blob: 559170464d7c50f43d6804d59dfd9f73e57a707f [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"
Anand Jainebb87652016-03-10 17:26:59 +080028#include "compression.h"
Chris Masondc17ff82008-01-08 15:46:30 -050029
Miao Xie6352b912012-09-06 04:01:51 -060030static struct kmem_cache *btrfs_ordered_extent_cache;
31
Chris Masone6dcd2d2008-07-17 12:53:50 -040032static u64 entry_end(struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -050033{
Chris Masone6dcd2d2008-07-17 12:53:50 -040034 if (entry->file_offset + entry->len < entry->file_offset)
35 return (u64)-1;
36 return entry->file_offset + entry->len;
Chris Masondc17ff82008-01-08 15:46:30 -050037}
38
Chris Masond352ac62008-09-29 15:18:18 -040039/* returns NULL if the insertion worked, or it returns the node it did find
40 * in the tree
41 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040042static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
43 struct rb_node *node)
Chris Masondc17ff82008-01-08 15:46:30 -050044{
Chris Masond3977122009-01-05 21:25:51 -050045 struct rb_node **p = &root->rb_node;
46 struct rb_node *parent = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040047 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -050048
Chris Masond3977122009-01-05 21:25:51 -050049 while (*p) {
Chris Masondc17ff82008-01-08 15:46:30 -050050 parent = *p;
Chris Masone6dcd2d2008-07-17 12:53:50 -040051 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050052
Chris Masone6dcd2d2008-07-17 12:53:50 -040053 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050054 p = &(*p)->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040055 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050056 p = &(*p)->rb_right;
57 else
58 return parent;
59 }
60
61 rb_link_node(node, parent, p);
62 rb_insert_color(node, root);
63 return NULL;
64}
65
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040066static void ordered_data_tree_panic(struct inode *inode, int errno,
67 u64 offset)
68{
69 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
70 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
David Sterba351fd352014-05-15 16:48:20 +020071 "%llu", offset);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -040072}
73
Chris Masond352ac62008-09-29 15:18:18 -040074/*
75 * look for a given offset in the tree, and if it can't be found return the
76 * first lesser offset
77 */
Chris Masone6dcd2d2008-07-17 12:53:50 -040078static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
79 struct rb_node **prev_ret)
Chris Masondc17ff82008-01-08 15:46:30 -050080{
Chris Masond3977122009-01-05 21:25:51 -050081 struct rb_node *n = root->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -050082 struct rb_node *prev = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -040083 struct rb_node *test;
84 struct btrfs_ordered_extent *entry;
85 struct btrfs_ordered_extent *prev_entry = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -050086
Chris Masond3977122009-01-05 21:25:51 -050087 while (n) {
Chris Masone6dcd2d2008-07-17 12:53:50 -040088 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Chris Masondc17ff82008-01-08 15:46:30 -050089 prev = n;
90 prev_entry = entry;
Chris Masondc17ff82008-01-08 15:46:30 -050091
Chris Masone6dcd2d2008-07-17 12:53:50 -040092 if (file_offset < entry->file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -050093 n = n->rb_left;
Chris Masone6dcd2d2008-07-17 12:53:50 -040094 else if (file_offset >= entry_end(entry))
Chris Masondc17ff82008-01-08 15:46:30 -050095 n = n->rb_right;
96 else
97 return n;
98 }
99 if (!prev_ret)
100 return NULL;
101
Chris Masond3977122009-01-05 21:25:51 -0500102 while (prev && file_offset >= entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400103 test = rb_next(prev);
104 if (!test)
105 break;
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 rb_node);
108 if (file_offset < entry_end(prev_entry))
109 break;
110
111 prev = test;
112 }
113 if (prev)
114 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
115 rb_node);
Chris Masond3977122009-01-05 21:25:51 -0500116 while (prev && file_offset < entry_end(prev_entry)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400117 test = rb_prev(prev);
118 if (!test)
119 break;
120 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
121 rb_node);
122 prev = test;
Chris Masondc17ff82008-01-08 15:46:30 -0500123 }
124 *prev_ret = prev;
125 return NULL;
126}
127
Chris Masond352ac62008-09-29 15:18:18 -0400128/*
129 * helper to check if a given offset is inside a given entry
130 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400131static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
Chris Masondc17ff82008-01-08 15:46:30 -0500132{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400133 if (file_offset < entry->file_offset ||
134 entry->file_offset + entry->len <= file_offset)
135 return 0;
136 return 1;
137}
138
Josef Bacik4b46fce2010-05-23 11:00:55 -0400139static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
140 u64 len)
141{
142 if (file_offset + len <= entry->file_offset ||
143 entry->file_offset + entry->len <= file_offset)
144 return 0;
145 return 1;
146}
147
Chris Masond352ac62008-09-29 15:18:18 -0400148/*
149 * look find the first ordered struct that has this offset, otherwise
150 * the first one less than this offset
151 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400152static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
153 u64 file_offset)
154{
155 struct rb_root *root = &tree->tree;
Chris Masonc87fb6f2011-01-31 19:54:59 -0500156 struct rb_node *prev = NULL;
Chris Masondc17ff82008-01-08 15:46:30 -0500157 struct rb_node *ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400158 struct btrfs_ordered_extent *entry;
159
160 if (tree->last) {
161 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
162 rb_node);
163 if (offset_in_entry(entry, file_offset))
164 return tree->last;
165 }
166 ret = __tree_search(root, file_offset, &prev);
Chris Masondc17ff82008-01-08 15:46:30 -0500167 if (!ret)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400168 ret = prev;
169 if (ret)
170 tree->last = ret;
Chris Masondc17ff82008-01-08 15:46:30 -0500171 return ret;
172}
173
Chris Masoneb84ae02008-07-17 13:53:27 -0400174/* allocate and add a new ordered_extent into the per-inode tree.
175 * file_offset is the logical offset in the file
176 *
177 * start is the disk block number of an extent already reserved in the
178 * extent allocation tree
179 *
180 * len is the length of the extent
181 *
Chris Masoneb84ae02008-07-17 13:53:27 -0400182 * The tree is given a single reference on the ordered extent that was
183 * inserted.
