3e4e5c227c0c33e72d93b0b9fdde24bb151870c7
[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/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/version.h>
33 #include "ctree.h"
34 #include "disk-io.h"
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "ioctl.h"
38 #include "print-tree.h"
39 #include "compat.h"
40
41
42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
43                                 struct page **prepared_pages,
44                                 const char __user * buf)
45 {
46         long page_fault = 0;
47         int i;
48         int offset = pos & (PAGE_CACHE_SIZE - 1);
49
50         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
51                 size_t count = min_t(size_t,
52                                      PAGE_CACHE_SIZE - offset, write_bytes);
53                 struct page *page = prepared_pages[i];
54                 fault_in_pages_readable(buf, count);
55
56                 /* Copy data from userspace to the current page */
57                 kmap(page);
58                 page_fault = __copy_from_user(page_address(page) + offset,
59                                               buf, count);
60                 /* Flush processor's dcache for this page */
61                 flush_dcache_page(page);
62                 kunmap(page);
63                 buf += count;
64                 write_bytes -= count;
65
66                 if (page_fault)
67                         break;
68         }
69         return page_fault ? -EFAULT : 0;
70 }
71
72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
73 {
74         size_t i;
75         for (i = 0; i < num_pages; i++) {
76                 if (!pages[i])
77                         break;
78                 unlock_page(pages[i]);
79                 mark_page_accessed(pages[i]);
80                 page_cache_release(pages[i]);
81         }
82 }
83
84 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
85                                 struct btrfs_root *root, struct inode *inode,
86                                 u64 offset, size_t size,
87                                 struct page **pages, size_t page_offset,
88                                 int num_pages)
89 {
90         struct btrfs_key key;
91         struct btrfs_path *path;
92         struct extent_buffer *leaf;
93         char *kaddr;
94         unsigned long ptr;
95         struct btrfs_file_extent_item *ei;
96         struct page *page;
97         u32 datasize;
98         int err = 0;
99         int ret;
100         int i;
101         ssize_t cur_size;
102
103         path = btrfs_alloc_path();
104         if (!path)
105                 return -ENOMEM;
106
107         btrfs_set_trans_block_group(trans, inode);
108
109         key.objectid = inode->i_ino;
110         key.offset = offset;
111         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
112
113         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
114         if (ret < 0) {
115                 err = ret;
116                 goto fail;
117         }
118         if (ret == 1) {
119                 struct btrfs_key found_key;
120
121                 if (path->slots[0] == 0)
122                         goto insert;
123
124                 path->slots[0]--;
125                 leaf = path->nodes[0];
126                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
127
128                 if (found_key.objectid != inode->i_ino)
129                         goto insert;
130
131                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
132                         goto insert;
133                 ei = btrfs_item_ptr(leaf, path->slots[0],
134                                     struct btrfs_file_extent_item);
135
136                 if (btrfs_file_extent_type(leaf, ei) !=
137                     BTRFS_FILE_EXTENT_INLINE) {
138                         goto insert;
139                 }
140                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
141                 ret = 0;
142         }
143         if (ret == 0) {
144                 u32 found_size;
145                 u64 found_end;
146
147                 leaf = path->nodes[0];
148                 ei = btrfs_item_ptr(leaf, path->slots[0],
149                                     struct btrfs_file_extent_item);
150
151                 if (btrfs_file_extent_type(leaf, ei) !=
152                     BTRFS_FILE_EXTENT_INLINE) {
153                         err = ret;
154                         btrfs_print_leaf(root, leaf);
155                         printk("found wasn't inline offset %Lu inode %lu\n",
156                                offset, inode->i_ino);
157                         goto fail;
158                 }
159                 found_size = btrfs_file_extent_inline_len(leaf,
160                                           btrfs_item_nr(leaf, path->slots[0]));
161                 found_end = key.offset + found_size;
162
163                 if (found_end < offset + size) {
164                         btrfs_release_path(root, path);
165                         ret = btrfs_search_slot(trans, root, &key, path,
166                                                 offset + size - found_end, 1);
167                         BUG_ON(ret != 0);
168
169                         ret = btrfs_extend_item(trans, root, path,
170                                                 offset + size - found_end);
171                         if (ret) {
172                                 err = ret;
173                                 goto fail;
174                         }
175                         leaf = path->nodes[0];
176                         ei = btrfs_item_ptr(leaf, path->slots[0],
177                                             struct btrfs_file_extent_item);
178                         inode->i_blocks += (offset + size - found_end) >> 9;
179                 }
180                 if (found_end < offset) {
