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