blob: e12441c7cf1d63ed8e7c22039187cf7d4a64e148 [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"
Miao Xie199c2a92013-05-15 07:48:23 +000027#include "disk-io.h"
Chris Masondc17ff82008-01-08 15:46:30 -050028
Miao Xie6352b912012-09-06 04:01:51 -060029static struct kmem_cache *btrfs_ordered_extent_cache;
30
Chris Masone6dcd2d2008-07-17 12:53:50 -040031static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050032{
Chris Masone6dcd2d2008-07-17 12:53:50 -040033 if (entry->file_offset + entry->len < entry->file_offset)
34 return (u64)-1;
35 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050036}
37
Chris Masond352ac62008-09-29 15:18:18 -040038/* returns NULL if the insertion worked, or it returns the node it did find
39 * in the tree
40 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040041static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
42 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050043{
Chris Masond3977122009-01-05 21:25:51 -050044 struct rb_node **p = &root->rb_node;
45 struct rb_node *parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040046 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050047
Chris Masond3977122009-01-05 21:25:51 -050048 while (*p) {
Chris Masondc17ff82008-01-08 15:46:30 -050049 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040050 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050051
Chris Masone6dcd2d2008-07-17 12:53:50 -040052 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050053 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040054 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050055 p = &(*p)->rb_right;
56 else
57 return parent;
58 }
59
60 rb_link_node(node, parent, p);
61 rb_insert_color(node, root);
62 return NULL;
63}
64
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040065static void ordered_data_tree_panic(struct inode *inode, int errno,
66 u64 offset)
67{
68 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
69 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
David Sterba351fd352014-05-15 16:48:20 +020070 "%llu", offset);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040071}
72
Chris Masond352ac62008-09-29 15:18:18 -040073/*
74 * look for a given offset in the tree, and if it can't be found return the
75 * first lesser offset
76 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040077static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
78 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050079{
Chris Masond3977122009-01-05 21:25:51 -050080 struct rb_node *n = root->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -050081 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040082 struct rb_node *test;
83 struct btrfs_ordered_extent *entry;
84 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050085
Chris Masond3977122009-01-05 21:25:51 -050086 while (n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040087 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050088 prev = n;
89 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050090
Chris Masone6dcd2d2008-07-17 12:53:50 -040091 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050092 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040093 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050094 n = n->rb_right;
95 else
96 return n;
97 }
98 if (!prev_ret)
99 return NULL;
100
Chris Masond3977122009-01-05 21:25:51 -0500101 while (prev && file_offset >= entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400102 test = rb_next(prev);
103 if (!test)
104 break;
105 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 rb_node);
107 if (file_offset < entry_end(prev_entry))
108 break;
109
110 prev = test;
111 }
112 if (prev)
113 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
114 rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500115 while (prev && file_offset < entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400116 test = rb_prev(prev);
117 if (!test)
118 break;
119 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
120 rb_node);
121 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500122 }
123 *prev_ret = prev;
124 return NULL;
125}
126
Chris Masond352ac62008-09-29 15:18:18 -0400127/*
128 * helper to check if a given offset is inside a given entry
129 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400130static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500131{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400132 if (file_offset < entry->file_offset ||
133 entry->file_offset + entry->len <= file_offset)
134 return 0;
135 return 1;
136}
137
Josef Bacik4b46fce2010-05-23 11:00:55 -0400138static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
139 u64 len)
140{
141 if (file_offset + len <= entry->file_offset ||
142 entry->file_offset + entry->len <= file_offset)
143 return 0;
144 return 1;
145}
146
Chris Masond352ac62008-09-29 15:18:18 -0400147/*
148 * look find the first ordered struct that has this offset, otherwise
149 * the first one less than this offset
150 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400151static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
152 u64 file_offset)
153{
154 struct rb_root *root = &tree->tree;
Chris Masonc87fb6f2011-01-31 19:54:59 -0500155 struct rb_node *prev = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -0500156 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400157 struct btrfs_ordered_extent *entry;
158
159 if (tree->last) {
160 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
161 rb_node);
162 if (offset_in_entry(entry, file_offset))
163 return tree->last;
164 }
165 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500166 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400167 ret = prev;
168 if (ret)
169 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500170 return ret;
171}
172
Chris Masoneb84ae02008-07-17 13:53:27 -0400173/* allocate and add a new ordered_extent into the per-inode tree.
174 * file_offset is the logical offset in the file
175 *
176 * start is the disk block number of an extent already reserved in the
177 * extent allocation tree
178 *
179 * len is the length of the extent
180 *
Chris Masoneb84ae02008-07-17 13:53:27 -0400181 * The tree is given a single reference on the ordered extent that was
182 * inserted.
