Fix btrfs_del_ordered_inode to allow forcing the drop during unlinks
[linux-3.10.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 "ordered-data.h"
38 #include "ioctl.h"
39 #include "print-tree.h"
40 #include "compat.h"
41
42
43 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
44                                 struct page **prepared_pages,
45                                 const char __user * buf)
46 {
47         long page_fault = 0;
48         int i;
49         int offset = pos & (PAGE_CACHE_SIZE - 1);
50
51         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
52                 size_t count = min_t(size_t,
53                                      PAGE_CACHE_SIZE - offset, write_bytes);
54                 struct page *page = prepared_pages[i];
55                 fault_in_pages_readable(buf, count);
56
57                 /* Copy data from userspace to the current page */
58                 kmap(page);
59                 page_fault = __copy_from_user(page_address(page) + offset,
60                                               buf, count);
61                 /* Flush processor's dcache for this page */
62                 flush_dcache_page(page);
63                 kunmap(page);
64                 buf += count;
65                 write_bytes -= count;
66
67                 if (page_fault)
68                         break;
69         }
70         return page_fault ? -EFAULT : 0;
71 }
72
73 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
74 {
75         size_t i;
76         for (i = 0; i < num_pages; i++) {
77                 if (!pages[i])
78                         break;
79                 unlock_page(pages[i]);
80                 mark_page_accessed(pages[i]);
81                 page_cache_release(pages[i]);
82         }
83 }
84
85 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
86                                 struct btrfs_root *root, struct inode *inode,
87                                 u64 offset, size_t size,
88                                 struct page **pages, size_t page_offset,
89                                 int num_pages)
90 {
91         struct btrfs_key key;
92         struct btrfs_path *path;
93         struct extent_buffer *leaf;
94         char *kaddr;
95         unsigned long ptr;
96         struct btrfs_file_extent_item *ei;
97         struct page *page;
98         u32 datasize;
99         int err = 0;
100         int ret;
101         int i;
102         ssize_t cur_size;
103
104         path = btrfs_alloc_path();
105         if (!path)
106                 return -ENOMEM;
107
108         btrfs_set_trans_block_group(trans, inode);
109
110         key.objectid = inode->i_ino;
111         key.offset = offset;
112         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
113
114         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
115         if (ret < 0) {
116                 err = ret;
117                 goto fail;
118         }
119         if (ret == 1) {
120                 struct btrfs_key found_key;
121
122                 if (path->slots[0] == 0)
123                         goto insert;
124
125                 path->slots[0]--;
126                 leaf = path->nodes[0];
127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
128
129                 if (found_key.objectid != inode->i_ino)
130                         goto insert;
131
132                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
133                         goto insert;
134                 ei = btrfs_item_ptr(leaf, path->slots[0],
135                                     struct btrfs_file_extent_item);
136
137                 if (btrfs_file_extent_type(leaf, ei) !=
138                     BTRFS_FILE_EXTENT_INLINE) {
139                         goto insert;
140                 }
141                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
142                 ret = 0;
143         }
144         if (ret == 0) {
145                 u32 found_size;
146                 u64 found_end;
147
148                 leaf = path->nodes[0];
149                 ei = btrfs_item_ptr(leaf, path->slots[0],
150                                     struct btrfs_file_extent_item);
151
152                 if (btrfs_file_extent_type(leaf, ei) !=
153                     BTRFS_FILE_EXTENT_INLINE) {
154                         err = ret;
155                         btrfs_print_leaf(root, leaf);
156                         printk("found wasn't inline offset %Lu inode %lu\n",
157                                offset, inode->i_ino);
158                         goto fail;
159                 }
160                 found_size = btrfs_file_extent_inline_len(leaf,
161                                           btrfs_item_nr(leaf, path->slots[0]));
162                 found_end = key.offset + found_size;
163
164                 if (found_end < offset + size) {
165                         btrfs_release_path(root, path);
166                         ret = btrfs_search_slot(trans, root, &key, path,
167                                                 offset + size - found_end, 1);
168                         BUG_ON(ret != 0);
169
170                         ret = btrfs_extend_item(trans, root, path,
171                                                 offset + size - found_end);
172                         if (ret) {
173                                 err = ret;
174                                 goto fail;
175                         }
176                         leaf = path->nodes[0];
177                         ei = btrfs_item_ptr(leaf, path->slots[0],
178                                             struct btrfs_file_extent_item);
179                         inode->i_blocks += (offset + size - found_end) >> 9;
180                 }
181                 if (found_end < offset) {
182                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
183                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
184                 }
185         } else {
186 insert:
