Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / fs / btrfs / file.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/fs.h>
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/version.h>
33 #include "ctree.h"
34 #include "disk-io.h"
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "ioctl.h"
38 #include "print-tree.h"
39 #include "tree-log.h"
40 #include "locking.h"
41 #include "compat.h"
42
43
44 /* simple helper to fault in pages and copy.  This should go away
45  * and be replaced with calls into generic code.
46  */
47 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
48                                          int write_bytes,
49                                          struct page **prepared_pages,
50                                          const char __user *buf)
51 {
52         long page_fault = 0;
53         int i;
54         int offset = pos & (PAGE_CACHE_SIZE - 1);
55
56         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
57                 size_t count = min_t(size_t,
58                                      PAGE_CACHE_SIZE - offset, write_bytes);
59                 struct page *page = prepared_pages[i];
60                 fault_in_pages_readable(buf, count);
61
62                 /* Copy data from userspace to the current page */
63                 kmap(page);
64                 page_fault = __copy_from_user(page_address(page) + offset,
65                                               buf, count);
66                 /* Flush processor's dcache for this page */
67                 flush_dcache_page(page);
68                 kunmap(page);
69                 buf += count;
70                 write_bytes -= count;
71
72                 if (page_fault)
73                         break;
74         }
75         return page_fault ? -EFAULT : 0;
76 }
77
78 /*
79  * unlocks pages after btrfs_file_write is done with them
80  */
81 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
82 {
83         size_t i;
84         for (i = 0; i < num_pages; i++) {
85                 if (!pages[i])
86                         break;
87                 /* page checked is some magic around finding pages that
88                  * have been modified without going through btrfs_set_page_dirty
89                  * clear it here
90                  */
91                 ClearPageChecked(pages[i]);
92                 unlock_page(pages[i]);
93                 mark_page_accessed(pages[i]);
94                 page_cache_release(pages[i]);
95         }
96 }
97
98 /*
99  * after copy_from_user, pages need to be dirtied and we need to make
100  * sure holes are created between the current EOF and the start of
101  * any next extents (if required).
102  *
103  * this also makes the decision about creating an inline extent vs
104  * doing real data extents, marking pages dirty and delalloc as required.
105  */
106 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
107                                    struct btrfs_root *root,
108                                    struct file *file,
109                                    struct page **pages,
110                                    size_t num_pages,
111                                    loff_t pos,
112                                    size_t write_bytes)
113 {
114         int err = 0;
115         int i;
116         struct inode *inode = fdentry(file)->d_inode;
117         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
118         u64 hint_byte;
119         u64 num_bytes;
120         u64 start_pos;
121         u64 end_of_last_block;
122         u64 end_pos = pos + write_bytes;
123         loff_t isize = i_size_read(inode);
124
125         start_pos = pos & ~((u64)root->sectorsize - 1);
126         num_bytes = (write_bytes + pos - start_pos +
127                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
128
129         end_of_last_block = start_pos + num_bytes - 1;
130
131         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
132         trans = btrfs_join_transaction(root, 1);
133         if (!trans) {
134                 err = -ENOMEM;
135                 goto out_unlock;
136         }
137         btrfs_set_trans_block_group(trans, inode);
138         hint_byte = 0;
139
140         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
141
142         /* check for reserved extents on each page, we don't want
143          * to reset the delalloc bit on things that already have
144          * extents reserved.
145          */
146         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
147         for (i = 0; i < num_pages; i++) {
148                 struct page *p = pages[i];
149                 SetPageUptodate(p);
150                 ClearPageChecked(p);
151                 set_page_dirty(p);
152         }
153         if (end_pos > isize) {
154                 i_size_write(inode, end_pos);
155                 btrfs_update_inode(trans, root, inode);
156         }
157         err = btrfs_end_transaction(trans, root);
158 out_unlock:
159         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
160         return err;
161 }
162
163 /*
164  * this drops all the extents in the cache that intersect the range
165  * [start, end].  Existing extents are split as required.