184 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400185static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
186 u64 start, u64 len, u64 disk_len,
Li Zefan261507a02010-12-17 14:21:50 +0800187 int type, int dio, int compress_type)
Chris Masondc17ff82008-01-08 15:46:30 -0500188{
Miao Xie199c2a92013-05-15 07:48:23 +0000189 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500190 struct btrfs_ordered_inode_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400191 struct rb_node *node;
192 struct btrfs_ordered_extent *entry;
Chris Masondc17ff82008-01-08 15:46:30 -0500193
Chris Masone6dcd2d2008-07-17 12:53:50 -0400194 tree = &BTRFS_I(inode)->ordered_tree;
Miao Xie6352b912012-09-06 04:01:51 -0600195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
Chris Masondc17ff82008-01-08 15:46:30 -0500196 if (!entry)
197 return -ENOMEM;
198
Chris Masone6dcd2d2008-07-17 12:53:50 -0400199 entry->file_offset = file_offset;
200 entry->start = start;
201 entry->len = len;
Chris Masonc8b97812008-10-29 14:49:59 -0400202 entry->disk_len = disk_len;
Chris Mason8b62b722009-09-02 16:53:46 -0400203 entry->bytes_left = len;
Josef Bacik5fd02042012-05-02 14:00:54 -0400204 entry->inode = igrab(inode);
Li Zefan261507a02010-12-17 14:21:50 +0800205 entry->compress_type = compress_type;
Josef Bacik77cef2e2013-08-29 13:57:21 -0400206 entry->truncated_len = (u64)-1;
Yan Zhengd899e052008-10-30 14:25:28 -0400207 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
Yan Zheng80ff3852008-10-30 14:20:02 -0400208 set_bit(type, &entry->flags);
Chris Mason3eaa2882008-07-24 11:57:52 -0400209
Josef Bacik4b46fce2010-05-23 11:00:55 -0400210 if (dio)
211 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
212
Chris Masone6dcd2d2008-07-17 12:53:50 -0400213 /* one ref for the tree */
214 atomic_set(&entry->refs, 1);
215 init_waitqueue_head(&entry->wait);
216 INIT_LIST_HEAD(&entry->list);
Chris Mason3eaa2882008-07-24 11:57:52 -0400217 INIT_LIST_HEAD(&entry->root_extent_list);
Miao Xie9afab882012-10-25 09:41:36 +0000218 INIT_LIST_HEAD(&entry->work_list);
219 init_completion(&entry->completion);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400220 INIT_LIST_HEAD(&entry->log_list);
Josef Bacik50d9aa92014-11-21 14:52:38 -0500221 INIT_LIST_HEAD(&entry->trans_list);
Chris Masondc17ff82008-01-08 15:46:30 -0500222
liubo1abe9b82011-03-24 11:18:59 +0000223 trace_btrfs_ordered_extent_add(inode, entry);
224
Josef Bacik5fd02042012-05-02 14:00:54 -0400225 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400226 node = tree_insert(&tree->tree, file_offset,
227 &entry->rb_node);
Jeff Mahoney43c04fb2011-10-03 23:22:33 -0400228 if (node)
229 ordered_data_tree_panic(inode, -EEXIST, file_offset);
Josef Bacik5fd02042012-05-02 14:00:54 -0400230 spin_unlock_irq(&tree->lock);
Chris Masond3977122009-01-05 21:25:51 -0500231
Miao Xie199c2a92013-05-15 07:48:23 +0000232 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400233 list_add_tail(&entry->root_extent_list,
Miao Xie199c2a92013-05-15 07:48:23 +0000234 &root->ordered_extents);
235 root->nr_ordered_extents++;
236 if (root->nr_ordered_extents == 1) {
237 spin_lock(&root->fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root,
240 &root->fs_info->ordered_roots);
241 spin_unlock(&root->fs_info->ordered_root_lock);
242 }
243 spin_unlock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400244
Chris Masone6dcd2d2008-07-17 12:53:50 -0400245 return 0;
246}
Chris Masondc17ff82008-01-08 15:46:30 -0500247
Josef Bacik4b46fce2010-05-23 11:00:55 -0400248int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
249 u64 start, u64 len, u64 disk_len, int type)
250{
251 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800252 disk_len, type, 0,
253 BTRFS_COMPRESS_NONE);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400254}
255
256int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
257 u64 start, u64 len, u64 disk_len, int type)
258{
259 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
Li Zefan261507a02010-12-17 14:21:50 +0800260 disk_len, type, 1,
261 BTRFS_COMPRESS_NONE);
262}
263
264int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
265 u64 start, u64 len, u64 disk_len,
266 int type, int compress_type)
267{
268 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
269 disk_len, type, 0,
270 compress_type);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400271}
272
Chris Masoneb84ae02008-07-17 13:53:27 -0400273/*
274 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
Chris Mason3edf7d32008-07-18 06:17:13 -0400275 * when an ordered extent is finished. If the list covers more than one
276 * ordered extent, it is split across multiples.
Chris Masoneb84ae02008-07-17 13:53:27 -0400277 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100278void btrfs_add_ordered_sum(struct inode *inode,
279 struct btrfs_ordered_extent *entry,
280 struct btrfs_ordered_sum *sum)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400281{
282 struct btrfs_ordered_inode_tree *tree;
Chris Mason1b1e2132008-06-25 16:01:31 -0400283
Chris Masone6dcd2d2008-07-17 12:53:50 -0400284 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400285 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400286 list_add_tail(&sum->list, &entry->list);
Josef Bacik5fd02042012-05-02 14:00:54 -0400287 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400288}
289
Chris Masoneb84ae02008-07-17 13:53:27 -0400290/*
291 * this is used to account for finished IO across a given range
Chris Mason163cf092010-11-28 19:56:33 -0500292 * of the file. The IO may span ordered extents. If
293 * a given ordered_extent is completely done, 1 is returned, otherwise
294 * 0.
295 *
296 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
297 * to make sure this function only returns 1 once for a given ordered extent.
298 *
299 * file_offset is updated to one byte past the range that is recorded as
300 * complete. This allows you to walk forward in the file.
