unsigned long nr_pages;
1003
1004         async_cow = container_of(work, struct async_cow, work);
1005
1006         root = async_cow->root;
1007         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1008                 PAGE_CACHE_SHIFT;
1009
1010         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1011
1012         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1013             5 * 1024 * 1024 &&
1014             waitqueue_active(&root->fs_info->async_submit_wait))
1015                 wake_up(&root->fs_info->async_submit_wait);
1016
1017         if (async_cow->inode)
1018                 submit_compressed_extents(async_cow->inode, async_cow);
1019 }
1020
1021 static noinline void async_cow_free(struct btrfs_work *work)
1022 {
1023         struct async_cow *async_cow;
1024         async_cow = container_of(work, struct async_cow, work);
1025         if (async_cow->inode)
1026                 btrfs_add_delayed_iput(async_cow->inode);
1027         kfree(async_cow);
1028 }
1029
1030 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1031                                 u64 start, u64 end, int *page_started,
1032                                 unsigned long *nr_written)
1033 {
1034         struct async_cow *async_cow;
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         unsigned long nr_pages;
1037         u64 cur_end;
1038         int limit = 10 * 1024 * 1024;
1039
1040         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1041                          1, 0, NULL, GFP_NOFS);
1042         while (start < end) {
1043                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1044                 BUG_ON(!async_cow); /* -ENOMEM */
1045                 async_cow->inode = igrab(inode);
1046                 async_cow->root = root;
1047                 async_cow->locked_page = locked_page;
1048                 async_cow->start = start;
1049
1050                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1051                         cur_end = end;
1052                 else
1053                         cur_end = min(end, start + 512 * 1024 - 1);
1054
1055                 async_cow->end = cur_end;
1056                 INIT_LIST_HEAD(&async_cow->extents);
1057
1058                 async_cow->work.func = async_cow_start;
1059                 async_cow->work.ordered_func = async_cow_submit;
1060                 async_cow->work.ordered_free = async_cow_free;
1061                 async_cow->work.flags = 0;
1062
1063                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1064                         PAGE_CACHE_SHIFT;
1065                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1066
1067                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1068                                    &async_cow->work);
1069
1070                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1071                         wait_event(root->fs_info->async_submit_wait,
1072                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1073                             limit));
1074                 }
1075
1076                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1077                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1078                         wait_event(root->fs_info->async_submit_wait,
1079                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1080                            0));
1081                 }
1082
1083                 *nr_written += nr_pages;
1084                 start = cur_end + 1;
1085         }
1086         *page_started = 1;
1087         return 0;
1088 }
1089
1090 static noinline int csum_exist_in_range(struct btrfs_root *root,
1091                                         u64 bytenr, u64 num_bytes)
1092 {
1093         int ret;
1094         struct btrfs_ordered_sum *sums;
1095         LIST_HEAD(list);
1096
1097         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1098                                        bytenr + num_bytes - 1, &list, 0);
1099         if (ret == 0 && list_empty(&list))
1100                 return 0;
1101
1102         while (!list_empty(&list)) {
1103                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1104                 list_del(&sums->list);
1105                 kfree(sums);
1106         }
1107         return 1;
1108 }
1109
1110 /*
1111  * when nowcow writeback call back.  This checks for snapshots or COW copies
1112  * of the extents that exist in the file, and COWs the file as required.
1113  *
1114  * If no cow copies or snapshots exist, we write directly to the existing
1115  * blocks on disk
1116  */
1117 static noinline int run_delalloc_nocow(struct inode *inode,
1118                                        struct page *locked_page,
1119                               u64 start, u64 end, int *page_started, int force,
1120                               unsigned long *nr_written)
1121 {
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         struct btrfs_trans_handle *trans;
1124         struct extent_buffer *leaf;
1125         struct btrfs_path *path;
1126         struct btrfs_file_extent_item *fi;
1127         struct btrfs_key found_key;
1128         u64 cow_start;
1129         u64 cur_offset;
1130         u64 extent_end;
1131         u64 extent_offset;
1132         u64 disk_bytenr;
1133         u64 num_bytes;
1134         int extent_type;
1135         int ret, err;
1136         int type;
1137         int nocow;
1138         int check_prev = 1;
1139         bool nolock;
1140         u64 ino = btrfs_ino(inode);
1141
1142         path = btrfs_alloc_path();
1143         if (!path) {
1144                 extent_clear_unlock_delalloc(inode,
1145                              &BTRFS_I(inode)->io_tree,
1146                              start, end, locked_page,
1147                              EXTENT_CLEAR_UNLOCK_PAGE |
1148                              EXTENT_CLEAR_UNLOCK |
1149                              EXTENT_CLEAR_DELALLOC |
1150                              EXTENT_CLEAR_DIRTY |
1151                              EXTENT_SET_WRITEBACK |
1152                              EXTENT_END_WRITEBACK);
1153                 return -ENOMEM;
1154         }
1155
1156         nolock = btrfs_is_free_space_inode(inode);
1157
1158         if (nolock)
1159                 trans = btrfs_join_transaction_nolock(root);
1160         else
1161                 trans = btrfs_join_transaction(root);
1162
1163         if (IS_ERR(trans)) {
1164                 extent_clear_unlock_delalloc(inode,
1165                              &BTRFS_I(inode)->io_tree,
1166                              start, end, locked_page,
1167                              EXTENT_CLEAR_UNLOCK_PAGE |
1168                              EXTENT_CLEAR_UNLOCK |
1169                              EXTENT_CLEAR_DELALLOC |
1170                              EXTENT_CLEAR_DIRTY |
1171                              EXTENT_SET_WRITEBACK |
1172                              EXTENT_END_WRITEBACK);
1173                 btrfs_free_path(path);
1174                 return PTR_ERR(trans);
1175         }
1176
1177         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1178
1179         cow_start = (u64)-1;
1180         cur_offset = start;
1181         while (1) {
1182                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1183                                                cur_offset, 0);
1184                 if (ret < 0) {
1185                         btrfs_abort_transaction(trans, root, ret);
1186                         goto error;
1187                 }
1188                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1189                         leaf = path->nodes[0];
1190                         btrfs_item_key_to_cpu(leaf, &found_key,
1191                                               path->slots[0] - 1);
1192                         if (found_key.objectid == ino &&
1193                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1194                                 path->slots[0]--;
1195                 }
1196                 check_prev = 0;
1197 next_slot:
1198                 leaf = path->nodes[0];
1199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1200                         ret = btrfs_next_leaf(root, path);
1201                         if (ret < 0) {
1202                                 btrfs_abort_transaction(trans, root, ret);
1203                                 goto error;
1204                         }
1205                         if (ret > 0)
1206                                 break;
1207                         leaf = path->nodes[0];
1208                 }
1209
1210                 nocow = 0;
1211                 disk_bytenr = 0;
1212                 num_bytes = 0;
1213                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1214
1215                 if (found_key.objectid > ino ||
1216                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1217                     found_key.offset > end)
1218                         break;
1219
1220                 if (found_key.offset > cur_offset) {
1221                         extent_end = found_key.offset;
1222                         extent_type = 0;
1223                         goto out_check;
1224                 }
1225
1226                 fi = btrfs_item_ptr(leaf, path->slots[0],
1227                                     struct btrfs_file_extent_item);
1228                 extent_type = btrfs_file_extent_type(leaf, fi);
1229
1230                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1231                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1232                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1233                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1234                         extent_end = found_key.offset +
1235                                 btrfs_file_extent_num_bytes(leaf, fi);
1236                         if (extent_end <= start) {
1237                                 path->slots[0]++;
1238                                 goto next_slot;
1239                         }
1240                         if (disk_bytenr == 0)
1241                                 goto out_check;
1242                         if (btrfs_file_extent_compression(leaf, fi) ||
1243                             btrfs_file_extent_encryption(leaf, fi) ||
1244                             btrfs_file_extent_other_encoding(leaf, fi))
1245                                 goto out_check;
1246                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247                                 goto out_check;
1248                         if (btrfs_extent_readonly(root, disk_bytenr))
1249                                 goto out_check;
1250                         if (btrfs_cross_ref_exist(trans, root, ino,
1251                                                   found_key.offset -
1252                                                   extent_offset, disk_bytenr))
1253                                 goto out_check;
1254                         disk_bytenr += extent_offset;
1255                         disk_bytenr += cur_offset - found_key.offset;
1256                         num_bytes = min(end + 1, extent_end) - cur_offset;
1257                         /*
1258                          * force cow if csum exists in the range.