166  */
167 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
168                             int skip_pinned)
169 {
170         struct extent_map *em;
171         struct extent_map *split = NULL;
172         struct extent_map *split2 = NULL;
173         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
174         u64 len = end - start + 1;
175         int ret;
176         int testend = 1;
177         unsigned long flags;
178         int compressed = 0;
179
180         WARN_ON(end < start);
181         if (end == (u64)-1) {
182                 len = (u64)-1;
183                 testend = 0;
184         }
185         while (1) {
186                 if (!split)
187                         split = alloc_extent_map(GFP_NOFS);
188                 if (!split2)
189                         split2 = alloc_extent_map(GFP_NOFS);
190
191                 spin_lock(&em_tree->lock);
192                 em = lookup_extent_mapping(em_tree, start, len);
193                 if (!em) {
194                         spin_unlock(&em_tree->lock);
195                         break;
196                 }
197                 flags = em->flags;
198                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
199                         spin_unlock(&em_tree->lock);
200                         if (em->start <= start &&
201                             (!testend || em->start + em->len >= start + len)) {
202                                 free_extent_map(em);
203                                 break;
204                         }
205                         if (start < em->start) {
206                                 len = em->start - start;
207                         } else {
208                                 len = start + len - (em->start + em->len);
209                                 start = em->start + em->len;
210                         }
211                         free_extent_map(em);
212                         continue;
213                 }
214                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
215                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
216                 remove_extent_mapping(em_tree, em);
217
218                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
219                     em->start < start) {
220                         split->start = em->start;
221                         split->len = start - em->start;
222                         split->orig_start = em->orig_start;
223                         split->block_start = em->block_start;
224
225                         if (compressed)
226                                 split->block_len = em->block_len;
227                         else
228                                 split->block_len = split->len;
229
230                         split->bdev = em->bdev;
231                         split->flags = flags;
232                         ret = add_extent_mapping(em_tree, split);
233                         BUG_ON(ret);
234                         free_extent_map(split);
235                         split = split2;
236                         split2 = NULL;
237                 }
238                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
239                     testend && em->start + em->len > start + len) {
240                         u64 diff = start + len - em->start;
241
242                         split->start = start + len;
243                         split->len = em->start + em->len - (start + len);
244                         split->bdev = em->bdev;
245                         split->flags = flags;
246
247                         if (compressed) {
248                                 split->block_len = em->block_len;
249                                 split->block_start = em->block_start;
250                                 split->orig_start = em->orig_start;
251                         } else {
252                                 split->block_len = split->len;
253                                 split->block_start = em->block_start + diff;
254                                 split->orig_start = split->start;
255                         }
256
257                         ret = add_extent_mapping(em_tree, split);
258                         BUG_ON(ret);
259                         free_extent_map(split);
260                         split = NULL;
261                 }
262                 spin_unlock(&em_tree->lock);
263
264                 /* once for us */
265                 free_extent_map(em);
266                 /* once for the tree*/
267                 free_extent_map(em);
268         }
269         if (split)
270                 free_extent_map(split);
271         if (split2)
272                 free_extent_map(split2);
273         return 0;
274 }
275
276 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
277 {
278         return 0;
279 #if 0
280         struct btrfs_path *path;
281         struct btrfs_key found_key;
282         struct extent_buffer *leaf;
283         struct btrfs_file_extent_item *extent;
284         u64 last_offset = 0;
285         int nritems;
286         int slot;
287         int found_type;
288         int ret;
289         int err = 0;
290         u64 extent_end = 0;
291
292         path = btrfs_alloc_path();
293         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
294                                        last_offset, 0);
295         while (1) {
296                 nritems = btrfs_header_nritems(path->nodes[0]);
297                 if (path->slots[0] >= nritems) {
298                         ret = btrfs_next_leaf(root, path);
299                         if (ret)
300                                 goto out;
301                         nritems = btrfs_header_nritems(path->nodes[0]);
302                 }
303                 slot = path->slots[0];
304                 leaf = path->nodes[0];
305                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
306                 if (found_key.objectid != inode->i_ino)
307                         break;
308                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
309                         goto out;
310
311                 if (found_key.offset < last_offset) {
312                         WARN_ON(1);
313                         btrfs_print_leaf(root, leaf);
314                         printk(KERN_ERR "inode %lu found offset %llu "
315                                "expected %llu\n", inode->i_ino,
316                                (unsigned long long)found_key.offset,
317                                (unsigned long long)last_offset);
318                         err = 1;
319                         goto out;
320                 }
321                 extent = btrfs_item_ptr(leaf, slot,
322                                         struct btrfs_file_extent_item);
323                 found_type = btrfs_file_extent_type(leaf, extent);
324                 if (found_type == BTRFS_FILE_EXTENT_REG) {
325                         extent_end = found_key.offset +
326                              btrfs_file_extent_num_bytes(leaf, extent);
327                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
328                         struct btrfs_item *item;
329                         item = btrfs_item_nr(leaf, slot);
330                         extent_end = found_key.offset +
331                              btrfs_file_extent_inline_len(leaf, extent);
332                         extent_end = (extent_end + root->sectorsize - 1) &
333                                 ~((u64)root->sectorsize - 1);
334                 }
335                 last_offset = extent_end;
336                 path->slots[0]++;
337         }
338         if (0 && last_offset < inode->i_size) {
339                 WARN_ON(1);
340                 btrfs_print_leaf(root, leaf);
341                 printk(KERN_ERR "inode %lu found offset %llu size %llu\n",
342                        inode->i_ino, (unsigned long long)last_offset,
343                        (unsigned long long)inode->i_size);
344                 err = 1;
345
346         }
347 out:
348         btrfs_free_path(path);
349         return err;
350 #endif
351 }
352
353 /*
354  * this is very complex, but the basic idea is to drop all extents
355  * in the range start - end.  hint_block is filled in with a block number
356  * that would be a good hint to the block allocator for this file.
357  *
358  * If an extent intersects the range but is not entirely inside the range
359  * it is either truncated or split.  Anything entirely inside the range
360  * is deleted from the tree.
361  *
362  * inline_limit is used to tell this code which offsets in the file to keep
363  * if they contain inline extents.
