blob: 9e138cdc36c5eb7d66bf80dfc37829878eeaa6e2 [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
Chris Masondc17ff82008-01-08 15:46:30 -050019#include <linux/slab.h>
Chris Masond6bfde82008-04-30 13:59:35 -040020#include <linux/blkdev.h>
Chris Masonf4219502008-07-22 11:18:09 -040021#include <linux/writeback.h>
22#include <linux/pagevec.h>
Chris Masondc17ff82008-01-08 15:46:30 -050023#include "ctree.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040026#include "extent_io.h"
Chris Masondc17ff82008-01-08 15:46:30 -050027
Chris Masone6dcd2d2008-07-17 12:53:50 -040028static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050029{
Chris Masone6dcd2d2008-07-17 12:53:50 -040030 if (entry->file_offset + entry->len < entry->file_offset)
31 return (u64)-1;
32 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050033}
34
Chris Masond352ac62008-09-29 15:18:18 -040035/* returns NULL if the insertion worked, or it returns the node it did find
36 * in the tree
37 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040038static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
39 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050040{
Chris Masond3977122009-01-05 21:25:51 -050041 struct rb_node **p = &root->rb_node;
42 struct rb_node *parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040043 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050044
Chris Masond3977122009-01-05 21:25:51 -050045 while (*p) {
Chris Masondc17ff82008-01-08 15:46:30 -050046 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040047 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050048
Chris Masone6dcd2d2008-07-17 12:53:50 -040049 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050050 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040051 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050052 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
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040062static void ordered_data_tree_panic(struct inode *inode, int errno,
63 u64 offset)
64{
65 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
66 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
67 "%llu\n", (unsigned long long)offset);
68}
69
Chris Masond352ac62008-09-29 15:18:18 -040070/*
71 * look for a given offset in the tree, and if it can't be found return the
72 * first lesser offset
73 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040074static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
75 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050076{
Chris Masond3977122009-01-05 21:25:51 -050077 struct rb_node *n = root->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -050078 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040079 struct rb_node *test;
80 struct btrfs_ordered_extent *entry;
81 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050082
Chris Masond3977122009-01-05 21:25:51 -050083 while (n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040084 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050085 prev = n;
86 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050087
Chris Masone6dcd2d2008-07-17 12:53:50 -040088 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050089 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040090 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050091 n = n->rb_right;
92 else
93 return n;
94 }
95 if (!prev_ret)
96 return NULL;
97
Chris Masond3977122009-01-05 21:25:51 -050098 while (prev && file_offset >= entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040099 test = rb_next(prev);
100 if (!test)
101 break;
102 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
103 rb_node);
104 if (file_offset < entry_end(prev_entry))
105 break;
106
107 prev = test;
108 }
109 if (prev)
110 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
111 rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500112 while (prev && file_offset < entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400113 test = rb_prev(prev);
114 if (!test)
115 break;
116 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
117 rb_node);
118 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500119 }
120 *prev_ret = prev;
121 return NULL;
122}
123
Chris Masond352ac62008-09-29 15:18:18 -0400124/*
125 * helper to check if a given offset is inside a given entry
126 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400127static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500128{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400129 if (file_offset < entry->file_offset ||
130 entry->file_offset + entry->len <= file_offset)
131 return 0;
132 return 1;
133}
134
Josef Bacik4b46fce2010-05-23 11:00:55 -0400135static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
136 u64 len)
137{
138 if (file_offset + len <= entry->file_offset ||
139 entry->file_offset + entry->len <= file_offset)
140 return 0;
141 return 1;
142}
143
Chris Masond352ac62008-09-29 15:18:18 -0400144/*
145 * look find the first ordered struct that has this offset, otherwise
146 * the first one less than this offset
147 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400148static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
149 u64 file_offset)
150{
151 struct rb_root *root = &tree->tree;
Chris Masonc87fb6f2011-01-31 19:54:59 -0500152 struct rb_node *prev = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -0500153 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400154 struct btrfs_ordered_extent *entry;
155
156 if (tree->last) {
157 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
158 rb_node);
159 if (offset_in_entry(entry, file_offset))
160 return tree->last;
161 }
162 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500163 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400164 ret = prev;
165 if (ret)
166 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500167 return ret;
168}
169
Chris Masoneb84ae02008-07-17 13:53:27 -0400170/* allocate and add a new ordered_extent into the per-inode tree.
