Add a per-inode lock around btrfs_drop_extents
[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         int did_inline = 0;
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 < start_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 = (start_pos - last_pos_in_file + mask) & ~mask;
276                 if (hole_size > 0) {
277                         btrfs_wait_ordered_range(inode, last_pos_in_file,
278                                                  last_pos_in_file + hole_size);
279                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
280                         err = btrfs_drop_extents(trans, root, inode,
281                                                  last_pos_in_file,
282                                                  last_pos_in_file + hole_size,
283                                                  last_pos_in_file,
284                                                  &hint_byte);
285                         if (err)
286                                 goto failed;
287
288                         err = btrfs_insert_file_extent(trans, root,
289                                                        inode->i_ino,
290                                                        last_pos_in_file,
291                                                        0, 0, hole_size, 0);
292                         btrfs_drop_extent_cache(inode, last_pos_in_file,
293                                         last_pos_in_file + hole_size -1);
294                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
295                         btrfs_check_file(root, inode);
296                 }
297                 if (err)
298                         goto failed;
299         }
300
301         /*
302          * either allocate an extent for the new bytes or setup the key
303          * to show we are doing inline data in the extent
304          */
305         inline_size = end_pos;
306         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
307             inline_size > root->fs_info->max_inline ||
308             (inline_size & (root->sectorsize -1)) == 0 ||
309             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
310                 /* check for reserved extents on each page, we don't want
311                  * to reset the delalloc bit on things that already have
312                  * extents reserved.
313                  */
314                 set_extent_delalloc(io_tree, start_pos,
315                                     end_of_last_block, GFP_NOFS);
316                 for (i = 0; i < num_pages; i++) {
317                         struct page *p = pages[i];
318                         SetPageUptodate(p);
319                         ClearPageChecked(p);
320                         set_page_dirty(p);
321                 }
322         } else {
323                 u64 aligned_end;
324                 /* step one, delete the existing extents in this range */
325                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
326                         ~((u64)root->sectorsize - 1);
327                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
328                 err = btrfs_drop_extents(trans, root, inode, start_pos,
329                                          aligned_end, aligned_end, &hint_byte);
330                 if (err)
331                         goto failed;
332                 if (isize > inline_size)
333                         inline_size = min_t(u64, isize, aligned_end);
334                 inline_size -= start_pos;
335                 err = insert_inline_extent(trans, root, inode, start_pos,
336                                            inline_size, pages, 0, num_pages);
337                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
338                 BUG_ON(err);
339                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
340                 did_inline = 1;
341         }
342         if (end_pos > isize) {
343                 i_size_write(inode, end_pos);
344                 if (did_inline)
345                         BTRFS_I(inode)->disk_i_size = end_pos;
346                 btrfs_update_inode(trans, root, inode);
347         }
348 failed:
349         err = btrfs_end_transaction_throttle(trans, root);
350 out_unlock:
351         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
352         return err;
353 }
354
355 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
356 {
357         struct extent_map *em;
358         struct extent_map *split = NULL;
359         struct extent_map *split2 = NULL;
360         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
361         struct extent_map *tmp;
362         u64 len = end - start + 1;
363         u64 next_start;
364         int ret;
365         int testend = 1;
366
367         WARN_ON(end < start);
368         if (end == (u64)-1) {
369                 len = (u64)-1;
370                 testend = 0;
371         }
372         while(1) {
373                 if (!split)
374                         split = alloc_extent_map(GFP_NOFS);
375                 if (!split2)
376                         split2 = alloc_extent_map(GFP_NOFS);
377
378                 spin_lock(&em_tree->lock);
379                 em = lookup_extent_mapping(em_tree, start, len);
380                 if (!em) {
381                         spin_unlock(&em_tree->lock);
382                         break;
383                 }
384                 tmp = rb_entry(&em->rb_node, struct extent_map, rb_node);
