summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorGreg Hudson <ghudson@mit.edu>2009-08-28 16:00:54 +0000
committerGreg Hudson <ghudson@mit.edu>2009-08-28 16:00:54 +0000
commit85959def61b4117bf2bbd17f61173d10ef2e7811 (patch)
treed74776a94e7781cd0776dfb07b6b7eedc483d554
parentbb571c5754e838acd15cf8cd2eba535accbc7820 (diff)
downloadkrb5-85959def61b4117bf2bbd17f61173d10ef2e7811.tar.gz
krb5-85959def61b4117bf2bbd17f61173d10ef2e7811.tar.xz
krb5-85959def61b4117bf2bbd17f61173d10ef2e7811.zip
Disable the COPY_FIRST_CANONNAME workaround on Linux glibc 2.4 and
later, since it leaks memory on fixed glibc versions. We will still leak memory on glibc 2.3.4 through 2.3.6 (e.g. RHEL 4) but that's harder to detect. ticket: 6534 git-svn-id: svn://anonsvn.mit.edu/krb5/trunk@22643 dc483132-0cff-0310-8789-dd5450dbe970
-rw-r--r--src/util/support/fake-addrinfo.c21
1 files changed, 16 insertions, 5 deletions
diff --git a/src/util/support/fake-addrinfo.c b/src/util/support/fake-addrinfo.c
index 4b628bb67b..34ce7701bd 100644
--- a/src/util/support/fake-addrinfo.c
+++ b/src/util/support/fake-addrinfo.c
@@ -140,7 +140,15 @@ extern /*@dependent@*/ char *gai_strerror (int code) /*@*/;
#endif
#if defined (__linux__) && defined(HAVE_GETADDRINFO)
-# define COPY_FIRST_CANONNAME
+/* Define COPY_FIRST_CANONNAME for glibc 2.3 and prior. */
+#include <features.h>
+# ifdef __GLIBC_PREREQ
+# if ! __GLIBC_PREREQ(2, 4)
+# define COPY_FIRST_CANONNAME
+# endif
+# else
+# define COPY_FIRST_CANONNAME
+# endif
#endif
#ifdef _AIX
@@ -1157,7 +1165,7 @@ getaddrinfo (const char *name, const char *serv, const struct addrinfo *hint,
return aierr;
}
- /* Linux libc version 6 (libc-2.2.4.so on Debian) is broken.
+ /* Linux libc version 6 prior to 2.3.4 is broken.
RFC 2553 says that when AI_CANONNAME is set, the ai_canonname
flag of the first returned structure has the canonical name of
@@ -1188,9 +1196,12 @@ getaddrinfo (const char *name, const char *serv, const struct addrinfo *hint,
Ref: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=133668 .
Since it's dependent on the target hostname, it's hard to check
- for at configure time. Always do it on Linux for now. When
- they get around to fixing it, add a compile-time or run-time
- check for the glibc version in use.
+ for at configure time. The bug was fixed in glibc 2.3.4.
+ After the fix, the ai_canonname field is allocated, so our
+ workaround leaks memory. We disable the workaround for glibc
+ >= 2.4, but there is no easy way to test for glibc patch
+ versions, so we still leak memory under glibc 2.3.4 through
+ 2.3.6.
Some Windows documentation says that even when AI_CANONNAME is
set, the returned ai_canonname field can be null. The NetBSD