Is getnameinfo memory leaking confirmed?

As per question, I'm running into some memory leaking by getnameinfo. I'm using Ubuntu 12.04 (Linux scv 3.2.0-35-generic #55-Ubuntu SMP Wed Dec 5 17:42:16 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux) with gcc version 4.6.3.
I'm linking my server executable with g++ and so far valgrind hasn't reported an issue. I've then added a simple call to getnameinfo to print out what are the network name and port of connecting clients.
And I get the following:

==4425== 
==4425== HEAP SUMMARY:
==4425==     in use at exit: 10 bytes in 1 blocks
==4425==   total heap usage: 4,508 allocs, 4,507 frees, 134,939,153 bytes allocated
==4425== 
==4425== 10 bytes in 1 blocks are definitely lost in loss record 1 of 1
==4425==    at 0x4C2B6CD: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==4425==    by 0x50D7D71: strdup (strdup.c:43)
==4425==    by 0x1484B861: ???
==4425==    by 0x515B871: gethostbyaddr_r@@GLIBC_2.2.5 (getXXbyYY_r.c:256)
==4425==    by 0x5161D06: getnameinfo (getnameinfo.c:223)
==4425==    by 0x404175: solsrv_run (solsrv.c:381)
==4425==    by 0x404DAC: main (main.c:167)
==4425== 
==4425== LEAK SUMMARY:
==4425==    definitely lost: 10 bytes in 1 blocks
==4425==    indirectly lost: 0 bytes in 0 blocks
==4425==      possibly lost: 0 bytes in 0 blocks
==4425==    still reachable: 0 bytes in 0 blocks
==4425==         suppressed: 0 bytes in 0 blocks
==4425== 
==4425== For counts of detected and suppressed errors, rerun with: -v
==4425== ERROR SUMMARY: 12 errors from 11 contexts (suppressed: 2 from 2)

What am I doing wrong?
Code is simply as follows:

struct sockaddr addr;
socklen_t           addr_sz = sizeof(addr);
char        host[NI_MAXHOST],
            serv[NI_MAXSERV];
int infd = accept(srv_fd, (struct sockaddr*)&addr, &addr_sz);
if (infd == -1) {
    ... manage error on accept ...
}
if(getnameinfo((struct sockaddr *)&addr, addr_sz, host, NI_MAXHOST, serv, NI_MAXSERV, NI_NUMERICSERV)) {
    strncpy(host, "<unknown host>", NI_MAXHOST-1);
    strncpy(serv, "<unknown port>", NI_MAXSERV-1);
}

And there you have the leak...
I can confirm, that the leak is happening : for 6 clients connected valgrind found 60 bytes leaked (I guess the clients were connecting from the same host so if it's related to host name the growth is linear as expected).
Any idea?

Cheers

---

Eventually found the real leak.

When connecting to the server socket use name.local instead of localhost and/or the fully qualified name.
getnameinfo() will then leak.

I can reproduce the bug on 12.04 , 12.10 both x64 and x86.
If I connect specifying .local on the name it leaks.

Cheers

链接地址: http://www.djcxy.com/p/64944.html

上一篇: 由valgrind检测到堆栈粉碎

下一篇: getnameinfo内存泄漏是否确认？