xref: /linux/fs/btrfs/file.c (revision e27ecdd94d81e5bc3d1f68591701db5adb342f0d)
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 "ctree.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "ioctl.h"
37 #include "print-tree.h"
38 #include "tree-log.h"
39 #include "locking.h"
40 #include "compat.h"
41 
42 
43 /* simple helper to fault in pages and copy.  This should go away
44  * and be replaced with calls into generic code.
45  */
46 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
47 					 int write_bytes,
48 					 struct page **prepared_pages,
49 					 const char __user *buf)
50 {
51 	long page_fault = 0;
52 	int i;
53 	int offset = pos & (PAGE_CACHE_SIZE - 1);
54 
55 	for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
56 		size_t count = min_t(size_t,
57 				     PAGE_CACHE_SIZE - offset, write_bytes);
58 		struct page *page = prepared_pages[i];
59 		fault_in_pages_readable(buf, count);
60 
61 		/* Copy data from userspace to the current page */
62 		kmap(page);
63 		page_fault = __copy_from_user(page_address(page) + offset,
64 					      buf, count);
65 		/* Flush processor's dcache for this page */
66 		flush_dcache_page(page);
67 		kunmap(page);
68 		buf += count;
69 		write_bytes -= count;
70 
71 		if (page_fault)
72 			break;
73 	}
74 	return page_fault ? -EFAULT : 0;
75 }
76 
77 /*
78  * unlocks pages after btrfs_file_write is done with them
79  */
80 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
81 {
82 	size_t i;
83 	for (i = 0; i < num_pages; i++) {
84 		if (!pages[i])
85 			break;
86 		/* page checked is some magic around finding pages that
87 		 * have been modified without going through btrfs_set_page_dirty
88 		 * clear it here
89 		 */
90 		ClearPageChecked(pages[i]);
91 		unlock_page(pages[i]);
92 		mark_page_accessed(pages[i]);
93 		page_cache_release(pages[i]);
94 	}
95 }
96 
97 /*
98  * after copy_from_user, pages need to be dirtied and we need to make
99  * sure holes are created between the current EOF and the start of
100  * any next extents (if required).
101  *
102  * this also makes the decision about creating an inline extent vs
103  * doing real data extents, marking pages dirty and delalloc as required.
104  */
105 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
106 				   struct btrfs_root *root,
107 				   struct file *file,
108 				   struct page **pages,
109 				   size_t num_pages,
110 				   loff_t pos,
111 				   size_t write_bytes)
112 {
113 	int err = 0;
114 	int i;
115 	struct inode *inode = fdentry(file)->d_inode;
116 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
117 	u64 hint_byte;
118 	u64 num_bytes;
119 	u64 start_pos;
120 	u64 end_of_last_block;
121 	u64 end_pos = pos + write_bytes;
122 	loff_t isize = i_size_read(inode);
123 
124 	start_pos = pos & ~((u64)root->sectorsize - 1);
125 	num_bytes = (write_bytes + pos - start_pos +
126 		    root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
127 
128 	end_of_last_block = start_pos + num_bytes - 1;
129 
130 	lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
131 	trans = btrfs_join_transaction(root, 1);
132 	if (!trans) {
133 		err = -ENOMEM;
134 		goto out_unlock;
135 	}
136 	btrfs_set_trans_block_group(trans, inode);
137 	hint_byte = 0;
138 
139 	set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
140 
141 	/* check for reserved extents on each page, we don't want
142 	 * to reset the delalloc bit on things that already have
143 	 * extents reserved.
144 	 */
145 	btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
146 	for (i = 0; i < num_pages; i++) {
147 		struct page *p = pages[i];
148 		SetPageUptodate(p);
149 		ClearPageChecked(p);
150 		set_page_dirty(p);
151 	}
152 	if (end_pos > isize) {
153 		i_size_write(inode, end_pos);
154 		btrfs_update_inode(trans, root, inode);
155 	}
156 	err = btrfs_end_transaction(trans, root);
157 out_unlock:
158 	unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
159 	return err;
160 }
161 
162 /*
163  * this drops all the extents in the cache that intersect the range
164  * [start, end].  Existing extents are split as required.
