?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: Throttle operations if the reference cache gets too large
[linux-2.6/kvm.git] / fs / btrfs / file.c
blobded5281f8463533cd8346b588432ef17ca6e20b0
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 ClearPageChecked(pages[i]);
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 int did_inline = 0;
247 loff_t isize = i_size_read(inode);
248
249 start_pos = pos & ~((u64)root->sectorsize - 1);
250 num_bytes = (write_bytes + pos - start_pos +
251 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
252
253 end_of_last_block = start_pos + num_bytes - 1;
254
255 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
256 trans = btrfs_start_transaction(root, 1);
257 if (!trans) {
258 err = -ENOMEM;
259 goto out_unlock;
260 }
261 btrfs_set_trans_block_group(trans, inode);
262 hint_byte = 0;
263
264 if ((end_of_last_block & 4095) == 0) {
265 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
266 }
267 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
268
269 /* FIXME...EIEIO, ENOSPC and more */
270 /* insert any holes we need to create */
271 if (isize < start_pos) {
272 u64 last_pos_in_file;
273 u64 hole_size;
274 u64 mask = root->sectorsize - 1;
275 last_pos_in_file = (isize + mask) & ~mask;
276 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
277 if (hole_size > 0) {
278 btrfs_wait_ordered_range(inode, last_pos_in_file,
279 last_pos_in_file + hole_size);
280 mutex_lock(&BTRFS_I(inode)->extent_mutex);
281 err = btrfs_drop_extents(trans, root, inode,
282 last_pos_in_file,
283 last_pos_in_file + hole_size,
284 last_pos_in_file,
285 &hint_byte);
286 if (err)
287 goto failed;
288
289 err = btrfs_insert_file_extent(trans, root,
290 inode->i_ino,
291 last_pos_in_file,
292 0, 0, hole_size, 0);
293 btrfs_drop_extent_cache(inode, last_pos_in_file,
294 last_pos_in_file + hole_size -1);
295 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
296 btrfs_check_file(root, inode);
297 }
298 if (err)
299 goto failed;
300 }
301
302 /*
303 * either allocate an extent for the new bytes or setup the key
304 * to show we are doing inline data in the extent
305 */
306 inline_size = end_pos;
307 if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
308 inline_size > root->fs_info->max_inline ||
309 (inline_size & (root->sectorsize -1)) == 0 ||
310 inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
311 /* check for reserved extents on each page, we don't want
312 * to reset the delalloc bit on things that already have
313 * extents reserved.
314 */
315 set_extent_delalloc(io_tree, start_pos,
316 end_of_last_block, GFP_NOFS);
317 for (i = 0; i < num_pages; i++) {
318 struct page *p = pages[i];
319 SetPageUptodate(p);
320 ClearPageChecked(p);
321 set_page_dirty(p);
322 }
323 } else {
324 u64 aligned_end;
325 /* step one, delete the existing extents in this range */
326 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
327 ~((u64)root->sectorsize - 1);
328 mutex_lock(&BTRFS_I(inode)->extent_mutex);
329 err = btrfs_drop_extents(trans, root, inode, start_pos,
330 aligned_end, aligned_end, &hint_byte);
331 if (err)
332 goto failed;
333 if (isize > inline_size)
334 inline_size = min_t(u64, isize, aligned_end);
335 inline_size -= start_pos;
336 err = insert_inline_extent(trans, root, inode, start_pos,
337 inline_size, pages, 0, num_pages);
338 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
339 BUG_ON(err);
340 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
341 did_inline = 1;
342 }
343 if (end_pos > isize) {
344 i_size_write(inode, end_pos);
345 if (did_inline)
346 BTRFS_I(inode)->disk_i_size = end_pos;
347 btrfs_update_inode(trans, root, inode);
348 }
349 failed:
350 err = btrfs_end_transaction(trans, root);
351 out_unlock:
352 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
353 return err;
354 }
355
356 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
357 {
358 struct extent_map *em;
359 struct extent_map *split = NULL;
360 struct extent_map *split2 = NULL;
361 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
362 u64 len = end - start + 1;
363 int ret;
364 int testend = 1;
365