183 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400184static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
185 u64 start, u64 len, u64 disk_len,
Li Zefan261507a02010-12-17 14:21:50 +0800186 int type, int dio, int compress_type)
Chris Masondc17ff82008-01-08 15:46:30 -0500187{
Miao Xie199c2a92013-05-15 07:48:23 +0000188 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500189 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400190 struct rb_node *node;
191 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500192
Chris Masone6dcd2d2008-07-17 12:53:50 -0400193 tree = &BTRFS_I(inode)->ordered_tree;
Miao Xie6352b912012-09-06 04:01:51 -0600194 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500195 if (!entry)
196 return -ENOMEM;
197
Chris Masone6dcd2d2008-07-17 12:53:50 -0400198 entry->file_offset = file_offset;
199 entry->start = start;
200 entry->len = len;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400201 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
202 !(type == BTRFS_ORDERED_NOCOW))
203 entry->csum_bytes_left = disk_len;
Chris Masonc8b97812008-10-29 14:49:59 -0400204 entry->disk_len = disk_len;
Chris Mason8b62b722009-09-02 16:53:46 -0400205 entry->bytes_left = len;
Josef Bacik5fd02042012-05-02 14:00:54 -0400206 entry->inode = igrab(inode);
Li Zefan261507a02010-12-17 14:21:50 +0800207 entry->compress_type = compress_type;
Josef Bacik77cef2e2013-08-29 13:57:21 -0400208 entry->truncated_len = (u64)-1;
Yan Zhengd899e052008-10-30 14:25:28 -0400209 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400210 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400211
Josef Bacik4b46fce2010-05-23 11:00:55 -0400212 if (dio)
213 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
214
Chris Masone6dcd2d2008-07-17 12:53:50 -0400215 /* one ref for the tree */
216 atomic_set(&entry->refs, 1);
217 init_waitqueue_head(&entry->wait);
218 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400219 INIT_LIST_HEAD(&entry->root_extent_list);
Miao Xie9afab882012-10-25 09:41:36 +0000220 INIT_LIST_HEAD(&entry->work_list);
221 init_completion(&entry->completion);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400222 INIT_LIST_HEAD(&entry->log_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500223
liubo1abe9b82011-03-24 11:18:59 +0000224 trace_btrfs_ordered_extent_add(inode, entry);
225
Josef Bacik5fd02042012-05-02 14:00:54 -0400226 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400227 node = tree_insert(&tree->tree, file_offset,
228 &entry->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400229 if (node)
230 ordered_data_tree_panic(inode, -EEXIST, file_offset);
Josef Bacik5fd02042012-05-02 14:00:54 -0400231 spin_unlock_irq(&tree->lock);
Chris Masond3977122009-01-05 21:25:51 -0500232
Miao Xie199c2a92013-05-15 07:48:23 +0000233 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400234 list_add_tail(&entry->root_extent_list,
Miao Xie199c2a92013-05-15 07:48:23 +0000235 &root->ordered_extents);
236 root->nr_ordered_extents++;
237 if (root->nr_ordered_extents == 1) {
238 spin_lock(&root->fs_info->ordered_root_lock);
239 BUG_ON(!list_empty(&root->ordered_root));
240 list_add_tail(&root->ordered_root,
241 &root->fs_info->ordered_roots);
242 spin_unlock(&root->fs_info->ordered_root_lock);
243 }
244 spin_unlock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400245
Chris Masone6dcd2d2008-07-17 12:53:50 -0400246 return 0;
247}
Chris Masondc17ff82008-01-08 15:46:30 -0500248
Josef Bacik4b46fce2010-05-23 11:00:55 -0400249int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
250 u64 start, u64 len, u64 disk_len, int type)
251{
252 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800253 disk_len, type, 0,
254 BTRFS_COMPRESS_NONE);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400255}
256
257int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
258 u64 start, u64 len, u64 disk_len, int type)
259{
260 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800261 disk_len, type, 1,
262 BTRFS_COMPRESS_NONE);
263}
264
265int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
266 u64 start, u64 len, u64 disk_len,
267 int type, int compress_type)
268{
269 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
270 disk_len, type, 0,
271 compress_type);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400272}
273
Chris Masoneb84ae02008-07-17 13:53:27 -0400274/*
275 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400276 * when an ordered extent is finished. If the list covers more than one
277 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400278 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100279void btrfs_add_ordered_sum(struct inode *inode,
280 struct btrfs_ordered_extent *entry,
281 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400282{
283 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400284
Chris Masone6dcd2d2008-07-17 12:53:50 -0400285 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400286 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400287 list_add_tail(&sum->list, &entry->list);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400288 WARN_ON(entry->csum_bytes_left < sum->len);
289 entry->csum_bytes_left -= sum->len;
290 if (entry->csum_bytes_left == 0)
291 wake_up(&entry->wait);
Josef Bacik5fd02042012-05-02 14:00:54 -0400292 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400293}
294
Chris Masoneb84ae02008-07-17 13:53:27 -0400295/*
296 * this is used to account for finished IO across a given range
Chris Mason163cf092010-11-28 19:56:33 -0500297 * of the file. The IO may span ordered extents. If
298 * a given ordered_extent is completely done, 1 is returned, otherwise
299 * 0.
300 *
301 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
302 * to make sure this function only returns 1 once for a given ordered extent.
303 *
304 * file_offset is updated to one byte past the range that is recorded as
305 * complete. This allows you to walk forward in the file.
306 */
307int btrfs_dec_test_first_ordered_pending(struct inode *inode,
308 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400309 u64 *file_offset, u64 io_size, int uptodate)
Chris Mason163cf092010-11-28 19:56:33 -0500310{
311 struct btrfs_ordered_inode_tree *tree;
312 struct rb_node *node;
313 struct btrfs_ordered_extent *entry = NULL;
314 int ret;
Josef Bacik5fd02042012-05-02 14:00:54 -0400315 unsigned long flags;
Chris Mason163cf092010-11-28 19:56:33 -0500316 u64 dec_end;
317 u64 dec_start;
318 u64 to_dec;
319
320 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400321 spin_lock_irqsave(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500322 node = tree_search(tree, *file_offset);
323 if (!node) {
324 ret = 1;
325 goto out;
326 }
327
328 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
329 if (!offset_in_entry(entry, *file_offset)) {
330 ret = 1;
331 goto out;
332 }
333
334 dec_start = max(*file_offset, entry->file_offset);
335 dec_end = min(*file_offset + io_size, entry->file_offset +
336 entry->len);
337 *file_offset = dec_end;
338 if (dec_start > dec_end) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
Chris Mason163cf092010-11-28 19:56:33 -0500341 }
342 to_dec = dec_end - dec_start;
343 if (to_dec > entry->bytes_left) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500344 btrfs_crit(BTRFS_I(inode)->root->fs_info,
345 "bad ordered accounting left %llu size %llu",
346 entry->bytes_left, to_dec);
Chris Mason163cf092010-11-28 19:56:33 -0500347 }
348 entry->bytes_left -= to_dec;
Josef Bacik5fd02042012-05-02 14:00:54 -0400349 if (!uptodate)
350 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
351
Miao Xieaf7a6502014-03-06 13:54:56 +0800352 if (entry->bytes_left == 0) {
Chris Mason163cf092010-11-28 19:56:33 -0500353 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
Miao Xieaf7a6502014-03-06 13:54:56 +0800354 if (waitqueue_active(&entry->wait))
355 wake_up(&entry->wait);
356 } else {
Chris Mason163cf092010-11-28 19:56:33 -0500357 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800358 }
Chris Mason163cf092010-11-28 19:56:33 -0500359out:
360 if (!ret && cached && entry) {
361 *cached = entry;
362 atomic_inc(&entry->refs);
363 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400364 spin_unlock_irqrestore(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500365 return ret == 0;
366}
367
368/*
369 * this is used to account for finished IO across a given range
Chris Masoneb84ae02008-07-17 13:53:27 -0400370 * of the file. The IO should not span ordered extents. If
371 * a given ordered_extent is completely done, 1 is returned, otherwise
372 * 0.
