Re: (stat(...) == -1 || faccessat(...) == -1) && errno == EINTR ?!??

[Tobias Bading <tbading@web.de>]

 > You could use systemtap to patch the kernel to see if you can trap the
 > path where they're returning `EINTR`.
 > I recently came across a blog post:
 >
https://engineering.skroutz.gr/blog/uncovering-a-24-year-old-bug-in-the-linux-kernel/
 > where some people did that and found a bug in the TCP code. Very
 > impressive.

Thanks, very interesting read indeed. I haven't heard of systemtap
before, sounds like a very cool tool for kernel hackers. Since I'm only
doing userland stuff, I usually trust Valgrind to help me out when
things get fishy.

 > I don't know if this qualifies as "breaking userland" [...]

I'd say breaking stat() qualifies big time, but to be honest I don't
really have the time/nerve to read
https://pubs.opengroup.org/onlinepubs/9699919799/xrat/V4_xsh_chap02.html
in full. The "Signal Effects on Other Functions" paragraph seems
interesting, but my head explodes when I read sentences like

   "Therefore, implementations should generally make such functions
(including ones defined as extensions) interruptible."

   "Functions not mentioned explicitly as interruptible may be so on
some implementations, possibly as an extension where the function gives
an [EINTR] error."

English isn't my first language, so I must be reading this wrong or
interpreting it in the wrong context. If a POSIX implementation was
allowed to add EINTR to the list of possible error codes returned by a
function defined in the standard, what kind of standard would that be?
How about an implementation of malloc() which returns NULL and sets
errno to EINTR when a signal is raised while the calling thread was
asleep because it was waiting for a disk read from swapspace? XD

Tobias

---

On 14.02.21 23:04, Godmar Back wrote:
> I don't know if this qualifies as "breaking userland" or "this has
> never worked."
>
> You could use systemtap to patch the kernel to see if you can trap the
> path where they're returning `EINTR`.
> I recently came across a blog post:
> https://engineering.skroutz.gr/blog/uncovering-a-24-year-old-bug-in-the-linux-kernel/
> where some people did that and found a bug in the TCP code. Very
> impressive.