blob: d6f0806c682ff83fc11b019ac4f26d1f732acbec [file] [log] [blame]
Chris Masondc17ff82008-01-08 15:46:30 -05001/*
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/gfp.h>
20#include <linux/slab.h>
Chris Masond6bfde82008-04-30 13:59:35 -040021#include <linux/blkdev.h>
Chris Masonf4219502008-07-22 11:18:09 -040022#include <linux/writeback.h>
23#include <linux/pagevec.h>
Chris Masondc17ff82008-01-08 15:46:30 -050024#include "ctree.h"
25#include "transaction.h"
26#include "btrfs_inode.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040027#include "extent_io.h"
Chris Masondc17ff82008-01-08 15:46:30 -050028
Chris Masone6dcd2d2008-07-17 12:53:50 -040029static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050030{
Chris Masone6dcd2d2008-07-17 12:53:50 -040031 if (entry->file_offset + entry->len < entry->file_offset)
32 return (u64)-1;
33 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050034}
35
Chris Masond352ac62008-09-29 15:18:18 -040036/* returns NULL if the insertion worked, or it returns the node it did find
37 * in the tree
38 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040039static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
40 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050041{
Chris Masond3977122009-01-05 21:25:51 -050042 struct rb_node **p = &root->rb_node;
43 struct rb_node *parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040044 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050045
Chris Masond3977122009-01-05 21:25:51 -050046 while (*p) {
Chris Masondc17ff82008-01-08 15:46:30 -050047 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040048 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050049
Chris Masone6dcd2d2008-07-17 12:53:50 -040050 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050051 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040052 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050053 p = &(*p)->rb_right;
54 else
55 return parent;
56 }
57
58 rb_link_node(node, parent, p);
59 rb_insert_color(node, root);
60 return NULL;
61}
62
Chris Masond352ac62008-09-29 15:18:18 -040063/*
64 * look for a given offset in the tree, and if it can't be found return the
65 * first lesser offset
66 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040067static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
68 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050069{
Chris Masond3977122009-01-05 21:25:51 -050070 struct rb_node *n = root->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -050071 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040072 struct rb_node *test;
73 struct btrfs_ordered_extent *entry;
74 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050075
Chris Masond3977122009-01-05 21:25:51 -050076 while (n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040077 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050078 prev = n;
79 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050080
Chris Masone6dcd2d2008-07-17 12:53:50 -040081 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050082 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040083 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050084 n = n->rb_right;
85 else
86 return n;
87 }
88 if (!prev_ret)
89 return NULL;
90
Chris Masond3977122009-01-05 21:25:51 -050091 while (prev && file_offset >= entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040092 test = rb_next(prev);
93 if (!test)
94 break;
95 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
96 rb_node);
97 if (file_offset < entry_end(prev_entry))
98 break;
99
100 prev = test;
101 }
102 if (prev)
103 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
104 rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500105 while (prev && file_offset < entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400106 test = rb_prev(prev);
107 if (!test)
108 break;
109 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
110 rb_node);
111 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500112 }
113 *prev_ret = prev;
114 return NULL;
115}
116
Chris Masond352ac62008-09-29 15:18:18 -0400117/*
118 * helper to check if a given offset is inside a given entry
119 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400120static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500121{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400122 if (file_offset < entry->file_offset ||
123 entry->file_offset + entry->len <= file_offset)
124 return 0;
125 return 1;
126}
127
Chris Masond352ac62008-09-29 15:18:18 -0400128/*
129 * look find the first ordered struct that has this offset, otherwise
130 * the first one less than this offset
131 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400132static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
133 u64 file_offset)
134{
135 struct rb_root *root = &tree->tree;
Chris Masondc17ff82008-01-08 15:46:30 -0500136 struct rb_node *prev;
137 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400138 struct btrfs_ordered_extent *entry;
139
140 if (tree->last) {
141 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
142 rb_node);
143 if (offset_in_entry(entry, file_offset))
144 return tree->last;
145 }
146 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500147 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400148 ret = prev;
149 if (ret)
150 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500151 return ret;
152}
153
Chris Masoneb84ae02008-07-17 13:53:27 -0400154/* allocate and add a new ordered_extent into the per-inode tree.
