Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
v3.1
  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/slab.h>
 20#include <linux/blkdev.h>
 21#include <linux/writeback.h>
 22#include <linux/pagevec.h>
 23#include "ctree.h"
 24#include "transaction.h"
 25#include "btrfs_inode.h"
 26#include "extent_io.h"
 
 
 
 27
 28static u64 entry_end(struct btrfs_ordered_extent *entry)
 29{
 30	if (entry->file_offset + entry->len < entry->file_offset)
 31		return (u64)-1;
 32	return entry->file_offset + entry->len;
 33}
 34
 35/* returns NULL if the insertion worked, or it returns the node it did find
 36 * in the tree
 37 */
 38static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
 39				   struct rb_node *node)
 40{
 41	struct rb_node **p = &root->rb_node;
 42	struct rb_node *parent = NULL;
 43	struct btrfs_ordered_extent *entry;
 44
 45	while (*p) {
 46		parent = *p;
 47		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
 48
 49		if (file_offset < entry->file_offset)
 50			p = &(*p)->rb_left;
 51		else if (file_offset >= entry_end(entry))
 52			p = &(*p)->rb_right;
 53		else
 54			return parent;
 55	}
 56
 57	rb_link_node(node, parent, p);
 58	rb_insert_color(node, root);
 59	return NULL;
 60}
 61
 
 
 
 
 
 
 
 
 62/*
 63 * look for a given offset in the tree, and if it can't be found return the
 64 * first lesser offset
 65 */
 66static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
 67				     struct rb_node **prev_ret)
 68{
 69	struct rb_node *n = root->rb_node;
 70	struct rb_node *prev = NULL;
 71	struct rb_node *test;
 72	struct btrfs_ordered_extent *entry;
 73	struct btrfs_ordered_extent *prev_entry = NULL;
 74
 75	while (n) {
 76		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
 77		prev = n;
 78		prev_entry = entry;
 79
 80		if (file_offset < entry->file_offset)
 81			n = n->rb_left;
 82		else if (file_offset >= entry_end(entry))
 83			n = n->rb_right;
 84		else
 85			return n;
 86	}
 87	if (!prev_ret)
 88		return NULL;
 89
 90	while (prev && file_offset >= entry_end(prev_entry)) {
 91		test = rb_next(prev);
 92		if (!test)
 93			break;
 94		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 95				      rb_node);
 96		if (file_offset < entry_end(prev_entry))
 97			break;
 98
 99		prev = test;
100	}
101	if (prev)
102		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
103				      rb_node);
104	while (prev && file_offset < entry_end(prev_entry)) {
105		test = rb_prev(prev);
106		if (!test)
107			break;
108		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
109				      rb_node);
110		prev = test;
111	}
112	*prev_ret = prev;
113	return NULL;
114}
115
116/*
117 * helper to check if a given offset is inside a given entry
118 */
119static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
120{
121	if (file_offset < entry->file_offset ||
122	    entry->file_offset + entry->len <= file_offset)
123		return 0;
124	return 1;
125}
126
127static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
128			  u64 len)
129{
130	if (file_offset + len <= entry->file_offset ||
131	    entry->file_offset + entry->len <= file_offset)
132		return 0;
133	return 1;
134}
135
136/*
137 * look find the first ordered struct that has this offset, otherwise
138 * the first one less than this offset
139 */
140static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
141					  u64 file_offset)
142{
143	struct rb_root *root = &tree->tree;
144	struct rb_node *prev = NULL;
145	struct rb_node *ret;
146	struct btrfs_ordered_extent *entry;
147
148	if (tree->last) {
149		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
150				 rb_node);
151		if (offset_in_entry(entry, file_offset))
152			return tree->last;
153	}
154	ret = __tree_search(root, file_offset, &prev);
155	if (!ret)
156		ret = prev;
157	if (ret)
158		tree->last = ret;
159	return ret;
160}
161
162/* allocate and add a new ordered_extent into the per-inode tree.
163 * file_offset is the logical offset in the file
164 *
165 * start is the disk block number of an extent already reserved in the
166 * extent allocation tree
167 *
168 * len is the length of the extent
169 *
170 * The tree is given a single reference on the ordered extent that was
171 * inserted.
172 */
173static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
174				      u64 start, u64 len, u64 disk_len,
175				      int type, int dio, int compress_type)
176{
 
177	struct btrfs_ordered_inode_tree *tree;
178	struct rb_node *node;
179	struct btrfs_ordered_extent *entry;
180
181	tree = &BTRFS_I(inode)->ordered_tree;
182	entry = kzalloc(sizeof(*entry), GFP_NOFS);
183	if (!entry)
184		return -ENOMEM;
185
186	entry->file_offset = file_offset;
187	entry->start = start;
188	entry->len = len;
 
 
 
189	entry->disk_len = disk_len;
190	entry->bytes_left = len;
191	entry->inode = inode;
192	entry->compress_type = compress_type;
 
193	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
194		set_bit(type, &entry->flags);
195
196	if (dio)
197		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
198
199	/* one ref for the tree */
200	atomic_set(&entry->refs, 1);
201	init_waitqueue_head(&entry->wait);
202	INIT_LIST_HEAD(&entry->list);
203	INIT_LIST_HEAD(&entry->root_extent_list);
 
 
 
204
205	trace_btrfs_ordered_extent_add(inode, entry);
206
207	spin_lock(&tree->lock);
208	node = tree_insert(&tree->tree, file_offset,
209			   &entry->rb_node);
210	BUG_ON(node);
211	spin_unlock(&tree->lock);
 
212
213	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
214	list_add_tail(&entry->root_extent_list,
215		      &BTRFS_I(inode)->root->fs_info->ordered_extents);
216	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
 
 
 
 
 
 
 
 
217
218	BUG_ON(node);
219	return 0;
220}
221
222int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
223			     u64 start, u64 len, u64 disk_len, int type)
224{
225	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
226					  disk_len, type, 0,
227					  BTRFS_COMPRESS_NONE);
228}
229
230int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
231				 u64 start, u64 len, u64 disk_len, int type)
232{
233	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
234					  disk_len, type, 1,
235					  BTRFS_COMPRESS_NONE);
236}
237
238int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
239				      u64 start, u64 len, u64 disk_len,
240				      int type, int compress_type)
241{
242	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
243					  disk_len, type, 0,
244					  compress_type);
245}
246
247/*
248 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
249 * when an ordered extent is finished.  If the list covers more than one
250 * ordered extent, it is split across multiples.
251 */
252int btrfs_add_ordered_sum(struct inode *inode,
253			  struct btrfs_ordered_extent *entry,
254			  struct btrfs_ordered_sum *sum)
255{
256	struct btrfs_ordered_inode_tree *tree;
257
258	tree = &BTRFS_I(inode)->ordered_tree;
259	spin_lock(&tree->lock);
260	list_add_tail(&sum->list, &entry->list);
261	spin_unlock(&tree->lock);
262	return 0;
 
 
 
263}
264
265/*
266 * this is used to account for finished IO across a given range
267 * of the file.  The IO may span ordered extents.  If
268 * a given ordered_extent is completely done, 1 is returned, otherwise
269 * 0.
270 *
271 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
272 * to make sure this function only returns 1 once for a given ordered extent.
273 *
274 * file_offset is updated to one byte past the range that is recorded as
275 * complete.  This allows you to walk forward in the file.
276 */
277int btrfs_dec_test_first_ordered_pending(struct inode *inode,
278				   struct btrfs_ordered_extent **cached,
279				   u64 *file_offset, u64 io_size)
280{
281	struct btrfs_ordered_inode_tree *tree;
282	struct rb_node *node;
283	struct btrfs_ordered_extent *entry = NULL;
284	int ret;
 
285	u64 dec_end;
286	u64 dec_start;
287	u64 to_dec;
288
289	tree = &BTRFS_I(inode)->ordered_tree;
290	spin_lock(&tree->lock);
291	node = tree_search(tree, *file_offset);
292	if (!node) {
293		ret = 1;
294		goto out;
295	}
296
297	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
298	if (!offset_in_entry(entry, *file_offset)) {
299		ret = 1;
300		goto out;
301	}
302
303	dec_start = max(*file_offset, entry->file_offset);
304	dec_end = min(*file_offset + io_size, entry->file_offset +
305		      entry->len);
306	*file_offset = dec_end;
307	if (dec_start > dec_end) {
308		printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
309		       (unsigned long long)dec_start,
310		       (unsigned long long)dec_end);
311	}
312	to_dec = dec_end - dec_start;
313	if (to_dec > entry->bytes_left) {
314		printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
315		       (unsigned long long)entry->bytes_left,
316		       (unsigned long long)to_dec);
317	}
318	entry->bytes_left -= to_dec;
319	if (entry->bytes_left == 0)
 
