Btrfs: Add fallocate support v2
[safe/jmp/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 int noinline 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 void noinline 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 int noinline 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         if ((end_of_last_block & 4095) == 0) {
141                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
142         }
143         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
144
145         /* check for reserved extents on each page, we don't want
146          * to reset the delalloc bit on things that already have
147          * extents reserved.
148          */
149         btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
150         for (i = 0; i < num_pages; i++) {
151                 struct page *p = pages[i];
152                 SetPageUptodate(p);
153                 ClearPageChecked(p);
154                 set_page_dirty(p);
155         }
156         if (end_pos > isize) {
157                 i_size_write(inode, end_pos);
158                 btrfs_update_inode(trans, root, inode);
159         }
160         err = btrfs_end_transaction(trans, root);
161 out_unlock:
162         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
163         return err;
164 }
165
166 /*
167  * this drops all the extents in the cache that intersect the range
168  * [start, end].  Existing extents are split as required.
169  */
170 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
171                             int skip_pinned)
172 {
173         struct extent_map *em;
174         struct extent_map *split = NULL;
175         struct extent_map *split2 = NULL;
176         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
177         u64 len = end - start + 1;
178         int ret;
179         int testend = 1;
180         unsigned long flags;
181         int compressed = 0;
182
183         WARN_ON(end < start);
184         if (end == (u64)-1) {
185                 len = (u64)-1;
186                 testend = 0;
187         }
188         while(1) {
189                 if (!split)
190                         split = alloc_extent_map(GFP_NOFS);
191                 if (!split2)
192                         split2 = alloc_extent_map(GFP_NOFS);
193
194                 spin_lock(&em_tree->lock);
195                 em = lookup_extent_mapping(em_tree, start, len);
196                 if (!em) {
197                         spin_unlock(&em_tree->lock);
198                         break;
199                 }
200                 flags = em->flags;
201                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
202                         spin_unlock(&em_tree->lock);
203                         if (em->start <= start &&
204                             (!testend || em->start + em->len >= start + len)) {
205                                 free_extent_map(em);
206                                 break;
207                         }
208                         if (start < em->start) {
209                                 len = em->start - start;
210                         } else {
211                                 len = start + len - (em->start + em->len);
212                                 start = em->start + em->len;
213                         }
214                         free_extent_map(em);
215                         continue;
216                 }
217                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
218                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
219                 remove_extent_mapping(em_tree, em);
220
221                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
222                     em->start < start) {
223                         split->start = em->start;
224                         split->len = start - em->start;
225                         split->block_start = em->block_start;
226
227                         if (compressed)
228                                 split->block_len = em->block_len;
229                         else
230                                 split->block_len = split->len;
231
232                         split->bdev = em->bdev;
233                         split->flags = flags;
234                         ret = add_extent_mapping(em_tree, split);
235                         BUG_ON(ret);
236                         free_extent_map(split);
237                         split = split2;
238                         split2 = NULL;
239                 }
240                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
241                     testend && em->start + em->len > start + len) {
242                         u64 diff = start + len - em->start;
243
244                         split->start = start + len;
245                         split->len = em->start + em->len - (start + len);
246                         split->bdev = em->bdev;
247                         split->flags = flags;
248
249                         if (compressed) {
250                                 split->block_len = em->block_len;
251                                 split->block_start = em->block_start;
252                         } else {
253                                 split->block_len = split->len;
254                                 split->block_start = em->block_start + diff;
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("inode %lu found offset %Lu expected %Lu\n",
315                                inode->i_ino, found_key.offset, last_offset);
316                         err = 1;
317                         goto out;
318                 }
319                 extent = btrfs_item_ptr(leaf, slot,
320                                         struct btrfs_file_extent_item);
321                 found_type = btrfs_file_extent_type(leaf, extent);
322                 if (found_type == BTRFS_FILE_EXTENT_REG) {
323                         extent_end = found_key.offset +
324                              btrfs_file_extent_num_bytes(leaf, extent);
325                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
326                         struct btrfs_item *item;
327                         item = btrfs_item_nr(leaf, slot);
328                         extent_end = found_key.offset +
329                              btrfs_file_extent_inline_len(leaf, extent);
330                         extent_end = (extent_end + root->sectorsize - 1) &
331                                 ~((u64)root->sectorsize -1 );
332                 }
333                 last_offset = extent_end;
334                 path->slots[0]++;
335         }
336         if (0 && last_offset < inode->i_size) {
337                 WARN_ON(1);
338                 btrfs_print_leaf(root, leaf);
339                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
340                        last_offset, inode->i_size);
341                 err = 1;
342
343         }
344 out:
345         btrfs_free_path(path);
346         return err;
347 #endif
348 }
349
350 /*
351  * this is very complex, but the basic idea is to drop all extents
352  * in the range start - end.  hint_block is filled in with a block number
353  * that would be a good hint to the block allocator for this file.
