Bug 821643 – >=sys-apps/sandbox-3.0 breaks build on tmpfs

• Product: Portage Development
• Status: CONFIRMED ---

[Description Christian Bricart 2021-11-04 08:35:51 UTC]

build of i.e. www-client/firefox breaks at random points receiving an -EINTR (interrupted system call) when $PORTAGE_TMPDIR being a tmpfs.

it succeeds when $PORTAGE_TMPDIR is a disk backed ext4.
Also downgrading sandbox from 3.1 to 2.29 (3.0 is not enough) lets the build succeed also on tmpfs.

Note:
the system has CONFIG_SECURITY_YAMA Kernel option fully unset and yama ptrace behaviour is not only mitigated to "0" through sysctl knob.

Reproducible: Always

[Comment 1 Christian Bricart 2021-11-04 08:37:43 UTC]

Created attachment 748554 [details]
emerge --info

emerge --info output
otherwise, it's a fully up-to-date ~amd64 box

[Comment 2 Christian Bricart 2021-11-04 08:46:49 UTC]

I do have build logs produced with:
  SANDBOX_VERBOSE=yes SANDBOX_DEBUG=yes MAKEOPTS=-j1 emerge -1v www-client/firefox

for
 - failed build with sandbox-3.1 on tmpfs
 - successful build with sandbox-3.1 on ext4
 - successful build with sandbox-2.29 on tmpfs

but they all exceed the 1k upload size - even xz compressed..

[Comment 3 Christian Bricart 2021-11-04 09:05:40 UTC]

I've uploaded the logs to https://looking.at/bug821643/

[Comment 4 Florian Schmaus 2021-11-04 09:16:22 UTC]

Created attachment 748563 [details]
shortened build.log

I guess the most relevant part of the log is:

8 5:18.53   = note: ld.lld: error: failed to open /var/tmp/portage-tmpfs/portage/www-client/firefox-94.0/work/firefox_build/release/build/cubeb-sys-a7764e8ed09bab90/build_script_build-a7764e8ed09bab90: Interrupted system call
8 5:18.53           clang-13: error: linker command failed with exit code 1 (use -v to see invocation)

I've also attached a shortened build.log.

[Comment 5 Sam James 2021-11-04 13:26:59 UTC]

(In reply to Christian Bricart from comment #3)
> I've uploaded the logs to https://looking.at/bug821643/

Can you upload them here compressed? Thanks.

[Comment 6 Christian Bricart 2021-11-04 13:44:53 UTC]

(In reply to Sam James from comment #5)
> (In reply to Christian Bricart from comment #3)
> > I've uploaded the logs to https://looking.at/bug821643/
> 
> Can you upload them here compressed? Thanks.

as said, there are xz compressed and exceed the maximum permitted upload size by far, i.e 13MB .. so you consider me to rather upload ~40 split files..?

And alas, I am not able to produce shortened logs because I do not know what is of value within..

[Comment 7 SpanKY 2021-11-04 16:38:08 UTC]

let's put a pin in this for now.  i suspect this is just a dupe of other KI, so let's have you retest with a newer version once i have those fixed.  if it resolves your issues, then we can not worry about log transfer at all.