 
 
320		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
321	else
 
 
322		ret = 1;
 
323out:
324	if (!ret && cached && entry) {
325		*cached = entry;
326		atomic_inc(&entry->refs);
327	}
328	spin_unlock(&tree->lock);
329	return ret == 0;
330}
331
332/*
333 * this is used to account for finished IO across a given range
334 * of the file.  The IO should not span ordered extents.  If
335 * a given ordered_extent is completely done, 1 is returned, otherwise
336 * 0.
337 *
338 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
339 * to make sure this function only returns 1 once for a given ordered extent.
340 */
341int btrfs_dec_test_ordered_pending(struct inode *inode,
342				   struct btrfs_ordered_extent **cached,
343				   u64 file_offset, u64 io_size)
344{
345	struct btrfs_ordered_inode_tree *tree;
346	struct rb_node *node;
347	struct btrfs_ordered_extent *entry = NULL;
 
348	int ret;
349
350	tree = &BTRFS_I(inode)->ordered_tree;
351	spin_lock(&tree->lock);
 
 
 
 
 
352	node = tree_search(tree, file_offset);
353	if (!node) {
354		ret = 1;
355		goto out;
356	}
357
358	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 
359	if (!offset_in_entry(entry, file_offset)) {
360		ret = 1;
361		goto out;
362	}
363
364	if (io_size > entry->bytes_left) {
365		printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
366		       (unsigned long long)entry->bytes_left,
367		       (unsigned long long)io_size);
368	}
369	entry->bytes_left -= io_size;
370	if (entry->bytes_left == 0)
 
 
 
371		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
372	else
 
 
373		ret = 1;
 
374out:
375	if (!ret && cached && entry) {
376		*cached = entry;
377		atomic_inc(&entry->refs);
378	}
379	spin_unlock(&tree->lock);
380	return ret == 0;
381}
382
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
383/*
384 * used to drop a reference on an ordered extent.  This will free
385 * the extent if the last reference is dropped
386 */
387int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
388{
389	struct list_head *cur;
390	struct btrfs_ordered_sum *sum;
391
392	trace_btrfs_ordered_extent_put(entry->inode, entry);
393
394	if (atomic_dec_and_test(&entry->refs)) {
 
 
395		while (!list_empty(&entry->list)) {
396			cur = entry->list.next;
397			sum = list_entry(cur, struct btrfs_ordered_sum, list);
398			list_del(&sum->list);
399			kfree(sum);
400		}
401		kfree(entry);
402	}
403	return 0;
404}
405
406/*
407 * remove an ordered extent from the tree.  No references are dropped
408 * and you must wake_up entry->wait.  You must hold the tree lock
409 * while you call this function.
410 */
411static int __btrfs_remove_ordered_extent(struct inode *inode,
412				struct btrfs_ordered_extent *entry)
413{
414	struct btrfs_ordered_inode_tree *tree;
415	struct btrfs_root *root = BTRFS_I(inode)->root;
416	struct rb_node *node;
417
418	tree = &BTRFS_I(inode)->ordered_tree;
 
419	node = &entry->rb_node;
420	rb_erase(node, &tree->tree);
421	tree->last = NULL;
 
422	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 
423
424	spin_lock(&root->fs_info->ordered_extent_lock);
425	list_del_init(&entry->root_extent_list);
 
426
427	trace_btrfs_ordered_extent_remove(inode, entry);
428
429	/*
430	 * we have no more ordered extents for this inode and
431	 * no dirty pages.  We can safely remove it from the
432	 * list of ordered extents
433	 */
434	if (RB_EMPTY_ROOT(&tree->tree) &&
435	    !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 
436		list_del_init(&BTRFS_I(inode)->ordered_operations);
 
437	}
438	spin_unlock(&root->fs_info->ordered_extent_lock);
439
440	return 0;
 
 
 
 
 
 
 
441}
442
443/*
444 * remove an ordered extent from the tree.  No references are dropped
445 * but any waiters are woken.
446 */
447int btrfs_remove_ordered_extent(struct inode *inode,
448				struct btrfs_ordered_extent *entry)
449{
450	struct btrfs_ordered_inode_tree *tree;
451	int ret;
452
453	tree = &BTRFS_I(inode)->ordered_tree;
454	spin_lock(&tree->lock);
455	ret = __btrfs_remove_ordered_extent(inode, entry);
456	spin_unlock(&tree->lock);
457	wake_up(&entry->wait);
458
459	return ret;
 
 
460}
461
462/*
463 * wait for all the ordered extents in a root.  This is done when balancing
464 * space between drives.
465 */
466int btrfs_wait_ordered_extents(struct btrfs_root *root,
467			       int nocow_only, int delay_iput)
468{
469	struct list_head splice;
470	struct list_head *cur;
471	struct btrfs_ordered_extent *ordered;
472	struct inode *inode;
473
474	INIT_LIST_HEAD(&splice);
 
475
476	spin_lock(&root->fs_info->ordered_extent_lock);
477	list_splice_init(&root->fs_info->ordered_extents, &splice);
478	while (!list_empty(&splice)) {
479		cur = splice.next;
480		ordered = list_entry(cur, struct btrfs_ordered_extent,
481				     root_extent_list);
482		if (nocow_only &&
483		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
484		    !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
485			list_move(&ordered->root_extent_list,
486				  &root->fs_info->ordered_extents);
487			cond_resched_lock(&root->fs_info->ordered_extent_lock);
488			continue;
489		}
490
491		list_del_init(&ordered->root_extent_list);
492		atomic_inc(&ordered->refs);
 
493
494		/*
495		 * the inode may be getting freed (in sys_unlink path).
496		 */
497		inode = igrab(ordered->inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
498
499		spin_unlock(&root->fs_info->ordered_extent_lock);
 
500
501		if (inode) {
502			btrfs_start_ordered_extent(inode, ordered, 1);
503			btrfs_put_ordered_extent(ordered);
504			if (delay_iput)
505				btrfs_add_delayed_iput(inode);
506			else
507				iput(inode);
508		} else {
509			btrfs_put_ordered_extent(ordered);
510		}
511
512		spin_lock(&root->fs_info->ordered_extent_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
513	}
514	spin_unlock(&root->fs_info->ordered_extent_lock);
515	return 0;
 
516}
517
518/*
519 * this is used during transaction commit to write all the inodes
520 * added to the ordered operation list.  These files must be fully on
521 * disk before the transaction commits.
522 *
523 * we have two modes here, one is to just start the IO via filemap_flush
524 * and the other is to wait for all the io.  When we wait, we have an
525 * extra check to make sure the ordered operation list really is empty
526 * before we return
527 */
528int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
 
529{
530	struct btrfs_inode *btrfs_inode;
531	struct inode *inode;
 
532	struct list_head splice;
 
 
 
533
534	INIT_LIST_HEAD(&splice);
 
535
536	mutex_lock(&root->fs_info->ordered_operations_mutex);
537	spin_lock(&root->fs_info->ordered_extent_lock);
538again:
539	list_splice_init(&root->fs_info->ordered_operations, &splice);
540
541	while (!list_empty(&splice)) {
542		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
543				   ordered_operations);
544
545		inode = &btrfs_inode->vfs_inode;
546
547		list_del_init(&btrfs_inode->ordered_operations);
548
549		/*
550		 * the inode may be getting freed (in sys_unlink path).
551		 */
552		inode = igrab(inode);
 
 
553
554		if (!wait && inode) {
555			list_add_tail(&BTRFS_I(inode)->ordered_operations,
556			      &root->fs_info->ordered_operations);
557		}
558		spin_unlock(&root->fs_info->ordered_extent_lock);
559
560		if (inode) {
561			if (wait)
562				btrfs_wait_ordered_range(inode, 0, (u64)-1);
563			else
564				filemap_flush(inode->i_mapping);
565			btrfs_add_delayed_iput(inode);
 
 
 
 
 
566		}
 
 
 
567
568		cond_resched();
569		spin_lock(&root->fs_info->ordered_extent_lock);
570	}
571	if (wait && !list_empty(&root->fs_info->ordered_operations))
572		goto again;
573
574	spin_unlock(&root->fs_info->ordered_extent_lock);
575	mutex_unlock(&root->fs_info->ordered_operations_mutex);
576
577	return 0;
 