354  *
355  * If an extent intersects the range but is not entirely inside the range
356  * it is either truncated or split.  Anything entirely inside the range
357  * is deleted from the tree.
358  *
359  * inline_limit is used to tell this code which offsets in the file to keep
360  * if they contain inline extents.
361  */
362 int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
363                        struct btrfs_root *root, struct inode *inode,
364                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
365 {
366         u64 extent_end = 0;
367         u64 locked_end = end;
368         u64 search_start = start;
369         u64 leaf_start;
370         u64 ram_bytes = 0;
371         u8 compression = 0;
372         u8 encryption = 0;
373         u16 other_encoding = 0;
374         u64 root_gen;
375         u64 root_owner;
376         struct extent_buffer *leaf;
377         struct btrfs_file_extent_item *extent;
378         struct btrfs_path *path;
379         struct btrfs_key key;
380         struct btrfs_file_extent_item old;
381         int keep;
382         int slot;
383         int bookend;
384         int found_type = 0;
385         int found_extent;
386         int found_inline;
387         int recow;
388         int ret;
389
390         inline_limit = 0;
391         btrfs_drop_extent_cache(inode, start, end - 1, 0);
392
393         path = btrfs_alloc_path();
394         if (!path)
395                 return -ENOMEM;
396         while(1) {
397                 recow = 0;
398                 btrfs_release_path(root, path);
399                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
400                                                search_start, -1);
401                 if (ret < 0)
402                         goto out;
403                 if (ret > 0) {
404                         if (path->slots[0] == 0) {
405                                 ret = 0;
406                                 goto out;
407                         }
408                         path->slots[0]--;
409                 }
410 next_slot:
411                 keep = 0;
412                 bookend = 0;
413                 found_extent = 0;
414                 found_inline = 0;
415                 leaf_start = 0;
416                 root_gen = 0;
417                 root_owner = 0;
418                 extent = NULL;
419                 leaf = path->nodes[0];
420                 slot = path->slots[0];
421                 ret = 0;
422                 btrfs_item_key_to_cpu(leaf, &key, slot);
423                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
424                     key.offset >= end) {
425                         goto out;
426                 }
427                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
428                     key.objectid != inode->i_ino) {
429                         goto out;
430                 }
431                 if (recow) {
432                         search_start = key.offset;
433                         continue;
434                 }
435                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
436                         extent = btrfs_item_ptr(leaf, slot,
437                                                 struct btrfs_file_extent_item);
438                         found_type = btrfs_file_extent_type(leaf, extent);
439                         compression = btrfs_file_extent_compression(leaf,
440                                                                     extent);
441                         encryption = btrfs_file_extent_encryption(leaf,
442                                                                   extent);
443                         other_encoding = btrfs_file_extent_other_encoding(leaf,
444                                                                   extent);
445                         if (found_type == BTRFS_FILE_EXTENT_REG ||
446                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
447                                 extent_end =
448                                      btrfs_file_extent_disk_bytenr(leaf,
449                                                                    extent);
450                                 if (extent_end)
451                                         *hint_byte = extent_end;
452
453                                 extent_end = key.offset +
454                                      btrfs_file_extent_num_bytes(leaf, extent);
455                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
456                                                                 extent);
457                                 found_extent = 1;
458                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
459                                 found_inline = 1;
460                                 extent_end = key.offset +
461                                      btrfs_file_extent_inline_len(leaf, extent);
462                         }
463                 } else {
464                         extent_end = search_start;
465                 }
466
467                 /* we found nothing we can drop */
468                 if ((!found_extent && !found_inline) ||
469                     search_start >= extent_end) {
470                         int nextret;
471                         u32 nritems;
472                         nritems = btrfs_header_nritems(leaf);
473                         if (slot >= nritems - 1) {
474                                 nextret = btrfs_next_leaf(root, path);
475                                 if (nextret)
476                                         goto out;
477                                 recow = 1;
478                         } else {
