9d08096299675548651069e4aa613712b827ba67
[sfrench/cifs-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         if (err)
128                 return err;
129
130         for (i = 0; i < num_pages; i++) {
131                 struct page *p = pages[i];
132                 SetPageUptodate(p);
133                 ClearPageChecked(p);
134                 set_page_dirty(p);
135         }
136         if (end_pos > isize) {
137                 i_size_write(inode, end_pos);
138                 /* we've only changed i_size in ram, and we haven't updated
139                  * the disk i_size.  There is no need to log the inode
140                  * at this time.
141                  */
142         }
143         return err;
144 }
145
146 /*
147  * this drops all the extents in the cache that intersect the range
148  * [start, end].  Existing extents are split as required.
149  */
150 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
151                             int skip_pinned)
152 {
153         struct extent_map *em;
154         struct extent_map *split = NULL;
155         struct extent_map *split2 = NULL;
156         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
157         u64 len = end - start + 1;
158         int ret;
159         int testend = 1;
160         unsigned long flags;
161         int compressed = 0;
162
163         WARN_ON(end < start);
164         if (end == (u64)-1) {
165                 len = (u64)-1;
166                 testend = 0;
167         }
168         while (1) {
169                 if (!split)
170                         split = alloc_extent_map(GFP_NOFS);
171                 if (!split2)
172                         split2 = alloc_extent_map(GFP_NOFS);
173
174                 write_lock(&em_tree->lock);
175                 em = lookup_extent_mapping(em_tree, start, len);
176                 if (!em) {
177                         write_unlock(&em_tree->lock);
178                         break;
179                 }
180                 flags = em->flags;
181                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
182                         if (testend && em->start + em->len >= start + len) {
183                                 free_extent_map(em);
184                                 write_unlock(&em_tree->lock);
185                                 break;
186                         }
187                         start = em->start + em->len;
188                         if (testend)
189                                 len = start + len - (em->start + em->len);
190                         free_extent_map(em);
191                         write_unlock(&em_tree->lock);
192                         continue;
193                 }
194                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
195                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
196                 remove_extent_mapping(em_tree, em);
197
198                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
199                     em->start < start) {
200                         split->start = em->start;
201                         split->len = start - em->start;
202                         split->orig_start = em->orig_start;
203                         split->block_start = em->block_start;
204
205                         if (compressed)
206                                 split->block_len = em->block_len;
207                         else
208                                 split->block_len = split->len;
209
210                         split->bdev = em->bdev;
211                         split->flags = flags;
212                         ret = add_extent_mapping(em_tree, split);
213                         BUG_ON(ret);
214                         free_extent_map(split);
215                         split = split2;
216                         split2 = NULL;
217                 }
218                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
219                     testend && em->start + em->len > start + len) {
220                         u64 diff = start + len - em->start;
221
222                         split->start = start + len;
223                         split->len = em->start + em->len - (start + len);
224                         split->bdev = em->bdev;
225                         split->flags = flags;
226
227                         if (compressed) {
228                                 split->block_len = em->block_len;
229                                 split->block_start = em->block_start;
230                                 split->orig_start = em->orig_start;
231                         } else {
232                                 split->block_len = split->len;
233                                 split->block_start = em->block_start + diff;
234                                 split->orig_start = split->start;
235                         }
236
237                         ret = add_extent_mapping(em_tree, split);
238                         BUG_ON(ret);
239                         free_extent_map(split);
240                         split = NULL;
241                 }
242                 write_unlock(&em_tree->lock);
243
244                 /* once for us */
245                 free_extent_map(em);
246                 /* once for the tree*/
247                 free_extent_map(em);
248         }
249         if (split)
250                 free_extent_map(split);
251         if (split2)
252                 free_extent_map(split2);
253         return 0;
254 }
255
256 /*
257  * this is very complex, but the basic idea is to drop all extents
258  * in the range start - end.  hint_block is filled in with a block number
259  * that would be a good hint to the block allocator for this file.
260  *
261  * If an extent intersects the range but is not entirely inside the range
262  * it is either truncated or split.  Anything entirely inside the range
263  * is deleted from the tree.