301 */
302int btrfs_dec_test_first_ordered_pending(struct inode *inode,
303 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400304 u64 *file_offset, u64 io_size, int uptodate)
Chris Mason163cf092010-11-28 19:56:33 -0500305{
306 struct btrfs_ordered_inode_tree *tree;
307 struct rb_node *node;
308 struct btrfs_ordered_extent *entry = NULL;
309 int ret;
Josef Bacik5fd02042012-05-02 14:00:54 -0400310 unsigned long flags;
Chris Mason163cf092010-11-28 19:56:33 -0500311 u64 dec_end;
312 u64 dec_start;
313 u64 to_dec;
314
315 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400316 spin_lock_irqsave(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500317 node = tree_search(tree, *file_offset);
318 if (!node) {
319 ret = 1;
320 goto out;
321 }
322
323 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
324 if (!offset_in_entry(entry, *file_offset)) {
325 ret = 1;
326 goto out;
327 }
328
329 dec_start = max(*file_offset, entry->file_offset);
330 dec_end = min(*file_offset + io_size, entry->file_offset +
331 entry->len);
332 *file_offset = dec_end;
333 if (dec_start > dec_end) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500334 btrfs_crit(BTRFS_I(inode)->root->fs_info,
335 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
Chris Mason163cf092010-11-28 19:56:33 -0500336 }
337 to_dec = dec_end - dec_start;
338 if (to_dec > entry->bytes_left) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordered accounting left %llu size %llu",
341 entry->bytes_left, to_dec);
Chris Mason163cf092010-11-28 19:56:33 -0500342 }
343 entry->bytes_left -= to_dec;
Josef Bacik5fd02042012-05-02 14:00:54 -0400344 if (!uptodate)
345 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
346
Miao Xieaf7a6502014-03-06 13:54:56 +0800347 if (entry->bytes_left == 0) {
Chris Mason163cf092010-11-28 19:56:33 -0500348 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
David Sterbaa83342a2015-02-16 19:36:47 +0100349 /*
350 * Implicit memory barrier after test_and_set_bit
351 */
Miao Xieaf7a6502014-03-06 13:54:56 +0800352 if (waitqueue_active(&entry->wait))
353 wake_up(&entry->wait);
354 } else {
Chris Mason163cf092010-11-28 19:56:33 -0500355 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800356 }
Chris Mason163cf092010-11-28 19:56:33 -0500357out:
358 if (!ret && cached && entry) {
359 *cached = entry;
360 atomic_inc(&entry->refs);
361 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400362 spin_unlock_irqrestore(&tree->lock, flags);
Chris Mason163cf092010-11-28 19:56:33 -0500363 return ret == 0;
364}
365
366/*
367 * this is used to account for finished IO across a given range
Chris Masoneb84ae02008-07-17 13:53:27 -0400368 * of the file. The IO should not span ordered extents. If
369 * a given ordered_extent is completely done, 1 is returned, otherwise
370 * 0.
371 *
372 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
373 * to make sure this function only returns 1 once for a given ordered extent.
374 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400375int btrfs_dec_test_ordered_pending(struct inode *inode,
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000376 struct btrfs_ordered_extent **cached,
Josef Bacik5fd02042012-05-02 14:00:54 -0400377 u64 file_offset, u64 io_size, int uptodate)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400378{
379 struct btrfs_ordered_inode_tree *tree;
380 struct rb_node *node;
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000381 struct btrfs_ordered_extent *entry = NULL;
Josef Bacik5fd02042012-05-02 14:00:54 -0400382 unsigned long flags;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400383 int ret;
384
385 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400386 spin_lock_irqsave(&tree->lock, flags);
387 if (cached && *cached) {
388 entry = *cached;
389 goto have_entry;
390 }
391
Chris Masone6dcd2d2008-07-17 12:53:50 -0400392 node = tree_search(tree, file_offset);
393 if (!node) {
394 ret = 1;
395 goto out;
396 }
397
398 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -0400399have_entry:
Chris Masone6dcd2d2008-07-17 12:53:50 -0400400 if (!offset_in_entry(entry, file_offset)) {
401 ret = 1;
402 goto out;
403 }
404
Chris Mason8b62b722009-09-02 16:53:46 -0400405 if (io_size > entry->bytes_left) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500406 btrfs_crit(BTRFS_I(inode)->root->fs_info,
407 "bad ordered accounting left %llu size %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200408 entry->bytes_left, io_size);
Chris Mason8b62b722009-09-02 16:53:46 -0400409 }
410 entry->bytes_left -= io_size;
Josef Bacik5fd02042012-05-02 14:00:54 -0400411 if (!uptodate)
412 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
413
Miao Xieaf7a6502014-03-06 13:54:56 +0800414 if (entry->bytes_left == 0) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400415 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
David Sterbaa83342a2015-02-16 19:36:47 +0100416 /*
417 * Implicit memory barrier after test_and_set_bit
418 */
Miao Xieaf7a6502014-03-06 13:54:56 +0800419 if (waitqueue_active(&entry->wait))
420 wake_up(&entry->wait);
421 } else {
Chris Mason8b62b722009-09-02 16:53:46 -0400422 ret = 1;
Miao Xieaf7a6502014-03-06 13:54:56 +0800423 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400424out:
Josef Bacik5a1a3df2010-02-02 20:51:14 +0000425 if (!ret && cached && entry) {
426 *cached = entry;
427 atomic_inc(&entry->refs);
428 }
Josef Bacik5fd02042012-05-02 14:00:54 -0400429 spin_unlock_irqrestore(&tree->lock, flags);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400430 return ret == 0;
431}
432
Josef Bacik2ab28f32012-10-12 15:27:49 -0400433/* Needs to either be called under a log transaction or the log_mutex */
Miao Xie827463c2014-01-14 20:31:51 +0800434void btrfs_get_logged_extents(struct inode *inode,
Filipe Manana08702952014-11-13 17:00:35 +0000435 struct list_head *logged_list,
436 const loff_t start,
437 const loff_t end)
Josef Bacik2ab28f32012-10-12 15:27:49 -0400438{
439 struct btrfs_ordered_inode_tree *tree;
440 struct btrfs_ordered_extent *ordered;
441 struct rb_node *n;
Filipe Manana08702952014-11-13 17:00:35 +0000442 struct rb_node *prev;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400443
444 tree = &BTRFS_I(inode)->ordered_tree;
445 spin_lock_irq(&tree->lock);
Filipe Manana08702952014-11-13 17:00:35 +0000446 n = __tree_search(&tree->tree, end, &prev);
447 if (!n)
448 n = prev;
449 for (; n; n = rb_prev(n)) {
Josef Bacik2ab28f32012-10-12 15:27:49 -0400450 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
Filipe Manana08702952014-11-13 17:00:35 +0000451 if (ordered->file_offset > end)
452 continue;
453 if (entry_end(ordered) <= start)
454 break;
Filipe Manana4d884fc2015-02-09 17:17:43 +0000455 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
Josef Bacik50d9aa92014-11-21 14:52:38 -0500456 continue;
Filipe Manana08702952014-11-13 17:00:35 +0000457 list_add(&ordered->log_list, logged_list);
Miao Xie827463c2014-01-14 20:31:51 +0800458 atomic_inc(&ordered->refs);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400459 }
460 spin_unlock_irq(&tree->lock);
461}
462
Miao Xie827463c2014-01-14 20:31:51 +0800463void btrfs_put_logged_extents(struct list_head *logged_list)
464{
465 struct btrfs_ordered_extent *ordered;
466
467 while (!list_empty(logged_list)) {
468 ordered = list_first_entry(logged_list,
469 struct btrfs_ordered_extent,
470 log_list);
471 list_del_init(&ordered->log_list);
472 btrfs_put_ordered_extent(ordered);
473 }
474}
475
476void btrfs_submit_logged_extents(struct list_head *logged_list,
477 struct btrfs_root *log)
478{
479 int index = log->log_transid % 2;
480
481 spin_lock_irq(&log->log_extents_lock[index]);
482 list_splice_tail(logged_list, &log->logged_list[index]);
483 spin_unlock_irq(&log->log_extents_lock[index]);
484}
485
Josef Bacik50d9aa92014-11-21 14:52:38 -0500486void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
487 struct btrfs_root *log, u64 transid)
Josef Bacik2ab28f32012-10-12 15:27:49 -0400488{
489 struct btrfs_ordered_extent *ordered;
490 int index = transid % 2;
491
492 spin_lock_irq(&log->log_extents_lock[index]);
493 while (!list_empty(&log->logged_list[index])) {
Josef Bacik161c35492015-09-24 16:17:39 -0400494 struct inode *inode;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400495 ordered = list_first_entry(&log->logged_list[index],
496 struct btrfs_ordered_extent,
497 log_list);
498 list_del_init(&ordered->log_list);
Josef Bacik161c35492015-09-24 16:17:39 -0400499 inode = ordered->inode;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400500 spin_unlock_irq(&log->log_extents_lock[index]);
Liu Bo98ce2de2014-07-17 16:08:36 +0800501
502 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
503 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
Liu Bo98ce2de2014-07-17 16:08:36 +0800504 u64 start = ordered->file_offset;
505 u64 end = ordered->file_offset + ordered->len - 1;
506
507 WARN_ON(!inode);
508 filemap_fdatawrite_range(inode->i_mapping, start, end);
509 }
Josef Bacik2ab28f32012-10-12 15:27:49 -0400510 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
511 &ordered->flags));
Liu Bo98ce2de2014-07-17 16:08:36 +0800512
Filipe Manana7558c8b2015-04-17 17:08:37 +0100513 /*
Josef Bacik161c35492015-09-24 16:17:39 -0400514 * In order to keep us from losing our ordered extent
515 * information when committing the transaction we have to make
516 * sure that any logged extents are completed when we go to
517 * commit the transaction. To do this we simply increase the
518 * current transactions pending_ordered counter and decrement it
519 * when the ordered extent completes.