366 WARN_ON(end < start);
367 if (end == (u64)-1) {
368 len = (u64)-1;
369 testend = 0;
370 }
371 while(1) {
372 if (!split)
373 split = alloc_extent_map(GFP_NOFS);
374 if (!split2)
375 split2 = alloc_extent_map(GFP_NOFS);
376
377 spin_lock(&em_tree->lock);
378 em = lookup_extent_mapping(em_tree, start, len);
379 if (!em) {
380 spin_unlock(&em_tree->lock);
381 break;
382 }
383 if (test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
384 printk(KERN_CRIT "inode %lu trying to drop pinned "
385 "extent start %llu end %llu, em [%llu %llu]\n",
386 inode->i_ino,
387 (unsigned long long)start,
388 (unsigned long long)end,
389 (unsigned long long)em->start,
390 (unsigned long long)em->len);
391 }
392 remove_extent_mapping(em_tree, em);
393
394 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
395 em->start < start) {
396 split->start = em->start;
397 split->len = start - em->start;
398 split->block_start = em->block_start;
399 split->bdev = em->bdev;
400 split->flags = em->flags;
401 ret = add_extent_mapping(em_tree, split);
402 BUG_ON(ret);
403 free_extent_map(split);
404 split = split2;
405 split2 = NULL;
406 }
407 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
408 testend && em->start + em->len > start + len) {
409 u64 diff = start + len - em->start;
410
411 split->start = start + len;
412 split->len = em->start + em->len - (start + len);
413 split->bdev = em->bdev;
414 split->flags = em->flags;
415
416 split->block_start = em->block_start + diff;
417
418 ret = add_extent_mapping(em_tree, split);
419 BUG_ON(ret);
420 free_extent_map(split);
421 split = NULL;
422 }
423 spin_unlock(&em_tree->lock);
424
425 /* once for us */
426 free_extent_map(em);
427 /* once for the tree*/
428 free_extent_map(em);
429 }
430 if (split)
431 free_extent_map(split);
432 if (split2)
433 free_extent_map(split2);
434 return 0;
435 }
436
437 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
438 {
439 return 0;
440 #if 0
441 struct btrfs_path *path;
442 struct btrfs_key found_key;
443 struct extent_buffer *leaf;
444 struct btrfs_file_extent_item *extent;
445 u64 last_offset = 0;
446 int nritems;
447 int slot;
448 int found_type;
449 int ret;
450 int err = 0;
451 u64 extent_end = 0;
452
453 path = btrfs_alloc_path();
454 ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
455 last_offset, 0);
456 while(1) {
457 nritems = btrfs_header_nritems(path->nodes[0]);
458 if (path->slots[0] >= nritems) {
459 ret = btrfs_next_leaf(root, path);
460 if (ret)
461 goto out;
462 nritems = btrfs_header_nritems(path->nodes[0]);
463 }
464 slot = path->slots[0];
465 leaf = path->nodes[0];
466 btrfs_item_key_to_cpu(leaf, &found_key, slot);
467 if (found_key.objectid != inode->i_ino)
468 break;
469 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
470 goto out;
471
472 if (found_key.offset < last_offset) {
473 WARN_ON(1);
474 btrfs_print_leaf(root, leaf);
475 printk("inode %lu found offset %Lu expected %Lu\n",
476 inode->i_ino, found_key.offset, last_offset);
477 err = 1;
478 goto out;
479 }
480 extent = btrfs_item_ptr(leaf, slot,
481 struct btrfs_file_extent_item);
482 found_type = btrfs_file_extent_type(leaf, extent);
483 if (found_type == BTRFS_FILE_EXTENT_REG) {
484 extent_end = found_key.offset +
485 btrfs_file_extent_num_bytes(leaf, extent);
486 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
487 struct btrfs_item *item;
488 item = btrfs_item_nr(leaf, slot);
489 extent_end = found_key.offset +
490 btrfs_file_extent_inline_len(leaf, item);
491 extent_end = (extent_end + root->sectorsize - 1) &
492 ~((u64)root->sectorsize -1 );
493 }
494 last_offset = extent_end;
495 path->slots[0]++;
496 }
497 if (0 && last_offset < inode->i_size) {
498 WARN_ON(1);
499 btrfs_print_leaf(root, leaf);
500 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
501 last_offset, inode->i_size);
502 err = 1;
503
504 }
505 out:
506 btrfs_free_path(path);
507 return err;
508 #endif
509 }
510
511 /*
512 * this is very complex, but the basic idea is to drop all extents
513 * in the range start - end. hint_block is filled in with a block number
514 * that would be a good hint to the block allocator for this file.
