[Toybox] [PATCH] grep: fix two bugs found by hwasan.

[Rob Landley]

On 7/17/19 11:16 AM, enh wrote:
>> I committed a fix on top of yours, does it still pass your sanitizer?
> 
> yes, thanks. (a comment there about alignment would have helped... in
> 2019 it's rare to see anyone still care about alignment!)

Rare for _you_ to see. Cortex-M0 faults on unaligned access, and I believe ARM
estimated at launch they expected to sell 4.5 billion cortex-m chips in 2013
alone? (https://www.arm.com/zh/files/event/20130715_AES_Joseph.pdf) And they're
one of a dozen players in this space (albeit the biggest).

There are more nommu devices deployed every year than android devices, it's just
that the range between "high end" and "low end" got big enough there's not
nearly as much developer overlap as there used to be. Plus a decade and change
back the embedded community largely seems to have given up on the Linux
community (too toxic) and stopped trying to interact with them, so they're
either sticking with old versions (I still see 2.6 out in the wild a lot), using
forks maintained by companies like https://www.emcraft.com/, or doing posix RTOS
du jour (yesterday's was https://twitter.com/Sat_U/status/1151342993883815936).

> i'll send a patch that adds an ASAN=1 to the build scripts so it's
> easier to test like this. whenever hwasan finds a bug (the last one i
> had to fix was in BSD grep, amusingly enough) i try to reproduce and
> fix on the host for convenience and although it's not hard to do it's
> quite a long command line.

This works with the NDK toolchain?

> i was hoping to talk about valgrind -> asan -> hwasan [where we are
> now] -> ARM MTE [where we want to be] in the Google I/O 2019 C++ talk,
> so i could send you a link.

Yay link!

I'm scheduled to give a "busybox vs toybox" talk at ELC in a month because of
the 4 talk topics I submitted, that's the one they wanted for some reason?
(*shrug* Ok.)

> as it was, we only got hwasan into the
> Pixel 2 kernels for Q
> [https://source.android.com/devices/tech/debug/hwasan], and i didn't
> want to spend time on what's effectively still vaporware from an app
> developer perspective. but there are some slides by the folks doing
> the actual work from an LLVM conference last year:
> https://llvm.org/devmtg/2018-10/slides/Serebryany-Stepanov-Tsyrklevich-Memory-Tagging-Slides-LLVM-2018.pdf
> --- TL;DR: MTE is "always-on asan in production".

Cool.

I tend to be a bit leery of stuff that isn't upstream "yet". The "yet" for
squashfs dragged on for something like 5 years?
(https://lwn.net/Articles/563578/) I was all excited about fdpic support for arm
(https://s3.amazonaws.com/connect.linaro.org/sfo15/Presentations/09-24-Thursday/SFO15-406-%20ARM%20FDPIC%20Toolchains.pdf)
but as far as I know it still hasn't been merged upstream into gcc.

Everybody else in the embedded world just uses the forks and ignores upstream,
and are thus stuck on old versions. _I_ don't, but I'm weird in this context.
Funky hybrid of "the linux way" and "the embedded way" trying to serve two
masters. And now, of course, Android.

>> Rob

Still Rob