Bug 1888660

Summary: glibc: pthread_atfork handlers that call pthread_atfork deadlock
Product: Red Hat Enterprise Linux 8 Reporter: Paulo Andrade <pandrade>
Component: glibcAssignee: Arjun Shankar <ashankar>
Status: CLOSED ERRATA QA Contact: Martin Coufal <mcoufal>
Severity: medium Docs Contact: Jacob Taylor Valdez <jvaldez>
Priority: unspecified    
Version: 8.2CC: ashankar, codonell, dj, fweimer, jvaldez, mcoufal, mnewsome, pfrankli, sipoyare, skolosov, vikpatil
Target Milestone: rcKeywords: Bugfix, Patch, Triaged
Target Release: 8.0   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: glibc-2.28-206.el8 Doc Type: Bug Fix
Doc Text:
.Applications no longer deadlock when invoking `pthread_atfork` or `dclose` from fork handler callbacks Previously, applications invoked `pthread_atfork` handler callbacks while `glibc` had acquired an internal lock. As a result, registering fork handlers or calling `dclose` from a fork handler could deadlock applications. A different synchronization mechanism is now used to protect internal data structures while fork handlers are running. As a result, applications no longer deadlock when invoking `pthread_atfork` or `dclose` from fork handler callbacks.
 Story Points: ---
Clone Of: Environment:
Last Closed: 2022-11-08 10:43:11 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:
Deadline: 2022-07-04   
Attachments:
Description Flags
hang.c none

Description Paulo Andrade 2020-10-15 12:56:46 UTC 
Created attachment 1721818 [details]
hang.c

Previous versions of glibc used a different approach, now, if a thread calls
pthread_atfork from an atfork handler, it will dead lock, attempting to get
the lock, hold during run of the handler.

Looks similar, or same root cause, to issue reported at
https://sourceware.org/bugzilla/show_bug.cgi?id=24595
[[2.28 Regression]: Deadlock in atfork handler which calls dlclose]

The sample hang.c, compiled with:

$ gcc -pthread -o hang hang.c

will show the dead lock.
Comment 1 Florian Weimer 2020-10-15 14:17:17 UTC 
The problem is this:

static void __attribute__((constructor)) init(void)
{
    pthread_atfork(NULL, NULL, init);
}

This recursive call into the fork subsystem now deadlocks. An upstream patch has been posted:

https://sourceware.org/bugzilla/show_bug.cgi?id=24595#c1

There were some concerns about this patch. The key point will be not to call any callback functions while implementation locks are held.

With glibc, re-registering fork handlers after fork (in the subprocess) is not necessary because they are automatically inherited. This is likely not POSIX-compliant, but is long-standing glibc behavior, and it cannot be changed. So as a workaround, you can change this:

static void __attribute__((constructor)) init(void)
{
    pthread_atfork(NULL, NULL, init);
}

to:

static void handler(void)
{
    // Real fork handler here.
}

static void __attribute__((constructor)) init(void)
{
    pthread_atfork(NULL, NULL, handler);
}
Comment 8 Carlos O'Donell 2022-05-13 17:38:01 UTC 
Patches for this are now under review upstream:
https://patchwork.sourceware.org/project/glibc/patch/20220427134625.3759831-1-arjun@redhat.com/
Comment 23 errata-xmlrpc 2022-11-08 10:43:11 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 (glibc bug fix and enhancement update), 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/RHBA-2022:7684