373 *
374 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
375 * to make sure this function only returns 1 once for a given ordered extent.
376 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400377int btrfs_dec_test_ordered_pending(struct inode *inode,
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000378 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400379 u64 file_offset, u64 io_size, int uptodate)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400380{
381 struct btrfs_ordered_inode_tree *tree;
382 struct rb_node *node;
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000383 struct btrfs_ordered_extent *entry = NULL;
Josef Bacik5fd02042012-05-02 14:00:54 -0400384 unsigned long flags;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400385 int ret;
386
387 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400388 spin_lock_irqsave(&tree->lock, flags);
389 if (cached && *cached) {
390 entry = *cached;
391 goto have_entry;
392 }
393
Chris Masone6dcd2d2008-07-17 12:53:50 -0400394 node = tree_search(tree, file_offset);
395 if (!node) {
396 ret = 1;
397 goto out;
398 }
399
400 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400401have_entry:
Chris Masone6dcd2d2008-07-17 12:53:50 -0400402 if (!offset_in_entry(entry, file_offset)) {
403 ret = 1;
404 goto out;
405 }
406
Chris Mason8b62b722009-09-02 16:53:46 -0400407 if (io_size > entry->bytes_left) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500408 btrfs_crit(BTRFS_I(inode)->root->fs_info,
409 "bad ordered accounting left %llu size %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200410 entry->bytes_left, io_size);
Chris Mason8b62b722009-09-02 16:53:46 -0400411 }
412 entry->bytes_left -= io_size;
Josef Bacik5fd02042012-05-02 14:00:54 -0400413 if (!uptodate)
414 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
415
Miao Xieaf7a6502014-03-06 13:54:56 +0800416 if (entry->bytes_left == 0) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400417 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
Miao Xieaf7a6502014-03-06 13:54:56 +0800418 if (waitqueue_active(&entry->wait))
419 wake_up(&entry->wait);
420 } else {
Chris Mason8b62b722009-09-02 16:53:46 -0400421 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800422 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400423out:
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000424 if (!ret && cached && entry) {
425 *cached = entry;
426 atomic_inc(&entry->refs);
427 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400428 spin_unlock_irqrestore(&tree->lock, flags);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400429 return ret == 0;
430}
431
Josef Bacik2ab28f32012-10-12 15:27:49 -0400432/* Needs to either be called under a log transaction or the log_mutex */
Miao Xie827463c2014-01-14 20:31:51 +0800433void btrfs_get_logged_extents(struct inode *inode,
434 struct list_head *logged_list)
Josef Bacik2ab28f32012-10-12 15:27:49 -0400435{
436 struct btrfs_ordered_inode_tree *tree;
437 struct btrfs_ordered_extent *ordered;
438 struct rb_node *n;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400439
440 tree = &BTRFS_I(inode)->ordered_tree;
441 spin_lock_irq(&tree->lock);
442 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
443 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Miao Xie827463c2014-01-14 20:31:51 +0800444 if (!list_empty(&ordered->log_list))
445 continue;
446 list_add_tail(&ordered->log_list, logged_list);
447 atomic_inc(&ordered->refs);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400448 }
449 spin_unlock_irq(&tree->lock);
450}
451
Miao Xie827463c2014-01-14 20:31:51 +0800452void btrfs_put_logged_extents(struct list_head *logged_list)
453{
454 struct btrfs_ordered_extent *ordered;
455
456 while (!list_empty(logged_list)) {
457 ordered = list_first_entry(logged_list,
458 struct btrfs_ordered_extent,
459 log_list);
460 list_del_init(&ordered->log_list);
461 btrfs_put_ordered_extent(ordered);
462 }
463}
464
465void btrfs_submit_logged_extents(struct list_head *logged_list,
466 struct btrfs_root *log)
467{
468 int index = log->log_transid % 2;
469
470 spin_lock_irq(&log->log_extents_lock[index]);
471 list_splice_tail(logged_list, &log->logged_list[index]);
472 spin_unlock_irq(&log->log_extents_lock[index]);
473}
474
Josef Bacik2ab28f32012-10-12 15:27:49 -0400475void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
476{
477 struct btrfs_ordered_extent *ordered;
478 int index = transid % 2;
479
480 spin_lock_irq(&log->log_extents_lock[index]);
481 while (!list_empty(&log->logged_list[index])) {
482 ordered = list_first_entry(&log->logged_list[index],
483 struct btrfs_ordered_extent,
484 log_list);
485 list_del_init(&ordered->log_list);
486 spin_unlock_irq(&log->log_extents_lock[index]);
487 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
488 &ordered->flags));
489 btrfs_put_ordered_extent(ordered);
490 spin_lock_irq(&log->log_extents_lock[index]);
491 }
492 spin_unlock_irq(&log->log_extents_lock[index]);
493}
494
495void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
496{
497 struct btrfs_ordered_extent *ordered;
498 int index = transid % 2;
499
500 spin_lock_irq(&log->log_extents_lock[index]);
501 while (!list_empty(&log->logged_list[index])) {
502 ordered = list_first_entry(&log->logged_list[index],
503 struct btrfs_ordered_extent,
504 log_list);
505 list_del_init(&ordered->log_list);
506 spin_unlock_irq(&log->log_extents_lock[index]);
507 btrfs_put_ordered_extent(ordered);
508 spin_lock_irq(&log->log_extents_lock[index]);
509 }
510 spin_unlock_irq(&log->log_extents_lock[index]);
511}
512
Chris Masoneb84ae02008-07-17 13:53:27 -0400513/*
514 * used to drop a reference on an ordered extent. This will free
515 * the extent if the last reference is dropped
516 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100517void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400518{
Chris Masonba1da2f2008-07-17 12:54:15 -0400519 struct list_head *cur;
520 struct btrfs_ordered_sum *sum;
521
liubo1abe9b82011-03-24 11:18:59 +0000522 trace_btrfs_ordered_extent_put(entry->inode, entry);
523
Chris Masonba1da2f2008-07-17 12:54:15 -0400524 if (atomic_dec_and_test(&entry->refs)) {
Josef Bacik5fd02042012-05-02 14:00:54 -0400525 if (entry->inode)
526 btrfs_add_delayed_iput(entry->inode);
Chris Masond3977122009-01-05 21:25:51 -0500527 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400528 cur = entry->list.next;
529 sum = list_entry(cur, struct btrfs_ordered_sum, list);
530 list_del(&sum->list);
531 kfree(sum);
532 }
Miao Xie6352b912012-09-06 04:01:51 -0600533 kmem_cache_free(btrfs_ordered_extent_cache, entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400534 }
Chris Masondc17ff82008-01-08 15:46:30 -0500535}
536
Chris Masoneb84ae02008-07-17 13:53:27 -0400537/*
538 * remove an ordered extent from the tree. No references are dropped
Josef Bacik5fd02042012-05-02 14:00:54 -0400539 * and waiters are woken up.