171 * file_offset is the logical offset in the file
172 *
173 * start is the disk block number of an extent already reserved in the
174 * extent allocation tree
175 *
176 * len is the length of the extent
177 *
Chris Masoneb84ae02008-07-17 13:53:27 -0400178 * The tree is given a single reference on the ordered extent that was
179 * inserted.
180 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400181static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
182 u64 start, u64 len, u64 disk_len,
Li Zefan261507a02010-12-17 14:21:50 +0800183 int type, int dio, int compress_type)
Chris Masondc17ff82008-01-08 15:46:30 -0500184{
Chris Masondc17ff82008-01-08 15:46:30 -0500185 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400186 struct rb_node *node;
187 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500188
Chris Masone6dcd2d2008-07-17 12:53:50 -0400189 tree = &BTRFS_I(inode)->ordered_tree;
190 entry = kzalloc(sizeof(*entry), GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500191 if (!entry)
192 return -ENOMEM;
193
Chris Masone6dcd2d2008-07-17 12:53:50 -0400194 entry->file_offset = file_offset;
195 entry->start = start;
196 entry->len = len;
Chris Masonc8b97812008-10-29 14:49:59 -0400197 entry->disk_len = disk_len;
Chris Mason8b62b722009-09-02 16:53:46 -0400198 entry->bytes_left = len;
Josef Bacik5fd02042012-05-02 14:00:54 -0400199 entry->inode = igrab(inode);
Li Zefan261507a02010-12-17 14:21:50 +0800200 entry->compress_type = compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -0400201 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400202 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400203
Josef Bacik4b46fce2010-05-23 11:00:55 -0400204 if (dio)
205 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
206
Chris Masone6dcd2d2008-07-17 12:53:50 -0400207 /* one ref for the tree */
208 atomic_set(&entry->refs, 1);
209 init_waitqueue_head(&entry->wait);
210 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400211 INIT_LIST_HEAD(&entry->root_extent_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500212
liubo1abe9b82011-03-24 11:18:59 +0000213 trace_btrfs_ordered_extent_add(inode, entry);
214
Josef Bacik5fd02042012-05-02 14:00:54 -0400215 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400216 node = tree_insert(&tree->tree, file_offset,
217 &entry->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400218 if (node)
219 ordered_data_tree_panic(inode, -EEXIST, file_offset);
Josef Bacik5fd02042012-05-02 14:00:54 -0400220 spin_unlock_irq(&tree->lock);
Chris Masond3977122009-01-05 21:25:51 -0500221
Chris Mason3eaa2882008-07-24 11:57:52 -0400222 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
223 list_add_tail(&entry->root_extent_list,
224 &BTRFS_I(inode)->root->fs_info->ordered_extents);
225 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
226
Chris Masone6dcd2d2008-07-17 12:53:50 -0400227 return 0;
228}
Chris Masondc17ff82008-01-08 15:46:30 -0500229
Josef Bacik4b46fce2010-05-23 11:00:55 -0400230int btrfs_add_ordered_extent(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,
Li Zefan261507a02010-12-17 14:21:50 +0800234 disk_len, type, 0,
235 BTRFS_COMPRESS_NONE);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400236}
237
238int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
239 u64 start, u64 len, u64 disk_len, int type)
240{
241 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800242 disk_len, type, 1,
243 BTRFS_COMPRESS_NONE);
244}
245
246int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
247 u64 start, u64 len, u64 disk_len,
248 int type, int compress_type)
249{
250 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
251 disk_len, type, 0,
252 compress_type);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400253}
254
Chris Masoneb84ae02008-07-17 13:53:27 -0400255/*
256 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400257 * when an ordered extent is finished. If the list covers more than one
258 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400259 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100260void btrfs_add_ordered_sum(struct inode *inode,
261 struct btrfs_ordered_extent *entry,
262 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400263{
264 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400265
Chris Masone6dcd2d2008-07-17 12:53:50 -0400266 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400267 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400268 list_add_tail(&sum->list, &entry->list);
Josef Bacik5fd02042012-05-02 14:00:54 -0400269 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400270}
271
Chris Masoneb84ae02008-07-17 13:53:27 -0400272/*
273 * this is used to account for finished IO across a given range
Chris Mason163cf092010-11-28 19:56:33 -0500274 * of the file. The IO may span ordered extents. If
275 * a given ordered_extent is completely done, 1 is returned, otherwise
276 * 0.
277 *
278 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
279 * to make sure this function only returns 1 once for a given ordered extent.
280 *
281 * file_offset is updated to one byte past the range that is recorded as
282 * complete. This allows you to walk forward in the file.