385                 next_start = tmp->start;
386                 remove_extent_mapping(em_tree, em);
387
388                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
389                     em->start < start) {
390                         split->start = em->start;
391                         split->len = start - em->start;
392                         split->block_start = em->block_start;
393                         split->bdev = em->bdev;
394                         split->flags = em->flags;
395                         ret = add_extent_mapping(em_tree, split);
396                         BUG_ON(ret);
397                         free_extent_map(split);
398                         split = split2;
399                         split2 = NULL;
400                 }
401                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
402                     testend && em->start + em->len > start + len) {
403                         u64 diff = start + len - em->start;
404
405                         split->start = start + len;
406                         split->len = em->start + em->len - (start + len);
407                         split->bdev = em->bdev;
408                         split->flags = em->flags;
409
410                         split->block_start = em->block_start + diff;
411
412                         ret = add_extent_mapping(em_tree, split);
413                         BUG_ON(ret);
414                         free_extent_map(split);
415                         split = NULL;
416                 }
417                 spin_unlock(&em_tree->lock);
418
419                 /* once for us */
420                 free_extent_map(em);
421                 /* once for the tree*/
422                 free_extent_map(em);
423         }
424         if (split)
425                 free_extent_map(split);
426         if (split2)
427                 free_extent_map(split2);
428         return 0;
429 }
430
431 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
432 {
433         return 0;
434 #if 0
435         struct btrfs_path *path;
436         struct btrfs_key found_key;
437         struct extent_buffer *leaf;
438         struct btrfs_file_extent_item *extent;
439         u64 last_offset = 0;
440         int nritems;
441         int slot;
442         int found_type;
443         int ret;
444         int err = 0;
445         u64 extent_end = 0;
446
447         path = btrfs_alloc_path();
448         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
449                                        last_offset, 0);
450         while(1) {
451                 nritems = btrfs_header_nritems(path->nodes[0]);
452                 if (path->slots[0] >= nritems) {
453                         ret = btrfs_next_leaf(root, path);
454                         if (ret)
455                                 goto out;
456                         nritems = btrfs_header_nritems(path->nodes[0]);
457                 }
458                 slot = path->slots[0];
459                 leaf = path->nodes[0];
460                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
461                 if (found_key.objectid != inode->i_ino)
462                         break;
463                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
464                         goto out;
465
466                 if (found_key.offset < last_offset) {
467                         WARN_ON(1);
468                         btrfs_print_leaf(root, leaf);
469                         printk("inode %lu found offset %Lu expected %Lu\n",
470                                inode->i_ino, found_key.offset, last_offset);
471                         err = 1;
472                         goto out;
473                 }
474                 extent = btrfs_item_ptr(leaf, slot,
475                                         struct btrfs_file_extent_item);
476                 found_type = btrfs_file_extent_type(leaf, extent);
477                 if (found_type == BTRFS_FILE_EXTENT_REG) {
478                         extent_end = found_key.offset +
479                              btrfs_file_extent_num_bytes(leaf, extent);
480                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
481                         struct btrfs_item *item;
482                         item = btrfs_item_nr(leaf, slot);
483                         extent_end = found_key.offset +
484                              btrfs_file_extent_inline_len(leaf, item);
485                         extent_end = (extent_end + root->sectorsize - 1) &
486                                 ~((u64)root->sectorsize -1 );
487                 }
488                 last_offset = extent_end;
489                 path->slots[0]++;
490         }
491         if (0 && last_offset < inode->i_size) {
492                 WARN_ON(1);
493                 btrfs_print_leaf(root, leaf);
494                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
495                        last_offset, inode->i_size);
496                 err = 1;
497
498         }
499 out:
500         btrfs_free_path(path);
501         return err;
502 #endif
503 }
504
505 /*
506  * this is very complex, but the basic idea is to drop all extents
507  * in the range start - end.  hint_block is filled in with a block number
508  * that would be a good hint to the block allocator for this file.
509  *
510  * If an extent intersects the range but is not entirely inside the range
511  * it is either truncated or split.  Anything entirely inside the range
512  * is deleted from the tree.