1259                          * this ensure that csum for a given extent are
1260                          * either valid or do not exist.
1261                          */
1262                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263                                 goto out_check;
1264                         nocow = 1;
1265                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266                         extent_end = found_key.offset +
1267                                 btrfs_file_extent_inline_len(leaf, fi);
1268                         extent_end = ALIGN(extent_end, root->sectorsize);
1269                 } else {
1270                         BUG_ON(1);
1271                 }
1272 out_check:
1273                 if (extent_end <= start) {
1274                         path->slots[0]++;
1275                         goto next_slot;
1276                 }
1277                 if (!nocow) {
1278                         if (cow_start == (u64)-1)
1279                                 cow_start = cur_offset;
1280                         cur_offset = extent_end;
1281                         if (cur_offset > end)
1282                                 break;
1283                         path->slots[0]++;
1284                         goto next_slot;
1285                 }
1286
1287                 btrfs_release_path(path);
1288                 if (cow_start != (u64)-1) {
1289                         ret = cow_file_range(inode, locked_page, cow_start,
1290                                         found_key.offset - 1, page_started,
1291                                         nr_written, 1);
1292                         if (ret) {
1293                                 btrfs_abort_transaction(trans, root, ret);
1294                                 goto error;
1295                         }
1296                         cow_start = (u64)-1;
1297                 }
1298
1299                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300                         struct extent_map *em;
1301                         struct extent_map_tree *em_tree;
1302                         em_tree = &BTRFS_I(inode)->extent_tree;
1303                         em = alloc_extent_map();
1304                         BUG_ON(!em); /* -ENOMEM */
1305                         em->start = cur_offset;
1306                         em->orig_start = em->start;
1307                         em->len = num_bytes;
1308                         em->block_len = num_bytes;
1309                         em->block_start = disk_bytenr;
1310                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1311                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312                         while (1) {
1313                                 write_lock(&em_tree->lock);
1314                                 ret = add_extent_mapping(em_tree, em);
1315                                 write_unlock(&em_tree->lock);
1316                                 if (ret != -EEXIST) {
1317                                         free_extent_map(em);
1318                                         break;
1319                                 }
1320                                 btrfs_drop_extent_cache(inode, em->start,
1321                                                 em->start + em->len - 1, 0);
1322                         }
1323                         type = BTRFS_ORDERED_PREALLOC;
1324                 } else {
1325                         type = BTRFS_ORDERED_NOCOW;
1326                 }
1327
1328                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1329                                                num_bytes, num_bytes, type);
1330                 BUG_ON(ret); /* -ENOMEM */
1331
1332                 if (root->root_key.objectid ==
1333                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1334                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1335                                                       num_bytes);
1336                         if (ret) {
1337                                 btrfs_abort_transaction(trans, root, ret);
1338                                 goto error;
1339                         }
1340                 }
1341
1342                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1343                                 cur_offset, cur_offset + num_bytes - 1,
1344                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1345                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1346                                 EXTENT_SET_PRIVATE2);
1347                 cur_offset = extent_end;
1348                 if (cur_offset > end)
1349                         break;
1350         }
1351         btrfs_release_path(path);
1352
1353         if (cur_offset <= end && cow_start == (u64)-1) {
1354                 cow_start = cur_offset;
1355                 cur_offset = end;
1356         }
1357
1358         if (cow_start != (u64)-1) {
1359                 ret = cow_file_range(inode, locked_page, cow_start, end,
1360                                      page_started, nr_written, 1);
1361                 if (ret) {
1362                         btrfs_abort_transaction(trans, root, ret);
1363                         goto error;
1364                 }
1365         }
1366
1367 error:
1368         if (nolock) {
1369                 err = btrfs_end_transaction_nolock(trans, root);
1370         } else {
1371                 err = btrfs_end_transaction(trans, root);
1372         }
1373         if (!ret)
1374                 ret = err;
1375
1376         if (ret && cur_offset < end)
1377                 extent_clear_unlock_delalloc(inode,
1378                              &BTRFS_I(inode)->io_tree,
1379                              cur_offset, end, locked_page,
1380                              EXTENT_CLEAR_UNLOCK_PAGE |
1381                              EXTENT_CLEAR_UNLOCK |
1382                              EXTENT_CLEAR_DELALLOC |
1383                              EXTENT_CLEAR_DIRTY |
1384                              EXTENT_SET_WRITEBACK |
1385                              EXTENT_END_WRITEBACK);
1386
1387         btrfs_free_path(path);
1388         return ret;
1389 }
1390
1391 /*
1392  * extent_io.c call back to do delayed allocation processing
1393  */
1394 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1395                               u64 start, u64 end, int *page_started,
1396                               unsigned long *nr_written)
1397 {
1398         int ret;
1399         struct btrfs_root *root = BTRFS_I(inode)->root;
1400
1401         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1402                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1403                                          page_started, 1, nr_written);
1404         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1405                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1406                                          page_started, 0, nr_written);
1407         } else if (!btrfs_test_opt(root, COMPRESS) &&
1408                    !(BTRFS_I(inode)->force_compress) &&
1409                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1410                 ret = cow_file_range(inode, locked_page, start, end,
1411                                       page_started, nr_written, 1);
1412         } else {
1413                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1414                         &BTRFS_I(inode)->runtime_flags);
1415                 ret = cow_file_range_async(inode, locked_page, start, end,
1416                                            page_started, nr_written);
1417         }
1418         return ret;
1419 }
1420
1421 static void btrfs_split_extent_hook(struct inode *inode,
1422                                     struct extent_state *orig, u64 split)
1423 {
1424         /* not delalloc, ignore it */
1425         if (!(orig->state & EXTENT_DELALLOC))
1426                 return;
1427
1428         spin_lock(&BTRFS_I(inode)->lock);
1429         BTRFS_I(inode)->outstanding_extents++;
1430         spin_unlock(&BTRFS_I(inode)->lock);
1431 }
1432
1433 /*
1434  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1435  * extents so we can keep track of new extents that are just merged onto old
1436  * extents, such as when we are doing sequential writes, so we can properly
1437  * account for the metadata space we'll need.
1438  */
1439 static void btrfs_merge_extent_hook(struct inode *inode,
1440                                     struct extent_state *new,
1441                                     struct extent_state *other)
1442 {
1443         /* not delalloc, ignore it */
1444         if (!(other->state & EXTENT_DELALLOC))
1445                 return;
1446
1447         spin_lock(&BTRFS_I(inode)->lock);
1448         BTRFS_I(inode)->outstanding_extents--;
1449         spin_unlock(&BTRFS_I(inode)->lock);
1450 }
1451
1452 /*
1453  * extent_io.c set_bit_hook, used to track delayed allocation
1454  * bytes in this file, and to maintain the list of inodes that
1455  * have pending delalloc work to be done.