578}
579
580/*
581 * Used to start IO or wait for a given ordered extent to finish.
582 *
583 * If wait is one, this effectively waits on page writeback for all the pages
584 * in the extent, and it waits on the io completion code to insert
585 * metadata into the btree corresponding to the extent
586 */
587void btrfs_start_ordered_extent(struct inode *inode,
588				       struct btrfs_ordered_extent *entry,
589				       int wait)
590{
591	u64 start = entry->file_offset;
592	u64 end = start + entry->len - 1;
593
594	trace_btrfs_ordered_extent_start(inode, entry);
595
596	/*
597	 * pages in the range can be dirty, clean or writeback.  We
598	 * start IO on any dirty ones so the wait doesn't stall waiting
599	 * for pdflush to find them
600	 */
601	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
602		filemap_fdatawrite_range(inode->i_mapping, start, end);
603	if (wait) {
604		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
605						 &entry->flags));
606	}
607}
608
609/*
610 * Used to wait on ordered extents across a large range of bytes.
611 */
612int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
613{
 
614	u64 end;
615	u64 orig_end;
616	struct btrfs_ordered_extent *ordered;
617	int found;
618
619	if (start + len < start) {
620		orig_end = INT_LIMIT(loff_t);
621	} else {
622		orig_end = start + len - 1;
623		if (orig_end > INT_LIMIT(loff_t))
624			orig_end = INT_LIMIT(loff_t);
625	}
626again:
627	/* start IO across the range first to instantiate any delalloc
628	 * extents
629	 */
630	filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
631
632	/* The compression code will leave pages locked but return from
633	 * writepage without setting the page writeback.  Starting again
634	 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
 
 
 
 
 
 
 
 
 
 
 
635	 */
636	filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
637
638	filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 
 
 
 
 
 
 
639
640	end = orig_end;
641	found = 0;
642	while (1) {
643		ordered = btrfs_lookup_first_ordered_extent(inode, end);
644		if (!ordered)
645			break;
646		if (ordered->file_offset > orig_end) {
647			btrfs_put_ordered_extent(ordered);
648			break;
649		}
650		if (ordered->file_offset + ordered->len < start) {
651			btrfs_put_ordered_extent(ordered);
652			break;
653		}
654		found++;
655		btrfs_start_ordered_extent(inode, ordered, 1);
656		end = ordered->file_offset;
 
 
657		btrfs_put_ordered_extent(ordered);
658		if (end == 0 || end == start)
659			break;
660		end--;
661	}
662	if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
663			   EXTENT_DELALLOC, 0, NULL)) {
664		schedule_timeout(1);
665		goto again;
666	}
667	return 0;
668}
669
670/*
671 * find an ordered extent corresponding to file_offset.  return NULL if
672 * nothing is found, otherwise take a reference on the extent and return it
673 */
674struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
675							 u64 file_offset)
676{
677	struct btrfs_ordered_inode_tree *tree;
678	struct rb_node *node;
679	struct btrfs_ordered_extent *entry = NULL;
680
681	tree = &BTRFS_I(inode)->ordered_tree;
682	spin_lock(&tree->lock);
683	node = tree_search(tree, file_offset);
684	if (!node)
685		goto out;
686
687	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
688	if (!offset_in_entry(entry, file_offset))
689		entry = NULL;
690	if (entry)
691		atomic_inc(&entry->refs);
692out:
693	spin_unlock(&tree->lock);
694	return entry;
695}
696
697/* Since the DIO code tries to lock a wide area we need to look for any ordered
698 * extents that exist in the range, rather than just the start of the range.
699 */
700struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
701							u64 file_offset,
702							u64 len)
703{
704	struct btrfs_ordered_inode_tree *tree;
705	struct rb_node *node;
706	struct btrfs_ordered_extent *entry = NULL;
707
708	tree = &BTRFS_I(inode)->ordered_tree;
709	spin_lock(&tree->lock);
710	node = tree_search(tree, file_offset);
711	if (!node) {
712		node = tree_search(tree, file_offset + len);
713		if (!node)
714			goto out;
715	}
716
717	while (1) {
718		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
719		if (range_overlaps(entry, file_offset, len))
720			break;
721
722		if (entry->file_offset >= file_offset + len) {
723			entry = NULL;
724			break;
725		}
726		entry = NULL;
727		node = rb_next(node);
728		if (!node)
729			break;
730	}
731out:
732	if (entry)
733		atomic_inc(&entry->refs);
734	spin_unlock(&tree->lock);
735	return entry;
736}
737
738/*
739 * lookup and return any extent before 'file_offset'.  NULL is returned
740 * if none is found
741 */
742struct btrfs_ordered_extent *
743btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
744{
745	struct btrfs_ordered_inode_tree *tree;
746	struct rb_node *node;
747	struct btrfs_ordered_extent *entry = NULL;
748
749	tree = &BTRFS_I(inode)->ordered_tree;
750	spin_lock(&tree->lock);
751	node = tree_search(tree, file_offset);
752	if (!node)
753		goto out;
754
755	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
756	atomic_inc(&entry->refs);
757out:
758	spin_unlock(&tree->lock);
759	return entry;
760}
761
762/*
763 * After an extent is done, call this to conditionally update the on disk
764 * i_size.  i_size is updated to cover any fully written part of the file.
765 */
766int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
767				struct btrfs_ordered_extent *ordered)
768{
769	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
770	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
771	u64 disk_i_size;
772	u64 new_i_size;
773	u64 i_size_test;
774	u64 i_size = i_size_read(inode);
775	struct rb_node *node;
776	struct rb_node *prev = NULL;
777	struct btrfs_ordered_extent *test;
778	int ret = 1;
779
780	if (ordered)
 
781		offset = entry_end(ordered);
782	else
 
 
 
 
783		offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
784
785	spin_lock(&tree->lock);
786	disk_i_size = BTRFS_I(inode)->disk_i_size;
787
788	/* truncate file */
789	if (disk_i_size > i_size) {
790		BTRFS_I(inode)->disk_i_size = i_size;
791		ret = 0;
792		goto out;
793	}
794
795	/*
796	 * if the disk i_size is already at the inode->i_size, or
797	 * this ordered extent is inside the disk i_size, we're done
798	 */
799	if (disk_i_size == i_size || offset <= disk_i_size) {
800		goto out;
801	}
802
803	/*
804	 * we can't update the disk_isize if there are delalloc bytes
805	 * between disk_i_size and  this ordered extent
806	 */
807	if (test_range_bit(io_tree, disk_i_size, offset - 1,
808			   EXTENT_DELALLOC, 0, NULL)) {
809		goto out;
810	}
811	/*
812	 * walk backward from this ordered extent to disk_i_size.
813	 * if we find an ordered extent then we can't update disk i_size
814	 * yet
815	 */
816	if (ordered) {
817		node = rb_prev(&ordered->rb_node);
818	} else {
819		prev = tree_search(tree, offset);
820		/*
821		 * we insert file extents without involving ordered struct,
822		 * so there should be no ordered struct cover this offset
823		 */
824		if (prev) {
825			test = rb_entry(prev, struct btrfs_ordered_extent,
826					rb_node);
827			BUG_ON(offset_in_entry(test, offset));
828		}
829		node = prev;
830	}
831	while (node) {
832		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 
 
 
 
833		if (test->file_offset + test->len <= disk_i_size)
834			break;
835		if (test->file_offset >= i_size)
836			break;
837		if (test->file_offset >= disk_i_size)
 
 
 
 
 
 
 
 
 
 
 
 
838			goto out;
839		node = rb_prev(node);
840	}
841	new_i_size = min_t(u64, offset, i_size);
842
843	/*
844	 * at this point, we know we can safely update i_size to at least
845	 * the offset from this ordered extent.  But, we need to
846	 * walk forward and see if ios from higher up in the file have
847	 * finished.
848	 */
849	if (ordered) {
850		node = rb_next(&ordered->rb_node);
851	} else {
852		if (prev)
853			node = rb_next(prev);
854		else
855			node = rb_first(&tree->tree);
856	}
857	i_size_test = 0;
858	if (node) {
859		/*
860		 * do we have an area where IO might have finished
861		 * between our ordered extent and the next one.
862		 */
863		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
864		if (test->file_offset > offset)
865			i_size_test = test->file_offset;
866	} else {
867		i_size_test = i_size;
868	}
869
870	/*
871	 * i_size_test is the end of a region after this ordered
872	 * extent where there are no ordered extents.  As long as there
873	 * are no delalloc bytes in this area, it is safe to update
874	 * disk_i_size to the end of the region.
875	 */
876	if (i_size_test > offset &&
877	    !test_range_bit(io_tree, offset, i_size_test - 1,
878			    EXTENT_DELALLOC, 0, NULL)) {
879		new_i_size = min_t(u64, i_size_test, i_size);
880	}
881	BTRFS_I(inode)->disk_i_size = new_i_size;
882	ret = 0;
883out:
884	/*
885	 * we need to remove the ordered extent with the tree lock held
886	 * so that other people calling this function don't find our fully
887	 * processed ordered entry and skip updating the i_size
 
