blob: 5799bc46a30993a1cb477fb3a4dd9db23dcad6da [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 *
Chris Masoneb84ae02008-07-17 13:53:27 -0400162 * The tree is given a single reference on the ordered extent that was
163 * inserted.
164 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400165int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
Yan Zheng80ff3852008-10-30 14:20:02 -0400166 u64 start, u64 len, u64 disk_len, int type)
Chris Masondc17ff82008-01-08 15:46:30 -0500167{
Chris Masondc17ff82008-01-08 15:46:30 -0500168 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400169 struct rb_node *node;
170 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500171
Chris Masone6dcd2d2008-07-17 12:53:50 -0400172 tree = &BTRFS_I(inode)->ordered_tree;
173 entry = kzalloc(sizeof(*entry), GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500174 if (!entry)
175 return -ENOMEM;
176
Chris Masone6dcd2d2008-07-17 12:53:50 -0400177 mutex_lock(&tree->mutex);
178 entry->file_offset = file_offset;
179 entry->start = start;
180 entry->len = len;
Chris Masonc8b97812008-10-29 14:49:59 -0400181 entry->disk_len = disk_len;
Chris Mason8b62b722009-09-02 16:53:46 -0400182 entry->bytes_left = len;
Chris Mason3eaa2882008-07-24 11:57:52 -0400183 entry->inode = inode;
Yan Zhengd899e052008-10-30 14:25:28 -0400184 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400185 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400186
Chris Masone6dcd2d2008-07-17 12:53:50 -0400187 /* one ref for the tree */
188 atomic_set(&entry->refs, 1);
189 init_waitqueue_head(&entry->wait);
190 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400191 INIT_LIST_HEAD(&entry->root_extent_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500192
Chris Masone6dcd2d2008-07-17 12:53:50 -0400193 node = tree_insert(&tree->tree, file_offset,
194 &entry->rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500195 BUG_ON(node);
196
Chris Mason3eaa2882008-07-24 11:57:52 -0400197 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
198 list_add_tail(&entry->root_extent_list,
199 &BTRFS_I(inode)->root->fs_info->ordered_extents);
200 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
201
Chris Masone6dcd2d2008-07-17 12:53:50 -0400202 mutex_unlock(&tree->mutex);
203 BUG_ON(node);
204 return 0;
205}
Chris Masondc17ff82008-01-08 15:46:30 -0500206
Chris Masoneb84ae02008-07-17 13:53:27 -0400207/*
208 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400209 * when an ordered extent is finished. If the list covers more than one
210 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400211 */
Chris Mason3edf7d32008-07-18 06:17:13 -0400212int btrfs_add_ordered_sum(struct inode *inode,
213 struct btrfs_ordered_extent *entry,
214 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400215{
216 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400217
Chris Masone6dcd2d2008-07-17 12:53:50 -0400218 tree = &BTRFS_I(inode)->ordered_tree;
219 mutex_lock(&tree->mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400220 list_add_tail(&sum->list, &entry->list);
221 mutex_unlock(&tree->mutex);
222 return 0;
223}
224
Chris Masoneb84ae02008-07-17 13:53:27 -0400225/*
226 * this is used to account for finished IO across a given range
227 * of the file. The IO should not span ordered extents. If
228 * a given ordered_extent is completely done, 1 is returned, otherwise
229 * 0.
230 *
231 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
232 * to make sure this function only returns 1 once for a given ordered extent.