155 * file_offset is the logical offset in the file
156 *
157 * start is the disk block number of an extent already reserved in the
158 * extent allocation tree
159 *
160 * len is the length of the extent
161 *
162 * This also sets the EXTENT_ORDERED bit on the range in the inode.
163 *
164 * The tree is given a single reference on the ordered extent that was
165 * inserted.
166 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400167int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
Yan Zheng80ff3852008-10-30 14:20:02 -0400168 u64 start, u64 len, u64 disk_len, int type)
Chris Masondc17ff82008-01-08 15:46:30 -0500169{
Chris Masondc17ff82008-01-08 15:46:30 -0500170 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400171 struct rb_node *node;
172 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500173
Chris Masone6dcd2d2008-07-17 12:53:50 -0400174 tree = &BTRFS_I(inode)->ordered_tree;
175 entry = kzalloc(sizeof(*entry), GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500176 if (!entry)
177 return -ENOMEM;
178
Chris Masone6dcd2d2008-07-17 12:53:50 -0400179 mutex_lock(&tree->mutex);
180 entry->file_offset = file_offset;
181 entry->start = start;
182 entry->len = len;
Chris Masonc8b97812008-10-29 14:49:59 -0400183 entry->disk_len = disk_len;
Chris Mason3eaa2882008-07-24 11:57:52 -0400184 entry->inode = inode;
Yan Zhengd899e052008-10-30 14:25:28 -0400185 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400186 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400187
Chris Masone6dcd2d2008-07-17 12:53:50 -0400188 /* one ref for the tree */
189 atomic_set(&entry->refs, 1);
190 init_waitqueue_head(&entry->wait);
191 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400192 INIT_LIST_HEAD(&entry->root_extent_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500193
Chris Masone6dcd2d2008-07-17 12:53:50 -0400194 node = tree_insert(&tree->tree, file_offset,
195 &entry->rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500196 BUG_ON(node);
197
Chris Masone6dcd2d2008-07-17 12:53:50 -0400198 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
199 entry_end(entry) - 1, GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500200
Chris Mason3eaa2882008-07-24 11:57:52 -0400201 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
202 list_add_tail(&entry->root_extent_list,
203 &BTRFS_I(inode)->root->fs_info->ordered_extents);
204 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
205
Chris Masone6dcd2d2008-07-17 12:53:50 -0400206 mutex_unlock(&tree->mutex);
207 BUG_ON(node);
208 return 0;
209}
Chris Masondc17ff82008-01-08 15:46:30 -0500210
Chris Masoneb84ae02008-07-17 13:53:27 -0400211/*
212 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400213 * when an ordered extent is finished. If the list covers more than one
214 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400215 */
Chris Mason3edf7d32008-07-18 06:17:13 -0400216int btrfs_add_ordered_sum(struct inode *inode,
217 struct btrfs_ordered_extent *entry,
218 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400219{
220 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400221
Chris Masone6dcd2d2008-07-17 12:53:50 -0400222 tree = &BTRFS_I(inode)->ordered_tree;
223 mutex_lock(&tree->mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400224 list_add_tail(&sum->list, &entry->list);
225 mutex_unlock(&tree->mutex);
226 return 0;
227}
228
Chris Masoneb84ae02008-07-17 13:53:27 -0400229/*
230 * this is used to account for finished IO across a given range
231 * of the file. The IO should not span ordered extents. If
232 * a given ordered_extent is completely done, 1 is returned, otherwise
233 * 0.
234 *
235 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
236 * to make sure this function only returns 1 once for a given ordered extent.