 
888	 */
889	if (ordered)
890		__btrfs_remove_ordered_extent(inode, ordered);
891	spin_unlock(&tree->lock);
892	if (ordered)
893		wake_up(&ordered->wait);
894	return ret;
895}
896
897/*
898 * search the ordered extents for one corresponding to 'offset' and
899 * try to find a checksum.  This is used because we allow pages to
900 * be reclaimed before their checksum is actually put into the btree
901 */
902int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
903			   u32 *sum)
904{
905	struct btrfs_ordered_sum *ordered_sum;
906	struct btrfs_sector_sum *sector_sums;
907	struct btrfs_ordered_extent *ordered;
908	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
909	unsigned long num_sectors;
910	unsigned long i;
911	u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
912	int ret = 1;
913
914	ordered = btrfs_lookup_ordered_extent(inode, offset);
915	if (!ordered)
916		return 1;
917
918	spin_lock(&tree->lock);
919	list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
920		if (disk_bytenr >= ordered_sum->bytenr) {
921			num_sectors = ordered_sum->len / sectorsize;
922			sector_sums = ordered_sum->sums;
923			for (i = 0; i < num_sectors; i++) {
924				if (sector_sums[i].bytenr == disk_bytenr) {
925					*sum = sector_sums[i].sum;
926					ret = 0;
927					goto out;
928				}
929			}
 
 
 
 
930		}
931	}
932out:
933	spin_unlock(&tree->lock);
934	btrfs_put_ordered_extent(ordered);
935	return ret;
936}
937
938
939/*
940 * add a given inode to the list of inodes that must be fully on
941 * disk before a transaction commit finishes.
942 *
943 * This basically gives us the ext3 style data=ordered mode, and it is mostly
944 * used to make sure renamed files are fully on disk.
945 *
946 * It is a noop if the inode is already fully on disk.
947 *
948 * If trans is not null, we'll do a friendly check for a transaction that
949 * is already flushing things and force the IO down ourselves.
950 */
951int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
952				struct btrfs_root *root,
953				struct inode *inode)
954{
 
955	u64 last_mod;
956
957	last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
958
959	/*
960	 * if this file hasn't been changed since the last transaction
961	 * commit, we can safely return without doing anything
962	 */
963	if (last_mod < root->fs_info->last_trans_committed)
964		return 0;
965
966	/*
967	 * the transaction is already committing.  Just start the IO and
968	 * don't bother with all of this list nonsense
969	 */
970	if (trans && root->fs_info->running_transaction->blocked) {
971		btrfs_wait_ordered_range(inode, 0, (u64)-1);
972		return 0;
973	}
974
975	spin_lock(&root->fs_info->ordered_extent_lock);
976	if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
977		list_add_tail(&BTRFS_I(inode)->ordered_operations,
978			      &root->fs_info->ordered_operations);
979	}
980	spin_unlock(&root->fs_info->ordered_extent_lock);
 
 
 
 
 
 
 
 
 
 
981
982	return 0;
 
 
 
 
 
 
983}
v3.15
   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/slab.h>
  20#include <linux/blkdev.h>
  21#include <linux/writeback.h>
  22#include <linux/pagevec.h>
  23#include "ctree.h"
  24#include "transaction.h"
  25#include "btrfs_inode.h"
  26#include "extent_io.h"
  27#include "disk-io.h"
  28
  29static struct kmem_cache *btrfs_ordered_extent_cache;
  30
  31static u64 entry_end(struct btrfs_ordered_extent *entry)
  32{
  33	if (entry->file_offset + entry->len < entry->file_offset)
  34		return (u64)-1;
  35	return entry->file_offset + entry->len;
  36}
  37
  38/* returns NULL if the insertion worked, or it returns the node it did find
  39 * in the tree
  40 */
  41static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
  42				   struct rb_node *node)
  43{
  44	struct rb_node **p = &root->rb_node;
  45	struct rb_node *parent = NULL;
  46	struct btrfs_ordered_extent *entry;
  47
  48	while (*p) {
  49		parent = *p;
  50		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
  51
  52		if (file_offset < entry->file_offset)
  53			p = &(*p)->rb_left;
  54		else if (file_offset >= entry_end(entry))
  55			p = &(*p)->rb_right;
  56		else
  57			return parent;
  58	}
  59
  60	rb_link_node(node, parent, p);
  61	rb_insert_color(node, root);
  62	return NULL;
  63}
  64
  65static void ordered_data_tree_panic(struct inode *inode, int errno,
  66					       u64 offset)
  67{
  68	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
  69	btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
  70		    "%llu\n", offset);
  71}
  72
  73/*
  74 * look for a given offset in the tree, and if it can't be found return the
  75 * first lesser offset
  76 */
  77static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
  78				     struct rb_node **prev_ret)
  79{
  80	struct rb_node *n = root->rb_node;
  81	struct rb_node *prev = NULL;
  82	struct rb_node *test;
  83	struct btrfs_ordered_extent *entry;
  84	struct btrfs_ordered_extent *prev_entry = NULL;
  85
  86	while (n) {
  87		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
  88		prev = n;
  89		prev_entry = entry;
  90
  91		if (file_offset < entry->file_offset)
  92			n = n->rb_left;
  93		else if (file_offset >= entry_end(entry))
  94			n = n->rb_right;
  95		else
  96			return n;
  97	}
  98	if (!prev_ret)
  99		return NULL;
 100
 101	while (prev && file_offset >= entry_end(prev_entry)) {
 102		test = rb_next(prev);
 103		if (!test)
 104			break;
 105		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 106				      rb_node);
 107		if (file_offset < entry_end(prev_entry))
 108			break;
 109
 110		prev = test;
 111	}
 112	if (prev)
 113		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
 114				      rb_node);
 115	while (prev && file_offset < entry_end(prev_entry)) {
 116		test = rb_prev(prev);
 117		if (!test)
 118			break;
 119		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 120				      rb_node);
 121		prev = test;
 122	}
 123	*prev_ret = prev;
 124	return NULL;
 125}
 126
 127/*
 128 * helper to check if a given offset is inside a given entry
 129 */
 130static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
 131{
 132	if (file_offset < entry->file_offset ||
 133	    entry->file_offset + entry->len <= file_offset)
 134		return 0;
 135	return 1;
 136}
 137
 138static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
 139			  u64 len)
 140{
 141	if (file_offset + len <= entry->file_offset ||
 142	    entry->file_offset + entry->len <= file_offset)
 143		return 0;
 144	return 1;
 145}
 146
 147/*
 148 * look find the first ordered struct that has this offset, otherwise
 149 * the first one less than this offset
 150 */
 151static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
 152					  u64 file_offset)
 153{
 154	struct rb_root *root = &tree->tree;
 155	struct rb_node *prev = NULL;
 156	struct rb_node *ret;
 157	struct btrfs_ordered_extent *entry;
 158
 159	if (tree->last) {
 160		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
 161				 rb_node);
 162		if (offset_in_entry(entry, file_offset))
 163			return tree->last;
 164	}
 165	ret = __tree_search(root, file_offset, &prev);
 166	if (!ret)
 167		ret = prev;
 168	if (ret)
 169		tree->last = ret;
 170	return ret;
 171}
 172
 173/* allocate and add a new ordered_extent into the per-inode tree.
 174 * file_offset is the logical offset in the file
 175 *
 176 * start is the disk block number of an extent already reserved in the
 177 * extent allocation tree
 178 *
 179 * len is the length of the extent
 180 *
 181 * The tree is given a single reference on the ordered extent that was
 182 * inserted.
 183 */
 184static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 185				      u64 start, u64 len, u64 disk_len,
 186				      int type, int dio, int compress_type)
 187{
 188	struct btrfs_root *root = BTRFS_I(inode)->root;
 189	struct btrfs_ordered_inode_tree *tree;
 190	struct rb_node *node;
 191	struct btrfs_ordered_extent *entry;
 192
 193	tree = &BTRFS_I(inode)->ordered_tree;
 194	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
 195	if (!entry)
 196		return -ENOMEM;
 197
 198	entry->file_offset = file_offset;
 199	entry->start = start;
 200	entry->len = len;
 201	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
 202	    !(type == BTRFS_ORDERED_NOCOW))
 203		entry->csum_bytes_left = disk_len;
 204	entry->disk_len = disk_len;
 205	entry->bytes_left = len;
 206	entry->inode = igrab(inode);
 207	entry->compress_type = compress_type;
 208	entry->truncated_len = (u64)-1;
 209	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 210		set_bit(type, &entry->flags);
 211
 212	if (dio)
 213		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
 214
 215	/* one ref for the tree */
 216	atomic_set(&entry->refs, 1);
 217	init_waitqueue_head(&entry->wait);
 218	INIT_LIST_HEAD(&entry->list);
 219	INIT_LIST_HEAD(&entry->root_extent_list);
 220	INIT_LIST_HEAD(&entry->work_list);
 221	init_completion(&entry->completion);
 222	INIT_LIST_HEAD(&entry->log_list);
 223
 224	trace_btrfs_ordered_extent_add(inode, entry);
 225
 226	spin_lock_irq(&tree->lock);
 227	node = tree_insert(&tree->tree, file_offset,
 228			   &entry->rb_node);
 229	if (node)
 230		ordered_data_tree_panic(inode, -EEXIST, file_offset);
 231	spin_unlock_irq(&tree->lock);
 232
 233	spin_lock(&root->ordered_extent_lock);
 234	list_add_tail(&entry->root_extent_list,
 235		      &root->ordered_extents);
 236	root->nr_ordered_extents++;
 237	if (root->nr_ordered_extents == 1) {
 238		spin_lock(&root->fs_info->ordered_root_lock);
 239		BUG_ON(!list_empty(&root->ordered_root));
 240		list_add_tail(&root->ordered_root,
 241			      &root->fs_info->ordered_roots);
 242		spin_unlock(&root->fs_info->ordered_root_lock);
 243	}
 244	spin_unlock(&root->ordered_extent_lock);
 245
 