181                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
182                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
183                 }
184         } else {
185 insert:
186                 btrfs_release_path(root, path);
187                 datasize = offset + size - key.offset;
188                 inode->i_blocks += datasize >> 9;
189                 datasize = btrfs_file_extent_calc_inline_size(datasize);
190                 ret = btrfs_insert_empty_item(trans, root, path, &key,
191                                               datasize);
192                 if (ret) {
193                         err = ret;
194                         printk("got bad ret %d\n", ret);
195                         goto fail;
196                 }
197                 leaf = path->nodes[0];
198                 ei = btrfs_item_ptr(leaf, path->slots[0],
199                                     struct btrfs_file_extent_item);
200                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
201                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
202         }
203         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
204
205         cur_size = size;
206         i = 0;
207         while (size > 0) {
208                 page = pages[i];
209                 kaddr = kmap_atomic(page, KM_USER0);
210                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
211                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
212                 kunmap_atomic(kaddr, KM_USER0);
213                 page_offset = 0;
214                 ptr += cur_size;
215                 size -= cur_size;
216                 if (i >= num_pages) {
217                         printk("i %d num_pages %d\n", i, num_pages);
218                 }
219                 i++;
220         }
221         btrfs_mark_buffer_dirty(leaf);
222 fail:
223         btrfs_free_path(path);
224         return err;
225 }
226
227 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
228                                    struct btrfs_root *root,
229                                    struct file *file,
230                                    struct page **pages,
231                                    size_t num_pages,
232                                    loff_t pos,
233                                    size_t write_bytes)
234 {
235         int err = 0;
236         int i;
237         struct inode *inode = fdentry(file)->d_inode;
238         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
239         u64 hint_byte;
240         u64 num_bytes;
241         u64 start_pos;
242         u64 end_of_last_block;
243         u64 end_pos = pos + write_bytes;
244         u64 inline_size;
245         loff_t isize = i_size_read(inode);
246
247         start_pos = pos & ~((u64)root->sectorsize - 1);
248         num_bytes = (write_bytes + pos - start_pos +
249                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
250
251         end_of_last_block = start_pos + num_bytes - 1;
252
253         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
254         trans = btrfs_start_transaction(root, 1);
255         if (!trans) {
256                 err = -ENOMEM;
257                 goto out_unlock;
258         }
259         btrfs_set_trans_block_group(trans, inode);
260         hint_byte = 0;
261
262         if ((end_of_last_block & 4095) == 0) {
263                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
264         }
265         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
266
267         /* FIXME...EIEIO, ENOSPC and more */
268         /* insert any holes we need to create */
269         if (isize < start_pos) {
270                 u64 last_pos_in_file;
271                 u64 hole_size;
272                 u64 mask = root->sectorsize - 1;
273                 last_pos_in_file = (isize + mask) & ~mask;
274                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
275                 if (hole_size > 0) {
276                         btrfs_wait_ordered_range(inode, last_pos_in_file,
277                                                  last_pos_in_file + hole_size);
278                         err = btrfs_drop_extents(trans, root, inode,
279                                                  last_pos_in_file,
280                                                  last_pos_in_file + hole_size,
281                                                  last_pos_in_file,
282                                                  &hint_byte);
283                         if (err)
284                                 goto failed;
285
286                         err = btrfs_insert_file_extent(trans, root,
287                                                        inode->i_ino,
288                                                        last_pos_in_file,
289                                                        0, 0, hole_size, 0);
290                         btrfs_drop_extent_cache(inode, last_pos_in_file,
291                                         last_pos_in_file + hole_size -1);
292                         btrfs_check_file(root, inode);
293                 }
294                 if (err)
295                         goto failed;
296         }
297
298         /*
299          * either allocate an extent for the new bytes or setup the key
300          * to show we are doing inline data in the extent
301          */
302         inline_size = end_pos;
303         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
304             inline_size > root->fs_info->max_inline ||
305             (inline_size & (root->sectorsize -1)) == 0 ||
306             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
307                 /* check for reserved extents on each page, we don't want
308                  * to reset the delalloc bit on things that already have
309                  * extents reserved.