165  */
166 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
167 			    int skip_pinned)
168 {
169 	struct extent_map *em;
170 	struct extent_map *split = NULL;
171 	struct extent_map *split2 = NULL;
172 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
173 	u64 len = end - start + 1;
174 	int ret;
175 	int testend = 1;
176 	unsigned long flags;
177 	int compressed = 0;
178 
179 	WARN_ON(end < start);
180 	if (end == (u64)-1) {
181 		len = (u64)-1;
182 		testend = 0;
183 	}
184 	while (1) {
185 		if (!split)
186 			split = alloc_extent_map(GFP_NOFS);
187 		if (!split2)
188 			split2 = alloc_extent_map(GFP_NOFS);
189 
190 		spin_lock(&em_tree->lock);
191 		em = lookup_extent_mapping(em_tree, start, len);
192 		if (!em) {
193 			spin_unlock(&em_tree->lock);
194 			break;
195 		}
196 		flags = em->flags;
197 		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
198 			spin_unlock(&em_tree->lock);
199 			if (em->start <= start &&
200 			    (!testend || em->start + em->len >= start + len)) {
201 				free_extent_map(em);
202 				break;
203 			}
204 			if (start < em->start) {
205 				len = em->start - start;
206 			} else {
207 				len = start + len - (em->start + em->len);
208 				start = em->start + em->len;
209 			}
210 			free_extent_map(em);
211 			continue;
212 		}
213 		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
214 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
215 		remove_extent_mapping(em_tree, em);
216 
217 		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
218 		    em->start < start) {
219 			split->start = em->start;
220 			split->len = start - em->start;
221 			split->orig_start = em->orig_start;
222 			split->block_start = em->block_start;
223 
224 			if (compressed)
225 				split->block_len = em->block_len;
226 			else
227 				split->block_len = split->len;
228 
229 			split->bdev = em->bdev;
230 			split->flags = flags;
231 			ret = add_extent_mapping(em_tree, split);
232 			BUG_ON(ret);
233 			free_extent_map(split);
234 			split = split2;
235 			split2 = NULL;
236 		}
237 		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
238 		    testend && em->start + em->len > start + len) {
239 			u64 diff = start + len - em->start;
240 
241 			split->start = start + len;
242 			split->len = em->start + em->len - (start + len);
243 			split->bdev = em->bdev;
244 			split->flags = flags;
245 
246 			if (compressed) {
247 				split->block_len = em->block_len;
248 				split->block_start = em->block_start;
249 				split->orig_start = em->orig_start;
250 			} else {
251 				split->block_len = split->len;
252 				split->block_start = em->block_start + diff;
253 				split->orig_start = split->start;
254 			}
255 
256 			ret = add_extent_mapping(em_tree, split);
257 			BUG_ON(ret);
258 			free_extent_map(split);
259 			split = NULL;
260 		}
261 		spin_unlock(&em_tree->lock);
262 
263 		/* once for us */
264 		free_extent_map(em);
265 		/* once for the tree*/
266 		free_extent_map(em);
267 	}
268 	if (split)
269 		free_extent_map(split);
270 	if (split2)
271 		free_extent_map(split2);
272 	return 0;
273 }
274 
275 /*
276  * this is very complex, but the basic idea is to drop all extents
277  * in the range start - end.  hint_block is filled in with a block number
278  * that would be a good hint to the block allocator for this file.
279  *
280  * If an extent intersects the range but is not entirely inside the range
281  * it is either truncated or split.  Anything entirely inside the range
282  * is deleted from the tree.
283  *
284  * inline_limit is used to tell this code which offsets in the file to keep
285  * if they contain inline extents.
286  */
287 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
288 		       struct btrfs_root *root, struct inode *inode,
289 		       u64 start, u64 end, u64 locked_end,
290 		       u64 inline_limit, u64 *hint_byte)
291 {
292 	u64 extent_end = 0;
293 	u64 search_start = start;
294 	u64 ram_bytes = 0;
295 	u64 disk_bytenr = 0;
296 	u64 orig_locked_end = locked_end;
297 	u8 compression;
298 	u8 encryption;
299 	u16 other_encoding = 0;
300 	struct extent_buffer *leaf;
301 	struct btrfs_file_extent_item *extent;
302 	struct btrfs_path *path;
303 	struct btrfs_key key;
304 	struct btrfs_file_extent_item old;
305 	int keep;
306 	int slot;
307 	int bookend;
308 	int found_type = 0;
309 	int found_extent;
310 	int found_inline;
311 	int recow;
312 	int ret;
313 
314 	inline_limit = 0;
315 	btrfs_drop_extent_cache(inode, start, end - 1, 0);
316 
317 	path = btrfs_alloc_path();
318 	if (!path)
319 		return -ENOMEM;
320 	while (1) {
321 		recow = 0;
322 		btrfs_release_path(root, path);
323 		ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
324 					       search_start, -1);
325 		if (ret < 0)
326 			goto out;
327 		if (ret > 0) {
328 			if (path->slots[0] == 0) {