233 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400234int btrfs_dec_test_ordered_pending(struct inode *inode,
235 u64 file_offset, u64 io_size)
236{
237 struct btrfs_ordered_inode_tree *tree;
238 struct rb_node *node;
239 struct btrfs_ordered_extent *entry;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400240 int ret;
241
242 tree = &BTRFS_I(inode)->ordered_tree;
243 mutex_lock(&tree->mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400244 node = tree_search(tree, file_offset);
245 if (!node) {
246 ret = 1;
247 goto out;
248 }
249
250 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
251 if (!offset_in_entry(entry, file_offset)) {
252 ret = 1;
253 goto out;
254 }
255
Chris Mason8b62b722009-09-02 16:53:46 -0400256 if (io_size > entry->bytes_left) {
257 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
258 (unsigned long long)entry->bytes_left,
259 (unsigned long long)io_size);
260 }
261 entry->bytes_left -= io_size;
262 if (entry->bytes_left == 0)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400263 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
Chris Mason8b62b722009-09-02 16:53:46 -0400264 else
265 ret = 1;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400266out:
267 mutex_unlock(&tree->mutex);
268 return ret == 0;
269}
270
Chris Masoneb84ae02008-07-17 13:53:27 -0400271/*
272 * used to drop a reference on an ordered extent. This will free
273 * the extent if the last reference is dropped
274 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400275int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
276{
Chris Masonba1da2f2008-07-17 12:54:15 -0400277 struct list_head *cur;
278 struct btrfs_ordered_sum *sum;
279
280 if (atomic_dec_and_test(&entry->refs)) {
Chris Masond3977122009-01-05 21:25:51 -0500281 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400282 cur = entry->list.next;
283 sum = list_entry(cur, struct btrfs_ordered_sum, list);
284 list_del(&sum->list);
285 kfree(sum);
286 }
Chris Masondc17ff82008-01-08 15:46:30 -0500287 kfree(entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400288 }
Chris Masondc17ff82008-01-08 15:46:30 -0500289 return 0;
290}
291
Chris Masoneb84ae02008-07-17 13:53:27 -0400292/*
293 * remove an ordered extent from the tree. No references are dropped
294 * but, anyone waiting on this extent is woken up.
295 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400296int btrfs_remove_ordered_extent(struct inode *inode,
297 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500298{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400299 struct btrfs_ordered_inode_tree *tree;
Chris Masondc17ff82008-01-08 15:46:30 -0500300 struct rb_node *node;
301
Chris Masone6dcd2d2008-07-17 12:53:50 -0400302 tree = &BTRFS_I(inode)->ordered_tree;
303 mutex_lock(&tree->mutex);
304 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500305 rb_erase(node, &tree->tree);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400306 tree->last = NULL;
307 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400308
Josef Bacik32c00af2009-10-08 13:34:05 -0400309 spin_lock(&BTRFS_I(inode)->accounting_lock);
310 BTRFS_I(inode)->outstanding_extents--;
311 spin_unlock(&BTRFS_I(inode)->accounting_lock);
312 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
313 inode, 1);
314
Chris Mason3eaa2882008-07-24 11:57:52 -0400315 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
316 list_del_init(&entry->root_extent_list);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400317
318 /*
319 * we have no more ordered extents for this inode and
320 * no dirty pages. We can safely remove it from the
321 * list of ordered extents
322 */
323 if (RB_EMPTY_ROOT(&tree->tree) &&
324 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
325 list_del_init(&BTRFS_I(inode)->ordered_operations);
326 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400327 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
328
Chris Masone6dcd2d2008-07-17 12:53:50 -0400329 mutex_unlock(&tree->mutex);
330 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400331 return 0;
332}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400333
Chris Masond352ac62008-09-29 15:18:18 -0400334/*
335 * wait for all the ordered extents in a root. This is done when balancing
336 * space between drives.
337 */
Yan Zheng7ea394f2008-08-05 13:05:02 -0400338int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
Chris Mason3eaa2882008-07-24 11:57:52 -0400339{
340 struct list_head splice;
341 struct list_head *cur;
342 struct btrfs_ordered_extent *ordered;
343 struct inode *inode;
344
345 INIT_LIST_HEAD(&splice);
346
347 spin_lock(&root->fs_info->ordered_extent_lock);
348 list_splice_init(&root->fs_info->ordered_extents, &splice);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400349 while (!list_empty(&splice)) {
Chris Mason3eaa2882008-07-24 11:57:52 -0400350 cur = splice.next;
351 ordered = list_entry(cur, struct btrfs_ordered_extent,
352 root_extent_list);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400353 if (nocow_only &&
Yan Zhengd899e052008-10-30 14:25:28 -0400354 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
355 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400356 list_move(&ordered->root_extent_list,
357 &root->fs_info->ordered_extents);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400358 cond_resched_lock(&root->fs_info->ordered_extent_lock);
359 continue;
360 }
361
Chris Mason3eaa2882008-07-24 11:57:52 -0400362 list_del_init(&ordered->root_extent_list);
363 atomic_inc(&ordered->refs);
Chris Mason3eaa2882008-07-24 11:57:52 -0400364
365 /*
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400366 * the inode may be getting freed (in sys_unlink path).