187                 btrfs_release_path(root, path);
188                 datasize = offset + size - key.offset;
189                 inode->i_blocks += datasize >> 9;
190                 datasize = btrfs_file_extent_calc_inline_size(datasize);
191                 ret = btrfs_insert_empty_item(trans, root, path, &key,
192                                               datasize);
193                 if (ret) {
194                         err = ret;
195                         printk("got bad ret %d\n", ret);
196                         goto fail;
197                 }
198                 leaf = path->nodes[0];
199                 ei = btrfs_item_ptr(leaf, path->slots[0],
200                                     struct btrfs_file_extent_item);
201                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
202                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
203         }
204         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
205
206         cur_size = size;
207         i = 0;
208         while (size > 0) {
209                 page = pages[i];
210                 kaddr = kmap_atomic(page, KM_USER0);
211                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
212                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
213                 kunmap_atomic(kaddr, KM_USER0);
214                 page_offset = 0;
215                 ptr += cur_size;
216                 size -= cur_size;
217                 if (i >= num_pages) {
218                         printk("i %d num_pages %d\n", i, num_pages);
219                 }
220                 i++;
221         }
222         btrfs_mark_buffer_dirty(leaf);
223 fail:
224         btrfs_free_path(path);
225         return err;
226 }
227
228 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
229                                    struct btrfs_root *root,
230                                    struct file *file,
231                                    struct page **pages,
232                                    size_t num_pages,
233                                    loff_t pos,
234                                    size_t write_bytes)
235 {
236         int err = 0;
237         int i;
238         struct inode *inode = fdentry(file)->d_inode;
239         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
240         u64 hint_byte;
241         u64 num_bytes;
242         u64 start_pos;
243         u64 end_of_last_block;
244         u64 end_pos = pos + write_bytes;
245         u64 inline_size;
246         loff_t isize = i_size_read(inode);
247
248         start_pos = pos & ~((u64)root->sectorsize - 1);
249         num_bytes = (write_bytes + pos - start_pos +
250                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
251
252         end_of_last_block = start_pos + num_bytes - 1;
253
254         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
255         trans = btrfs_start_transaction(root, 1);
256         if (!trans) {
257                 err = -ENOMEM;
258                 goto out_unlock;
259         }
260         btrfs_set_trans_block_group(trans, inode);
261         hint_byte = 0;
262
263         if ((end_of_last_block & 4095) == 0) {
264                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
265         }
266         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
267
268         /* FIXME...EIEIO, ENOSPC and more */
269         /* insert any holes we need to create */
270         if (isize < end_pos) {
271                 u64 last_pos_in_file;
272                 u64 hole_size;
273                 u64 mask = root->sectorsize - 1;
274                 last_pos_in_file = (isize + mask) & ~mask;
275                 hole_size = (end_pos - last_pos_in_file + mask) & ~mask;
276                 if (last_pos_in_file < end_pos) {
277                         err = btrfs_drop_extents(trans, root, inode,
278                                                  last_pos_in_file,
279                                                  last_pos_in_file + hole_size,
280                                                  last_pos_in_file,
281                                                  &hint_byte);
282                         if (err)
283                                 goto failed;
284
285                         err = btrfs_insert_file_extent(trans, root,
286                                                        inode->i_ino,
287                                                        last_pos_in_file,
288                                                        0, 0, hole_size, 0);
289                         btrfs_drop_extent_cache(inode, last_pos_in_file,
290                                         last_pos_in_file + hole_size -1);
291                         btrfs_check_file(root, inode);
292                 }
293                 if (err)
294                         goto failed;
295         }
296
297         /*
298          * either allocate an extent for the new bytes or setup the key
299          * to show we are doing inline data in the extent
300          */
301         inline_size = end_pos;
302         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
303             inline_size > root->fs_info->max_inline ||
304             (inline_size & (root->sectorsize -1)) == 0 ||
305             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
306                 u64 last_end;
307
308                 for (i = 0; i < num_pages; i++) {
309                         struct page *p = pages[i];
310                         SetPageUptodate(p);
311                         set_page_dirty(p);
312                 }
313                 last_end = (u64)(pages[num_pages -1]->index) <<
314                                 PAGE_CACHE_SHIFT;
315                 last_end += PAGE_CACHE_SIZE - 1;
316                 set_extent_delalloc(io_tree, start_pos, end_of_last_block,
317                                  GFP_NOFS);
318                 btrfs_add_ordered_inode(inode);