283 */
284int btrfs_dec_test_first_ordered_pending(struct inode *inode,
285 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400286 u64 *file_offset, u64 io_size, int uptodate)
Chris Mason163cf092010-11-28 19:56:33 -0500287{
288 struct btrfs_ordered_inode_tree *tree;
289 struct rb_node *node;
290 struct btrfs_ordered_extent *entry = NULL;
291 int ret;
Josef Bacik5fd02042012-05-02 14:00:54 -0400292 unsigned long flags;
Chris Mason163cf092010-11-28 19:56:33 -0500293 u64 dec_end;
294 u64 dec_start;
295 u64 to_dec;
296
297 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400298 spin_lock_irqsave(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500299 node = tree_search(tree, *file_offset);
300 if (!node) {
301 ret = 1;
302 goto out;
303 }
304
305 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
306 if (!offset_in_entry(entry, *file_offset)) {
307 ret = 1;
308 goto out;
309 }
310
311 dec_start = max(*file_offset, entry->file_offset);
312 dec_end = min(*file_offset + io_size, entry->file_offset +
313 entry->len);
314 *file_offset = dec_end;
315 if (dec_start > dec_end) {
316 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
317 (unsigned long long)dec_start,
318 (unsigned long long)dec_end);
319 }
320 to_dec = dec_end - dec_start;
321 if (to_dec > entry->bytes_left) {
322 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
323 (unsigned long long)entry->bytes_left,
324 (unsigned long long)to_dec);
325 }
326 entry->bytes_left -= to_dec;
Josef Bacik5fd02042012-05-02 14:00:54 -0400327 if (!uptodate)
328 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
329
Chris Mason163cf092010-11-28 19:56:33 -0500330 if (entry->bytes_left == 0)
331 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
332 else
333 ret = 1;
334out:
335 if (!ret && cached && entry) {
336 *cached = entry;
337 atomic_inc(&entry->refs);
338 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400339 spin_unlock_irqrestore(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500340 return ret == 0;
341}
342
343/*
344 * this is used to account for finished IO across a given range
Chris Masoneb84ae02008-07-17 13:53:27 -0400345 * of the file. The IO should not span ordered extents. If
346 * a given ordered_extent is completely done, 1 is returned, otherwise
347 * 0.
348 *
349 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
350 * to make sure this function only returns 1 once for a given ordered extent.
351 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400352int btrfs_dec_test_ordered_pending(struct inode *inode,
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000353 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400354 u64 file_offset, u64 io_size, int uptodate)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400355{
356 struct btrfs_ordered_inode_tree *tree;
357 struct rb_node *node;
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000358 struct btrfs_ordered_extent *entry = NULL;
Josef Bacik5fd02042012-05-02 14:00:54 -0400359 unsigned long flags;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400360 int ret;
361
362 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400363 spin_lock_irqsave(&tree->lock, flags);
364 if (cached && *cached) {
365 entry = *cached;
366 goto have_entry;
367 }
368
Chris Masone6dcd2d2008-07-17 12:53:50 -0400369 node = tree_search(tree, file_offset);
370 if (!node) {
371 ret = 1;
372 goto out;
373 }
374
375 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400376have_entry:
Chris Masone6dcd2d2008-07-17 12:53:50 -0400377 if (!offset_in_entry(entry, file_offset)) {
378 ret = 1;
379 goto out;
380 }
381
Chris Mason8b62b722009-09-02 16:53:46 -0400382 if (io_size > entry->bytes_left) {
383 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
384 (unsigned long long)entry->bytes_left,
385 (unsigned long long)io_size);
386 }
387 entry->bytes_left -= io_size;
Josef Bacik5fd02042012-05-02 14:00:54 -0400388 if (!uptodate)
389 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
390
Chris Mason8b62b722009-09-02 16:53:46 -0400391 if (entry->bytes_left == 0)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400392 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
Chris Mason8b62b722009-09-02 16:53:46 -0400393 else
394 ret = 1;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400395out:
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000396 if (!ret && cached && entry) {
397 *cached = entry;
398 atomic_inc(&entry->refs);
399 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400400 spin_unlock_irqrestore(&tree->lock, flags);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400401 return ret == 0;
402}
403
Chris Masoneb84ae02008-07-17 13:53:27 -0400404/*
405 * used to drop a reference on an ordered extent. This will free
406 * the extent if the last reference is dropped
407 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100408void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400409{
Chris Masonba1da2f2008-07-17 12:54:15 -0400410 struct list_head *cur;
411 struct btrfs_ordered_sum *sum;
412
liubo1abe9b82011-03-24 11:18:59 +0000413 trace_btrfs_ordered_extent_put(entry->inode, entry);
414
Chris Masonba1da2f2008-07-17 12:54:15 -0400415 if (atomic_dec_and_test(&entry->refs)) {
Josef Bacik5fd02042012-05-02 14:00:54 -0400416 if (entry->inode)
417 btrfs_add_delayed_iput(entry->inode);
Chris Masond3977122009-01-05 21:25:51 -0500418 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400419 cur = entry->list.next;
420 sum = list_entry(cur, struct btrfs_ordered_sum, list);
421 list_del(&sum->list);
422 kfree(sum);
423 }
Chris Masondc17ff82008-01-08 15:46:30 -0500424 kfree(entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400425 }
Chris Masondc17ff82008-01-08 15:46:30 -0500426}
427
Chris Masoneb84ae02008-07-17 13:53:27 -0400428/*
429 * remove an ordered extent from the tree. No references are dropped
Josef Bacik5fd02042012-05-02 14:00:54 -0400430 * and waiters are woken up.