237 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400238int btrfs_dec_test_ordered_pending(struct inode *inode,
239 u64 file_offset, u64 io_size)
240{
241 struct btrfs_ordered_inode_tree *tree;
242 struct rb_node *node;
243 struct btrfs_ordered_extent *entry;
244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
245 int ret;
246
247 tree = &BTRFS_I(inode)->ordered_tree;
248 mutex_lock(&tree->mutex);
249 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
250 GFP_NOFS);
251 node = tree_search(tree, file_offset);
252 if (!node) {
253 ret = 1;
254 goto out;
255 }
256
257 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
258 if (!offset_in_entry(entry, file_offset)) {
259 ret = 1;
260 goto out;
261 }
262
263 ret = test_range_bit(io_tree, entry->file_offset,
264 entry->file_offset + entry->len - 1,
265 EXTENT_ORDERED, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400266 if (ret == 0)
267 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
268out:
269 mutex_unlock(&tree->mutex);
270 return ret == 0;
271}
272
Chris Masoneb84ae02008-07-17 13:53:27 -0400273/*
274 * used to drop a reference on an ordered extent. This will free
275 * the extent if the last reference is dropped
276 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400277int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
278{
Chris Masonba1da2f2008-07-17 12:54:15 -0400279 struct list_head *cur;
280 struct btrfs_ordered_sum *sum;
281
282 if (atomic_dec_and_test(&entry->refs)) {
Chris Masond3977122009-01-05 21:25:51 -0500283 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400284 cur = entry->list.next;
285 sum = list_entry(cur, struct btrfs_ordered_sum, list);
286 list_del(&sum->list);
287 kfree(sum);
288 }
Chris Masondc17ff82008-01-08 15:46:30 -0500289 kfree(entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400290 }
Chris Masondc17ff82008-01-08 15:46:30 -0500291 return 0;
292}
293
Chris Masoneb84ae02008-07-17 13:53:27 -0400294/*
295 * remove an ordered extent from the tree. No references are dropped
296 * but, anyone waiting on this extent is woken up.
297 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400298int btrfs_remove_ordered_extent(struct inode *inode,
299 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500300{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400301 struct btrfs_ordered_inode_tree *tree;
Chris Masondc17ff82008-01-08 15:46:30 -0500302 struct rb_node *node;
303
Chris Masone6dcd2d2008-07-17 12:53:50 -0400304 tree = &BTRFS_I(inode)->ordered_tree;
305 mutex_lock(&tree->mutex);
306 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500307 rb_erase(node, &tree->tree);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400308 tree->last = NULL;
309 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400310
311 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
312 list_del_init(&entry->root_extent_list);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400313
314 /*
315 * we have no more ordered extents for this inode and
316 * no dirty pages. We can safely remove it from the
317 * list of ordered extents
318 */
319 if (RB_EMPTY_ROOT(&tree->tree) &&
320 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
321 list_del_init(&BTRFS_I(inode)->ordered_operations);
322 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400323 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
324
Chris Masone6dcd2d2008-07-17 12:53:50 -0400325 mutex_unlock(&tree->mutex);
326 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400327 return 0;
328}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400329
Chris Masond352ac62008-09-29 15:18:18 -0400330/*
331 * wait for all the ordered extents in a root. This is done when balancing
332 * space between drives.
333 */
Yan Zheng7ea394f2008-08-05 13:05:02 -0400334int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
Chris Mason3eaa2882008-07-24 11:57:52 -0400335{
336 struct list_head splice;
337 struct list_head *cur;
338 struct btrfs_ordered_extent *ordered;
339 struct inode *inode;
340
341 INIT_LIST_HEAD(&splice);
342
343 spin_lock(&root->fs_info->ordered_extent_lock);
344 list_splice_init(&root->fs_info->ordered_extents, &splice);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400345 while (!list_empty(&splice)) {
Chris Mason3eaa2882008-07-24 11:57:52 -0400346 cur = splice.next;
347 ordered = list_entry(cur, struct btrfs_ordered_extent,
348 root_extent_list);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400349 if (nocow_only &&
Yan Zhengd899e052008-10-30 14:25:28 -0400350 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
351 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400352 list_move(&ordered->root_extent_list,
353 &root->fs_info->ordered_extents);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400354 cond_resched_lock(&root->fs_info->ordered_extent_lock);
355 continue;
356 }
357
Chris Mason3eaa2882008-07-24 11:57:52 -0400358 list_del_init(&ordered->root_extent_list);
359 atomic_inc(&ordered->refs);
Chris Mason3eaa2882008-07-24 11:57:52 -0400360
361 /*
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400362 * the inode may be getting freed (in sys_unlink path).