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(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         u64 len = end - start + 1;
354         int ret;
355         int testend = 1;
356
357         if (end == (u64)-1) {
358                 len = (u64)-1;
359                 testend = 0;
360         }
361         while(1) {
362                 if (!split)
363                         split = alloc_extent_map(GFP_NOFS);
364                 if (!split2)
365                         split2 = alloc_extent_map(GFP_NOFS);
366
367                 spin_lock(&em_tree->lock);
368                 em = lookup_extent_mapping(em_tree, start, len);
369                 if (!em) {
370                         spin_unlock(&em_tree->lock);
371                         break;
372                 }
373                 remove_extent_mapping(em_tree, em);
374
375                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
376                     em->start < start) {
377                         split->start = em->start;
378                         split->len = start - em->start;
379                         split->block_start = em->block_start;
380                         split->bdev = em->bdev;
381                         split->flags = em->flags;
382                         ret = add_extent_mapping(em_tree, split);
383                         BUG_ON(ret);
384                         free_extent_map(split);
385                         split = split2;
386                         split2 = NULL;
387                 }
388                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
389                     testend && em->start + em->len > start + len) {
390                         u64 diff = start + len - em->start;
391
392                         split->start = start + len;
393                         split->len = em->start + em->len - (start + len);
394                         split->bdev = em->bdev;
395                         split->flags = em->flags;
396
397                         split->block_start = em->block_start + diff;
398
399                         ret = add_extent_mapping(em_tree, split);
400                         BUG_ON(ret);
401                         free_extent_map(split);
402                         split = NULL;
403                 }
404                 spin_unlock(&em_tree->lock);
405
406                 /* once for us */
407                 free_extent_map(em);
408                 /* once for the tree*/
409                 free_extent_map(em);
410         }
411         if (split)
412                 free_extent_map(split);
413         if (split2)
414                 free_extent_map(split2);
415         return 0;
416 }
417
418 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
419 {
420         return 0;
421 #if 0
422         struct btrfs_path *path;
423         struct btrfs_key found_key;
424         struct extent_buffer *leaf;
425         struct btrfs_file_extent_item *extent;
426         u64 last_offset = 0;
427         int nritems;
428         int slot;
429         int found_type;
430         int ret;
431         int err = 0;
432         u64 extent_end = 0;
433
434         path = btrfs_alloc_path();
435         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
436                                        last_offset, 0);
437         while(1) {
438                 nritems = btrfs_header_nritems(path->nodes[0]);
439                 if (path->slots[0] >= nritems) {
440                         ret = btrfs_next_leaf(root, path);
441                         if (ret)
442                                 goto out;
443                         nritems = btrfs_header_nritems(path->nodes[0]);
444                 }
445                 slot = path->slots[0];
446                 leaf = path->nodes[0];
447                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
448                 if (found_key.objectid != inode->i_ino)
449                         break;
450                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
451                         goto out;
452
453                 if (found_key.offset < last_offset) {
454                         WARN_ON(1);
455                         btrfs_print_leaf(root, leaf);
456                         printk("inode %lu found offset %Lu expected %Lu\n",
457                                inode->i_ino, found_key.offset, last_offset);
458                         err = 1;
459                         goto out;
460                 }
461                 extent = btrfs_item_ptr(leaf, slot,
462                                         struct btrfs_file_extent_item);
463                 found_type = btrfs_file_extent_type(leaf, extent);
464                 if (found_type == BTRFS_FILE_EXTENT_REG) {
465                         extent_end = found_key.offset +
466                              btrfs_file_extent_num_bytes(leaf, extent);
467                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
468                         struct btrfs_item *item;
469                         item = btrfs_item_nr(leaf, slot);
470                         extent_end = found_key.offset +
471                              btrfs_file_extent_inline_len(leaf, item);
472                         extent_end = (extent_end + root->sectorsize - 1) &
473                                 ~((u64)root->sectorsize -1 );
474                 }
475                 last_offset = extent_end;
476                 path->slots[0]++;
477         }
478         if (0 && last_offset < inode->i_size) {
479                 WARN_ON(1);
480                 btrfs_print_leaf(root, leaf);
481                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
482                        last_offset, inode->i_size);
483                 err = 1;
484
485         }
486 out:
487         btrfs_free_path(path);
488         return err;
489 #endif
490 }
491
492 /*
493  * this is very complex, but the basic idea is to drop all extents
494  * in the range start - end.  hint_block is filled in with a block number
495  * that would be a good hint to the block allocator for this file.