Chris Masoneb84ae02008-07-17 13:53:27 -0400540 */
Josef Bacik5fd02042012-05-02 14:00:54 -0400541void btrfs_remove_ordered_extent(struct inode *inode,
542 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500543{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400544 struct btrfs_ordered_inode_tree *tree;
Josef Bacik287a0ab2010-03-19 18:07:23 +0000545 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500546 struct rb_node *node;
547
Chris Masone6dcd2d2008-07-17 12:53:50 -0400548 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400549 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400550 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500551 rb_erase(node, &tree->tree);
Filipe David Borba Manana1b8e7e42013-11-22 18:54:58 +0000552 if (tree->last == node)
553 tree->last = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400554 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Josef Bacik5fd02042012-05-02 14:00:54 -0400555 spin_unlock_irq(&tree->lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400556
Miao Xie199c2a92013-05-15 07:48:23 +0000557 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400558 list_del_init(&entry->root_extent_list);
Miao Xie199c2a92013-05-15 07:48:23 +0000559 root->nr_ordered_extents--;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400560
liubo1abe9b82011-03-24 11:18:59 +0000561 trace_btrfs_ordered_extent_remove(inode, entry);
562
Chris Mason5a3f23d2009-03-31 13:27:11 -0400563 /*
564 * we have no more ordered extents for this inode and
565 * no dirty pages. We can safely remove it from the
566 * list of ordered extents
567 */
568 if (RB_EMPTY_ROOT(&tree->tree) &&
569 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
Josef Bacik93858762013-10-28 09:13:25 -0400570 spin_lock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400571 list_del_init(&BTRFS_I(inode)->ordered_operations);
Josef Bacik93858762013-10-28 09:13:25 -0400572 spin_unlock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400573 }
Miao Xie199c2a92013-05-15 07:48:23 +0000574
575 if (!root->nr_ordered_extents) {
576 spin_lock(&root->fs_info->ordered_root_lock);
577 BUG_ON(list_empty(&root->ordered_root));
578 list_del_init(&root->ordered_root);
579 spin_unlock(&root->fs_info->ordered_root_lock);
580 }
581 spin_unlock(&root->ordered_extent_lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400582 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400583}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400584
Qu Wenruod458b052014-02-28 10:46:19 +0800585static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
Miao Xie9afab882012-10-25 09:41:36 +0000586{
587 struct btrfs_ordered_extent *ordered;
588
589 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
590 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
591 complete(&ordered->completion);
592}
593
Chris Masond352ac62008-09-29 15:18:18 -0400594/*
595 * wait for all the ordered extents in a root. This is done when balancing
596 * space between drives.
597 */
Miao Xie31f3d252014-03-06 13:55:02 +0800598int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
Chris Mason3eaa2882008-07-24 11:57:52 -0400599{
Miao Xie9afab882012-10-25 09:41:36 +0000600 struct list_head splice, works;
Miao Xie9afab882012-10-25 09:41:36 +0000601 struct btrfs_ordered_extent *ordered, *next;
Miao Xieb0244192013-11-04 23:13:25 +0800602 int count = 0;
Chris Mason3eaa2882008-07-24 11:57:52 -0400603
604 INIT_LIST_HEAD(&splice);
Miao Xie9afab882012-10-25 09:41:36 +0000605 INIT_LIST_HEAD(&works);
Chris Mason3eaa2882008-07-24 11:57:52 -0400606
Miao Xie31f3d252014-03-06 13:55:02 +0800607 mutex_lock(&root->ordered_extent_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000608 spin_lock(&root->ordered_extent_lock);
609 list_splice_init(&root->ordered_extents, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800610 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000611 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
612 root_extent_list);
613 list_move_tail(&ordered->root_extent_list,
614 &root->ordered_extents);
Miao Xie199c2a92013-05-15 07:48:23 +0000615 atomic_inc(&ordered->refs);
616 spin_unlock(&root->ordered_extent_lock);
617
Qu Wenruoa44903a2014-02-28 10:46:09 +0800618 btrfs_init_work(&ordered->flush_work,
619 btrfs_run_ordered_extent_work, NULL, NULL);
Miao Xie199c2a92013-05-15 07:48:23 +0000620 list_add_tail(&ordered->work_list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +0800621 btrfs_queue_work(root->fs_info->flush_workers,
622 &ordered->flush_work);
Miao Xie199c2a92013-05-15 07:48:23 +0000623
Miao Xie9afab882012-10-25 09:41:36 +0000624 cond_resched();
Miao Xie199c2a92013-05-15 07:48:23 +0000625 spin_lock(&root->ordered_extent_lock);
Miao Xieb0244192013-11-04 23:13:25 +0800626 if (nr != -1)
627 nr--;
628 count++;
Chris Mason3eaa2882008-07-24 11:57:52 -0400629 }
Miao Xieb0244192013-11-04 23:13:25 +0800630 list_splice_tail(&splice, &root->ordered_extents);
Miao Xie199c2a92013-05-15 07:48:23 +0000631 spin_unlock(&root->ordered_extent_lock);
Miao Xie9afab882012-10-25 09:41:36 +0000632
633 list_for_each_entry_safe(ordered, next, &works, work_list) {
634 list_del_init(&ordered->work_list);
635 wait_for_completion(&ordered->completion);
Miao Xie9afab882012-10-25 09:41:36 +0000636 btrfs_put_ordered_extent(ordered);
Miao Xie9afab882012-10-25 09:41:36 +0000637 cond_resched();
638 }
Miao Xie31f3d252014-03-06 13:55:02 +0800639 mutex_unlock(&root->ordered_extent_mutex);
Miao Xieb0244192013-11-04 23:13:25 +0800640
641 return count;
Chris Mason3eaa2882008-07-24 11:57:52 -0400642}
643
Miao Xieb0244192013-11-04 23:13:25 +0800644void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