515 *
516 * If an extent intersects the range but is not entirely inside the range
517 * it is either truncated or split. Anything entirely inside the range
518 * is deleted from the tree.
519 */
520 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
521 struct btrfs_root *root, struct inode *inode,
522 u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
523 {
524 u64 extent_end = 0;
525 u64 search_start = start;
526 struct extent_buffer *leaf;
527 struct btrfs_file_extent_item *extent;
528 struct btrfs_path *path;
529 struct btrfs_key key;
530 struct btrfs_file_extent_item old;
531 int keep;
532 int slot;
533 int bookend;
534 int found_type;
535 int found_extent;
536 int found_inline;
537 int recow;
538 int ret;
539
540 btrfs_drop_extent_cache(inode, start, end - 1);
541
542 path = btrfs_alloc_path();
543 if (!path)
544 return -ENOMEM;
545 while(1) {
546 recow = 0;
547 btrfs_release_path(root, path);
548 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
549 search_start, -1);
550 if (ret < 0)
551 goto out;
552 if (ret > 0) {
553 if (path->slots[0] == 0) {
554 ret = 0;
555 goto out;
556 }
557 path->slots[0]--;
558 }
559 next_slot:
560 keep = 0;
561 bookend = 0;
562 found_extent = 0;
563 found_inline = 0;
564 extent = NULL;
565 leaf = path->nodes[0];
566 slot = path->slots[0];
567 ret = 0;
568 btrfs_item_key_to_cpu(leaf, &key, slot);
569 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
570 key.offset >= end) {
571 goto out;
572 }
573 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
574 key.objectid != inode->i_ino) {
575 goto out;
576 }
577 if (recow) {
578 search_start = key.offset;
579 continue;
580 }
581 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
582 extent = btrfs_item_ptr(leaf, slot,
583 struct btrfs_file_extent_item);
584 found_type = btrfs_file_extent_type(leaf, extent);
585 if (found_type == BTRFS_FILE_EXTENT_REG) {
586 extent_end =
587 btrfs_file_extent_disk_bytenr(leaf,
588 extent);
589 if (extent_end)
590 *hint_byte = extent_end;
591
592 extent_end = key.offset +
593 btrfs_file_extent_num_bytes(leaf, extent);
594 found_extent = 1;
595 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
596 struct btrfs_item *item;
597 item = btrfs_item_nr(leaf, slot);
598 found_inline = 1;
599 extent_end = key.offset +
600 btrfs_file_extent_inline_len(leaf, item);
601 }
602 } else {
603 extent_end = search_start;
604 }
605
606 /* we found nothing we can drop */
607 if ((!found_extent && !found_inline) ||
608 search_start >= extent_end) {
609 int nextret;
610 u32 nritems;
611 nritems = btrfs_header_nritems(leaf);
612 if (slot >= nritems - 1) {
613 nextret = btrfs_next_leaf(root, path);
614 if (nextret)
615 goto out;
616 recow = 1;
617 } else {
618 path->slots[0]++;
619 }
620 goto next_slot;
621 }
622
623 if (found_inline) {
624 u64 mask = root->sectorsize - 1;
625 search_start = (extent_end + mask) & ~mask;
626 } else
627 search_start = extent_end;
628 if (end <= extent_end && start >= key.offset && found_inline) {
629 *hint_byte = EXTENT_MAP_INLINE;
630 continue;
631 }
632 if (end < extent_end && end >= key.offset) {
633 if (found_extent) {
634 u64 disk_bytenr =
635 btrfs_file_extent_disk_bytenr(leaf, extent);
636 u64 disk_num_bytes =
637 btrfs_file_extent_disk_num_bytes(leaf,
