Btrfs: deal with short returns from copy_from_user
[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 <linux/slab.h>
32 #include "ctree.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "ioctl.h"
37 #include "print-tree.h"
38 #include "tree-log.h"
39 #include "locking.h"
40 #include "compat.h"
41
42
43 /* simple helper to fault in pages and copy.  This should go away
44  * and be replaced with calls into generic code.
45  */
46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
47                                          int write_bytes,
48                                          struct page **prepared_pages,
49                                          struct iov_iter *i)
50 {
51         size_t copied = 0;
52         int pg = 0;
53         int offset = pos & (PAGE_CACHE_SIZE - 1);
54         int total_copied = 0;
55
56         while (write_bytes > 0) {
57                 size_t count = min_t(size_t,
58                                      PAGE_CACHE_SIZE - offset, write_bytes);
59                 struct page *page = prepared_pages[pg];
60                 /*
61                  * Copy data from userspace to the current page
62                  *
63                  * Disable pagefault to avoid recursive lock since
64                  * the pages are already locked
65                  */
66                 pagefault_disable();
67                 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
68                 pagefault_enable();
69
70                 /* Flush processor's dcache for this page */
71                 flush_dcache_page(page);
72
73                 /*
74                  * if we get a partial write, we can end up with
75                  * partially up to date pages.  These add
76                  * a lot of complexity, so make sure they don't
77                  * happen by forcing this copy to be retried.
78                  *
79                  * The rest of the btrfs_file_write code will fall
80                  * back to page at a time copies after we return 0.
81                  */
82                 if (!PageUptodate(page) && copied < count)
83                         copied = 0;
84
85                 iov_iter_advance(i, copied);
86                 write_bytes -= copied;
87                 total_copied += copied;
88
89                 /* Return to btrfs_file_aio_write to fault page */
90                 if (unlikely(copied == 0)) {
91                         break;
92                 }
93
94                 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
95                         offset += copied;
96                 } else {
97                         pg++;
98                         offset = 0;
99                 }
100         }
101         return total_copied;
102 }
103
104 /*
105  * unlocks pages after btrfs_file_write is done with them
106  */
107 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
108 {
109         size_t i;
110         for (i = 0; i < num_pages; i++) {
111                 if (!pages[i])
112                         break;
113                 /* page checked is some magic around finding pages that
114                  * have been modified without going through btrfs_set_page_dirty
115                  * clear it here
116                  */
117                 ClearPageChecked(pages[i]);
118                 unlock_page(pages[i]);
119                 mark_page_accessed(pages[i]);
120                 page_cache_release(pages[i]);
121         }
122 }
123
124 /*
125  * after copy_from_user, pages need to be dirtied and we need to make
126  * sure holes are created between the current EOF and the start of
127  * any next extents (if required).
128  *
129  * this also makes the decision about creating an inline extent vs
130  * doing real data extents, marking pages dirty and delalloc as required.
131  */
132 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
133                                    struct btrfs_root *root,
134                                    struct file *file,
135                                    struct page **pages,
136                                    size_t num_pages,
137                                    loff_t pos,
138                                    size_t write_bytes)
139 {
140         int err = 0;
141         int i;
142         struct inode *inode = fdentry(file)->d_inode;
143         u64 num_bytes;
144         u64 start_pos;
145         u64 end_of_last_block;
146         u64 end_pos = pos + write_bytes;
147         loff_t isize = i_size_read(inode);
148
149         start_pos = pos & ~((u64)root->sectorsize - 1);
150         num_bytes = (write_bytes + pos - start_pos +
151                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
152
153         end_of_last_block = start_pos + num_bytes - 1;
154         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
155                                         NULL);
156         BUG_ON(err);
157
158         for (i = 0; i < num_pages; i++) {
159                 struct page *p = pages[i];
160                 SetPageUptodate(p);
161                 ClearPageChecked(p);
162                 set_page_dirty(p);
163         }
164         if (end_pos > isize) {
165                 i_size_write(inode, end_pos);
166                 /* we've only changed i_size in ram, and we haven't updated
167                  * the disk i_size.  There is no need to log the inode
168                  * at this time.
169                  */
170         }
171         return 0;
172 }
173
174 /*
175  * this drops all the extents in the cache that intersect the range
176  * [start, end].  Existing extents are split as required.