479                                 path->slots[0]++;
480                         }
481                         goto next_slot;
482                 }
483
484                 if (end <= extent_end && start >= key.offset && found_inline)
485                         *hint_byte = EXTENT_MAP_INLINE;
486
487                 if (found_extent) {
488                         read_extent_buffer(leaf, &old, (unsigned long)extent,
489                                            sizeof(old));
490                         root_gen = btrfs_header_generation(leaf);
491                         root_owner = btrfs_header_owner(leaf);
492                         leaf_start = leaf->start;
493                 }
494
495                 if (end < extent_end && end >= key.offset) {
496                         bookend = 1;
497                         if (found_inline && start <= key.offset)
498                                 keep = 1;
499                 }
500
501                 if (bookend && found_extent && locked_end < extent_end) {
502                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
503                                         locked_end, extent_end - 1, GFP_NOFS);
504                         if (!ret) {
505                                 btrfs_release_path(root, path);
506                                 lock_extent(&BTRFS_I(inode)->io_tree,
507                                         locked_end, extent_end - 1, GFP_NOFS);
508                                 locked_end = extent_end;
509                                 continue;
510                         }
511                         locked_end = extent_end;
512                 }
513
514                 if (found_inline) {
515                         u64 mask = root->sectorsize - 1;
516                         search_start = (extent_end + mask) & ~mask;
517                 } else
518                         search_start = extent_end;
519
520                 /* truncate existing extent */
521                 if (start > key.offset) {
522                         u64 new_num;
523                         u64 old_num;
524                         keep = 1;
525                         WARN_ON(start & (root->sectorsize - 1));
526                         if (found_extent) {
527                                 new_num = start - key.offset;
528                                 old_num = btrfs_file_extent_num_bytes(leaf,
529                                                                       extent);
530                                 *hint_byte =
531                                         btrfs_file_extent_disk_bytenr(leaf,
532                                                                       extent);
533                                 if (btrfs_file_extent_disk_bytenr(leaf,
534                                                                   extent)) {
535                                         inode_sub_bytes(inode, old_num -
536                                                         new_num);
537                                 }
538                                 btrfs_set_file_extent_num_bytes(leaf, extent,
539                                                                 new_num);
540                                 btrfs_mark_buffer_dirty(leaf);
541                         } else if (key.offset < inline_limit &&
542                                    (end > extent_end) &&
543                                    (inline_limit < extent_end)) {
544                                 u32 new_size;
545                                 new_size = btrfs_file_extent_calc_inline_size(
546                                                    inline_limit - key.offset);
547                                 inode_sub_bytes(inode, extent_end -
548                                                 inline_limit);
549                                 btrfs_truncate_item(trans, root, path,
550                                                     new_size, 1);
551                         }
552                 }
553                 /* delete the entire extent */
554                 if (!keep) {
555                         if (found_inline)
556                                 inode_sub_bytes(inode, extent_end -
557                                                 key.offset);
558                         ret = btrfs_del_item(trans, root, path);
559                         /* TODO update progress marker and return */
560                         BUG_ON(ret);
561                         extent = NULL;
562                         btrfs_release_path(root, path);
563                         /* the extent will be freed later */
564                 }
565                 if (bookend && found_inline && start <= key.offset) {
566                         u32 new_size;
567                         new_size = btrfs_file_extent_calc_inline_size(
568                                                    extent_end - end);
569                         inode_sub_bytes(inode, end - key.offset);
570                         ret = btrfs_truncate_item(trans, root, path,
571                                                   new_size, 0);
572                         BUG_ON(ret);
573                 }
574                 /* create bookend, splitting the extent in two */
575                 if (bookend && found_extent) {
576                         u64 disk_bytenr;
577                         struct btrfs_key ins;
578                         ins.objectid = inode->i_ino;
579                         ins.offset = end;
580                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
581                         btrfs_release_path(root, path);
582                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