638 extent);
639 read_extent_buffer(leaf, &old,
640 (unsigned long)extent,
641 sizeof(old));
642 if (disk_bytenr != 0) {
643 ret = btrfs_inc_extent_ref(trans, root,
644 disk_bytenr, disk_num_bytes,
645 root->root_key.objectid,
646 trans->transid,
647 key.objectid, end);
648 BUG_ON(ret);
649 }
650 }
651 bookend = 1;
652 if (found_inline && start <= key.offset)
653 keep = 1;
654 }
655 /* truncate existing extent */
656 if (start > key.offset) {
657 u64 new_num;
658 u64 old_num;
659 keep = 1;
660 WARN_ON(start & (root->sectorsize - 1));
661 if (found_extent) {
662 new_num = start - key.offset;
663 old_num = btrfs_file_extent_num_bytes(leaf,
664 extent);
665 *hint_byte =
666 btrfs_file_extent_disk_bytenr(leaf,
667 extent);
668 if (btrfs_file_extent_disk_bytenr(leaf,
669 extent)) {
670 dec_i_blocks(inode, old_num - new_num);
671 }
672 btrfs_set_file_extent_num_bytes(leaf, extent,
673 new_num);
674 btrfs_mark_buffer_dirty(leaf);
675 } else if (key.offset < inline_limit &&
676 (end > extent_end) &&
677 (inline_limit < extent_end)) {
678 u32 new_size;
679 new_size = btrfs_file_extent_calc_inline_size(
680 inline_limit - key.offset);
681 dec_i_blocks(inode, (extent_end - key.offset) -
682 (inline_limit - key.offset));
683 btrfs_truncate_item(trans, root, path,
684 new_size, 1);
685 }
686 }
687 /* delete the entire extent */
688 if (!keep) {
689 u64 disk_bytenr = 0;
690 u64 disk_num_bytes = 0;
691 u64 extent_num_bytes = 0;
692 u64 root_gen;
693 u64 root_owner;
694
695 root_gen = btrfs_header_generation(leaf);
696 root_owner = btrfs_header_owner(leaf);
697 if (found_extent) {
698 disk_bytenr =
699 btrfs_file_extent_disk_bytenr(leaf,
700 extent);
701 disk_num_bytes =
702 btrfs_file_extent_disk_num_bytes(leaf,
703 extent);
704 extent_num_bytes =
705 btrfs_file_extent_num_bytes(leaf, extent);
706 *hint_byte =
707 btrfs_file_extent_disk_bytenr(leaf,
708 extent);
709 }
710 ret = btrfs_del_item(trans, root, path);
711 /* TODO update progress marker and return */
712 BUG_ON(ret);
713 btrfs_release_path(root, path);
714 extent = NULL;
715 if (found_extent && disk_bytenr != 0) {
716 dec_i_blocks(inode, extent_num_bytes);
717 ret = btrfs_free_extent(trans, root,
718 disk_bytenr,
719 disk_num_bytes,
720 root_owner,
721 root_gen, inode->i_ino,
722 key.offset, 0);
723 }
724
725 BUG_ON(ret);
726 if (!bookend && search_start >= end) {
727 ret = 0;
728 goto out;
729 }
730 if (!bookend)
731 continue;
732 }
733 if (bookend && found_inline && start <= key.offset) {
734 u32 new_size;
735 new_size = btrfs_file_extent_calc_inline_size(
736 extent_end - end);
737 dec_i_blocks(inode, (extent_end - key.offset) -
738 (extent_end - end));
739 btrfs_truncate_item(trans, root, path, new_size, 0);
740 }
741 /* create bookend, splitting the extent in two */
742 if (bookend && found_extent) {
743 struct btrfs_key ins;
744 ins.objectid = inode->i_ino;
745 ins.offset = end;
746 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
747 btrfs_release_path(root, path);
748 ret = btrfs_insert_empty_item(trans, root, path, &ins,
749 sizeof(*extent));
750
751 leaf = path->nodes[0];
752 if (ret) {
753 btrfs_print_leaf(root, leaf);
754 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);
755 }
756 BUG_ON(ret);