310                  */
311                 set_extent_delalloc(io_tree, start_pos,
312                                     end_of_last_block, GFP_NOFS);
313                 for (i = 0; i < num_pages; i++) {
314                         struct page *p = pages[i];
315                         SetPageUptodate(p);
316                         ClearPageChecked(p);
317                         set_page_dirty(p);
318                 }
319         } else {
320                 u64 aligned_end;
321                 /* step one, delete the existing extents in this range */
322                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
323                         ~((u64)root->sectorsize - 1);
324                 err = btrfs_drop_extents(trans, root, inode, start_pos,
325                                          aligned_end, aligned_end, &hint_byte);
326                 if (err)
327                         goto failed;
328                 if (isize > inline_size)
329                         inline_size = min_t(u64, isize, aligned_end);
330                 inline_size -= start_pos;
331                 err = insert_inline_extent(trans, root, inode, start_pos,
332                                            inline_size, pages, 0, num_pages);
333                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
334                 BUG_ON(err);
335         }
336         if (end_pos > isize) {
337                 i_size_write(inode, end_pos);
338                 btrfs_update_inode(trans, root, inode);
339         }
340 failed:
341         err = btrfs_end_transaction_throttle(trans, root);
342 out_unlock:
343         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
344         return err;
345 }
346
347 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
348 {
349         struct extent_map *em;
350         struct extent_map *split = NULL;
351         struct extent_map *split2 = NULL;
352         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
353         struct extent_map *tmp;
354         u64 len = end - start + 1;
355         u64 next_start;
356         int ret;
357         int testend = 1;
358
359         WARN_ON(end < start);
360         if (end == (u64)-1) {
361                 len = (u64)-1;
362                 testend = 0;
363         }
364         while(1) {
365                 if (!split)
366                         split = alloc_extent_map(GFP_NOFS);
367                 if (!split2)
368                         split2 = alloc_extent_map(GFP_NOFS);
369
370                 spin_lock(&em_tree->lock);
371                 em = lookup_extent_mapping(em_tree, start, len);
372                 if (!em) {
373                         spin_unlock(&em_tree->lock);
374                         break;
375                 }
376                 tmp = rb_entry(&em->rb_node, struct extent_map, rb_node);
377                 next_start = tmp->start;
378                 remove_extent_mapping(em_tree, em);
379
380                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
381                     em->start < start) {
382                         split->start = em->start;
383                         split->len = start - em->start;
384                         split->block_start = em->block_start;
385                         split->bdev = em->bdev;
386                         split->flags = em->flags;
387                         ret = add_extent_mapping(em_tree, split);
388                         BUG_ON(ret);
389                         free_extent_map(split);
390                         split = split2;
391                         split2 = NULL;
392                 }
393                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
394                     testend && em->start + em->len > start + len) {
395                         u64 diff = start + len - em->start;
396
397                         split->start = start + len;
398                         split->len = em->start + em->len - (start + len);
399                         split->bdev = em->bdev;
400                         split->flags = em->flags;
401
402                         split->block_start = em->block_start + diff;
403
404                         ret = add_extent_mapping(em_tree, split);
405                         BUG_ON(ret);
406                         free_extent_map(split);
407                         split = NULL;
408                 }
409                 spin_unlock(&em_tree->lock);
410
411                 /* once for us */
412                 free_extent_map(em);
413                 /* once for the tree*/
414                 free_extent_map(em);
415         }
416         if (split)
417                 free_extent_map(split);
418         if (split2)
419                 free_extent_map(split2);
420         return 0;
421 }
422
423 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
424 {
425         return 0;
426 #if 0
427         struct btrfs_path *path;
428         struct btrfs_key found_key;
429         struct extent_buffer *leaf;
430         struct btrfs_file_extent_item *extent;
431         u64 last_offset = 0;
432         int nritems;
433         int slot;
434         int found_type;
435         int ret;
436         int err = 0;
437         u64 extent_end = 0;
438
439         path = btrfs_alloc_path();
440         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
441                                        last_offset, 0);
442         while(1) {
443                 nritems = btrfs_header_nritems(path->nodes[0]);
444                 if (path->slots[0] >= nritems) {
445                         ret = btrfs_next_leaf(root, path);
446                         if (ret)
447                                 goto out;
448                         nritems = btrfs_header_nritems(path->nodes[0]);
449                 }
450                 slot = path->slots[0];