Filipe Manana7558c8b2015-04-17 17:08:37 +0100520 */
Josef Bacik161c35492015-09-24 16:17:39 -0400521 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
522 struct btrfs_ordered_inode_tree *tree;
Filipe Manana7558c8b2015-04-17 17:08:37 +0100523
Josef Bacik161c35492015-09-24 16:17:39 -0400524 tree = &BTRFS_I(inode)->ordered_tree;
525 spin_lock_irq(&tree->lock);
526 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
527 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
528 atomic_inc(&trans->transaction->pending_ordered);
529 }
530 spin_unlock_irq(&tree->lock);
531 }
532 btrfs_put_ordered_extent(ordered);
Josef Bacik2ab28f32012-10-12 15:27:49 -0400533 spin_lock_irq(&log->log_extents_lock[index]);
534 }
535 spin_unlock_irq(&log->log_extents_lock[index]);
536}
537
538void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
539{
540 struct btrfs_ordered_extent *ordered;
541 int index = transid % 2;
542
543 spin_lock_irq(&log->log_extents_lock[index]);
544 while (!list_empty(&log->logged_list[index])) {
545 ordered = list_first_entry(&log->logged_list[index],
546 struct btrfs_ordered_extent,
547 log_list);
548 list_del_init(&ordered->log_list);
549 spin_unlock_irq(&log->log_extents_lock[index]);
550 btrfs_put_ordered_extent(ordered);
551 spin_lock_irq(&log->log_extents_lock[index]);
552 }
553 spin_unlock_irq(&log->log_extents_lock[index]);
554}
555
Chris Masoneb84ae02008-07-17 13:53:27 -0400556/*
557 * used to drop a reference on an ordered extent. This will free
558 * the extent if the last reference is dropped
559 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100560void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400561{
Chris Masonba1da2f2008-07-17 12:54:15 -0400562 struct list_head *cur;
563 struct btrfs_ordered_sum *sum;
564
liubo1abe9b82011-03-24 11:18:59 +0000565 trace_btrfs_ordered_extent_put(entry->inode, entry);
566
Chris Masonba1da2f2008-07-17 12:54:15 -0400567 if (atomic_dec_and_test(&entry->refs)) {
Filipe Manana61de7182015-07-01 12:13:10 +0100568 ASSERT(list_empty(&entry->log_list));
569 ASSERT(list_empty(&entry->trans_list));
570 ASSERT(list_empty(&entry->root_extent_list));
571 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
Josef Bacik5fd02042012-05-02 14:00:54 -0400572 if (entry->inode)
573 btrfs_add_delayed_iput(entry->inode);
Chris Masond3977122009-01-05 21:25:51 -0500574 while (!list_empty(&entry->list)) {
Chris Masonba1da2f2008-07-17 12:54:15 -0400575 cur = entry->list.next;
576 sum = list_entry(cur, struct btrfs_ordered_sum, list);
577 list_del(&sum->list);
578 kfree(sum);
579 }
Miao Xie6352b912012-09-06 04:01:51 -0600580 kmem_cache_free(btrfs_ordered_extent_cache, entry);
Chris Masonba1da2f2008-07-17 12:54:15 -0400581 }
Chris Masondc17ff82008-01-08 15:46:30 -0500582}
583
Chris Masoneb84ae02008-07-17 13:53:27 -0400584/*
585 * remove an ordered extent from the tree. No references are dropped
Josef Bacik5fd02042012-05-02 14:00:54 -0400586 * and waiters are woken up.
Chris Masoneb84ae02008-07-17 13:53:27 -0400587 */
Josef Bacik5fd02042012-05-02 14:00:54 -0400588void btrfs_remove_ordered_extent(struct inode *inode,
589 struct btrfs_ordered_extent *entry)
Chris Masondc17ff82008-01-08 15:46:30 -0500590{
Chris Masone6dcd2d2008-07-17 12:53:50 -0400591 struct btrfs_ordered_inode_tree *tree;
Josef Bacik287a0ab2010-03-19 18:07:23 +0000592 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondc17ff82008-01-08 15:46:30 -0500593 struct rb_node *node;
Josef Bacik161c35492015-09-24 16:17:39 -0400594 bool dec_pending_ordered = false;
Chris Masondc17ff82008-01-08 15:46:30 -0500595
Chris Masone6dcd2d2008-07-17 12:53:50 -0400596 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400597 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400598 node = &entry->rb_node;
Chris Masondc17ff82008-01-08 15:46:30 -0500599 rb_erase(node, &tree->tree);
Filipe Manana61de7182015-07-01 12:13:10 +0100600 RB_CLEAR_NODE(node);
Filipe David Borba Manana1b8e7e42013-11-22 18:54:58 +0000601 if (tree->last == node)
602 tree->last = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400603 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
Josef Bacik161c35492015-09-24 16:17:39 -0400604 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
605 dec_pending_ordered = true;
Josef Bacik5fd02042012-05-02 14:00:54 -0400606 spin_unlock_irq(&tree->lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400607
Josef Bacik161c35492015-09-24 16:17:39 -0400608 /*
609 * The current running transaction is waiting on us, we need to let it
610 * know that we're complete and wake it up.