583                                                       sizeof(*extent));
584                         BUG_ON(ret);
585
586                         leaf = path->nodes[0];
587                         extent = btrfs_item_ptr(leaf, path->slots[0],
588                                                 struct btrfs_file_extent_item);
589                         write_extent_buffer(leaf, &old,
590                                             (unsigned long)extent, sizeof(old));
591
592                         btrfs_set_file_extent_compression(leaf, extent,
593                                                           compression);
594                         btrfs_set_file_extent_encryption(leaf, extent,
595                                                          encryption);
596                         btrfs_set_file_extent_other_encoding(leaf, extent,
597                                                              other_encoding);
598                         btrfs_set_file_extent_offset(leaf, extent,
599                                     le64_to_cpu(old.offset) + end - key.offset);
600                         WARN_ON(le64_to_cpu(old.num_bytes) <
601                                 (extent_end - end));
602                         btrfs_set_file_extent_num_bytes(leaf, extent,
603                                                         extent_end - end);
604
605                         /*
606                          * set the ram bytes to the size of the full extent
607                          * before splitting.  This is a worst case flag,
608                          * but its the best we can do because we don't know
609                          * how splitting affects compression
610                          */
611                         btrfs_set_file_extent_ram_bytes(leaf, extent,
612                                                         ram_bytes);
613                         btrfs_set_file_extent_type(leaf, extent, found_type);
614
615                         btrfs_mark_buffer_dirty(path->nodes[0]);
616
617                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
618                         if (disk_bytenr != 0) {
619                                 ret = btrfs_inc_extent_ref(trans, root,
620                                                 disk_bytenr,
621                                                 le64_to_cpu(old.disk_num_bytes),
622                                                 leaf->start,
623                                                 root->root_key.objectid,
624                                                 trans->transid, ins.objectid);
625                                 BUG_ON(ret);
626                         }
627                         btrfs_release_path(root, path);
628                         if (disk_bytenr != 0) {
629                                 inode_add_bytes(inode, extent_end - end);
630                         }
631                 }
632
633                 if (found_extent && !keep) {
634                         u64 disk_bytenr = le64_to_cpu(old.disk_bytenr);
635
636                         if (disk_bytenr != 0) {
637                                 inode_sub_bytes(inode,
638                                                 le64_to_cpu(old.num_bytes));
639                                 ret = btrfs_free_extent(trans, root,
640                                                 disk_bytenr,
641                                                 le64_to_cpu(old.disk_num_bytes),
642                                                 leaf_start, root_owner,
643                                                 root_gen, key.objectid, 0);
644                                 BUG_ON(ret);
645                                 *hint_byte = disk_bytenr;
646                         }
647                 }
648
649                 if (search_start >= end) {
650                         ret = 0;
651                         goto out;
652                 }
653         }
654 out:
655         btrfs_free_path(path);
656         if (locked_end > end) {
657                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
658                               GFP_NOFS);
659         }
660         btrfs_check_file(root, inode);
661         return ret;
662 }
663
664 static int extent_mergeable(struct extent_buffer *leaf, int slot,
665                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
666 {
667         struct btrfs_file_extent_item *fi;
668         struct btrfs_key key;
669         u64 extent_end;
670
671         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
672                 return 0;
673
674         btrfs_item_key_to_cpu(leaf, &key, slot);
675         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
676                 return 0;
677
678         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
679         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
680             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
681             btrfs_file_extent_compression(leaf, fi) ||
682             btrfs_file_extent_encryption(leaf, fi) ||
683             btrfs_file_extent_other_encoding(leaf, fi))
684                 return 0;
685
686         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
687         if ((*start && *start != key.offset) || (*end && *end != extent_end))
688                 return 0;
689
690         *start = key.offset;
691         *end = extent_end;
692         return 1;
693 }
694
695 /*
696  * Mark extent in the range start - end as written.
697  *
698  * This changes extent type from 'pre-allocated' to 'regular'. If only
699  * part of extent is marked as written, the extent will be split into
700  * two or three.