177  */
178 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
179                             int skip_pinned)
180 {
181         struct extent_map *em;
182         struct extent_map *split = NULL;
183         struct extent_map *split2 = NULL;
184         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
185         u64 len = end - start + 1;
186         int ret;
187         int testend = 1;
188         unsigned long flags;
189         int compressed = 0;
190
191         WARN_ON(end < start);
192         if (end == (u64)-1) {
193                 len = (u64)-1;
194                 testend = 0;
195         }
196         while (1) {
197                 if (!split)
198                         split = alloc_extent_map(GFP_NOFS);
199                 if (!split2)
200                         split2 = alloc_extent_map(GFP_NOFS);
201                 BUG_ON(!split || !split2);
202
203                 write_lock(&em_tree->lock);
204                 em = lookup_extent_mapping(em_tree, start, len);
205                 if (!em) {
206                         write_unlock(&em_tree->lock);
207                         break;
208                 }
209                 flags = em->flags;
210                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
211                         if (testend && em->start + em->len >= start + len) {
212                                 free_extent_map(em);
213                                 write_unlock(&em_tree->lock);
214                                 break;
215                         }
216                         start = em->start + em->len;
217                         if (testend)
218                                 len = start + len - (em->start + em->len);
219                         free_extent_map(em);
220                         write_unlock(&em_tree->lock);
221                         continue;
222                 }
223                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
224                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
225                 remove_extent_mapping(em_tree, em);
226
227                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
228                     em->start < start) {
229                         split->start = em->start;
230                         split->len = start - em->start;
231                         split->orig_start = em->orig_start;
232                         split->block_start = em->block_start;
233
234                         if (compressed)
235                                 split->block_len = em->block_len;
236                         else
237                                 split->block_len = split->len;
238
239                         split->bdev = em->bdev;
240                         split->flags = flags;
241                         split->compress_type = em->compress_type;
242                         ret = add_extent_mapping(em_tree, split);
243                         BUG_ON(ret);
244                         free_extent_map(split);
245                         split = split2;
246                         split2 = NULL;
247                 }
248                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
249                     testend && em->start + em->len > start + len) {
250                         u64 diff = start + len - em->start;
251
252                         split->start = start + len;
253                         split->len = em->start + em->len - (start + len);
254                         split->bdev = em->bdev;
255                         split->flags = flags;
256                         split->compress_type = em->compress_type;
257
258                         if (compressed) {
259                                 split->block_len = em->block_len;
260                                 split->block_start = em->block_start;
261                                 split->orig_start = em->orig_start;
262                         } else {
263                                 split->block_len = split->len;
264                                 split->block_start = em->block_start + diff;
265                                 split->orig_start = split->start;
266                         }
267
268                         ret = add_extent_mapping(em_tree, split);
269                         BUG_ON(ret);
270                         free_extent_map(split);
271                         split = NULL;
272                 }
273                 write_unlock(&em_tree->lock);
274
275                 /* once for us */
276                 free_extent_map(em);
277                 /* once for the tree*/
278                 free_extent_map(em);
279         }
280         if (split)
281                 free_extent_map(split);
282         if (split2)
283                 free_extent_map(split2);
284         return 0;
285 }
286
287 /*
288  * this is very complex, but the basic idea is to drop all extents
289  * in the range start - end.  hint_block is filled in with a block number
290  * that would be a good hint to the block allocator for this file.
291  *
292  * If an extent intersects the range but is not entirely inside the range
293  * it is either truncated or split.  Anything entirely inside the range
294  * is deleted from the tree.
295  */
296 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
297                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
298 {
299         struct btrfs_root *root = BTRFS_I(inode)->root;
300         struct extent_buffer *leaf;
301         struct btrfs_file_extent_item *fi;
302         struct btrfs_path *path;
303         struct btrfs_key key;
304         struct btrfs_key new_key;
305         u64 search_start = start;
306         u64 disk_bytenr = 0;
307         u64 num_bytes = 0;
308         u64 extent_offset = 0;
309         u64 extent_end = 0;
310         int del_nr = 0;
311         int del_slot = 0;
312         int extent_type;
313         int recow;
314         int ret;
315
316         if (drop_cache)
317                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
318
319         path = btrfs_alloc_path();
320         if (!path)
321                 return -ENOMEM;
322
323         while (1) {
324                 recow = 0;
325                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
326                                                search_start, -1);
327                 if (ret < 0)
328                         break;
329                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
330                         leaf = path->nodes[0];
331                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
332                         if (key.objectid == inode->i_ino &&
333                             key.type == BTRFS_EXTENT_DATA_KEY)
334                                 path->slots[0]--;
335                 }
336                 ret = 0;
337 next_slot:
338                 leaf = path->nodes[0];
339                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
340                         BUG_ON(del_nr > 0);
341                         ret = btrfs_next_leaf(root, path);
342                         if (ret < 0)
343                                 break;
344                         if (ret > 0) {
345                                 ret = 0;
346                                 break;
347                         }
348                         leaf = path->nodes[0];
349                         recow = 1;
350                 }
351
352                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
353                 if (key.objectid > inode->i_ino ||
354                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
355                         break;
356
357                 fi = btrfs_item_ptr(leaf, path->slots[0],
358                                     struct btrfs_file_extent_item);
359                 extent_type = btrfs_file_extent_type(leaf, fi);
360
361                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
362                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
363                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
364                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
365                         extent_offset = btrfs_file_extent_offset(leaf, fi);
366                         extent_end = key.offset +
367                                 btrfs_file_extent_num_bytes(leaf, fi);
368                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
369                         extent_end = key.offset +
370                                 btrfs_file_extent_inline_len(leaf, fi);
371                 } else {
372                         WARN_ON(1);
373                         extent_end = search_start;
374                 }
375
376                 if (extent_end <= search_start) {
377                         path->slots[0]++;
