Re: [PATCH -tip 2/4] [BUGFIX]perf: Fix strlist__parse_list to handle const string

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2010/12/22 3:23), Arnaldo Carvalho de Melo wrote:
> Em Fri, Dec 17, 2010 at 10:12:06PM +0900, Masami Hiramatsu escreveu:
>> Fix strlist__parse_list to handle const string. Without this patch,
>> strlist__parse_list() causes SEGV when caller passes a constant
>> string.
>  
>> +++ b/tools/perf/util/strlist.c
>> @@ -136,13 +136,19 @@ static int strlist__parse_list_entry(struct strlist *self, const char *s)
>>  
>>  int strlist__parse_list(struct strlist *self, const char *s)
>>  {
>> -	char *sep;
>> +	char *sep, *tmp;
>>  	int err;
>>  
>> +	/* This method requires strdup, because this changes given string */
>> +	if (!self->dupstr)
>> +		return -EINVAL;
>> +
> 
> Why is the above check needed if you solved the problem by strnduping at
> each separator?

Without dupstr, strlist(str_node__delete) will not release
allocated buffer... And we have to release it if the parsed
string started with "file://" after loading file.

(Ah, and there is a same problem in strlist__load() too...)

> Wouldn't be better to just stop changing the string by passing the
> length to strlist__parse_list_entry, etc?

Yeah, it could be, but in that case, we need changing code
wider.
I think this is the simplest solution. :)

(Of course, we have to check dupstr flag on top of strlist__load too.)

Thank you,

>>  	while ((sep = strchr(s, ',')) != NULL) {
>> -		*sep = '\0';
>> -		err = strlist__parse_list_entry(self, s);
>> -		*sep = ',';
>> +		tmp = strndup(s, sep - s);
>> +		if (tmp == NULL)
>> +			return -ENOMEM;
>> +		err = strlist__parse_list_entry(self, tmp);
>> +		free(tmp);
>>  		if (err != 0)
>>  			return err;
>>  		s = sep + 1;

-- 
Masami HIRAMATSU
2nd Dept. Linux Technology Center
Hitachi, Ltd., Systems Development Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com