Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
[linux-2.6.git] / fs / btrfs / file.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/statfs.h>
30 #include <linux/compat.h>
31 #include "ctree.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "ioctl.h"
36 #include "print-tree.h"
37 #include "tree-log.h"
38 #include "locking.h"
39 #include "compat.h"
40
41
42 /* simple helper to fault in pages and copy.  This should go away
43  * and be replaced with calls into generic code.
44  */
45 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
46                                          int write_bytes,
47                                          struct page **prepared_pages,
48                                          const char __user *buf)
49 {
50         long page_fault = 0;
51         int i;
52         int offset = pos & (PAGE_CACHE_SIZE - 1);
53
54         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
55                 size_t count = min_t(size_t,
56                                      PAGE_CACHE_SIZE - offset, write_bytes);
57                 struct page *page = prepared_pages[i];
58                 fault_in_pages_readable(buf, count);
59
60                 /* Copy data from userspace to the current page */
61                 kmap(page);
62                 page_fault = __copy_from_user(page_address(page) + offset,
63                                               buf, count);
64                 /* Flush processor's dcache for this page */
65                 flush_dcache_page(page);
66                 kunmap(page);
67                 buf += count;
68                 write_bytes -= count;
69
70                 if (page_fault)
71                         break;
72         }
73         return page_fault ? -EFAULT : 0;
74 }
75
76 /*
77  * unlocks pages after btrfs_file_write is done with them
78  */
79 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
80 {
81         size_t i;
82         for (i = 0; i < num_pages; i++) {
83                 if (!pages[i])
84                         break;
85                 /* page checked is some magic around finding pages that
86                  * have been modified without going through btrfs_set_page_dirty
87                  * clear it here
88                  */
89                 ClearPageChecked(pages[i]);
90                 unlock_page(pages[i]);
91                 mark_page_accessed(pages[i]);
92                 page_cache_release(pages[i]);
93         }
94 }
95
96 /*
97  * after copy_from_user, pages need to be dirtied and we need to make
98  * sure holes are created between the current EOF and the start of
99  * any next extents (if required).
100  *
101  * this also makes the decision about creating an inline extent vs
102  * doing real data extents, marking pages dirty and delalloc as required.
103  */
104 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
105                                    struct btrfs_root *root,
106                                    struct file *file,
107                                    struct page **pages,
108                                    size_t num_pages,
109                                    loff_t pos,
110                                    size_t write_bytes)
111 {
112         int err = 0;
113         int i;
114         struct inode *inode = fdentry(file)->d_inode;
115         u64 num_bytes;
116         u64 start_pos;
117         u64 end_of_last_block;
118         u64 end_pos = pos + write_bytes;
119         loff_t isize = i_size_read(inode);
120
121         start_pos = pos & ~((u64)root->sectorsize - 1);
122         num_bytes = (write_bytes + pos - start_pos +
123                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
124
125         end_of_last_block = start_pos + num_bytes - 1;
126         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
127         if (err)
128                 return err;
129
130         for (i = 0; i < num_pages; i++) {
131                 struct page *p = pages[i];
132                 SetPageUptodate(p);
133                 ClearPageChecked(p);
134                 set_page_dirty(p);
135         }
136         if (end_pos > isize) {
137                 i_size_write(inode, end_pos);
138                 /* we've only changed i_size in ram, and we haven't updated
139                  * the disk i_size.  There is no need to log the inode
140                  * at this time.
141                  */
142         }
143         return err;
144 }
145
146 /*
147  * this drops all the extents in the cache that intersect the range
148  * [start, end].  Existing extents are split as required.