264  */
265 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
266                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
267 {
268         struct btrfs_root *root = BTRFS_I(inode)->root;
269         struct extent_buffer *leaf;
270         struct btrfs_file_extent_item *fi;
271         struct btrfs_path *path;
272         struct btrfs_key key;
273         struct btrfs_key new_key;
274         u64 search_start = start;
275         u64 disk_bytenr = 0;
276         u64 num_bytes = 0;
277         u64 extent_offset = 0;
278         u64 extent_end = 0;
279         int del_nr = 0;
280         int del_slot = 0;
281         int extent_type;
282         int recow;
283         int ret;
284
285         if (drop_cache)
286                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
287
288         path = btrfs_alloc_path();
289         if (!path)
290                 return -ENOMEM;
291
292         while (1) {
293                 recow = 0;
294                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
295                                                search_start, -1);
296                 if (ret < 0)
297                         break;
298                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
299                         leaf = path->nodes[0];
300                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
301                         if (key.objectid == inode->i_ino &&
302                             key.type == BTRFS_EXTENT_DATA_KEY)
303                                 path->slots[0]--;
304                 }
305                 ret = 0;
306 next_slot:
307                 leaf = path->nodes[0];
308                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
309                         BUG_ON(del_nr > 0);
310                         ret = btrfs_next_leaf(root, path);
311                         if (ret < 0)
312                                 break;
313                         if (ret > 0) {
314                                 ret = 0;
315                                 break;
316                         }
317                         leaf = path->nodes[0];
318                         recow = 1;
319                 }
320
321                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
322                 if (key.objectid > inode->i_ino ||
323                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
324                         break;
325
326                 fi = btrfs_item_ptr(leaf, path->slots[0],
327                                     struct btrfs_file_extent_item);
328                 extent_type = btrfs_file_extent_type(leaf, fi);
329
330                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
331                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
332                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
333                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
334                         extent_offset = btrfs_file_extent_offset(leaf, fi);
335                         extent_end = key.offset +
336                                 btrfs_file_extent_num_bytes(leaf, fi);
337                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
338                         extent_end = key.offset +
339                                 btrfs_file_extent_inline_len(leaf, fi);
340                 } else {
341                         WARN_ON(1);
342                         extent_end = search_start;
343                 }
344
345                 if (extent_end <= search_start) {
346                         path->slots[0]++;
347                         goto next_slot;
348                 }
349
350                 search_start = max(key.offset, start);
351                 if (recow) {
352                         btrfs_release_path(root, path);
353                         continue;
354                 }
355
356                 /*
357                  *     | - range to drop - |
358                  *  | -------- extent -------- |
359                  */
360                 if (start > key.offset && end < extent_end) {
361                         BUG_ON(del_nr > 0);
362                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
363
364                         memcpy(&new_key, &key, sizeof(new_key));
365                         new_key.offset = start;
366                         ret = btrfs_duplicate_item(trans, root, path,
367                                                    &new_key);
368                         if (ret == -EAGAIN) {
369                                 btrfs_release_path(root, path);
370                                 continue;
371                         }
372                         if (ret < 0)
373                                 break;
374
375                         leaf = path->nodes[0];
376                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
377                                             struct btrfs_file_extent_item);
378                         btrfs_set_file_extent_num_bytes(leaf, fi,
379                                                         start - key.offset);
380
381                         fi = btrfs_item_ptr(leaf, path->slots[0],
382                                             struct btrfs_file_extent_item);
383
384                         extent_offset += start - key.offset;
385                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
386                         btrfs_set_file_extent_num_bytes(leaf, fi,
387                                                         extent_end - start);
388                         btrfs_mark_buffer_dirty(leaf);
389
390                         if (disk_bytenr > 0) {
391                                 ret = btrfs_inc_extent_ref(trans, root,
392                                                 disk_bytenr, num_bytes, 0,
393                                                 root->root_key.objectid,
394                                                 new_key.objectid,
395                                                 start - extent_offset);
396                                 BUG_ON(ret);
397                                 *hint_byte = disk_bytenr;
398                         }
399                         key.offset = start;
400                 }
401                 /*