Chris Masoneb84ae02008-07-17 13:53:27 -0400431 */
Josef Bacik5fd02042012-05-02 14:00:54 -0400432void btrfs_remove_ordered_extent(struct inode *inode,
433 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500434{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400435 struct btrfs_ordered_inode_tree *tree;
Josef Bacik287a0ab2010-03-19 18:07:23 +0000436 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500437 struct rb_node *node;
438
Chris Masone6dcd2d2008-07-17 12:53:50 -0400439 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400440 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400441 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500442 rb_erase(node, &tree->tree);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400443 tree->last = NULL;
444 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Josef Bacik5fd02042012-05-02 14:00:54 -0400445 spin_unlock_irq(&tree->lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400446
Josef Bacik287a0ab2010-03-19 18:07:23 +0000447 spin_lock(&root->fs_info->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400448 list_del_init(&entry->root_extent_list);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400449
liubo1abe9b82011-03-24 11:18:59 +0000450 trace_btrfs_ordered_extent_remove(inode, entry);
451
Chris Mason5a3f23d2009-03-31 13:27:11 -0400452 /*
453 * we have no more ordered extents for this inode and
454 * no dirty pages. We can safely remove it from the
455 * list of ordered extents
456 */
457 if (RB_EMPTY_ROOT(&tree->tree) &&
458 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
459 list_del_init(&BTRFS_I(inode)->ordered_operations);
460 }
Josef Bacik287a0ab2010-03-19 18:07:23 +0000461 spin_unlock(&root->fs_info->ordered_extent_lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400462 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400463}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400464
Chris Masond352ac62008-09-29 15:18:18 -0400465/*
466 * wait for all the ordered extents in a root. This is done when balancing
467 * space between drives.
468 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100469void btrfs_wait_ordered_extents(struct btrfs_root *root,
470 int nocow_only, int delay_iput)
Chris Mason3eaa2882008-07-24 11:57:52 -0400471{
472 struct list_head splice;
473 struct list_head *cur;
474 struct btrfs_ordered_extent *ordered;
475 struct inode *inode;
476
477 INIT_LIST_HEAD(&splice);
478
479 spin_lock(&root->fs_info->ordered_extent_lock);
480 list_splice_init(&root->fs_info->ordered_extents, &splice);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400481 while (!list_empty(&splice)) {
Chris Mason3eaa2882008-07-24 11:57:52 -0400482 cur = splice.next;
483 ordered = list_entry(cur, struct btrfs_ordered_extent,
484 root_extent_list);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400485 if (nocow_only &&
Yan Zhengd899e052008-10-30 14:25:28 -0400486 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
487 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400488 list_move(&ordered->root_extent_list,
489 &root->fs_info->ordered_extents);
Yan Zheng7ea394f2008-08-05 13:05:02 -0400490 cond_resched_lock(&root->fs_info->ordered_extent_lock);
491 continue;
492 }
493
Chris Mason3eaa2882008-07-24 11:57:52 -0400494 list_del_init(&ordered->root_extent_list);
495 atomic_inc(&ordered->refs);
Chris Mason3eaa2882008-07-24 11:57:52 -0400496
497 /*
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400498 * the inode may be getting freed (in sys_unlink path).