496  *
497  * If an extent intersects the range but is not entirely inside the range
498  * it is either truncated or split.  Anything entirely inside the range
499  * is deleted from the tree.
500  */
501 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
502                        struct btrfs_root *root, struct inode *inode,
503                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
504 {
505         u64 extent_end = 0;
506         u64 search_start = start;
507         struct extent_buffer *leaf;
508         struct btrfs_file_extent_item *extent;
509         struct btrfs_path *path;
510         struct btrfs_key key;
511         struct btrfs_file_extent_item old;
512         int keep;
513         int slot;
514         int bookend;
515         int found_type;
516         int found_extent;
517         int found_inline;
518         int recow;
519         int ret;
520
521         btrfs_drop_extent_cache(inode, start, end - 1);
522
523         path = btrfs_alloc_path();
524         if (!path)
525                 return -ENOMEM;
526         while(1) {
527                 recow = 0;
528                 btrfs_release_path(root, path);
529                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
530                                                search_start, -1);
531                 if (ret < 0)
532                         goto out;
533                 if (ret > 0) {
534                         if (path->slots[0] == 0) {
535                                 ret = 0;
536                                 goto out;
537                         }
538                         path->slots[0]--;
539                 }
540 next_slot:
541                 keep = 0;
542                 bookend = 0;
543                 found_extent = 0;
544                 found_inline = 0;
545                 extent = NULL;
546                 leaf = path->nodes[0];
547                 slot = path->slots[0];
548                 ret = 0;
549                 btrfs_item_key_to_cpu(leaf, &key, slot);
550                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
551                     key.offset >= end) {
552                         goto out;
553                 }
554                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
555                     key.objectid != inode->i_ino) {
556                         goto out;
557                 }
558                 if (recow) {
559                         search_start = key.offset;
560                         continue;
561                 }
562                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
563                         extent = btrfs_item_ptr(leaf, slot,
564                                                 struct btrfs_file_extent_item);
565                         found_type = btrfs_file_extent_type(leaf, extent);
566                         if (found_type == BTRFS_FILE_EXTENT_REG) {
567                                 extent_end =
568                                      btrfs_file_extent_disk_bytenr(leaf,
569                                                                    extent);
570                                 if (extent_end)
571                                         *hint_byte = extent_end;
572
573                                 extent_end = key.offset +
574                                      btrfs_file_extent_num_bytes(leaf, extent);
575                                 found_extent = 1;
576                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
577                                 struct btrfs_item *item;
578                                 item = btrfs_item_nr(leaf, slot);
579                                 found_inline = 1;
580                                 extent_end = key.offset +
581                                      btrfs_file_extent_inline_len(leaf, item);
582                         }
583                 } else {
584                         extent_end = search_start;
585                 }
586
587                 /* we found nothing we can drop */
588                 if ((!found_extent && !found_inline) ||
589                     search_start >= extent_end) {
590                         int nextret;
591                         u32 nritems;
592                         nritems = btrfs_header_nritems(leaf);
593                         if (slot >= nritems - 1) {
594                                 nextret = btrfs_next_leaf(root, path);
595                                 if (nextret)
596                                         goto out;
597                                 recow = 1;
598                         } else {
599                                 path->slots[0]++;
600                         }
601                         goto next_slot;
602                 }
603
604                 if (found_inline) {
605                         u64 mask = root->sectorsize - 1;
606                         search_start = (extent_end + mask) & ~mask;
607                 } else
608                         search_start = extent_end;
609                 if (end <= extent_end && start >= key.offset && found_inline) {
610                         *hint_byte = EXTENT_MAP_INLINE;