451                 leaf = path->nodes[0];
452                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
453                 if (found_key.objectid != inode->i_ino)
454                         break;
455                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
456                         goto out;
457
458                 if (found_key.offset < last_offset) {
459                         WARN_ON(1);
460                         btrfs_print_leaf(root, leaf);
461                         printk("inode %lu found offset %Lu expected %Lu\n",
462                                inode->i_ino, found_key.offset, last_offset);
463                         err = 1;
464                         goto out;
465                 }
466                 extent = btrfs_item_ptr(leaf, slot,
467                                         struct btrfs_file_extent_item);
468                 found_type = btrfs_file_extent_type(leaf, extent);
469                 if (found_type == BTRFS_FILE_EXTENT_REG) {
470                         extent_end = found_key.offset +
471                              btrfs_file_extent_num_bytes(leaf, extent);
472                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
473                         struct btrfs_item *item;
474                         item = btrfs_item_nr(leaf, slot);
475                         extent_end = found_key.offset +
476                              btrfs_file_extent_inline_len(leaf, item);
477                         extent_end = (extent_end + root->sectorsize - 1) &
478                                 ~((u64)root->sectorsize -1 );
479                 }
480                 last_offset = extent_end;
481                 path->slots[0]++;
482         }
483         if (0 && last_offset < inode->i_size) {
484                 WARN_ON(1);
485                 btrfs_print_leaf(root, leaf);
486                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
487                        last_offset, inode->i_size);
488                 err = 1;
489
490         }
491 out:
492         btrfs_free_path(path);
493         return err;
494 #endif
495 }
496
497 /*
498  * this is very complex, but the basic idea is to drop all extents
499  * in the range start - end.  hint_block is filled in with a block number
500  * that would be a good hint to the block allocator for this file.
501  *
502  * If an extent intersects the range but is not entirely inside the range
503  * it is either truncated or split.  Anything entirely inside the range
504  * is deleted from the tree.
505  */
506 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
507                        struct btrfs_root *root, struct inode *inode,
508                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
509 {
510         u64 extent_end = 0;
511         u64 search_start = start;
512         struct extent_buffer *leaf;
513         struct btrfs_file_extent_item *extent;
514         struct btrfs_path *path;
515         struct btrfs_key key;
516         struct btrfs_file_extent_item old;
517         int keep;
518         int slot;
519         int bookend;
520         int found_type;
521         int found_extent;
522         int found_inline;
523         int recow;
524         int ret;
525
526         btrfs_drop_extent_cache(inode, start, end - 1);
527
528         path = btrfs_alloc_path();
529         if (!path)
530                 return -ENOMEM;
531         while(1) {
532                 recow = 0;
533                 btrfs_release_path(root, path);
534                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
535                                                search_start, -1);
536                 if (ret < 0)
537                         goto out;
538                 if (ret > 0) {
539                         if (path->slots[0] == 0) {
540                                 ret = 0;
541                                 goto out;
542                         }
543                         path->slots[0]--;
544                 }
545 next_slot:
546                 keep = 0;
547                 bookend = 0;
548                 found_extent = 0;
549                 found_inline = 0;
550                 extent = NULL;
551                 leaf = path->nodes[0];
552                 slot = path->slots[0];
553                 ret = 0;
554                 btrfs_item_key_to_cpu(leaf, &key, slot);
555                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
556                     key.offset >= end) {
557                         goto out;
558                 }
559                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
560                     key.objectid != inode->i_ino) {
561                         goto out;
562                 }
563                 if (recow) {
564                         search_start = key.offset;
565                         continue;
566                 }
567                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
568                         extent = btrfs_item_ptr(leaf, slot,
569                                                 struct btrfs_file_extent_item);
570                         found_type = btrfs_file_extent_type(leaf, extent);
571                         if (found_type == BTRFS_FILE_EXTENT_REG) {
572                                 extent_end =
573                                      btrfs_file_extent_disk_bytenr(leaf,
574                                                                    extent);
575                                 if (extent_end)
576                                         *hint_byte = extent_end;
577
578                                 extent_end = key.offset +
579                                      btrfs_file_extent_num_bytes(leaf, extent);
580                                 found_extent = 1;