1456  */
1457 static void btrfs_set_bit_hook(struct inode *inode,
1458                                struct extent_state *state, int *bits)
1459 {
1460
1461         /*
1462          * set_bit and clear bit hooks normally require _irqsave/restore
1463          * but in this case, we are only testing for the DELALLOC
1464          * bit, which is only set or cleared with irqs on
1465          */
1466         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1467                 struct btrfs_root *root = BTRFS_I(inode)->root;
1468                 u64 len = state->end + 1 - state->start;
1469                 bool do_list = !btrfs_is_free_space_inode(inode);
1470
1471                 if (*bits & EXTENT_FIRST_DELALLOC) {
1472                         *bits &= ~EXTENT_FIRST_DELALLOC;
1473                 } else {
1474                         spin_lock(&BTRFS_I(inode)->lock);
1475                         BTRFS_I(inode)->outstanding_extents++;
1476                         spin_unlock(&BTRFS_I(inode)->lock);
1477                 }
1478
1479                 spin_lock(&root->fs_info->delalloc_lock);
1480                 BTRFS_I(inode)->delalloc_bytes += len;
1481                 root->fs_info->delalloc_bytes += len;
1482                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1483                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1484                                       &root->fs_info->delalloc_inodes);
1485                 }
1486                 spin_unlock(&root->fs_info->delalloc_lock);
1487         }
1488 }
1489
1490 /*
1491  * extent_io.c clear_bit_hook, see set_bit_hook for why
1492  */
1493 static void btrfs_clear_bit_hook(struct inode *inode,
1494                                  struct extent_state *state, int *bits)
1495 {
1496         /*
1497          * set_bit and clear bit hooks normally require _irqsave/restore
1498          * but in this case, we are only testing for the DELALLOC
1499          * bit, which is only set or cleared with irqs on
1500          */
1501         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1502                 struct btrfs_root *root = BTRFS_I(inode)->root;
1503                 u64 len = state->end + 1 - state->start;
1504                 bool do_list = !btrfs_is_free_space_inode(inode);
1505
1506                 if (*bits & EXTENT_FIRST_DELALLOC) {
1507                         *bits &= ~EXTENT_FIRST_DELALLOC;
1508                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1509                         spin_lock(&BTRFS_I(inode)->lock);
1510                         BTRFS_I(inode)->outstanding_extents--;
1511                         spin_unlock(&BTRFS_I(inode)->lock);
1512                 }
1513
1514                 if (*bits & EXTENT_DO_ACCOUNTING)
1515                         btrfs_delalloc_release_metadata(inode, len);
1516
1517                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1518                     && do_list)
1519                         btrfs_free_reserved_data_space(inode, len);
1520
1521                 spin_lock(&root->fs_info->delalloc_lock);
1522                 root->fs_info->delalloc_bytes -= len;
1523                 BTRFS_I(inode)->delalloc_bytes -= len;
1524
1525                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1526                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1528                 }
1529                 spin_unlock(&root->fs_info->delalloc_lock);
1530         }
1531 }
1532
1533 /*
1534  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1535  * we don't create bios that span stripes or chunks
1536  */
1537 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1538                          size_t size, struct bio *bio,
1539                          unsigned long bio_flags)
1540 {
1541         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1542         struct btrfs_mapping_tree *map_tree;
1543         u64 logical = (u64)bio->bi_sector << 9;
1544         u64 length = 0;
1545         u64 map_length;
1546         int ret;
1547
1548         if (bio_flags & EXTENT_BIO_COMPRESSED)
1549                 return 0;
1550
1551         length = bio->bi_size;
1552         map_tree = &root->fs_info->mapping_tree;
1553         map_length = length;
1554         ret = btrfs_map_block(map_tree, READ, logical,
1555                               &map_length, NULL, 0);
1556         /* Will always return 0 or 1 with map_multi == NULL */
1557         BUG_ON(ret < 0);
1558         if (map_length < length + size)
1559                 return 1;
1560         return 0;
1561 }
1562
1563 /*
1564  * in order to insert checksums into the metadata in large chunks,
1565  * we wait until bio submission time.   All the pages in the bio are
1566  * checksummed and sums are attached onto the ordered extent record.
1567  *
1568  * At IO completion time the cums attached on the ordered extent record
1569  * are inserted into the btree
1570  */
1571 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1572                                     struct bio *bio, int mirror_num,
1573                                     unsigned long bio_flags,
1574                                     u64 bio_offset)
1575 {
1576         struct btrfs_root *root = BTRFS_I(inode)->root;
1577         int ret = 0;
1578
1579         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1580         BUG_ON(ret); /* -ENOMEM */
1581         return 0;
1582 }
1583
1584 /*
1585  * in order to insert checksums into the metadata in large chunks,
1586  * we wait until bio submission time.   All the pages in the bio are
1587  * checksummed and sums are attached onto the ordered extent record.
1588  *
1589  * At IO completion time the cums attached on the ordered extent record
1590  * are inserted into the btree
1591  */
1592 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1593                           int mirror_num, unsigned long bio_flags,
1594                           u64 bio_offset)
1595 {
1596         struct btrfs_root *root = BTRFS_I(inode)->root;
1597         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1598 }
1599
1600 /*
1601  * extent_io.c submission hook. This does the right thing for csum calculation
1602  * on write, or reading the csums from the tree before a read
1603  */
1604 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1605                           int mirror_num, unsigned long bio_flags,
1606                           u64 bio_offset)
1607 {
1608         struct btrfs_root *root = BTRFS_I(inode)->root;
1609         int ret = 0;
1610         int skip_sum;
1611         int metadata = 0;
1612
1613         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1614
1615         if (btrfs_is_free_space_inode(inode))
1616                 metadata = 2;
1617
1618         if (!(rw & REQ_WRITE)) {
1619                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1620                 if (ret)
1621                         return ret;
1622
1623                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1624                         return btrfs_submit_compressed_read(inode, bio,
1625                                                     mirror_num, bio_flags);
1626                 } else if (!skip_sum) {
1627                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1628                         if (ret)
1629                                 return ret;
1630                 }
1631                 goto mapit;
1632         } else if (!skip_sum) {
1633                 /* csum items have already been cloned */
1634                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1635                         goto mapit;
1636                 /* we're doing a write, do the async checksumming */
1637                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1638                                    inode, rw, bio, mirror_num,
1639                                    bio_flags, bio_offset,
1640                                    __btrfs_submit_bio_start,
1641                                    __btrfs_submit_bio_done);
1642         }
1643
1644 mapit:
1645         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1646 }
1647
1648 /*
1649  * given a list of ordered sums record them in the inode.  This happens
1650  * at IO completion time based on sums calculated at bio submission time.
1651  */
1652 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1653                              struct inode *inode, u64 file_offset,
1654                              struct list_head *list)
1655 {
1656         struct btrfs_ordered_sum *sum;
1657
1658         list_for_each_entry(sum, list, list) {
1659                 btrfs_csum_file_blocks(trans,
1660                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1661         }
1662         return 0;
1663 }
1664
1665 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1666                               struct extent_state **cached_state)
1667 {
1668         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1669                 WARN_ON(1);
1670         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1671                                    cached_state, GFP_NOFS);
1672 }
1673
1674 /* see btrfs_writepage_start_hook for details on why this is required */
1675 struct btrfs_writepage_fixup {
1676         struct page *page;
1677         struct btrfs_work work;
1678 };
1679
1680 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1681 {
1682         struct btrfs_writepage_fixup *fixup;
1683         struct btrfs_ordered_extent *ordered;
1684         struct extent_state *cached_state = NULL;
1685         struct page *page;
1686         struct inode *inode;
1687         u64 page_start;
1688         u64 page_end;
1689         int ret;
1690
1691         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1692         page = fixup->page;
1693 again:
1694         lock_page(page);
1695         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1696                 ClearPageChecked(page);
1697                 goto out_page;
1698         }
1699
1700         inode = page->mapping->host;
1701         page_start = page_offset(page);
1702         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1703
1704         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1705                          &cached_state);
1706
1707         /* already ordered? We're done */
1708         if (PagePrivate2(page))
1709                 goto out;
1710
1711         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1712         if (ordered) {
1713                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1714                                      page_end, &cached_state, GFP_NOFS);
1715                 unlock_page(page);
1716                 btrfs_start_ordered_extent(inode, ordered, 1);
1717                 btrfs_put_ordered_extent(ordered);
1718                 goto again;
1719         }
1720
1721         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1722         if (ret) {
1723                 mapping_set_error(page->mapping, ret);
1724                 end_extent_writepage(page, ret, page_start, page_end);
1725                 ClearPageChecked(page);
1726                 goto out;
1727          }
1728
1729         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1730         ClearPageChecked(page);
1731         set_page_dirty(page);
1732 out:
1733         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1734                              &cached_state, GFP_NOFS);
1735 out_page:
1736         unlock_page(page);
1737         page_cache_release(page);
1738         kfree(fixup);
1739 }
1740
1741 /*
1742  * There are a few paths in the higher layers of the kernel that directly
1743  * set the page dirty bit without asking the filesystem if it is a
1744  * good idea.  This causes problems because we want to make sure COW
1745  * properly happens and the data=ordered rules are followed.
1746  *
1747  * In our case any range that doesn't have the ORDERED bit set
1748  * hasn't been properly setup for IO.  We kick off an async process
1749  * to fix it up.  The async helper will wait for ordered extents, set
1750  * the delalloc bit and make it safe to write the page.