701  */
702 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
703                               struct btrfs_root *root,
704                               struct inode *inode, u64 start, u64 end)
705 {
706         struct extent_buffer *leaf;
707         struct btrfs_path *path;
708         struct btrfs_file_extent_item *fi;
709         struct btrfs_key key;
710         u64 bytenr;
711         u64 num_bytes;
712         u64 extent_end;
713         u64 extent_offset;
714         u64 other_start;
715         u64 other_end;
716         u64 split = start;
717         u64 locked_end = end;
718         int extent_type;
719         int split_end = 1;
720         int ret;
721
722         btrfs_drop_extent_cache(inode, start, end - 1, 0);
723
724         path = btrfs_alloc_path();
725         BUG_ON(!path);
726 again:
727         key.objectid = inode->i_ino;
728         key.type = BTRFS_EXTENT_DATA_KEY;
729         if (split == start)
730                 key.offset = split;
731         else
732                 key.offset = split - 1;
733
734         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
735         if (ret > 0 && path->slots[0] > 0)
736                 path->slots[0]--;
737
738         leaf = path->nodes[0];
739         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
740         BUG_ON(key.objectid != inode->i_ino ||
741                key.type != BTRFS_EXTENT_DATA_KEY);
742         fi = btrfs_item_ptr(leaf, path->slots[0],
743                             struct btrfs_file_extent_item);
744         extent_type = btrfs_file_extent_type(leaf, fi);
745         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
746         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
747         BUG_ON(key.offset > start || extent_end < end);
748
749         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
750         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
751         extent_offset = btrfs_file_extent_offset(leaf, fi);
752
753         if (key.offset == start)
754                 split = end;
755
756         if (key.offset == start && extent_end == end) {
757                 int del_nr = 0;
758                 int del_slot = 0;
759                 u64 leaf_owner = btrfs_header_owner(leaf);
760                 u64 leaf_gen = btrfs_header_generation(leaf);
761                 other_start = end;
762                 other_end = 0;
763                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
764                                      bytenr, &other_start, &other_end)) {
765                         extent_end = other_end;
766                         del_slot = path->slots[0] + 1;
767                         del_nr++;
768                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
769                                                 leaf->start, leaf_owner,
770                                                 leaf_gen, inode->i_ino, 0);
771                         BUG_ON(ret);
772                 }
773                 other_start = 0;
774                 other_end = start;
775                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
776                                      bytenr, &other_start, &other_end)) {
777                         key.offset = other_start;
778                         del_slot = path->slots[0];
779                         del_nr++;
780                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
781                                                 leaf->start, leaf_owner,
782                                                 leaf_gen, inode->i_ino, 0);
783                         BUG_ON(ret);
784                 }
785                 split_end = 0;
786                 if (del_nr == 0) {
787                         btrfs_set_file_extent_type(leaf, fi,
788                                                    BTRFS_FILE_EXTENT_REG);
789                         goto done;
790                 }
791
792                 fi = btrfs_item_ptr(leaf, del_slot - 1,
793                                     struct btrfs_file_extent_item);
794                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
795                 btrfs_set_file_extent_num_bytes(leaf, fi,
796                                                 extent_end - key.offset);
797                 btrfs_mark_buffer_dirty(leaf);
798
799                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
800                 BUG_ON(ret);
801                 goto done;
802         } else if (split == start) {
803                 if (locked_end < extent_end) {
804                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
805                                         locked_end, extent_end - 1, GFP_NOFS);
806                         if (!ret) {
807                                 btrfs_release_path(root, path);
808                                 lock_extent(&BTRFS_I(inode)->io_tree,
809                                         locked_end, extent_end - 1, GFP_NOFS);
810                                 locked_end = extent_end;
811                                 goto again;
812                         }
813                         locked_end = extent_end;
814                 }
815                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
816                 extent_offset += split - key.offset;
817         } else  {
818                 BUG_ON(key.offset != start);
819                 btrfs_set_file_extent_offset(leaf, fi, extent_offset +
820                                              split - key.offset);
821                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
822                 key.offset = split;
823                 btrfs_set_item_key_safe(trans, root, path, &key);
824                 extent_end = split;
825         }
826
827         if (extent_end == end) {
828                 split_end = 0;
829                 extent_type = BTRFS_FILE_EXTENT_REG;
830         }
831         if (extent_end == end && split == start) {
832                 other_start = end;
833                 other_end = 0;
834                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
835                                      bytenr, &other_start, &other_end)) {
836                         path->slots[0]++;
837                         fi = btrfs_item_ptr(leaf, path->slots[0],
838                                             struct btrfs_file_extent_item);
839                         key.offset = split;
840                         btrfs_set_item_key_safe(trans, root, path, &key);
841                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
842                         btrfs_set_file_extent_num_bytes(leaf, fi,
843                                                         other_end - split);
844                         goto done;
845                 }
846         }
847         if (extent_end == end && split == end) {
848                 other_start = 0;
849                 other_end = start;
850                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
851                                      bytenr, &other_start, &other_end)) {
852                         path->slots[0]--;
853                         fi = btrfs_item_ptr(leaf, path->slots[0],
854                                             struct btrfs_file_extent_item);
855                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
856                                                         other_start);
857                         goto done;
858                 }
859         }
860
861         btrfs_mark_buffer_dirty(leaf);
862         btrfs_release_path(root, path);
863
864         key.offset = start;
865         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
866         BUG_ON(ret);
867
868         leaf = path->nodes[0];
869         fi = btrfs_item_ptr(leaf, path->slots[0],
870                             struct btrfs_file_extent_item);
871         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
872         btrfs_set_file_extent_type(leaf, fi, extent_type);
873         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
874         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
875         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
876         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
877         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
878         btrfs_set_file_extent_compression(leaf, fi, 0);
879         btrfs_set_file_extent_encryption(leaf, fi, 0);
880         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
881
882         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
883                                    leaf->start, root->root_key.objectid,
884                                    trans->transid, inode->i_ino);
885         BUG_ON(ret);
886 done:
887         btrfs_mark_buffer_dirty(leaf);
888         btrfs_release_path(root, path);
889         if (split_end && split == start) {
890                 split = end;
891                 goto again;
892         }
893         if (locked_end > end) {
894                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
895                               GFP_NOFS);
896         }
897         btrfs_free_path(path);
898         return 0;
899 }
900
901 /*
902  * this gets pages into the page cache and locks them down, it also properly
903  * waits for data=ordered extents to finish before allowing the pages to be
904  * modified.