Chris Mason3eaa2882008-07-24 11:57:52 -0400367 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400368 inode = igrab(ordered->inode);
369
Chris Mason3eaa2882008-07-24 11:57:52 -0400370 spin_unlock(&root->fs_info->ordered_extent_lock);
371
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400372 if (inode) {
373 btrfs_start_ordered_extent(inode, ordered, 1);
374 btrfs_put_ordered_extent(ordered);
375 iput(inode);
376 } else {
377 btrfs_put_ordered_extent(ordered);
378 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400379
380 spin_lock(&root->fs_info->ordered_extent_lock);
381 }
382 spin_unlock(&root->fs_info->ordered_extent_lock);
383 return 0;
384}
385
Chris Masoneb84ae02008-07-17 13:53:27 -0400386/*
Chris Mason5a3f23d2009-03-31 13:27:11 -0400387 * this is used during transaction commit to write all the inodes
388 * added to the ordered operation list. These files must be fully on
389 * disk before the transaction commits.
390 *
391 * we have two modes here, one is to just start the IO via filemap_flush
392 * and the other is to wait for all the io. When we wait, we have an
393 * extra check to make sure the ordered operation list really is empty
394 * before we return
395 */
396int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
397{
398 struct btrfs_inode *btrfs_inode;
399 struct inode *inode;
400 struct list_head splice;
401
402 INIT_LIST_HEAD(&splice);
403
404 mutex_lock(&root->fs_info->ordered_operations_mutex);
405 spin_lock(&root->fs_info->ordered_extent_lock);
406again:
407 list_splice_init(&root->fs_info->ordered_operations, &splice);
408
409 while (!list_empty(&splice)) {
410 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
411 ordered_operations);
412
413 inode = &btrfs_inode->vfs_inode;
414
415 list_del_init(&btrfs_inode->ordered_operations);
416
417 /*
418 * the inode may be getting freed (in sys_unlink path).
419 */
420 inode = igrab(inode);
421
422 if (!wait && inode) {
423 list_add_tail(&BTRFS_I(inode)->ordered_operations,
424 &root->fs_info->ordered_operations);
425 }
426 spin_unlock(&root->fs_info->ordered_extent_lock);
427
428 if (inode) {
429 if (wait)
430 btrfs_wait_ordered_range(inode, 0, (u64)-1);
431 else
432 filemap_flush(inode->i_mapping);
433 iput(inode);
434 }
435
436 cond_resched();
437 spin_lock(&root->fs_info->ordered_extent_lock);
438 }
439 if (wait && !list_empty(&root->fs_info->ordered_operations))
440 goto again;
441
442 spin_unlock(&root->fs_info->ordered_extent_lock);
443 mutex_unlock(&root->fs_info->ordered_operations_mutex);
444
445 return 0;
446}
447
448/*
Chris Masoneb84ae02008-07-17 13:53:27 -0400449 * Used to start IO or wait for a given ordered extent to finish.