1751  */
1752 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1753 {
1754         struct inode *inode = page->mapping->host;
1755         struct btrfs_writepage_fixup *fixup;
1756         struct btrfs_root *root = BTRFS_I(inode)->root;
1757
1758         /* this page is properly in the ordered list */
1759         if (TestClearPagePrivate2(page))
1760                 return 0;
1761
1762         if (PageChecked(page))
1763                 return -EAGAIN;
1764
1765         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1766         if (!fixup)
1767                 return -EAGAIN;
1768
1769         SetPageChecked(page);
1770         page_cache_get(page);
1771         fixup->work.func = btrfs_writepage_fixup_worker;
1772         fixup->page = page;
1773         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1774         return -EBUSY;
1775 }
1776
1777 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1778                                        struct inode *inode, u64 file_pos,
1779                                        u64 disk_bytenr, u64 disk_num_bytes,
1780                                        u64 num_bytes, u64 ram_bytes,
1781                                        u8 compression, u8 encryption,
1782                                        u16 other_encoding, int extent_type)
1783 {
1784         struct btrfs_root *root = BTRFS_I(inode)->root;
1785         struct btrfs_file_extent_item *fi;
1786         struct btrfs_path *path;
1787         struct extent_buffer *leaf;
1788         struct btrfs_key ins;
1789         u64 hint;
1790         int ret;
1791
1792         path = btrfs_alloc_path();
1793         if (!path)
1794                 return -ENOMEM;
1795
1796         path->leave_spinning = 1;
1797
1798         /*
1799          * we may be replacing one extent in the tree with another.
1800          * The new extent is pinned in the extent map, and we don't want
1801          * to drop it from the cache until it is completely in the btree.
1802          *
1803          * So, tell btrfs_drop_extents to leave this extent in the cache.
1804          * the caller is expected to unpin it and allow it to be merged
1805          * with the others.
1806          */
1807         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1808                                  &hint, 0);
1809         if (ret)
1810                 goto out;
1811
1812         ins.objectid = btrfs_ino(inode);
1813         ins.offset = file_pos;
1814         ins.type = BTRFS_EXTENT_DATA_KEY;
1815         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1816         if (ret)
1817                 goto out;
1818         leaf = path->nodes[0];
1819         fi = btrfs_item_ptr(leaf, path->slots[0],
1820                             struct btrfs_file_extent_item);
1821         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1822         btrfs_set_file_extent_type(leaf, fi, extent_type);
1823         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1824         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1825         btrfs_set_file_extent_offset(leaf, fi, 0);
1826         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1827         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1828         btrfs_set_file_extent_compression(leaf, fi, compression);
1829         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1830         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1831
1832         btrfs_unlock_up_safe(path, 1);
1833         btrfs_set_lock_blocking(leaf);
1834
1835         btrfs_mark_buffer_dirty(leaf);
1836
1837         inode_add_bytes(inode, num_bytes);
1838
1839         ins.objectid = disk_bytenr;
1840         ins.offset = disk_num_bytes;
1841         ins.type = BTRFS_EXTENT_ITEM_KEY;
1842         ret = btrfs_alloc_reserved_file_extent(trans, root,
1843                                         root->root_key.objectid,
1844                                         btrfs_ino(inode), file_pos, &ins);
1845 out:
1846         btrfs_free_path(path);
1847
1848         return ret;
1849 }
1850
1851 /*
1852  * helper function for btrfs_finish_ordered_io, this
1853  * just reads in some of the csum leaves to prime them into ram
1854  * before we start the transaction.  It limits the amount of btree
1855  * reads required while inside the transaction.
1856  */
1857 /* as ordered data IO finishes, this gets called so we can finish
1858  * an ordered extent if the range of bytes in the file it covers are
1859  * fully written.
1860  */
1861 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1862 {
1863         struct inode *inode = ordered_extent->inode;
1864         struct btrfs_root *root = BTRFS_I(inode)->root;
1865         struct btrfs_trans_handle *trans = NULL;
1866         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1867         struct extent_state *cached_state = NULL;
1868         int compress_type = 0;
1869         int ret;
1870         bool nolock;
1871
1872         nolock = btrfs_is_free_space_inode(inode);
1873
1874         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1875                 ret = -EIO;
1876                 goto out;
1877         }
1878
1879         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1880                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1881                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1882                 if (!ret) {
1883                         if (nolock)
1884                                 trans = btrfs_join_transaction_nolock(root);
1885                         else
1886                                 trans = btrfs_join_transaction(root);
1887                         if (IS_ERR(trans))
1888                                 return PTR_ERR(trans);
1889                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1890                         ret = btrfs_update_inode_fallback(trans, root, inode);
1891                         if (ret) /* -ENOMEM or corruption */
1892                                 btrfs_abort_transaction(trans, root, ret);
1893                 }
1894                 goto out;
1895         }
1896
1897         lock_extent_bits(io_tree, ordered_extent->file_offset,
1898                          ordered_extent->file_offset + ordered_extent->len - 1,
1899                          0, &cached_state);
1900
1901         if (nolock)
1902                 trans = btrfs_join_transaction_nolock(root);
1903         else
1904                 trans = btrfs_join_transaction(root);
1905         if (IS_ERR(trans)) {
1906                 ret = PTR_ERR(trans);
1907                 trans = NULL;
1908                 goto out_unlock;
1909         }
1910         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1911
1912         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1913                 compress_type = ordered_extent->compress_type;
1914         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1915                 BUG_ON(compress_type);
1916                 ret = btrfs_mark_extent_written(trans, inode,
1917                                                 ordered_extent->file_offset,
1918                                                 ordered_extent->file_offset +
1919                                                 ordered_extent->len);
1920         } else {
1921                 BUG_ON(root == root->fs_info->tree_root);
1922                 ret = insert_reserved_file_extent(trans, inode,
1923                                                 ordered_extent->file_offset,
1924                                                 ordered_extent->start,
1925                                                 ordered_extent->disk_len,
1926                                                 ordered_extent->len,
1927                                                 ordered_extent->len,
1928                                                 compress_type, 0, 0,
1929                                                 BTRFS_FILE_EXTENT_REG);
1930                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1931                                    ordered_extent->file_offset,
1932                                    ordered_extent->len);
1933         }
1934
1935         if (ret < 0) {
1936                 btrfs_abort_transaction(trans, root, ret);
1937                 goto out_unlock;
1938         }
1939
1940         add_pending_csums(trans, inode, ordered_extent->file_offset,
1941                           &ordered_extent->list);
1942
1943         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1944         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1945                 ret = btrfs_update_inode_fallback(trans, root, inode);
1946                 if (ret) { /* -ENOMEM or corruption */
1947                         btrfs_abort_transaction(trans, root, ret);
1948                         goto out_unlock;
1949                 }
1950         }
1951         ret = 0;
1952 out_unlock:
1953         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1954                              ordered_extent->file_offset +
1955                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1956 out:
1957         if (root != root->fs_info->tree_root)
1958                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1959         if (trans) {
1960                 if (nolock)
1961                         btrfs_end_transaction_nolock(trans, root);
1962                 else
1963                         btrfs_end_transaction(trans, root);
1964         }
1965
1966         if (ret)
1967                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1968                                       ordered_extent->file_offset +
1969                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1970
1971         /*
1972          * This needs to be dont to make sure anybody waiting knows we are done
1973          * upating everything for this ordered extent.
1974          */
1975         btrfs_remove_ordered_extent(inode, ordered_extent);
1976
1977         /* once for us */
1978         btrfs_put_ordered_extent(ordered_extent);
1979         /* once for the tree */
1980         btrfs_put_ordered_extent(ordered_extent);
1981
1982         return ret;
1983 }
1984
1985 static void finish_ordered_fn(struct btrfs_work *work)
1986 {
1987         struct btrfs_ordered_extent *ordered_extent;
1988         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1989         btrfs_finish_ordered_io(ordered_extent);
1990 }
1991
1992 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1993                                 struct extent_state *state, int uptodate)
1994 {
1995         struct inode *inode = page->mapping->host;
1996         struct btrfs_root *root = BTRFS_I(inode)->root;
1997         struct btrfs_ordered_extent *ordered_extent = NULL;
1998         struct btrfs_workers *workers;
1999
2000         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2001
2002         ClearPagePrivate2(page);
2003         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2004                                             end - start + 1, uptodate))
2005                 return 0;
2006
2007         ordered_extent->work.func = finish_ordered_fn;
2008         ordered_extent->work.flags = 0;
2009
2010         if (btrfs_is_free_space_inode(inode))
2011                 workers = &root->fs_info->endio_freespace_worker;
2012         else
2013                 workers = &root->fs_info->endio_write_workers;
2014         btrfs_queue_worker(workers, &ordered_extent->work);
2015
2016         return 0;
2017 }
2018
2019 /*
2020  * when reads are done, we need to check csums to verify the data is correct
2021  * if there's a match, we allow the bio to finish.  If not, the code in
2022  * extent_io.c will try to find good copies for us.