149  */
150 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
151                             int skip_pinned)
152 {
153         struct extent_map *em;
154         struct extent_map *split = NULL;
155         struct extent_map *split2 = NULL;
156         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
157         u64 len = end - start + 1;
158         int ret;
159         int testend = 1;
160         unsigned long flags;
161         int compressed = 0;
162
163         WARN_ON(end < start);
164         if (end == (u64)-1) {
165                 len = (u64)-1;
166                 testend = 0;
167         }
168         while (1) {
169                 if (!split)
170                         split = alloc_extent_map(GFP_NOFS);
171                 if (!split2)
172                         split2 = alloc_extent_map(GFP_NOFS);
173
174                 write_lock(&em_tree->lock);
175                 em = lookup_extent_mapping(em_tree, start, len);
176                 if (!em) {
177                         write_unlock(&em_tree->lock);
178                         break;
179                 }
180                 flags = em->flags;
181                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
182                         if (em->start <= start &&
183                             (!testend || em->start + em->len >= start + len)) {
184                                 free_extent_map(em);
185                                 write_unlock(&em_tree->lock);
186                                 break;
187                         }
188                         if (start < em->start) {
189                                 len = em->start - start;
190                         } else {
191                                 len = start + len - (em->start + em->len);
192                                 start = em->start + em->len;
193                         }
194                         free_extent_map(em);
195                         write_unlock(&em_tree->lock);
196                         continue;
197                 }
198                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
199                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
200                 remove_extent_mapping(em_tree, em);
201
202                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
203                     em->start < start) {
204                         split->start = em->start;
205                         split->len = start - em->start;
206                         split->orig_start = em->orig_start;
207                         split->block_start = em->block_start;
208
209                         if (compressed)
210                                 split->block_len = em->block_len;
211                         else
212                                 split->block_len = split->len;
213
214                         split->bdev = em->bdev;
215                         split->flags = flags;
216                         ret = add_extent_mapping(em_tree, split);
217                         BUG_ON(ret);
218                         free_extent_map(split);
219                         split = split2;
220                         split2 = NULL;
221                 }
222                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
223                     testend && em->start + em->len > start + len) {
224                         u64 diff = start + len - em->start;
225
226                         split->start = start + len;
227                         split->len = em->start + em->len - (start + len);
228                         split->bdev = em->bdev;
229                         split->flags = flags;
230
231                         if (compressed) {
232                                 split->block_len = em->block_len;
233                                 split->block_start = em->block_start;
234                                 split->orig_start = em->orig_start;
235                         } else {
236                                 split->block_len = split->len;
237                                 split->block_start = em->block_start + diff;
238                                 split->orig_start = split->start;
239                         }
240
241                         ret = add_extent_mapping(em_tree, split);
242                         BUG_ON(ret);
243                         free_extent_map(split);
244                         split = NULL;
245                 }
246                 write_unlock(&em_tree->lock);
247
248                 /* once for us */
249                 free_extent_map(em);
250                 /* once for the tree*/
251                 free_extent_map(em);
252         }
253         if (split)
254                 free_extent_map(split);
255         if (split2)
256                 free_extent_map(split2);
257         return 0;
258 }
259
260 /*
261  * this is very complex, but the basic idea is to drop all extents
262  * in the range start - end.  hint_block is filled in with a block number
263  * that would be a good hint to the block allocator for this file.
264  *
265  * If an extent intersects the range but is not entirely inside the range
266  * it is either truncated or split.  Anything entirely inside the range
267  * is deleted from the tree.
268  *
269  * inline_limit is used to tell this code which offsets in the file to keep
270  * if they contain inline extents.
271  */
272 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
273                        struct btrfs_root *root, struct inode *inode,
274                        u64 start, u64 end, u64 locked_end,
275                        u64 inline_limit, u64 *hint_byte, int drop_cache)
276 {
277         u64 extent_end = 0;
278         u64 search_start = start;
279         u64 ram_bytes = 0;
280         u64 disk_bytenr = 0;
281         u64 orig_locked_end = locked_end;
282         u8 compression;
283         u8 encryption;
284         u16 other_encoding = 0;
285         struct extent_buffer *leaf;
286         struct btrfs_file_extent_item *extent;
287         struct btrfs_path *path;
288         struct btrfs_key key;
289         struct btrfs_file_extent_item old;
290         int keep;
291         int slot;
292         int bookend;
293         int found_type = 0;
294         int found_extent;
295         int found_inline;
296         int recow;
297         int ret;
298
299         inline_limit = 0;
300         if (drop_cache)
301                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
302
303         path = btrfs_alloc_path();
304         if (!path)
305                 return -ENOMEM;
306         while (1) {
307                 recow = 0;
308                 btrfs_release_path(root, path);
309                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
310                                                search_start, -1);
311                 if (ret < 0)
312                         goto out;
313                 if (ret > 0) {
314                         if (path->slots[0] == 0) {
315                                 ret = 0;
316                                 goto out;
317                         }
318                         path->slots[0]--;
319                 }
320 next_slot:
321                 keep = 0;
322                 bookend = 0;
323                 found_extent = 0;
324                 found_inline = 0;
325                 compression = 0;
326                 encryption = 0;
327                 extent = NULL;
328                 leaf = path->nodes[0];
329                 slot = path->slots[0];
330                 ret = 0;
331                 btrfs_item_key_to_cpu(leaf, &key, slot);
332                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