402                  *  | ---- range to drop ----- |
403                  *      | -------- extent -------- |
404                  */
405                 if (start <= key.offset && end < extent_end) {
406                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
407
408                         memcpy(&new_key, &key, sizeof(new_key));
409                         new_key.offset = end;
410                         btrfs_set_item_key_safe(trans, root, path, &new_key);
411
412                         extent_offset += end - key.offset;
413                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
414                         btrfs_set_file_extent_num_bytes(leaf, fi,
415                                                         extent_end - end);
416                         btrfs_mark_buffer_dirty(leaf);
417                         if (disk_bytenr > 0) {
418                                 inode_sub_bytes(inode, end - key.offset);
419                                 *hint_byte = disk_bytenr;
420                         }
421                         break;
422                 }
423
424                 search_start = extent_end;
425                 /*
426                  *       | ---- range to drop ----- |
427                  *  | -------- extent -------- |
428                  */
429                 if (start > key.offset && end >= extent_end) {
430                         BUG_ON(del_nr > 0);
431                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
432
433                         btrfs_set_file_extent_num_bytes(leaf, fi,
434                                                         start - key.offset);
435                         btrfs_mark_buffer_dirty(leaf);
436                         if (disk_bytenr > 0) {
437                                 inode_sub_bytes(inode, extent_end - start);
438                                 *hint_byte = disk_bytenr;
439                         }
440                         if (end == extent_end)
441                                 break;
442
443                         path->slots[0]++;
444                         goto next_slot;
445                 }
446
447                 /*
448                  *  | ---- range to drop ----- |
449                  *    | ------ extent ------ |
450                  */
451                 if (start <= key.offset && end >= extent_end) {
452                         if (del_nr == 0) {
453                                 del_slot = path->slots[0];
454                                 del_nr = 1;
455                         } else {
456                                 BUG_ON(del_slot + del_nr != path->slots[0]);
457                                 del_nr++;
458                         }
459
460                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
461                                 inode_sub_bytes(inode,
462                                                 extent_end - key.offset);
463                                 extent_end = ALIGN(extent_end,
464                                                    root->sectorsize);
465                         } else if (disk_bytenr > 0) {
466                                 ret = btrfs_free_extent(trans, root,
467                                                 disk_bytenr, num_bytes, 0,
468                                                 root->root_key.objectid,
469                                                 key.objectid, key.offset -
470                                                 extent_offset);
471                                 BUG_ON(ret);
472                                 inode_sub_bytes(inode,
473                                                 extent_end - key.offset);
474                                 *hint_byte = disk_bytenr;
475                         }
476
477                         if (end == extent_end)
478                                 break;
479
480                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
481                                 path->slots[0]++;
482                                 goto next_slot;
483                         }
484
485                         ret = btrfs_del_items(trans, root, path, del_slot,
486                                               del_nr);
487                         BUG_ON(ret);
488
489                         del_nr = 0;
490                         del_slot = 0;
491
492                         btrfs_release_path(root, path);
493                         continue;
494                 }
495
496                 BUG_ON(1);
497         }
498
499         if (del_nr > 0) {
500                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
501                 BUG_ON(ret);
502         }
503
504         btrfs_free_path(path);
505         return ret;
506 }
507
508 static int extent_mergeable(struct extent_buffer *leaf, int slot,
509                             u64 objectid, u64 bytenr, u64 orig_offset,
510                             u64 *start, u64 *end)
511 {
512         struct btrfs_file_extent_item *fi;
513         struct btrfs_key key;
514         u64 extent_end;
515
516         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
517                 return 0;
518
519         btrfs_item_key_to_cpu(leaf, &key, slot);
520         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
521                 return 0;
522
523         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
524         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
525             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
526             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
527             btrfs_file_extent_compression(leaf, fi) ||
528             btrfs_file_extent_encryption(leaf, fi) ||
529             btrfs_file_extent_other_encoding(leaf, fi))
530                 return 0;
531
532         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
533         if ((*start && *start != key.offset) || (*end && *end != extent_end))
534                 return 0;
535
536         *start = key.offset;
537         *end = extent_end;
538         return 1;
539 }
540
541 /*
542  * Mark extent in the range start - end as written.
543  *
544  * This changes extent type from 'pre-allocated' to 'regular'. If only
545  * part of extent is marked as written, the extent will be split into
546  * two or three.