2023  */
2024 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2025                                struct extent_state *state, int mirror)
2026 {
2027         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2028         struct inode *inode = page->mapping->host;
2029         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2030         char *kaddr;
2031         u64 private = ~(u32)0;
2032         int ret;
2033         struct btrfs_root *root = BTRFS_I(inode)->root;
2034         u32 csum = ~(u32)0;
2035
2036         if (PageChecked(page)) {
2037                 ClearPageChecked(page);
2038                 goto good;
2039         }
2040
2041         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2042                 goto good;
2043
2044         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2045             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2046                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2047                                   GFP_NOFS);
2048                 return 0;
2049         }
2050
2051         if (state && state->start == start) {
2052                 private = state->private;
2053                 ret = 0;
2054         } else {
2055                 ret = get_state_private(io_tree, start, &private);
2056         }
2057         kaddr = kmap_atomic(page);
2058         if (ret)
2059                 goto zeroit;
2060
2061         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2062         btrfs_csum_final(csum, (char *)&csum);
2063         if (csum != private)
2064                 goto zeroit;
2065
2066         kunmap_atomic(kaddr);
2067 good:
2068         return 0;
2069
2070 zeroit:
2071         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2072                        "private %llu\n",
2073                        (unsigned long long)btrfs_ino(page->mapping->host),
2074                        (unsigned long long)start, csum,
2075                        (unsigned long long)private);
2076         memset(kaddr + offset, 1, end - start + 1);
2077         flush_dcache_page(page);
2078         kunmap_atomic(kaddr);
2079         if (private == 0)
2080                 return 0;
2081         return -EIO;
2082 }
2083
2084 struct delayed_iput {
2085         struct list_head list;
2086         struct inode *inode;
2087 };
2088
2089 /* JDM: If this is fs-wide, why can't we add a pointer to
2090  * btrfs_inode instead and avoid the allocation? */
2091 void btrfs_add_delayed_iput(struct inode *inode)
2092 {
2093         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2094         struct delayed_iput *delayed;
2095
2096         if (atomic_add_unless(&inode->i_count, -1, 1))
2097                 return;
2098
2099         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2100         delayed->inode = inode;
2101
2102         spin_lock(&fs_info->delayed_iput_lock);
2103         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2104         spin_unlock(&fs_info->delayed_iput_lock);
2105 }
2106
2107 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2108 {
2109         LIST_HEAD(list);
2110         struct btrfs_fs_info *fs_info = root->fs_info;
2111         struct delayed_iput *delayed;
2112         int empty;
2113
2114         spin_lock(&fs_info->delayed_iput_lock);
2115         empty = list_empty(&fs_info->delayed_iputs);
2116         spin_unlock(&fs_info->delayed_iput_lock);
2117         if (empty)
2118                 return;
2119
2120         down_read(&root->fs_info->cleanup_work_sem);
2121         spin_lock(&fs_info->delayed_iput_lock);
2122         list_splice_init(&fs_info->delayed_iputs, &list);
2123         spin_unlock(&fs_info->delayed_iput_lock);
2124
2125         while (!list_empty(&list)) {
2126                 delayed = list_entry(list.next, struct delayed_iput, list);
2127                 list_del(&delayed->list);
2128                 iput(delayed->inode);
2129                 kfree(delayed);
2130         }
2131         up_read(&root->fs_info->cleanup_work_sem);
2132 }
2133
2134 enum btrfs_orphan_cleanup_state {
2135         ORPHAN_CLEANUP_STARTED  = 1,
2136         ORPHAN_CLEANUP_DONE     = 2,
2137 };
2138
2139 /*
2140  * This is called in transaction commit time. If there are no orphan
2141  * files in the subvolume, it removes orphan item and frees block_rsv
2142  * structure.
2143  */
2144 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2145                               struct btrfs_root *root)
2146 {
2147         struct btrfs_block_rsv *block_rsv;
2148         int ret;
2149
2150         if (atomic_read(&root->orphan_inodes) ||
2151             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2152                 return;
2153
2154         spin_lock(&root->orphan_lock);
2155         if (atomic_read(&root->orphan_inodes)) {
2156                 spin_unlock(&root->orphan_lock);
2157                 return;
2158         }
2159
2160         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2161                 spin_unlock(&root->orphan_lock);
2162                 return;
2163         }
2164
2165         block_rsv = root->orphan_block_rsv;
2166         root->orphan_block_rsv = NULL;
2167         spin_unlock(&root->orphan_lock);
2168
2169         if (root->orphan_item_inserted &&
2170             btrfs_root_refs(&root->root_item) > 0) {
2171                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172                                             root->root_key.objectid);
2173                 BUG_ON(ret);
2174                 root->orphan_item_inserted = 0;
2175         }
2176
2177         if (block_rsv) {
2178                 WARN_ON(block_rsv->size > 0);
2179                 btrfs_free_block_rsv(root, block_rsv);
2180         }
2181 }
2182
2183 /*
2184  * This creates an orphan entry for the given inode in case something goes
2185  * wrong in the middle of an unlink/truncate.
2186  *
2187  * NOTE: caller of this function should reserve 5 units of metadata for
2188  *       this function.
2189  */
2190 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2191 {
2192         struct btrfs_root *root = BTRFS_I(inode)->root;
2193         struct btrfs_block_rsv *block_rsv = NULL;
2194         int reserve = 0;
2195         int insert = 0;
2196         int ret;
2197
2198         if (!root->orphan_block_rsv) {
2199                 block_rsv = btrfs_alloc_block_rsv(root);
2200                 if (!block_rsv)
2201                         return -ENOMEM;
2202         }
2203
2204         spin_lock(&root->orphan_lock);
2205         if (!root->orphan_block_rsv) {
2206                 root->orphan_block_rsv = block_rsv;
2207         } else if (block_rsv) {
2208                 btrfs_free_block_rsv(root, block_rsv);
2209                 block_rsv = NULL;
2210         }
2211
2212         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2213                               &BTRFS_I(inode)->runtime_flags)) {
2214 #if 0
2215                 /*
2216                  * For proper ENOSPC handling, we should do orphan
2217                  * cleanup when mounting. But this introduces backward
2218                  * compatibility issue.
2219                  */
2220                 if (!xchg(&root->orphan_item_inserted, 1))
2221                         insert = 2;
2222                 else
2223                         insert = 1;
2224 #endif
2225                 insert = 1;
2226                 atomic_dec(&root->orphan_inodes);
2227         }
2228
2229         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2230                               &BTRFS_I(inode)->runtime_flags))
2231                 reserve = 1;
2232         spin_unlock(&root->orphan_lock);
2233
2234         /* grab metadata reservation from transaction handle */
2235         if (reserve) {
2236                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2237                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2238         }
2239
2240         /* insert an orphan item to track this unlinked/truncated file */
2241         if (insert >= 1) {
2242                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2243                 if (ret && ret != -EEXIST) {
2244                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2245                                   &BTRFS_I(inode)->runtime_flags);
2246                         btrfs_abort_transaction(trans, root, ret);
2247                         return ret;
2248                 }
2249                 ret = 0;
2250         }
2251
2252         /* insert an orphan item to track subvolume contains orphan files */
2253         if (insert >= 2) {
2254                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2255                                                root->root_key.objectid);
2256                 if (ret && ret != -EEXIST) {
2257                         btrfs_abort_transaction(trans, root, ret);
2258                         return ret;
2259                 }
2260         }
2261         return 0;
2262 }
2263
2264 /*
2265  * We have done the truncate/delete so we can go ahead and remove the orphan
2266  * item for this particular inode.
2267  */
2268 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2269 {
2270         struct btrfs_root *root = BTRFS_I(inode)->root;
2271         int delete_item = 0;
2272         int release_rsv = 0;
2273         int ret = 0;
2274
2275         spin_lock(&root->orphan_lock);
2276         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2277                                &BTRFS_I(inode)->runtime_flags))
2278                 delete_item = 1;
2279
2280         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2281                                &BTRFS_I(inode)->runtime_flags))
2282                 release_rsv = 1;
2283         spin_unlock(&root->orphan_lock);
2284
2285         if (trans && delete_item) {
2286                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2287                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2288         }
2289
2290         if (release_rsv) {
2291                 btrfs_orphan_release_metadata(inode);
2292                 atomic_dec(&root->orphan_inodes);
2293         }
2294
2295         return 0;
2296 }
2297
2298 /*
2299  * this cleans up any orphans that may be left on the list from the last use
2300  * of this root.