329 				ret = 0;
330 				goto out;
331 			}
332 			path->slots[0]--;
333 		}
334 next_slot:
335 		keep = 0;
336 		bookend = 0;
337 		found_extent = 0;
338 		found_inline = 0;
339 		compression = 0;
340 		encryption = 0;
341 		extent = NULL;
342 		leaf = path->nodes[0];
343 		slot = path->slots[0];
344 		ret = 0;
345 		btrfs_item_key_to_cpu(leaf, &key, slot);
346 		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
347 		    key.offset >= end) {
348 			goto out;
349 		}
350 		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
351 		    key.objectid != inode->i_ino) {
352 			goto out;
353 		}
354 		if (recow) {
355 			search_start = max(key.offset, start);
356 			continue;
357 		}
358 		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
359 			extent = btrfs_item_ptr(leaf, slot,
360 						struct btrfs_file_extent_item);
361 			found_type = btrfs_file_extent_type(leaf, extent);
362 			compression = btrfs_file_extent_compression(leaf,
363 								    extent);
364 			encryption = btrfs_file_extent_encryption(leaf,
365 								  extent);
366 			other_encoding = btrfs_file_extent_other_encoding(leaf,
367 								  extent);
368 			if (found_type == BTRFS_FILE_EXTENT_REG ||
369 			    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
370 				extent_end =
371 				     btrfs_file_extent_disk_bytenr(leaf,
372 								   extent);
373 				if (extent_end)
374 					*hint_byte = extent_end;
375 
376 				extent_end = key.offset +
377 				     btrfs_file_extent_num_bytes(leaf, extent);
378 				ram_bytes = btrfs_file_extent_ram_bytes(leaf,
379 								extent);
380 				found_extent = 1;
381 			} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
382 				found_inline = 1;
383 				extent_end = key.offset +
384 				     btrfs_file_extent_inline_len(leaf, extent);
385 			}
386 		} else {
387 			extent_end = search_start;
388 		}
389 
390 		/* we found nothing we can drop */
391 		if ((!found_extent && !found_inline) ||
392 		    search_start >= extent_end) {
393 			int nextret;
394 			u32 nritems;
395 			nritems = btrfs_header_nritems(leaf);
396 			if (slot >= nritems - 1) {
397 				nextret = btrfs_next_leaf(root, path);
398 				if (nextret)
399 					goto out;
400 				recow = 1;
401 			} else {
402 				path->slots[0]++;
403 			}
404 			goto next_slot;
405 		}
406 
407 		if (end <= extent_end && start >= key.offset && found_inline)
408 			*hint_byte = EXTENT_MAP_INLINE;
409 
410 		if (found_extent) {
411 			read_extent_buffer(leaf, &old, (unsigned long)extent,
412 					   sizeof(old));
413 		}
414 
415 		if (end < extent_end && end >= key.offset) {
416 			bookend = 1;
417 			if (found_inline && start <= key.offset)
418 				keep = 1;
419 		}
420 
421 		if (bookend && found_extent) {
422 			if (locked_end < extent_end) {
423 				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
424 						locked_end, extent_end - 1,
425 						GFP_NOFS);
426 				if (!ret) {
427 					btrfs_release_path(root, path);
428 					lock_extent(&BTRFS_I(inode)->io_tree,
429 						locked_end, extent_end - 1,
430 						GFP_NOFS);
431 					locked_end = extent_end;
432 					continue;
433 				}
434 				locked_end = extent_end;
435 			}
436 			disk_bytenr = le64_to_cpu(old.disk_bytenr);
437 			if (disk_bytenr != 0) {
438 				ret = btrfs_inc_extent_ref(trans, root,
439 					   disk_bytenr,
440 					   le64_to_cpu(old.disk_num_bytes), 0,
441 					   root->root_key.objectid,
442 					   key.objectid, key.offset -
443 					   le64_to_cpu(old.offset));
444 				BUG_ON(ret);
445 			}
446 		}
447 
448 		if (found_inline) {
449 			u64 mask = root->sectorsize - 1;
450 			search_start = (extent_end + mask) & ~mask;
451 		} else
452 			search_start = extent_end;
453 
454 		/* truncate existing extent */
455 		if (start > key.offset) {
456 			u64 new_num;
457 			u64 old_num;
458 			keep = 1;
459 			WARN_ON(start & (root->sectorsize - 1));
460 			if (found_extent) {
461 				new_num = start - key.offset;
462 				old_num = btrfs_file_extent_num_bytes(leaf,
463 								      extent);
464 				*hint_byte =
465 					btrfs_file_extent_disk_bytenr(leaf,
466 								      extent);
467 				if (btrfs_file_extent_disk_bytenr(leaf,
468 								  extent)) {
469 					inode_sub_bytes(inode, old_num -
470 							new_num);
471 				}
472 				btrfs_set_file_extent_num_bytes(leaf,
473 							extent, new_num);
474 				btrfs_mark_buffer_dirty(leaf);
475 			} else if (key.offset < inline_limit &&
476 				   (end > extent_end) &&
477 				   (inline_limit < extent_end)) {
478 				u32 new_size;
479 				new_size = btrfs_file_extent_calc_inline_size(
480 						   inline_limit - key.offset);
481 				inode_sub_bytes(inode, extent_end -
482 						inline_limit);
483 				btrfs_set_file_extent_ram_bytes(leaf, extent,
484 							new_size);
485 				if (!compression && !encryption) {
486 					btrfs_truncate_item(trans, root, path,