364  */
365 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
366                        struct btrfs_root *root, struct inode *inode,
367                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
368 {
369         u64 extent_end = 0;
370         u64 locked_end = end;
371         u64 search_start = start;
372         u64 leaf_start;
373         u64 ram_bytes = 0;
374         u64 orig_parent = 0;
375         u64 disk_bytenr = 0;
376         u8 compression;
377         u8 encryption;
378         u16 other_encoding = 0;
379         u64 root_gen;
380         u64 root_owner;
381         struct extent_buffer *leaf;
382         struct btrfs_file_extent_item *extent;
383         struct btrfs_path *path;
384         struct btrfs_key key;
385         struct btrfs_file_extent_item old;
386         int keep;
387         int slot;
388         int bookend;
389         int found_type = 0;
390         int found_extent;
391         int found_inline;
392         int recow;
393         int ret;
394
395         inline_limit = 0;
396         btrfs_drop_extent_cache(inode, start, end - 1, 0);
397
398         path = btrfs_alloc_path();
399         if (!path)
400                 return -ENOMEM;
401         while (1) {
402                 recow = 0;
403                 btrfs_release_path(root, path);
404                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
405                                                search_start, -1);
406                 if (ret < 0)
407                         goto out;
408                 if (ret > 0) {
409                         if (path->slots[0] == 0) {
410                                 ret = 0;
411                                 goto out;
412                         }
413                         path->slots[0]--;
414                 }
415 next_slot:
416                 keep = 0;
417                 bookend = 0;
418                 found_extent = 0;
419                 found_inline = 0;
420                 leaf_start = 0;
421                 root_gen = 0;
422                 root_owner = 0;
423                 compression = 0;
424                 encryption = 0;
425                 extent = NULL;
426                 leaf = path->nodes[0];
427                 slot = path->slots[0];
428                 ret = 0;
429                 btrfs_item_key_to_cpu(leaf, &key, slot);
430                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
431                     key.offset >= end) {
432                         goto out;
433                 }
434                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
435                     key.objectid != inode->i_ino) {
436                         goto out;
437                 }
438                 if (recow) {
439                         search_start = max(key.offset, start);
440                         continue;
441                 }
442                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
443                         extent = btrfs_item_ptr(leaf, slot,
444                                                 struct btrfs_file_extent_item);
445                         found_type = btrfs_file_extent_type(leaf, extent);
446                         compression = btrfs_file_extent_compression(leaf,
447                                                                     extent);
448                         encryption = btrfs_file_extent_encryption(leaf,
449                                                                   extent);
450                         other_encoding = btrfs_file_extent_other_encoding(leaf,
451                                                                   extent);
452                         if (found_type == BTRFS_FILE_EXTENT_REG ||
453                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
454                                 extent_end =
455                                      btrfs_file_extent_disk_bytenr(leaf,
456                                                                    extent);
457                                 if (extent_end)
458                                         *hint_byte = extent_end;
459
460                                 extent_end = key.offset +
461                                      btrfs_file_extent_num_bytes(leaf, extent);
462                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
463                                                                 extent);
464                                 found_extent = 1;
465                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
466                                 found_inline = 1;
467                                 extent_end = key.offset +
468                                      btrfs_file_extent_inline_len(leaf, extent);
469                         }
470                 } else {
471                         extent_end = search_start;
472                 }
473
474                 /* we found nothing we can drop */
475                 if ((!found_extent && !found_inline) ||
476                     search_start >= extent_end) {
477                         int nextret;
478                         u32 nritems;
479                         nritems = btrfs_header_nritems(leaf);
480                         if (slot >= nritems - 1) {
481                                 nextret = btrfs_next_leaf(root, path);
482                                 if (nextret)
483                                         goto out;
484                                 recow = 1;
485                         } else {
486                                 path->slots[0]++;
487                         }
488                         goto next_slot;
489                 }
490
491                 if (end <= extent_end && start >= key.offset && found_inline)
492                         *hint_byte = EXTENT_MAP_INLINE;
493
494                 if (found_extent) {
495                         read_extent_buffer(leaf, &old, (unsigned long)extent,
496                                            sizeof(old));
497                         root_gen = btrfs_header_generation(leaf);
498                         root_owner = btrfs_header_owner(leaf);
499                         leaf_start = leaf->start;
500                 }
501
502                 if (end < extent_end && end >= key.offset) {
503                         bookend = 1;
504                         if (found_inline && start <= key.offset)
505                                 keep = 1;
506                 }
507
508                 if (bookend && found_extent) {
509                         if (locked_end < extent_end) {
510                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
511                                                 locked_end, extent_end - 1,
512                                                 GFP_NOFS);
513                                 if (!ret) {
514                                         btrfs_release_path(root, path);
515                                         lock_extent(&BTRFS_I(inode)->io_tree,
516                                                 locked_end, extent_end - 1,
517                                                 GFP_NOFS);
518                                         locked_end = extent_end;
519                                         continue;
520                                 }
521                                 locked_end = extent_end;
522                         }
523                         orig_parent = path->nodes[0]->start;
524                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
525                         if (disk_bytenr != 0) {
526                                 ret = btrfs_inc_extent_ref(trans, root,
527                                            disk_bytenr,