905  */
906 static int noinline prepare_pages(struct btrfs_root *root, struct file *file,
907                          struct page **pages, size_t num_pages,
908                          loff_t pos, unsigned long first_index,
909                          unsigned long last_index, size_t write_bytes)
910 {
911         int i;
912         unsigned long index = pos >> PAGE_CACHE_SHIFT;
913         struct inode *inode = fdentry(file)->d_inode;
914         int err = 0;
915         u64 start_pos;
916         u64 last_pos;
917
918         start_pos = pos & ~((u64)root->sectorsize - 1);
919         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
920
921         if (start_pos > inode->i_size) {
922                 err = btrfs_cont_expand(inode, start_pos);
923                 if (err)
924                         return err;
925         }
926
927         memset(pages, 0, num_pages * sizeof(struct page *));
928 again:
929         for (i = 0; i < num_pages; i++) {
930                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
931                 if (!pages[i]) {
932                         err = -ENOMEM;
933                         BUG_ON(1);
934                 }
935                 wait_on_page_writeback(pages[i]);
936         }
937         if (start_pos < inode->i_size) {
938                 struct btrfs_ordered_extent *ordered;
939                 lock_extent(&BTRFS_I(inode)->io_tree,
940                             start_pos, last_pos - 1, GFP_NOFS);
941                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
942                 if (ordered &&
943                     ordered->file_offset + ordered->len > start_pos &&
944                     ordered->file_offset < last_pos) {
945                         btrfs_put_ordered_extent(ordered);
946                         unlock_extent(&BTRFS_I(inode)->io_tree,
947                                       start_pos, last_pos - 1, GFP_NOFS);
948                         for (i = 0; i < num_pages; i++) {
949                                 unlock_page(pages[i]);
950                                 page_cache_release(pages[i]);
951                         }
952                         btrfs_wait_ordered_range(inode, start_pos,
953                                                  last_pos - start_pos);
954                         goto again;
955                 }
956                 if (ordered)
957                         btrfs_put_ordered_extent(ordered);
958
959                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
960                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
961                                   GFP_NOFS);
962                 unlock_extent(&BTRFS_I(inode)->io_tree,
963                               start_pos, last_pos - 1, GFP_NOFS);
964         }
965         for (i = 0; i < num_pages; i++) {
966                 clear_page_dirty_for_io(pages[i]);
967                 set_page_extent_mapped(pages[i]);
968                 WARN_ON(!PageLocked(pages[i]));
969         }
970         return 0;
971 }
972
973 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
974                                 size_t count, loff_t *ppos)
975 {
976         loff_t pos;
977         loff_t start_pos;
978         ssize_t num_written = 0;
979         ssize_t err = 0;
980         int ret = 0;
981         struct inode *inode = fdentry(file)->d_inode;
982         struct btrfs_root *root = BTRFS_I(inode)->root;
983         struct page **pages = NULL;
984         int nrptrs;
985         struct page *pinned[2];
986         unsigned long first_index;
987         unsigned long last_index;
988         int will_write;
989
990         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
991                       (file->f_flags & O_DIRECT));
992
993         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
994                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
995         pinned[0] = NULL;
996         pinned[1] = NULL;
997
998         pos = *ppos;
999         start_pos = pos;
1000
1001         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1002         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1003         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1004         if (err)
1005                 goto out_nolock;
1006         if (count == 0)
1007                 goto out_nolock;
1008
1009         err = file_remove_suid(file);
1010         if (err)
1011                 goto out_nolock;
1012         file_update_time(file);
1013
1014         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1015
1016         mutex_lock(&inode->i_mutex);
1017         first_index = pos >> PAGE_CACHE_SHIFT;
1018         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1019
1020         /*
1021          * if this is a nodatasum mount, force summing off for the inode
1022          * all the time.  That way a later mount with summing on won't
1023          * get confused
1024          */
1025         if (btrfs_test_opt(root, NODATASUM))
1026                 btrfs_set_flag(inode, NODATASUM);
1027
1028         /*
1029          * there are lots of better ways to do this, but this code
1030          * makes sure the first and last page in the file range are
1031          * up to date and ready for cow
1032          */
1033         if ((pos & (PAGE_CACHE_SIZE - 1))) {