487 							    new_size, 1);
488 				}
489 			}
490 		}
491 		/* delete the entire extent */
492 		if (!keep) {
493 			if (found_inline)
494 				inode_sub_bytes(inode, extent_end -
495 						key.offset);
496 			ret = btrfs_del_item(trans, root, path);
497 			/* TODO update progress marker and return */
498 			BUG_ON(ret);
499 			extent = NULL;
500 			btrfs_release_path(root, path);
501 			/* the extent will be freed later */
502 		}
503 		if (bookend && found_inline && start <= key.offset) {
504 			u32 new_size;
505 			new_size = btrfs_file_extent_calc_inline_size(
506 						   extent_end - end);
507 			inode_sub_bytes(inode, end - key.offset);
508 			btrfs_set_file_extent_ram_bytes(leaf, extent,
509 							new_size);
510 			if (!compression && !encryption)
511 				ret = btrfs_truncate_item(trans, root, path,
512 							  new_size, 0);
513 			BUG_ON(ret);
514 		}
515 		/* create bookend, splitting the extent in two */
516 		if (bookend && found_extent) {
517 			struct btrfs_key ins;
518 			ins.objectid = inode->i_ino;
519 			ins.offset = end;
520 			btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
521 
522 			btrfs_release_path(root, path);
523 			path->leave_spinning = 1;
524 			ret = btrfs_insert_empty_item(trans, root, path, &ins,
525 						      sizeof(*extent));
526 			BUG_ON(ret);
527 
528 			leaf = path->nodes[0];
529 			extent = btrfs_item_ptr(leaf, path->slots[0],
530 						struct btrfs_file_extent_item);
531 			write_extent_buffer(leaf, &old,
532 					    (unsigned long)extent, sizeof(old));
533 
534 			btrfs_set_file_extent_compression(leaf, extent,
535 							  compression);
536 			btrfs_set_file_extent_encryption(leaf, extent,
537 							 encryption);
538 			btrfs_set_file_extent_other_encoding(leaf, extent,
539 							     other_encoding);
540 			btrfs_set_file_extent_offset(leaf, extent,
541 				    le64_to_cpu(old.offset) + end - key.offset);
542 			WARN_ON(le64_to_cpu(old.num_bytes) <
543 				(extent_end - end));
544 			btrfs_set_file_extent_num_bytes(leaf, extent,
545 							extent_end - end);
546 
547 			/*
548 			 * set the ram bytes to the size of the full extent
549 			 * before splitting.  This is a worst case flag,
550 			 * but its the best we can do because we don't know
551 			 * how splitting affects compression
552 			 */
553 			btrfs_set_file_extent_ram_bytes(leaf, extent,
554 							ram_bytes);
555 			btrfs_set_file_extent_type(leaf, extent, found_type);
556 
557 			btrfs_unlock_up_safe(path, 1);
558 			btrfs_mark_buffer_dirty(path->nodes[0]);
559 			btrfs_set_lock_blocking(path->nodes[0]);
560 
561 			path->leave_spinning = 0;
562 			btrfs_release_path(root, path);
563 			if (disk_bytenr != 0)
564 				inode_add_bytes(inode, extent_end - end);
565 		}
566 
567 		if (found_extent && !keep) {
568 			u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
569 
570 			if (old_disk_bytenr != 0) {
571 				inode_sub_bytes(inode,
572 						le64_to_cpu(old.num_bytes));
573 				ret = btrfs_free_extent(trans, root,
574 						old_disk_bytenr,
575 						le64_to_cpu(old.disk_num_bytes),
576 						0, root->root_key.objectid,
577 						key.objectid, key.offset -
578 						le64_to_cpu(old.offset));
579 				BUG_ON(ret);
580 				*hint_byte = old_disk_bytenr;
581 			}
582 		}
583 
584 		if (search_start >= end) {
585 			ret = 0;
586 			goto out;
587 		}
588 	}
589 out:
590 	btrfs_free_path(path);
591 	if (locked_end > orig_locked_end) {
592 		unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
593 			      locked_end - 1, GFP_NOFS);
594 	}
595 	return ret;
596 }
597 
598 static int extent_mergeable(struct extent_buffer *leaf, int slot,
599 			    u64 objectid, u64 bytenr, u64 *start, u64 *end)
600 {
601 	struct btrfs_file_extent_item *fi;
602 	struct btrfs_key key;
603 	u64 extent_end;
604 
605 	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
606 		return 0;
607 
608 	btrfs_item_key_to_cpu(leaf, &key, slot);
609 	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
610 		return 0;
611 
612 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
613 	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
614 	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
615 	    btrfs_file_extent_compression(leaf, fi) ||
616 	    btrfs_file_extent_encryption(leaf, fi) ||
617 	    btrfs_file_extent_other_encoding(leaf, fi))
618 		return 0;
619 
620 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
621 	if ((*start && *start != key.offset) || (*end && *end != extent_end))
622 		return 0;
623 
624 	*start = key.offset;
625 	*end = extent_end;
626 	return 1;
627 }
628 
629 /*
630  * Mark extent in the range start - end as written.
631  *
632  * This changes extent type from 'pre-allocated' to 'regular'. If only
633  * part of extent is marked as written, the extent will be split into
634  * two or three.