528                                            le64_to_cpu(old.disk_num_bytes),
529                                            orig_parent, root->root_key.objectid,
530                                            trans->transid, inode->i_ino);
531                                 BUG_ON(ret);
532                         }
533                 }
534
535                 if (found_inline) {
536                         u64 mask = root->sectorsize - 1;
537                         search_start = (extent_end + mask) & ~mask;
538                 } else
539                         search_start = extent_end;
540
541                 /* truncate existing extent */
542                 if (start > key.offset) {
543                         u64 new_num;
544                         u64 old_num;
545                         keep = 1;
546                         WARN_ON(start & (root->sectorsize - 1));
547                         if (found_extent) {
548                                 new_num = start - key.offset;
549                                 old_num = btrfs_file_extent_num_bytes(leaf,
550                                                                       extent);
551                                 *hint_byte =
552                                         btrfs_file_extent_disk_bytenr(leaf,
553                                                                       extent);
554                                 if (btrfs_file_extent_disk_bytenr(leaf,
555                                                                   extent)) {
556                                         inode_sub_bytes(inode, old_num -
557                                                         new_num);
558                                 }
559                                 btrfs_set_file_extent_num_bytes(leaf,
560                                                         extent, new_num);
561                                 btrfs_mark_buffer_dirty(leaf);
562                         } else if (key.offset < inline_limit &&
563                                    (end > extent_end) &&
564                                    (inline_limit < extent_end)) {
565                                 u32 new_size;
566                                 new_size = btrfs_file_extent_calc_inline_size(
567                                                    inline_limit - key.offset);
568                                 inode_sub_bytes(inode, extent_end -
569                                                 inline_limit);
570                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
571                                                         new_size);
572                                 if (!compression && !encryption) {
573                                         btrfs_truncate_item(trans, root, path,
574                                                             new_size, 1);
575                                 }
576                         }
577                 }
578                 /* delete the entire extent */
579                 if (!keep) {
580                         if (found_inline)
581                                 inode_sub_bytes(inode, extent_end -
582                                                 key.offset);
583                         ret = btrfs_del_item(trans, root, path);
584                         /* TODO update progress marker and return */
585                         BUG_ON(ret);
586                         extent = NULL;
587                         btrfs_release_path(root, path);
588                         /* the extent will be freed later */
589                 }
590                 if (bookend && found_inline && start <= key.offset) {
591                         u32 new_size;
592                         new_size = btrfs_file_extent_calc_inline_size(
593                                                    extent_end - end);
594                         inode_sub_bytes(inode, end - key.offset);
595                         btrfs_set_file_extent_ram_bytes(leaf, extent,
596                                                         new_size);
597                         if (!compression && !encryption)
598                                 ret = btrfs_truncate_item(trans, root, path,
599                                                           new_size, 0);
600                         BUG_ON(ret);
601                 }
602                 /* create bookend, splitting the extent in two */
603                 if (bookend && found_extent) {
604                         struct btrfs_key ins;
605                         ins.objectid = inode->i_ino;
606                         ins.offset = end;
607                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
608
609                         btrfs_release_path(root, path);
610                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
611                                                       sizeof(*extent));
612                         BUG_ON(ret);
613
614                         leaf = path->nodes[0];
615                         extent = btrfs_item_ptr(leaf, path->slots[0],
616                                                 struct btrfs_file_extent_item);
617                         write_extent_buffer(leaf, &old,
618                                             (unsigned long)extent, sizeof(old));
619
620                         btrfs_set_file_extent_compression(leaf, extent,
621                                                           compression);
622                         btrfs_set_file_extent_encryption(leaf, extent,
623                                                          encryption);
624                         btrfs_set_file_extent_other_encoding(leaf, extent,
625                                                              other_encoding);
626                         btrfs_set_file_extent_offset(leaf, extent,
627                                     le64_to_cpu(old.offset) + end - key.offset);
628                         WARN_ON(le64_to_cpu(old.num_bytes) <
629                                 (extent_end - end));
630                         btrfs_set_file_extent_num_bytes(leaf, extent,
631                                                         extent_end - end);
632
633                         /*
634                          * set the ram bytes to the size of the full extent
635                          * before splitting.  This is a worst case flag,
636                          * but its the best we can do because we don't know
637                          * how splitting affects compression
638                          */
639                         btrfs_set_file_extent_ram_bytes(leaf, extent,
640                                                         ram_bytes);
641                         btrfs_set_file_extent_type(leaf, extent, found_type);
642
643                         btrfs_mark_buffer_dirty(path->nodes[0]);
644
645                         if (disk_bytenr != 0) {
646                                 ret = btrfs_update_extent_ref(trans, root,
647                                                 disk_bytenr, orig_parent,
648                                                 leaf->start,
649                                                 root->root_key.objectid,
650                                                 trans->transid, ins.objectid);
651
652                                 BUG_ON(ret);
653                         }
654                         btrfs_release_path(root, path);
655                         if (disk_bytenr != 0)
656                                 inode_add_bytes(inode, extent_end - end);
657                 }
658
659                 if (found_extent && !keep) {