450 *
451 * If wait is one, this effectively waits on page writeback for all the pages
452 * in the extent, and it waits on the io completion code to insert
453 * metadata into the btree corresponding to the extent
454 */
455void btrfs_start_ordered_extent(struct inode *inode,
456 struct btrfs_ordered_extent *entry,
457 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400458{
459 u64 start = entry->file_offset;
460 u64 end = start + entry->len - 1;
461
Chris Masoneb84ae02008-07-17 13:53:27 -0400462 /*
463 * pages in the range can be dirty, clean or writeback. We
464 * start IO on any dirty ones so the wait doesn't stall waiting
465 * for pdflush to find them
466 */
Christoph Hellwig8aa38c32009-10-01 12:58:30 -0400467 filemap_fdatawrite_range(inode->i_mapping, start, end);
Chris Masonc8b97812008-10-29 14:49:59 -0400468 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400469 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
470 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400471 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400472}
473
Chris Masoneb84ae02008-07-17 13:53:27 -0400474/*
475 * Used to wait on ordered extents across a large range of bytes.
476 */
Chris Masoncb843a62008-10-03 12:30:02 -0400477int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400478{
479 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400480 u64 orig_end;
481 u64 wait_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400482 struct btrfs_ordered_extent *ordered;
Chris Mason8b62b722009-09-02 16:53:46 -0400483 int found;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400484
Chris Masone5a22172008-07-18 20:42:20 -0400485 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400486 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400487 } else {
488 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400489 if (orig_end > INT_LIMIT(loff_t))
490 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400491 }
Chris Masonf4219502008-07-22 11:18:09 -0400492 wait_end = orig_end;
Chris Mason4a096752008-07-21 10:29:44 -0400493again:
Chris Masone5a22172008-07-18 20:42:20 -0400494 /* start IO across the range first to instantiate any delalloc
495 * extents
496 */
Christoph Hellwig8aa38c32009-10-01 12:58:30 -0400497 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
Chris Masone5a22172008-07-18 20:42:20 -0400498
Chris Mason771ed682008-11-06 22:02:51 -0500499 /* The compression code will leave pages locked but return from
500 * writepage without setting the page writeback. Starting again
501 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
502 */
Christoph Hellwig8aa38c32009-10-01 12:58:30 -0400503 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
Chris Mason771ed682008-11-06 22:02:51 -0500504
Christoph Hellwig8aa38c32009-10-01 12:58:30 -0400505 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
Chris Masonf4219502008-07-22 11:18:09 -0400506
507 end = orig_end;
Chris Mason8b62b722009-09-02 16:53:46 -0400508 found = 0;
Chris Masond3977122009-01-05 21:25:51 -0500509 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400510 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500511 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400512 break;
Chris Masone5a22172008-07-18 20:42:20 -0400513 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400514 btrfs_put_ordered_extent(ordered);
515 break;
516 }
517 if (ordered->file_offset + ordered->len < start) {
518 btrfs_put_ordered_extent(ordered);
519 break;
520 }
Chris Mason8b62b722009-09-02 16:53:46 -0400521 found++;
Chris Masone5a22172008-07-18 20:42:20 -0400522 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400523 end = ordered->file_offset;
524 btrfs_put_ordered_extent(ordered);
Chris Masone5a22172008-07-18 20:42:20 -0400525 if (end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400526 break;
527 end--;
528 }
Chris Mason8b62b722009-09-02 16:53:46 -0400529 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
530 EXTENT_DELALLOC, 0, NULL)) {
Chris Mason771ed682008-11-06 22:02:51 -0500531 schedule_timeout(1);
Chris Mason4a096752008-07-21 10:29:44 -0400532 goto again;
533 }
Chris Masoncb843a62008-10-03 12:30:02 -0400534 return 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400535}
536
Chris Masoneb84ae02008-07-17 13:53:27 -0400537/*
538 * find an ordered extent corresponding to file_offset. return NULL if
539 * nothing is found, otherwise take a reference on the extent and return it
540 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400541struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
542 u64 file_offset)
543{
544 struct btrfs_ordered_inode_tree *tree;
545 struct rb_node *node;
546 struct btrfs_ordered_extent *entry = NULL;
547
548 tree = &BTRFS_I(inode)->ordered_tree;
549 mutex_lock(&tree->mutex);
550 node = tree_search(tree, file_offset);
551 if (!node)
552 goto out;
553
554 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
555 if (!offset_in_entry(entry, file_offset))
556 entry = NULL;
557 if (entry)
558 atomic_inc(&entry->refs);
559out:
560 mutex_unlock(&tree->mutex);
561 return entry;
562}
563
Chris Masoneb84ae02008-07-17 13:53:27 -0400564/*
565 * lookup and return any extent before 'file_offset'. NULL is returned
566 * if none is found
567 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400568struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500569btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400570{
571 struct btrfs_ordered_inode_tree *tree;
572 struct rb_node *node;
573 struct btrfs_ordered_extent *entry = NULL;
574
575 tree = &BTRFS_I(inode)->ordered_tree;
576 mutex_lock(&tree->mutex);
577 node = tree_search(tree, file_offset);
578 if (!node)
579 goto out;
580
581 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
582 atomic_inc(&entry->refs);
583out:
584 mutex_unlock(&tree->mutex);
585 return entry;
586}
Chris Masondbe674a2008-07-17 12:54:05 -0400587
Chris Masoneb84ae02008-07-17 13:53:27 -0400588/*
589 * After an extent is done, call this to conditionally update the on disk
590 * i_size. i_size is updated to cover any fully written part of the file.