635  */
636 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
637 			      struct btrfs_root *root,
638 			      struct inode *inode, u64 start, u64 end)
639 {
640 	struct extent_buffer *leaf;
641 	struct btrfs_path *path;
642 	struct btrfs_file_extent_item *fi;
643 	struct btrfs_key key;
644 	u64 bytenr;
645 	u64 num_bytes;
646 	u64 extent_end;
647 	u64 orig_offset;
648 	u64 other_start;
649 	u64 other_end;
650 	u64 split = start;
651 	u64 locked_end = end;
652 	int extent_type;
653 	int split_end = 1;
654 	int ret;
655 
656 	btrfs_drop_extent_cache(inode, start, end - 1, 0);
657 
658 	path = btrfs_alloc_path();
659 	BUG_ON(!path);
660 again:
661 	key.objectid = inode->i_ino;
662 	key.type = BTRFS_EXTENT_DATA_KEY;
663 	if (split == start)
664 		key.offset = split;
665 	else
666 		key.offset = split - 1;
667 
668 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
669 	if (ret > 0 && path->slots[0] > 0)
670 		path->slots[0]--;
671 
672 	leaf = path->nodes[0];
673 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
674 	BUG_ON(key.objectid != inode->i_ino ||
675 	       key.type != BTRFS_EXTENT_DATA_KEY);
676 	fi = btrfs_item_ptr(leaf, path->slots[0],
677 			    struct btrfs_file_extent_item);
678 	extent_type = btrfs_file_extent_type(leaf, fi);
679 	BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
680 	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
681 	BUG_ON(key.offset > start || extent_end < end);
682 
683 	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
684 	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
685 	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
686 
687 	if (key.offset == start)
688 		split = end;
689 
690 	if (key.offset == start && extent_end == end) {
691 		int del_nr = 0;
692 		int del_slot = 0;
693 		other_start = end;
694 		other_end = 0;
695 		if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
696 				     bytenr, &other_start, &other_end)) {
697 			extent_end = other_end;
698 			del_slot = path->slots[0] + 1;
699 			del_nr++;
700 			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
701 						0, root->root_key.objectid,
702 						inode->i_ino, orig_offset);
703 			BUG_ON(ret);
704 		}
705 		other_start = 0;
706 		other_end = start;
707 		if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
708 				     bytenr, &other_start, &other_end)) {
709 			key.offset = other_start;
710 			del_slot = path->slots[0];
711 			del_nr++;
712 			ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
713 						0, root->root_key.objectid,
714 						inode->i_ino, orig_offset);
715 			BUG_ON(ret);
716 		}
717 		split_end = 0;
718 		if (del_nr == 0) {
719 			btrfs_set_file_extent_type(leaf, fi,
720 						   BTRFS_FILE_EXTENT_REG);
721 			goto done;
722 		}
723 
724 		fi = btrfs_item_ptr(leaf, del_slot - 1,
725 				    struct btrfs_file_extent_item);
726 		btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
727 		btrfs_set_file_extent_num_bytes(leaf, fi,
728 						extent_end - key.offset);
729 		btrfs_mark_buffer_dirty(leaf);
730 
731 		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
732 		BUG_ON(ret);
733 		goto release;
734 	} else if (split == start) {
735 		if (locked_end < extent_end) {
736 			ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
737 					locked_end, extent_end - 1, GFP_NOFS);
738 			if (!ret) {
739 				btrfs_release_path(root, path);
740 				lock_extent(&BTRFS_I(inode)->io_tree,
741 					locked_end, extent_end - 1, GFP_NOFS);
742 				locked_end = extent_end;
743 				goto again;
744 			}
745 			locked_end = extent_end;
746 		}
747 		btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
748 	} else  {
749 		BUG_ON(key.offset != start);
750 		key.offset = split;
751 		btrfs_set_file_extent_offset(leaf, fi, key.offset -
752 					     orig_offset);
753 		btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
754 		btrfs_set_item_key_safe(trans, root, path, &key);
755 		extent_end = split;
756 	}
757 
758 	if (extent_end == end) {
759 		split_end = 0;
760 		extent_type = BTRFS_FILE_EXTENT_REG;
761 	}
762 	if (extent_end == end && split == start) {
763 		other_start = end;
764 		other_end = 0;
765 		if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
766 				     bytenr, &other_start, &other_end)) {
767 			path->slots[0]++;
768 			fi = btrfs_item_ptr(leaf, path->slots[0],
769 					    struct btrfs_file_extent_item);
770 			key.offset = split;
771 			btrfs_set_item_key_safe(trans, root, path, &key);
772 			btrfs_set_file_extent_offset(leaf, fi, key.offset -
773 						     orig_offset);
774 			btrfs_set_file_extent_num_bytes(leaf, fi,
775 							other_end - split);
776 			goto done;
777 		}
778 	}
779 	if (extent_end == end && split == end) {
780 		other_start = 0;
781 		other_end = start;
782 		if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
783 				     bytenr, &other_start, &other_end)) {
784 			path->slots[0]--;
785 			fi = btrfs_item_ptr(leaf, path->slots[0],
786 					    struct btrfs_file_extent_item);
787 			btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
788 							other_start);
789 			goto done;
790 		}
791 	}
792 
793 	btrfs_mark_buffer_dirty(leaf);
794 
795 	ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
796 				   root->root_key.objectid,
797 				   inode->i_ino, orig_offset);
798 	BUG_ON(ret);
799 	btrfs_release_path(root, path);
800 
801 	key.offset = start;
802 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
803 	BUG_ON(ret);
804 
805 	leaf = path->nodes[0];
806 	fi = btrfs_item_ptr(leaf, path->slots[0],
807 			    struct btrfs_file_extent_item);
808 	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
809 	btrfs_set_file_extent_type(leaf, fi, extent_type);
810 	btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
811 	btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
812 	btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
813 	btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
814 	btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
815 	btrfs_set_file_extent_compression(leaf, fi, 0);
816 	btrfs_set_file_extent_encryption(leaf, fi, 0);
817 	btrfs_set_file_extent_other_encoding(leaf, fi, 0);
818 done:
819 	btrfs_mark_buffer_dirty(leaf);
820 
821 release:
822 	btrfs_release_path(root, path);
823 	if (split_end && split == start) {
824 		split = end;
825 		goto again;
826 	}
827 	if (locked_end > end) {
828 		unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
829 			      GFP_NOFS);
830 	}
831 	btrfs_free_path(path);
832 	return 0;
833 }
834 
835 /*
836  * this gets pages into the page cache and locks them down, it also properly
837  * waits for data=ordered extents to finish before allowing the pages to be
838  * modified.