Chris Mason3eaa2882008-07-24 11:57:52 -0400499 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400500 inode = igrab(ordered->inode);
501
Chris Mason3eaa2882008-07-24 11:57:52 -0400502 spin_unlock(&root->fs_info->ordered_extent_lock);
503
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400504 if (inode) {
505 btrfs_start_ordered_extent(inode, ordered, 1);
506 btrfs_put_ordered_extent(ordered);
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000507 if (delay_iput)
508 btrfs_add_delayed_iput(inode);
509 else
510 iput(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400511 } else {
512 btrfs_put_ordered_extent(ordered);
513 }
Chris Mason3eaa2882008-07-24 11:57:52 -0400514
515 spin_lock(&root->fs_info->ordered_extent_lock);
516 }
517 spin_unlock(&root->fs_info->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400518}
519
Chris Masoneb84ae02008-07-17 13:53:27 -0400520/*
Chris Mason5a3f23d2009-03-31 13:27:11 -0400521 * this is used during transaction commit to write all the inodes
522 * added to the ordered operation list. These files must be fully on
523 * disk before the transaction commits.
524 *
525 * we have two modes here, one is to just start the IO via filemap_flush
526 * and the other is to wait for all the io. When we wait, we have an
527 * extra check to make sure the ordered operation list really is empty
528 * before we return
529 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100530void btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
Chris Mason5a3f23d2009-03-31 13:27:11 -0400531{
532 struct btrfs_inode *btrfs_inode;
533 struct inode *inode;
534 struct list_head splice;
535
536 INIT_LIST_HEAD(&splice);
537
538 mutex_lock(&root->fs_info->ordered_operations_mutex);
539 spin_lock(&root->fs_info->ordered_extent_lock);
540again:
541 list_splice_init(&root->fs_info->ordered_operations, &splice);
542
543 while (!list_empty(&splice)) {
544 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
545 ordered_operations);
546
547 inode = &btrfs_inode->vfs_inode;
548
549 list_del_init(&btrfs_inode->ordered_operations);
550
551 /*
552 * the inode may be getting freed (in sys_unlink path).
553 */
554 inode = igrab(inode);
555
556 if (!wait && inode) {
557 list_add_tail(&BTRFS_I(inode)->ordered_operations,
558 &root->fs_info->ordered_operations);
559 }
560 spin_unlock(&root->fs_info->ordered_extent_lock);
561
562 if (inode) {
563 if (wait)
564 btrfs_wait_ordered_range(inode, 0, (u64)-1);
565 else
566 filemap_flush(inode->i_mapping);
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000567 btrfs_add_delayed_iput(inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400568 }
569
570 cond_resched();
571 spin_lock(&root->fs_info->ordered_extent_lock);
572 }
573 if (wait && !list_empty(&root->fs_info->ordered_operations))
574 goto again;
575
576 spin_unlock(&root->fs_info->ordered_extent_lock);
577 mutex_unlock(&root->fs_info->ordered_operations_mutex);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400578}
579
580/*
Chris Masoneb84ae02008-07-17 13:53:27 -0400581 * 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)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400590{
591 u64 start = entry->file_offset;
592 u64 end = start + entry->len - 1;
593
liubo1abe9b82011-03-24 11:18:59 +0000594 trace_btrfs_ordered_extent_start(inode, entry);
595
Chris Masoneb84ae02008-07-17 13:53:27 -0400596 /*
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 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400601 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
602 filemap_fdatawrite_range(inode->i_mapping, start, end);
Chris Masonc8b97812008-10-29 14:49:59 -0400603 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400604 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
605 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400606 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400607}
608
Chris Masoneb84ae02008-07-17 13:53:27 -0400609/*
610 * Used to wait on ordered extents across a large range of bytes.