581                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
582                                 struct btrfs_item *item;
583                                 item = btrfs_item_nr(leaf, slot);
584                                 found_inline = 1;
585                                 extent_end = key.offset +
586                                      btrfs_file_extent_inline_len(leaf, item);
587                         }
588                 } else {
589                         extent_end = search_start;
590                 }
591
592                 /* we found nothing we can drop */
593                 if ((!found_extent && !found_inline) ||
594                     search_start >= extent_end) {
595                         int nextret;
596                         u32 nritems;
597                         nritems = btrfs_header_nritems(leaf);
598                         if (slot >= nritems - 1) {
599                                 nextret = btrfs_next_leaf(root, path);
600                                 if (nextret)
601                                         goto out;
602                                 recow = 1;
603                         } else {
604                                 path->slots[0]++;
605                         }
606                         goto next_slot;
607                 }
608
609                 if (found_inline) {
610                         u64 mask = root->sectorsize - 1;
611                         search_start = (extent_end + mask) & ~mask;
612                 } else
613                         search_start = extent_end;
614                 if (end <= extent_end && start >= key.offset && found_inline) {
615                         *hint_byte = EXTENT_MAP_INLINE;
616                         continue;
617                 }
618                 if (end < extent_end && end >= key.offset) {
619                         if (found_extent) {
620                                 u64 disk_bytenr =
621                                     btrfs_file_extent_disk_bytenr(leaf, extent);
622                                 u64 disk_num_bytes =
623                                     btrfs_file_extent_disk_num_bytes(leaf,
624                                                                       extent);
625                                 read_extent_buffer(leaf, &old,
626                                                    (unsigned long)extent,
627                                                    sizeof(old));
628                                 if (disk_bytenr != 0) {
629                                         ret = btrfs_inc_extent_ref(trans, root,
630                                                  disk_bytenr, disk_num_bytes,
631                                                  root->root_key.objectid,
632                                                  trans->transid,
633                                                  key.objectid, end);
634                                         BUG_ON(ret);
635                                 }
636                         }
637                         bookend = 1;
638                         if (found_inline && start <= key.offset)
639                                 keep = 1;
640                 }
641                 /* truncate existing extent */
642                 if (start > key.offset) {
643                         u64 new_num;
644                         u64 old_num;
645                         keep = 1;
646                         WARN_ON(start & (root->sectorsize - 1));
647                         if (found_extent) {
648                                 new_num = start - key.offset;
649                                 old_num = btrfs_file_extent_num_bytes(leaf,
650                                                                       extent);
651                                 *hint_byte =
652                                         btrfs_file_extent_disk_bytenr(leaf,
653                                                                       extent);
654                                 if (btrfs_file_extent_disk_bytenr(leaf,
655                                                                   extent)) {
656                                         dec_i_blocks(inode, old_num - new_num);
657                                 }
658                                 btrfs_set_file_extent_num_bytes(leaf, extent,
659                                                                 new_num);
660                                 btrfs_mark_buffer_dirty(leaf);
661                         } else if (key.offset < inline_limit &&
662                                    (end > extent_end) &&
663                                    (inline_limit < extent_end)) {
664                                 u32 new_size;
665                                 new_size = btrfs_file_extent_calc_inline_size(
666                                                    inline_limit - key.offset);
667                                 dec_i_blocks(inode, (extent_end - key.offset) -
668                                         (inline_limit - key.offset));
669                                 btrfs_truncate_item(trans, root, path,
670                                                     new_size, 1);
671                         }
672                 }
673                 /* delete the entire extent */
674                 if (!keep) {
675                         u64 disk_bytenr = 0;
676                         u64 disk_num_bytes = 0;
677                         u64 extent_num_bytes = 0;
678                         u64 root_gen;
679                         u64 root_owner;
680
681                         root_gen = btrfs_header_generation(leaf);
682                         root_owner = btrfs_header_owner(leaf);
683                         if (found_extent) {
684                                 disk_bytenr =