Chris Mason3eaa2882008-07-24 11:57:52 -0400363 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400364 inode = igrab(ordered->inode);
365
Chris Mason3eaa2882008-07-24 11:57:52 -0400366 spin_unlock(&root->fs_info->ordered_extent_lock);
367
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400368 if (inode) {
369 btrfs_start_ordered_extent(inode, ordered, 1);
370 btrfs_put_ordered_extent(ordered);
371 iput(inode);
372 } else {
373 btrfs_put_ordered_extent(ordered);
374 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400375
376 spin_lock(&root->fs_info->ordered_extent_lock);
377 }
378 spin_unlock(&root->fs_info->ordered_extent_lock);
379 return 0;
380}
381
Chris Masoneb84ae02008-07-17 13:53:27 -0400382/*
Chris Mason5a3f23d2009-03-31 13:27:11 -0400383 * this is used during transaction commit to write all the inodes
384 * added to the ordered operation list. These files must be fully on
385 * disk before the transaction commits.
386 *
387 * we have two modes here, one is to just start the IO via filemap_flush
388 * and the other is to wait for all the io. When we wait, we have an
389 * extra check to make sure the ordered operation list really is empty
390 * before we return
391 */
392int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
393{
394 struct btrfs_inode *btrfs_inode;
395 struct inode *inode;
396 struct list_head splice;
397
398 INIT_LIST_HEAD(&splice);
399
400 mutex_lock(&root->fs_info->ordered_operations_mutex);
401 spin_lock(&root->fs_info->ordered_extent_lock);
402again:
403 list_splice_init(&root->fs_info->ordered_operations, &splice);
404
405 while (!list_empty(&splice)) {
406 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
407 ordered_operations);
408
409 inode = &btrfs_inode->vfs_inode;
410
411 list_del_init(&btrfs_inode->ordered_operations);
412
413 /*
414 * the inode may be getting freed (in sys_unlink path).
415 */
416 inode = igrab(inode);
417
418 if (!wait && inode) {
419 list_add_tail(&BTRFS_I(inode)->ordered_operations,
420 &root->fs_info->ordered_operations);
421 }
422 spin_unlock(&root->fs_info->ordered_extent_lock);
423
424 if (inode) {
425 if (wait)
426 btrfs_wait_ordered_range(inode, 0, (u64)-1);
427 else
428 filemap_flush(inode->i_mapping);
429 iput(inode);
430 }
431
432 cond_resched();
433 spin_lock(&root->fs_info->ordered_extent_lock);
434 }
435 if (wait && !list_empty(&root->fs_info->ordered_operations))
436 goto again;
437
438 spin_unlock(&root->fs_info->ordered_extent_lock);
439 mutex_unlock(&root->fs_info->ordered_operations_mutex);
440
441 return 0;
442}
443
444/*
Chris Masoneb84ae02008-07-17 13:53:27 -0400445 * Used to start IO or wait for a given ordered extent to finish.
446 *
447 * If wait is one, this effectively waits on page writeback for all the pages
448 * in the extent, and it waits on the io completion code to insert
449 * metadata into the btree corresponding to the extent
450 */
451void btrfs_start_ordered_extent(struct inode *inode,
452 struct btrfs_ordered_extent *entry,
453 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400454{
455 u64 start = entry->file_offset;
456 u64 end = start + entry->len - 1;
457
Chris Masoneb84ae02008-07-17 13:53:27 -0400458 /*
459 * pages in the range can be dirty, clean or writeback. We
460 * start IO on any dirty ones so the wait doesn't stall waiting
461 * for pdflush to find them
462 */
Chris Mason771ed682008-11-06 22:02:51 -0500463 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL);
Chris Masonc8b97812008-10-29 14:49:59 -0400464 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400465 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
466 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400467 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400468}
469
Chris Masoneb84ae02008-07-17 13:53:27 -0400470/*
471 * Used to wait on ordered extents across a large range of bytes.