513  */
514 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
515                        struct btrfs_root *root, struct inode *inode,
516                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
517 {
518         u64 extent_end = 0;
519         u64 search_start = start;
520         struct extent_buffer *leaf;
521         struct btrfs_file_extent_item *extent;
522         struct btrfs_path *path;
523         struct btrfs_key key;
524         struct btrfs_file_extent_item old;
525         int keep;
526         int slot;
527         int bookend;
528         int found_type;
529         int found_extent;
530         int found_inline;
531         int recow;
532         int ret;
533
534         btrfs_drop_extent_cache(inode, start, end - 1);
535
536         path = btrfs_alloc_path();
537         if (!path)
538                 return -ENOMEM;
539         while(1) {
540                 recow = 0;
541                 btrfs_release_path(root, path);
542                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
543                                                search_start, -1);
544                 if (ret < 0)
545                         goto out;
546                 if (ret > 0) {
547                         if (path->slots[0] == 0) {
548                                 ret = 0;
549                                 goto out;
550                         }
551                         path->slots[0]--;
552                 }
553 next_slot:
554                 keep = 0;
555                 bookend = 0;
556                 found_extent = 0;
557                 found_inline = 0;
558                 extent = NULL;
559                 leaf = path->nodes[0];
560                 slot = path->slots[0];
561                 ret = 0;
562                 btrfs_item_key_to_cpu(leaf, &key, slot);
563                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
564                     key.offset >= end) {
565                         goto out;
566                 }
567                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
568                     key.objectid != inode->i_ino) {
569                         goto out;
570                 }
571                 if (recow) {
572                         search_start = key.offset;
573                         continue;
574                 }
575                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
576                         extent = btrfs_item_ptr(leaf, slot,
577                                                 struct btrfs_file_extent_item);
578                         found_type = btrfs_file_extent_type(leaf, extent);
579                         if (found_type == BTRFS_FILE_EXTENT_REG) {
580                                 extent_end =
581                                      btrfs_file_extent_disk_bytenr(leaf,
582                                                                    extent);
583                                 if (extent_end)
584                                         *hint_byte = extent_end;
585
586                                 extent_end = key.offset +
587                                      btrfs_file_extent_num_bytes(leaf, extent);
588                                 found_extent = 1;
589                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
590                                 struct btrfs_item *item;
591                                 item = btrfs_item_nr(leaf, slot);
592                                 found_inline = 1;
593                                 extent_end = key.offset +
594                                      btrfs_file_extent_inline_len(leaf, item);
595                         }
596                 } else {
597                         extent_end = search_start;
598                 }
599
600                 /* we found nothing we can drop */
601                 if ((!found_extent && !found_inline) ||
602                     search_start >= extent_end) {
603                         int nextret;
604                         u32 nritems;
605                         nritems = btrfs_header_nritems(leaf);
606                         if (slot >= nritems - 1) {
607                                 nextret = btrfs_next_leaf(root, path);
608                                 if (nextret)
609                                         goto out;
610                                 recow = 1;
611                         } else {
612                                 path->slots[0]++;
613                         }
614                         goto next_slot;
615                 }
616
617                 if (found_inline) {
618                         u64 mask = root->sectorsize - 1;
619                         search_start = (extent_end + mask) & ~mask;
620                 } else
621                         search_start = extent_end;
622                 if (end <= extent_end && start >= key.offset && found_inline) {
623                         *hint_byte = EXTENT_MAP_INLINE;
624                         continue;
625                 }
626                 if (end < extent_end && end >= key.offset) {
627                         if (found_extent) {
628                                 u64 disk_bytenr =
629                                     btrfs_file_extent_disk_bytenr(leaf, extent);
