[dpdk-dev] [PATCH] net/failsafe: fix parameters parsing

[Gaëtan Rivet]

On Thu, Aug 17, 2017 at 03:54:23PM +0000, Matan Azrad wrote:
> 
> 
> > -----Original Message-----
> > From: Gaëtan Rivet [mailto:gaetan.rivet at 6wind.com]
> > Sent: Thursday, August 17, 2017 6:26 PM
> > To: Matan Azrad <matan at mellanox.com>
> > Cc: dev at dpdk.org; Ori Kam <orika at mellanox.com>; stable at dpdk.org
> > Subject: Re: [PATCH] net/failsafe: fix parameters parsing
> > 
> > Hi Matan,
> > 
> > Thanks for spotting the bug.
> > 
> > On Thu, Aug 17, 2017 at 05:19:41PM +0300, Matan Azrad wrote:
> > > The corrupted code used wrongly snprintf return value as the number of
> > > characters actually copied, in spite of the meanning is the number of
> > > characters which would be generated for the given input.
> > >
> > > It caused to remain zerod bytes between the failsafe command line non
> > > sub device parameters indicates end of string.
> > >
> > > Hence, when rte_kvargs_parse tried to parse all parameters, it got end
> > > of string after the first one and the others weren't parsed.
> > >
> > > So, if the mac parameters was the first in command line it was taken
> > > while hotplug_poll was left default, and vice versa.
> > >
> > > The fix updates the buffer index by dedicated variable contains the
> > > copy size, by the way uses memcpy instead of snprintf which is good
> > > enouth for this copy scenario.
> > 
> > Actually snprintf should still be used.
> > 
> Why?
> We just want to copy from buffer to buffer, no needs snprintf overheads
> If actually we are not using complicated format.
> 

snprintf plays nice with strings and ensure that it will be terminated
by a '\0'. It is generally preferable, particularly considering that the
parameter string we are building here is meant to be parsed by another
lib (rte_kvargs).

> > >
> > > Fixes: a46f8d584eb8 ("net/failsafe: add fail-safe PMD")
> > > Cc: stable at dpdk.org
> > >
> > > Signed-off-by: Matan Azrad <matan at mellanox.com>
> > > ---
> > >  drivers/net/failsafe/failsafe_args.c | 14 ++++++++------
> > >  1 file changed, 8 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/drivers/net/failsafe/failsafe_args.c
> > > b/drivers/net/failsafe/failsafe_args.c
> > > index 1f22416..0a98b04 100644
> > > --- a/drivers/net/failsafe/failsafe_args.c
> > > +++ b/drivers/net/failsafe/failsafe_args.c
> > > @@ -271,7 +271,7 @@ static int
> > >  fs_remove_sub_devices_definition(char params[DEVARGS_MAXLEN])  {
> > >  	char buffer[DEVARGS_MAXLEN] = {0};
> > > -	size_t a, b;
> > > +	size_t a, b, temp;
> > 
> > temp is not descriptive enough.
> 
> What are about a, b, i ?
>  
> > You are declaring this variable here because you want to re-use it instead of
> > `start`. This is an overreach however, this fix must be restricted to the actual
> > bug.
> 
> temp is helping to address the original bug, don't we want to reuse variable 
> Instead of 2  if statement variables, maybe other name for all?
> 

The compiler should be able to see that their use does not overlap.

The issue is that temp is meant to describe a length limit in one branch
and be a marker of a starting point in another. Thus both variable
should be named differently to make the intent clear.

When declaring temp (or saved_val) in the higher-scope, you are thus
forced to use generic, non-descriptive name precisely because you are
sharing the variable between two different uses.

> Something like:
> a=> start
> b=> end
> i => next
> temp=> saved_val
>  
> 
> > 
> > >  	int i;
> > >
> > >  	a = 0;
> > > @@ -286,12 +286,14 @@ fs_remove_sub_devices_definition(char
> > params[DEVARGS_MAXLEN])
> > >  			ERROR("Invalid parameter");
> > >  			return -EINVAL;
> > >  		}
> > > -		if (params[b] == ',' || params[b] == '\0')
> > 
> > Declare the temporary variable in this scope, with a descriptive name such as
> > "len", "length", "param_len" or something close.
> > 
> > > -			i += snprintf(&buffer[i], b - a + 1, "%s", &params[a]);
> > > -		if (params[b] == '(') {
> > > -			size_t start = b;
> > > +		if (params[b] == ',' || params[b] == '\0') {
> > > +			temp = b - a + 1;
> > 
> > The value here should be "b - a".
> > If a != 0 however, then we are parsing a new parameter and buffer already
> > contains at least one. A comma should be added to separate the two.
> > 
> > An example might clarify what I mean:
> > 
> >                 if (params[b] == ',' || params[b] == '\0') {
> >                         size_t param_len = b - a;
> > 
> >                         if (a)
> >                                 param_len += 1;
> >                         snprintf(&buffer[i], param_len + 1, "%s%s",
> >                                  a ? "," : "", &params[a]);
> >                         i += param_len;
> >                 }
> > 
> > The conditionals about `a` are ugly however, if you find a better way to write
> > those you are most welcome :).
> > 
> > > +			memcpy(&buffer[i], &params[a], temp);
> > > +			i += temp;
> > > +		} else if (params[b] == '(') {
> > > +			temp = b;
> > >  			b += closing_paren(&params[b]);
> > > -			if (b == start)
> > 
> 
> I think the last comma is harmless for next parse
> But I can just change the last coma to '\0' in the end of function(if exists).
> But, This solves another issue, don't it?  Maybe in different patch?
> 

It might be harmless, but I prefer having a clean output.
Which other issue? Handling the commas seems in line with
properly copying regular parameters.

You could certainly check for the presence of a comma, but you already
see how ugly the code flow would be ("if it exists"), and you will keep a
dirty string between two loops, with possibly other branches wanting to
return on error.

I prefer having a compact, self-contained logic addressing kvargs which
should not leak conceptual gotcha that the next programmer would need to
be wary of when trying to touch the code in other parts of the function.

> > The changes should be limited to the actual bug. No need to change this.
> > 
> > > +			if (b == temp)
> > >  				return -EINVAL;
> > >  			b += 1;
> > >  			if (params[b] == '\0')
> > > --
> > > 2.7.4
> > >
> > 
> > Thanks again for the debug and sorry for being nitpicky.
> > 
> > --
> > Gaëtan Rivet
> > 6WIND
> 
> Regards,
> Matan Azrad.

-- 
Gaëtan Rivet
6WIND