472 */
Chris Masoncb843a62008-10-03 12:30:02 -0400473int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400474{
475 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400476 u64 orig_end;
477 u64 wait_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400478 struct btrfs_ordered_extent *ordered;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400479
Chris Masone5a22172008-07-18 20:42:20 -0400480 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400481 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400482 } else {
483 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400484 if (orig_end > INT_LIMIT(loff_t))
485 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400486 }
Chris Masonf4219502008-07-22 11:18:09 -0400487 wait_end = orig_end;
Chris Mason4a096752008-07-21 10:29:44 -0400488again:
Chris Masone5a22172008-07-18 20:42:20 -0400489 /* start IO across the range first to instantiate any delalloc
490 * extents
491 */
Chris Masonffbd5172009-04-20 15:50:09 -0400492 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
Chris Masone5a22172008-07-18 20:42:20 -0400493
Chris Mason771ed682008-11-06 22:02:51 -0500494 /* The compression code will leave pages locked but return from
495 * writepage without setting the page writeback. Starting again
496 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
497 */
498 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
499
Chris Masonf4219502008-07-22 11:18:09 -0400500 btrfs_wait_on_page_writeback_range(inode->i_mapping,
501 start >> PAGE_CACHE_SHIFT,
502 orig_end >> PAGE_CACHE_SHIFT);
503
504 end = orig_end;
Chris Masond3977122009-01-05 21:25:51 -0500505 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400506 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500507 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400508 break;
Chris Masone5a22172008-07-18 20:42:20 -0400509 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400510 btrfs_put_ordered_extent(ordered);
511 break;
512 }
513 if (ordered->file_offset + ordered->len < start) {
514 btrfs_put_ordered_extent(ordered);
515 break;
516 }
Chris Masone5a22172008-07-18 20:42:20 -0400517 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400518 end = ordered->file_offset;
519 btrfs_put_ordered_extent(ordered);
Chris Masone5a22172008-07-18 20:42:20 -0400520 if (end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400521 break;
522 end--;
523 }
Chris Mason4a096752008-07-21 10:29:44 -0400524 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
525 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
Chris Mason771ed682008-11-06 22:02:51 -0500526 schedule_timeout(1);
Chris Mason4a096752008-07-21 10:29:44 -0400527 goto again;
528 }
Chris Masoncb843a62008-10-03 12:30:02 -0400529 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400530}
531
Chris Masoneb84ae02008-07-17 13:53:27 -0400532/*
533 * find an ordered extent corresponding to file_offset. return NULL if
534 * nothing is found, otherwise take a reference on the extent and return it
535 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400536struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
537 u64 file_offset)
538{
539 struct btrfs_ordered_inode_tree *tree;
540 struct rb_node *node;
541 struct btrfs_ordered_extent *entry = NULL;
542
543 tree = &BTRFS_I(inode)->ordered_tree;
544 mutex_lock(&tree->mutex);
545 node = tree_search(tree, file_offset);
546 if (!node)
547 goto out;
548
549 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
550 if (!offset_in_entry(entry, file_offset))
551 entry = NULL;
552 if (entry)
553 atomic_inc(&entry->refs);
554out:
555 mutex_unlock(&tree->mutex);
556 return entry;
557}
558
Chris Masoneb84ae02008-07-17 13:53:27 -0400559/*
560 * lookup and return any extent before 'file_offset'. NULL is returned
561 * if none is found
562 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400563struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500564btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400565{
566 struct btrfs_ordered_inode_tree *tree;
567 struct rb_node *node;
568 struct btrfs_ordered_extent *entry = NULL;
569
570 tree = &BTRFS_I(inode)->ordered_tree;
571 mutex_lock(&tree->mutex);
572 node = tree_search(tree, file_offset);
573 if (!node)
574 goto out;
575
576 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
577 atomic_inc(&entry->refs);
578out:
579 mutex_unlock(&tree->mutex);
580 return entry;
581}
Chris Masondbe674a2008-07-17 12:54:05 -0400582
Chris Masoneb84ae02008-07-17 13:53:27 -0400583/*
584 * After an extent is done, call this to conditionally update the on disk
585 * i_size. i_size is updated to cover any fully written part of the file.