660                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
661
662                         if (old_disk_bytenr != 0) {
663                                 inode_sub_bytes(inode,
664                                                 le64_to_cpu(old.num_bytes));
665                                 ret = btrfs_free_extent(trans, root,
666                                                 old_disk_bytenr,
667                                                 le64_to_cpu(old.disk_num_bytes),
668                                                 leaf_start, root_owner,
669                                                 root_gen, key.objectid, 0);
670                                 BUG_ON(ret);
671                                 *hint_byte = old_disk_bytenr;
672                         }
673                 }
674
675                 if (search_start >= end) {
676                         ret = 0;
677                         goto out;
678                 }
679         }
680 out:
681         btrfs_free_path(path);
682         if (locked_end > end) {
683                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
684                               GFP_NOFS);
685         }
686         btrfs_check_file(root, inode);
687         return ret;
688 }
689
690 static int extent_mergeable(struct extent_buffer *leaf, int slot,
691                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
692 {
693         struct btrfs_file_extent_item *fi;
694         struct btrfs_key key;
695         u64 extent_end;
696
697         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
698                 return 0;
699
700         btrfs_item_key_to_cpu(leaf, &key, slot);
701         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
702                 return 0;
703
704         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
705         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
706             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
707             btrfs_file_extent_compression(leaf, fi) ||
708             btrfs_file_extent_encryption(leaf, fi) ||
709             btrfs_file_extent_other_encoding(leaf, fi))
710                 return 0;
711
712         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
713         if ((*start && *start != key.offset) || (*end && *end != extent_end))
714                 return 0;
715
716         *start = key.offset;
717         *end = extent_end;
718         return 1;
719 }
720
721 /*
722  * Mark extent in the range start - end as written.
723  *
724  * This changes extent type from 'pre-allocated' to 'regular'. If only
725  * part of extent is marked as written, the extent will be split into
726  * two or three.
727  */
728 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
729                               struct btrfs_root *root,
730                               struct inode *inode, u64 start, u64 end)
731 {
732         struct extent_buffer *leaf;
733         struct btrfs_path *path;
734         struct btrfs_file_extent_item *fi;
735         struct btrfs_key key;
736         u64 bytenr;
737         u64 num_bytes;
738         u64 extent_end;
739         u64 extent_offset;
740         u64 other_start;
741         u64 other_end;
742         u64 split = start;
743         u64 locked_end = end;
744         u64 orig_parent;
745         int extent_type;
746         int split_end = 1;
747         int ret;
748
749         btrfs_drop_extent_cache(inode, start, end - 1, 0);
750
751         path = btrfs_alloc_path();
752         BUG_ON(!path);
753 again:
754         key.objectid = inode->i_ino;
755         key.type = BTRFS_EXTENT_DATA_KEY;
756         if (split == start)
757                 key.offset = split;
758         else
759                 key.offset = split - 1;
760
761         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
762         if (ret > 0 && path->slots[0] > 0)
763                 path->slots[0]--;
764
765         leaf = path->nodes[0];
766         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
767         BUG_ON(key.objectid != inode->i_ino ||
768                key.type != BTRFS_EXTENT_DATA_KEY);
769         fi = btrfs_item_ptr(leaf, path->slots[0],
770                             struct btrfs_file_extent_item);
771         extent_type = btrfs_file_extent_type(leaf, fi);
772         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
773         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
774         BUG_ON(key.offset > start || extent_end < end);
775
776         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
777         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
778         extent_offset = btrfs_file_extent_offset(leaf, fi);
779
780         if (key.offset == start)
781                 split = end;
782
783         if (key.offset == start && extent_end == end) {
784                 int del_nr = 0;
785                 int del_slot = 0;
786                 u64 leaf_owner = btrfs_header_owner(leaf);
787                 u64 leaf_gen = btrfs_header_generation(leaf);
788                 other_start = end;
789                 other_end = 0;
790                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
791                                      bytenr, &other_start, &other_end)) {
792                         extent_end = other_end;
793                         del_slot = path->slots[0] + 1;
794                         del_nr++;
795                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
796                                                 leaf->start, leaf_owner,
797                                                 leaf_gen, inode->i_ino, 0);
798                         BUG_ON(ret);
799                 }
800                 other_start = 0;
801                 other_end = start;
802                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
803                                      bytenr, &other_start, &other_end)) {
804                         key.offset = other_start;
805                         del_slot = path->slots[0];
806                         del_nr++;
807                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
808                                                 leaf->start, leaf_owner,
809                                                 leaf_gen, inode->i_ino, 0);
810                         BUG_ON(ret);
811                 }
812                 split_end = 0;
813                 if (del_nr == 0) {
814                         btrfs_set_file_extent_type(leaf, fi,
815                                                    BTRFS_FILE_EXTENT_REG);
816                         goto done;
817                 }
818
819                 fi = btrfs_item_ptr(leaf, del_slot - 1,
820                                     struct btrfs_file_extent_item);
821                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
822                 btrfs_set_file_extent_num_bytes(leaf, fi,
823                                                 extent_end - key.offset);
824                 btrfs_mark_buffer_dirty(leaf);
825
826                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
827                 BUG_ON(ret);
828                 goto done;
829         } else if (split == start) {
830                 if (locked_end < extent_end) {
831                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
832                                         locked_end, extent_end - 1, GFP_NOFS);
833                         if (!ret) {
834                                 btrfs_release_path(root, path);
835                                 lock_extent(&BTRFS_I(inode)->io_tree,
836                                         locked_end, extent_end - 1, GFP_NOFS);
