Re: [PATCH 1/2] remote.c: don't leak the base branch name in format_tracking_info
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Aug 10, 2014, 23:03 UTC
- Message-ID
- <xmqq7g2g7xrv.fsf@gitster.dls.corp.google.com>
- In-Reply-To
- <53E78C59.2010704@gmail.com>
Stefan Beller <stefanbeller@gmail.com> writes:
Show 29 quoted lines
> On 10.08.2014 15:57, Stefan Beller wrote:
>> Found by scan.coverity.com (Id: 1127809)
>>
>> Signed-off-by: Stefan Beller <stefanbeller@gmail.com>
>> ---
>> remote.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/remote.c b/remote.c
>> index 3d6c86a..2c1458f 100644
>> --- a/remote.c
>> +++ b/remote.c
>> @@ -1983,6 +1983,7 @@ int format_tracking_info(struct branch *branch, struct strbuf *sb)
>> strbuf_addf(sb,
>> _(" (use \"git pull\" to merge the remote branch into yours)\n"));
>> }
>> + free(base);
>> return 1;
>> }
>>
>>
>
> Upon testing this one again, I get a warning
> remote.c: In function ‘format_tracking_info’:
> remote.c:1986:2: warning: passing argument 1 of ‘free’ discards ‘const’ qualifier from pointer target type [enabled by default]
> free(base);
> ^
> ...
> Please ignore this patch.It is perfectly fine to cast it to (char *) in this case, I think.
The real culprit is that the functionà reuses the same "base" (which is a pointer into a constant region of memory) to receive the new copy allocated by shorten_unambiguous_ref(); the piece of memory returned by the callee belongs to this function, and it is perfectly fine if this function modifies the contents of it (which it doesn't happen to do). Storing the returned value to a variable of type "const char *" does not absolve it from the responsibility to free it (hence your patch).