547  */
548 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
549                               struct inode *inode, u64 start, u64 end)
550 {
551         struct btrfs_root *root = BTRFS_I(inode)->root;
552         struct extent_buffer *leaf;
553         struct btrfs_path *path;
554         struct btrfs_file_extent_item *fi;
555         struct btrfs_key key;
556         struct btrfs_key new_key;
557         u64 bytenr;
558         u64 num_bytes;
559         u64 extent_end;
560         u64 orig_offset;
561         u64 other_start;
562         u64 other_end;
563         u64 split;
564         int del_nr = 0;
565         int del_slot = 0;
566         int recow;
567         int ret;
568
569         btrfs_drop_extent_cache(inode, start, end - 1, 0);
570
571         path = btrfs_alloc_path();
572         BUG_ON(!path);
573 again:
574         recow = 0;
575         split = start;
576         key.objectid = inode->i_ino;
577         key.type = BTRFS_EXTENT_DATA_KEY;
578         key.offset = split;
579
580         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
581         if (ret > 0 && path->slots[0] > 0)
582                 path->slots[0]--;
583
584         leaf = path->nodes[0];
585         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
586         BUG_ON(key.objectid != inode->i_ino ||
587                key.type != BTRFS_EXTENT_DATA_KEY);
588         fi = btrfs_item_ptr(leaf, path->slots[0],
589                             struct btrfs_file_extent_item);
590         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
591                BTRFS_FILE_EXTENT_PREALLOC);
592         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
593         BUG_ON(key.offset > start || extent_end < end);
594
595         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
596         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
597         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
598         memcpy(&new_key, &key, sizeof(new_key));
599
600         if (start == key.offset && end < extent_end) {
601                 other_start = 0;
602                 other_end = start;
603                 if (extent_mergeable(leaf, path->slots[0] - 1,
604                                      inode->i_ino, bytenr, orig_offset,
605                                      &other_start, &other_end)) {
606                         new_key.offset = end;
607                         btrfs_set_item_key_safe(trans, root, path, &new_key);
608                         fi = btrfs_item_ptr(leaf, path->slots[0],
609                                             struct btrfs_file_extent_item);
610                         btrfs_set_file_extent_num_bytes(leaf, fi,
611                                                         extent_end - end);
612                         btrfs_set_file_extent_offset(leaf, fi,
613                                                      end - orig_offset);
614                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
615                                             struct btrfs_file_extent_item);
616                         btrfs_set_file_extent_num_bytes(leaf, fi,
617                                                         end - other_start);
618                         btrfs_mark_buffer_dirty(leaf);
619                         goto out;
620                 }
621         }
622
623         if (start > key.offset && end == extent_end) {
624                 other_start = end;
625                 other_end = 0;
626                 if (extent_mergeable(leaf, path->slots[0] + 1,
627                                      inode->i_ino, bytenr, orig_offset,
628                                      &other_start, &other_end)) {
629                         fi = btrfs_item_ptr(leaf, path->slots[0],
630                                             struct btrfs_file_extent_item);
631                         btrfs_set_file_extent_num_bytes(leaf, fi,
632                                                         start - key.offset);
633                         path->slots[0]++;
634                         new_key.offset = start;
635                         btrfs_set_item_key_safe(trans, root, path, &new_key);
636
637                         fi = btrfs_item_ptr(leaf, path->slots[0],
638                                             struct btrfs_file_extent_item);
639                         btrfs_set_file_extent_num_bytes(leaf, fi,
640                                                         other_end - start);
641                         btrfs_set_file_extent_offset(leaf, fi,
642                                                      start - orig_offset);
643                         btrfs_mark_buffer_dirty(leaf);
644                         goto out;
645                 }
646         }
647
648         while (start > key.offset || end < extent_end) {
649                 if (key.offset == start)
650                         split = end;
651
652                 new_key.offset = split;
653                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
654                 if (ret == -EAGAIN) {
655                         btrfs_release_path(root, path);
656                         goto again;
657                 }
658                 BUG_ON(ret < 0);
659
660                 leaf = path->nodes[0];
661                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
662                                     struct btrfs_file_extent_item);
663                 btrfs_set_file_extent_num_bytes(leaf, fi,
664                                                 split - key.offset);