630                                 u64 disk_num_bytes =
631                                     btrfs_file_extent_disk_num_bytes(leaf,
632                                                                       extent);
633                                 read_extent_buffer(leaf, &old,
634                                                    (unsigned long)extent,
635                                                    sizeof(old));
636                                 if (disk_bytenr != 0) {
637                                         ret = btrfs_inc_extent_ref(trans, root,
638                                                  disk_bytenr, disk_num_bytes,
639                                                  root->root_key.objectid,
640                                                  trans->transid,
641                                                  key.objectid, end);
642                                         BUG_ON(ret);
643                                 }
644                         }
645                         bookend = 1;
646                         if (found_inline && start <= key.offset)
647                                 keep = 1;
648                 }
649                 /* truncate existing extent */
650                 if (start > key.offset) {
651                         u64 new_num;
652                         u64 old_num;
653                         keep = 1;
654                         WARN_ON(start & (root->sectorsize - 1));
655                         if (found_extent) {
656                                 new_num = start - key.offset;
657                                 old_num = btrfs_file_extent_num_bytes(leaf,
658                                                                       extent);
659                                 *hint_byte =
660                                         btrfs_file_extent_disk_bytenr(leaf,
661                                                                       extent);
662                                 if (btrfs_file_extent_disk_bytenr(leaf,
663                                                                   extent)) {
664                                         dec_i_blocks(inode, old_num - new_num);
665                                 }
666                                 btrfs_set_file_extent_num_bytes(leaf, extent,
667                                                                 new_num);
668                                 btrfs_mark_buffer_dirty(leaf);
669                         } else if (key.offset < inline_limit &&
670                                    (end > extent_end) &&
671                                    (inline_limit < extent_end)) {
672                                 u32 new_size;
673                                 new_size = btrfs_file_extent_calc_inline_size(
674                                                    inline_limit - key.offset);
675                                 dec_i_blocks(inode, (extent_end - key.offset) -
676                                         (inline_limit - key.offset));
677                                 btrfs_truncate_item(trans, root, path,
678                                                     new_size, 1);
679                         }
680                 }
681                 /* delete the entire extent */
682                 if (!keep) {
683                         u64 disk_bytenr = 0;
684                         u64 disk_num_bytes = 0;
685                         u64 extent_num_bytes = 0;
686                         u64 root_gen;
687                         u64 root_owner;
688
689                         root_gen = btrfs_header_generation(leaf);
690                         root_owner = btrfs_header_owner(leaf);
691                         if (found_extent) {
692                                 disk_bytenr =
693                                       btrfs_file_extent_disk_bytenr(leaf,
694                                                                      extent);
695                                 disk_num_bytes =
696                                       btrfs_file_extent_disk_num_bytes(leaf,
697                                                                        extent);
698                                 extent_num_bytes =
699                                       btrfs_file_extent_num_bytes(leaf, extent);
700                                 *hint_byte =
701                                         btrfs_file_extent_disk_bytenr(leaf,
702                                                                       extent);
703                         }
704                         ret = btrfs_del_item(trans, root, path);
705                         /* TODO update progress marker and return */
706                         BUG_ON(ret);
707                         btrfs_release_path(root, path);
708                         extent = NULL;
709                         if (found_extent && disk_bytenr != 0) {
710                                 dec_i_blocks(inode, extent_num_bytes);
711                                 ret = btrfs_free_extent(trans, root,
712                                                 disk_bytenr,
713                                                 disk_num_bytes,
714                                                 root_owner,
715                                                 root_gen, inode->i_ino,