Miao Xie199c2a92013-05-15 07:48:23 +0000645{
646 struct btrfs_root *root;
647 struct list_head splice;
Miao Xieb0244192013-11-04 23:13:25 +0800648 int done;
Miao Xie199c2a92013-05-15 07:48:23 +0000649
650 INIT_LIST_HEAD(&splice);
651
Miao Xie8b9d83c2014-03-06 13:54:55 +0800652 mutex_lock(&fs_info->ordered_operations_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000653 spin_lock(&fs_info->ordered_root_lock);
654 list_splice_init(&fs_info->ordered_roots, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800655 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000656 root = list_first_entry(&splice, struct btrfs_root,
657 ordered_root);
658 root = btrfs_grab_fs_root(root);
659 BUG_ON(!root);
660 list_move_tail(&root->ordered_root,
661 &fs_info->ordered_roots);
662 spin_unlock(&fs_info->ordered_root_lock);
663
Miao Xie31f3d252014-03-06 13:55:02 +0800664 done = btrfs_wait_ordered_extents(root, nr);
Miao Xie199c2a92013-05-15 07:48:23 +0000665 btrfs_put_fs_root(root);
666
667 spin_lock(&fs_info->ordered_root_lock);
Miao Xieb0244192013-11-04 23:13:25 +0800668 if (nr != -1) {
669 nr -= done;
670 WARN_ON(nr < 0);
671 }
Miao Xie199c2a92013-05-15 07:48:23 +0000672 }
Miao Xie931aa872013-11-14 17:33:21 +0800673 list_splice_tail(&splice, &fs_info->ordered_roots);
Miao Xie199c2a92013-05-15 07:48:23 +0000674 spin_unlock(&fs_info->ordered_root_lock);
Miao Xie8b9d83c2014-03-06 13:54:55 +0800675 mutex_unlock(&fs_info->ordered_operations_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000676}
677
Chris Masoneb84ae02008-07-17 13:53:27 -0400678/*
Chris Mason5a3f23d2009-03-31 13:27:11 -0400679 * this is used during transaction commit to write all the inodes
680 * added to the ordered operation list. These files must be fully on
681 * disk before the transaction commits.
682 *
683 * we have two modes here, one is to just start the IO via filemap_flush
684 * and the other is to wait for all the io. When we wait, we have an
685 * extra check to make sure the ordered operation list really is empty
686 * before we return
687 */
Josef Bacik569e0f32013-02-13 11:09:14 -0500688int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
689 struct btrfs_root *root, int wait)
Chris Mason5a3f23d2009-03-31 13:27:11 -0400690{
691 struct btrfs_inode *btrfs_inode;
692 struct inode *inode;
Josef Bacik569e0f32013-02-13 11:09:14 -0500693 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400694 struct list_head splice;
Miao Xie25287e02012-10-25 09:31:03 +0000695 struct list_head works;
696 struct btrfs_delalloc_work *work, *next;
697 int ret = 0;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400698
699 INIT_LIST_HEAD(&splice);
Miao Xie25287e02012-10-25 09:31:03 +0000700 INIT_LIST_HEAD(&works);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400701
Josef Bacik9ffba8c2013-08-14 11:33:56 -0400702 mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000703 spin_lock(&root->fs_info->ordered_root_lock);
Josef Bacik569e0f32013-02-13 11:09:14 -0500704 list_splice_init(&cur_trans->ordered_operations, &splice);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400705 while (!list_empty(&splice)) {
706 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
707 ordered_operations);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400708 inode = &btrfs_inode->vfs_inode;
709
710 list_del_init(&btrfs_inode->ordered_operations);
711
712 /*
713 * the inode may be getting freed (in sys_unlink path).
714 */
715 inode = igrab(inode);
Miao Xie25287e02012-10-25 09:31:03 +0000716 if (!inode)
717 continue;
Miao Xie5b947f12013-01-22 10:52:04 +0000718
719 if (!wait)
720 list_add_tail(&BTRFS_I(inode)->ordered_operations,
Josef Bacik569e0f32013-02-13 11:09:14 -0500721 &cur_trans->ordered_operations);
Miao Xie199c2a92013-05-15 07:48:23 +0000722 spin_unlock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400723
Miao Xie25287e02012-10-25 09:31:03 +0000724 work = btrfs_alloc_delalloc_work(inode, wait, 1);
725 if (!work) {
Miao Xie199c2a92013-05-15 07:48:23 +0000726 spin_lock(&root->fs_info->ordered_root_lock);
Miao Xie25287e02012-10-25 09:31:03 +0000727 if (list_empty(&BTRFS_I(inode)->ordered_operations))
728 list_add_tail(&btrfs_inode->ordered_operations,
729 &splice);
Miao Xie25287e02012-10-25 09:31:03 +0000730 list_splice_tail(&splice,
Josef Bacik569e0f32013-02-13 11:09:14 -0500731 &cur_trans->ordered_operations);
Miao Xie199c2a92013-05-15 07:48:23 +0000732 spin_unlock(&root->fs_info->ordered_root_lock);
Miao Xie25287e02012-10-25 09:31:03 +0000733 ret = -ENOMEM;
734 goto out;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400735 }
Miao Xie25287e02012-10-25 09:31:03 +0000736 list_add_tail(&work->list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +0800737 btrfs_queue_work(root->fs_info->flush_workers,
738 &work->work);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400739
740 cond_resched();
Miao Xie199c2a92013-05-15 07:48:23 +0000741 spin_lock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -0400742 }
Miao Xie199c2a92013-05-15 07:48:23 +0000743 spin_unlock(&root->fs_info->ordered_root_lock);
Miao Xie25287e02012-10-25 09:31:03 +0000744out:
745 list_for_each_entry_safe(work, next, &works, list) {
746 list_del_init(&work->list);
747 btrfs_wait_and_free_delalloc_work(work);
748 }
Josef Bacik9ffba8c2013-08-14 11:33:56 -0400749 mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
Miao Xie25287e02012-10-25 09:31:03 +0000750 return ret;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400751}
752
753/*
Chris Masoneb84ae02008-07-17 13:53:27 -0400754 * Used to start IO or wait for a given ordered extent to finish.