2301  */
2302 int btrfs_orphan_cleanup(struct btrfs_root *root)
2303 {
2304         struct btrfs_path *path;
2305         struct extent_buffer *leaf;
2306         struct btrfs_key key, found_key;
2307         struct btrfs_trans_handle *trans;
2308         struct inode *inode;
2309         u64 last_objectid = 0;
2310         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2311
2312         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2313                 return 0;
2314
2315         path = btrfs_alloc_path();
2316         if (!path) {
2317                 ret = -ENOMEM;
2318                 goto out;
2319         }
2320         path->reada = -1;
2321
2322         key.objectid = BTRFS_ORPHAN_OBJECTID;
2323         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2324         key.offset = (u64)-1;
2325
2326         while (1) {
2327                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2328                 if (ret < 0)
2329                         goto out;
2330
2331                 /*
2332                  * if ret == 0 means we found what we were searching for, which
2333                  * is weird, but possible, so only screw with path if we didn't
2334                  * find the key and see if we have stuff that matches
2335                  */
2336                 if (ret > 0) {
2337                         ret = 0;
2338                         if (path->slots[0] == 0)
2339                                 break;
2340                         path->slots[0]--;
2341                 }
2342
2343                 /* pull out the item */
2344                 leaf = path->nodes[0];
2345                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2346
2347                 /* make sure the item matches what we want */
2348                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2349                         break;
2350                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2351                         break;
2352
2353                 /* release the path since we're done with it */
2354                 btrfs_release_path(path);
2355
2356                 /*
2357                  * this is where we are basically btrfs_lookup, without the
2358                  * crossing root thing.  we store the inode number in the
2359                  * offset of the orphan item.
2360                  */
2361
2362                 if (found_key.offset == last_objectid) {
2363                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2364                                "stopping orphan cleanup\n");
2365                         ret = -EINVAL;
2366                         goto out;
2367                 }
2368
2369                 last_objectid = found_key.offset;
2370
2371                 found_key.objectid = found_key.offset;
2372                 found_key.type = BTRFS_INODE_ITEM_KEY;
2373                 found_key.offset = 0;
2374                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2375                 ret = PTR_RET(inode);
2376                 if (ret && ret != -ESTALE)
2377                         goto out;
2378
2379                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2380                         struct btrfs_root *dead_root;
2381                         struct btrfs_fs_info *fs_info = root->fs_info;
2382                         int is_dead_root = 0;
2383
2384                         /*
2385                          * this is an orphan in the tree root. Currently these
2386                          * could come from 2 sources:
2387                          *  a) a snapshot deletion in progress
2388                          *  b) a free space cache inode
2389                          * We need to distinguish those two, as the snapshot
2390                          * orphan must not get deleted.
2391                          * find_dead_roots already ran before us, so if this
2392                          * is a snapshot deletion, we should find the root
2393                          * in the dead_roots list
2394                          */
2395                         spin_lock(&fs_info->trans_lock);
2396                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2397                                             root_list) {
2398                                 if (dead_root->root_key.objectid ==
2399                                     found_key.objectid) {
2400                                         is_dead_root = 1;
2401                                         break;
2402                                 }
2403                         }
2404                         spin_unlock(&fs_info->trans_lock);
2405                         if (is_dead_root) {
2406                                 /* prevent this orphan from being found again */
2407                                 key.offset = found_key.objectid - 1;
2408                                 continue;
2409                         }
2410                 }
2411                 /*
2412                  * Inode is already gone but the orphan item is still there,
2413                  * kill the orphan item.
2414                  */
2415                 if (ret == -ESTALE) {
2416                         trans = btrfs_start_transaction(root, 1);
2417                         if (IS_ERR(trans)) {
2418                                 ret = PTR_ERR(trans);
2419                                 goto out;
2420                         }
2421                         printk(KERN_ERR "auto deleting %Lu\n",
2422                                found_key.objectid);
2423                         ret = btrfs_del_orphan_item(trans, root,
2424                                                     found_key.objectid);
2425                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2426                         btrfs_end_transaction(trans, root);
2427                         continue;
2428                 }
2429
2430                 /*
2431                  * add this inode to the orphan list so btrfs_orphan_del does
2432                  * the proper thing when we hit it
2433                  */
2434                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2435                         &BTRFS_I(inode)->runtime_flags);
2436
2437                 /* if we have links, this was a truncate, lets do that */
2438                 if (inode->i_nlink) {
2439                         if (!S_ISREG(inode->i_mode)) {
2440                                 WARN_ON(1);
2441                                 iput(inode);
2442                                 continue;
2443                         }
2444                         nr_truncate++;
2445                         ret = btrfs_truncate(inode);
2446                 } else {
2447                         nr_unlink++;
2448                 }
2449
2450                 /* this will do delete_inode and everything for us */
2451                 iput(inode);
2452                 if (ret)
2453                         goto out;
2454         }
2455         /* release the path since we're done with it */
2456         btrfs_release_path(path);
2457
2458         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2459
2460         if (root->orphan_block_rsv)
2461                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2462                                         (u64)-1);
2463
2464         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2465                 trans = btrfs_join_transaction(root);
2466                 if (!IS_ERR(trans))
2467                         btrfs_end_transaction(trans, root);
2468         }
2469
2470         if (nr_unlink)
2471                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2472         if (nr_truncate)
2473                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2474
2475 out:
2476         if (ret)
2477                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2478         btrfs_free_path(path);
2479         return ret;
2480 }
2481
2482 /*
2483  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2484  * don't find any xattrs, we know there can't be any acls.
2485  *
2486  * slot is the slot the inode is in, objectid is the objectid of the inode
2487  */
2488 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2489                                           int slot, u64 objectid)
2490 {
2491         u32 nritems = btrfs_header_nritems(leaf);
2492         struct btrfs_key found_key;
2493         int scanned = 0;
2494
2495         slot++;
2496         while (slot < nritems) {
2497                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2498
2499                 /* we found a different objectid, there must not be acls */
2500                 if (found_key.objectid != objectid)
2501                         return 0;
2502
2503                 /* we found an xattr, assume we've got an acl */
2504                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2505                         return 1;
2506
2507                 /*
2508                  * we found a key greater than an xattr key, there can't
2509                  * be any acls later on
2510                  */
2511                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2512                         return 0;
2513
2514                 slot++;
2515                 scanned++;
2516
2517                 /*
2518                  * it goes inode, inode backrefs, xattrs, extents,
2519                  * so if there are a ton of hard links to an inode there can
2520                  * be a lot of backrefs.  Don't waste time searching too hard,
2521                  * this is just an optimization
2522                  */
2523                 if (scanned >= 8)
2524                         break;
2525         }
2526         /* we hit the end of the leaf before we found an xattr or
2527          * something larger than an xattr.  We have to assume the inode
2528          * has acls
2529          */
2530         return 1;
2531 }
2532
2533 /*
2534  * read an inode from the btree into the in-memory inode
2535  */
2536 static void btrfs_read_locked_inode(struct inode *inode)
2537 {
2538         struct btrfs_path *path;
2539         struct extent_buffer *leaf;
2540         struct btrfs_inode_item *inode_item;
2541         struct btrfs_timespec *tspec;
2542         struct btrfs_root *root = BTRFS_I(inode)->root;
2543         struct btrfs_key location;
2544         int maybe_acls;
2545         u32 rdev;
2546         int ret;
2547         bool filled = false;
2548
2549         ret = btrfs_fill_inode(inode, &rdev);
2550         if (!ret)
2551                 filled = true;
2552
2553         path = btrfs_alloc_path();
2554         if (!path)
2555                 goto make_bad;
2556
2557         path->leave_spinning = 1;
2558         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2559
2560         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2561         if (ret)
2562                 goto make_bad;
2563
2564         leaf = path->nodes[0];
2565
2566         if (filled)
2567                 goto cache_acl;
2568
2569         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2570                                     struct btrfs_inode_item);
2571         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2572         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2573         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2574         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2575         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2576
2577         tspec = btrfs_inode_atime(inode_item);
2578         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2579         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2580
2581         tspec = btrfs_inode_mtime(inode_item);