1034                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1035                 if (!PageUptodate(pinned[0])) {
1036                         ret = btrfs_readpage(NULL, pinned[0]);
1037                         BUG_ON(ret);
1038                         wait_on_page_locked(pinned[0]);
1039                 } else {
1040                         unlock_page(pinned[0]);
1041                 }
1042         }
1043         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
1044                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1045                 if (!PageUptodate(pinned[1])) {
1046                         ret = btrfs_readpage(NULL, pinned[1]);
1047                         BUG_ON(ret);
1048                         wait_on_page_locked(pinned[1]);
1049                 } else {
1050                         unlock_page(pinned[1]);
1051                 }
1052         }
1053
1054         while(count > 0) {
1055                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1056                 size_t write_bytes = min(count, nrptrs *
1057                                         (size_t)PAGE_CACHE_SIZE -
1058                                          offset);
1059                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1060                                         PAGE_CACHE_SHIFT;
1061
1062                 WARN_ON(num_pages > nrptrs);
1063                 memset(pages, 0, sizeof(pages));
1064
1065                 ret = btrfs_check_free_space(root, write_bytes, 0);
1066                 if (ret)
1067                         goto out;
1068
1069                 ret = prepare_pages(root, file, pages, num_pages,
1070                                     pos, first_index, last_index,
1071                                     write_bytes);
1072                 if (ret)
1073                         goto out;
1074
1075                 ret = btrfs_copy_from_user(pos, num_pages,
1076                                            write_bytes, pages, buf);
1077                 if (ret) {
1078                         btrfs_drop_pages(pages, num_pages);
1079                         goto out;
1080                 }
1081
1082                 ret = dirty_and_release_pages(NULL, root, file, pages,
1083                                               num_pages, pos, write_bytes);
1084                 btrfs_drop_pages(pages, num_pages);
1085                 if (ret)
1086                         goto out;
1087
1088                 if (will_write) {
1089                         btrfs_fdatawrite_range(inode->i_mapping, pos,
1090                                                pos + write_bytes - 1,
1091                                                WB_SYNC_NONE);
1092                 } else {
1093                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1094                                                            num_pages);
1095                         if (num_pages <
1096                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1097                                 btrfs_btree_balance_dirty(root, 1);
1098                         btrfs_throttle(root);
1099                 }
1100
1101                 buf += write_bytes;
1102                 count -= write_bytes;
1103                 pos += write_bytes;
1104                 num_written += write_bytes;
1105
1106                 cond_resched();
1107         }
1108 out:
1109         mutex_unlock(&inode->i_mutex);
1110
1111 out_nolock:
1112         kfree(pages);
1113         if (pinned[0])
1114                 page_cache_release(pinned[0]);
1115         if (pinned[1])
1116                 page_cache_release(pinned[1]);
1117         *ppos = pos;
1118
1119         if (num_written > 0 && will_write) {
1120                 struct btrfs_trans_handle *trans;
1121
1122                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1123                 if (err)
1124                         num_written = err;
1125
1126                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1127                         trans = btrfs_start_transaction(root, 1);
1128                         ret = btrfs_log_dentry_safe(trans, root,
1129                                                     file->f_dentry);
1130                         if (ret == 0) {
1131                                 btrfs_sync_log(trans, root);
1132                                 btrfs_end_transaction(trans, root);
1133                         } else {
1134                                 btrfs_commit_transaction(trans, root);
1135                         }
1136                 }
1137                 if (file->f_flags & O_DIRECT) {
1138                         invalidate_mapping_pages(inode->i_mapping,
1139                               start_pos >> PAGE_CACHE_SHIFT,
1140                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1141                 }
1142         }
1143         current->backing_dev_info = NULL;
1144         return num_written ? num_written : err;
1145 }
1146
1147 int btrfs_release_file(struct inode * inode, struct file * filp)
1148 {
1149         if (filp->private_data)
1150                 btrfs_ioctl_trans_end(filp);
1151         return 0;
1152 }
1153
1154 /*
1155  * fsync call for both files and directories.  This logs the inode into
1156  * the tree log instead of forcing full commits whenever possible.