716                                                 key.offset, 0);
717                         }
718
719                         BUG_ON(ret);
720                         if (!bookend && search_start >= end) {
721                                 ret = 0;
722                                 goto out;
723                         }
724                         if (!bookend)
725                                 continue;
726                 }
727                 if (bookend && found_inline && start <= key.offset) {
728                         u32 new_size;
729                         new_size = btrfs_file_extent_calc_inline_size(
730                                                    extent_end - end);
731                         dec_i_blocks(inode, (extent_end - key.offset) -
732                                         (extent_end - end));
733                         btrfs_truncate_item(trans, root, path, new_size, 0);
734                 }
735                 /* create bookend, splitting the extent in two */
736                 if (bookend && found_extent) {
737                         struct btrfs_key ins;
738                         ins.objectid = inode->i_ino;
739                         ins.offset = end;
740                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
741                         btrfs_release_path(root, path);
742                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
743                                                       sizeof(*extent));
744
745                         leaf = path->nodes[0];
746                         if (ret) {
747                                 btrfs_print_leaf(root, leaf);
748                                 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);
749                         }
750                         BUG_ON(ret);
751                         extent = btrfs_item_ptr(leaf, path->slots[0],
752                                                 struct btrfs_file_extent_item);
753                         write_extent_buffer(leaf, &old,
754                                             (unsigned long)extent, sizeof(old));
755
756                         btrfs_set_file_extent_offset(leaf, extent,
757                                     le64_to_cpu(old.offset) + end - key.offset);
758                         WARN_ON(le64_to_cpu(old.num_bytes) <
759                                 (extent_end - end));
760                         btrfs_set_file_extent_num_bytes(leaf, extent,
761                                                         extent_end - end);
762                         btrfs_set_file_extent_type(leaf, extent,
763                                                    BTRFS_FILE_EXTENT_REG);
764
765                         btrfs_mark_buffer_dirty(path->nodes[0]);
766                         if (le64_to_cpu(old.disk_bytenr) != 0) {
767                                 inode->i_blocks +=
768                                       btrfs_file_extent_num_bytes(leaf,
769                                                                   extent) >> 9;
770                         }
771                         ret = 0;
772                         goto out;
773                 }
774         }
775 out:
776         btrfs_free_path(path);
777         btrfs_check_file(root, inode);
778         return ret;
779 }
780
781 /*
782  * this gets pages into the page cache and locks them down
783  */
784 static int prepare_pages(struct btrfs_root *root, struct file *file,
785                          struct page **pages, size_t num_pages,
786                          loff_t pos, unsigned long first_index,
787                          unsigned long last_index, size_t write_bytes)
788 {
789         int i;
790         unsigned long index = pos >> PAGE_CACHE_SHIFT;
791         struct inode *inode = fdentry(file)->d_inode;
792         int err = 0;
793         u64 start_pos;
794         u64 last_pos;
795
796         start_pos = pos & ~((u64)root->sectorsize - 1);
797         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
798
799         memset(pages, 0, num_pages * sizeof(struct page *));
800 again:
801         for (i = 0; i < num_pages; i++) {
802                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
803                 if (!pages[i]) {
804                         err = -ENOMEM;
805                         BUG_ON(1);
806                 }
807                 wait_on_page_writeback(pages[i]);
808         }
809         if (start_pos < inode->i_size) {
810                 struct btrfs_ordered_extent *ordered;
811                 lock_extent(&BTRFS_I(inode)->io_tree,
812                             start_pos, last_pos - 1, GFP_NOFS);
813                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
814                 if (ordered &&
815                     ordered->file_offset + ordered->len > start_pos &&
816                     ordered->file_offset < last_pos) {
817                         btrfs_put_ordered_extent(ordered);
818                         unlock_extent(&BTRFS_I(inode)->io_tree,
819                                       start_pos, last_pos - 1, GFP_NOFS);
820                         for (i = 0; i < num_pages; i++) {
821                                 unlock_page(pages[i]);
822                                 page_cache_release(pages[i]);
