libsa: Allow a tftp retry if a sendrecv returns ETIMEDOUT

[sbruno]

Summary

Allow a sendrecv retry on ETIMEDOUT

After https://svnweb.freebsd.org/base?view=revision&revision=318320 sendrecv returns ETIMEDOUT a lot more often causing TFTP of files to fail. Retrying when there is no data available seems to fix this for my testcase. However, it may be masking something else.

[sbruno]

stand/libsa/tftp.c

507

Having the actual return code was useful, so I left it in.

624

"debug" didn't exist in the tree and invoking this on every packet is pointless in debug mode.

[tsoome]

nice catch.

[kevans]

Seems reasonable to me

[imp]

stand/libsa/tftp.c

644

I'd be tempted to limit this to 10 retries or something like that...

[imp]

stand/libsa/tftp.c

644–649

On second thought, maybe just
if (errno == EAGAIN) {
cc = 0;
errno = 0;
}

and then letting it fall through to the current retry logic might be a lot less disruptive.

[imp]

stand/libsa/tftp.c

649

Or even simpler:
if (cc != -1 || (errno != 0 && errno != EAGAIN))

[sbruno]

Use @kevans version of the patch from private email. Add a debug statement so
I can detect the ETIMEDOUT condition, which does get hit and resumes TFTP without
failing.

[kevans]

Heh, sorry, I ended up creating another review with a different version of that patch: https://reviews.freebsd.org/D14350 -- it turned out there was also a case over in net.c that needed fixing, and we have some timing weirdness happening in the tftp bits to go with it.

[sbruno]

In D14332#300876, @kevans wrote:

Heh, sorry, I ended up creating another review with a different version of that patch: https://reviews.freebsd.org/D14350 -- it turned out there was also a case over in net.c that needed fixing, and we have some timing weirdness happening in the tftp bits to go with it.

Ok, let me abandon this and switch over to D14350

[imp]

stand/libsa/tftp.c

619

this can be removed.