685                                       btrfs_file_extent_disk_bytenr(leaf,
686                                                                      extent);
687                                 disk_num_bytes =
688                                       btrfs_file_extent_disk_num_bytes(leaf,
689                                                                        extent);
690                                 extent_num_bytes =
691                                       btrfs_file_extent_num_bytes(leaf, extent);
692                                 *hint_byte =
693                                         btrfs_file_extent_disk_bytenr(leaf,
694                                                                       extent);
695                         }
696                         ret = btrfs_del_item(trans, root, path);
697                         /* TODO update progress marker and return */
698                         BUG_ON(ret);
699                         btrfs_release_path(root, path);
700                         extent = NULL;
701                         if (found_extent && disk_bytenr != 0) {
702                                 dec_i_blocks(inode, extent_num_bytes);
703                                 ret = btrfs_free_extent(trans, root,
704                                                 disk_bytenr,
705                                                 disk_num_bytes,
706                                                 root_owner,
707                                                 root_gen, inode->i_ino,
708                                                 key.offset, 0);
709                         }
710
711                         BUG_ON(ret);
712                         if (!bookend && search_start >= end) {
713                                 ret = 0;
714                                 goto out;
715                         }
716                         if (!bookend)
717                                 continue;
718                 }
719                 if (bookend && found_inline && start <= key.offset) {
720                         u32 new_size;
721                         new_size = btrfs_file_extent_calc_inline_size(
722                                                    extent_end - end);
723                         dec_i_blocks(inode, (extent_end - key.offset) -
724                                         (extent_end - end));
725                         btrfs_truncate_item(trans, root, path, new_size, 0);
726                 }
727                 /* create bookend, splitting the extent in two */
728                 if (bookend && found_extent) {
729                         struct btrfs_key ins;
730                         ins.objectid = inode->i_ino;
731                         ins.offset = end;
732                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
733                         btrfs_release_path(root, path);
734                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
735                                                       sizeof(*extent));
736
737                         leaf = path->nodes[0];
738                         if (ret) {
739                                 btrfs_print_leaf(root, leaf);
740                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
741                         }
742                         BUG_ON(ret);
743                         extent = btrfs_item_ptr(leaf, path->slots[0],
744                                                 struct btrfs_file_extent_item);
745                         write_extent_buffer(leaf, &old,
746                                             (unsigned long)extent, sizeof(old));
747
748                         btrfs_set_file_extent_offset(leaf, extent,
749                                     le64_to_cpu(old.offset) + end - key.offset);
750                         WARN_ON(le64_to_cpu(old.num_bytes) <
751                                 (extent_end - end));
752                         btrfs_set_file_extent_num_bytes(leaf, extent,
753                                                         extent_end - end);
754                         btrfs_set_file_extent_type(leaf, extent,
755                                                    BTRFS_FILE_EXTENT_REG);
756
757                         btrfs_mark_buffer_dirty(path->nodes[0]);
758                         if (le64_to_cpu(old.disk_bytenr) != 0) {
759                                 inode->i_blocks +=
760                                       btrfs_file_extent_num_bytes(leaf,
761                                                                   extent) >> 9;
762                         }
763                         ret = 0;
764                         goto out;
765                 }
766         }
767 out:
768         btrfs_free_path(path);
769         btrfs_check_file(root, inode);
770         return ret;
771 }
772
773 /*
774  * this gets pages into the page cache and locks them down
775  */
776 static int prepare_pages(struct btrfs_root *root, struct file *file,
777                          struct page **pages, size_t num_pages,
778                          loff_t pos, unsigned long first_index,
779                          unsigned long last_index, size_t write_bytes)
780 {
781         int i;
782         unsigned long index = pos >> PAGE_CACHE_SHIFT;
783         struct inode *inode = fdentry(file)->d_inode;
784         int err = 0;
785         u64 start_pos;
786         u64 last_pos;
787
788         start_pos = pos & ~((u64)root->sectorsize - 1);
789         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
790
791         memset(pages, 0, num_pages * sizeof(struct page *));
792 again:
793         for (i = 0; i < num_pages; i++) {
794                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
795                 if (!pages[i]) {
796                         err = -ENOMEM;