665
666                 fi = btrfs_item_ptr(leaf, path->slots[0],
667                                     struct btrfs_file_extent_item);
668
669                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
670                 btrfs_set_file_extent_num_bytes(leaf, fi,
671                                                 extent_end - split);
672                 btrfs_mark_buffer_dirty(leaf);
673
674                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
675                                            root->root_key.objectid,
676                                            inode->i_ino, orig_offset);
677                 BUG_ON(ret);
678
679                 if (split == start) {
680                         key.offset = start;
681                 } else {
682                         BUG_ON(start != key.offset);
683                         path->slots[0]--;
684                         extent_end = end;
685                 }
686                 recow = 1;
687         }
688
689         other_start = end;
690         other_end = 0;
691         if (extent_mergeable(leaf, path->slots[0] + 1,
692                              inode->i_ino, bytenr, orig_offset,
693                              &other_start, &other_end)) {
694                 if (recow) {
695                         btrfs_release_path(root, path);
696                         goto again;
697                 }
698                 extent_end = other_end;
699                 del_slot = path->slots[0] + 1;
700                 del_nr++;
701                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
702                                         0, root->root_key.objectid,
703                                         inode->i_ino, orig_offset);
704                 BUG_ON(ret);
705         }
706         other_start = 0;
707         other_end = start;
708         if (extent_mergeable(leaf, path->slots[0] - 1,
709                              inode->i_ino, bytenr, orig_offset,
710                              &other_start, &other_end)) {
711                 if (recow) {
712                         btrfs_release_path(root, path);
713                         goto again;
714                 }
715                 key.offset = other_start;
716                 del_slot = path->slots[0];
717                 del_nr++;
718                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
719                                         0, root->root_key.objectid,
720                                         inode->i_ino, orig_offset);
721                 BUG_ON(ret);
722         }
723         fi = btrfs_item_ptr(leaf, path->slots[0],
724                            struct btrfs_file_extent_item);
725         if (del_nr == 0) {
726                 btrfs_set_file_extent_type(leaf, fi,
727                                            BTRFS_FILE_EXTENT_REG);
728                 btrfs_mark_buffer_dirty(leaf);
729         } else {
730                 btrfs_set_file_extent_type(leaf, fi,
731                                            BTRFS_FILE_EXTENT_REG);
732                 btrfs_set_file_extent_num_bytes(leaf, fi,
733                                                 extent_end - key.offset);
734                 btrfs_mark_buffer_dirty(leaf);
735
736                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
737                 BUG_ON(ret);
738         }
739 out:
740         btrfs_free_path(path);
741         return 0;
742 }
743
744 /*
745  * this gets pages into the page cache and locks them down, it also properly
746  * waits for data=ordered extents to finish before allowing the pages to be
747  * modified.
748  */
749 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
750                          struct page **pages, size_t num_pages,
751                          loff_t pos, unsigned long first_index,
752                          unsigned long last_index, size_t write_bytes)
753 {
754         int i;
755         unsigned long index = pos >> PAGE_CACHE_SHIFT;
756         struct inode *inode = fdentry(file)->d_inode;
757         int err = 0;
758         u64 start_pos;
759         u64 last_pos;
760
761         start_pos = pos & ~((u64)root->sectorsize - 1);
762         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
763
764         if (start_pos > inode->i_size) {
765                 err = btrfs_cont_expand(inode, start_pos);
766                 if (err)
767                         return err;
768         }
769
770         memset(pages, 0, num_pages * sizeof(struct page *));
771 again:
772         for (i = 0; i < num_pages; i++) {
773                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
774                 if (!pages[i]) {
775                         err = -ENOMEM;
776                         BUG_ON(1);
777                 }
778                 wait_on_page_writeback(pages[i]);
779         }
780         if (start_pos < inode->i_size) {
781                 struct btrfs_ordered_extent *ordered;
782                 lock_extent(&BTRFS_I(inode)->io_tree,
783                             start_pos, last_pos - 1, GFP_NOFS);
784                 ordered = btrfs_lookup_first_ordered_extent(inode,
785                                                             last_pos - 1);
786                 if (ordered &&
787                     ordered->file_offset + ordered->len > start_pos &&
788                     ordered->file_offset < last_pos) {
789                         btrfs_put_ordered_extent(ordered);
790                         unlock_extent(&BTRFS_I(inode)->io_tree,