757 extent = btrfs_item_ptr(leaf, path->slots[0],
758 struct btrfs_file_extent_item);
759 write_extent_buffer(leaf, &old,
760 (unsigned long)extent, sizeof(old));
761
762 btrfs_set_file_extent_offset(leaf, extent,
763 le64_to_cpu(old.offset) + end - key.offset);
764 WARN_ON(le64_to_cpu(old.num_bytes) <
765 (extent_end - end));
766 btrfs_set_file_extent_num_bytes(leaf, extent,
767 extent_end - end);
768 btrfs_set_file_extent_type(leaf, extent,
769 BTRFS_FILE_EXTENT_REG);
770
771 btrfs_mark_buffer_dirty(path->nodes[0]);
772 if (le64_to_cpu(old.disk_bytenr) != 0) {
773 inode->i_blocks +=
774 btrfs_file_extent_num_bytes(leaf,
775 extent) >> 9;
776 }
777 ret = 0;
778 goto out;
779 }
780 }
781 out:
782 btrfs_free_path(path);
783 btrfs_check_file(root, inode);
784 return ret;
785 }
786
787 /*
788 * this gets pages into the page cache and locks them down
789 */
790 static int prepare_pages(struct btrfs_root *root, struct file *file,
791 struct page **pages, size_t num_pages,
792 loff_t pos, unsigned long first_index,
793 unsigned long last_index, size_t write_bytes)
794 {
795 int i;
796 unsigned long index = pos >> PAGE_CACHE_SHIFT;
797 struct inode *inode = fdentry(file)->d_inode;
798 int err = 0;
799 u64 start_pos;
800 u64 last_pos;
801
802 start_pos = pos & ~((u64)root->sectorsize - 1);
803 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
804
805 memset(pages, 0, num_pages * sizeof(struct page *));
806 again:
807 for (i = 0; i < num_pages; i++) {
808 pages[i] = grab_cache_page(inode->i_mapping, index + i);
809 if (!pages[i]) {
810 err = -ENOMEM;
811 BUG_ON(1);
812 }
813 wait_on_page_writeback(pages[i]);
814 }
815 if (start_pos < inode->i_size) {
816 struct btrfs_ordered_extent *ordered;
817 lock_extent(&BTRFS_I(inode)->io_tree,
818 start_pos, last_pos - 1, GFP_NOFS);
819 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
820 if (ordered &&
821 ordered->file_offset + ordered->len > start_pos &&
822 ordered->file_offset < last_pos) {
823 btrfs_put_ordered_extent(ordered);
824 unlock_extent(&BTRFS_I(inode)->io_tree,
825 start_pos, last_pos - 1, GFP_NOFS);
826 for (i = 0; i < num_pages; i++) {
827 unlock_page(pages[i]);
828 page_cache_release(pages[i]);
829 }
830 btrfs_wait_ordered_range(inode, start_pos,
831 last_pos - start_pos);
832 goto again;
833 }
834 if (ordered)
835 btrfs_put_ordered_extent(ordered);
836
837 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
838 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
839 GFP_NOFS);
840 unlock_extent(&BTRFS_I(inode)->io_tree,
841 start_pos, last_pos - 1, GFP_NOFS);
842 }
843 for (i = 0; i < num_pages; i++) {
844 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
845 ClearPageDirty(pages[i]);
846 #else
847 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
848 #endif
849 set_page_extent_mapped(pages[i]);
850 WARN_ON(!PageLocked(pages[i]));
851 }
852 return 0;
853 }
854
855 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
856 size_t count, loff_t *ppos)
857 {
858 loff_t pos;
859 loff_t start_pos;
860 ssize_t num_written = 0;
861 ssize_t err = 0;
862 int ret = 0;
863 struct inode *inode = fdentry(file)->d_inode;
864 struct btrfs_root *root = BTRFS_I(inode)->root;
865 struct page **pages = NULL;
866 int nrptrs;
867 struct page *pinned[2];