611                         continue;
612                 }
613                 if (end < extent_end && end >= key.offset) {
614                         if (found_extent) {
615                                 u64 disk_bytenr =
616                                     btrfs_file_extent_disk_bytenr(leaf, extent);
617                                 u64 disk_num_bytes =
618                                     btrfs_file_extent_disk_num_bytes(leaf,
619                                                                       extent);
620                                 read_extent_buffer(leaf, &old,
621                                                    (unsigned long)extent,
622                                                    sizeof(old));
623                                 if (disk_bytenr != 0) {
624                                         ret = btrfs_inc_extent_ref(trans, root,
625                                                  disk_bytenr, disk_num_bytes,
626                                                  root->root_key.objectid,
627                                                  trans->transid,
628                                                  key.objectid, end);
629                                         BUG_ON(ret);
630                                 }
631                         }
632                         bookend = 1;
633                         if (found_inline && start <= key.offset)
634                                 keep = 1;
635                 }
636                 /* truncate existing extent */
637                 if (start > key.offset) {
638                         u64 new_num;
639                         u64 old_num;
640                         keep = 1;
641                         WARN_ON(start & (root->sectorsize - 1));
642                         if (found_extent) {
643                                 new_num = start - key.offset;
644                                 old_num = btrfs_file_extent_num_bytes(leaf,
645                                                                       extent);
646                                 *hint_byte =
647                                         btrfs_file_extent_disk_bytenr(leaf,
648                                                                       extent);
649                                 if (btrfs_file_extent_disk_bytenr(leaf,
650                                                                   extent)) {
651                                         dec_i_blocks(inode, old_num - new_num);
652                                 }
653                                 btrfs_set_file_extent_num_bytes(leaf, extent,
654                                                                 new_num);
655                                 btrfs_mark_buffer_dirty(leaf);
656                         } else if (key.offset < inline_limit &&
657                                    (end > extent_end) &&
658                                    (inline_limit < extent_end)) {
659                                 u32 new_size;
660                                 new_size = btrfs_file_extent_calc_inline_size(
661                                                    inline_limit - key.offset);
662                                 dec_i_blocks(inode, (extent_end - key.offset) -
663                                         (inline_limit - key.offset));
664                                 btrfs_truncate_item(trans, root, path,
665                                                     new_size, 1);
666                         }
667                 }
668                 /* delete the entire extent */
669                 if (!keep) {
670                         u64 disk_bytenr = 0;
671                         u64 disk_num_bytes = 0;
672                         u64 extent_num_bytes = 0;
673                         u64 root_gen;
674                         u64 root_owner;
675
676                         root_gen = btrfs_header_generation(leaf);
677                         root_owner = btrfs_header_owner(leaf);
678                         if (found_extent) {
679                                 disk_bytenr =
680                                       btrfs_file_extent_disk_bytenr(leaf,
681                                                                      extent);
682                                 disk_num_bytes =
683                                       btrfs_file_extent_disk_num_bytes(leaf,
684                                                                        extent);
685                                 extent_num_bytes =
686                                       btrfs_file_extent_num_bytes(leaf, extent);
687                                 *hint_byte =
688                                         btrfs_file_extent_disk_bytenr(leaf,
689                                                                       extent);
690                         }
691                         ret = btrfs_del_item(trans, root, path);
692                         /* TODO update progress marker and return */
693                         BUG_ON(ret);
694                         btrfs_release_path(root, path);
695                         extent = NULL;