1157  *
1158  * It needs to call filemap_fdatawait so that all ordered extent updates are
1159  * in the metadata btree are up to date for copying to the log.
1160  *
1161  * It drops the inode mutex before doing the tree log commit.  This is an
1162  * important optimization for directories because holding the mutex prevents
1163  * new operations on the dir while we write to disk.
1164  */
1165 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1166 {
1167         struct inode *inode = dentry->d_inode;
1168         struct btrfs_root *root = BTRFS_I(inode)->root;
1169         int ret = 0;
1170         struct btrfs_trans_handle *trans;
1171
1172         /*
1173          * check the transaction that last modified this inode
1174          * and see if its already been committed
1175          */
1176         if (!BTRFS_I(inode)->last_trans)
1177                 goto out;
1178
1179         mutex_lock(&root->fs_info->trans_mutex);
1180         if (BTRFS_I(inode)->last_trans <=
1181             root->fs_info->last_trans_committed) {
1182                 BTRFS_I(inode)->last_trans = 0;
1183                 mutex_unlock(&root->fs_info->trans_mutex);
1184                 goto out;
1185         }
1186         mutex_unlock(&root->fs_info->trans_mutex);
1187
1188         root->fs_info->tree_log_batch++;
1189         filemap_fdatawait(inode->i_mapping);
1190         root->fs_info->tree_log_batch++;
1191
1192         /*
1193          * ok we haven't committed the transaction yet, lets do a commit
1194          */
1195         if (file->private_data)
1196                 btrfs_ioctl_trans_end(file);
1197
1198         trans = btrfs_start_transaction(root, 1);
1199         if (!trans) {
1200                 ret = -ENOMEM;
1201                 goto out;
1202         }
1203
1204         ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1205         if (ret < 0) {
1206                 goto out;
1207         }
1208
1209         /* we've logged all the items and now have a consistent
1210          * version of the file in the log.  It is possible that
1211          * someone will come in and modify the file, but that's
1212          * fine because the log is consistent on disk, and we
1213          * have references to all of the file's extents
1214          *
1215          * It is possible that someone will come in and log the
1216          * file again, but that will end up using the synchronization
1217          * inside btrfs_sync_log to keep things safe.
1218          */
1219         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
1220
1221         if (ret > 0) {
1222                 ret = btrfs_commit_transaction(trans, root);
1223         } else {
1224                 btrfs_sync_log(trans, root);
1225                 ret = btrfs_end_transaction(trans, root);
1226         }
1227         mutex_lock(&file->f_dentry->d_inode->i_mutex);
1228 out:
1229         return ret > 0 ? EIO : ret;
1230 }
1231
1232 static struct vm_operations_struct btrfs_file_vm_ops = {
1233         .fault          = filemap_fault,
1234         .page_mkwrite   = btrfs_page_mkwrite,
1235 };
1236
1237 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1238 {
1239         vma->vm_ops = &btrfs_file_vm_ops;
1240         file_accessed(filp);
1241         return 0;
1242 }
1243
1244 struct file_operations btrfs_file_operations = {
1245         .llseek         = generic_file_llseek,
1246         .read           = do_sync_read,
1247         .aio_read       = generic_file_aio_read,
1248         .splice_read    = generic_file_splice_read,
1249         .write          = btrfs_file_write,
1250         .mmap           = btrfs_file_mmap,
1251         .open           = generic_file_open,
1252         .release        = btrfs_release_file,
1253         .fsync          = btrfs_sync_file,
1254         .unlocked_ioctl = btrfs_ioctl,
1255 #ifdef CONFIG_COMPAT
1256         .compat_ioctl   = btrfs_ioctl,
1257 #endif
1258 };