611 */
612 if (dec_pending_ordered) {
613 struct btrfs_transaction *trans;
614
615 /*
616 * The checks for trans are just a formality, it should be set,
617 * but if it isn't we don't want to deref/assert under the spin
618 * lock, so be nice and check if trans is set, but ASSERT() so
619 * if it isn't set a developer will notice.
620 */
621 spin_lock(&root->fs_info->trans_lock);
622 trans = root->fs_info->running_transaction;
623 if (trans)
624 atomic_inc(&trans->use_count);
625 spin_unlock(&root->fs_info->trans_lock);
626
627 ASSERT(trans);
628 if (trans) {
629 if (atomic_dec_and_test(&trans->pending_ordered))
630 wake_up(&trans->pending_wait);
631 btrfs_put_transaction(trans);
632 }
633 }
634
Miao Xie199c2a92013-05-15 07:48:23 +0000635 spin_lock(&root->ordered_extent_lock);
Chris Mason3eaa2882008-07-24 11:57:52 -0400636 list_del_init(&entry->root_extent_list);
Miao Xie199c2a92013-05-15 07:48:23 +0000637 root->nr_ordered_extents--;
Chris Mason5a3f23d2009-03-31 13:27:11 -0400638
liubo1abe9b82011-03-24 11:18:59 +0000639 trace_btrfs_ordered_extent_remove(inode, entry);
640
Miao Xie199c2a92013-05-15 07:48:23 +0000641 if (!root->nr_ordered_extents) {
642 spin_lock(&root->fs_info->ordered_root_lock);
643 BUG_ON(list_empty(&root->ordered_root));
644 list_del_init(&root->ordered_root);
645 spin_unlock(&root->fs_info->ordered_root_lock);
646 }
647 spin_unlock(&root->ordered_extent_lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400648 wake_up(&entry->wait);
Chris Mason81d7ed22008-04-25 08:51:48 -0400649}
Chris Masone6dcd2d2008-07-17 12:53:50 -0400650
Qu Wenruod458b052014-02-28 10:46:19 +0800651static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
Miao Xie9afab882012-10-25 09:41:36 +0000652{
653 struct btrfs_ordered_extent *ordered;
654
655 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
656 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
657 complete(&ordered->completion);
658}
659
Chris Masond352ac62008-09-29 15:18:18 -0400660/*
661 * wait for all the ordered extents in a root. This is done when balancing
662 * space between drives.
663 */
Filipe Manana578def72016-04-26 15:36:38 +0100664int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
665 const u64 range_start, const u64 range_len)
Chris Mason3eaa2882008-07-24 11:57:52 -0400666{
Filipe Manana578def72016-04-26 15:36:38 +0100667 LIST_HEAD(splice);
668 LIST_HEAD(skipped);
669 LIST_HEAD(works);
Miao Xie9afab882012-10-25 09:41:36 +0000670 struct btrfs_ordered_extent *ordered, *next;
Miao Xieb0244192013-11-04 23:13:25 +0800671 int count = 0;
Filipe Manana578def72016-04-26 15:36:38 +0100672 const u64 range_end = range_start + range_len;
Chris Mason3eaa2882008-07-24 11:57:52 -0400673
Miao Xie31f3d252014-03-06 13:55:02 +0800674 mutex_lock(&root->ordered_extent_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000675 spin_lock(&root->ordered_extent_lock);
676 list_splice_init(&root->ordered_extents, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800677 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000678 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
679 root_extent_list);
Filipe Manana578def72016-04-26 15:36:38 +0100680
681 if (range_end <= ordered->start ||
682 ordered->start + ordered->disk_len <= range_start) {
683 list_move_tail(&ordered->root_extent_list, &skipped);
684 cond_resched_lock(&root->ordered_extent_lock);
685 continue;
686 }
687
Miao Xie199c2a92013-05-15 07:48:23 +0000688 list_move_tail(&ordered->root_extent_list,
689 &root->ordered_extents);
Miao Xie199c2a92013-05-15 07:48:23 +0000690 atomic_inc(&ordered->refs);
691 spin_unlock(&root->ordered_extent_lock);
692
Qu Wenruoa44903a2014-02-28 10:46:09 +0800693 btrfs_init_work(&ordered->flush_work,
Liu Bo9e0af232014-08-15 23:36:53 +0800694 btrfs_flush_delalloc_helper,
Qu Wenruoa44903a2014-02-28 10:46:09 +0800695 btrfs_run_ordered_extent_work, NULL, NULL);
Miao Xie199c2a92013-05-15 07:48:23 +0000696 list_add_tail(&ordered->work_list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +0800697 btrfs_queue_work(root->fs_info->flush_workers,
698 &ordered->flush_work);
Miao Xie199c2a92013-05-15 07:48:23 +0000699
Miao Xie9afab882012-10-25 09:41:36 +0000700 cond_resched();
Miao Xie199c2a92013-05-15 07:48:23 +0000701 spin_lock(&root->ordered_extent_lock);
Miao Xieb0244192013-11-04 23:13:25 +0800702 if (nr != -1)
703 nr--;
704 count++;
Chris Mason3eaa2882008-07-24 11:57:52 -0400705 }
Filipe Manana578def72016-04-26 15:36:38 +0100706 list_splice_tail(&skipped, &root->ordered_extents);
Miao Xieb0244192013-11-04 23:13:25 +0800707 list_splice_tail(&splice, &root->ordered_extents);
Miao Xie199c2a92013-05-15 07:48:23 +0000708 spin_unlock(&root->ordered_extent_lock);
Miao Xie9afab882012-10-25 09:41:36 +0000709
710 list_for_each_entry_safe(ordered, next, &works, work_list) {
711 list_del_init(&ordered->work_list);
712 wait_for_completion(&ordered->completion);
Miao Xie9afab882012-10-25 09:41:36 +0000713 btrfs_put_ordered_extent(ordered);
Miao Xie9afab882012-10-25 09:41:36 +0000714 cond_resched();
715 }
Miao Xie31f3d252014-03-06 13:55:02 +0800716 mutex_unlock(&root->ordered_extent_mutex);
Miao Xieb0244192013-11-04 23:13:25 +0800717
718 return count;
Chris Mason3eaa2882008-07-24 11:57:52 -0400719}
720
Filipe Manana578def72016-04-26 15:36:38 +0100721void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
722 const u64 range_start, const u64 range_len)
Miao Xie199c2a92013-05-15 07:48:23 +0000723{
724 struct btrfs_root *root;
725 struct list_head splice;
Miao Xieb0244192013-11-04 23:13:25 +0800726 int done;
Miao Xie199c2a92013-05-15 07:48:23 +0000727
728 INIT_LIST_HEAD(&splice);
729
Miao Xie8b9d83c2014-03-06 13:54:55 +0800730 mutex_lock(&fs_info->ordered_operations_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000731 spin_lock(&fs_info->ordered_root_lock);
732 list_splice_init(&fs_info->ordered_roots, &splice);
Miao Xieb0244192013-11-04 23:13:25 +0800733 while (!list_empty(&splice) && nr) {
Miao Xie199c2a92013-05-15 07:48:23 +0000734 root = list_first_entry(&splice, struct btrfs_root,
735 ordered_root);
736 root = btrfs_grab_fs_root(root);
737 BUG_ON(!root);
738 list_move_tail(&root->ordered_root,
739 &fs_info->ordered_roots);
740 spin_unlock(&fs_info->ordered_root_lock);
741
Filipe Manana578def72016-04-26 15:36:38 +0100742 done = btrfs_wait_ordered_extents(root, nr,
743 range_start, range_len);
Miao Xie199c2a92013-05-15 07:48:23 +0000744 btrfs_put_fs_root(root);
745
746 spin_lock(&fs_info->ordered_root_lock);
