1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
|
From ff1eae7f882b8f12380e0c35a9a73b672583cd4c Mon Sep 17 00:00:00 2001
From: N Balachandran <nbalacha@redhat.com>
Date: Tue, 01 Oct 2019 17:37:15 +0530
Subject: [PATCH] cluster/dht: Correct fd processing loop
The fd processing loops in the
dht_migration_complete_check_task and the
dht_rebalance_inprogress_task functions were unsafe
and could cause an open to be sent on an already freed
fd. This has been fixed.
> Change-Id: I0a3c7d2fba314089e03dfd704f9dceb134749540
> Fixes: bz#1757399
> Signed-off-by: N Balachandran <nbalacha@redhat.com>
> (cherry picked from commit 9b15867070b0cc241ab165886292ecffc3bc0aed)
Change-Id: I0a3c7d2fba314089e03dfd704f9dceb134749540
Fixes: bz#1786983
Signed-off-by: Mohit Agrawal <moagrawa@redhat.com>
---
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index acad493..4f7370d 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -1290,6 +1290,7 @@
fd_t *tmp = NULL;
uint64_t tmp_miginfo = 0;
dht_migrate_info_t *miginfo = NULL;
+ gf_boolean_t skip_open = _gf_false;
int open_failed = 0;
this = THIS;
@@ -1428,24 +1429,34 @@
* the loop will cause the destruction of the fd. So we need to
* iterate the list safely because iter_fd cannot be trusted.
*/
- list_for_each_entry_safe(iter_fd, tmp, &inode->fd_list, inode_list)
- {
- if (fd_is_anonymous(iter_fd))
- continue;
-
- if (dht_fd_open_on_dst(this, iter_fd, dst_node))
- continue;
-
+ iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+ while (&iter_fd->inode_list != (&inode->fd_list)) {
+ if (fd_is_anonymous(iter_fd) ||
+ (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+ if (!tmp) {
+ iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+ inode_list);
+ continue;
+ }
+ skip_open = _gf_true;
+ }
/* We need to release the inode->lock before calling
* syncop_open() to avoid possible deadlocks. However this
* can cause the iter_fd to be released by other threads.
* To avoid this, we take a reference before releasing the
* lock.
*/
- __fd_ref(iter_fd);
+ fd_ref(iter_fd);
UNLOCK(&inode->lock);
+ if (tmp) {
+ fd_unref(tmp);
+ tmp = NULL;
+ }
+ if (skip_open)
+ goto next;
+
/* flags for open are stripped down to allow following the
* new location of the file, otherwise we can get EEXIST or
* truncate the file again as rebalance is moving the data */
@@ -1467,9 +1478,11 @@
dht_fd_ctx_set(this, iter_fd, dst_node);
}
- fd_unref(iter_fd);
-
+ next:
LOCK(&inode->lock);
+ skip_open = _gf_false;
+ tmp = iter_fd;
+ iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
}
SYNCTASK_SETID(frame->root->uid, frame->root->gid);
@@ -1482,6 +1495,10 @@
unlock:
UNLOCK(&inode->lock);
+ if (tmp) {
+ fd_unref(tmp);
+ tmp = NULL;
+ }
out:
if (dict) {
@@ -1563,6 +1580,7 @@
int open_failed = 0;
uint64_t tmp_miginfo = 0;
dht_migrate_info_t *miginfo = NULL;
+ gf_boolean_t skip_open = _gf_false;
this = THIS;
frame = data;
@@ -1683,24 +1701,40 @@
* the loop will cause the destruction of the fd. So we need to
* iterate the list safely because iter_fd cannot be trusted.
*/
- list_for_each_entry_safe(iter_fd, tmp, &inode->fd_list, inode_list)
- {
- if (fd_is_anonymous(iter_fd))
- continue;
-
- if (dht_fd_open_on_dst(this, iter_fd, dst_node))
- continue;
-
+ iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+ while (&iter_fd->inode_list != (&inode->fd_list)) {
/* We need to release the inode->lock before calling
* syncop_open() to avoid possible deadlocks. However this
* can cause the iter_fd to be released by other threads.
* To avoid this, we take a reference before releasing the
* lock.
*/
- __fd_ref(iter_fd);
+ if (fd_is_anonymous(iter_fd) ||
+ (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+ if (!tmp) {
+ iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+ inode_list);
+ continue;
+ }
+ skip_open = _gf_true;
+ }
+
+ /* Yes, this is ugly but there isn't a cleaner way to do this
+ * the fd_ref is an atomic increment so not too bad. We want to
+ * reduce the number of inode locks and unlocks.
+ */
+
+ fd_ref(iter_fd);
UNLOCK(&inode->lock);
+ if (tmp) {
+ fd_unref(tmp);
+ tmp = NULL;
+ }
+ if (skip_open)
+ goto next;
+
/* flags for open are stripped down to allow following the
* new location of the file, otherwise we can get EEXIST or
* truncate the file again as rebalance is moving the data */
@@ -1721,9 +1755,11 @@
dht_fd_ctx_set(this, iter_fd, dst_node);
}
- fd_unref(iter_fd);
-
+ next:
LOCK(&inode->lock);
+ skip_open = _gf_false;
+ tmp = iter_fd;
+ iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
}
SYNCTASK_SETID(frame->root->uid, frame->root->gid);
@@ -1731,6 +1767,10 @@
unlock:
UNLOCK(&inode->lock);
+ if (tmp) {
+ fd_unref(tmp);
+ tmp = NULL;
+ }
if (open_failed) {
ret = -1;
goto out;
|