755 *
756 * If wait is one, this effectively waits on page writeback for all the pages
757 * in the extent, and it waits on the io completion code to insert
758 * metadata into the btree corresponding to the extent
759 */
760void btrfs_start_ordered_extent(struct inode *inode,
761 struct btrfs_ordered_extent *entry,
762 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400763{
764 u64 start = entry->file_offset;
765 u64 end = start + entry->len - 1;
766
liubo1abe9b82011-03-24 11:18:59 +0000767 trace_btrfs_ordered_extent_start(inode, entry);
768
Chris Masoneb84ae02008-07-17 13:53:27 -0400769 /*
770 * pages in the range can be dirty, clean or writeback. We
771 * start IO on any dirty ones so the wait doesn't stall waiting
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300772 * for the flusher thread to find them
Chris Masoneb84ae02008-07-17 13:53:27 -0400773 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400774 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
775 filemap_fdatawrite_range(inode->i_mapping, start, end);
Chris Masonc8b97812008-10-29 14:49:59 -0400776 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400777 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
778 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400779 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400780}
781
Chris Masoneb84ae02008-07-17 13:53:27 -0400782/*
783 * Used to wait on ordered extents across a large range of bytes.
784 */
Josef Bacik0ef8b722013-10-25 16:13:35 -0400785int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400786{
Josef Bacik0ef8b722013-10-25 16:13:35 -0400787 int ret = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400788 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400789 u64 orig_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400790 struct btrfs_ordered_extent *ordered;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400791
Chris Masone5a22172008-07-18 20:42:20 -0400792 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400793 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400794 } else {
795 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400796 if (orig_end > INT_LIMIT(loff_t))
797 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400798 }
Josef Bacik551ebb22012-04-23 14:41:09 -0400799
Chris Masone5a22172008-07-18 20:42:20 -0400800 /* start IO across the range first to instantiate any delalloc
801 * extents
802 */
Josef Bacik0ef8b722013-10-25 16:13:35 -0400803 ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
804 if (ret)
805 return ret;
Josef Bacik7ddf5a42012-06-08 15:26:47 -0400806 /*
807 * So with compression we will find and lock a dirty page and clear the
808 * first one as dirty, setup an async extent, and immediately return
809 * with the entire range locked but with nobody actually marked with
810 * writeback. So we can't just filemap_write_and_wait_range() and
811 * expect it to work since it will just kick off a thread to do the
812 * actual work. So we need to call filemap_fdatawrite_range _again_
813 * since it will wait on the page lock, which won't be unlocked until
814 * after the pages have been marked as writeback and so we're good to go
815 * from there. We have to do this otherwise we'll miss the ordered
816 * extents and that results in badness. Please Josef, do not think you
817 * know better and pull this out at some point in the future, it is
818 * right and you are wrong.
819 */
820 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
Josef Bacik0ef8b722013-10-25 16:13:35 -0400821 &BTRFS_I(inode)->runtime_flags)) {
822 ret = filemap_fdatawrite_range(inode->i_mapping, start,
823 orig_end);
824 if (ret)
825 return ret;
826 }
827 ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
828 if (ret)
829 return ret;
Chris Masonf4219502008-07-22 11:18:09 -0400830
831 end = orig_end;
Chris Masond3977122009-01-05 21:25:51 -0500832 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400833 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500834 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400835 break;
Chris Masone5a22172008-07-18 20:42:20 -0400836 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400837 btrfs_put_ordered_extent(ordered);
838 break;
839 }
Filipe David Borba Mananab52abf12013-11-06 15:12:40 +0000840 if (ordered->file_offset + ordered->len <= start) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400841 btrfs_put_ordered_extent(ordered);
842 break;
843 }
Chris Masone5a22172008-07-18 20:42:20 -0400844 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400845 end = ordered->file_offset;
Josef Bacik0ef8b722013-10-25 16:13:35 -0400846 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
847 ret = -EIO;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400848 btrfs_put_ordered_extent(ordered);
Josef Bacik0ef8b722013-10-25 16:13:35 -0400849 if (ret || end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400850 break;
851 end--;
852 }
Josef Bacik0ef8b722013-10-25 16:13:35 -0400853 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400854}
855
Chris Masoneb84ae02008-07-17 13:53:27 -0400856/*
857 * find an ordered extent corresponding to file_offset. return NULL if
858 * nothing is found, otherwise take a reference on the extent and return it
859 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400860struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
861 u64 file_offset)
862{
863 struct btrfs_ordered_inode_tree *tree;
864 struct rb_node *node;
865 struct btrfs_ordered_extent *entry = NULL;
866
867 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400868 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400869 node = tree_search(tree, file_offset);
870 if (!node)
871 goto out;
872
873 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
874 if (!offset_in_entry(entry, file_offset))
875 entry = NULL;
876 if (entry)
877 atomic_inc(&entry->refs);
878out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400879 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400880 return entry;
881}
882
Josef Bacik4b46fce2010-05-23 11:00:55 -0400883/* Since the DIO code tries to lock a wide area we need to look for any ordered
884 * extents that exist in the range, rather than just the start of the range.