378                         goto next_slot;
379                 }
380
381                 search_start = max(key.offset, start);
382                 if (recow) {
383                         btrfs_release_path(root, path);
384                         continue;
385                 }
386
387                 /*
388                  *     | - range to drop - |
389                  *  | -------- extent -------- |
390                  */
391                 if (start > key.offset && end < extent_end) {
392                         BUG_ON(del_nr > 0);
393                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
394
395                         memcpy(&new_key, &key, sizeof(new_key));
396                         new_key.offset = start;
397                         ret = btrfs_duplicate_item(trans, root, path,
398                                                    &new_key);
399                         if (ret == -EAGAIN) {
400                                 btrfs_release_path(root, path);
401                                 continue;
402                         }
403                         if (ret < 0)
404                                 break;
405
406                         leaf = path->nodes[0];
407                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
408                                             struct btrfs_file_extent_item);
409                         btrfs_set_file_extent_num_bytes(leaf, fi,
410                                                         start - key.offset);
411
412                         fi = btrfs_item_ptr(leaf, path->slots[0],
413                                             struct btrfs_file_extent_item);
414
415                         extent_offset += start - key.offset;
416                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
417                         btrfs_set_file_extent_num_bytes(leaf, fi,
418                                                         extent_end - start);
419                         btrfs_mark_buffer_dirty(leaf);
420
421                         if (disk_bytenr > 0) {
422                                 ret = btrfs_inc_extent_ref(trans, root,
423                                                 disk_bytenr, num_bytes, 0,
424                                                 root->root_key.objectid,
425                                                 new_key.objectid,
426                                                 start - extent_offset);
427                                 BUG_ON(ret);
428                                 *hint_byte = disk_bytenr;
429                         }
430                         key.offset = start;
431                 }
432                 /*
433                  *  | ---- range to drop ----- |
434                  *      | -------- extent -------- |
435                  */
436                 if (start <= key.offset && end < extent_end) {
437                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
438
439                         memcpy(&new_key, &key, sizeof(new_key));
440                         new_key.offset = end;
441                         btrfs_set_item_key_safe(trans, root, path, &new_key);
442
443                         extent_offset += end - key.offset;
444                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
445                         btrfs_set_file_extent_num_bytes(leaf, fi,
446                                                         extent_end - end);
447                         btrfs_mark_buffer_dirty(leaf);
448                         if (disk_bytenr > 0) {
449                                 inode_sub_bytes(inode, end - key.offset);
450                                 *hint_byte = disk_bytenr;
451                         }
452                         break;
453                 }
454
455                 search_start = extent_end;
456                 /*
457                  *       | ---- range to drop ----- |
458                  *  | -------- extent -------- |
459                  */
460                 if (start > key.offset && end >= extent_end) {
461                         BUG_ON(del_nr > 0);
462                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
463
464                         btrfs_set_file_extent_num_bytes(leaf, fi,
465                                                         start - key.offset);
466                         btrfs_mark_buffer_dirty(leaf);
467                         if (disk_bytenr > 0) {
468                                 inode_sub_bytes(inode, extent_end - start);
469                                 *hint_byte = disk_bytenr;
470                         }
471                         if (end == extent_end)
472                                 break;
473
474                         path->slots[0]++;
475                         goto next_slot;
476                 }
477
478                 /*
479                  *  | ---- range to drop ----- |
480                  *    | ------ extent ------ |
481                  */
482                 if (start <= key.offset && end >= extent_end) {
483                         if (del_nr == 0) {
484                                 del_slot = path->slots[0];
485                                 del_nr = 1;
486                         } else {
487                                 BUG_ON(del_slot + del_nr != path->slots[0]);
488                                 del_nr++;
489                         }
490
491                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
492                                 inode_sub_bytes(inode,
493                                                 extent_end - key.offset);
494                                 extent_end = ALIGN(extent_end,
495                                                    root->sectorsize);
496                         } else if (disk_bytenr > 0) {
497                                 ret = btrfs_free_extent(trans, root,
498                                                 disk_bytenr, num_bytes, 0,
499                                                 root->root_key.objectid,
500                                                 key.objectid, key.offset -
501                                                 extent_offset);
502                                 BUG_ON(ret);
503                                 inode_sub_bytes(inode,
504                                                 extent_end - key.offset);
505                                 *hint_byte = disk_bytenr;
506                         }
507
508                         if (end == extent_end)
509                                 break;
510
511                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
512                                 path->slots[0]++;
513                                 goto next_slot;
514                         }
515
516                         ret = btrfs_del_items(trans, root, path, del_slot,
517                                               del_nr);
518                         BUG_ON(ret);
519
520                         del_nr = 0;
521                         del_slot = 0;
522
523                         btrfs_release_path(root, path);
524                         continue;
525                 }
526
527                 BUG_ON(1);
528         }
529
530         if (del_nr > 0) {
531                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
532                 BUG_ON(ret);
533         }
534
535         btrfs_free_path(path);
536         return ret;
537 }
538
539 static int extent_mergeable(struct extent_buffer *leaf, int slot,
540                             u64 objectid, u64 bytenr, u64 orig_offset,
541                             u64 *start, u64 *end)
542 {
543         struct btrfs_file_extent_item *fi;
544         struct btrfs_key key;
545         u64 extent_end;
546
547         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
548                 return 0;
549
550         btrfs_item_key_to_cpu(leaf, &key, slot);
551         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
552                 return 0;
553
554         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
555         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
556             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
557             btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
558             btrfs_file_extent_compression(leaf, fi) ||
559             btrfs_file_extent_encryption(leaf, fi) ||
560             btrfs_file_extent_other_encoding(leaf, fi))
561                 return 0;
562
563         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
564         if ((*start && *start != key.offset) || (*end && *end != extent_end))
565                 return 0;
566
567         *start = key.offset;
568         *end = extent_end;
569         return 1;
570 }
571
572 /*
573  * Mark extent in the range start - end as written.
