blob: 55a49d72f6870c30b559ea4abf0219152ba4ef90 (
plain)
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
|
From cf2104d4d991361c53f6e2fea93b69de10cd654b Mon Sep 17 00:00:00 2001
From: Federico Simoncelli <fsimonce@redhat.com>
Date: Sat, 15 Nov 2014 14:14:04 +0000
Subject: [PATCH] common: do not unlock rwlock on destruction
According to pthread_rwlock_unlock(3p):
Results are undefined if the read-write lock rwlock is not held
by the calling thread.
and:
https://sourceware.org/bugzilla/show_bug.cgi?id=17561
Calling pthread_rwlock_unlock on an rwlock which is not locked
is undefined.
calling pthread_rwlock_unlock on RWLock destruction could cause
an unknown behavior for two reasons:
- the lock is acquired by another thread (undefined)
- the lock is not acquired (undefined)
Moreover since glibc-2.20 calling pthread_rwlock_unlock on a
rwlock that is not locked results in a SIGILL that kills the
application.
This patch removes the pthread_rwlock_unlock call on destruction
and replaces it with an assertion to check that the RWLock is
not in use.
Any code that relied on the implicit release is now going to
break the assertion, e.g.:
{
RWLock l;
l.get(for_write);
} // implicit release, wrong.
Signed-off-by: Federico Simoncelli <fsimonce@redhat.com>
---
src/common/RWLock.h | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/src/common/RWLock.h b/src/common/RWLock.h
index e647e17..6f0ab8e 100644
--- a/src/common/RWLock.h
+++ b/src/common/RWLock.h
@@ -46,7 +46,9 @@ class RWLock
return (nwlock.read() > 0);
}
virtual ~RWLock() {
- pthread_rwlock_unlock(&L);
+ // The following check is racy but we are about to destroy
+ // the object and we assume that there are no other users.
+ assert(!is_locked());
pthread_rwlock_destroy(&L);
}
|
GitWeb