837                                 locked_end = extent_end;
838                                 goto again;
839                         }
840                         locked_end = extent_end;
841                 }
842                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
843                 extent_offset += split - key.offset;
844         } else  {
845                 BUG_ON(key.offset != start);
846                 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
847                                              split - key.offset);
848                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
849                 key.offset = split;
850                 btrfs_set_item_key_safe(trans, root, path, &key);
851                 extent_end = split;
852         }
853
854         if (extent_end == end) {
855                 split_end = 0;
856                 extent_type = BTRFS_FILE_EXTENT_REG;
857         }
858         if (extent_end == end && split == start) {
859                 other_start = end;
860                 other_end = 0;
861                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
862                                      bytenr, &other_start, &other_end)) {
863                         path->slots[0]++;
864                         fi = btrfs_item_ptr(leaf, path->slots[0],
865                                             struct btrfs_file_extent_item);
866                         key.offset = split;
867                         btrfs_set_item_key_safe(trans, root, path, &key);
868                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
869                         btrfs_set_file_extent_num_bytes(leaf, fi,
870                                                         other_end - split);
871                         goto done;
872                 }
873         }
874         if (extent_end == end && split == end) {
875                 other_start = 0;
876                 other_end = start;
877                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
878                                      bytenr, &other_start, &other_end)) {
879                         path->slots[0]--;
880                         fi = btrfs_item_ptr(leaf, path->slots[0],
881                                             struct btrfs_file_extent_item);
882                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
883                                                         other_start);
884                         goto done;
885                 }
886         }
887
888         btrfs_mark_buffer_dirty(leaf);
889
890         orig_parent = leaf->start;
891         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
892                                    orig_parent, root->root_key.objectid,
893                                    trans->transid, inode->i_ino);
894         BUG_ON(ret);
895         btrfs_release_path(root, path);
896
897         key.offset = start;
898         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
899         BUG_ON(ret);
900
901         leaf = path->nodes[0];
902         fi = btrfs_item_ptr(leaf, path->slots[0],
903                             struct btrfs_file_extent_item);
904         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
905         btrfs_set_file_extent_type(leaf, fi, extent_type);
906         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
907         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
908         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
909         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
910         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
911         btrfs_set_file_extent_compression(leaf, fi, 0);
912         btrfs_set_file_extent_encryption(leaf, fi, 0);
913         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
914
915         if (orig_parent != leaf->start) {
916                 ret = btrfs_update_extent_ref(trans, root, bytenr,
917                                               orig_parent, leaf->start,
918                                               root->root_key.objectid,
919                                               trans->transid, inode->i_ino);
920                 BUG_ON(ret);
921         }
922 done:
923         btrfs_mark_buffer_dirty(leaf);
924         btrfs_release_path(root, path);
925         if (split_end && split == start) {
926                 split = end;
927                 goto again;
928         }
929         if (locked_end > end) {
930                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
931                               GFP_NOFS);
932         }
933         btrfs_free_path(path);
934         return 0;
935 }
936
937 /*
938  * this gets pages into the page cache and locks them down, it also properly
939  * waits for data=ordered extents to finish before allowing the pages to be
940  * modified.
941  */
942 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
943                          struct page **pages, size_t num_pages,
944                          loff_t pos, unsigned long first_index,
945                          unsigned long last_index, size_t write_bytes)
946 {
947         int i;
948         unsigned long index = pos >> PAGE_CACHE_SHIFT;
949         struct inode *inode = fdentry(file)->d_inode;
950         int err = 0;
951         u64 start_pos;
952         u64 last_pos;
953
954         start_pos = pos & ~((u64)root->sectorsize - 1);
955         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
956
957         if (start_pos > inode->i_size) {
958                 err = btrfs_cont_expand(inode, start_pos);
959                 if (err)
960                         return err;
961         }
962
963         memset(pages, 0, num_pages * sizeof(struct page *));
964 again:
965         for (i = 0; i < num_pages; i++) {
966                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
967                 if (!pages[i]) {
968                         err = -ENOMEM;
969                         BUG_ON(1);
970                 }
971                 wait_on_page_writeback(pages[i]);
972         }
973         if (start_pos < inode->i_size) {
974                 struct btrfs_ordered_extent *ordered;
975                 lock_extent(&BTRFS_I(inode)->io_tree,
976                             start_pos, last_pos - 1, GFP_NOFS);
977                 ordered = btrfs_lookup_first_ordered_extent(inode,
978                                                             last_pos - 1);
979                 if (ordered &&
980                     ordered->file_offset + ordered->len > start_pos &&
981                     ordered->file_offset < last_pos) {
982                         btrfs_put_ordered_extent(ordered);
983                         unlock_extent(&BTRFS_I(inode)->io_tree,
984                                       start_pos, last_pos - 1, GFP_NOFS);
985                         for (i = 0; i < num_pages; i++) {
986                                 unlock_page(pages[i]);
987                                 page_cache_release(pages[i]);
988                         }
989                         btrfs_wait_ordered_range(inode, start_pos,
990                                                  last_pos - start_pos);
991                         goto again;
992                 }
993                 if (ordered)
994                         btrfs_put_ordered_extent(ordered);
995
996                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
997                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
998                                   GFP_NOFS);
999                 unlock_extent(&BTRFS_I(inode)->io_tree,
1000                               start_pos, last_pos - 1, GFP_NOFS);
