enosys terminates the user session on Azure 4.13

[Po-Hsu Lin]

Identical to bug 1755358, user session will be terminated with running the enosys test in the ubuntu_stress_smoke test.

This happens on an Azure instance Standard-H16mr, looks like this is not affecting running the test remotely from the jenkins server, just occurs when I'm running this manually.

please find the attachment for the syslog.

ProblemType: Bug
DistroRelease: Ubuntu 16.04
Package: linux-image-4.13.0-1012-azure 4.13.0-1012.15
ProcVersionSignature: User Name 4.13.0-1012.15-username 4.13.13
Uname: Linux 4.13.0-1012-azure x86_64
ApportVersion: 2.20.1-0ubuntu2.15
Architecture: amd64
Date: Thu Mar 22 06:37:01 2018
ProcEnviron:
 TERM=xterm-256color
 PATH=(custom, no user)
 XDG_RUNTIME_DIR=<set>
 LANG=en_US.UTF-8
 SHELL=/bin/bash
SourcePackage: linux-azure
UpgradeStatus: No upgrade log present (probably fresh install)

[Kleber Sacilotto de Souza (kleber-souza)]

It seems that the stress-ng log messages from the enosys testcase are:

Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test stress-ng: info: [14237] dispatching hogs: 4 enosys
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test stress-ng: info: [14237] cache allocate: using built-in defaults as unable to determine cache details
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test kernel: [ 513.698028] signal_fault: 5 callbacks suppressed
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test kernel: [ 513.698030] stress-ng-enosy[14972] bad frame in x32 rt_sigreturn frame:00007ffc115313c0 ip:40128f5c28f5c28f sp:500cd8a6c9bbb800 orax:ffffffffffffffff
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test kernel: [ 513.699810] stress-ng-enosy[14993] bad frame in x32 rt_sigreturn frame:00007ffc115313c0 ip:40128f5c28f5c28f sp:500cd8a6c9bbb800 orax:ffffffffffffffff
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test kernel: [ 513.700422] stress-ng-enosy[15001] bad frame in x32 rt_sigreturn frame:00007ffc115313c0 ip:40128f5c28f5c28f sp:500cd8a6c9bbb800 orax:ffffffffffffffff
Mar 22 06:27:10 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test kernel: [ 513.702243] stress-ng-enosy[15022] bad frame in x32 rt_sigreturn frame:00007ffc115313c0 ip:40128f5c28f5c28f sp:500cd8a6c9bbb800 orax:ffffffffffffffff
Mar 22 06:27:13 x-la-azure-4-13-0-Standard-H16mr-stress-smoke-test stress-ng: info: [14237] successful run completed in 2.96s

[Colin Ian King (colin-king)]

"bad frame in x32 rt_sigreturn" kernel messages are OK, it's because the enosys stressor is calling an x32 rt_sigreturn() in the incorrect context and the kernel flags this up as a misbehaving application.

[Colin Ian King (colin-king)]

I thought the test exiting early issues were fixed with commits:

http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=d0e42d6315403fd35f681275cc32ec1e466d96c0
http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=ed8729f09e13887de97a530ed293de9086bff67c

Is this still breaking? If so, if I can get access to a host and I can try and debug it further.

[Po-Hsu Lin (cypressyew)]

Hi Colin, yes I can still reproduce this (enosys terminates my SSH session) on Standard_H16mr

I will send you the info for this system.

[Colin Ian King (colin-king)]

This occurs because we are calling aliases of vhangup that terminate the session. Added mask'd detection of this system call.

Fix committed: http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=94e40091651659cf08a1ba11965e56102d15d87a

[Po-Hsu Lin (cypressyew)]

Verified on the very same node, and this fix works.
Thanks!