823                         }
824                         btrfs_wait_ordered_range(inode, start_pos,
825                                                  last_pos - start_pos);
826                         goto again;
827                 }
828                 if (ordered)
829                         btrfs_put_ordered_extent(ordered);
830
831                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
832                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
833                                   GFP_NOFS);
834                 unlock_extent(&BTRFS_I(inode)->io_tree,
835                               start_pos, last_pos - 1, GFP_NOFS);
836         }
837         for (i = 0; i < num_pages; i++) {
838 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
839                 ClearPageDirty(pages[i]);
840 #else
841                 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
842 #endif
843                 set_page_extent_mapped(pages[i]);
844                 WARN_ON(!PageLocked(pages[i]));
845         }
846         return 0;
847 }
848
849 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
850                                 size_t count, loff_t *ppos)
851 {
852         loff_t pos;
853         loff_t start_pos;
854         ssize_t num_written = 0;
855         ssize_t err = 0;
856         int ret = 0;
857         struct inode *inode = fdentry(file)->d_inode;
858         struct btrfs_root *root = BTRFS_I(inode)->root;
859         struct page **pages = NULL;
860         int nrptrs;
861         struct page *pinned[2];
862         unsigned long first_index;
863         unsigned long last_index;
864
865         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
866                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
867         pinned[0] = NULL;
868         pinned[1] = NULL;
869
870         pos = *ppos;
871         start_pos = pos;
872
873         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
874         current->backing_dev_info = inode->i_mapping->backing_dev_info;
875         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
876         if (err)
877                 goto out_nolock;
878         if (count == 0)
879                 goto out_nolock;
880 #ifdef REMOVE_SUID_PATH
881         err = remove_suid(&file->f_path);
882 #else
883         err = remove_suid(fdentry(file));
884 #endif
885         if (err)
886                 goto out_nolock;
887         file_update_time(file);
888
889         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
890
891         mutex_lock(&inode->i_mutex);
892         first_index = pos >> PAGE_CACHE_SHIFT;
893         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
894
895         /*
896          * if this is a nodatasum mount, force summing off for the inode
897          * all the time.  That way a later mount with summing on won't
898          * get confused
899          */
900         if (btrfs_test_opt(root, NODATASUM))
901                 btrfs_set_flag(inode, NODATASUM);
902
903         /*
904          * there are lots of better ways to do this, but this code
905          * makes sure the first and last page in the file range are
906          * up to date and ready for cow
907          */
908         if ((pos & (PAGE_CACHE_SIZE - 1))) {
909                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
910                 if (!PageUptodate(pinned[0])) {
911                         ret = btrfs_readpage(NULL, pinned[0]);
912                         BUG_ON(ret);
913                         wait_on_page_locked(pinned[0]);
914                 } else {
915                         unlock_page(pinned[0]);
916                 }
917         }
918         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
919                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
920                 if (!PageUptodate(pinned[1])) {
921                         ret = btrfs_readpage(NULL, pinned[1]);
922                         BUG_ON(ret);
923                         wait_on_page_locked(pinned[1]);
924                 } else {
925                         unlock_page(pinned[1]);
926                 }
927         }
928
929         while(count > 0) {
930                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
931                 size_t write_bytes = min(count, nrptrs *
932                                         (size_t)PAGE_CACHE_SIZE -
933                                          offset);
934                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
935                                         PAGE_CACHE_SHIFT;
936
937                 WARN_ON(num_pages > nrptrs);
938                 memset(pages, 0, sizeof(pages));
939
940                 ret = btrfs_check_free_space(root, write_bytes, 0);
941                 if (ret)
942                         goto out;
943
944                 ret = prepare_pages(root, file, pages, num_pages,
945                                     pos, first_index, last_index,
946                                     write_bytes);
947                 if (ret)
948                         goto out;
949
950                 ret = btrfs_copy_from_user(pos, num_pages,
951                                            write_bytes, pages, buf);
952                 if (ret) {