1001         }
1002         for (i = 0; i < num_pages; i++) {
1003                 clear_page_dirty_for_io(pages[i]);
1004                 set_page_extent_mapped(pages[i]);
1005                 WARN_ON(!PageLocked(pages[i]));
1006         }
1007         return 0;
1008 }
1009
1010 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1011                                 size_t count, loff_t *ppos)
1012 {
1013         loff_t pos;
1014         loff_t start_pos;
1015         ssize_t num_written = 0;
1016         ssize_t err = 0;
1017         int ret = 0;
1018         struct inode *inode = fdentry(file)->d_inode;
1019         struct btrfs_root *root = BTRFS_I(inode)->root;
1020         struct page **pages = NULL;
1021         int nrptrs;
1022         struct page *pinned[2];
1023         unsigned long first_index;
1024         unsigned long last_index;
1025         int will_write;
1026
1027         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
1028                       (file->f_flags & O_DIRECT));
1029
1030         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
1031                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
1032         pinned[0] = NULL;
1033         pinned[1] = NULL;
1034
1035         pos = *ppos;
1036         start_pos = pos;
1037
1038         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1039         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1040         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1041         if (err)
1042                 goto out_nolock;
1043         if (count == 0)
1044                 goto out_nolock;
1045
1046         err = file_remove_suid(file);
1047         if (err)
1048                 goto out_nolock;
1049         file_update_time(file);
1050
1051         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1052
1053         mutex_lock(&inode->i_mutex);
1054         BTRFS_I(inode)->sequence++;
1055         first_index = pos >> PAGE_CACHE_SHIFT;
1056         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1057
1058         /*
1059          * there are lots of better ways to do this, but this code
1060          * makes sure the first and last page in the file range are
1061          * up to date and ready for cow
1062          */
1063         if ((pos & (PAGE_CACHE_SIZE - 1))) {
1064                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1065                 if (!PageUptodate(pinned[0])) {
1066                         ret = btrfs_readpage(NULL, pinned[0]);
1067                         BUG_ON(ret);
1068                         wait_on_page_locked(pinned[0]);
1069                 } else {
1070                         unlock_page(pinned[0]);
1071                 }
1072         }
1073         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1074                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1075                 if (!PageUptodate(pinned[1])) {
1076                         ret = btrfs_readpage(NULL, pinned[1]);
1077                         BUG_ON(ret);
1078                         wait_on_page_locked(pinned[1]);
1079                 } else {
1080                         unlock_page(pinned[1]);
1081                 }
1082         }
1083
1084         while (count > 0) {
1085                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1086                 size_t write_bytes = min(count, nrptrs *
1087                                         (size_t)PAGE_CACHE_SIZE -
1088                                          offset);
1089                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1090                                         PAGE_CACHE_SHIFT;
1091
1092                 WARN_ON(num_pages > nrptrs);
1093                 memset(pages, 0, sizeof(struct page *) * nrptrs);
1094
1095                 ret = btrfs_check_free_space(root, write_bytes, 0);
1096                 if (ret)
1097                         goto out;
1098
1099                 ret = prepare_pages(root, file, pages, num_pages,
1100                                     pos, first_index, last_index,
1101                                     write_bytes);
1102                 if (ret)
1103                         goto out;
1104
1105                 ret = btrfs_copy_from_user(pos, num_pages,
1106                                            write_bytes, pages, buf);
1107                 if (ret) {
1108                         btrfs_drop_pages(pages, num_pages);
1109                         goto out;
1110                 }
1111
1112                 ret = dirty_and_release_pages(NULL, root, file, pages,
1113                                               num_pages, pos, write_bytes);
1114                 btrfs_drop_pages(pages, num_pages);
1115                 if (ret)
1116                         goto out;
1117
1118                 if (will_write) {
1119                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1120                                                pos + write_bytes - 1,
1121                                                WB_SYNC_NONE);
1122                 } else {
1123                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1124                                                            num_pages);
1125                         if (num_pages <
1126                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1127                                 btrfs_btree_balance_dirty(root, 1);
1128                         btrfs_throttle(root);
1129                 }
1130
1131                 buf += write_bytes;
1132                 count -= write_bytes;
1133                 pos += write_bytes;
1134                 num_written += write_bytes;
1135
1136                 cond_resched();
1137         }
1138 out:
1139         mutex_unlock(&inode->i_mutex);
1140
1141 out_nolock:
1142         kfree(pages);
1143         if (pinned[0])
1144                 page_cache_release(pinned[0]);
1145         if (pinned[1])
1146                 page_cache_release(pinned[1]);
1147         *ppos = pos;
1148
1149         if (num_written > 0 && will_write) {
1150                 struct btrfs_trans_handle *trans;
1151
1152                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1153                 if (err)
1154                         num_written = err;
1155
1156                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1157                         trans = btrfs_start_transaction(root, 1);
1158                         ret = btrfs_log_dentry_safe(trans, root,
1159                                                     file->f_dentry);
1160                         if (ret == 0) {
1161                                 btrfs_sync_log(trans, root);
1162                                 btrfs_end_transaction(trans, root);
1163                         } else {
1164                                 btrfs_commit_transaction(trans, root);
1165                         }
1166                 }
1167                 if (file->f_flags & O_DIRECT) {
1168                         invalidate_mapping_pages(inode->i_mapping,
1169                               start_pos >> PAGE_CACHE_SHIFT,
1170                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1171                 }
1172         }
1173         current->backing_dev_info = NULL;
1174         return num_written ? num_written : err;
1175 }
1176
1177 int btrfs_release_file(struct inode *inode, struct file *filp)
1178 {
1179         if (filp->private_data)
1180                 btrfs_ioctl_trans_end(filp);
1181         return 0;
1182 }
1183
1184 /*
1185  * fsync call for both files and directories.  This logs the inode into
1186  * the tree log instead of forcing full commits whenever possible.