885 */
886struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
887 u64 file_offset,
888 u64 len)
889{
890 struct btrfs_ordered_inode_tree *tree;
891 struct rb_node *node;
892 struct btrfs_ordered_extent *entry = NULL;
893
894 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400895 spin_lock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400896 node = tree_search(tree, file_offset);
897 if (!node) {
898 node = tree_search(tree, file_offset + len);
899 if (!node)
900 goto out;
901 }
902
903 while (1) {
904 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
905 if (range_overlaps(entry, file_offset, len))
906 break;
907
908 if (entry->file_offset >= file_offset + len) {
909 entry = NULL;
910 break;
911 }
912 entry = NULL;
913 node = rb_next(node);
914 if (!node)
915 break;
916 }
917out:
918 if (entry)
919 atomic_inc(&entry->refs);
Josef Bacik5fd02042012-05-02 14:00:54 -0400920 spin_unlock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400921 return entry;
922}
923
Chris Masoneb84ae02008-07-17 13:53:27 -0400924/*
925 * lookup and return any extent before 'file_offset'. NULL is returned
926 * if none is found
927 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400928struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500929btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400930{
931 struct btrfs_ordered_inode_tree *tree;
932 struct rb_node *node;
933 struct btrfs_ordered_extent *entry = NULL;
934
935 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400936 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400937 node = tree_search(tree, file_offset);
938 if (!node)
939 goto out;
940
941 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
942 atomic_inc(&entry->refs);
943out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400944 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400945 return entry;
946}
Chris Masondbe674a2008-07-17 12:54:05 -0400947
Chris Masoneb84ae02008-07-17 13:53:27 -0400948/*
949 * After an extent is done, call this to conditionally update the on disk
950 * i_size. i_size is updated to cover any fully written part of the file.
951 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000952int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
Chris Masondbe674a2008-07-17 12:54:05 -0400953 struct btrfs_ordered_extent *ordered)
954{
955 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Masondbe674a2008-07-17 12:54:05 -0400956 u64 disk_i_size;
957 u64 new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +0000958 u64 i_size = i_size_read(inode);
Chris Masondbe674a2008-07-17 12:54:05 -0400959 struct rb_node *node;
Yan, Zhengc2167752009-11-12 09:34:21 +0000960 struct rb_node *prev = NULL;
Chris Masondbe674a2008-07-17 12:54:05 -0400961 struct btrfs_ordered_extent *test;
Yan, Zhengc2167752009-11-12 09:34:21 +0000962 int ret = 1;
963
Josef Bacik5fd02042012-05-02 14:00:54 -0400964 spin_lock_irq(&tree->lock);
Josef Bacik77cef2e2013-08-29 13:57:21 -0400965 if (ordered) {
966 offset = entry_end(ordered);
967 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
968 offset = min(offset,
969 ordered->file_offset +
970 ordered->truncated_len);
971 } else {
972 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
973 }
Chris Masondbe674a2008-07-17 12:54:05 -0400974 disk_i_size = BTRFS_I(inode)->disk_i_size;
975
Yan, Zhengc2167752009-11-12 09:34:21 +0000976 /* truncate file */
977 if (disk_i_size > i_size) {
978 BTRFS_I(inode)->disk_i_size = i_size;
979 ret = 0;
980 goto out;
981 }
982
Chris Masondbe674a2008-07-17 12:54:05 -0400983 /*
984 * if the disk i_size is already at the inode->i_size, or
985 * this ordered extent is inside the disk i_size, we're done
986 */
Josef Bacik5d1f4022013-01-30 14:17:31 -0500987 if (disk_i_size == i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400988 goto out;
Josef Bacik5d1f4022013-01-30 14:17:31 -0500989
990 /*
991 * We still need to update disk_i_size if outstanding_isize is greater
992 * than disk_i_size.
993 */
994 if (offset <= disk_i_size &&
995 (!ordered || ordered->outstanding_isize <= disk_i_size))
996 goto out;
Chris Masondbe674a2008-07-17 12:54:05 -0400997
998 /*
Chris Masondbe674a2008-07-17 12:54:05 -0400999 * walk backward from this ordered extent to disk_i_size.
1000 * if we find an ordered extent then we can't update disk i_size
1001 * yet
1002 */
Yan, Zhengc2167752009-11-12 09:34:21 +00001003 if (ordered) {
1004 node = rb_prev(&ordered->rb_node);
1005 } else {
1006 prev = tree_search(tree, offset);
1007 /*
1008 * we insert file extents without involving ordered struct,
1009 * so there should be no ordered struct cover this offset
1010 */
1011 if (prev) {
1012 test = rb_entry(prev, struct btrfs_ordered_extent,
1013 rb_node);
1014 BUG_ON(offset_in_entry(test, offset));
1015 }
1016 node = prev;
1017 }
Josef Bacik5fd02042012-05-02 14:00:54 -04001018 for (; node; node = rb_prev(node)) {
Chris Masondbe674a2008-07-17 12:54:05 -04001019 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -04001020
1021 /* We treat this entry as if it doesnt exist */
1022 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1023 continue;
Chris Masondbe674a2008-07-17 12:54:05 -04001024 if (test->file_offset + test->len <= disk_i_size)
1025 break;
Yan, Zhengc2167752009-11-12 09:34:21 +00001026 if (test->file_offset >= i_size)
Chris Masondbe674a2008-07-17 12:54:05 -04001027 break;
Josef Bacik59fe4f42013-01-30 14:31:31 -05001028 if (entry_end(test) > disk_i_size) {
Miao Xieb9a8cc52012-09-06 04:01:21 -06001029 /*
1030 * we don't update disk_i_size now, so record this
1031 * undealt i_size. Or we will not know the real
1032 * i_size.
1033 */
1034 if (test->outstanding_isize < offset)
1035 test->outstanding_isize = offset;
1036 if (ordered &&
1037 ordered->outstanding_isize >
1038 test->outstanding_isize)
1039 test->outstanding_isize =
1040 ordered->outstanding_isize;
Chris Masondbe674a2008-07-17 12:54:05 -04001041 goto out;
Miao Xieb9a8cc52012-09-06 04:01:21 -06001042 }
Chris Masondbe674a2008-07-17 12:54:05 -04001043 }
Yan, Zhengc2167752009-11-12 09:34:21 +00001044 new_i_size = min_t(u64, offset, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -04001045
1046 /*
Miao Xieb9a8cc52012-09-06 04:01:21 -06001047 * Some ordered extents may completed before the current one, and
1048 * we hold the real i_size in ->outstanding_isize.