791                                       start_pos, last_pos - 1, GFP_NOFS);
792                         for (i = 0; i < num_pages; i++) {
793                                 unlock_page(pages[i]);
794                                 page_cache_release(pages[i]);
795                         }
796                         btrfs_wait_ordered_range(inode, start_pos,
797                                                  last_pos - start_pos);
798                         goto again;
799                 }
800                 if (ordered)
801                         btrfs_put_ordered_extent(ordered);
802
803                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
804                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
805                                   EXTENT_DO_ACCOUNTING,
806                                   GFP_NOFS);
807                 unlock_extent(&BTRFS_I(inode)->io_tree,
808                               start_pos, last_pos - 1, GFP_NOFS);
809         }
810         for (i = 0; i < num_pages; i++) {
811                 clear_page_dirty_for_io(pages[i]);
812                 set_page_extent_mapped(pages[i]);
813                 WARN_ON(!PageLocked(pages[i]));
814         }
815         return 0;
816 }
817
818 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
819                                 size_t count, loff_t *ppos)
820 {
821         loff_t pos;
822         loff_t start_pos;
823         ssize_t num_written = 0;
824         ssize_t err = 0;
825         int ret = 0;
826         struct inode *inode = fdentry(file)->d_inode;
827         struct btrfs_root *root = BTRFS_I(inode)->root;
828         struct page **pages = NULL;
829         int nrptrs;
830         struct page *pinned[2];
831         unsigned long first_index;
832         unsigned long last_index;
833         int will_write;
834
835         will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
836                       (file->f_flags & O_DIRECT));
837
838         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
839                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
840         pinned[0] = NULL;
841         pinned[1] = NULL;
842
843         pos = *ppos;
844         start_pos = pos;
845
846         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
847
848         /* do the reserve before the mutex lock in case we have to do some
849          * flushing.  We wouldn't deadlock, but this is more polite.
850          */
851         err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
852         if (err)
853                 goto out_nolock;
854
855         mutex_lock(&inode->i_mutex);
856
857         current->backing_dev_info = inode->i_mapping->backing_dev_info;
858         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
859         if (err)
860                 goto out;
861
862         if (count == 0)
863                 goto out;
864
865         err = file_remove_suid(file);
866         if (err)
867                 goto out;
868
869         file_update_time(file);
870
871         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
872
873         /* generic_write_checks can change our pos */
874         start_pos = pos;
875
876         BTRFS_I(inode)->sequence++;
877         first_index = pos >> PAGE_CACHE_SHIFT;
878         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
879
880         /*
881          * there are lots of better ways to do this, but this code
882          * makes sure the first and last page in the file range are
883          * up to date and ready for cow
884          */
885         if ((pos & (PAGE_CACHE_SIZE - 1))) {
886                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
887                 if (!PageUptodate(pinned[0])) {
888                         ret = btrfs_readpage(NULL, pinned[0]);
889                         BUG_ON(ret);
890                         wait_on_page_locked(pinned[0]);
891                 } else {
892                         unlock_page(pinned[0]);
893                 }
894         }
895         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
896                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
897                 if (!PageUptodate(pinned[1])) {
898                         ret = btrfs_readpage(NULL, pinned[1]);
899                         BUG_ON(ret);
900                         wait_on_page_locked(pinned[1]);
901                 } else {
902                         unlock_page(pinned[1]);
903                 }
904         }
905
906         while (count > 0) {
907                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
908                 size_t write_bytes = min(count, nrptrs *
909                                         (size_t)PAGE_CACHE_SIZE -
910                                          offset);
911                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
912                                         PAGE_CACHE_SHIFT;
913
914                 WARN_ON(num_pages > nrptrs);
915                 memset(pages, 0, sizeof(struct page *) * nrptrs);
916
917                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
918                 if (ret)
919                         goto out;
920
921                 ret = prepare_pages(root, file, pages, num_pages,
922                                     pos, first_index, last_index,
923                                     write_bytes);
924                 if (ret) {
925                         btrfs_free_reserved_data_space(root, inode,
926                                                        write_bytes);