333                     key.offset >= end) {
334                         goto out;
335                 }
336                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
337                     key.objectid != inode->i_ino) {
338                         goto out;
339                 }
340                 if (recow) {
341                         search_start = max(key.offset, start);
342                         continue;
343                 }
344                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
345                         extent = btrfs_item_ptr(leaf, slot,
346                                                 struct btrfs_file_extent_item);
347                         found_type = btrfs_file_extent_type(leaf, extent);
348                         compression = btrfs_file_extent_compression(leaf,
349                                                                     extent);
350                         encryption = btrfs_file_extent_encryption(leaf,
351                                                                   extent);
352                         other_encoding = btrfs_file_extent_other_encoding(leaf,
353                                                                   extent);
354                         if (found_type == BTRFS_FILE_EXTENT_REG ||
355                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
356                                 extent_end =
357                                      btrfs_file_extent_disk_bytenr(leaf,
358                                                                    extent);
359                                 if (extent_end)
360                                         *hint_byte = extent_end;
361
362                                 extent_end = key.offset +
363                                      btrfs_file_extent_num_bytes(leaf, extent);
364                                 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
365                                                                 extent);
366                                 found_extent = 1;
367                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
368                                 found_inline = 1;
369                                 extent_end = key.offset +
370                                      btrfs_file_extent_inline_len(leaf, extent);
371                         }
372                 } else {
373                         extent_end = search_start;
374                 }
375
376                 /* we found nothing we can drop */
377                 if ((!found_extent && !found_inline) ||
378                     search_start >= extent_end) {
379                         int nextret;
380                         u32 nritems;
381                         nritems = btrfs_header_nritems(leaf);
382                         if (slot >= nritems - 1) {
383                                 nextret = btrfs_next_leaf(root, path);
384                                 if (nextret)
385                                         goto out;
386                                 recow = 1;
387                         } else {
388                                 path->slots[0]++;
389                         }
390                         goto next_slot;
391                 }
392
393                 if (end <= extent_end && start >= key.offset && found_inline)
394                         *hint_byte = EXTENT_MAP_INLINE;
395
396                 if (found_extent) {
397                         read_extent_buffer(leaf, &old, (unsigned long)extent,
398                                            sizeof(old));
399                 }
400
401                 if (end < extent_end && end >= key.offset) {
402                         bookend = 1;
403                         if (found_inline && start <= key.offset)
404                                 keep = 1;
405                 }
406
407                 if (bookend && found_extent) {
408                         if (locked_end < extent_end) {
409                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
410                                                 locked_end, extent_end - 1,
411                                                 GFP_NOFS);
412                                 if (!ret) {
413                                         btrfs_release_path(root, path);
414                                         lock_extent(&BTRFS_I(inode)->io_tree,
415                                                 locked_end, extent_end - 1,
416                                                 GFP_NOFS);
417                                         locked_end = extent_end;
418                                         continue;
419                                 }
420                                 locked_end = extent_end;
421                         }
422                         disk_bytenr = le64_to_cpu(old.disk_bytenr);
423                         if (disk_bytenr != 0) {
424                                 ret = btrfs_inc_extent_ref(trans, root,
425                                            disk_bytenr,
426                                            le64_to_cpu(old.disk_num_bytes), 0,
427                                            root->root_key.objectid,
428                                            key.objectid, key.offset -
429                                            le64_to_cpu(old.offset));
430                                 BUG_ON(ret);
431                         }
432                 }
433
434                 if (found_inline) {
435                         u64 mask = root->sectorsize - 1;
436                         search_start = (extent_end + mask) & ~mask;
437                 } else
438                         search_start = extent_end;
439
440                 /* truncate existing extent */
441                 if (start > key.offset) {
442                         u64 new_num;
443                         u64 old_num;
444                         keep = 1;
445                         WARN_ON(start & (root->sectorsize - 1));
446                         if (found_extent) {
447                                 new_num = start - key.offset;
448                                 old_num = btrfs_file_extent_num_bytes(leaf,
449                                                                       extent);
450                                 *hint_byte =
451                                         btrfs_file_extent_disk_bytenr(leaf,
452                                                                       extent);
453                                 if (btrfs_file_extent_disk_bytenr(leaf,
454                                                                   extent)) {
455                                         inode_sub_bytes(inode, old_num -
456                                                         new_num);
457                                 }
458                                 btrfs_set_file_extent_num_bytes(leaf,
459                                                         extent, new_num);
460                                 btrfs_mark_buffer_dirty(leaf);
461                         } else if (key.offset < inline_limit &&
462                                    (end > extent_end) &&
463                                    (inline_limit < extent_end)) {
464                                 u32 new_size;
465                                 new_size = btrfs_file_extent_calc_inline_size(
466                                                    inline_limit - key.offset);
467                                 inode_sub_bytes(inode, extent_end -
468                                                 inline_limit);