2582         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584
2585         tspec = btrfs_inode_ctime(inode_item);
2586         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588
2589         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2590         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2591         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2592         inode->i_generation = BTRFS_I(inode)->generation;
2593         inode->i_rdev = 0;
2594         rdev = btrfs_inode_rdev(leaf, inode_item);
2595
2596         BTRFS_I(inode)->index_cnt = (u64)-1;
2597         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2598 cache_acl:
2599         /*
2600          * try to precache a NULL acl entry for files that don't have
2601          * any xattrs or acls
2602          */
2603         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2604                                            btrfs_ino(inode));
2605         if (!maybe_acls)
2606                 cache_no_acl(inode);
2607
2608         btrfs_free_path(path);
2609
2610         switch (inode->i_mode & S_IFMT) {
2611         case S_IFREG:
2612                 inode->i_mapping->a_ops = &btrfs_aops;
2613                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2614                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2615                 inode->i_fop = &btrfs_file_operations;
2616                 inode->i_op = &btrfs_file_inode_operations;
2617                 break;
2618         case S_IFDIR:
2619                 inode->i_fop = &btrfs_dir_file_operations;
2620                 if (root == root->fs_info->tree_root)
2621                         inode->i_op = &btrfs_dir_ro_inode_operations;
2622                 else
2623                         inode->i_op = &btrfs_dir_inode_operations;
2624                 break;
2625         case S_IFLNK:
2626                 inode->i_op = &btrfs_symlink_inode_operations;
2627                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2628                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2629                 break;
2630         default:
2631                 inode->i_op = &btrfs_special_inode_operations;
2632                 init_special_inode(inode, inode->i_mode, rdev);
2633                 break;
2634         }
2635
2636         btrfs_update_iflags(inode);
2637         return;
2638
2639 make_bad:
2640         btrfs_free_path(path);
2641         make_bad_inode(inode);
2642 }
2643
2644 /*
2645  * given a leaf and an inode, copy the inode fields into the leaf
2646  */
2647 static void fill_inode_item(struct btrfs_trans_handle *trans,
2648                             struct extent_buffer *leaf,
2649                             struct btrfs_inode_item *item,
2650                             struct inode *inode)
2651 {
2652         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2653         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2654         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2655         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2656         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2657
2658         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2659                                inode->i_atime.tv_sec);
2660         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2661                                 inode->i_atime.tv_nsec);
2662
2663         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2664                                inode->i_mtime.tv_sec);
2665         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2666                                 inode->i_mtime.tv_nsec);
2667
2668         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2669                                inode->i_ctime.tv_sec);
2670         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2671                                 inode->i_ctime.tv_nsec);
2672
2673         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2674         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2675         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2676         btrfs_set_inode_transid(leaf, item, trans->transid);
2677         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2678         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2679         btrfs_set_inode_block_group(leaf, item, 0);
2680 }
2681
2682 /*
2683  * copy everything in the in-memory inode into the btree.
2684  */
2685 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2686                                 struct btrfs_root *root, struct inode *inode)
2687 {
2688         struct btrfs_inode_item *inode_item;
2689         struct btrfs_path *path;
2690         struct extent_buffer *leaf;
2691         int ret;
2692
2693         path = btrfs_alloc_path();
2694         if (!path)
2695                 return -ENOMEM;
2696
2697         path->leave_spinning = 1;
2698         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2699                                  1);
2700         if (ret) {
2701                 if (ret > 0)
2702                         ret = -ENOENT;
2703                 goto failed;
2704         }
2705
2706         btrfs_unlock_up_safe(path, 1);
2707         leaf = path->nodes[0];
2708         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2709                                     struct btrfs_inode_item);
2710
2711         fill_inode_item(trans, leaf, inode_item, inode);
2712         btrfs_mark_buffer_dirty(leaf);
2713         btrfs_set_inode_last_trans(trans, inode);
2714         ret = 0;
2715 failed:
2716         btrfs_free_path(path);
2717         return ret;
2718 }
2719
2720 /*
2721  * copy everything in the in-memory inode into the btree.
2722  */
2723 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2724                                 struct btrfs_root *root, struct inode *inode)
2725 {
2726         int ret;
2727
2728         /*
2729          * If the inode is a free space inode, we can deadlock during commit
2730          * if we put it into the delayed code.
2731          *
2732          * The data relocation inode should also be directly updated
2733          * without delay
2734          */
2735         if (!btrfs_is_free_space_inode(inode)
2736             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737                 btrfs_update_root_times(trans, root);
2738
2739                 ret = btrfs_delayed_update_inode(trans, root, inode);
2740                 if (!ret)
2741                         btrfs_set_inode_last_trans(trans, inode);
2742                 return ret;
2743         }
2744
2745         return btrfs_update_inode_item(trans, root, inode);
2746 }
2747
2748 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2749                                 struct btrfs_root *root, struct inode *inode)
2750 {
2751         int ret;
2752
2753         ret = btrfs_update_inode(trans, root, inode);
2754         if (ret == -ENOSPC)
2755                 return btrfs_update_inode_item(trans, root, inode);
2756         return ret;
2757 }
2758
2759 /*
2760  * unlink helper that gets used here in inode.c and in the tree logging
2761  * recovery code.  It remove a link in a directory with a given name, and
2762  * also drops the back refs in the inode to the directory
2763  */
2764 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2765                                 struct btrfs_root *root,
2766                                 struct inode *dir, struct inode *inode,
2767                                 const char *name, int name_len)
2768 {
2769         struct btrfs_path *path;
2770         int ret = 0;
2771         struct extent_buffer *leaf;
2772         struct btrfs_dir_item *di;
2773         struct btrfs_key key;
2774         u64 index;
2775         u64 ino = btrfs_ino(inode);
2776         u64 dir_ino = btrfs_ino(dir);
2777
2778         path = btrfs_alloc_path();
2779         if (!path) {
2780                 ret = -ENOMEM;
2781                 goto out;
2782         }
2783
2784         path->leave_spinning = 1;
2785         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2786                                     name, name_len, -1);
2787         if (IS_ERR(di)) {
2788                 ret = PTR_ERR(di);
2789                 goto err;
2790         }
2791         if (!di) {
2792                 ret = -ENOENT;
2793                 goto err;
2794         }
2795         leaf = path->nodes[0];
2796         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2797         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2798         if (ret)
2799                 goto err;
2800         btrfs_release_path(path);
2801
2802         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2803                                   dir_ino, &index);
2804         if (ret) {
2805                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2806                        "inode %llu parent %llu\n", name_len, name,
2807                        (unsigned long long)ino, (unsigned long long)dir_ino);
2808                 btrfs_abort_transaction(trans, root, ret);
2809                 goto err;
2810         }
2811
2812         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2813         if (ret) {
2814                 btrfs_abort_transaction(trans, root, ret);
2815                 goto err;
2816         }
2817
2818         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2819                                          inode, dir_ino);
2820         if (ret != 0 && ret != -ENOENT) {
2821                 btrfs_abort_transaction(trans, root, ret);
2822                 goto err;
2823         }
2824
2825         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2826                                            dir, index);
2827         if (ret == -ENOENT)
2828                 ret = 0;
2829 err:
2830         btrfs_free_path(path);
2831         if (ret)
2832                 goto out;
2833
2834         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2835         inode_inc_iversion(inode);
2836         inode_inc_iversion(dir);
2837         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2838         ret = btrfs_update_inode(trans, root, dir);
2839 out:
2840         return ret;
2841 }
2842
2843 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2844                        struct btrfs_root *root,
2845                        struct inode *dir, struct inode *inode,
2846                        const char *name, int name_len)
2847 {
2848         int ret;
2849         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2850         if (!ret) {
2851                 btrfs_drop_nlink(inode);
2852                 ret = btrfs_update_inode(trans, root, inode);
2853         }
2854         return ret;
2855 }
2856                 
2857
2858 /* helper to check if there is any shared block in the path */
2859 static int check_path_shared(struct btrfs_root *root,
2860                              struct btrfs_path *path)
2861 {
2862         struct extent_buffer *eb;
2863         int level;
2864         u64 refs = 1;
2865
2866         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2867                 int ret;
2868
2869                 if (!path->nodes[level])
2870                         break;
2871                 eb = path->nodes[level];
2872                 if (!btrfs_block_can_be_shared(root, eb))
2873                         continue;
2874                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2875                                                &refs, NULL);
2876                 if (refs > 1)
2877                         return 1;
2878         }
2879         return 0;
2880 }
2881
2882 /*
2883  * helper to start transaction for unlink and rmdir.