574  *
575  * This changes extent type from 'pre-allocated' to 'regular'. If only
576  * part of extent is marked as written, the extent will be split into
577  * two or three.
578  */
579 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
580                               struct inode *inode, u64 start, u64 end)
581 {
582         struct btrfs_root *root = BTRFS_I(inode)->root;
583         struct extent_buffer *leaf;
584         struct btrfs_path *path;
585         struct btrfs_file_extent_item *fi;
586         struct btrfs_key key;
587         struct btrfs_key new_key;
588         u64 bytenr;
589         u64 num_bytes;
590         u64 extent_end;
591         u64 orig_offset;
592         u64 other_start;
593         u64 other_end;
594         u64 split;
595         int del_nr = 0;
596         int del_slot = 0;
597         int recow;
598         int ret;
599
600         btrfs_drop_extent_cache(inode, start, end - 1, 0);
601
602         path = btrfs_alloc_path();
603         BUG_ON(!path);
604 again:
605         recow = 0;
606         split = start;
607         key.objectid = inode->i_ino;
608         key.type = BTRFS_EXTENT_DATA_KEY;
609         key.offset = split;
610
611         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
612         if (ret > 0 && path->slots[0] > 0)
613                 path->slots[0]--;
614
615         leaf = path->nodes[0];
616         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
617         BUG_ON(key.objectid != inode->i_ino ||
618                key.type != BTRFS_EXTENT_DATA_KEY);
619         fi = btrfs_item_ptr(leaf, path->slots[0],
620                             struct btrfs_file_extent_item);
621         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
622                BTRFS_FILE_EXTENT_PREALLOC);
623         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
624         BUG_ON(key.offset > start || extent_end < end);
625
626         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
627         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
628         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
629         memcpy(&new_key, &key, sizeof(new_key));
630
631         if (start == key.offset && end < extent_end) {
632                 other_start = 0;
633                 other_end = start;
634                 if (extent_mergeable(leaf, path->slots[0] - 1,
635                                      inode->i_ino, bytenr, orig_offset,
636                                      &other_start, &other_end)) {
637                         new_key.offset = end;
638                         btrfs_set_item_key_safe(trans, root, path, &new_key);
639                         fi = btrfs_item_ptr(leaf, path->slots[0],
640                                             struct btrfs_file_extent_item);
641                         btrfs_set_file_extent_num_bytes(leaf, fi,
642                                                         extent_end - end);
643                         btrfs_set_file_extent_offset(leaf, fi,
644                                                      end - orig_offset);
645                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
646                                             struct btrfs_file_extent_item);
647                         btrfs_set_file_extent_num_bytes(leaf, fi,
648                                                         end - other_start);
649                         btrfs_mark_buffer_dirty(leaf);
650                         goto out;
651                 }
652         }
653
654         if (start > key.offset && end == extent_end) {
655                 other_start = end;
656                 other_end = 0;
657                 if (extent_mergeable(leaf, path->slots[0] + 1,
658                                      inode->i_ino, bytenr, orig_offset,
659                                      &other_start, &other_end)) {
660                         fi = btrfs_item_ptr(leaf, path->slots[0],
661                                             struct btrfs_file_extent_item);
662                         btrfs_set_file_extent_num_bytes(leaf, fi,
663                                                         start - key.offset);
664                         path->slots[0]++;
665                         new_key.offset = start;
666                         btrfs_set_item_key_safe(trans, root, path, &new_key);
667
668                         fi = btrfs_item_ptr(leaf, path->slots[0],
669                                             struct btrfs_file_extent_item);
670                         btrfs_set_file_extent_num_bytes(leaf, fi,
671                                                         other_end - start);
672                         btrfs_set_file_extent_offset(leaf, fi,
673                                                      start - orig_offset);
674                         btrfs_mark_buffer_dirty(leaf);
675                         goto out;
676                 }
677         }
678
679         while (start > key.offset || end < extent_end) {
680                 if (key.offset == start)
681                         split = end;
682
683                 new_key.offset = split;
684                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
685                 if (ret == -EAGAIN) {
686                         btrfs_release_path(root, path);
687                         goto again;
688                 }
689                 BUG_ON(ret < 0);
690
691                 leaf = path->nodes[0];
692                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
693                                     struct btrfs_file_extent_item);
694                 btrfs_set_file_extent_num_bytes(leaf, fi,
695                                                 split - key.offset);
696
697                 fi = btrfs_item_ptr(leaf, path->slots[0],
698                                     struct btrfs_file_extent_item);
699
700                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
701                 btrfs_set_file_extent_num_bytes(leaf, fi,
702                                                 extent_end - split);