469                                 btrfs_set_file_extent_ram_bytes(leaf, extent,
470                                                         new_size);
471                                 if (!compression && !encryption) {
472                                         btrfs_truncate_item(trans, root, path,
473                                                             new_size, 1);
474                                 }
475                         }
476                 }
477                 /* delete the entire extent */
478                 if (!keep) {
479                         if (found_inline)
480                                 inode_sub_bytes(inode, extent_end -
481                                                 key.offset);
482                         ret = btrfs_del_item(trans, root, path);
483                         /* TODO update progress marker and return */
484                         BUG_ON(ret);
485                         extent = NULL;
486                         btrfs_release_path(root, path);
487                         /* the extent will be freed later */
488                 }
489                 if (bookend && found_inline && start <= key.offset) {
490                         u32 new_size;
491                         new_size = btrfs_file_extent_calc_inline_size(
492                                                    extent_end - end);
493                         inode_sub_bytes(inode, end - key.offset);
494                         btrfs_set_file_extent_ram_bytes(leaf, extent,
495                                                         new_size);
496                         if (!compression && !encryption)
497                                 ret = btrfs_truncate_item(trans, root, path,
498                                                           new_size, 0);
499                         BUG_ON(ret);
500                 }
501                 /* create bookend, splitting the extent in two */
502                 if (bookend && found_extent) {
503                         struct btrfs_key ins;
504                         ins.objectid = inode->i_ino;
505                         ins.offset = end;
506                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
507
508                         btrfs_release_path(root, path);
509                         path->leave_spinning = 1;
510                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
511                                                       sizeof(*extent));
512                         BUG_ON(ret);
513
514                         leaf = path->nodes[0];
515                         extent = btrfs_item_ptr(leaf, path->slots[0],
516                                                 struct btrfs_file_extent_item);
517                         write_extent_buffer(leaf, &old,
518                                             (unsigned long)extent, sizeof(old));
519
520                         btrfs_set_file_extent_compression(leaf, extent,
521                                                           compression);
522                         btrfs_set_file_extent_encryption(leaf, extent,
523                                                          encryption);
524                         btrfs_set_file_extent_other_encoding(leaf, extent,
525                                                              other_encoding);
526                         btrfs_set_file_extent_offset(leaf, extent,
527                                     le64_to_cpu(old.offset) + end - key.offset);
528                         WARN_ON(le64_to_cpu(old.num_bytes) <
529                                 (extent_end - end));
530                         btrfs_set_file_extent_num_bytes(leaf, extent,
531                                                         extent_end - end);
532
533                         /*
534                          * set the ram bytes to the size of the full extent
535                          * before splitting.  This is a worst case flag,
536                          * but its the best we can do because we don't know
537                          * how splitting affects compression
538                          */
539                         btrfs_set_file_extent_ram_bytes(leaf, extent,
540                                                         ram_bytes);
541                         btrfs_set_file_extent_type(leaf, extent, found_type);
542
543                         btrfs_unlock_up_safe(path, 1);
544                         btrfs_mark_buffer_dirty(path->nodes[0]);
545                         btrfs_set_lock_blocking(path->nodes[0]);
546
547                         path->leave_spinning = 0;
548                         btrfs_release_path(root, path);
549                         if (disk_bytenr != 0)
550                                 inode_add_bytes(inode, extent_end - end);
551                 }
552
553                 if (found_extent && !keep) {
554                         u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
555
556                         if (old_disk_bytenr != 0) {
557                                 inode_sub_bytes(inode,
558                                                 le64_to_cpu(old.num_bytes));
559                                 ret = btrfs_free_extent(trans, root,
560                                                 old_disk_bytenr,
561                                                 le64_to_cpu(old.disk_num_bytes),
562                                                 0, root->root_key.objectid,
563                                                 key.objectid, key.offset -
564                                                 le64_to_cpu(old.offset));
565                                 BUG_ON(ret);
566                                 *hint_byte = old_disk_bytenr;
567                         }
568                 }
569
570                 if (search_start >= end) {
571                         ret = 0;
572                         goto out;
573                 }
574         }
575 out:
576         btrfs_free_path(path);
577         if (locked_end > orig_locked_end) {
578                 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
579                               locked_end - 1, GFP_NOFS);
580         }
581         return ret;
582 }
583
584 static int extent_mergeable(struct extent_buffer *leaf, int slot,
585                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
586 {
587         struct btrfs_file_extent_item *fi;
588         struct btrfs_key key;
589         u64 extent_end;
590
591         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
592                 return 0;
593
594         btrfs_item_key_to_cpu(leaf, &key, slot);
595         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
596                 return 0;
597
598         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
599         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
600             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
601             btrfs_file_extent_compression(leaf, fi) ||
602             btrfs_file_extent_encryption(leaf, fi) ||
603             btrfs_file_extent_other_encoding(leaf, fi))
604                 return 0;
605
606         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
607         if ((*start && *start != key.offset) || (*end && *end != extent_end))
608                 return 0;
609
610         *start = key.offset;
611         *end = extent_end;
612         return 1;
613 }
614
615 /*
616  * Mark extent in the range start - end as written.
617  *
618  * This changes extent type from 'pre-allocated' to 'regular'. If only
619  * part of extent is marked as written, the extent will be split into
620  * two or three.