611 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100612void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400613{
614 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400615 u64 orig_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400616 struct btrfs_ordered_extent *ordered;
Chris Mason8b62b722009-09-02 16:53:46 -0400617 int found;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400618
Chris Masone5a22172008-07-18 20:42:20 -0400619 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400620 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400621 } else {
622 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400623 if (orig_end > INT_LIMIT(loff_t))
624 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400625 }
Josef Bacik551ebb22012-04-23 14:41:09 -0400626
Chris Masone5a22172008-07-18 20:42:20 -0400627 /* start IO across the range first to instantiate any delalloc
628 * extents
629 */
Josef Bacik551ebb22012-04-23 14:41:09 -0400630 filemap_write_and_wait_range(inode->i_mapping, start, orig_end);
Chris Masonf4219502008-07-22 11:18:09 -0400631
632 end = orig_end;
Chris Mason8b62b722009-09-02 16:53:46 -0400633 found = 0;
Chris Masond3977122009-01-05 21:25:51 -0500634 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400635 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500636 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400637 break;
Chris Masone5a22172008-07-18 20:42:20 -0400638 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400639 btrfs_put_ordered_extent(ordered);
640 break;
641 }
642 if (ordered->file_offset + ordered->len < start) {
643 btrfs_put_ordered_extent(ordered);
644 break;
645 }
Chris Mason8b62b722009-09-02 16:53:46 -0400646 found++;
Chris Masone5a22172008-07-18 20:42:20 -0400647 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400648 end = ordered->file_offset;
649 btrfs_put_ordered_extent(ordered);
Chris Masone5a22172008-07-18 20:42:20 -0400650 if (end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400651 break;
652 end--;
653 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400654}
655
Chris Masoneb84ae02008-07-17 13:53:27 -0400656/*
657 * find an ordered extent corresponding to file_offset. return NULL if
658 * nothing is found, otherwise take a reference on the extent and return it
659 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400660struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
661 u64 file_offset)
662{
663 struct btrfs_ordered_inode_tree *tree;
664 struct rb_node *node;
665 struct btrfs_ordered_extent *entry = NULL;
666
667 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400668 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400669 node = tree_search(tree, file_offset);
670 if (!node)
671 goto out;
672
673 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
674 if (!offset_in_entry(entry, file_offset))
675 entry = NULL;
676 if (entry)
677 atomic_inc(&entry->refs);
678out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400679 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400680 return entry;
681}
682
Josef Bacik4b46fce2010-05-23 11:00:55 -0400683/* Since the DIO code tries to lock a wide area we need to look for any ordered
684 * extents that exist in the range, rather than just the start of the range.
685 */
686struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
687 u64 file_offset,
688 u64 len)
689{
690 struct btrfs_ordered_inode_tree *tree;
691 struct rb_node *node;
692 struct btrfs_ordered_extent *entry = NULL;
693
694 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400695 spin_lock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400696 node = tree_search(tree, file_offset);
697 if (!node) {
698 node = tree_search(tree, file_offset + len);
699 if (!node)
700 goto out;
701 }
702
703 while (1) {
704 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
705 if (range_overlaps(entry, file_offset, len))
706 break;
707
708 if (entry->file_offset >= file_offset + len) {
709 entry = NULL;
710 break;
711 }
712 entry = NULL;
713 node = rb_next(node);
714 if (!node)
715 break;
716 }
717out:
718 if (entry)
719 atomic_inc(&entry->refs);
Josef Bacik5fd02042012-05-02 14:00:54 -0400720 spin_unlock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400721 return entry;
722}
723
Chris Masoneb84ae02008-07-17 13:53:27 -0400724/*
725 * lookup and return any extent before 'file_offset'. NULL is returned
726 * if none is found
727 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400728struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500729btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400730{
731 struct btrfs_ordered_inode_tree *tree;
732 struct rb_node *node;
733 struct btrfs_ordered_extent *entry = NULL;
734
735 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400736 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400737 node = tree_search(tree, file_offset);
738 if (!node)
739 goto out;
740
741 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
742 atomic_inc(&entry->refs);
743out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400744 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400745 return entry;
746}
Chris Masondbe674a2008-07-17 12:54:05 -0400747
Chris Masoneb84ae02008-07-17 13:53:27 -0400748/*
749 * After an extent is done, call this to conditionally update the on disk
750 * i_size. i_size is updated to cover any fully written part of the file.