703                 btrfs_mark_buffer_dirty(leaf);
704
705                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
706                                            root->root_key.objectid,
707                                            inode->i_ino, orig_offset);
708                 BUG_ON(ret);
709
710                 if (split == start) {
711                         key.offset = start;
712                 } else {
713                         BUG_ON(start != key.offset);
714                         path->slots[0]--;
715                         extent_end = end;
716                 }
717                 recow = 1;
718         }
719
720         other_start = end;
721         other_end = 0;
722         if (extent_mergeable(leaf, path->slots[0] + 1,
723                              inode->i_ino, bytenr, orig_offset,
724                              &other_start, &other_end)) {
725                 if (recow) {
726                         btrfs_release_path(root, path);
727                         goto again;
728                 }
729                 extent_end = other_end;
730                 del_slot = path->slots[0] + 1;
731                 del_nr++;
732                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
733                                         0, root->root_key.objectid,
734                                         inode->i_ino, orig_offset);
735                 BUG_ON(ret);
736         }
737         other_start = 0;
738         other_end = start;
739         if (extent_mergeable(leaf, path->slots[0] - 1,
740                              inode->i_ino, bytenr, orig_offset,
741                              &other_start, &other_end)) {
742                 if (recow) {
743                         btrfs_release_path(root, path);
744                         goto again;
745                 }
746                 key.offset = other_start;
747                 del_slot = path->slots[0];
748                 del_nr++;
749                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
750                                         0, root->root_key.objectid,
751                                         inode->i_ino, orig_offset);
752                 BUG_ON(ret);
753         }
754         if (del_nr == 0) {
755                 fi = btrfs_item_ptr(leaf, path->slots[0],
756                            struct btrfs_file_extent_item);
757                 btrfs_set_file_extent_type(leaf, fi,
758                                            BTRFS_FILE_EXTENT_REG);
759                 btrfs_mark_buffer_dirty(leaf);
760         } else {
761                 fi = btrfs_item_ptr(leaf, del_slot - 1,
762                            struct btrfs_file_extent_item);
763                 btrfs_set_file_extent_type(leaf, fi,
764                                            BTRFS_FILE_EXTENT_REG);
765                 btrfs_set_file_extent_num_bytes(leaf, fi,
766                                                 extent_end - key.offset);
767                 btrfs_mark_buffer_dirty(leaf);
768
769                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
770                 BUG_ON(ret);
771         }
772 out:
773         btrfs_free_path(path);
774         return 0;
775 }
776
777 /*
778  * on error we return an unlocked page and the error value
779  * on success we return a locked page and 0
780  */
781 static int prepare_uptodate_page(struct page *page, u64 pos)
782 {
783         int ret = 0;
784
785         if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
786                 ret = btrfs_readpage(NULL, page);
787                 if (ret)
788                         return ret;
789                 lock_page(page);
790                 if (!PageUptodate(page)) {
791                         unlock_page(page);
792                         return -EIO;
793                 }
794         }
795         return 0;
796 }
797
798 /*
799  * this gets pages into the page cache and locks them down, it also properly
800  * waits for data=ordered extents to finish before allowing the pages to be
801  * modified.
802  */
803 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
804                          struct page **pages, size_t num_pages,
805                          loff_t pos, unsigned long first_index,
806                          unsigned long last_index, size_t write_bytes)
807 {
808         struct extent_state *cached_state = NULL;
809         int i;
810         unsigned long index = pos >> PAGE_CACHE_SHIFT;
811         struct inode *inode = fdentry(file)->d_inode;
812         int err = 0;
813         int faili = 0;
814         u64 start_pos;
815         u64 last_pos;
816
817         start_pos = pos & ~((u64)root->sectorsize - 1);
818         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
819
820         if (start_pos > inode->i_size) {
821                 err = btrfs_cont_expand(inode, start_pos);
822                 if (err)
823                         return err;
824         }
825
826         memset(pages, 0, num_pages * sizeof(struct page *));
827 again:
828         for (i = 0; i < num_pages; i++) {
829                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
830                 if (!pages[i]) {
831                         faili = i - 1;
832                         err = -ENOMEM;
833                         goto fail;
834                 }
835
836                 if (i == 0)
837                         err = prepare_uptodate_page(pages[i], pos);
838                 if (i == num_pages - 1)
839                         err = prepare_uptodate_page(pages[i],
840                                                     pos + write_bytes);
841                 if (err) {
842                         page_cache_release(pages[i]);
843                         faili = i - 1;
844                         goto fail;
845                 }
846                 wait_on_page_writeback(pages[i]);
847         }
848         err = 0;
849         if (start_pos < inode->i_size) {
850                 struct btrfs_ordered_extent *ordered;