696                         if (found_extent && disk_bytenr != 0) {
697                                 dec_i_blocks(inode, extent_num_bytes);
698                                 ret = btrfs_free_extent(trans, root,
699                                                 disk_bytenr,
700                                                 disk_num_bytes,
701                                                 root_owner,
702                                                 root_gen, inode->i_ino,
703                                                 key.offset, 0);
704                         }
705
706                         BUG_ON(ret);
707                         if (!bookend && search_start >= end) {
708                                 ret = 0;
709                                 goto out;
710                         }
711                         if (!bookend)
712                                 continue;
713                 }
714                 if (bookend && found_inline && start <= key.offset) {
715                         u32 new_size;
716                         new_size = btrfs_file_extent_calc_inline_size(
717                                                    extent_end - end);
718                         dec_i_blocks(inode, (extent_end - key.offset) -
719                                         (extent_end - end));
720                         btrfs_truncate_item(trans, root, path, new_size, 0);
721                 }
722                 /* create bookend, splitting the extent in two */
723                 if (bookend && found_extent) {
724                         struct btrfs_key ins;
725                         ins.objectid = inode->i_ino;
726                         ins.offset = end;
727                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
728                         btrfs_release_path(root, path);
729                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
730                                                       sizeof(*extent));
731
732                         leaf = path->nodes[0];
733                         if (ret) {
734                                 btrfs_print_leaf(root, leaf);
735                                 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);
736                         }
737                         BUG_ON(ret);
738                         extent = btrfs_item_ptr(leaf, path->slots[0],
739                                                 struct btrfs_file_extent_item);
740                         write_extent_buffer(leaf, &old,
741                                             (unsigned long)extent, sizeof(old));
742
743                         btrfs_set_file_extent_offset(leaf, extent,
744                                     le64_to_cpu(old.offset) + end - key.offset);
745                         WARN_ON(le64_to_cpu(old.num_bytes) <
746                                 (extent_end - end));
747                         btrfs_set_file_extent_num_bytes(leaf, extent,
748                                                         extent_end - end);
749                         btrfs_set_file_extent_type(leaf, extent,
750                                                    BTRFS_FILE_EXTENT_REG);
751
752                         btrfs_mark_buffer_dirty(path->nodes[0]);
753                         if (le64_to_cpu(old.disk_bytenr) != 0) {
754                                 inode->i_blocks +=
755                                       btrfs_file_extent_num_bytes(leaf,
756                                                                   extent) >> 9;
757                         }
758                         ret = 0;
759                         goto out;
760                 }
761         }
762 out:
763         btrfs_free_path(path);
764         btrfs_check_file(root, inode);
765         return ret;
766 }
767
768 /*
769  * this gets pages into the page cache and locks them down
770  */
771 static int prepare_pages(struct btrfs_root *root, struct file *file,
772                          struct page **pages, size_t num_pages,
773                          loff_t pos, unsigned long first_index,
774                          unsigned long last_index, size_t write_bytes)
775 {
776         int i;
777         unsigned long index = pos >> PAGE_CACHE_SHIFT;
778         struct inode *inode = fdentry(file)->d_inode;
779         int err = 0;
780         u64 start_pos;
781
782         start_pos = pos & ~((u64)root->sectorsize - 1);
783
784         memset(pages, 0, num_pages * sizeof(struct page *));
785
786         for (i = 0; i < num_pages; i++) {
787                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
788                 if (!pages[i]) {
789                         err = -ENOMEM;
790                         BUG_ON(1);
791                 }
792 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
793                 ClearPageDirty(pages[i]);
794 #else
795                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
796 #endif
797                 wait_on_page_writeback(pages[i]);
798                 set_page_extent_mapped(pages[i]);
799                 WARN_ON(!PageLocked(pages[i]));
800         }
