Is struct addrinfo **res allocated when getaddrinfo() returns a non-zero value?

Question

I am having some memory leaking troubles in an embedded application and while looking in the code, I see I do not freeaddrinfo() when getaddrinfo() returns non-zero:

s = getaddrinfo(hostname, port, &hints, &result);
if (s != 0) {
  log_error();
} else {
  // do stuff
  freeaddrinfo(result);
}

Could this lead to memory leaks? I have tried to look into the man page, but it doesn't state it explicitly.

Did you try `valgrind`? There's a big chance to help you finding the bug - it can show you the exact place, a memory is allocated and not freed later. — Kiril Kirov, Aug 20 '12 at 09:44
@KirilKirov I know, but this is an embedded device, and valgrind is not available. (I am happy to have a debugger ;)) — Bart Friederichs, Aug 20 '12 at 09:45
It's actually very easy to test: Set the result pointer to `NULL`, call the function with arguments you know will fail, and see if the pointer is still `NULL` after. If not then you should probably free it. — Some programmer dude, Aug 20 '12 at 09:46

score 5 · Accepted Answer · answered Aug 20 '12 at 09:48

The specification doesn't say that result isn't assigned if it fails, so it appears that a conforming implementation could do so.

Why not free it unconditionally?

result = 0;
s = getaddrinfo(hostname, port, &hints, &result);
if (s) {
    // log error
} else {
    // success
}
if (result) {
    freeaddrinfo(result);
}

Ideally you'd be able to call freeaddrinfo(result) without checking whether result is NULL, but although it's implied by the standard that that's OK I wouldn't count on it.

That's the solution I am leaning towards yes. – Bart Friederichs Aug 20 '12 at 09:49 — Bart Friederichs, Aug 20 '12 at 09:49

score 2 · Answer 2 · answered Mar 31 '14 at 09:39

While it's true that the standard is a little bit vague about that, the only interpretation that makes sense is to treat the error as an alternative to a result. This is how other libc functions work and this is how getaddrinfo() is used in the examples even in the standard.

See: http://pubs.opengroup.org/onlinepubs/9699919799/functions/freeaddrinfo.html

I don't think there is a need to invent ways to get around that unless you're dealing with a known broken implementation. POSIX functions don't require you to zero the pointers used for output unless specified otherwise.

From the above link:

Upon successful return of getaddrinfo(), the location to which res points shall refer to a linked list of addrinfo structures, each of which shall specify a socket address and information for use in creating a socket with which to use that socket address.

It's pretty clear that the standard is written with an assumption that in other cases, the res variable is not being touched.

Is struct addrinfo **res allocated when getaddrinfo() returns a non-zero value?

2 Answers2