Miao Xieb0244192013-11-04 23:13:25 +0800747 if (nr != -1) {
748 nr -= done;
749 WARN_ON(nr < 0);
750 }
Miao Xie199c2a92013-05-15 07:48:23 +0000751 }
Miao Xie931aa872013-11-14 17:33:21 +0800752 list_splice_tail(&splice, &fs_info->ordered_roots);
Miao Xie199c2a92013-05-15 07:48:23 +0000753 spin_unlock(&fs_info->ordered_root_lock);
Miao Xie8b9d83c2014-03-06 13:54:55 +0800754 mutex_unlock(&fs_info->ordered_operations_mutex);
Miao Xie199c2a92013-05-15 07:48:23 +0000755}
756
Chris Masoneb84ae02008-07-17 13:53:27 -0400757/*
758 * Used to start IO or wait for a given ordered extent to finish.
759 *
760 * If wait is one, this effectively waits on page writeback for all the pages
761 * in the extent, and it waits on the io completion code to insert
762 * metadata into the btree corresponding to the extent
763 */
764void btrfs_start_ordered_extent(struct inode *inode,
765 struct btrfs_ordered_extent *entry,
766 int wait)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400767{
768 u64 start = entry->file_offset;
769 u64 end = start + entry->len - 1;
770
liubo1abe9b82011-03-24 11:18:59 +0000771 trace_btrfs_ordered_extent_start(inode, entry);
772
Chris Masoneb84ae02008-07-17 13:53:27 -0400773 /*
774 * pages in the range can be dirty, clean or writeback. We
775 * start IO on any dirty ones so the wait doesn't stall waiting
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300776 * for the flusher thread to find them
Chris Masoneb84ae02008-07-17 13:53:27 -0400777 */
Josef Bacik4b46fce2010-05-23 11:00:55 -0400778 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
779 filemap_fdatawrite_range(inode->i_mapping, start, end);
Chris Masonc8b97812008-10-29 14:49:59 -0400780 if (wait) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400781 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
782 &entry->flags));
Chris Masonc8b97812008-10-29 14:49:59 -0400783 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400784}
785
Chris Masoneb84ae02008-07-17 13:53:27 -0400786/*
787 * Used to wait on ordered extents across a large range of bytes.
788 */
Josef Bacik0ef8b722013-10-25 16:13:35 -0400789int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400790{
Josef Bacik0ef8b722013-10-25 16:13:35 -0400791 int ret = 0;
Filipe Manana28aeeac2015-05-05 19:03:10 +0100792 int ret_wb = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400793 u64 end;
Chris Masone5a22172008-07-18 20:42:20 -0400794 u64 orig_end;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400795 struct btrfs_ordered_extent *ordered;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400796
Chris Masone5a22172008-07-18 20:42:20 -0400797 if (start + len < start) {
Chris Masonf4219502008-07-22 11:18:09 -0400798 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400799 } else {
800 orig_end = start + len - 1;
Chris Masonf4219502008-07-22 11:18:09 -0400801 if (orig_end > INT_LIMIT(loff_t))
802 orig_end = INT_LIMIT(loff_t);
Chris Masone5a22172008-07-18 20:42:20 -0400803 }
Josef Bacik551ebb22012-04-23 14:41:09 -0400804
Chris Masone5a22172008-07-18 20:42:20 -0400805 /* start IO across the range first to instantiate any delalloc
806 * extents
807 */
Filipe Manana728404d2014-10-10 09:43:11 +0100808 ret = btrfs_fdatawrite_range(inode, start, orig_end);
Josef Bacik0ef8b722013-10-25 16:13:35 -0400809 if (ret)
810 return ret;
Filipe Manana728404d2014-10-10 09:43:11 +0100811
Filipe Manana28aeeac2015-05-05 19:03:10 +0100812 /*
813 * If we have a writeback error don't return immediately. Wait first
814 * for any ordered extents that haven't completed yet. This is to make
815 * sure no one can dirty the same page ranges and call writepages()
816 * before the ordered extents complete - to avoid failures (-EEXIST)
817 * when adding the new ordered extents to the ordered tree.
818 */
819 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
Chris Masonf4219502008-07-22 11:18:09 -0400820
821 end = orig_end;
Chris Masond3977122009-01-05 21:25:51 -0500822 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400823 ordered = btrfs_lookup_first_ordered_extent(inode, end);
Chris Masond3977122009-01-05 21:25:51 -0500824 if (!ordered)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400825 break;
Chris Masone5a22172008-07-18 20:42:20 -0400826 if (ordered->file_offset > orig_end) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400827 btrfs_put_ordered_extent(ordered);
828 break;
829 }
Filipe David Borba Mananab52abf12013-11-06 15:12:40 +0000830 if (ordered->file_offset + ordered->len <= start) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400831 btrfs_put_ordered_extent(ordered);
832 break;
833 }
Chris Masone5a22172008-07-18 20:42:20 -0400834 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400835 end = ordered->file_offset;
Josef Bacik0ef8b722013-10-25 16:13:35 -0400836 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
837 ret = -EIO;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400838 btrfs_put_ordered_extent(ordered);
Josef Bacik0ef8b722013-10-25 16:13:35 -0400839 if (ret || end == 0 || end == start)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400840 break;
841 end--;
842 }
Filipe Manana28aeeac2015-05-05 19:03:10 +0100843 return ret_wb ? ret_wb : ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400844}
845
Chris Masoneb84ae02008-07-17 13:53:27 -0400846/*
847 * find an ordered extent corresponding to file_offset. return NULL if
848 * nothing is found, otherwise take a reference on the extent and return it
849 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400850struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
851 u64 file_offset)
852{
853 struct btrfs_ordered_inode_tree *tree;
854 struct rb_node *node;
855 struct btrfs_ordered_extent *entry = NULL;
856
857 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400858 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400859 node = tree_search(tree, file_offset);
860 if (!node)
861 goto out;
862
863 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
864 if (!offset_in_entry(entry, file_offset))
865 entry = NULL;
866 if (entry)
867 atomic_inc(&entry->refs);
868out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400869 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400870 return entry;
871}
872
Josef Bacik4b46fce2010-05-23 11:00:55 -0400873/* Since the DIO code tries to lock a wide area we need to look for any ordered
874 * extents that exist in the range, rather than just the start of the range.