 246	return 0;
 247}
 248
 249int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 250			     u64 start, u64 len, u64 disk_len, int type)
 251{
 252	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 253					  disk_len, type, 0,
 254					  BTRFS_COMPRESS_NONE);
 255}
 256
 257int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
 258				 u64 start, u64 len, u64 disk_len, int type)
 259{
 260	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 261					  disk_len, type, 1,
 262					  BTRFS_COMPRESS_NONE);
 263}
 264
 265int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
 266				      u64 start, u64 len, u64 disk_len,
 267				      int type, int compress_type)
 268{
 269	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 270					  disk_len, type, 0,
 271					  compress_type);
 272}
 273
 274/*
 275 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 276 * when an ordered extent is finished.  If the list covers more than one
 277 * ordered extent, it is split across multiples.
 278 */
 279void btrfs_add_ordered_sum(struct inode *inode,
 280			   struct btrfs_ordered_extent *entry,
 281			   struct btrfs_ordered_sum *sum)
 282{
 283	struct btrfs_ordered_inode_tree *tree;
 284
 285	tree = &BTRFS_I(inode)->ordered_tree;
 286	spin_lock_irq(&tree->lock);
 287	list_add_tail(&sum->list, &entry->list);
 288	WARN_ON(entry->csum_bytes_left < sum->len);
 289	entry->csum_bytes_left -= sum->len;
 290	if (entry->csum_bytes_left == 0)
 291		wake_up(&entry->wait);
 292	spin_unlock_irq(&tree->lock);
 293}
 294
 295/*
 296 * this is used to account for finished IO across a given range
 297 * of the file.  The IO may span ordered extents.  If
 298 * a given ordered_extent is completely done, 1 is returned, otherwise
 299 * 0.
 300 *
 301 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 302 * to make sure this function only returns 1 once for a given ordered extent.
 303 *
 304 * file_offset is updated to one byte past the range that is recorded as
 305 * complete.  This allows you to walk forward in the file.
 306 */
 307int btrfs_dec_test_first_ordered_pending(struct inode *inode,
 308				   struct btrfs_ordered_extent **cached,
 309				   u64 *file_offset, u64 io_size, int uptodate)
 310{
 311	struct btrfs_ordered_inode_tree *tree;
 312	struct rb_node *node;
 313	struct btrfs_ordered_extent *entry = NULL;
 314	int ret;
 315	unsigned long flags;
 316	u64 dec_end;
 317	u64 dec_start;
 318	u64 to_dec;
 319
 320	tree = &BTRFS_I(inode)->ordered_tree;
 321	spin_lock_irqsave(&tree->lock, flags);
 322	node = tree_search(tree, *file_offset);
 323	if (!node) {
 324		ret = 1;
 325		goto out;
 326	}
 327
 328	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 329	if (!offset_in_entry(entry, *file_offset)) {
 330		ret = 1;
 331		goto out;
 332	}
 333
 334	dec_start = max(*file_offset, entry->file_offset);
 335	dec_end = min(*file_offset + io_size, entry->file_offset +
 336		      entry->len);
 337	*file_offset = dec_end;
 338	if (dec_start > dec_end) {
 339		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 340			"bad ordering dec_start %llu end %llu", dec_start, dec_end);
 