851                 lock_extent_bits(&BTRFS_I(inode)->io_tree,
852                                  start_pos, last_pos - 1, 0, &cached_state,
853                                  GFP_NOFS);
854                 ordered = btrfs_lookup_first_ordered_extent(inode,
855                                                             last_pos - 1);
856                 if (ordered &&
857                     ordered->file_offset + ordered->len > start_pos &&
858                     ordered->file_offset < last_pos) {
859                         btrfs_put_ordered_extent(ordered);
860                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
861                                              start_pos, last_pos - 1,
862                                              &cached_state, GFP_NOFS);
863                         for (i = 0; i < num_pages; i++) {
864                                 unlock_page(pages[i]);
865                                 page_cache_release(pages[i]);
866                         }
867                         btrfs_wait_ordered_range(inode, start_pos,
868                                                  last_pos - start_pos);
869                         goto again;
870                 }
871                 if (ordered)
872                         btrfs_put_ordered_extent(ordered);
873
874                 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
875                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
876                                   EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
877                                   GFP_NOFS);
878                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
879                                      start_pos, last_pos - 1, &cached_state,
880                                      GFP_NOFS);
881         }
882         for (i = 0; i < num_pages; i++) {
883                 clear_page_dirty_for_io(pages[i]);
884                 set_page_extent_mapped(pages[i]);
885                 WARN_ON(!PageLocked(pages[i]));
886         }
887         return 0;
888 fail:
889         while (faili >= 0) {
890                 unlock_page(pages[faili]);
891                 page_cache_release(pages[faili]);
892                 faili--;
893         }
894         return err;
895
896 }
897
898 static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
899                                     const struct iovec *iov,
900                                     unsigned long nr_segs, loff_t pos)
901 {
902         struct file *file = iocb->ki_filp;
903         struct inode *inode = fdentry(file)->d_inode;
904         struct btrfs_root *root = BTRFS_I(inode)->root;
905         struct page **pages = NULL;
906         struct iov_iter i;
907         loff_t *ppos = &iocb->ki_pos;
908         loff_t start_pos;
909         ssize_t num_written = 0;
910         ssize_t err = 0;
911         size_t count;
912         size_t ocount;
913         int ret = 0;
914         int nrptrs;
915         unsigned long first_index;
916         unsigned long last_index;
917         int will_write;
918         int buffered = 0;
919         int copied = 0;
920         int dirty_pages = 0;
921
922         will_write = ((file->f_flags & O_DSYNC) || IS_SYNC(inode) ||
923                       (file->f_flags & O_DIRECT));
924
925         start_pos = pos;
926
927         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
928
929         mutex_lock(&inode->i_mutex);
930
931         err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
932         if (err)
933                 goto out;
934         count = ocount;
935
936         current->backing_dev_info = inode->i_mapping->backing_dev_info;
937         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
938         if (err)
939                 goto out;
940
941         if (count == 0)
942                 goto out;
943
944         err = file_remove_suid(file);
945         if (err)
946                 goto out;
947
948         /*
949          * If BTRFS flips readonly due to some impossible error
950          * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
951          * although we have opened a file as writable, we have
952          * to stop this write operation to ensure FS consistency.
953          */
954         if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
955                 err = -EROFS;
956                 goto out;
957         }
958
959         file_update_time(file);
960         BTRFS_I(inode)->sequence++;
961
962         if (unlikely(file->f_flags & O_DIRECT)) {
963                 num_written = generic_file_direct_write(iocb, iov, &nr_segs,
964                                                         pos, ppos, count,
965                                                         ocount);
966                 /*
967                  * the generic O_DIRECT will update in-memory i_size after the
968                  * DIOs are done.  But our endio handlers that update the on
969                  * disk i_size never update past the in memory i_size.  So we
970                  * need one more update here to catch any additions to the
971                  * file
972                  */
973                 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
974                         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
975                         mark_inode_dirty(inode);
976                 }
977
978                 if (num_written < 0) {
979                         ret = num_written;
980                         num_written = 0;
981                         goto out;
982                 } else if (num_written == count) {
983                         /* pick up pos changes done by the generic code */
984                         pos = *ppos;
985                         goto out;
986                 }
987                 /*
988                  * We are going to do buffered for the rest of the range, so we
989                  * need to make sure to invalidate the buffered pages when we're
990                  * done.