801         if (start_pos < inode->i_size) {
802                 u64 last_pos;
803                 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
804                 lock_extent(&BTRFS_I(inode)->io_tree,
805                             start_pos, last_pos - 1, GFP_NOFS);
806                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
807                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
808                                   GFP_NOFS);
809                 unlock_extent(&BTRFS_I(inode)->io_tree,
810                               start_pos, last_pos - 1, GFP_NOFS);
811         }
812         return 0;
813 }
814
815 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
816                                 size_t count, loff_t *ppos)
817 {
818         loff_t pos;
819         loff_t start_pos;
820         ssize_t num_written = 0;
821         ssize_t err = 0;
822         int ret = 0;
823         struct inode *inode = fdentry(file)->d_inode;
824         struct btrfs_root *root = BTRFS_I(inode)->root;
825         struct page **pages = NULL;
826         int nrptrs;
827         struct page *pinned[2];
828         unsigned long first_index;
829         unsigned long last_index;
830
831         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
832                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
833         pinned[0] = NULL;
834         pinned[1] = NULL;
835
836         pos = *ppos;
837         start_pos = pos;
838
839         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
840         current->backing_dev_info = inode->i_mapping->backing_dev_info;
841         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
842         if (err)
843                 goto out_nolock;
844         if (count == 0)
845                 goto out_nolock;
846 #ifdef REMOVE_SUID_PATH
847         err = remove_suid(&file->f_path);
848 #else
849         err = remove_suid(fdentry(file));
850 #endif
851         if (err)
852                 goto out_nolock;
853         file_update_time(file);
854
855         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
856
857         mutex_lock(&inode->i_mutex);
858         first_index = pos >> PAGE_CACHE_SHIFT;
859         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
860
861         /*
862          * if this is a nodatasum mount, force summing off for the inode
863          * all the time.  That way a later mount with summing on won't
864          * get confused
865          */
866         if (btrfs_test_opt(root, NODATASUM))
867                 btrfs_set_flag(inode, NODATASUM);
868
869         /*
870          * there are lots of better ways to do this, but this code
871          * makes sure the first and last page in the file range are
872          * up to date and ready for cow
873          */
874         if ((pos & (PAGE_CACHE_SIZE - 1))) {
875                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
876                 if (!PageUptodate(pinned[0])) {
877                         ret = btrfs_readpage(NULL, pinned[0]);
878                         BUG_ON(ret);
879                         wait_on_page_locked(pinned[0]);
880                 } else {
881                         unlock_page(pinned[0]);
882                 }
883         }
884         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
885                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
886                 if (!PageUptodate(pinned[1])) {
887                         ret = btrfs_readpage(NULL, pinned[1]);
888                         BUG_ON(ret);
889                         wait_on_page_locked(pinned[1]);
890                 } else {
891                         unlock_page(pinned[1]);
892                 }
893         }
894
895         while(count > 0) {
896                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
897                 size_t write_bytes = min(count, nrptrs *
898                                         (size_t)PAGE_CACHE_SIZE -
899                                          offset);
900                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
901                                         PAGE_CACHE_SHIFT;
902
903                 WARN_ON(num_pages > nrptrs);
904                 memset(pages, 0, sizeof(pages));
905
906                 ret = btrfs_check_free_space(root, write_bytes, 0);
907                 if (ret)
908                         goto out;
909
910                 ret = prepare_pages(root, file, pages, num_pages,
911                                     pos, first_index, last_index,
912                                     write_bytes);
913                 if (ret)
914                         goto out;
915
916                 ret = btrfs_copy_from_user(pos, num_pages,
917                                            write_bytes, pages, buf);
918                 if (ret) {
919                         btrfs_drop_pages(pages, num_pages);
920                         goto out;
921                 }
922
923                 ret = dirty_and_release_pages(NULL, root, file, pages,
924                                               num_pages, pos, write_bytes);