927                         goto out;
928                 }
929
930                 ret = btrfs_copy_from_user(pos, num_pages,
931                                            write_bytes, pages, buf);
932                 if (ret) {
933                         btrfs_free_reserved_data_space(root, inode,
934                                                        write_bytes);
935                         btrfs_drop_pages(pages, num_pages);
936                         goto out;
937                 }
938
939                 ret = dirty_and_release_pages(NULL, root, file, pages,
940                                               num_pages, pos, write_bytes);
941                 btrfs_drop_pages(pages, num_pages);
942                 if (ret) {
943                         btrfs_free_reserved_data_space(root, inode,
944                                                        write_bytes);
945                         goto out;
946                 }
947
948                 if (will_write) {
949                         filemap_fdatawrite_range(inode->i_mapping, pos,
950                                                  pos + write_bytes - 1);
951                 } else {
952                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
953                                                            num_pages);
954                         if (num_pages <
955                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
956                                 btrfs_btree_balance_dirty(root, 1);
957                         btrfs_throttle(root);
958                 }
959
960                 buf += write_bytes;
961                 count -= write_bytes;
962                 pos += write_bytes;
963                 num_written += write_bytes;
964
965                 cond_resched();
966         }
967 out:
968         mutex_unlock(&inode->i_mutex);
969         if (ret)
970                 err = ret;
971         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
972
973 out_nolock:
974         kfree(pages);
975         if (pinned[0])
976                 page_cache_release(pinned[0]);
977         if (pinned[1])
978                 page_cache_release(pinned[1]);
979         *ppos = pos;
980
981         /*
982          * we want to make sure fsync finds this change
983          * but we haven't joined a transaction running right now.
984          *
985          * Later on, someone is sure to update the inode and get the
986          * real transid recorded.
987          *
988          * We set last_trans now to the fs_info generation + 1,
989          * this will either be one more than the running transaction
990          * or the generation used for the next transaction if there isn't
991          * one running right now.
992          */
993         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
994
995         if (num_written > 0 && will_write) {
996                 struct btrfs_trans_handle *trans;
997
998                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
999                 if (err)
1000                         num_written = err;
1001
1002                 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1003                         trans = btrfs_start_transaction(root, 1);
1004                         ret = btrfs_log_dentry_safe(trans, root,
1005                                                     file->f_dentry);
1006                         if (ret == 0) {
1007                                 ret = btrfs_sync_log(trans, root);
1008                                 if (ret == 0)
1009                                         btrfs_end_transaction(trans, root);
1010                                 else
1011                                         btrfs_commit_transaction(trans, root);
1012                         } else if (ret != BTRFS_NO_LOG_SYNC) {
1013                                 btrfs_commit_transaction(trans, root);
1014                         } else {
1015                                 btrfs_end_transaction(trans, root);
1016                         }
1017                 }
1018                 if (file->f_flags & O_DIRECT) {
1019                         invalidate_mapping_pages(inode->i_mapping,
1020                               start_pos >> PAGE_CACHE_SHIFT,
1021                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1022                 }
1023         }
1024         current->backing_dev_info = NULL;
1025         return num_written ? num_written : err;
1026 }
1027
1028 int btrfs_release_file(struct inode *inode, struct file *filp)
1029 {
1030         /*
1031          * ordered_data_close is set by settattr when we are about to truncate
1032          * a file from a non-zero size to a zero size.  This tries to
1033          * flush down new bytes that may have been written if the
1034          * application were using truncate to replace a file in place.
1035          */
1036         if (BTRFS_I(inode)->ordered_data_close) {
1037                 BTRFS_I(inode)->ordered_data_close = 0;
1038                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1039                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1040                         filemap_flush(inode->i_mapping);
1041         }
1042         if (filp->private_data)
1043                 btrfs_ioctl_trans_end(filp);
1044         return 0;
1045 }
1046
1047 /*
1048  * fsync call for both files and directories.  This logs the inode into
1049  * the tree log instead of forcing full commits whenever possible.