839  */
840 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
841 			 struct page **pages, size_t num_pages,
842 			 loff_t pos, unsigned long first_index,
843 			 unsigned long last_index, size_t write_bytes)
844 {
845 	int i;
846 	unsigned long index = pos >> PAGE_CACHE_SHIFT;
847 	struct inode *inode = fdentry(file)->d_inode;
848 	int err = 0;
849 	u64 start_pos;
850 	u64 last_pos;
851 
852 	start_pos = pos & ~((u64)root->sectorsize - 1);
853 	last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
854 
855 	if (start_pos > inode->i_size) {
856 		err = btrfs_cont_expand(inode, start_pos);
857 		if (err)
858 			return err;
859 	}
860 
861 	memset(pages, 0, num_pages * sizeof(struct page *));
862 again:
863 	for (i = 0; i < num_pages; i++) {
864 		pages[i] = grab_cache_page(inode->i_mapping, index + i);
865 		if (!pages[i]) {
866 			err = -ENOMEM;
867 			BUG_ON(1);
868 		}
869 		wait_on_page_writeback(pages[i]);
870 	}
871 	if (start_pos < inode->i_size) {
872 		struct btrfs_ordered_extent *ordered;
873 		lock_extent(&BTRFS_I(inode)->io_tree,
874 			    start_pos, last_pos - 1, GFP_NOFS);
875 		ordered = btrfs_lookup_first_ordered_extent(inode,
876 							    last_pos - 1);
877 		if (ordered &&
878 		    ordered->file_offset + ordered->len > start_pos &&
879 		    ordered->file_offset < last_pos) {
880 			btrfs_put_ordered_extent(ordered);
881 			unlock_extent(&BTRFS_I(inode)->io_tree,
882 				      start_pos, last_pos - 1, GFP_NOFS);
883 			for (i = 0; i < num_pages; i++) {
884 				unlock_page(pages[i]);
885 				page_cache_release(pages[i]);
886 			}
887 			btrfs_wait_ordered_range(inode, start_pos,
888 						 last_pos - start_pos);
889 			goto again;
890 		}
891 		if (ordered)
892 			btrfs_put_ordered_extent(ordered);
893 
894 		clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
895 				  last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
896 				  GFP_NOFS);
897 		unlock_extent(&BTRFS_I(inode)->io_tree,
898 			      start_pos, last_pos - 1, GFP_NOFS);
899 	}
900 	for (i = 0; i < num_pages; i++) {
901 		clear_page_dirty_for_io(pages[i]);
902 		set_page_extent_mapped(pages[i]);
903 		WARN_ON(!PageLocked(pages[i]));
904 	}
905 	return 0;
906 }
907 
908 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
909 				size_t count, loff_t *ppos)
910 {
911 	loff_t pos;
912 	loff_t start_pos;
913 	ssize_t num_written = 0;
914 	ssize_t err = 0;
915 	int ret = 0;
916 	struct inode *inode = fdentry(file)->d_inode;
917 	struct btrfs_root *root = BTRFS_I(inode)->root;
918 	struct page **pages = NULL;
919 	int nrptrs;
920 	struct page *pinned[2];
921 	unsigned long first_index;
922 	unsigned long last_index;
923 	int will_write;
924 
925 	will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
926 		      (file->f_flags & O_DIRECT));
927 
928 	nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
929 		     PAGE_CACHE_SIZE / (sizeof(struct page *)));
930 	pinned[0] = NULL;
931 	pinned[1] = NULL;
932 
933 	pos = *ppos;
934 	start_pos = pos;
935 
936 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
937 	current->backing_dev_info = inode->i_mapping->backing_dev_info;
938 	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
939 	if (err)
940 		goto out_nolock;
941 	if (count == 0)
942 		goto out_nolock;
943 
944 	err = file_remove_suid(file);
945 	if (err)
946 		goto out_nolock;
947 	file_update_time(file);
948 
949 	pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
950 
951 	mutex_lock(&inode->i_mutex);
952 	BTRFS_I(inode)->sequence++;
953 	first_index = pos >> PAGE_CACHE_SHIFT;
954 	last_index = (pos + count) >> PAGE_CACHE_SHIFT;
955 
956 	/*
957 	 * there are lots of better ways to do this, but this code
958 	 * makes sure the first and last page in the file range are
959 	 * up to date and ready for cow
960 	 */
961 	if ((pos & (PAGE_CACHE_SIZE - 1))) {
962 		pinned[0] = grab_cache_page(inode->i_mapping, first_index);
963 		if (!PageUptodate(pinned[0])) {
964 			ret = btrfs_readpage(NULL, pinned[0]);
965 			BUG_ON(ret);
966 			wait_on_page_locked(pinned[0]);
967 		} else {
968 			unlock_page(pinned[0]);
969 		}
970 	}
971 	if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
972 		pinned[1] = grab_cache_page(inode->i_mapping, last_index);