925                 btrfs_drop_pages(pages, num_pages);
926                 if (ret)
927                         goto out;
928
929                 buf += write_bytes;
930                 count -= write_bytes;
931                 pos += write_bytes;
932                 num_written += write_bytes;
933
934                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
935                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
936                         btrfs_btree_balance_dirty(root, 1);
937                 btrfs_throttle(root);
938                 cond_resched();
939         }
940 out:
941         mutex_unlock(&inode->i_mutex);
942
943 out_nolock:
944         kfree(pages);
945         if (pinned[0])
946                 page_cache_release(pinned[0]);
947         if (pinned[1])
948                 page_cache_release(pinned[1]);
949         *ppos = pos;
950
951         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
952                 err = sync_page_range(inode, inode->i_mapping,
953                                       start_pos, num_written);
954                 if (err < 0)
955                         num_written = err;
956         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
957 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
958                 do_sync_file_range(file, start_pos,
959                                       start_pos + num_written - 1,
960                                       SYNC_FILE_RANGE_WRITE |
961                                       SYNC_FILE_RANGE_WAIT_AFTER);
962 #else
963                 do_sync_mapping_range(inode->i_mapping, start_pos,
964                                       start_pos + num_written - 1,
965                                       SYNC_FILE_RANGE_WRITE |
966                                       SYNC_FILE_RANGE_WAIT_AFTER);
967 #endif
968                 invalidate_mapping_pages(inode->i_mapping,
969                       start_pos >> PAGE_CACHE_SHIFT,
970                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
971         }
972         current->backing_dev_info = NULL;
973         btrfs_ordered_throttle(root, inode);
974         return num_written ? num_written : err;
975 }
976
977 int btrfs_release_file(struct inode * inode, struct file * filp)
978 {
979         btrfs_del_ordered_inode(inode, 0);
980         if (filp->private_data)
981                 btrfs_ioctl_trans_end(filp);
982         return 0;
983 }
984
985 static int btrfs_sync_file(struct file *file,
986                            struct dentry *dentry, int datasync)
987 {
988         struct inode *inode = dentry->d_inode;
989         struct btrfs_root *root = BTRFS_I(inode)->root;
990         int ret = 0;
991         struct btrfs_trans_handle *trans;
992
993         /*
994          * check the transaction that last modified this inode
995          * and see if its already been committed
996          */
997         if (!BTRFS_I(inode)->last_trans)
998                 goto out;
999
1000         mutex_lock(&root->fs_info->trans_mutex);
1001         if (BTRFS_I(inode)->last_trans <=
1002             root->fs_info->last_trans_committed) {
1003                 BTRFS_I(inode)->last_trans = 0;
1004                 mutex_unlock(&root->fs_info->trans_mutex);
1005                 goto out;
1006         }
1007         mutex_unlock(&root->fs_info->trans_mutex);
1008
1009         /*
1010          * ok we haven't committed the transaction yet, lets do a commit
1011          */
1012         if (file->private_data)
1013                 btrfs_ioctl_trans_end(file);
1014
1015         trans = btrfs_start_transaction(root, 1);
1016         if (!trans) {
1017                 ret = -ENOMEM;
1018                 goto out;
1019         }
1020         ret = btrfs_commit_transaction(trans, root);
1021 out:
1022         return ret > 0 ? EIO : ret;
1023 }
1024
1025 static struct vm_operations_struct btrfs_file_vm_ops = {
1026 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1027         .nopage         = filemap_nopage,
1028         .populate       = filemap_populate,
1029 #else
1030         .fault          = filemap_fault,
1031 #endif
1032         .page_mkwrite   = btrfs_page_mkwrite,
1033 };
1034
1035 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1036 {
1037         vma->vm_ops = &btrfs_file_vm_ops;
1038         file_accessed(filp);
1039         return 0;
1040 }
1041
1042 struct file_operations btrfs_file_operations = {
1043         .llseek         = generic_file_llseek,
1044         .read           = do_sync_read,
1045         .aio_read       = generic_file_aio_read,
1046         .splice_read    = generic_file_splice_read,
1047 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1048         .sendfile       = generic_file_sendfile,
1049 #endif
1050         .write          = btrfs_file_write,
1051         .mmap           = btrfs_file_mmap,
1052         .open           = generic_file_open,
1053         .release        = btrfs_release_file,
1054         .fsync          = btrfs_sync_file,
1055         .unlocked_ioctl = btrfs_ioctl,
1056 #ifdef CONFIG_COMPAT
1057         .compat_ioctl   = btrfs_ioctl,
1058 #endif
1059 };
1060