Bug 1710539

Summary:	glibc: pthread_create vs EPERM error and stack cleanup.
Product:	Red Hat Enterprise Linux 8	Reporter:	Andrew Mike <amike>
Component:	glibc	Assignee:	Siddhesh Poyarekar <sipoyare>
Status:	CLOSED WONTFIX	QA Contact:	qe-baseos-tools-bugs
Severity:	low	Docs Contact:
Priority:	low
Version:	8.2	CC:	alanm, ashankar, brclark, codonell, dj, fweimer, jwright, mkolbas, mnewsome, pfrankli, sipoyare
Target Milestone:	rc	Keywords:	Patch, Triaged
Target Release:	8.2
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2022-11-18 14:10:43 UTC	Type:	---
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description Andrew Mike 2019-05-15 17:41:09 UTC

This bug was initially created as a copy of Bug #1678421

I am copying this bug because the customer requested 
us to refile the bug against RHEL 7.

If clone returns EPERM the stack is deallocated. After clone is
successful, glibc might return EPERM if it fails to scheduling policies
and/or attributes. On the later case, the thread stack is not released
(no call to __deallocate_stack).

  User wants to know a defined behaviour when pthread_create returns EPERM.

  If the kernel returns EPERM for the clone call, all data is deallocated?
If glibc returns EPERM on subsequent syscalls it would need to create the
thread detached and explicitly call pthread_detach?

Comment 2 Florian Weimer 2019-05-21 12:56:17 UTC

(In reply to Andrew Mike from comment #0)
> This bug was initially created as a copy of Bug #1678421
> 
> I am copying this bug because the customer requested 
> us to refile the bug against RHEL 7.
> 
> If clone returns EPERM the stack is deallocated. After clone is
> successful, glibc might return EPERM if it fails to scheduling policies
> and/or attributes. On the later case, the thread stack is not released
> (no call to __deallocate_stack).
> 
>   User wants to know a defined behaviour when pthread_create returns EPERM.
> 
>   If the kernel returns EPERM for the clone call, all data is deallocated?

Yes, if the clone system call fails, we deallocate the previously allocated resources in the current implementation.

> If glibc returns EPERM on subsequent syscalls it would need to create the
> thread detached and explicitly call pthread_detach?

No, if the start routine never runs, pthread_detach would be the wrong thing to do because it does not report the caller of pthread_create.  Something like this approach is needed, but needs implementation upstream and careful review there:

  <https://sourceware.org/bugzilla/show_bug.cgi?id=19511#c4>

Comment 3 Carlos O'Donell 2019-06-18 19:15:58 UTC

Red Hat Enterprise Linux 7 is entering Maintenance Phase Support 1 this year and only Urgent priority bug fixes will be considered. This issue is not urgent and we are moving this to Red Hat Enterprise Linux 8 for further consideration.

Comment 7 Florian Weimer 2021-10-21 13:53:44 UTC

The core fix is this:

commit 02189e8fb00c3c7f4e67476e21011a22c5dee707
Author: Adhemerval Zanella <adhemerval.zanella>
Date:   Fri May 21 14:19:23 2021 -0300

    nptl: Deallocate the thread stack on setup failure (BZ #19511)

But given of the amount of development that went into 2.34, I doubt it will apply cleanly to our 2.28-based tree.

Comment 14 Siddhesh Poyarekar 2022-11-18 14:10:43 UTC

This fix needs a number of backports that significantly increase the risk of regressions, making it unsuitable for RHEL-8.