 341	}
 342	to_dec = dec_end - dec_start;
 343	if (to_dec > entry->bytes_left) {
 344		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 345			"bad ordered accounting left %llu size %llu",
 346			entry->bytes_left, to_dec);
 347	}
 348	entry->bytes_left -= to_dec;
 349	if (!uptodate)
 350		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
 351
 352	if (entry->bytes_left == 0) {
 353		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 354		if (waitqueue_active(&entry->wait))
 355			wake_up(&entry->wait);
 356	} else {
 357		ret = 1;
 358	}
 359out:
 360	if (!ret && cached && entry) {
 361		*cached = entry;
 362		atomic_inc(&entry->refs);
 363	}
 364	spin_unlock_irqrestore(&tree->lock, flags);
 365	return ret == 0;
 366}
 367
 368/*
 369 * this is used to account for finished IO across a given range
 370 * of the file.  The IO should not span ordered extents.  If
 371 * a given ordered_extent is completely done, 1 is returned, otherwise
 372 * 0.
 373 *
 374 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 375 * to make sure this function only returns 1 once for a given ordered extent.
 376 */
 377int btrfs_dec_test_ordered_pending(struct inode *inode,
 378				   struct btrfs_ordered_extent **cached,
 379				   u64 file_offset, u64 io_size, int uptodate)
 380{
 381	struct btrfs_ordered_inode_tree *tree;
 382	struct rb_node *node;
 383	struct btrfs_ordered_extent *entry = NULL;
 384	unsigned long flags;
 385	int ret;
 386
 387	tree = &BTRFS_I(inode)->ordered_tree;
 388	spin_lock_irqsave(&tree->lock, flags);
 389	if (cached && *cached) {
 390		entry = *cached;
 391		goto have_entry;
 392	}
 393
 394	node = tree_search(tree, file_offset);
 395	if (!node) {
 396		ret = 1;
 397		goto out;
 398	}
 399
 400	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 401have_entry:
 402	if (!offset_in_entry(entry, file_offset)) {
 403		ret = 1;
 404		goto out;
 405	}
 406
 407	if (io_size > entry->bytes_left) {
 408		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 409			   "bad ordered accounting left %llu size %llu",
 410		       entry->bytes_left, io_size);
 411	}
 412	entry->bytes_left -= io_size;
 413	if (!uptodate)
 414		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
 415
 416	if (entry->bytes_left == 0) {
 417		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 418		if (waitqueue_active(&entry->wait))
 419			wake_up(&entry->wait);
 420	} else {
 421		ret = 1;
 422	}
 423out:
 424	if (!ret && cached && entry) {
 425		*cached = entry;
 426		atomic_inc(&entry->refs);
 427	}
 428	spin_unlock_irqrestore(&tree->lock, flags);
 429	return ret == 0;
 430}
 431
 432/* Needs to either be called under a log transaction or the log_mutex */
 433void btrfs_get_logged_extents(struct inode *inode,
 434			      struct list_head *logged_list)
 435{
 436	struct btrfs_ordered_inode_tree *tree;
 437	struct btrfs_ordered_extent *ordered;
 438	struct rb_node *n;
 439
 440	tree = &BTRFS_I(inode)->ordered_tree;
 441	spin_lock_irq(&tree->lock);
 442	for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
 443		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
 444		if (!list_empty(&ordered->log_list))
 445			continue;
 446		list_add_tail(&ordered->log_list, logged_list);
 447		atomic_inc(&ordered->refs);
 448	}
 449	spin_unlock_irq(&tree->lock);
 450}
 451
 452void btrfs_put_logged_extents(struct list_head *logged_list)
 453{
 454	struct btrfs_ordered_extent *ordered;
 455
 456	while (!list_empty(logged_list)) {
 457		ordered = list_first_entry(logged_list,
 458					   struct btrfs_ordered_extent,
 459					   log_list);
 460		list_del_init(&ordered->log_list);
 461		btrfs_put_ordered_extent(ordered);
 462	}
 463}
 464
 465void btrfs_submit_logged_extents(struct list_head *logged_list,
 466				 struct btrfs_root *log)
 467{
 468	int index = log->log_transid % 2;
 469
 470	spin_lock_irq(&log->log_extents_lock[index]);
 471	list_splice_tail(logged_list, &log->logged_list[index]);
 472	spin_unlock_irq(&log->log_extents_lock[index]);
 473}
 474
 475void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
 476{
 477	struct btrfs_ordered_extent *ordered;
 478	int index = transid % 2;
 479
 480	spin_lock_irq(&log->log_extents_lock[index]);
 481	while (!list_empty(&log->logged_list[index])) {
 482		ordered = list_first_entry(&log->logged_list[index],
 483					   struct btrfs_ordered_extent,
 484					   log_list);
 485		list_del_init(&ordered->log_list);
 486		spin_unlock_irq(&log->log_extents_lock[index]);
 487		wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
 488						   &ordered->flags));
 489		btrfs_put_ordered_extent(ordered);
 490		spin_lock_irq(&log->log_extents_lock[index]);
 491	}
 492	spin_unlock_irq(&log->log_extents_lock[index]);
 493}
 494
 495void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
 496{
 497	struct btrfs_ordered_extent *ordered;
 498	int index = transid % 2;
 499
 500	spin_lock_irq(&log->log_extents_lock[index]);
 501	while (!list_empty(&log->logged_list[index])) {
 502		ordered = list_first_entry(&log->logged_list[index],
 503					   struct btrfs_ordered_extent,
 504					   log_list);
 505		list_del_init(&ordered->log_list);
 506		spin_unlock_irq(&log->log_extents_lock[index]);
 507		btrfs_put_ordered_extent(ordered);
 508		spin_lock_irq(&log->log_extents_lock[index]);
 509	}
 510	spin_unlock_irq(&log->log_extents_lock[index]);
 511}
 512
 513/*
 514 * used to drop a reference on an ordered extent.  This will free
 515 * the extent if the last reference is dropped
 516 */
 517void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 518{
 519	struct list_head *cur;
 520	struct btrfs_ordered_sum *sum;
 521
 522	trace_btrfs_ordered_extent_put(entry->inode, entry);
 523
 524	if (atomic_dec_and_test(&entry->refs)) {
 525		if (entry->inode)
 526			btrfs_add_delayed_iput(entry->inode);
 527		while (!list_empty(&entry->list)) {
 528			cur = entry->list.next;
 529			sum = list_entry(cur, struct btrfs_ordered_sum, list);
 530			list_del(&sum->list);
 531			kfree(sum);
 532		}
 533		kmem_cache_free(btrfs_ordered_extent_cache, entry);
 534	}
 
 535}
 536
 537/*
 538 * remove an ordered extent from the tree.  No references are dropped
 539 * and waiters are woken up.
 
 540 */
 541void btrfs_remove_ordered_extent(struct inode *inode,
 542				 struct btrfs_ordered_extent *entry)
 543{
 544	struct btrfs_ordered_inode_tree *tree;
 545	struct btrfs_root *root = BTRFS_I(inode)->root;
 546	struct rb_node *node;
 547
 548	tree = &BTRFS_I(inode)->ordered_tree;
 549	spin_lock_irq(&tree->lock);
 550	node = &entry->rb_node;
 551	rb_erase(node, &tree->tree);
 552	if (tree->last == node)
 553		tree->last = NULL;
 554	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 555	spin_unlock_irq(&tree->lock);
 556
 557	spin_lock(&root->ordered_extent_lock);
 558	list_del_init(&entry->root_extent_list);
 559	root->nr_ordered_extents--;
 560
 561	trace_btrfs_ordered_extent_remove(inode, entry);
 562
 563	/*
 564	 * we have no more ordered extents for this inode and
 565	 * no dirty pages.  We can safely remove it from the
 566	 * list of ordered extents
 567	 */
 568	if (RB_EMPTY_ROOT(&tree->tree) &&
 569	    !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 570		spin_lock(&root->fs_info->ordered_root_lock);
 571		list_del_init(&BTRFS_I(inode)->ordered_operations);
 572		spin_unlock(&root->fs_info->ordered_root_lock);
 573	}
 
 574
 575	if (!root->nr_ordered_extents) {
 576		spin_lock(&root->fs_info->ordered_root_lock);
 577		BUG_ON(list_empty(&root->ordered_root));
 578		list_del_init(&root->ordered_root);
 579		spin_unlock(&root->fs_info->ordered_root_lock);
 580	}
 581	spin_unlock(&root->ordered_extent_lock);
 582	wake_up(&entry->wait);
 583}
 584
 585static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
 
 
 
 
 
 586{
 587	struct btrfs_ordered_extent *ordered;
 
 
 
 
 
 
 
 588
 589	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
 590	btrfs_start_ordered_extent(ordered->inode, ordered, 1);
 591	complete(&ordered->completion);
 592}
 593
 594/*
 595 * wait for all the ordered extents in a root.  This is done when balancing
 596 * space between drives.
 597 */
 598int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
 
 599{
 600	struct list_head splice, works;
 601	struct btrfs_ordered_extent *ordered, *next;
 602	int count = 0;
 
 603
 604	INIT_LIST_HEAD(&splice);
 605	INIT_LIST_HEAD(&works);
 606
 607	mutex_lock(&root->ordered_extent_mutex);
 608	spin_lock(&root->ordered_extent_lock);
 609	list_splice_init(&root->ordered_extents, &splice);
 610	while (!list_empty(&splice) && nr) {
 611		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
 612					   root_extent_list);
 613		list_move_tail(&ordered->root_extent_list,
 614			       &root->ordered_extents);
 
 
 
 
 
 
 
 
 615		atomic_inc(&ordered->refs);
 616		spin_unlock(&root->ordered_extent_lock);
 617
 618		btrfs_init_work(&ordered->flush_work,
 619				btrfs_run_ordered_extent_work, NULL, NULL);
 620		list_add_tail(&ordered->work_list, &works);
 621		btrfs_queue_work(root->fs_info->flush_workers,
 622				 &ordered->flush_work);
 623
 624		cond_resched();
 625		spin_lock(&root->ordered_extent_lock);
 626		if (nr != -1)
 627			nr--;
 628		count++;
 629	}
 630	list_splice_tail(&splice, &root->ordered_extents);
 631	spin_unlock(&root->ordered_extent_lock);
 632
 633	list_for_each_entry_safe(ordered, next, &works, work_list) {
 634		list_del_init(&ordered->work_list);
 635		wait_for_completion(&ordered->completion);
 636		btrfs_put_ordered_extent(ordered);
 637		cond_resched();
 638	}
 639	mutex_unlock(&root->ordered_extent_mutex);
 640
 641	return count;
 642}
 643
 644void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
 645{
 646	struct btrfs_root *root;
 647	struct list_head splice;
 648	int done;
 
 
 
 
 