868 unsigned long first_index;
869 unsigned long last_index;
870
871 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
872 PAGE_CACHE_SIZE / (sizeof(struct page *)));
873 pinned[0] = NULL;
874 pinned[1] = NULL;
875
876 pos = *ppos;
877 start_pos = pos;
878
879 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
880 current->backing_dev_info = inode->i_mapping->backing_dev_info;
881 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
882 if (err)
883 goto out_nolock;
884 if (count == 0)
885 goto out_nolock;
886 #ifdef REMOVE_SUID_PATH
887 err = remove_suid(&file->f_path);
888 #else
889 err = remove_suid(fdentry(file));
890 #endif
891 if (err)
892 goto out_nolock;
893 file_update_time(file);
894
895 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
896
897 mutex_lock(&inode->i_mutex);
898 first_index = pos >> PAGE_CACHE_SHIFT;
899 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
900
901 /*
902 * if this is a nodatasum mount, force summing off for the inode
903 * all the time. That way a later mount with summing on won't
904 * get confused
905 */
906 if (btrfs_test_opt(root, NODATASUM))
907 btrfs_set_flag(inode, NODATASUM);
908
909 /*
910 * there are lots of better ways to do this, but this code
911 * makes sure the first and last page in the file range are
912 * up to date and ready for cow
913 */
914 if ((pos & (PAGE_CACHE_SIZE - 1))) {
915 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
916 if (!PageUptodate(pinned[0])) {
917 ret = btrfs_readpage(NULL, pinned[0]);
918 BUG_ON(ret);
919 wait_on_page_locked(pinned[0]);
920 } else {
921 unlock_page(pinned[0]);
922 }
923 }
924 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
925 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
926 if (!PageUptodate(pinned[1])) {
927 ret = btrfs_readpage(NULL, pinned[1]);
928 BUG_ON(ret);
929 wait_on_page_locked(pinned[1]);
930 } else {
931 unlock_page(pinned[1]);
932 }
933 }
934
935 while(count > 0) {
936 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
937 size_t write_bytes = min(count, nrptrs *
938 (size_t)PAGE_CACHE_SIZE -
939 offset);
940 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
941 PAGE_CACHE_SHIFT;
942
943 WARN_ON(num_pages > nrptrs);
944 memset(pages, 0, sizeof(pages));
945
946 ret = btrfs_check_free_space(root, write_bytes, 0);
947 if (ret)
948 goto out;
949
950 ret = prepare_pages(root, file, pages, num_pages,
951 pos, first_index, last_index,
952 write_bytes);
953 if (ret)
954 goto out;
955
956 ret = btrfs_copy_from_user(pos, num_pages,
957 write_bytes, pages, buf);
958 if (ret) {
959 btrfs_drop_pages(pages, num_pages);
960 goto out;
961 }
962
963 ret = dirty_and_release_pages(NULL, root, file, pages,
964 num_pages, pos, write_bytes);
965 btrfs_drop_pages(pages, num_pages);
966 if (ret)
967 goto out;
968
969 buf += write_bytes;
970 count -= write_bytes;
971 pos += write_bytes;
972 num_written += write_bytes;
973
974 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
975 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
976 btrfs_btree_balance_dirty(root, 1);
977 btrfs_throttle(root);
978 cond_resched();
979 }
980 out:
981 mutex_unlock(&inode->i_mutex);
982
983 out_nolock:
984 kfree(pages);
985 if (pinned[0])
986 page_cache_release(pinned[0]);
987 if (pinned[1])