751 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000752int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
Chris Masondbe674a2008-07-17 12:54:05 -0400753 struct btrfs_ordered_extent *ordered)
754{
755 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Masondbe674a2008-07-17 12:54:05 -0400756 u64 disk_i_size;
757 u64 new_i_size;
758 u64 i_size_test;
Yan, Zhengc2167752009-11-12 09:34:21 +0000759 u64 i_size = i_size_read(inode);
Chris Masondbe674a2008-07-17 12:54:05 -0400760 struct rb_node *node;
Yan, Zhengc2167752009-11-12 09:34:21 +0000761 struct rb_node *prev = NULL;
Chris Masondbe674a2008-07-17 12:54:05 -0400762 struct btrfs_ordered_extent *test;
Yan, Zhengc2167752009-11-12 09:34:21 +0000763 int ret = 1;
764
765 if (ordered)
766 offset = entry_end(ordered);
Yan, Zhenga038fab02009-12-28 05:01:58 +0000767 else
768 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
Chris Masondbe674a2008-07-17 12:54:05 -0400769
Josef Bacik5fd02042012-05-02 14:00:54 -0400770 spin_lock_irq(&tree->lock);
Chris Masondbe674a2008-07-17 12:54:05 -0400771 disk_i_size = BTRFS_I(inode)->disk_i_size;
772
Yan, Zhengc2167752009-11-12 09:34:21 +0000773 /* truncate file */
774 if (disk_i_size > i_size) {
775 BTRFS_I(inode)->disk_i_size = i_size;
776 ret = 0;
777 goto out;
778 }
779
Chris Masondbe674a2008-07-17 12:54:05 -0400780 /*
781 * if the disk i_size is already at the inode->i_size, or
782 * this ordered extent is inside the disk i_size, we're done
783 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000784 if (disk_i_size == i_size || offset <= disk_i_size) {
Chris Masondbe674a2008-07-17 12:54:05 -0400785 goto out;
786 }
787
788 /*
Chris Masondbe674a2008-07-17 12:54:05 -0400789 * walk backward from this ordered extent to disk_i_size.
790 * if we find an ordered extent then we can't update disk i_size
791 * yet
792 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000793 if (ordered) {
794 node = rb_prev(&ordered->rb_node);
795 } else {
796 prev = tree_search(tree, offset);
797 /*
798 * we insert file extents without involving ordered struct,
799 * so there should be no ordered struct cover this offset
800 */
801 if (prev) {
802 test = rb_entry(prev, struct btrfs_ordered_extent,
803 rb_node);
804 BUG_ON(offset_in_entry(test, offset));
805 }
806 node = prev;
807 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400808 for (; node; node = rb_prev(node)) {
Chris Masondbe674a2008-07-17 12:54:05 -0400809 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400810
811 /* We treat this entry as if it doesnt exist */
812 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
813 continue;
Chris Masondbe674a2008-07-17 12:54:05 -0400814 if (test->file_offset + test->len <= disk_i_size)
815 break;
Yan, Zhengc2167752009-11-12 09:34:21 +0000816 if (test->file_offset >= i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400817 break;
818 if (test->file_offset >= disk_i_size)
819 goto out;
820 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000821 new_i_size = min_t(u64, offset, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -0400822
823 /*
824 * at this point, we know we can safely update i_size to at least
825 * the offset from this ordered extent. But, we need to
826 * walk forward and see if ios from higher up in the file have
827 * finished.
828 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000829 if (ordered) {
830 node = rb_next(&ordered->rb_node);
831 } else {
832 if (prev)
833 node = rb_next(prev);
834 else
835 node = rb_first(&tree->tree);
836 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400837
838 /*
839 * We are looking for an area between our current extent and the next
840 * ordered extent to update the i_size to. There are 3 cases here
841 *
842 * 1) We don't actually have anything and we can update to i_size.
843 * 2) We have stuff but they already did their i_size update so again we
844 * can just update to i_size.
845 * 3) We have an outstanding ordered extent so the most we can update
846 * our disk_i_size to is the start of the next offset.
847 */
848 i_size_test = i_size;
849 for (; node; node = rb_next(node)) {
Chris Masondbe674a2008-07-17 12:54:05 -0400850 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400851
852 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
853 continue;
854 if (test->file_offset > offset) {
Yan Zhengb48652c2008-08-04 23:23:47 -0400855 i_size_test = test->file_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -0400856 break;
857 }
Chris Masondbe674a2008-07-17 12:54:05 -0400858 }
859
860 /*
861 * i_size_test is the end of a region after this ordered
Josef Bacik4e899152012-05-02 13:52:09 -0400862 * extent where there are no ordered extents, we can safely set
863 * disk_i_size to this.
Chris Masondbe674a2008-07-17 12:54:05 -0400864 */
Josef Bacik4e899152012-05-02 13:52:09 -0400865 if (i_size_test > offset)
Yan, Zhengc2167752009-11-12 09:34:21 +0000866 new_i_size = min_t(u64, i_size_test, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -0400867 BTRFS_I(inode)->disk_i_size = new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +0000868 ret = 0;
Chris Masondbe674a2008-07-17 12:54:05 -0400869out:
Yan, Zhengc2167752009-11-12 09:34:21 +0000870 /*
Josef Bacik5fd02042012-05-02 14:00:54 -0400871 * We need to do this because we can't remove ordered extents until
872 * after the i_disk_size has been updated and then the inode has been
873 * updated to reflect the change, so we need to tell anybody who finds
874 * this ordered extent that we've already done all the real work, we
875 * just haven't completed all the other work.