875 */
876struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
877 u64 file_offset,
878 u64 len)
879{
880 struct btrfs_ordered_inode_tree *tree;
881 struct rb_node *node;
882 struct btrfs_ordered_extent *entry = NULL;
883
884 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400885 spin_lock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400886 node = tree_search(tree, file_offset);
887 if (!node) {
888 node = tree_search(tree, file_offset + len);
889 if (!node)
890 goto out;
891 }
892
893 while (1) {
894 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
895 if (range_overlaps(entry, file_offset, len))
896 break;
897
898 if (entry->file_offset >= file_offset + len) {
899 entry = NULL;
900 break;
901 }
902 entry = NULL;
903 node = rb_next(node);
904 if (!node)
905 break;
906 }
907out:
908 if (entry)
909 atomic_inc(&entry->refs);
Josef Bacik5fd02042012-05-02 14:00:54 -0400910 spin_unlock_irq(&tree->lock);
Josef Bacik4b46fce2010-05-23 11:00:55 -0400911 return entry;
912}
913
Filipe Mananab659ef02015-03-31 14:16:52 +0100914bool btrfs_have_ordered_extents_in_range(struct inode *inode,
915 u64 file_offset,
916 u64 len)
917{
918 struct btrfs_ordered_extent *oe;
919
920 oe = btrfs_lookup_ordered_range(inode, file_offset, len);
921 if (oe) {
922 btrfs_put_ordered_extent(oe);
923 return true;
924 }
925 return false;
926}
927
Chris Masoneb84ae02008-07-17 13:53:27 -0400928/*
929 * lookup and return any extent before 'file_offset'. NULL is returned
930 * if none is found
931 */
Chris Masone6dcd2d2008-07-17 12:53:50 -0400932struct btrfs_ordered_extent *
Chris Masond3977122009-01-05 21:25:51 -0500933btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
Chris Masone6dcd2d2008-07-17 12:53:50 -0400934{
935 struct btrfs_ordered_inode_tree *tree;
936 struct rb_node *node;
937 struct btrfs_ordered_extent *entry = NULL;
938
939 tree = &BTRFS_I(inode)->ordered_tree;
Josef Bacik5fd02042012-05-02 14:00:54 -0400940 spin_lock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400941 node = tree_search(tree, file_offset);
942 if (!node)
943 goto out;
944
945 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
946 atomic_inc(&entry->refs);
947out:
Josef Bacik5fd02042012-05-02 14:00:54 -0400948 spin_unlock_irq(&tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400949 return entry;
950}
Chris Masondbe674a2008-07-17 12:54:05 -0400951
Chris Masoneb84ae02008-07-17 13:53:27 -0400952/*
953 * After an extent is done, call this to conditionally update the on disk
954 * i_size. i_size is updated to cover any fully written part of the file.
955 */
Yan, Zhengc2167752009-11-12 09:34:21 +0000956int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
Chris Masondbe674a2008-07-17 12:54:05 -0400957 struct btrfs_ordered_extent *ordered)
958{
959 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Masondbe674a2008-07-17 12:54:05 -0400960 u64 disk_i_size;
961 u64 new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +0000962 u64 i_size = i_size_read(inode);
Chris Masondbe674a2008-07-17 12:54:05 -0400963 struct rb_node *node;
Yan, Zhengc2167752009-11-12 09:34:21 +0000964 struct rb_node *prev = NULL;
Chris Masondbe674a2008-07-17 12:54:05 -0400965 struct btrfs_ordered_extent *test;
Yan, Zhengc2167752009-11-12 09:34:21 +0000966 int ret = 1;
967
Josef Bacik5fd02042012-05-02 14:00:54 -0400968 spin_lock_irq(&tree->lock);
Josef Bacik77cef2e2013-08-29 13:57:21 -0400969 if (ordered) {
970 offset = entry_end(ordered);
971 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
972 offset = min(offset,
973 ordered->file_offset +
974 ordered->truncated_len);
975 } else {
976 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
977 }
Chris Masondbe674a2008-07-17 12:54:05 -0400978 disk_i_size = BTRFS_I(inode)->disk_i_size;
979
Yan, Zhengc2167752009-11-12 09:34:21 +0000980 /* truncate file */
981 if (disk_i_size > i_size) {
982 BTRFS_I(inode)->disk_i_size = i_size;
983 ret = 0;
984 goto out;
985 }
986
Chris Masondbe674a2008-07-17 12:54:05 -0400987 /*
988 * if the disk i_size is already at the inode->i_size, or
989 * this ordered extent is inside the disk i_size, we're done
990 */
Josef Bacik5d1f4022013-01-30 14:17:31 -0500991 if (disk_i_size == i_size)
Chris Masondbe674a2008-07-17 12:54:05 -0400992 goto out;
Josef Bacik5d1f4022013-01-30 14:17:31 -0500993
994 /*
995 * We still need to update disk_i_size if outstanding_isize is greater
996 * than disk_i_size.
997 */
998 if (offset <= disk_i_size &&
999 (!ordered || ordered->outstanding_isize <= disk_i_size))
1000 goto out;
Chris Masondbe674a2008-07-17 12:54:05 -04001001
1002 /*
Chris Masondbe674a2008-07-17 12:54:05 -04001003 * walk backward from this ordered extent to disk_i_size.