586 */
Chris Masondbe674a2008-07-17 12:54:05 -0400587int btrfs_ordered_update_i_size(struct inode *inode,
588 struct btrfs_ordered_extent *ordered)
589{
590 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
591 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
592 u64 disk_i_size;
593 u64 new_i_size;
594 u64 i_size_test;
595 struct rb_node *node;
596 struct btrfs_ordered_extent *test;
597
598 mutex_lock(&tree->mutex);
599 disk_i_size = BTRFS_I(inode)->disk_i_size;
600
601 /*
602 * if the disk i_size is already at the inode->i_size, or
603 * this ordered extent is inside the disk i_size, we're done
604 */
605 if (disk_i_size >= inode->i_size ||
606 ordered->file_offset + ordered->len <= disk_i_size) {
607 goto out;
608 }
609
610 /*
611 * we can't update the disk_isize if there are delalloc bytes
612 * between disk_i_size and this ordered extent
613 */
614 if (test_range_bit(io_tree, disk_i_size,
615 ordered->file_offset + ordered->len - 1,
616 EXTENT_DELALLOC, 0)) {
617 goto out;
618 }
619 /*
620 * walk backward from this ordered extent to disk_i_size.
621 * if we find an ordered extent then we can't update disk i_size
622 * yet
623 */
Chris Masonba1da2f2008-07-17 12:54:15 -0400624 node = &ordered->rb_node;
Chris Masond3977122009-01-05 21:25:51 -0500625 while (1) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400626 node = rb_prev(node);
Chris Masondbe674a2008-07-17 12:54:05 -0400627 if (!node)
628 break;
629 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
630 if (test->file_offset + test->len <= disk_i_size)
631 break;
632 if (test->file_offset >= inode->i_size)
633 break;
634 if (test->file_offset >= disk_i_size)
635 goto out;
636 }
637 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
638
639 /*
640 * at this point, we know we can safely update i_size to at least
641 * the offset from this ordered extent. But, we need to
642 * walk forward and see if ios from higher up in the file have
643 * finished.
644 */
645 node = rb_next(&ordered->rb_node);
646 i_size_test = 0;
647 if (node) {
648 /*
649 * do we have an area where IO might have finished
650 * between our ordered extent and the next one.
651 */
652 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500653 if (test->file_offset > entry_end(ordered))
Yan Zhengb48652c2008-08-04 23:23:47 -0400654 i_size_test = test->file_offset;
Chris Masondbe674a2008-07-17 12:54:05 -0400655 } else {
656 i_size_test = i_size_read(inode);
657 }
658
659 /*
660 * i_size_test is the end of a region after this ordered
661 * extent where there are no ordered extents. As long as there
662 * are no delalloc bytes in this area, it is safe to update
663 * disk_i_size to the end of the region.
664 */
665 if (i_size_test > entry_end(ordered) &&
Yan Zhengb48652c2008-08-04 23:23:47 -0400666 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
Chris Masondbe674a2008-07-17 12:54:05 -0400667 EXTENT_DELALLOC, 0)) {
668 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
669 }
670 BTRFS_I(inode)->disk_i_size = new_i_size;
671out:
672 mutex_unlock(&tree->mutex);
673 return 0;
674}
Chris Masonba1da2f2008-07-17 12:54:15 -0400675
Chris Masoneb84ae02008-07-17 13:53:27 -0400676/*
677 * search the ordered extents for one corresponding to 'offset' and
678 * try to find a checksum. This is used because we allow pages to
679 * be reclaimed before their checksum is actually put into the btree
680 */
Chris Masond20f7042008-12-08 16:58:54 -0500681int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
682 u32 *sum)
Chris Masonba1da2f2008-07-17 12:54:15 -0400683{
684 struct btrfs_ordered_sum *ordered_sum;
685 struct btrfs_sector_sum *sector_sums;
686 struct btrfs_ordered_extent *ordered;
687 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -0400688 unsigned long num_sectors;
689 unsigned long i;
690 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -0400691 int ret = 1;
Chris Masonba1da2f2008-07-17 12:54:15 -0400692
693 ordered = btrfs_lookup_ordered_extent(inode, offset);
694 if (!ordered)
695 return 1;
696
697 mutex_lock(&tree->mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500698 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Chris Masond20f7042008-12-08 16:58:54 -0500699 if (disk_bytenr >= ordered_sum->bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400700 num_sectors = ordered_sum->len / sectorsize;
Chris Masoned98b562008-07-22 23:06:42 -0400701 sector_sums = ordered_sum->sums;
Chris Mason3edf7d32008-07-18 06:17:13 -0400702 for (i = 0; i < num_sectors; i++) {
Chris Masond20f7042008-12-08 16:58:54 -0500703 if (sector_sums[i].bytenr == disk_bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400704 *sum = sector_sums[i].sum;
705 ret = 0;
706 goto out;
707 }
708 }
Chris Masonba1da2f2008-07-17 12:54:15 -0400709 }
710 }
711out:
712 mutex_unlock(&tree->mutex);
Chris Mason89642222008-07-24 09:41:53 -0400713 btrfs_put_ordered_extent(ordered);
Chris Masonba1da2f2008-07-17 12:54:15 -0400714 return ret;
715}
716
Chris Masonf4219502008-07-22 11:18:09 -0400717
718/**
719 * taken from mm/filemap.c because it isn't exported
720 *
721 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
722 * @mapping: address space structure to write
723 * @start: offset in bytes where the range starts
724 * @end: offset in bytes where the range ends (inclusive)
725 * @sync_mode: enable synchronous operation
726 *
727 * Start writeback against all of a mapping's dirty pages that lie
728 * within the byte offsets <start, end> inclusive.