621  */
622 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
623                               struct btrfs_root *root,
624                               struct inode *inode, u64 start, u64 end)
625 {
626         struct extent_buffer *leaf;
627         struct btrfs_path *path;
628         struct btrfs_file_extent_item *fi;
629         struct btrfs_key key;
630         u64 bytenr;
631         u64 num_bytes;
632         u64 extent_end;
633         u64 orig_offset;
634         u64 other_start;
635         u64 other_end;
636         u64 split = start;
637         u64 locked_end = end;
638         int extent_type;
639         int split_end = 1;
640         int ret;
641
642         btrfs_drop_extent_cache(inode, start, end - 1, 0);
643
644         path = btrfs_alloc_path();
645         BUG_ON(!path);
646 again:
647         key.objectid = inode->i_ino;
648         key.type = BTRFS_EXTENT_DATA_KEY;
649         if (split == start)
650                 key.offset = split;
651         else
652                 key.offset = split - 1;
653
654         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
655         if (ret > 0 && path->slots[0] > 0)
656                 path->slots[0]--;
657
658         leaf = path->nodes[0];
659         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
660         BUG_ON(key.objectid != inode->i_ino ||
661                key.type != BTRFS_EXTENT_DATA_KEY);
662         fi = btrfs_item_ptr(leaf, path->slots[0],
663                             struct btrfs_file_extent_item);
664         extent_type = btrfs_file_extent_type(leaf, fi);
665         BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
666         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
667         BUG_ON(key.offset > start || extent_end < end);
668
669         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
670         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
671         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
672
673         if (key.offset == start)
674                 split = end;
675
676         if (key.offset == start && extent_end == end) {
677                 int del_nr = 0;
678                 int del_slot = 0;
679                 other_start = end;
680                 other_end = 0;
681                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
682                                      bytenr, &other_start, &other_end)) {
683                         extent_end = other_end;
684                         del_slot = path->slots[0] + 1;
685                         del_nr++;
686                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
687                                                 0, root->root_key.objectid,
688                                                 inode->i_ino, orig_offset);
689                         BUG_ON(ret);
690                 }
691                 other_start = 0;
692                 other_end = start;
693                 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
694                                      bytenr, &other_start, &other_end)) {
695                         key.offset = other_start;
696                         del_slot = path->slots[0];
697                         del_nr++;
698                         ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
699                                                 0, root->root_key.objectid,
700                                                 inode->i_ino, orig_offset);
701                         BUG_ON(ret);
702                 }
703                 split_end = 0;
704                 if (del_nr == 0) {
705                         btrfs_set_file_extent_type(leaf, fi,
706                                                    BTRFS_FILE_EXTENT_REG);
707                         goto done;
708                 }
709
710                 fi = btrfs_item_ptr(leaf, del_slot - 1,
711                                     struct btrfs_file_extent_item);
712                 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
713                 btrfs_set_file_extent_num_bytes(leaf, fi,
714                                                 extent_end - key.offset);
715                 btrfs_mark_buffer_dirty(leaf);
716
717                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
718                 BUG_ON(ret);
719                 goto release;
720         } else if (split == start) {
721                 if (locked_end < extent_end) {
722                         ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
723                                         locked_end, extent_end - 1, GFP_NOFS);
724                         if (!ret) {
725                                 btrfs_release_path(root, path);
726                                 lock_extent(&BTRFS_I(inode)->io_tree,
727                                         locked_end, extent_end - 1, GFP_NOFS);
728                                 locked_end = extent_end;
729                                 goto again;
730                         }
731                         locked_end = extent_end;
732                 }
733                 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
734         } else  {
735                 BUG_ON(key.offset != start);
736                 key.offset = split;
737                 btrfs_set_file_extent_offset(leaf, fi, key.offset -
738                                              orig_offset);
739                 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
740                 btrfs_set_item_key_safe(trans, root, path, &key);
741                 extent_end = split;
742         }
743
744         if (extent_end == end) {
745                 split_end = 0;
746                 extent_type = BTRFS_FILE_EXTENT_REG;
747         }
748         if (extent_end == end && split == start) {
749                 other_start = end;
750                 other_end = 0;
751                 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
752                                      bytenr, &other_start, &other_end)) {
753                         path->slots[0]++;
754                         fi = btrfs_item_ptr(leaf, path->slots[0],
755                                             struct btrfs_file_extent_item);
756                         key.offset = split;
757                         btrfs_set_item_key_safe(trans, root, path, &key);
758                         btrfs_set_file_extent_offset(leaf, fi, key.offset -
759                                                      orig_offset);
760                         btrfs_set_file_extent_num_bytes(leaf, fi,
761                                                         other_end - split);
762                         goto done;
763                 }
764         }
765         if (extent_end == end && split == end) {
766                 other_start = 0;
767                 other_end = start;
768                 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
769                                      bytenr, &other_start, &other_end)) {
770                         path->slots[0]--;
771                         fi = btrfs_item_ptr(leaf, path->slots[0],
772                                             struct btrfs_file_extent_item);
773                         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
774                                                         other_start);
775                         goto done;
776                 }
777         }
778
779         btrfs_mark_buffer_dirty(leaf);
780
781         ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
782                                    root->root_key.objectid,
783                                    inode->i_ino, orig_offset);
784         BUG_ON(ret);
785         btrfs_release_path(root, path);
786
787         key.offset = start;
788         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
789         BUG_ON(ret);
790
791         leaf = path->nodes[0];
792         fi = btrfs_item_ptr(leaf, path->slots[0],
793                             struct btrfs_file_extent_item);
794         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
795         btrfs_set_file_extent_type(leaf, fi, extent_type);
796         btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
797         btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
798         btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
799         btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
800         btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
801         btrfs_set_file_extent_compression(leaf, fi, 0);
802         btrfs_set_file_extent_encryption(leaf, fi, 0);
803         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
804 done:
805         btrfs_mark_buffer_dirty(leaf);
806
807 release:
808         btrfs_release_path(root, path);
809         if (split_end && split == start) {
810                 split = end;
811                 goto again;
812         }
813         if (locked_end > end) {
814                 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
815                               GFP_NOFS);
816         }
817         btrfs_free_path(path);
818         return 0;
819 }
820
821 /*
822  * this gets pages into the page cache and locks them down, it also properly
823  * waits for data=ordered extents to finish before allowing the pages to be
824  * modified.