973 		if (!PageUptodate(pinned[1])) {
974 			ret = btrfs_readpage(NULL, pinned[1]);
975 			BUG_ON(ret);
976 			wait_on_page_locked(pinned[1]);
977 		} else {
978 			unlock_page(pinned[1]);
979 		}
980 	}
981 
982 	while (count > 0) {
983 		size_t offset = pos & (PAGE_CACHE_SIZE - 1);
984 		size_t write_bytes = min(count, nrptrs *
985 					(size_t)PAGE_CACHE_SIZE -
986 					 offset);
987 		size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
988 					PAGE_CACHE_SHIFT;
989 
990 		WARN_ON(num_pages > nrptrs);
991 		memset(pages, 0, sizeof(struct page *) * nrptrs);
992 
993 		ret = btrfs_check_data_free_space(root, inode, write_bytes);
994 		if (ret)
995 			goto out;
996 
997 		ret = prepare_pages(root, file, pages, num_pages,
998 				    pos, first_index, last_index,
999 				    write_bytes);
1000 		if (ret) {
1001 			btrfs_free_reserved_data_space(root, inode,
1002 						       write_bytes);
1003 			goto out;
1004 		}
1005 
1006 		ret = btrfs_copy_from_user(pos, num_pages,
1007 					   write_bytes, pages, buf);
1008 		if (ret) {
1009 			btrfs_free_reserved_data_space(root, inode,
1010 						       write_bytes);
1011 			btrfs_drop_pages(pages, num_pages);
1012 			goto out;
1013 		}
1014 
1015 		ret = dirty_and_release_pages(NULL, root, file, pages,
1016 					      num_pages, pos, write_bytes);
1017 		btrfs_drop_pages(pages, num_pages);
1018 		if (ret) {
1019 			btrfs_free_reserved_data_space(root, inode,
1020 						       write_bytes);
1021 			goto out;
1022 		}
1023 
1024 		if (will_write) {
1025 			btrfs_fdatawrite_range(inode->i_mapping, pos,
1026 					       pos + write_bytes - 1,
1027 					       WB_SYNC_ALL);
1028 		} else {
1029 			balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1030 							   num_pages);
1031 			if (num_pages <
1032 			    (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1033 				btrfs_btree_balance_dirty(root, 1);
1034 			btrfs_throttle(root);
1035 		}
1036 
1037 		buf += write_bytes;
1038 		count -= write_bytes;
1039 		pos += write_bytes;
1040 		num_written += write_bytes;
1041 
1042 		cond_resched();
1043 	}
1044 out:
1045 	mutex_unlock(&inode->i_mutex);
1046 	if (ret)
1047 		err = ret;
1048 
1049 out_nolock:
1050 	kfree(pages);
1051 	if (pinned[0])
1052 		page_cache_release(pinned[0]);
1053 	if (pinned[1])
1054 		page_cache_release(pinned[1]);
1055 	*ppos = pos;
1056 
1057 	/*
1058 	 * we want to make sure fsync finds this change
1059 	 * but we haven't joined a transaction running right now.
1060 	 *
1061 	 * Later on, someone is sure to update the inode and get the
1062 	 * real transid recorded.
1063 	 *
1064 	 * We set last_trans now to the fs_info generation + 1,
1065 	 * this will either be one more than the running transaction
1066 	 * or the generation used for the next transaction if there isn't
1067 	 * one running right now.
1068 	 */
1069 	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1070 
1071 	if (num_written > 0 && will_write) {
1072 		struct btrfs_trans_handle *trans;
1073 
1074 		err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1075 		if (err)
1076 			num_written = err;
1077 
1078 		if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1079 			trans = btrfs_start_transaction(root, 1);
1080 			ret = btrfs_log_dentry_safe(trans, root,
1081 						    file->f_dentry);
1082 			if (ret == 0) {
1083 				ret = btrfs_sync_log(trans, root);
1084 				if (ret == 0)
1085 					btrfs_end_transaction(trans, root);
1086 				else
1087 					btrfs_commit_transaction(trans, root);
1088 			} else {
1089 				btrfs_commit_transaction(trans, root);
1090 			}
1091 		}
1092 		if (file->f_flags & O_DIRECT) {
1093 			invalidate_mapping_pages(inode->i_mapping,
1094 			      start_pos >> PAGE_CACHE_SHIFT,
1095 			     (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1096 		}
1097 	}
1098 	current->backing_dev_info = NULL;
1099 	return num_written ? num_written : err;
1100 }
1101 
1102 int btrfs_release_file(struct inode *inode, struct file *filp)
1103 {
1104 	/*
1105 	 * ordered_data_close is set by settattr when we are about to truncate
1106 	 * a file from a non-zero size to a zero size.  This tries to
1107 	 * flush down new bytes that may have been written if the
1108 	 * application were using truncate to replace a file in place.