Chris Masondbe674a2008-07-17 12:54:05 -04001049 */
Miao Xieb9a8cc52012-09-06 04:01:21 -06001050 if (ordered && ordered->outstanding_isize > new_i_size)
1051 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -04001052 BTRFS_I(inode)->disk_i_size = new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +00001053 ret = 0;
Chris Masondbe674a2008-07-17 12:54:05 -04001054out:
Yan, Zhengc2167752009-11-12 09:34:21 +00001055 /*
Josef Bacik5fd02042012-05-02 14:00:54 -04001056 * We need to do this because we can't remove ordered extents until
1057 * after the i_disk_size has been updated and then the inode has been
1058 * updated to reflect the change, so we need to tell anybody who finds
1059 * this ordered extent that we've already done all the real work, we
1060 * just haven't completed all the other work.
Yan, Zhengc2167752009-11-12 09:34:21 +00001061 */
1062 if (ordered)
Josef Bacik5fd02042012-05-02 14:00:54 -04001063 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1064 spin_unlock_irq(&tree->lock);
Yan, Zhengc2167752009-11-12 09:34:21 +00001065 return ret;
Chris Masondbe674a2008-07-17 12:54:05 -04001066}
Chris Masonba1da2f2008-07-17 12:54:15 -04001067
Chris Masoneb84ae02008-07-17 13:53:27 -04001068/*
1069 * search the ordered extents for one corresponding to 'offset' and
1070 * try to find a checksum. This is used because we allow pages to
1071 * be reclaimed before their checksum is actually put into the btree
1072 */
Chris Masond20f7042008-12-08 16:58:54 -05001073int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
Miao Xiee4100d92013-04-05 07:20:56 +00001074 u32 *sum, int len)
Chris Masonba1da2f2008-07-17 12:54:15 -04001075{
1076 struct btrfs_ordered_sum *ordered_sum;
Chris Masonba1da2f2008-07-17 12:54:15 -04001077 struct btrfs_ordered_extent *ordered;
1078 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -04001079 unsigned long num_sectors;
1080 unsigned long i;
1081 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Miao Xiee4100d92013-04-05 07:20:56 +00001082 int index = 0;
Chris Masonba1da2f2008-07-17 12:54:15 -04001083
1084 ordered = btrfs_lookup_ordered_extent(inode, offset);
1085 if (!ordered)
Miao Xiee4100d92013-04-05 07:20:56 +00001086 return 0;
Chris Masonba1da2f2008-07-17 12:54:15 -04001087
Josef Bacik5fd02042012-05-02 14:00:54 -04001088 spin_lock_irq(&tree->lock);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001089 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Miao Xiee4100d92013-04-05 07:20:56 +00001090 if (disk_bytenr >= ordered_sum->bytenr &&
1091 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1092 i = (disk_bytenr - ordered_sum->bytenr) >>
1093 inode->i_sb->s_blocksize_bits;
Miao Xiee4100d92013-04-05 07:20:56 +00001094 num_sectors = ordered_sum->len >>
1095 inode->i_sb->s_blocksize_bits;
Miao Xief51a4a12013-06-19 10:36:09 +08001096 num_sectors = min_t(int, len - index, num_sectors - i);
1097 memcpy(sum + index, ordered_sum->sums + i,
1098 num_sectors);
1099
1100 index += (int)num_sectors;
1101 if (index == len)
1102 goto out;
1103 disk_bytenr += num_sectors * sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -04001104 }
1105 }
1106out:
Josef Bacik5fd02042012-05-02 14:00:54 -04001107 spin_unlock_irq(&tree->lock);
Chris Mason89642222008-07-24 09:41:53 -04001108 btrfs_put_ordered_extent(ordered);
Miao Xiee4100d92013-04-05 07:20:56 +00001109 return index;
Chris Masonba1da2f2008-07-17 12:54:15 -04001110}
1111
Chris Masonf4219502008-07-22 11:18:09 -04001112
Chris Mason5a3f23d2009-03-31 13:27:11 -04001113/*
1114 * add a given inode to the list of inodes that must be fully on
1115 * disk before a transaction commit finishes.
1116 *
1117 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1118 * used to make sure renamed files are fully on disk.
1119 *
1120 * It is a noop if the inode is already fully on disk.
1121 *
1122 * If trans is not null, we'll do a friendly check for a transaction that
1123 * is already flushing things and force the IO down ourselves.
1124 */
Jeff Mahoney143bede2012-03-01 14:56:26 +01001125void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1126 struct btrfs_root *root, struct inode *inode)
Chris Mason5a3f23d2009-03-31 13:27:11 -04001127{
Josef Bacik569e0f32013-02-13 11:09:14 -05001128 struct btrfs_transaction *cur_trans = trans->transaction;
Chris Mason5a3f23d2009-03-31 13:27:11 -04001129 u64 last_mod;
1130
1131 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1132
1133 /*
1134 * if this file hasn't been changed since the last transaction
1135 * commit, we can safely return without doing anything
1136 */
chandan5ede8592013-10-14 18:44:39 +05301137 if (last_mod <= root->fs_info->last_trans_committed)
Jeff Mahoney143bede2012-03-01 14:56:26 +01001138 return;
Chris Mason5a3f23d2009-03-31 13:27:11 -04001139
Miao Xie199c2a92013-05-15 07:48:23 +00001140 spin_lock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -04001141 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1142 list_add_tail(&BTRFS_I(inode)->ordered_operations,
Josef Bacik569e0f32013-02-13 11:09:14 -05001143 &cur_trans->ordered_operations);
Chris Mason5a3f23d2009-03-31 13:27:11 -04001144 }
Miao Xie199c2a92013-05-15 07:48:23 +00001145 spin_unlock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -04001146}
Miao Xie6352b912012-09-06 04:01:51 -06001147
1148int __init ordered_data_init(void)
1149{
1150 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1151 sizeof(struct btrfs_ordered_extent), 0,
1152 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1153 NULL);
1154 if (!btrfs_ordered_extent_cache)
1155 return -ENOMEM;
Miao Xie25287e02012-10-25 09:31:03 +00001156
Miao Xie6352b912012-09-06 04:01:51 -06001157 return 0;
1158}
1159
1160void ordered_data_exit(void)
1161{
1162 if (btrfs_ordered_extent_cache)
1163 kmem_cache_destroy(btrfs_ordered_extent_cache);
1164}