825  */
826 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
827                          struct page **pages, size_t num_pages,
828                          loff_t pos, unsigned long first_index,
829                          unsigned long last_index, size_t write_bytes)
830 {
831         int i;
832         unsigned long index = pos >> PAGE_CACHE_SHIFT;
833         struct inode *inode = fdentry(file)->d_inode;
834         int err = 0;
835         u64 start_pos;
836         u64 last_pos;
837
838         start_pos = pos & ~((u64)root->sectorsize - 1);
839         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
840
841         if (start_pos > inode->i_size) {
842                 err = btrfs_cont_expand(inode, start_pos);
843                 if (err)
844                         return err;
845         }
846
847         memset(pages, 0, num_pages * sizeof(struct page *));
848 again:
849         for (i = 0; i < num_pages; i++) {
850                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
851                 if (!pages[i]) {
852                         err = -ENOMEM;
853                         BUG_ON(1);
854                 }
855                 wait_on_page_writeback(pages[i]);
856         }
857         if (start_pos < inode->i_size) {
858                 struct btrfs_ordered_extent *ordered;
859                 lock_extent(&BTRFS_I(inode)->io_tree,
860                             start_pos, last_pos - 1, GFP_NOFS);
861                 ordered = btrfs_lookup_first_ordered_extent(inode,
862                                                             last_pos - 1);
863                 if (ordered &&
864                     ordered->file_offset + ordered->len > start_pos &&
865                     ordered->file_offset < last_pos) {
866                         btrfs_put_ordered_extent(ordered);
867                         unlock_extent(&BTRFS_I(inode)->io_tree,
868                                       start_pos, last_pos - 1, GFP_NOFS);
869                         for (i = 0; i < num_pages; i++) {
870                                 unlock_page(pages[i]);
871                                 page_cache_release(pages[i]);
872                         }
873                         btrfs_wait_ordered_range(inode, start_pos,
874                                                  last_pos - start_pos);
875                         goto again;
876                 }
877                 if (ordered)
878                         btrfs_put_ordered_extent(ordered);
879
880                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
881                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
882                                   EXTENT_DO_ACCOUNTING,
883                                   GFP_NOFS);
884                 unlock_extent(&BTRFS_I(inode)->io_tree,
885                               start_pos, last_pos - 1, GFP_NOFS);
886         }
887         for (i = 0; i < num_pages; i++) {
888                 clear_page_dirty_for_io(pages[i]);
889                 set_page_extent_mapped(pages[i]);
890                 WARN_ON(!PageLocked(pages[i]));
891         }
892         return 0;
893 }
894
895 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
896                                 size_t count, loff_t *ppos)
897 {
898         loff_t pos;
899         loff_t start_pos;
900         ssize_t num_written = 0;
901         ssize_t err = 0;
902         int ret = 0;
903         struct inode *inode = fdentry(file)->d_inode;
904         struct btrfs_root *root = BTRFS_I(inode)->root;
905         struct page **pages = NULL;
906         int nrptrs;
907         struct page *pinned[2];
908         unsigned long first_index;
909         unsigned long last_index;
910         int will_write;
911
912         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
913                       (file->f_flags & O_DIRECT));
914
915         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
916                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
917         pinned[0] = NULL;
918         pinned[1] = NULL;
919
920         pos = *ppos;
921         start_pos = pos;
922
923         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
924
925         /* do the reserve before the mutex lock in case we have to do some
926          * flushing.  We wouldn't deadlock, but this is more polite.