1109 	 */
1110 	if (BTRFS_I(inode)->ordered_data_close) {
1111 		BTRFS_I(inode)->ordered_data_close = 0;
1112 		btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1113 		if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1114 			filemap_flush(inode->i_mapping);
1115 	}
1116 	if (filp->private_data)
1117 		btrfs_ioctl_trans_end(filp);
1118 	return 0;
1119 }
1120 
1121 /*
1122  * fsync call for both files and directories.  This logs the inode into
1123  * the tree log instead of forcing full commits whenever possible.
1124  *
1125  * It needs to call filemap_fdatawait so that all ordered extent updates are
1126  * in the metadata btree are up to date for copying to the log.
1127  *
1128  * It drops the inode mutex before doing the tree log commit.  This is an
1129  * important optimization for directories because holding the mutex prevents
1130  * new operations on the dir while we write to disk.
1131  */
1132 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1133 {
1134 	struct inode *inode = dentry->d_inode;
1135 	struct btrfs_root *root = BTRFS_I(inode)->root;
1136 	int ret = 0;
1137 	struct btrfs_trans_handle *trans;
1138 
1139 	/*
1140 	 * check the transaction that last modified this inode
1141 	 * and see if its already been committed
1142 	 */
1143 	if (!BTRFS_I(inode)->last_trans)
1144 		goto out;
1145 
1146 	mutex_lock(&root->fs_info->trans_mutex);
1147 	if (BTRFS_I(inode)->last_trans <=
1148 	    root->fs_info->last_trans_committed) {
1149 		BTRFS_I(inode)->last_trans = 0;
1150 		mutex_unlock(&root->fs_info->trans_mutex);
1151 		goto out;
1152 	}
1153 	mutex_unlock(&root->fs_info->trans_mutex);
1154 
1155 	root->log_batch++;
1156 	filemap_fdatawrite(inode->i_mapping);
1157 	btrfs_wait_ordered_range(inode, 0, (u64)-1);
1158 	root->log_batch++;
1159 
1160 	if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1161 		goto out;
1162 	/*
1163 	 * ok we haven't committed the transaction yet, lets do a commit
1164 	 */
1165 	if (file && file->private_data)
1166 		btrfs_ioctl_trans_end(file);
1167 
1168 	trans = btrfs_start_transaction(root, 1);
1169 	if (!trans) {
1170 		ret = -ENOMEM;
1171 		goto out;
1172 	}
1173 
1174 	ret = btrfs_log_dentry_safe(trans, root, dentry);
1175 	if (ret < 0)
1176 		goto out;
1177 
1178 	/* we've logged all the items and now have a consistent
1179 	 * version of the file in the log.  It is possible that
1180 	 * someone will come in and modify the file, but that's
1181 	 * fine because the log is consistent on disk, and we
1182 	 * have references to all of the file's extents
1183 	 *
1184 	 * It is possible that someone will come in and log the
1185 	 * file again, but that will end up using the synchronization
1186 	 * inside btrfs_sync_log to keep things safe.
1187 	 */
1188 	mutex_unlock(&dentry->d_inode->i_mutex);
1189 
1190 	if (ret > 0) {
1191 		ret = btrfs_commit_transaction(trans, root);
1192 	} else {
1193 		ret = btrfs_sync_log(trans, root);
1194 		if (ret == 0)
1195 			ret = btrfs_end_transaction(trans, root);
1196 		else
1197 			ret = btrfs_commit_transaction(trans, root);
1198 	}
1199 	mutex_lock(&dentry->d_inode->i_mutex);
1200 out:
1201 	return ret > 0 ? EIO : ret;
1202 }
1203 
1204 static struct vm_operations_struct btrfs_file_vm_ops = {
1205 	.fault		= filemap_fault,
1206 	.page_mkwrite	= btrfs_page_mkwrite,
1207 };
1208 
1209 static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
1210 {
1211 	vma->vm_ops = &btrfs_file_vm_ops;
1212 	file_accessed(filp);
1213 	return 0;
1214 }
1215 
1216 struct file_operations btrfs_file_operations = {
1217 	.llseek		= generic_file_llseek,
1218 	.read		= do_sync_read,
1219 	.aio_read       = generic_file_aio_read,
1220 	.splice_read	= generic_file_splice_read,
1221 	.write		= btrfs_file_write,
1222 	.mmap		= btrfs_file_mmap,
1223 	.open		= generic_file_open,
1224 	.release	= btrfs_release_file,
1225 	.fsync		= btrfs_sync_file,
1226 	.unlocked_ioctl	= btrfs_ioctl,
1227 #ifdef CONFIG_COMPAT
1228 	.compat_ioctl	= btrfs_ioctl,
1229 #endif
1230 };
1231