991                  */
992                 buffered = 1;
993                 pos += num_written;
994         }
995
996         iov_iter_init(&i, iov, nr_segs, count, num_written);
997         nrptrs = min((iov_iter_count(&i) + PAGE_CACHE_SIZE - 1) /
998                      PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
999                      (sizeof(struct page *)));
1000         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1001         if (!pages) {
1002                 ret = -ENOMEM;
1003                 goto out;
1004         }
1005
1006         /* generic_write_checks can change our pos */
1007         start_pos = pos;
1008
1009         first_index = pos >> PAGE_CACHE_SHIFT;
1010         last_index = (pos + iov_iter_count(&i)) >> PAGE_CACHE_SHIFT;
1011
1012         while (iov_iter_count(&i) > 0) {
1013                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1014                 size_t write_bytes = min(iov_iter_count(&i),
1015                                          nrptrs * (size_t)PAGE_CACHE_SIZE -
1016                                          offset);
1017                 size_t num_pages = (write_bytes + offset +
1018                                     PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1019
1020                 WARN_ON(num_pages > nrptrs);
1021                 memset(pages, 0, sizeof(struct page *) * nrptrs);
1022
1023                 /*
1024                  * Fault pages before locking them in prepare_pages
1025                  * to avoid recursive lock
1026                  */
1027                 if (unlikely(iov_iter_fault_in_readable(&i, write_bytes))) {
1028                         ret = -EFAULT;
1029                         goto out;
1030                 }
1031
1032                 ret = btrfs_delalloc_reserve_space(inode,
1033                                         num_pages << PAGE_CACHE_SHIFT);
1034                 if (ret)
1035                         goto out;
1036
1037                 ret = prepare_pages(root, file, pages, num_pages,
1038                                     pos, first_index, last_index,
1039                                     write_bytes);
1040                 if (ret) {
1041                         btrfs_delalloc_release_space(inode,
1042                                         num_pages << PAGE_CACHE_SHIFT);
1043                         goto out;
1044                 }
1045
1046                 copied = btrfs_copy_from_user(pos, num_pages,
1047                                            write_bytes, pages, &i);
1048
1049                 /*
1050                  * if we have trouble faulting in the pages, fall
1051                  * back to one page at a time
1052                  */
1053                 if (copied < write_bytes)
1054                         nrptrs = 1;
1055
1056                 if (copied == 0)
1057                         dirty_pages = 0;
1058                 else
1059                         dirty_pages = (copied + offset +
1060                                        PAGE_CACHE_SIZE - 1) >>
1061                                        PAGE_CACHE_SHIFT;
1062
1063                 if (num_pages > dirty_pages) {
1064                         if (copied > 0)
1065                                 atomic_inc(
1066                                         &BTRFS_I(inode)->outstanding_extents);
1067                         btrfs_delalloc_release_space(inode,
1068                                         (num_pages - dirty_pages) <<
1069                                         PAGE_CACHE_SHIFT);
1070                 }
1071
1072                 if (copied > 0) {
1073                         dirty_and_release_pages(NULL, root, file, pages,
1074                                                 dirty_pages, pos, copied);
1075                 }
1076
1077                 btrfs_drop_pages(pages, num_pages);
1078
1079                 if (copied > 0) {
1080                         if (will_write) {
1081                                 filemap_fdatawrite_range(inode->i_mapping, pos,
1082                                                          pos + copied - 1);
1083                         } else {
1084                                 balance_dirty_pages_ratelimited_nr(
1085                                                         inode->i_mapping,
1086                                                         dirty_pages);
1087                                 if (dirty_pages <
1088                                 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1089                                         btrfs_btree_balance_dirty(root, 1);
1090                                 btrfs_throttle(root);
1091                         }
1092                 }
1093
1094                 pos += copied;
1095                 num_written += copied;
1096
1097                 cond_resched();
1098         }
1099 out:
1100         mutex_unlock(&inode->i_mutex);
1101         if (ret)
1102                 err = ret;
1103
1104         kfree(pages);
1105         *ppos = pos;
1106
1107         /*
1108          * we want to make sure fsync finds this change
1109          * but we haven't joined a transaction running right now.
1110          *
1111          * Later on, someone is sure to update the inode and get the
1112          * real transid recorded.
1113          *
1114          * We set last_trans now to the fs_info generation + 1,
1115          * this will either be one more than the running transaction
1116          * or the generation used for the next transaction if there isn't
1117          * one running right now.
1118          */
1119         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1120
1121         if (num_written > 0 && will_write) {
1122                 struct btrfs_trans_handle *trans;
1123
1124                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1125                 if (err)
1126                         num_written = err;
1127
1128                 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
1129                         trans = btrfs_start_transaction(root, 0);
1130                         if (IS_ERR(trans)) {
1131                                 num_written = PTR_ERR(trans);
1132                                 goto done;
1133                         }
1134                         mutex_lock(&inode->i_mutex);
1135                         ret = btrfs_log_dentry_safe(trans, root,
1136                                                     file->f_dentry);
1137                         mutex_unlock(&inode->i_mutex);
1138                         if (ret == 0) {
1139                                 ret = btrfs_sync_log(trans, root);
1140                                 if (ret == 0)
1141                                         btrfs_end_transaction(trans, root);
1142                                 else
1143                                         btrfs_commit_transaction(trans, root);
1144                         } else if (ret != BTRFS_NO_LOG_SYNC) {
1145                                 btrfs_commit_transaction(trans, root);
1146                         } else {
1147                                 btrfs_end_transaction(trans, root);
1148                         }
1149                 }
1150                 if (file->f_flags & O_DIRECT && buffered) {
1151                         invalidate_mapping_pages(inode->i_mapping,
1152                               start_pos >> PAGE_CACHE_SHIFT,
1153                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1154                 }
1155         }
1156 done:
1157         current->backing_dev_info = NULL;
1158         return num_written ? num_written : err;
1159 }
1160
1161 int btrfs_release_file(struct inode *inode, struct file *filp)
1162 {
1163         /*
1164          * ordered_data_close is set by settattr when we are about to truncate
1165          * a file from a non-zero size to a zero size.  This tries to
1166          * flush down new bytes that may have been written if the
1167          * application were using truncate to replace a file in place.