927          */
928         err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
929         if (err)
930                 goto out_nolock;
931
932         mutex_lock(&inode->i_mutex);
933
934         current->backing_dev_info = inode->i_mapping->backing_dev_info;
935         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
936         if (err)
937                 goto out;
938
939         if (count == 0)
940                 goto out;
941
942         err = file_remove_suid(file);
943         if (err)
944                 goto out;
945
946         file_update_time(file);
947
948         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
949
950         /* generic_write_checks can change our pos */
951         start_pos = pos;
952
953         BTRFS_I(inode)->sequence++;
954         first_index = pos >> PAGE_CACHE_SHIFT;
955         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
956
957         /*
958          * there are lots of better ways to do this, but this code
959          * makes sure the first and last page in the file range are
960          * up to date and ready for cow
961          */
962         if ((pos & (PAGE_CACHE_SIZE - 1))) {
963                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
964                 if (!PageUptodate(pinned[0])) {
965                         ret = btrfs_readpage(NULL, pinned[0]);
966                         BUG_ON(ret);
967                         wait_on_page_locked(pinned[0]);
968                 } else {
969                         unlock_page(pinned[0]);
970                 }
971         }
972         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
973                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
974                 if (!PageUptodate(pinned[1])) {
975                         ret = btrfs_readpage(NULL, pinned[1]);
976                         BUG_ON(ret);
977                         wait_on_page_locked(pinned[1]);
978                 } else {
979                         unlock_page(pinned[1]);
980                 }
981         }
982
983         while (count > 0) {
984                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
985                 size_t write_bytes = min(count, nrptrs *
986                                         (size_t)PAGE_CACHE_SIZE -
987                                          offset);
988                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
989                                         PAGE_CACHE_SHIFT;
990
991                 WARN_ON(num_pages > nrptrs);
992                 memset(pages, 0, sizeof(struct page *) * nrptrs);
993
994                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
995                 if (ret)
996                         goto out;
997
998                 ret = prepare_pages(root, file, pages, num_pages,
999                                     pos, first_index, last_index,
1000                                     write_bytes);
1001                 if (ret) {
1002                         btrfs_free_reserved_data_space(root, inode,
1003                                                        write_bytes);
1004                         goto out;
1005                 }
1006
1007                 ret = btrfs_copy_from_user(pos, num_pages,
1008                                            write_bytes, pages, buf);
1009                 if (ret) {
1010                         btrfs_free_reserved_data_space(root, inode,
1011                                                        write_bytes);
1012                         btrfs_drop_pages(pages, num_pages);
1013                         goto out;
1014                 }
1015
1016                 ret = dirty_and_release_pages(NULL, root, file, pages,
1017                                               num_pages, pos, write_bytes);
1018                 btrfs_drop_pages(pages, num_pages);
1019                 if (ret) {
1020                         btrfs_free_reserved_data_space(root, inode,
1021                                                        write_bytes);
1022                         goto out;
1023                 }
1024
1025                 if (will_write) {
1026                         filemap_fdatawrite_range(inode->i_mapping, pos,
1027                                                  pos + write_bytes - 1);
1028                 } else {
1029                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1030                                                            num_pages);
1031                         if (num_pages <
1032                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1033                                 btrfs_btree_balance_dirty(root, 1);
1034                         btrfs_throttle(root);
1035                 }
1036
1037                 buf += write_bytes;
1038                 count -= write_bytes;
1039                 pos += write_bytes;
1040                 num_written += write_bytes;
1041
1042                 cond_resched();
1043         }
1044 out:
1045         mutex_unlock(&inode->i_mutex);
1046         if (ret)
1047                 err = ret;
1048         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1049
1050 out_nolock:
1051         kfree(pages);
1052         if (pinned[0])
1053                 page_cache_release(pinned[0]);
1054         if (pinned[1])
1055                 page_cache_release(pinned[1]);
1056         *ppos = pos;
1057
1058         /*
1059          * we want to make sure fsync finds this change
1060          * but we haven't joined a transaction running right now.
1061          *
1062          * Later on, someone is sure to update the inode and get the
1063          * real transid recorded.
1064          *
1065          * We set last_trans now to the fs_info generation + 1,
1066          * this will either be one more than the running transaction
1067          * or the generation used for the next transaction if there isn't
1068          * one running right now.
1069          */
1070         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1071
1072         if (num_written > 0 && will_write) {
1073                 struct btrfs_trans_handle *trans;
1074
1075                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1076                 if (err)
1077                         num_written = err;
1078
1079                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1080                         trans = btrfs_start_transaction(root, 1);
1081                         ret = btrfs_log_dentry_safe(trans, root,
1082                                                     file->f_dentry);
1083                         if (ret == 0) {
1084                                 ret = btrfs_sync_log(trans, root);
1085                                 if (ret == 0)
1086                                         btrfs_end_transaction(trans, root);
1087                                 else
1088                                         btrfs_commit_transaction(trans, root);
1089                         } else if (ret != BTRFS_NO_LOG_SYNC) {
1090                                 btrfs_commit_transaction(trans, root);
1091                         } else {
1092                                 btrfs_end_transaction(trans, root);
1093                         }
1094                 }
1095                 if (file->f_flags & O_DIRECT) {
1096                         invalidate_mapping_pages(inode->i_mapping,
1097                               start_pos >> PAGE_CACHE_SHIFT,
1098                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1099                 }
1100         }
1101         current->backing_dev_info = NULL;
1102         return num_written ? num_written : err;
1103 }
1104
1105 int btrfs_release_file(struct inode *inode, struct file *filp)
1106 {
1107         /*
1108          * ordered_data_close is set by settattr when we are about to truncate
1109          * a file from a non-zero size to a zero size.  This tries to
1110          * flush down new bytes that may have been written if the
1111          * application were using truncate to replace a file in place.