591 */
Chris Masondbe674a2008-07-17 12:54:05 -0400592int btrfs_ordered_update_i_size(struct inode *inode,
593 struct btrfs_ordered_extent *ordered)
594{
595 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
596 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
597 u64 disk_i_size;
598 u64 new_i_size;
599 u64 i_size_test;
600 struct rb_node *node;
601 struct btrfs_ordered_extent *test;
602
603 mutex_lock(&tree->mutex);
604 disk_i_size = BTRFS_I(inode)->disk_i_size;
605
606 /*
607 * if the disk i_size is already at the inode->i_size, or
608 * this ordered extent is inside the disk i_size, we're done
609 */
610 if (disk_i_size >= inode->i_size ||
611 ordered->file_offset + ordered->len <= disk_i_size) {
612 goto out;
613 }
614
615 /*
616 * we can't update the disk_isize if there are delalloc bytes
617 * between disk_i_size and this ordered extent
618 */
619 if (test_range_bit(io_tree, disk_i_size,
620 ordered->file_offset + ordered->len - 1,
Chris Mason9655d292009-09-02 15:22:30 -0400621 EXTENT_DELALLOC, 0, NULL)) {
Chris Masondbe674a2008-07-17 12:54:05 -0400622 goto out;
623 }
624 /*
625 * walk backward from this ordered extent to disk_i_size.
626 * if we find an ordered extent then we can't update disk i_size
627 * yet
628 */
Chris Masonba1da2f2008-07-17 12:54:15 -0400629 node = &ordered->rb_node;
Chris Masond3977122009-01-05 21:25:51 -0500630 while (1) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400631 node = rb_prev(node);
Chris Masondbe674a2008-07-17 12:54:05 -0400632 if (!node)
633 break;
634 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
635 if (test->file_offset + test->len <= disk_i_size)
636 break;
637 if (test->file_offset >= inode->i_size)
638 break;
639 if (test->file_offset >= disk_i_size)
640 goto out;
641 }
642 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
643
644 /*
645 * at this point, we know we can safely update i_size to at least
646 * the offset from this ordered extent. But, we need to
647 * walk forward and see if ios from higher up in the file have
648 * finished.
649 */
650 node = rb_next(&ordered->rb_node);
651 i_size_test = 0;
652 if (node) {
653 /*
654 * do we have an area where IO might have finished
655 * between our ordered extent and the next one.
656 */
657 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500658 if (test->file_offset > entry_end(ordered))
Yan Zhengb48652c2008-08-04 23:23:47 -0400659 i_size_test = test->file_offset;
Chris Masondbe674a2008-07-17 12:54:05 -0400660 } else {
661 i_size_test = i_size_read(inode);
662 }
663
664 /*
665 * i_size_test is the end of a region after this ordered
666 * extent where there are no ordered extents. As long as there
667 * are no delalloc bytes in this area, it is safe to update
668 * disk_i_size to the end of the region.