1168          */
1169         if (BTRFS_I(inode)->ordered_data_close) {
1170                 BTRFS_I(inode)->ordered_data_close = 0;
1171                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1172                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1173                         filemap_flush(inode->i_mapping);
1174         }
1175         if (filp->private_data)
1176                 btrfs_ioctl_trans_end(filp);
1177         return 0;
1178 }
1179
1180 /*
1181  * fsync call for both files and directories.  This logs the inode into
1182  * the tree log instead of forcing full commits whenever possible.
1183  *
1184  * It needs to call filemap_fdatawait so that all ordered extent updates are
1185  * in the metadata btree are up to date for copying to the log.
1186  *
1187  * It drops the inode mutex before doing the tree log commit.  This is an
1188  * important optimization for directories because holding the mutex prevents
1189  * new operations on the dir while we write to disk.
1190  */
1191 int btrfs_sync_file(struct file *file, int datasync)
1192 {
1193         struct dentry *dentry = file->f_path.dentry;
1194         struct inode *inode = dentry->d_inode;
1195         struct btrfs_root *root = BTRFS_I(inode)->root;
1196         int ret = 0;
1197         struct btrfs_trans_handle *trans;
1198
1199
1200         /* we wait first, since the writeback may change the inode */
1201         root->log_batch++;
1202         /* the VFS called filemap_fdatawrite for us */
1203         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1204         root->log_batch++;
1205
1206         /*
1207          * check the transaction that last modified this inode
1208          * and see if its already been committed
1209          */
1210         if (!BTRFS_I(inode)->last_trans)
1211                 goto out;
1212
1213         /*
1214          * if the last transaction that changed this file was before
1215          * the current transaction, we can bail out now without any
1216          * syncing
1217          */
1218         mutex_lock(&root->fs_info->trans_mutex);
1219         if (BTRFS_I(inode)->last_trans <=
1220             root->fs_info->last_trans_committed) {
1221                 BTRFS_I(inode)->last_trans = 0;
1222                 mutex_unlock(&root->fs_info->trans_mutex);
1223                 goto out;
1224         }
1225         mutex_unlock(&root->fs_info->trans_mutex);
1226
1227         /*
1228          * ok we haven't committed the transaction yet, lets do a commit
1229          */
1230         if (file->private_data)
1231                 btrfs_ioctl_trans_end(file);
1232
1233         trans = btrfs_start_transaction(root, 0);
1234         if (IS_ERR(trans)) {
1235                 ret = PTR_ERR(trans);
1236                 goto out;
1237         }
1238
1239         ret = btrfs_log_dentry_safe(trans, root, dentry);
1240         if (ret < 0)
1241                 goto out;
1242
1243         /* we've logged all the items and now have a consistent
1244          * version of the file in the log.  It is possible that
1245          * someone will come in and modify the file, but that's
1246          * fine because the log is consistent on disk, and we
1247          * have references to all of the file's extents
1248          *
1249          * It is possible that someone will come in and log the
1250          * file again, but that will end up using the synchronization
1251          * inside btrfs_sync_log to keep things safe.
1252          */
1253         mutex_unlock(&dentry->d_inode->i_mutex);
1254
1255         if (ret != BTRFS_NO_LOG_SYNC) {
1256                 if (ret > 0) {
1257                         ret = btrfs_commit_transaction(trans, root);
1258                 } else {
1259                         ret = btrfs_sync_log(trans, root);
1260                         if (ret == 0)
1261                                 ret = btrfs_end_transaction(trans, root);
1262                         else
1263                                 ret = btrfs_commit_transaction(trans, root);
1264                 }
1265         } else {
1266                 ret = btrfs_end_transaction(trans, root);
1267         }
1268         mutex_lock(&dentry->d_inode->i_mutex);
1269 out:
1270         return ret > 0 ? -EIO : ret;
1271 }
1272
1273 static const struct vm_operations_struct btrfs_file_vm_ops = {
1274         .fault          = filemap_fault,
1275         .page_mkwrite   = btrfs_page_mkwrite,
1276 };
1277
1278 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1279 {
1280         struct address_space *mapping = filp->f_mapping;
1281
1282         if (!mapping->a_ops->readpage)
1283                 return -ENOEXEC;
1284
1285         file_accessed(filp);
1286         vma->vm_ops = &btrfs_file_vm_ops;
1287         vma->vm_flags |= VM_CAN_NONLINEAR;
1288
1289         return 0;
1290 }
1291
1292 const struct file_operations btrfs_file_operations = {
1293         .llseek         = generic_file_llseek,
1294         .read           = do_sync_read,
1295         .write          = do_sync_write,
1296         .aio_read       = generic_file_aio_read,
1297         .splice_read    = generic_file_splice_read,
1298         .aio_write      = btrfs_file_aio_write,
1299         .mmap           = btrfs_file_mmap,
1300         .open           = generic_file_open,
1301         .release        = btrfs_release_file,
1302         .fsync          = btrfs_sync_file,
1303         .unlocked_ioctl = btrfs_ioctl,
1304 #ifdef CONFIG_COMPAT
1305         .compat_ioctl   = btrfs_ioctl,
1306 #endif
1307 };