1112          */
1113         if (BTRFS_I(inode)->ordered_data_close) {
1114                 BTRFS_I(inode)->ordered_data_close = 0;
1115                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1116                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1117                         filemap_flush(inode->i_mapping);
1118         }
1119         if (filp->private_data)
1120                 btrfs_ioctl_trans_end(filp);
1121         return 0;
1122 }
1123
1124 /*
1125  * fsync call for both files and directories.  This logs the inode into
1126  * the tree log instead of forcing full commits whenever possible.
1127  *
1128  * It needs to call filemap_fdatawait so that all ordered extent updates are
1129  * in the metadata btree are up to date for copying to the log.
1130  *
1131  * It drops the inode mutex before doing the tree log commit.  This is an
1132  * important optimization for directories because holding the mutex prevents
1133  * new operations on the dir while we write to disk.
1134  */
1135 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1136 {
1137         struct inode *inode = dentry->d_inode;
1138         struct btrfs_root *root = BTRFS_I(inode)->root;
1139         int ret = 0;
1140         struct btrfs_trans_handle *trans;
1141
1142
1143         /* we wait first, since the writeback may change the inode */
1144         root->log_batch++;
1145         /* the VFS called filemap_fdatawrite for us */
1146         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1147         root->log_batch++;
1148
1149         /*
1150          * check the transaction that last modified this inode
1151          * and see if its already been committed
1152          */
1153         if (!BTRFS_I(inode)->last_trans)
1154                 goto out;
1155
1156         /*
1157          * if the last transaction that changed this file was before
1158          * the current transaction, we can bail out now without any
1159          * syncing
1160          */
1161         mutex_lock(&root->fs_info->trans_mutex);
1162         if (BTRFS_I(inode)->last_trans <=
1163             root->fs_info->last_trans_committed) {
1164                 BTRFS_I(inode)->last_trans = 0;
1165                 mutex_unlock(&root->fs_info->trans_mutex);
1166                 goto out;
1167         }
1168         mutex_unlock(&root->fs_info->trans_mutex);
1169
1170         /*
1171          * ok we haven't committed the transaction yet, lets do a commit
1172          */
1173         if (file && file->private_data)
1174                 btrfs_ioctl_trans_end(file);
1175
1176         trans = btrfs_start_transaction(root, 1);
1177         if (!trans) {
1178                 ret = -ENOMEM;
1179                 goto out;
1180         }
1181
1182         ret = btrfs_log_dentry_safe(trans, root, dentry);
1183         if (ret < 0)
1184                 goto out;
1185
1186         /* we've logged all the items and now have a consistent
1187          * version of the file in the log.  It is possible that
1188          * someone will come in and modify the file, but that's
1189          * fine because the log is consistent on disk, and we
1190          * have references to all of the file's extents
1191          *
1192          * It is possible that someone will come in and log the
1193          * file again, but that will end up using the synchronization
1194          * inside btrfs_sync_log to keep things safe.
1195          */
1196         mutex_unlock(&dentry->d_inode->i_mutex);
1197
1198         if (ret != BTRFS_NO_LOG_SYNC) {
1199                 if (ret > 0) {
1200                         ret = btrfs_commit_transaction(trans, root);
1201                 } else {
1202                         ret = btrfs_sync_log(trans, root);
1203                         if (ret == 0)
1204                                 ret = btrfs_end_transaction(trans, root);
1205                         else
1206                                 ret = btrfs_commit_transaction(trans, root);
1207                 }
1208         } else {
1209                 ret = btrfs_end_transaction(trans, root);
1210         }
1211         mutex_lock(&dentry->d_inode->i_mutex);
1212 out:
1213         return ret > 0 ? EIO : ret;
1214 }
1215
1216 static const struct vm_operations_struct btrfs_file_vm_ops = {
1217         .fault          = filemap_fault,
1218         .page_mkwrite   = btrfs_page_mkwrite,
1219 };
1220
1221 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1222 {
1223         vma->vm_ops = &btrfs_file_vm_ops;
1224         file_accessed(filp);
1225         return 0;
1226 }
1227
1228 const struct file_operations btrfs_file_operations = {
1229         .llseek         = generic_file_llseek,
1230         .read           = do_sync_read,
1231         .aio_read       = generic_file_aio_read,
1232         .splice_read    = generic_file_splice_read,
1233         .write          = btrfs_file_write,
1234         .mmap           = btrfs_file_mmap,
1235         .open           = generic_file_open,
1236         .release        = btrfs_release_file,
1237         .fsync          = btrfs_sync_file,
1238         .unlocked_ioctl = btrfs_ioctl,
1239 #ifdef CONFIG_COMPAT
1240         .compat_ioctl   = btrfs_ioctl,
1241 #endif
1242 };