Re: [dpdk-dev] [PATCH v2] lib/cmdline: init CLI parsing memory

[Olivier Matz <olivier.matz@6wind.com>]

On Thu, Jan 18, 2018 at 04:29:59AM +0000, Xueming(Steven) Li wrote:
> 
> 
> > -----Original Message-----
> > From: Olivier Matz [mailto:olivier.matz@6wind.com]
> > Sent: Tuesday, January 16, 2018 8:46 PM
> > To: Xueming(Steven) Li <xuemingl@mellanox.com>
> > Cc: Adrien Mazarguil <adrien.mazarguil@6wind.com>; dev@dpdk.org
> > Subject: Re: [PATCH v2] lib/cmdline: init CLI parsing memory
> > 
> > Hi Xueming,
> > 
> > Sorry for the late response.
> > 
> > On Tue, Dec 26, 2017 at 12:57:41PM +0000, Xueming(Steven) Li wrote:
> > > HI Olivier,
> > >
> > > By reading p1 comments carefully, looks like the pointer to result
> > > buffer issue not resolved by result copy. How about this:
> > >
> > > @@ -263,6 +263,7 @@
> > >  #ifdef RTE_LIBRTE_CMDLINE_DEBUG
> > >  	char debug_buf[BUFSIZ];
> > >  #endif
> > > +	char *result_buf = result.buf;
> > >
> > >  	if (!cl || !buf)
> > >  		return CMDLINE_PARSE_BAD_ARGS;
> > > @@ -312,16 +313,13 @@
> > >  		debug_printf("INST %d\n", inst_num);
> > >
> > >  		/* fully parsed */
> > > -		tok = match_inst(inst, buf, 0, tmp_result.buf,
> > > -				 sizeof(tmp_result.buf));
> > > +		tok = match_inst(inst, buf, 0, result_buf, sizeof(result.buf));
> > 
> > If we don't use tmp_result, it is maybe better to also replace
> > sizeof(result.buf) by CMDLINE_PARSE_RESULT_BUFSIZE
> 
> Thanks, would like to send out new version soon
> 
> > 
> > >
> > >  		if (tok > 0) /* we matched at least one token */
> > >  			err = CMDLINE_PARSE_BAD_ARGS;
> > >
> > >  		else if (!tok) {
> > >  			debug_printf("INST fully parsed\n");
> > > -			memcpy(&result, &tmp_result,
> > > -			       sizeof(result));
> > >  			/* skip spaces */
> > >  			while (isblank2(*curbuf)) {
> > >  				curbuf++;
> > > @@ -332,6 +330,7 @@
> > >  				if (!f) {
> > >  					memcpy(&f, &inst->f, sizeof(f));
> > >  					memcpy(&data, &inst->data, sizeof(data));
> > > +					result_buf = tmp_result.buf;
> > >  				}
> > >  				else {
> > >  					/* more than 1 inst matches */
> > >
> > 
> > 
> > I guess the issue you are talking about is the one described in
> > "cmdline: fix dynamic tokens parsing" in my previous description?
> > 
> > I think this patch is ok, and is probably better than the initial
> > suggestion, because it avoids to copy the buffer. However, I don't
> > understand why the previous patch was wrong, can you detail?
> 
> Let me quote your last email:
> "  When using dynamic tokens, the result buffer contains pointers
>    to some location inside the result buffer. When the content of
>    the temporary buffer is copied in the final one, these pointers
>    still point to the temporary buffer."
> If pointer exist in buffer, the new copy still pint to the old copy
> which very probably being changed.

In patch v2, I still think it was correct because there were 2 copies:

  1/ tok = match_inst(inst, buf, 0, result.buf, sizeof(result.buf));

     result is set, it may contain pointers to itself

  2/ result_ok = result;

     result is copied in result_ok
     result_ok may contain pointer to result

  3/ other calls to match_inst(inst, buf, 0, result...)

     this can clobber the result buffer

  4/ result = result_ok;   // before calling f()

     restores the content of result as it was in 1/
     it may contain pointers to itself, which is valid

Was there another problem I missed?


Anyway, I think your v3 is better because it avoids buffer copies.
If it's ok for you, please send a v4 with the small updated regarding
sizeof(result.buf) -> CMDLINE_PARSE_RESULT_BUFSIZE.

Thanks,
Olivier