Yan, Zhengc2167752009-11-12 09:34:21 +0000876 */
877 if (ordered)
Josef Bacik5fd02042012-05-02 14:00:54 -0400878 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
879 spin_unlock_irq(&tree->lock);
Yan, Zhengc2167752009-11-12 09:34:21 +0000880 return ret;
Chris Masondbe674a2008-07-17 12:54:05 -0400881}
Chris Masonba1da2f2008-07-17 12:54:15 -0400882
Chris Masoneb84ae02008-07-17 13:53:27 -0400883/*
884 * search the ordered extents for one corresponding to 'offset' and
885 * try to find a checksum. This is used because we allow pages to
886 * be reclaimed before their checksum is actually put into the btree
887 */
Chris Masond20f7042008-12-08 16:58:54 -0500888int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
889 u32 *sum)
Chris Masonba1da2f2008-07-17 12:54:15 -0400890{
891 struct btrfs_ordered_sum *ordered_sum;
892 struct btrfs_sector_sum *sector_sums;
893 struct btrfs_ordered_extent *ordered;
894 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -0400895 unsigned long num_sectors;
896 unsigned long i;
897 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -0400898 int ret = 1;
Chris Masonba1da2f2008-07-17 12:54:15 -0400899
900 ordered = btrfs_lookup_ordered_extent(inode, offset);
901 if (!ordered)
902 return 1;
903
Josef Bacik5fd02042012-05-02 14:00:54 -0400904 spin_lock_irq(&tree->lock);
Qinghuang Fengc6e30872009-01-21 10:59:08 -0500905 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Chris Masond20f7042008-12-08 16:58:54 -0500906 if (disk_bytenr >= ordered_sum->bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400907 num_sectors = ordered_sum->len / sectorsize;
Chris Masoned98b562008-07-22 23:06:42 -0400908 sector_sums = ordered_sum->sums;
Chris Mason3edf7d32008-07-18 06:17:13 -0400909 for (i = 0; i < num_sectors; i++) {
Chris Masond20f7042008-12-08 16:58:54 -0500910 if (sector_sums[i].bytenr == disk_bytenr) {
Chris Mason3edf7d32008-07-18 06:17:13 -0400911 *sum = sector_sums[i].sum;
912 ret = 0;
913 goto out;
914 }
915 }
Chris Masonba1da2f2008-07-17 12:54:15 -0400916 }
917 }
918out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400919 spin_unlock_irq(&tree->lock);
Chris Mason89642222008-07-24 09:41:53 -0400920 btrfs_put_ordered_extent(ordered);
Chris Masonba1da2f2008-07-17 12:54:15 -0400921 return ret;
922}
923
Chris Masonf4219502008-07-22 11:18:09 -0400924
Chris Mason5a3f23d2009-03-31 13:27:11 -0400925/*
926 * add a given inode to the list of inodes that must be fully on
927 * disk before a transaction commit finishes.
928 *
929 * This basically gives us the ext3 style data=ordered mode, and it is mostly
930 * used to make sure renamed files are fully on disk.
931 *
932 * It is a noop if the inode is already fully on disk.
933 *
934 * If trans is not null, we'll do a friendly check for a transaction that
935 * is already flushing things and force the IO down ourselves.
936 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100937void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
938 struct btrfs_root *root, struct inode *inode)
Chris Mason5a3f23d2009-03-31 13:27:11 -0400939{
940 u64 last_mod;
941
942 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
943
944 /*
945 * if this file hasn't been changed since the last transaction
946 * commit, we can safely return without doing anything
947 */
948 if (last_mod < root->fs_info->last_trans_committed)
Jeff Mahoney143bede2012-03-01 14:56:26 +0100949 return;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400950
951 /*
952 * the transaction is already committing. Just start the IO and
953 * don't bother with all of this list nonsense
954 */
955 if (trans && root->fs_info->running_transaction->blocked) {
956 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100957 return;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400958 }
959
960 spin_lock(&root->fs_info->ordered_extent_lock);
961 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
962 list_add_tail(&BTRFS_I(inode)->ordered_operations,
963 &root->fs_info->ordered_operations);
964 }
965 spin_unlock(&root->fs_info->ordered_extent_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400966}