988 page_cache_release(pinned[1]);
989 *ppos = pos;
990
991 if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
992 err = sync_page_range(inode, inode->i_mapping,
993 start_pos, num_written);
994 if (err < 0)
995 num_written = err;
996 } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
997 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
998 do_sync_file_range(file, start_pos,
999 start_pos + num_written - 1,
1000 SYNC_FILE_RANGE_WRITE |
1001 SYNC_FILE_RANGE_WAIT_AFTER);
1002 #else
1003 do_sync_mapping_range(inode->i_mapping, start_pos,
1004 start_pos + num_written - 1,
1005 SYNC_FILE_RANGE_WRITE |
1006 SYNC_FILE_RANGE_WAIT_AFTER);
1007 #endif
1008 invalidate_mapping_pages(inode->i_mapping,
1009 start_pos >> PAGE_CACHE_SHIFT,
1010 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1011 }
1012 current->backing_dev_info = NULL;
1013 return num_written ? num_written : err;
1014 }
1015
1016 int btrfs_release_file(struct inode * inode, struct file * filp)
1017 {
1018 if (filp->private_data)
1019 btrfs_ioctl_trans_end(filp);
1020 return 0;
1021 }
1022
1023 static int btrfs_sync_file(struct file *file,
1024 struct dentry *dentry, int datasync)
1025 {
1026 struct inode *inode = dentry->d_inode;
1027 struct btrfs_root *root = BTRFS_I(inode)->root;
1028 int ret = 0;
1029 struct btrfs_trans_handle *trans;
1030
1031 /*
1032 * check the transaction that last modified this inode
1033 * and see if its already been committed
1034 */
1035 if (!BTRFS_I(inode)->last_trans)
1036 goto out;
1037
1038 mutex_lock(&root->fs_info->trans_mutex);
1039 if (BTRFS_I(inode)->last_trans <=
1040 root->fs_info->last_trans_committed) {
1041 BTRFS_I(inode)->last_trans = 0;
1042 mutex_unlock(&root->fs_info->trans_mutex);
1043 goto out;
1044 }
1045 mutex_unlock(&root->fs_info->trans_mutex);
1046
1047 /*
1048 * ok we haven't committed the transaction yet, lets do a commit
1049 */
1050 if (file->private_data)
1051 btrfs_ioctl_trans_end(file);
1052
1053 trans = btrfs_start_transaction(root, 1);
1054 if (!trans) {
1055 ret = -ENOMEM;
1056 goto out;
1057 }
1058 ret = btrfs_commit_transaction(trans, root);
1059 out:
1060 return ret > 0 ? EIO : ret;
1061 }
1062
1063 static struct vm_operations_struct btrfs_file_vm_ops = {
1064 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1065 .nopage = filemap_nopage,
1066 .populate = filemap_populate,
1067 #else
1068 .fault = filemap_fault,
1069 #endif
1070 .page_mkwrite = btrfs_page_mkwrite,
1071 };
1072
1073 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1074 {
1075 vma->vm_ops = &btrfs_file_vm_ops;
1076 file_accessed(filp);
1077 return 0;
1078 }
1079
1080 struct file_operations btrfs_file_operations = {
1081 .llseek = generic_file_llseek,
1082 .read = do_sync_read,
1083 .aio_read = generic_file_aio_read,
1084 .splice_read = generic_file_splice_read,
1085 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1086 .sendfile = generic_file_sendfile,
1087 #endif
1088 .write = btrfs_file_write,
1089 .mmap = btrfs_file_mmap,
1090 .open = generic_file_open,
1091 .release = btrfs_release_file,
1092 .fsync = btrfs_sync_file,
1093 .unlocked_ioctl = btrfs_ioctl,
1094 #ifdef CONFIG_COMPAT
1095 .compat_ioctl = btrfs_ioctl,
1096 #endif
1097 };
1098
kernel-based virtual machine