[dpdk-stable] [PATCH] net/failsafe: fix exec parameter parsing error flow

[Gaëtan Rivet]

On Wed, Aug 30, 2017 at 06:11:47AM +0000, Matan Azrad wrote:
> Hi Gaetan
> 
> > -----Original Message-----
> > From: Gaëtan Rivet [mailto:gaetan.rivet at 6wind.com]
> > Sent: Tuesday, August 29, 2017 7:34 PM
> > To: Matan Azrad <matan at mellanox.com>
> > Cc: dev at dpdk.org; Raslan Darawsheh <rasland at mellanox.com>;
> > stable at dpdk.org
> > Subject: Re: [PATCH] net/failsafe: fix exec parameter parsing error flow
> > 
> > Hi Matan,
> > 
> > On Tue, Aug 29, 2017 at 05:59:08PM +0300, Matan Azrad wrote:
> > > The corrupted code returns success value in case of the execution
> > > process output stream is empty(EOF).
> > > It causes to segmentation fault while failsafe polls this command line
> > > again, than gets success and tries to do hotplug add to the sub device
> > > by uninitialized pointer dereferencing.
> > >
> > 
> > This is a bug and should be fixed, thanks.
> > 

Actually I am unable to reproduce this bug.

Do you have a fail-safe command line that would showcase this behavior?

> > > Morever, when the output is not empty but uncorrect, failsafe returns
> > > error for its probe function while the expected behavior is to do
> > > polling until the output is correct.
> > >
> > 
> > The expected behavior is for the fail-safe to return an error if the execution
> > of the given command returns an error.
> > 
> > The intention is that users writing such script would be able to output a blank
> > lines in case there is nothing to probe, but still remain aware of issues during
> > the execution of the command.
> > 
> > The fail-safe ignores errors pertaining to absent devices due to its nature.
> > This does not mean that it should ignore all errors and try to keep on going
> > while everything else is on fire.
> > 
> > The contract with the user is that "blank line" without other errors means
> > "absent device". Garbled output or return code != 0 means runtime error
> > and should be thrown to the user / application.
> > 
> 
> OK, good, I would have signed this contract :)
> 
> What's about if the parsing is not empty and out with error in the polling process?
> I think in current code failsafe just continues normally and tries again on next polling time.
> Because of this code I thought that if error occurs we should poll it again...
> 

It depends whether the fail-safe has already been initialized or not.
During the initialization phase, any errors other than -ENODEV means
that it must stop and force the user to look into it.

When initialization has finished, if polling errors occurs, the
fail-safe will try to minimize service disruption to the potentially
existing sub-devices. It thus discards the error and will try again
later.

> Can you please add it (the contract) in failsafe documentation for exec parameter?
> 
> > > The fix changes the return value to be -ENODEV for this sub device in
> > > the two cases.
> > > By this way, failsafe tries to parse this sub device parameter by exec
> > > method until the output is correct.
> > >
> > 
> > The issue is that this portion of the code will be heavily modified anyway. The
> > errno handling is erroneous and must be fixed, which is in conflict with your
> > patch.
> > 
> > I will send the intended fix shortly, referencing this patch and the issue your
> > highlighted, but both patch won't be compatible.
> > 
> 
> Good, no problems.
> 
> > > Fixes: a0194d828100 ("net/failsafe: add flexible device definition")
> > > Fixes: 35ffe4208140 ("net/failsafe: fix missing pclose after popen")
> > > Cc: stable at dpdk.org
> > >
> > > Signed-off-by: Matan Azrad <matan at mellanox.com>
> > > ---
> > >  drivers/net/failsafe/failsafe_args.c | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/failsafe/failsafe_args.c
> > > b/drivers/net/failsafe/failsafe_args.c
> > > index 645c885..61c55df 100644
> > > --- a/drivers/net/failsafe/failsafe_args.c
> > > +++ b/drivers/net/failsafe/failsafe_args.c
> > > @@ -157,12 +157,16 @@ fs_execute_cmd(struct sub_device *sdev, char
> > *cmdline)
> > >  	ret = fs_parse_device(sdev, output);
> > >  	if (ret) {
> > >  		ERROR("Parsing device '%s' failed", output);
> > > +		ret = -ENODEV;
> 
> Remove the above line for probe function error report.
> 
> > >  		goto ret_pclose;
> > >  	}
> > >  ret_pclose:
> > >  	pclose_ret = pclose(fp);
> > >  	if (pclose_ret) {
> > > -		pclose_ret = errno;
> > > +		if (errno == 0)
> > > +			errno = -(pclose_ret = ret);
> > > +		else
> > > +			pclose_ret = errno;
> > >  		ERROR("pclose: %s", strerror(errno));
> > >  		errno = old_err;
> > >  		return pclose_ret;
> > > --
> > > 2.7.4
> > >
> > 
> > Best regards,
> > --
> > Gaëtan Rivet
> > 6WIND
> 
> Thanks,
> Matan Azrad

-- 
Gaëtan Rivet
6WIND