 649
 650	INIT_LIST_HEAD(&splice);
 651
 652	mutex_lock(&fs_info->ordered_operations_mutex);
 653	spin_lock(&fs_info->ordered_root_lock);
 654	list_splice_init(&fs_info->ordered_roots, &splice);
 655	while (!list_empty(&splice) && nr) {
 656		root = list_first_entry(&splice, struct btrfs_root,
 657					ordered_root);
 658		root = btrfs_grab_fs_root(root);
 659		BUG_ON(!root);
 660		list_move_tail(&root->ordered_root,
 661			       &fs_info->ordered_roots);
 662		spin_unlock(&fs_info->ordered_root_lock);
 663
 664		done = btrfs_wait_ordered_extents(root, nr);
 665		btrfs_put_fs_root(root);
 666
 667		spin_lock(&fs_info->ordered_root_lock);
 668		if (nr != -1) {
 669			nr -= done;
 670			WARN_ON(nr < 0);
 671		}
 672	}
 673	list_splice_tail(&splice, &fs_info->ordered_roots);
 674	spin_unlock(&fs_info->ordered_root_lock);
 675	mutex_unlock(&fs_info->ordered_operations_mutex);
 676}
 677
 678/*
 679 * this is used during transaction commit to write all the inodes
 680 * added to the ordered operation list.  These files must be fully on
 681 * disk before the transaction commits.
 682 *
 683 * we have two modes here, one is to just start the IO via filemap_flush
 684 * and the other is to wait for all the io.  When we wait, we have an
 685 * extra check to make sure the ordered operation list really is empty
 686 * before we return
 687 */
 688int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
 689				 struct btrfs_root *root, int wait)
 690{
 691	struct btrfs_inode *btrfs_inode;
 692	struct inode *inode;
 693	struct btrfs_transaction *cur_trans = trans->transaction;
 694	struct list_head splice;
 695	struct list_head works;
 696	struct btrfs_delalloc_work *work, *next;
 697	int ret = 0;
 698
 699	INIT_LIST_HEAD(&splice);
 700	INIT_LIST_HEAD(&works);
 701
 702	mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
 703	spin_lock(&root->fs_info->ordered_root_lock);
 704	list_splice_init(&cur_trans->ordered_operations, &splice);
 
 
 705	while (!list_empty(&splice)) {
 706		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 707				   ordered_operations);
 
 708		inode = &btrfs_inode->vfs_inode;
 709
 710		list_del_init(&btrfs_inode->ordered_operations);
 711
 712		/*
 713		 * the inode may be getting freed (in sys_unlink path).
 714		 */
 715		inode = igrab(inode);
 716		if (!inode)
 717			continue;
 718
 719		if (!wait)
 720			list_add_tail(&BTRFS_I(inode)->ordered_operations,
 721				      &cur_trans->ordered_operations);
 722		spin_unlock(&root->fs_info->ordered_root_lock);
 
 723
 724		work = btrfs_alloc_delalloc_work(inode, wait, 1);
 725		if (!work) {
 726			spin_lock(&root->fs_info->ordered_root_lock);
 727			if (list_empty(&BTRFS_I(inode)->ordered_operations))
 728				list_add_tail(&btrfs_inode->ordered_operations,
 729					      &splice);
 730			list_splice_tail(&splice,
 731					 &cur_trans->ordered_operations);
 732			spin_unlock(&root->fs_info->ordered_root_lock);
 733			ret = -ENOMEM;
 734			goto out;
 735		}
 736		list_add_tail(&work->list, &works);
 737		btrfs_queue_work(root->fs_info->flush_workers,
 738				 &work->work);
 739
 740		cond_resched();
 741		spin_lock(&root->fs_info->ordered_root_lock);
 742	}
 743	spin_unlock(&root->fs_info->ordered_root_lock);
 744out:
 745	list_for_each_entry_safe(work, next, &works, list) {
 746		list_del_init(&work->list);
 747		btrfs_wait_and_free_delalloc_work(work);
 748	}
 749	mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
 750	return ret;
 751}
 752
 753/*
 754 * Used to start IO or wait for a given ordered extent to finish.
 755 *
 756 * If wait is one, this effectively waits on page writeback for all the pages
 757 * in the extent, and it waits on the io completion code to insert
 758 * metadata into the btree corresponding to the extent
 759 */
 760void btrfs_start_ordered_extent(struct inode *inode,
 761				       struct btrfs_ordered_extent *entry,
 762				       int wait)
 763{
 764	u64 start = entry->file_offset;
 765	u64 end = start + entry->len - 1;
 766
 767	trace_btrfs_ordered_extent_start(inode, entry);
 768
 769	/*
 770	 * pages in the range can be dirty, clean or writeback.  We
 771	 * start IO on any dirty ones so the wait doesn't stall waiting
 772	 * for the flusher thread to find them
 773	 */
 774	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
 775		filemap_fdatawrite_range(inode->i_mapping, start, end);
 776	if (wait) {
 777		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 778						 &entry->flags));
 779	}
 780}
 781
 782/*
 783 * Used to wait on ordered extents across a large range of bytes.
 784 */
 785int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 786{
 787	int ret = 0;
 788	u64 end;
 789	u64 orig_end;
 790	struct btrfs_ordered_extent *ordered;
 
 791
 792	if (start + len < start) {
 793		orig_end = INT_LIMIT(loff_t);
 794	} else {
 795		orig_end = start + len - 1;
 796		if (orig_end > INT_LIMIT(loff_t))
 797			orig_end = INT_LIMIT(loff_t);
 798	}
 799
 800	/* start IO across the range first to instantiate any delalloc
 801	 * extents
 802	 */
 803	ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 804	if (ret)
 805		return ret;
 806	/*
 807	 * So with compression we will find and lock a dirty page and clear the
 808	 * first one as dirty, setup an async extent, and immediately return
 809	 * with the entire range locked but with nobody actually marked with
 810	 * writeback.  So we can't just filemap_write_and_wait_range() and
 811	 * expect it to work since it will just kick off a thread to do the
 812	 * actual work.  So we need to call filemap_fdatawrite_range _again_
 813	 * since it will wait on the page lock, which won't be unlocked until
 814	 * after the pages have been marked as writeback and so we're good to go
 815	 * from there.  We have to do this otherwise we'll miss the ordered
 816	 * extents and that results in badness.  Please Josef, do not think you
 817	 * know better and pull this out at some point in the future, it is
 818	 * right and you are wrong.
 819	 */
 820	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
 821		     &BTRFS_I(inode)->runtime_flags)) {
 822		ret = filemap_fdatawrite_range(inode->i_mapping, start,
 823					       orig_end);
 824		if (ret)
 825			return ret;
 826	}
 827	ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 828	if (ret)
 829		return ret;
 830
 831	end = orig_end;
 
 832	while (1) {
 833		ordered = btrfs_lookup_first_ordered_extent(inode, end);
 834		if (!ordered)
 835			break;
 836		if (ordered->file_offset > orig_end) {
 837			btrfs_put_ordered_extent(ordered);
 838			break;
 839		}
 840		if (ordered->file_offset + ordered->len <= start) {
 841			btrfs_put_ordered_extent(ordered);
 842			break;
 843		}
 
 844		btrfs_start_ordered_extent(inode, ordered, 1);
 845		end = ordered->file_offset;
 846		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
 847			ret = -EIO;
 848		btrfs_put_ordered_extent(ordered);
 849		if (ret || end == 0 || end == start)
 850			break;
 851		end--;
 852	}
 853	return ret;
 
 
 
 
 