729 *
730 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
731 * opposed to a regular memory cleansing writeback. The difference between
732 * these two operations is that if a dirty page/buffer is encountered, it must
733 * be waited upon, and not just skipped over.
734 */
735int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
736 loff_t end, int sync_mode)
737{
738 struct writeback_control wbc = {
739 .sync_mode = sync_mode,
740 .nr_to_write = mapping->nrpages * 2,
741 .range_start = start,
742 .range_end = end,
743 .for_writepages = 1,
744 };
745 return btrfs_writepages(mapping, &wbc);
746}
747
748/**
749 * taken from mm/filemap.c because it isn't exported
750 *
751 * wait_on_page_writeback_range - wait for writeback to complete
752 * @mapping: target address_space
753 * @start: beginning page index
754 * @end: ending page index
755 *
756 * Wait for writeback to complete against pages indexed by start->end
757 * inclusive
758 */
759int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
760 pgoff_t start, pgoff_t end)
761{
762 struct pagevec pvec;
763 int nr_pages;
764 int ret = 0;
765 pgoff_t index;
766
767 if (end < start)
768 return 0;
769
770 pagevec_init(&pvec, 0);
771 index = start;
772 while ((index <= end) &&
773 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
774 PAGECACHE_TAG_WRITEBACK,
775 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
776 unsigned i;
777
778 for (i = 0; i < nr_pages; i++) {
779 struct page *page = pvec.pages[i];
780
781 /* until radix tree lookup accepts end_index */
782 if (page->index > end)
783 continue;
784
785 wait_on_page_writeback(page);
786 if (PageError(page))
787 ret = -EIO;
788 }
789 pagevec_release(&pvec);
790 cond_resched();
791 }
792
793 /* Check for outstanding write errors */
794 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
795 ret = -ENOSPC;
796 if (test_and_clear_bit(AS_EIO, &mapping->flags))
797 ret = -EIO;
798
799 return ret;
800}
Chris Mason5a3f23d2009-03-31 13:27:11 -0400801
802/*
803 * add a given inode to the list of inodes that must be fully on
804 * disk before a transaction commit finishes.
805 *
806 * This basically gives us the ext3 style data=ordered mode, and it is mostly
807 * used to make sure renamed files are fully on disk.
808 *
809 * It is a noop if the inode is already fully on disk.
810 *
811 * If trans is not null, we'll do a friendly check for a transaction that
812 * is already flushing things and force the IO down ourselves.
813 */
814int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 struct inode *inode)
817{
818 u64 last_mod;
819
820 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
821
822 /*
823 * if this file hasn't been changed since the last transaction
824 * commit, we can safely return without doing anything
825 */
826 if (last_mod < root->fs_info->last_trans_committed)
827 return 0;
828
829 /*
830 * the transaction is already committing. Just start the IO and
831 * don't bother with all of this list nonsense
832 */
833 if (trans && root->fs_info->running_transaction->blocked) {
834 btrfs_wait_ordered_range(inode, 0, (u64)-1);
835 return 0;
836 }
837
838 spin_lock(&root->fs_info->ordered_extent_lock);
839 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
840 list_add_tail(&BTRFS_I(inode)->ordered_operations,
841 &root->fs_info->ordered_operations);
842 }
843 spin_unlock(&root->fs_info->ordered_extent_lock);
844
845 return 0;
846}