After installing Redhat 6.0 on a production server, I noticed a problem with syslogd. syslogd was growing in memory size proportional to the amount of data received. This is a box that is doing lots and lots of syslogging from lots of hosts. Well after some investigation, and examining of the source code, I have isolated the bug. And I am including a sample program that shows the bug. It appears that if you call gethostbyaddr, and if the address is not in /etc/hosts, the gethostbyaddr call leaks memory. If the address is in /etc/hosts (which is what we did to solve the problem for now), it does not leak. The program below will call gethostbyaddr 100,000 times, and then pause so you can take a look at the size of the process. Under 5.2 the process does not grow in size. Under 6.0 the process grows rapidly. test.c ---------------------------- #include <stdio.h> #include <sys/socket.h> #include <netinet/in.h> #include <netdb.h> void main() { unsigned long n = 0x24b445cf; int i; struct hostent *hp = 0, *ohp = 0; struct sockaddr_in f; struct sockaddr_in *f2 = &f; f.sin_addr.s_addr = 0xCF45B424; f.sin_family = AF_INET; for (i=0; i<100000; i++) { hp = gethostbyaddr((char *) &n, 4, 2); if (ohp != hp) printf("HP CHANGED (%d != %d)\n", ohp, hp); ohp = hp; if (!(i % 1000)) printf("."); fflush(stdout); } printf("Press <RETURN>... "); getc(stdin); }
*** This bug has been marked as a duplicate of 3560 ***