797                         BUG_ON(1);
798                 }
799                 wait_on_page_writeback(pages[i]);
800         }
801         if (start_pos < inode->i_size) {
802                 struct btrfs_ordered_extent *ordered;
803                 lock_extent(&BTRFS_I(inode)->io_tree,
804                             start_pos, last_pos - 1, GFP_NOFS);
805                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
806                 if (ordered &&
807                     ordered->file_offset + ordered->len > start_pos &&
808                     ordered->file_offset < last_pos) {
809                         btrfs_put_ordered_extent(ordered);
810                         unlock_extent(&BTRFS_I(inode)->io_tree,
811                                       start_pos, last_pos - 1, GFP_NOFS);
812                         for (i = 0; i < num_pages; i++) {
813                                 unlock_page(pages[i]);
814                                 page_cache_release(pages[i]);
815                         }
816                         btrfs_wait_ordered_range(inode, start_pos,
817                                                  last_pos - start_pos);
818                         goto again;
819                 }
820                 if (ordered)
821                         btrfs_put_ordered_extent(ordered);
822
823                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
824                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
825                                   GFP_NOFS);
826                 unlock_extent(&BTRFS_I(inode)->io_tree,
827                               start_pos, last_pos - 1, GFP_NOFS);
828         }
829         for (i = 0; i < num_pages; i++) {
830 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
831                 ClearPageDirty(pages[i]);
832 #else
833                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
834 #endif
835                 set_page_extent_mapped(pages[i]);
836                 WARN_ON(!PageLocked(pages[i]));
837         }
838         return 0;
839 }
840
841 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
842                                 size_t count, loff_t *ppos)
843 {
844         loff_t pos;
845         loff_t start_pos;
846         ssize_t num_written = 0;
847         ssize_t err = 0;
848         int ret = 0;
849         struct inode *inode = fdentry(file)->d_inode;
850         struct btrfs_root *root = BTRFS_I(inode)->root;
851         struct page **pages = NULL;
852         int nrptrs;
853         struct page *pinned[2];
854         unsigned long first_index;
855         unsigned long last_index;
856
857         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
858                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
859         pinned[0] = NULL;
860         pinned[1] = NULL;
861
862         pos = *ppos;
863         start_pos = pos;
864
865         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
866         current->backing_dev_info = inode->i_mapping->backing_dev_info;
867         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
868         if (err)
869                 goto out_nolock;
870         if (count == 0)
871                 goto out_nolock;
872 #ifdef REMOVE_SUID_PATH
873         err = remove_suid(&file->f_path);
874 #else
875         err = remove_suid(fdentry(file));
876 #endif
877         if (err)
878                 goto out_nolock;
879         file_update_time(file);
880
881         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
882
883         mutex_lock(&inode->i_mutex);
884         first_index = pos >> PAGE_CACHE_SHIFT;
885         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
886
887         /*
888          * if this is a nodatasum mount, force summing off for the inode
889          * all the time.  That way a later mount with summing on won't
890          * get confused
891          */
892         if (btrfs_test_opt(root, NODATASUM))
893                 btrfs_set_flag(inode, NODATASUM);
894
895         /*
896          * there are lots of better ways to do this, but this code
897          * makes sure the first and last page in the file range are
898          * up to date and ready for cow
899          */
900         if ((pos & (PAGE_CACHE_SIZE - 1))) {
901                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
902                 if (!PageUptodate(pinned[0])) {
903                         ret = btrfs_readpage(NULL, pinned[0]);
904                         BUG_ON(ret);
905                         wait_on_page_locked(pinned[0]);
906                 } else {
907                         unlock_page(pinned[0]);
908                 }
909         }
910         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
911                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
912                 if (!PageUptodate(pinned[1])) {
913                         ret = btrfs_readpage(NULL, pinned[1]);
914                         BUG_ON(ret);
915                         wait_on_page_locked(pinned[1]);
916                 } else {
917                         unlock_page(pinned[1]);
918                 }
919         }
920
921         while(count > 0) {
922                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
923                 size_t write_bytes = min(count, nrptrs *
924                                         (size_t)PAGE_CACHE_SIZE -
925                                          offset);
926                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
927                                         PAGE_CACHE_SHIFT;
928
929                 WARN_ON(num_pages > nrptrs);
930                 memset(pages, 0, sizeof(pages));
931
932                 ret = btrfs_check_free_space(root, write_bytes, 0);
933                 if (ret)