1187  *
1188  * It needs to call filemap_fdatawait so that all ordered extent updates are
1189  * in the metadata btree are up to date for copying to the log.
1190  *
1191  * It drops the inode mutex before doing the tree log commit.  This is an
1192  * important optimization for directories because holding the mutex prevents
1193  * new operations on the dir while we write to disk.
1194  */
1195 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1196 {
1197         struct inode *inode = dentry->d_inode;
1198         struct btrfs_root *root = BTRFS_I(inode)->root;
1199         int ret = 0;
1200         struct btrfs_trans_handle *trans;
1201
1202         /*
1203          * check the transaction that last modified this inode
1204          * and see if its already been committed
1205          */
1206         if (!BTRFS_I(inode)->last_trans)
1207                 goto out;
1208
1209         mutex_lock(&root->fs_info->trans_mutex);
1210         if (BTRFS_I(inode)->last_trans <=
1211             root->fs_info->last_trans_committed) {
1212                 BTRFS_I(inode)->last_trans = 0;
1213                 mutex_unlock(&root->fs_info->trans_mutex);
1214                 goto out;
1215         }
1216         mutex_unlock(&root->fs_info->trans_mutex);
1217
1218         root->fs_info->tree_log_batch++;
1219         filemap_fdatawrite(inode->i_mapping);
1220         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1221         root->fs_info->tree_log_batch++;
1222
1223         /*
1224          * ok we haven't committed the transaction yet, lets do a commit
1225          */
1226         if (file->private_data)
1227                 btrfs_ioctl_trans_end(file);
1228
1229         trans = btrfs_start_transaction(root, 1);
1230         if (!trans) {
1231                 ret = -ENOMEM;
1232                 goto out;
1233         }
1234
1235         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1236         if (ret < 0)
1237                 goto out;
1238
1239         /* we've logged all the items and now have a consistent
1240          * version of the file in the log.  It is possible that
1241          * someone will come in and modify the file, but that's
1242          * fine because the log is consistent on disk, and we
1243          * have references to all of the file's extents
1244          *
1245          * It is possible that someone will come in and log the
1246          * file again, but that will end up using the synchronization
1247          * inside btrfs_sync_log to keep things safe.
1248          */
1249         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1250
1251         if (ret > 0) {
1252                 ret = btrfs_commit_transaction(trans, root);
1253         } else {
1254                 btrfs_sync_log(trans, root);
1255                 ret = btrfs_end_transaction(trans, root);
1256         }
1257         mutex_lock(&file->f_dentry->d_inode->i_mutex);
1258 out:
1259         return ret > 0 ? EIO : ret;
1260 }
1261
1262 static struct vm_operations_struct btrfs_file_vm_ops = {
1263         .fault          = filemap_fault,
1264         .page_mkwrite   = btrfs_page_mkwrite,
1265 };
1266
1267 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1268 {
1269         vma->vm_ops = &btrfs_file_vm_ops;
1270         file_accessed(filp);
1271         return 0;
1272 }
1273
1274 struct file_operations btrfs_file_operations = {
1275         .llseek         = generic_file_llseek,
1276         .read           = do_sync_read,
1277         .aio_read       = generic_file_aio_read,
1278         .splice_read    = generic_file_splice_read,
1279         .write          = btrfs_file_write,
1280         .mmap           = btrfs_file_mmap,
1281         .open           = generic_file_open,
1282         .release        = btrfs_release_file,
1283         .fsync          = btrfs_sync_file,
1284         .unlocked_ioctl = btrfs_ioctl,
1285 #ifdef CONFIG_COMPAT
1286         .compat_ioctl   = btrfs_ioctl,
1287 #endif
1288 };