[PATCH v2] arc4random: simplify design for better safety

[Jason A. Donenfeld]

On Tue, Jul 26, 2022 at 08:42:51PM +0200, Jason A. Donenfeld wrote:
> Hi Mark,
> 
> On Tue, Jul 26, 2022 at 09:51:03AM -0700, Mark Harris wrote:
> > > Thanks. It looks like INTERNAL_SYSCALL_CALL just returns the errno as-is
> > > as a return value, right? I'll adjust the code to account for that.
> > 
> > Yes INTERNAL_SYSCALL_CALL just returns the negated errno value that it
> > gets from the Linux kernel, but only on Linux does
> > __getrandom_nocancel use that.  The Hurd and generic implementations
> > set errno on error.  Previously the only call to this function did not
> > care about the specific error value so it didn't matter.  Since you
> > are now using the error value in generic code, __getrandom_nocancel
> > should be changed on Linux to set errno like most other _nocancel
> > calls, and then it should go back to checking errno here.
> > 
> > And as Adhemerval mentioned, you only added a Linux implementation of
> > __ppoll_infinity_nocancel, but are calling it from generic code.
> 
> Okay, I'll switch this to use INLINE_SYSCALL_CALL, so that it sets
> errno, and then will use the normal TEMP_FAILURE_RETRY macro for EINTR.
> 
> > Also, by the way your patches cc'd directly to me get quarantined
> > because DKIM signature verification failed.  The non-patch messages
> > pass DKIM and are fine.
> 
> That sure is odd. The emails are all going through the MTA. rspamd bug?
> OpenSMTPD bug? Hmm...

It's because LICENSE has a ^L in it, which I guess doesn't go over well
with OpenSMPTD or rspamd or kernel.org's smtp server or some combination
thereof...

I just posted v5, by the way, in case it's in your spam folder.

Jason