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