669 */
670 if (i_size_test > entry_end(ordered) &&
Yan Zhengb48652c2008-08-04 23:23:47 -0400671 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
Chris Mason9655d292009-09-02 15:22:30 -0400672 EXTENT_DELALLOC, 0, NULL)) {
Chris Masondbe674a2008-07-17 12:54:05 -0400673 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
674 }
675 BTRFS_I(inode)->disk_i_size = new_i_size;
676out:
677 mutex_unlock(&tree->mutex);
678 return 0;
679}
Chris Masonba1da2f2008-07-17 12:54:15 -0400680
Chris Masoneb84ae02008-07-17 13:53:27 -0400681/*
682 * search the ordered extents for one corresponding to 'offset' and
683 * try to find a checksum. This is used because we allow pages to
684 * be reclaimed before their checksum is actually put into the btree
685 */
Chris Masond20f7042008-12-08 16:58:54 -0500686int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
687 u32 *sum)
Chris Masonba1da2f2008-07-17 12:54:15 -0400688{
689 struct btrfs_ordered_sum *ordered_sum;
690 struct btrfs_sector_sum *sector_sums;
691 struct btrfs_ordered_extent *ordered;
692 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -0400693 unsigned long num_sectors;
694 unsigned long i;
695 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -0400696 int ret = 1;
Chris Masonba1da2f2008-07-17 12:54:15 -0400697
698 ordered = btrfs_lookup_ordered_extent(inode, offset);
699 if (!ordered)
700 return 1;
701
702 mutex_lock(&tree->mutex);
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500703 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Chris Masond20f7042008-12-08 16:58:54 -0500704 if (disk_bytenr >= ordered_sum->bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400705 num_sectors = ordered_sum->len / sectorsize;
Chris Masoned98b562008-07-22 23:06:42 -0400706 sector_sums = ordered_sum->sums;
Chris Mason3edf7d32008-07-18 06:17:13 -0400707 for (i = 0; i < num_sectors; i++) {
Chris Masond20f7042008-12-08 16:58:54 -0500708 if (sector_sums[i].bytenr == disk_bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400709 *sum = sector_sums[i].sum;
710 ret = 0;
711 goto out;
712 }
713 }
Chris Masonba1da2f2008-07-17 12:54:15 -0400714 }
715 }
716out:
717 mutex_unlock(&tree->mutex);
Chris Mason89642222008-07-24 09:41:53 -0400718 btrfs_put_ordered_extent(ordered);
Chris Masonba1da2f2008-07-17 12:54:15 -0400719 return ret;
720}
721
Chris Masonf4219502008-07-22 11:18:09 -0400722
Chris Mason5a3f23d2009-03-31 13:27:11 -0400723/*
724 * add a given inode to the list of inodes that must be fully on
725 * disk before a transaction commit finishes.
726 *
727 * This basically gives us the ext3 style data=ordered mode, and it is mostly
728 * used to make sure renamed files are fully on disk.
729 *
730 * It is a noop if the inode is already fully on disk.
731 *
732 * If trans is not null, we'll do a friendly check for a transaction that
733 * is already flushing things and force the IO down ourselves.
734 */
735int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
736 struct btrfs_root *root,
737 struct inode *inode)
738{
739 u64 last_mod;
740
741 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
742
743 /*
744 * if this file hasn't been changed since the last transaction
745 * commit, we can safely return without doing anything
746 */
747 if (last_mod < root->fs_info->last_trans_committed)
748 return 0;
749
750 /*
751 * the transaction is already committing. Just start the IO and
752 * don't bother with all of this list nonsense
753 */
754 if (trans && root->fs_info->running_transaction->blocked) {
755 btrfs_wait_ordered_range(inode, 0, (u64)-1);
756 return 0;
757 }
758
759 spin_lock(&root->fs_info->ordered_extent_lock);
760 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
761 list_add_tail(&BTRFS_I(inode)->ordered_operations,
762 &root->fs_info->ordered_operations);
763 }
764 spin_unlock(&root->fs_info->ordered_extent_lock);
765
766 return 0;
767}