1004 * if we find an ordered extent then we can't update disk i_size
1005 * yet
1006 */
Yan, Zhengc2167752009-11-12 09:34:21 +00001007 if (ordered) {
1008 node = rb_prev(&ordered->rb_node);
1009 } else {
1010 prev = tree_search(tree, offset);
1011 /*
1012 * we insert file extents without involving ordered struct,
1013 * so there should be no ordered struct cover this offset
1014 */
1015 if (prev) {
1016 test = rb_entry(prev, struct btrfs_ordered_extent,
1017 rb_node);
1018 BUG_ON(offset_in_entry(test, offset));
1019 }
1020 node = prev;
1021 }
Josef Bacik5fd02042012-05-02 14:00:54 -04001022 for (; node; node = rb_prev(node)) {
Chris Masondbe674a2008-07-17 12:54:05 -04001023 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
Josef Bacik5fd02042012-05-02 14:00:54 -04001024
Adam Buchbinderbb7ab3b2016-03-04 11:23:12 -08001025 /* We treat this entry as if it doesn't exist */
Josef Bacik5fd02042012-05-02 14:00:54 -04001026 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1027 continue;
Chris Masondbe674a2008-07-17 12:54:05 -04001028 if (test->file_offset + test->len <= disk_i_size)
1029 break;
Yan, Zhengc2167752009-11-12 09:34:21 +00001030 if (test->file_offset >= i_size)
Chris Masondbe674a2008-07-17 12:54:05 -04001031 break;
Josef Bacik59fe4f42013-01-30 14:31:31 -05001032 if (entry_end(test) > disk_i_size) {
Miao Xieb9a8cc52012-09-06 04:01:21 -06001033 /*
1034 * we don't update disk_i_size now, so record this
1035 * undealt i_size. Or we will not know the real
1036 * i_size.
1037 */
1038 if (test->outstanding_isize < offset)
1039 test->outstanding_isize = offset;
1040 if (ordered &&
1041 ordered->outstanding_isize >
1042 test->outstanding_isize)
1043 test->outstanding_isize =
1044 ordered->outstanding_isize;
Chris Masondbe674a2008-07-17 12:54:05 -04001045 goto out;
Miao Xieb9a8cc52012-09-06 04:01:21 -06001046 }
Chris Masondbe674a2008-07-17 12:54:05 -04001047 }
Yan, Zhengc2167752009-11-12 09:34:21 +00001048 new_i_size = min_t(u64, offset, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -04001049
1050 /*
Miao Xieb9a8cc52012-09-06 04:01:21 -06001051 * Some ordered extents may completed before the current one, and
1052 * we hold the real i_size in ->outstanding_isize.
Chris Masondbe674a2008-07-17 12:54:05 -04001053 */
Miao Xieb9a8cc52012-09-06 04:01:21 -06001054 if (ordered && ordered->outstanding_isize > new_i_size)
1055 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
Chris Masondbe674a2008-07-17 12:54:05 -04001056 BTRFS_I(inode)->disk_i_size = new_i_size;
Yan, Zhengc2167752009-11-12 09:34:21 +00001057 ret = 0;
Chris Masondbe674a2008-07-17 12:54:05 -04001058out:
Yan, Zhengc2167752009-11-12 09:34:21 +00001059 /*
Josef Bacik5fd02042012-05-02 14:00:54 -04001060 * We need to do this because we can't remove ordered extents until
1061 * after the i_disk_size has been updated and then the inode has been
1062 * updated to reflect the change, so we need to tell anybody who finds
1063 * this ordered extent that we've already done all the real work, we
1064 * just haven't completed all the other work.
Yan, Zhengc2167752009-11-12 09:34:21 +00001065 */
1066 if (ordered)
Josef Bacik5fd02042012-05-02 14:00:54 -04001067 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1068 spin_unlock_irq(&tree->lock);
Yan, Zhengc2167752009-11-12 09:34:21 +00001069 return ret;
Chris Masondbe674a2008-07-17 12:54:05 -04001070}
Chris Masonba1da2f2008-07-17 12:54:15 -04001071
Chris Masoneb84ae02008-07-17 13:53:27 -04001072/*
1073 * search the ordered extents for one corresponding to 'offset' and
1074 * try to find a checksum. This is used because we allow pages to
1075 * be reclaimed before their checksum is actually put into the btree
1076 */
Chris Masond20f7042008-12-08 16:58:54 -05001077int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
Miao Xiee4100d92013-04-05 07:20:56 +00001078 u32 *sum, int len)
Chris Masonba1da2f2008-07-17 12:54:15 -04001079{
1080 struct btrfs_ordered_sum *ordered_sum;
Chris Masonba1da2f2008-07-17 12:54:15 -04001081 struct btrfs_ordered_extent *ordered;
1082 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
Chris Mason3edf7d32008-07-18 06:17:13 -04001083 unsigned long num_sectors;
1084 unsigned long i;
1085 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
Miao Xiee4100d92013-04-05 07:20:56 +00001086 int index = 0;
Chris Masonba1da2f2008-07-17 12:54:15 -04001087
1088 ordered = btrfs_lookup_ordered_extent(inode, offset);
1089 if (!ordered)
Miao Xiee4100d92013-04-05 07:20:56 +00001090 return 0;
Chris Masonba1da2f2008-07-17 12:54:15 -04001091
Josef Bacik5fd02042012-05-02 14:00:54 -04001092 spin_lock_irq(&tree->lock);
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001093 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
Miao Xiee4100d92013-04-05 07:20:56 +00001094 if (disk_bytenr >= ordered_sum->bytenr &&
1095 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1096 i = (disk_bytenr - ordered_sum->bytenr) >>
1097 inode->i_sb->s_blocksize_bits;
Miao Xiee4100d92013-04-05 07:20:56 +00001098 num_sectors = ordered_sum->len >>
1099 inode->i_sb->s_blocksize_bits;
Miao Xief51a4a12013-06-19 10:36:09 +08001100 num_sectors = min_t(int, len - index, num_sectors - i);
1101 memcpy(sum + index, ordered_sum->sums + i,
1102 num_sectors);
1103
1104 index += (int)num_sectors;
1105 if (index == len)
1106 goto out;
1107 disk_bytenr += num_sectors * sectorsize;
Chris Masonba1da2f2008-07-17 12:54:15 -04001108 }
1109 }
1110out:
Josef Bacik5fd02042012-05-02 14:00:54 -04001111 spin_unlock_irq(&tree->lock);
Chris Mason89642222008-07-24 09:41:53 -04001112 btrfs_put_ordered_extent(ordered);
Miao Xiee4100d92013-04-05 07:20:56 +00001113 return index;
Chris Masonba1da2f2008-07-17 12:54:15 -04001114}
1115
Miao Xie6352b912012-09-06 04:01:51 -06001116int __init ordered_data_init(void)
1117{
1118 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1119 sizeof(struct btrfs_ordered_extent), 0,
1120 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1121 NULL);
1122 if (!btrfs_ordered_extent_cache)
1123 return -ENOMEM;
Miao Xie25287e02012-10-25 09:31:03 +00001124
Miao Xie6352b912012-09-06 04:01:51 -06001125 return 0;
1126}
1127
1128void ordered_data_exit(void)
1129{
Kinglong Mee5598e902016-01-29 21:36:35 +08001130 kmem_cache_destroy(btrfs_ordered_extent_cache);
Miao Xie6352b912012-09-06 04:01:51 -06001131}