Bug 1793984

Summary:	[RHEL 8][s390x] Restore modified SIGILL signal handler during libcrypto library initialisation
Product:	Red Hat Enterprise Linux 8	Reporter:	Eugene Syromiatnikov <esyr>
Component:	openssl	Assignee:	Tomas Mraz <tmraz>
Status:	CLOSED ERRATA	QA Contact:	Ondrej Moriš <omoris>
Severity:	medium	Docs Contact:
Priority:	medium
Version:	8.2	CC:	emachado, fche, mjw, omoris, ssorce, szidek
Target Milestone:	rc	Keywords:	Triaged
Target Release:	8.3	Flags:	pm-rhel: mirror+
Hardware:	s390x
OS:	Linux
Whiteboard:
Fixed In Version:	openssl-1.1.1c-13.el8	Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:
Clones:	1874594 (view as bug list)		Environment:
Last Closed:	2020-04-28 16:52:04 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:
Bug Blocks:	1780068, 1873552, 1874594

Description Eugene Syromiatnikov 2020-01-22 12:40:31 UTC

Description of problem:
libcrypto on s390x changes SIGILL handler during library initialisation (specifically, in the crypto/s390xcap.c:OPENSSL_cpuid_setup() function) and doesn't restore it back.

Version-Release number of selected component (if applicable):
openssl-libs-1.1.1c-10.el8.s390x

How reproducible:
Always

Steps to Reproduce:
See the reproducer in bug 1780068.

Actual results:
sigign.test from strace's test suite fails.

Expected results:
sigign.test from strace's test suite passes.

Additional info:
sigign.test runs (among other things) the command "../set_sigign 1 $SIGNO ../../strace -e none ../check_sigign 1 $SIGNO" for the appropriate values of $SIGNO; check_sigign checks that it has the signal handler the set_sigign has set; it is failing on RHEL 8.2.0 (and not on RHEL 8.1.1) for SIGNO=4 (SIGILL), as outlined by bug 1780068.

The change between RHEL 8.1.1 and RHEL 8.2.0 is that the latter now has debuginfod package, the libdebuginfod library of which is pulled by libdw during the library load (libdw is used in strace for stack unwinding purposes). The libdebuginfod, in turn, has to initialise libcurl early (during library load), and that, in turn, pulls in libcrypto.

The issue seems to be fixed upstream in commit c17d60ea293746d7cd06a910ced446edbb6c1eba ("s390x assembly pack: fix restoring of SIGILL action").

Kudos to Frank Ch. Eigler and Mark Wielaard for helping triaging the issue.

Comment 1 Tomas Mraz 2020-01-22 13:00:06 UTC

This is not a regression in RHEL-8.2 package, the bug was there from start.

Comment 2 Eugene Syromiatnikov 2020-01-22 13:15:17 UTC

Well, this bug has caused a regression in strace between 8.1.1 and 8.2.0 environments.

Comment 3 Tomas Mraz 2020-01-22 13:47:58 UTC

Yes, but not by a change in OpenSSL.

Comment 9 Simo Sorce 2020-01-27 08:10:28 UTC

What is the impact of this bug?
Are there workarounds?

Comment 10 Frank Ch. Eigler 2020-01-27 12:24:24 UTC

(In reply to Simo Sorce from comment #9)
> What is the impact of this bug?

Applications that are very sensitive to signal handler configuration may fail.  It is unknown how many of these exist, outside the testsuite item that led us to discover it in RHEL.

> Are there workarounds?

None that I know of.

Comment 11 Tomas Mraz 2020-01-27 12:47:27 UTC

Yes, but in general it is not a regression within RHEL-8 updates. It is a regression compared to RHEL-7 only. Within RHEL-8 updates it is a regression of this particular strace use-case only.

Comment 12 Eugene Syromiatnikov 2020-01-27 13:51:44 UTC

(In reply to Tomas Mraz from comment #11)
> Yes, but in general it is not a regression within RHEL-8 updates. It is a
> regression compared to RHEL-7 only.

I don't see how it's not a regression between RHEL 8.1 and RHEL 8.2, as debuginfod is in RHEL 8.2 and not in RHEL 8.1; as a result, strace's signal setting transparency for SIGILL was preserved in 8.1 and is not preserver in 8.2.

The same is probably the case for other clients of libdw/debuginfod.

Comment 13 Tomas Mraz 2020-01-27 16:36:49 UTC

I was unclear a little - it is not regression for usages that do not use libdw/debuginfod.

Comment 14 Eugene Syromiatnikov 2020-02-10 12:23:37 UTC

Clearing needinfo, as the information requested in comment 9 has been provided in comment 10.

Comment 21 errata-xmlrpc 2020-04-28 16:52:04 UTC

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2020:1840

Comment 22 Fedora Update System 2020-05-29 00:57:12 UTC

FEDORA-EPEL-2020-ff94ccbdec has been pushed to the Fedora EPEL 7 stable repository.
If problem still persists, please make note of it in this bug report.