 854}
 855
 856/*
 857 * find an ordered extent corresponding to file_offset.  return NULL if
 858 * nothing is found, otherwise take a reference on the extent and return it
 859 */
 860struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 861							 u64 file_offset)
 862{
 863	struct btrfs_ordered_inode_tree *tree;
 864	struct rb_node *node;
 865	struct btrfs_ordered_extent *entry = NULL;
 866
 867	tree = &BTRFS_I(inode)->ordered_tree;
 868	spin_lock_irq(&tree->lock);
 869	node = tree_search(tree, file_offset);
 870	if (!node)
 871		goto out;
 872
 873	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 874	if (!offset_in_entry(entry, file_offset))
 875		entry = NULL;
 876	if (entry)
 877		atomic_inc(&entry->refs);
 878out:
 879	spin_unlock_irq(&tree->lock);
 880	return entry;
 881}
 882
 883/* Since the DIO code tries to lock a wide area we need to look for any ordered
 884 * extents that exist in the range, rather than just the start of the range.
 885 */
 886struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
 887							u64 file_offset,
 888							u64 len)
 889{
 890	struct btrfs_ordered_inode_tree *tree;
 891	struct rb_node *node;
 892	struct btrfs_ordered_extent *entry = NULL;
 893
 894	tree = &BTRFS_I(inode)->ordered_tree;
 895	spin_lock_irq(&tree->lock);
 896	node = tree_search(tree, file_offset);
 897	if (!node) {
 898		node = tree_search(tree, file_offset + len);
 899		if (!node)
 900			goto out;
 901	}
 902
 903	while (1) {
 904		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 905		if (range_overlaps(entry, file_offset, len))
 906			break;
 907
 908		if (entry->file_offset >= file_offset + len) {
 909			entry = NULL;
 910			break;
 911		}
 912		entry = NULL;
 913		node = rb_next(node);
 914		if (!node)
 915			break;
 916	}
 917out:
 918	if (entry)
 919		atomic_inc(&entry->refs);
 920	spin_unlock_irq(&tree->lock);
 921	return entry;
 922}
 923
 924/*
 925 * lookup and return any extent before 'file_offset'.  NULL is returned
 926 * if none is found
 927 */
 928struct btrfs_ordered_extent *
 929btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 930{
 931	struct btrfs_ordered_inode_tree *tree;
 932	struct rb_node *node;
 933	struct btrfs_ordered_extent *entry = NULL;
 934
 935	tree = &BTRFS_I(inode)->ordered_tree;
 936	spin_lock_irq(&tree->lock);
 937	node = tree_search(tree, file_offset);
 938	if (!node)
 939		goto out;
 940
 941	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 942	atomic_inc(&entry->refs);
 943out:
 944	spin_unlock_irq(&tree->lock);
 945	return entry;
 946}
 947
 948/*
 949 * After an extent is done, call this to conditionally update the on disk
 950 * i_size.  i_size is updated to cover any fully written part of the file.
 951 */
 952int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 953				struct btrfs_ordered_extent *ordered)
 954{
 955	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 
 956	u64 disk_i_size;
 957	u64 new_i_size;
 
 958	u64 i_size = i_size_read(inode);
 959	struct rb_node *node;
 960	struct rb_node *prev = NULL;
 961	struct btrfs_ordered_extent *test;
 962	int ret = 1;
 963
 964	spin_lock_irq(&tree->lock);
 965	if (ordered) {
 966		offset = entry_end(ordered);
 967		if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
 968			offset = min(offset,
 969				     ordered->file_offset +
 970				     ordered->truncated_len);
 971	} else {
 972		offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 973	}
 
 974	disk_i_size = BTRFS_I(inode)->disk_i_size;
 975
 976	/* truncate file */
 977	if (disk_i_size > i_size) {
 978		BTRFS_I(inode)->disk_i_size = i_size;
 979		ret = 0;
 980		goto out;
 981	}
 982
 983	/*
 984	 * if the disk i_size is already at the inode->i_size, or
 985	 * this ordered extent is inside the disk i_size, we're done
 986	 */
 987	if (disk_i_size == i_size)
 988		goto out;
 
 989
 990	/*
 991	 * We still need to update disk_i_size if outstanding_isize is greater
 992	 * than disk_i_size.
 993	 */
 994	if (offset <= disk_i_size &&
 995	    (!ordered || ordered->outstanding_isize <= disk_i_size))
 996		goto out;
 997
 998	/*
 999	 * walk backward from this ordered extent to disk_i_size.
1000	 * if we find an ordered extent then we can't update disk i_size
1001	 * yet
1002	 */
1003	if (ordered) {
1004		node = rb_prev(&ordered->rb_node);
1005	} else {
1006		prev = tree_search(tree, offset);
1007		/*
1008		 * we insert file extents without involving ordered struct,
1009		 * so there should be no ordered struct cover this offset
1010		 */
1011		if (prev) {
1012			test = rb_entry(prev, struct btrfs_ordered_extent,
1013					rb_node);
1014			BUG_ON(offset_in_entry(test, offset));
1015		}
1016		node = prev;
1017	}
1018	for (; node; node = rb_prev(node)) {
1019		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1020
1021		/* We treat this entry as if it doesnt exist */
1022		if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1023			continue;
1024		if (test->file_offset + test->len <= disk_i_size)
1025			break;
1026		if (test->file_offset >= i_size)
1027			break;
1028		if (entry_end(test) > disk_i_size) {
1029			/*
1030			 * we don't update disk_i_size now, so record this
1031			 * undealt i_size. Or we will not know the real
1032			 * i_size.
1033			 */
1034			if (test->outstanding_isize < offset)
1035				test->outstanding_isize = offset;
1036			if (ordered &&
1037			    ordered->outstanding_isize >
1038			    test->outstanding_isize)
1039				test->outstanding_isize =
1040						ordered->outstanding_isize;
1041			goto out;
1042		}
1043	}
1044	new_i_size = min_t(u64, offset, i_size);
1045
1046	/*
1047	 * Some ordered extents may completed before the current one, and
1048	 * we hold the real i_size in ->outstanding_isize.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1049	 */
1050	if (ordered && ordered->outstanding_isize > new_i_size)
1051		new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
 
 
 
1052	BTRFS_I(inode)->disk_i_size = new_i_size;
1053	ret = 0;
1054out:
1055	/*
1056	 * We need to do this because we can't remove ordered extents until
1057	 * after the i_disk_size has been updated and then the inode has been
1058	 * updated to reflect the change, so we need to tell anybody who finds
1059	 * this ordered extent that we've already done all the real work, we
1060	 * just haven't completed all the other work.
1061	 */
1062	if (ordered)
1063		set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1064	spin_unlock_irq(&tree->lock);
 
 
1065	return ret;
1066}
1067
1068/*
1069 * search the ordered extents for one corresponding to 'offset' and
1070 * try to find a checksum.  This is used because we allow pages to
1071 * be reclaimed before their checksum is actually put into the btree
1072 */
1073int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1074			   u32 *sum, int len)
1075{
1076	struct btrfs_ordered_sum *ordered_sum;
 
1077	struct btrfs_ordered_extent *ordered;
1078	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1079	unsigned long num_sectors;
1080	unsigned long i;
1081	u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1082	int index = 0;
1083
1084	ordered = btrfs_lookup_ordered_extent(inode, offset);
1085	if (!ordered)
1086		return 0;
1087
1088	spin_lock_irq(&tree->lock);
1089	list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1090		if (disk_bytenr >= ordered_sum->bytenr &&
1091		    disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1092			i = (disk_bytenr - ordered_sum->bytenr) >>
1093			    inode->i_sb->s_blocksize_bits;
1094			num_sectors = ordered_sum->len >>
1095				      inode->i_sb->s_blocksize_bits;
1096			num_sectors = min_t(int, len - index, num_sectors - i);
1097			memcpy(sum + index, ordered_sum->sums + i,
1098			       num_sectors);
1099
1100			index += (int)num_sectors;
1101			if (index == len)
1102				goto out;
1103			disk_bytenr += num_sectors * sectorsize;
1104		}
1105	}
1106out:
1107	spin_unlock_irq(&tree->lock);
1108	btrfs_put_ordered_extent(ordered);
1109	return index;
1110}
1111
1112
1113/*
1114 * add a given inode to the list of inodes that must be fully on
1115 * disk before a transaction commit finishes.
1116 *
1117 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1118 * used to make sure renamed files are fully on disk.
1119 *
1120 * It is a noop if the inode is already fully on disk.
1121 *
1122 * If trans is not null, we'll do a friendly check for a transaction that
1123 * is already flushing things and force the IO down ourselves.
1124 */
1125void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1126				 struct btrfs_root *root, struct inode *inode)
 
1127{
1128	struct btrfs_transaction *cur_trans = trans->transaction;
1129	u64 last_mod;
1130
1131	last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1132
1133	/*
1134	 * if this file hasn't been changed since the last transaction
1135	 * commit, we can safely return without doing anything
1136	 */
1137	if (last_mod <= root->fs_info->last_trans_committed)
1138		return;
1139
1140	spin_lock(&root->fs_info->ordered_root_lock);
 
 
 
 
 
 
 
 
 
1141	if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1142		list_add_tail(&BTRFS_I(inode)->ordered_operations,
1143			      &cur_trans->ordered_operations);
1144	}
1145	spin_unlock(&root->fs_info->ordered_root_lock);
1146}
1147
1148int __init ordered_data_init(void)
1149{
1150	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1151				     sizeof(struct btrfs_ordered_extent), 0,
1152				     SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1153				     NULL);
1154	if (!btrfs_ordered_extent_cache)
1155		return -ENOMEM;
1156
1157	return 0;
1158}
1159
1160void ordered_data_exit(void)
1161{
1162	if (btrfs_ordered_extent_cache)
1163		kmem_cache_destroy(btrfs_ordered_extent_cache);
1164}