2884  *
2885  * unlink and rmdir are special in btrfs, they do not always free space.
2886  * so in enospc case, we should make sure they will free space before
2887  * allowing them to use the global metadata reservation.
2888  */
2889 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2890                                                        struct dentry *dentry)
2891 {
2892         struct btrfs_trans_handle *trans;
2893         struct btrfs_root *root = BTRFS_I(dir)->root;
2894         struct btrfs_path *path;
2895         struct btrfs_inode_ref *ref;
2896         struct btrfs_dir_item *di;
2897         struct inode *inode = dentry->d_inode;
2898         u64 index;
2899         int check_link = 1;
2900         int err = -ENOSPC;
2901         int ret;
2902         u64 ino = btrfs_ino(inode);
2903         u64 dir_ino = btrfs_ino(dir);
2904
2905         /*
2906          * 1 for the possible orphan item
2907          * 1 for the dir item
2908          * 1 for the dir index
2909          * 1 for the inode ref
2910          * 1 for the inode ref in the tree log
2911          * 2 for the dir entries in the log
2912          * 1 for the inode
2913          */
2914         trans = btrfs_start_transaction(root, 8);
2915         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2916                 return trans;
2917
2918         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2919                 return ERR_PTR(-ENOSPC);
2920
2921         /* check if there is someone else holds reference */
2922         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2923                 return ERR_PTR(-ENOSPC);
2924
2925         if (atomic_read(&inode->i_count) > 2)
2926                 return ERR_PTR(-ENOSPC);
2927
2928         if (xchg(&root->fs_info->enospc_unlink, 1))
2929                 return ERR_PTR(-ENOSPC);
2930
2931         path = btrfs_alloc_path();
2932         if (!path) {
2933                 root->fs_info->enospc_unlink = 0;
2934                 return ERR_PTR(-ENOMEM);
2935         }
2936
2937         /* 1 for the orphan item */
2938         trans = btrfs_start_transaction(root, 1);
2939         if (IS_ERR(trans)) {
2940                 btrfs_free_path(path);
2941                 root->fs_info->enospc_unlink = 0;
2942                 return trans;
2943         }
2944
2945         path->skip_locking = 1;
2946         path->search_commit_root = 1;
2947
2948         ret = btrfs_lookup_inode(trans, root, path,
2949                                 &BTRFS_I(dir)->location, 0);
2950         if (ret < 0) {
2951                 err = ret;
2952                 goto out;
2953         }
2954         if (ret == 0) {
2955                 if (check_path_shared(root, path))
2956                         goto out;
2957         } else {
2958                 check_link = 0;
2959         }
2960         btrfs_release_path(path);
2961
2962         ret = btrfs_lookup_inode(trans, root, path,
2963                                 &BTRFS_I(inode)->location, 0);
2964         if (ret < 0) {
2965                 err = ret;
2966                 goto out;
2967         }
2968         if (ret == 0) {
2969                 if (check_path_shared(root, path))
2970                         goto out;
2971         } else {
2972                 check_link = 0;
2973         }
2974         btrfs_release_path(path);
2975
2976         if (ret == 0 && S_ISREG(inode->i_mode)) {
2977                 ret = btrfs_lookup_file_extent(trans, root, path,
2978                                                ino, (u64)-1, 0);
2979                 if (ret < 0) {
2980                         err = ret;
2981                         goto out;
2982                 }
2983                 BUG_ON(ret == 0); /* Corruption */
2984                 if (check_path_shared(root, path))
2985                         goto out;
2986                 btrfs_release_path(path);
2987         }
2988
2989         if (!check_link) {
2990                 err = 0;
2991                 goto out;
2992         }
2993
2994         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2995                                 dentry->d_name.name, dentry->d_name.len, 0);
2996         if (IS_ERR(di)) {
2997                 err = PTR_ERR(di);
2998                 goto out;
2999         }
3000         if (di) {
3001                 if (check_path_shared(root, path))
3002                         goto out;
3003         } else {
3004                 err = 0;
3005                 goto out;
3006         }
3007         btrfs_release_path(path);
3008
3009         ref = btrfs_lookup_inode_ref(trans, root, path,
3010                                 dentry->d_name.name, dentry->d_name.len,
3011                                 ino, dir_ino, 0);
3012         if (IS_ERR(ref)) {
3013                 err = PTR_ERR(ref);
3014                 goto out;
3015         }
3016         BUG_ON(!ref); /* Logic error */
3017         if (check_path_shared(root, path))
3018                 goto out;
3019         index = btrfs_inode_ref_index(path->nodes[0], ref);
3020         btrfs_release_path(path);
3021
3022         /*
3023          * This is a commit root search, if we can lookup inode item and other
3024          * relative items in the commit root, it means the transaction of
3025          * dir/file creation has been committed, and the dir index item that we
3026          * delay to insert has also been inserted into the commit root. So
3027          * we needn't worry about the delayed insertion of the dir index item
3028          * here.
3029          */
3030         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3031                                 dentry->d_name.name, dentry->d_name.len, 0);
3032         if (IS_ERR(di)) {
3033                 err = PTR_ERR(di);
3034                 goto out;
3035         }
3036         BUG_ON(ret == -ENOENT);
3037         if (check_path_shared(root, path))
3038                 goto out;
3039
3040         err = 0;
3041 out:
3042         btrfs_free_path(path);
3043         /* Migrate the orphan reservation over */
3044         if (!err)
3045                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3046                                 &root->fs_info->global_block_rsv,
3047                                 trans->bytes_reserved);
3048
3049         if (err) {
3050                 btrfs_end_transaction(trans, root);
3051                 root->fs_info->enospc_unlink = 0;
3052                 return ERR_PTR(err);
3053         }
3054
3055         trans->block_rsv = &root->fs_info->global_block_rsv;
3056         return trans;
3057 }
3058
3059 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3060                                struct btrfs_root *root)
3061 {
3062         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3063                 btrfs_block_rsv_release(root, trans->block_rsv,
3064                                         trans->bytes_reserved);
3065                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3066                 BUG_ON(!root->fs_info->enospc_unlink);
3067                 root->fs_info->enospc_unlink = 0;
3068         }
3069         btrfs_end_transaction(trans, root);
3070 }
3071
3072 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3073 {
3074         struct btrfs_root *root = BTRFS_I(dir)->root;
3075         struct btrfs_trans_handle *trans;
3076         struct inode *inode = dentry->d_inode;
3077         int ret;
3078         unsigned long nr = 0;
3079
3080         trans = __unlink_start_trans(dir, dentry);
3081         if (IS_ERR(trans))
3082                 return PTR_ERR(trans);
3083
3084         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3085
3086         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3087                                  dentry->d_name.name, dentry->d_name.len);
3088         if (ret)
3089                 goto out;
3090
3091         if (inode->i_nlink == 0) {
3092                 ret = btrfs_orphan_add(trans, inode);
3093                 if (ret)
3094                         goto out;
3095         }
3096
3097 out:
3098         nr = trans->blocks_used;
3099         __unlink_end_trans(trans, root);
3100         btrfs_btree_balance_dirty(root, nr);
3101         return ret;
3102 }
3103
3104 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3105                         struct btrfs_root *root,
3106                         struct inode *dir, u64 objectid,
3107                         const char *name, int name_len)
3108 {
3109         struct btrfs_path *path;
3110         struct extent_buffer *leaf;
3111         struct btrfs_dir_item *di;
3112         struct btrfs_key key;
3113         u64 index;
3114         int ret;
3115         u64 dir_ino = btrfs_ino(dir);
3116
3117         path = btrfs_alloc_path();
3118         if (!path)
3119                 return -ENOMEM;
3120
3121         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3122                                    name, name_len, -1);
3123         if (IS_ERR_OR_NULL(di)) {
3124                 if (!di)
3125                         ret = -ENOENT;
3126                 else
3127                         ret = PTR_ERR(di);
3128                 goto out;
3129         }
3130
3131         leaf = path->nodes[0];
3132         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3133         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3134         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3135         if (ret) {
3136                 btrfs_abort_transaction(trans, root, ret);
3137                 goto out;
3138         }
3139         btrfs_release_path(path);
3140