953                         btrfs_drop_pages(pages, num_pages);
954                         goto out;
955                 }
956
957                 ret = dirty_and_release_pages(NULL, root, file, pages,
958                                               num_pages, pos, write_bytes);
959                 btrfs_drop_pages(pages, num_pages);
960                 if (ret)
961                         goto out;
962
963                 buf += write_bytes;
964                 count -= write_bytes;
965                 pos += write_bytes;
966                 num_written += write_bytes;
967
968                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
969                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
970                         btrfs_btree_balance_dirty(root, 1);
971                 cond_resched();
972         }
973 out:
974         mutex_unlock(&inode->i_mutex);
975
976 out_nolock:
977         kfree(pages);
978         if (pinned[0])
979                 page_cache_release(pinned[0]);
980         if (pinned[1])
981                 page_cache_release(pinned[1]);
982         *ppos = pos;
983
984         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
985                 err = sync_page_range(inode, inode->i_mapping,
986                                       start_pos, num_written);
987                 if (err < 0)
988                         num_written = err;
989         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
990 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
991                 do_sync_file_range(file, start_pos,
992                                       start_pos + num_written - 1,
993                                       SYNC_FILE_RANGE_WRITE |
994                                       SYNC_FILE_RANGE_WAIT_AFTER);
995 #else
996                 do_sync_mapping_range(inode->i_mapping, start_pos,
997                                       start_pos + num_written - 1,
998                                       SYNC_FILE_RANGE_WRITE |
999                                       SYNC_FILE_RANGE_WAIT_AFTER);
1000 #endif
1001                 invalidate_mapping_pages(inode->i_mapping,
1002                       start_pos >> PAGE_CACHE_SHIFT,
1003                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1004         }
1005         current->backing_dev_info = NULL;
1006         return num_written ? num_written : err;
1007 }
1008
1009 int btrfs_release_file(struct inode * inode, struct file * filp)
1010 {
1011         if (filp->private_data)
1012                 btrfs_ioctl_trans_end(filp);
1013         return 0;
1014 }
1015
1016 static int btrfs_sync_file(struct file *file,
1017                            struct dentry *dentry, int datasync)
1018 {
1019         struct inode *inode = dentry->d_inode;
1020         struct btrfs_root *root = BTRFS_I(inode)->root;
1021         int ret = 0;
1022         struct btrfs_trans_handle *trans;
1023
1024         /*
1025          * check the transaction that last modified this inode
1026          * and see if its already been committed
1027          */
1028         if (!BTRFS_I(inode)->last_trans)
1029                 goto out;
1030
1031         mutex_lock(&root->fs_info->trans_mutex);
1032         if (BTRFS_I(inode)->last_trans <=
1033             root->fs_info->last_trans_committed) {
1034                 BTRFS_I(inode)->last_trans = 0;
1035                 mutex_unlock(&root->fs_info->trans_mutex);
1036                 goto out;
1037         }
1038         mutex_unlock(&root->fs_info->trans_mutex);
1039
1040         /*
1041          * ok we haven't committed the transaction yet, lets do a commit
1042          */
1043         if (file->private_data)
1044                 btrfs_ioctl_trans_end(file);
1045
1046         trans = btrfs_start_transaction(root, 1);
1047         if (!trans) {
1048                 ret = -ENOMEM;
1049                 goto out;
1050         }
1051         ret = btrfs_commit_transaction(trans, root);
1052 out:
1053         return ret > 0 ? EIO : ret;
1054 }
1055
1056 static struct vm_operations_struct btrfs_file_vm_ops = {
1057 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1058         .nopage         = filemap_nopage,
1059         .populate       = filemap_populate,
1060 #else
1061         .fault          = filemap_fault,
1062 #endif
1063         .page_mkwrite   = btrfs_page_mkwrite,
1064 };
1065
1066 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1067 {
1068         vma->vm_ops = &btrfs_file_vm_ops;
1069         file_accessed(filp);
1070         return 0;
1071 }
1072
1073 struct file_operations btrfs_file_operations = {
1074         .llseek         = generic_file_llseek,
1075         .read           = do_sync_read,
1076         .aio_read       = generic_file_aio_read,
1077         .splice_read    = generic_file_splice_read,
1078 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1079         .sendfile       = generic_file_sendfile,
1080 #endif
1081         .write          = btrfs_file_write,
1082         .mmap           = btrfs_file_mmap,
1083         .open           = generic_file_open,
1084         .release        = btrfs_release_file,
1085         .fsync          = btrfs_sync_file,
1086         .unlocked_ioctl = btrfs_ioctl,
1087 #ifdef CONFIG_COMPAT
1088         .compat_ioctl   = btrfs_ioctl,
1089 #endif
1090 };
1091