1050  *
1051  * It needs to call filemap_fdatawait so that all ordered extent updates are
1052  * in the metadata btree are up to date for copying to the log.
1053  *
1054  * It drops the inode mutex before doing the tree log commit.  This is an
1055  * important optimization for directories because holding the mutex prevents
1056  * new operations on the dir while we write to disk.
1057  */
1058 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1059 {
1060         struct inode *inode = dentry->d_inode;
1061         struct btrfs_root *root = BTRFS_I(inode)->root;
1062         int ret = 0;
1063         struct btrfs_trans_handle *trans;
1064
1065
1066         /* we wait first, since the writeback may change the inode */
1067         root->log_batch++;
1068         /* the VFS called filemap_fdatawrite for us */
1069         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1070         root->log_batch++;
1071
1072         /*
1073          * check the transaction that last modified this inode
1074          * and see if its already been committed
1075          */
1076         if (!BTRFS_I(inode)->last_trans)
1077                 goto out;
1078
1079         /*
1080          * if the last transaction that changed this file was before
1081          * the current transaction, we can bail out now without any
1082          * syncing
1083          */
1084         mutex_lock(&root->fs_info->trans_mutex);
1085         if (BTRFS_I(inode)->last_trans <=
1086             root->fs_info->last_trans_committed) {
1087                 BTRFS_I(inode)->last_trans = 0;
1088                 mutex_unlock(&root->fs_info->trans_mutex);
1089                 goto out;
1090         }
1091         mutex_unlock(&root->fs_info->trans_mutex);
1092
1093         /*
1094          * ok we haven't committed the transaction yet, lets do a commit
1095          */
1096         if (file && file->private_data)
1097                 btrfs_ioctl_trans_end(file);
1098
1099         trans = btrfs_start_transaction(root, 1);
1100         if (!trans) {
1101                 ret = -ENOMEM;
1102                 goto out;
1103         }
1104
1105         ret = btrfs_log_dentry_safe(trans, root, dentry);
1106         if (ret < 0)
1107                 goto out;
1108
1109         /* we've logged all the items and now have a consistent
1110          * version of the file in the log.  It is possible that
1111          * someone will come in and modify the file, but that's
1112          * fine because the log is consistent on disk, and we
1113          * have references to all of the file's extents
1114          *
1115          * It is possible that someone will come in and log the
1116          * file again, but that will end up using the synchronization
1117          * inside btrfs_sync_log to keep things safe.
1118          */
1119         mutex_unlock(&dentry->d_inode->i_mutex);
1120
1121         if (ret != BTRFS_NO_LOG_SYNC) {
1122                 if (ret > 0) {
1123                         ret = btrfs_commit_transaction(trans, root);
1124                 } else {
1125                         ret = btrfs_sync_log(trans, root);
1126                         if (ret == 0)
1127                                 ret = btrfs_end_transaction(trans, root);
1128                         else
1129                                 ret = btrfs_commit_transaction(trans, root);
1130                 }
1131         } else {
1132                 ret = btrfs_end_transaction(trans, root);
1133         }
1134         mutex_lock(&dentry->d_inode->i_mutex);
1135 out:
1136         return ret > 0 ? -EIO : ret;
1137 }
1138
1139 static const struct vm_operations_struct btrfs_file_vm_ops = {
1140         .fault          = filemap_fault,
1141         .page_mkwrite   = btrfs_page_mkwrite,
1142 };
1143
1144 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1145 {
1146         vma->vm_ops = &btrfs_file_vm_ops;
1147         file_accessed(filp);
1148         return 0;
1149 }
1150
1151 const struct file_operations btrfs_file_operations = {
1152         .llseek         = generic_file_llseek,
1153         .read           = do_sync_read,
1154         .aio_read       = generic_file_aio_read,
1155         .splice_read    = generic_file_splice_read,
1156         .write          = btrfs_file_write,
1157         .mmap           = btrfs_file_mmap,
1158         .open           = generic_file_open,
1159         .release        = btrfs_release_file,
1160         .fsync          = btrfs_sync_file,
1161         .unlocked_ioctl = btrfs_ioctl,
1162 #ifdef CONFIG_COMPAT
1163         .compat_ioctl   = btrfs_ioctl,
1164 #endif
1165 };