[PATCH] memcmp() error in gas/dwarf2dbg.c

[Jakub Jelinek]

On Fri, Feb 13, 2004 at 12:21:30PM +0100, Andreas Jaeger wrote:
> > --- binutils-2.14.90.0.8/gas/dwarf2dbg..c.orig	2004-02-13 11:55:05.470239719 +0100
> > +++ binutils-2.14.90.0.8/gas/dwarf2dbg.c	2004-02-13 11:57:23.679576129 +0100
> 
> Note this patch applies also to current CVS, Hannes noticed it with a
> segmentation fault of gas due to accessing beyond the allocated memory.
> 
> > @@ -339,7 +339,7 @@ get_filenum (const char *filename, unsig
> >  {
> >    static unsigned int last_used, last_used_dir_len;
> >    const char *file;
> > -  size_t dir_len;
> > +  size_t dir_len, tmp_len;
> >    unsigned int i, dir;
> >  
> >    if (num == 0 && last_used)
> > @@ -372,8 +372,9 @@ get_filenum (const char *filename, unsig
> >      {
> >        --dir_len;
> >        for (dir = 1; dir < dirs_in_use; ++dir)
> > -	if (memcmp (filename, dirs[dir], dir_len) == 0
> > -	    && dirs[dir][dir_len] == '\0')
> > +	tmp_len = strlen(dirs[dir]) < dir_len?strlen(dirs[dir]):dir_len;
> > +	if (memcmp (filename, dirs[dir], tmp_len) == 0
> > +	    && dirs[dir][tmp_len] == '\0')
> >  	  break;
> >  
> >        if (dir >= dirs_in_use)

I very much doubt it has been tested at all.
Previously it has been doing if (...) break; in the for cycle, now
it does just tmp_len = in the for cycle and if (...) break; afterwards.
Also, calling strlen twice is a waste, relying on compiler to optimize
it out is wrong.  And doing even one strlen in the cycle is too expensive.
I believe using
strncmp (filename, dirs[dir], dir_len) == 0 && dirs[dir][dir_len] == '\0'
would be much better.

	Jakub