934                         goto out;
935
936                 ret = prepare_pages(root, file, pages, num_pages,
937                                     pos, first_index, last_index,
938                                     write_bytes);
939                 if (ret)
940                         goto out;
941
942                 ret = btrfs_copy_from_user(pos, num_pages,
943                                            write_bytes, pages, buf);
944                 if (ret) {
945                         btrfs_drop_pages(pages, num_pages);
946                         goto out;
947                 }
948
949                 ret = dirty_and_release_pages(NULL, root, file, pages,
950                                               num_pages, pos, write_bytes);
951                 btrfs_drop_pages(pages, num_pages);
952                 if (ret)
953                         goto out;
954
955                 buf += write_bytes;
956                 count -= write_bytes;
957                 pos += write_bytes;
958                 num_written += write_bytes;
959
960                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
961                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
962                         btrfs_btree_balance_dirty(root, 1);
963                 cond_resched();
964         }
965 out:
966         mutex_unlock(&inode->i_mutex);
967
968 out_nolock:
969         kfree(pages);
970         if (pinned[0])
971                 page_cache_release(pinned[0]);
972         if (pinned[1])
973                 page_cache_release(pinned[1]);
974         *ppos = pos;
975
976         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
977                 err = sync_page_range(inode, inode->i_mapping,
978                                       start_pos, num_written);
979                 if (err < 0)
980                         num_written = err;
981         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
982 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
983                 do_sync_file_range(file, start_pos,
984                                       start_pos + num_written - 1,
985                                       SYNC_FILE_RANGE_WRITE |
986                                       SYNC_FILE_RANGE_WAIT_AFTER);
987 #else
988                 do_sync_mapping_range(inode->i_mapping, start_pos,
989                                       start_pos + num_written - 1,
990                                       SYNC_FILE_RANGE_WRITE |
991                                       SYNC_FILE_RANGE_WAIT_AFTER);
992 #endif
993                 invalidate_mapping_pages(inode->i_mapping,
994                       start_pos >> PAGE_CACHE_SHIFT,
995                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
996         }
997         current->backing_dev_info = NULL;
998         return num_written ? num_written : err;
999 }
1000
1001 int btrfs_release_file(struct inode * inode, struct file * filp)
1002 {
1003         if (filp->private_data)
1004                 btrfs_ioctl_trans_end(filp);
1005         return 0;
1006 }
1007
1008 static int btrfs_sync_file(struct file *file,
1009                            struct dentry *dentry, int datasync)
1010 {
1011         struct inode *inode = dentry->d_inode;
1012         struct btrfs_root *root = BTRFS_I(inode)->root;
1013         int ret = 0;
1014         struct btrfs_trans_handle *trans;
1015
1016         /*
1017          * check the transaction that last modified this inode
1018          * and see if its already been committed
1019          */
1020         if (!BTRFS_I(inode)->last_trans)
1021                 goto out;
1022
1023         mutex_lock(&root->fs_info->trans_mutex);
1024         if (BTRFS_I(inode)->last_trans <=
1025             root->fs_info->last_trans_committed) {
1026                 BTRFS_I(inode)->last_trans = 0;
1027                 mutex_unlock(&root->fs_info->trans_mutex);
1028                 goto out;
1029         }
1030         mutex_unlock(&root->fs_info->trans_mutex);
1031
1032         /*
1033          * ok we haven't committed the transaction yet, lets do a commit
1034          */
1035         if (file->private_data)
1036                 btrfs_ioctl_trans_end(file);
1037
1038         trans = btrfs_start_transaction(root, 1);
1039         if (!trans) {
1040                 ret = -ENOMEM;
1041                 goto out;
1042         }
1043         ret = btrfs_commit_transaction(trans, root);
1044 out:
1045         return ret > 0 ? EIO : ret;
1046 }
1047
1048 static struct vm_operations_struct btrfs_file_vm_ops = {
1049 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1050         .nopage         = filemap_nopage,
1051         .populate       = filemap_populate,
1052 #else
1053         .fault          = filemap_fault,
1054 #endif
1055         .page_mkwrite   = btrfs_page_mkwrite,
1056 };
1057
1058 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1059 {
1060         vma->vm_ops = &btrfs_file_vm_ops;
1061         file_accessed(filp);
1062         return 0;
1063 }
1064
1065 struct file_operations btrfs_file_operations = {
1066         .llseek         = generic_file_llseek,
1067         .read           = do_sync_read,
1068         .aio_read       = generic_file_aio_read,
1069         .splice_read    = generic_file_splice_read,
1070 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1071         .sendfile       = generic_file_sendfile,
1072 #endif
1073         .write          = btrfs_file_write,
1074         .mmap           = btrfs_file_mmap,
1075         .open           = generic_file_open,
1076         .release        = btrfs_release_file,
1077         .fsync          = btrfs_sync_file,
1078         .unlocked_ioctl = btrfs_ioctl,
1079 #ifdef CONFIG_COMPAT
1080         .compat_ioctl   = btrfs_ioctl,
1081 #endif
1082 };
1083