[libvirt] [PATCH 2/2] build: fix getcwd portability problems

Matthias Bolte matthias.bolte at googlemail.com
Fri Apr 29 18:19:38 UTC 2011


2011/4/29 Eric Blake <eblake at redhat.com>:
> On 04/29/2011 11:48 AM, Matthias Bolte wrote:
>>> +++ b/tests/testutils.c
>>> @@ -478,7 +478,6 @@ int virtTestMain(int argc,
>>>                  int (*func)(void))
>>>  {
>>>     int ret;
>>> -    char cwd[PATH_MAX];
>>>  #if TEST_OOM
>>>     int approxAlloc = 0;
>>>     int n;
>>> @@ -491,7 +490,7 @@ int virtTestMain(int argc,
>>>
>>>     abs_srcdir = getenv("abs_srcdir");
>>>     if (!abs_srcdir)
>>> -        abs_srcdir = getcwd(cwd, sizeof(cwd));
>>> +        abs_srcdir = getcwd(NULL, 0);
>>>     if (!abs_srcdir)
>>>         exit(EXIT_AM_HARDFAIL);
>>
>> Now you have created a memory leak (not a critical one, that's true),
>> because abs_srcdir can be malloc'ed and you missed to free it.
>
> It's technically only a leak if someone overwrites abs_srcdir with
> different contents without freeing it first, since there is a global
> variable that still tracks the pointer through the point of program
> exit().  Valgrind reports this type of open-ended allocation as "still
> reachable", rather than "definitely lost".

I know, but I like valgrind clean software :)

> But, to make valgrind even quieter, yes, I can fix things up to free the
> memory if it was not read from getenv and before returning from
> virtTestMain.

Thanks.

>>
>> ACK, with that memory leak fixed.
>
> Pushed with this addition:
>
> diff --git i/tests/testutils.c w/tests/testutils.c
> index 91035a2..ae73530 100644
> --- i/tests/testutils.c
> +++ w/tests/testutils.c
> @@ -478,6 +478,7 @@ int virtTestMain(int argc,
>                  int (*func)(void))
>  {
>     int ret;
> +    bool abs_srcdir_cleanup = falseb;

As I saw this line, I hoped you didn't push it with that typo... and
you didn't :)

Now I'll go and rebase my stack usage cleanup patch.

Matthias




More information about the libvir-list mailing list