RHEL Engineering is moving the tracking of its product development work on RHEL 6 through RHEL 9 to Red Hat Jira (issues.redhat.com). If you're a Red Hat customer, please continue to file support cases via the Red Hat customer portal. If you're not, please head to the "RHEL project" in Red Hat Jira and file new tickets here. Individual Bugzilla bugs in the statuses "NEW", "ASSIGNED", and "POST" are being migrated throughout September 2023. Bugs of Red Hat partners with an assigned Engineering Partner Manager (EPM) are migrated in late September as per pre-agreed dates. Bugs against components "kernel", "kernel-rt", and "kpatch" are only migrated if still in "NEW" or "ASSIGNED". If you cannot log in to RH Jira, please consult article #7032570. That failing, please send an e-mail to the RH Jira admins at rh-issues@redhat.com to troubleshoot your issue as a user management inquiry. The email creates a ServiceNow ticket with Red Hat. Individual Bugzilla bugs that are migrated will be moved to status "CLOSED", resolution "MIGRATED", and set with "MigratedToJIRA" in "Keywords". The link to the successor Jira issue will be found under "Links", have a little "two-footprint" icon next to it, and direct you to the "RHEL project" in Red Hat Jira (issue links are of type "https://issues.redhat.com/browse/RHEL-XXXX", where "X" is a digit). This same link will be available in a blue banner at the top of the page informing you that that bug has been migrated.
Bug 1939607 - hang because of incorrect accounting of readers in vattr rwlock
Summary: hang because of incorrect accounting of readers in vattr rwlock
Keywords:
Status: CLOSED ERRATA
Alias: None
Product: Red Hat Enterprise Linux 8
Classification: Red Hat
Component: 389-ds-base
Version: 8.4
Hardware: Unspecified
OS: Unspecified
unspecified
unspecified
Target Milestone: rc
: ---
Assignee: thierry bordaz
QA Contact: RHDS QE
URL:
Whiteboard: sync-to-jira
Depends On:
Blocks: 2018257
TreeView+ depends on / blocked
 
Reported: 2021-03-16 17:06 UTC by thierry bordaz
Modified: 2021-11-09 22:27 UTC (History)
4 users (show)

Fixed In Version: 389-ds-1.4-8050020210514191740-d5c171fc
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
: 2018257 (view as bug list)
Environment:
Last Closed: 2021-11-09 18:11:20 UTC
Type: Bug
Target Upstream Version:
Embargoed:

Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Github 389ds 389-ds-base issues 4667 0 None open incorrect accounting of readers in vattr rwlock 2021-03-16 17:14:20 UTC
Red Hat Product Errata RHBA-2021:4203 0 None None None 2021-11-09 18:12:06 UTC

Description thierry bordaz 2021-03-16 17:06:35 UTC 
Description of problem:

The hang occurs when a thread (cos cache rebuild) tries to acquire vattr rwlock (the_map->lock) in write. As it remains readers, this thread is stopped but because of priority of writers it blocks others SRCH threads that try to acquire the rwlock in read.

The hang should finished when the readers threads that acquired the lock before writer thread (cos cache) release the lock.
The problem is that there is no others readers threads. The backtrace is only showing readers that are waiting for the writers.

The backtrace is showing 5 readers and 1 writers but the lock is showing 14 readers. So some readers, that complete their task, has not released the lock

(gdb) print *the_map->lock
$41 = {__data = {__readers = 14, __writers = 0, __wrphase_futex = 2, __writers_futex = 1, __pad3 = 0, __pad4 = 0, __cur_writer = 0, __shared = 0, __rwelision = 0 '\000',
    __pad1 = "\000\000\000\000\000\000", __pad2 = 0, __flags = 2},
  __size = "\016\000\000\000\000\000\000\000\002\000\000\000\001", '\000' , "\002\000\000\000\000\000\000", __align = 14}


Related ticket:
#51068 That gives priority to writers and hang later readers
#49873 That acquires the map rwlock at the operation level using a per cpu variable

Version-Release number of selected component (if applicable):
Likely the bug exists since 1.4.1.2 but is more prone to happen since 1.4.3.8

How reproducible:
No identified simple testcase at the moment
ATM it occurs 1% time with freeipa tests


Actual results:
DS hang

Expected results:
DS should not hang
Comment 1 thierry bordaz 2021-03-17 13:18:30 UTC 
The first analyse was wrong. There is no vattr lock leak. Actually a pthread rwlock test program shows the exact same lock dump with 
 - T1 reader holding the lock
 - T2 writer waiting for T1
 - T3 reader waiting for T2
 - T4 reader waiting for T2

(gdb) print *the_map->lock
$41 = {__data = {__readers = 14, __writers = 0, __wrphase_futex = 2, __writers_futex = 1, __pad3 = 0, __pad4 = 0, __cur_writer = 0, __shared = 0, __rwelision = 0 '\000',
    __pad1 = "\000\000\000\000\000\000", __pad2 = 0, __flags = 2},
  __size = "\016\000\000\000\000\000\000\000\002\000\000\000\001", '\000' , "\002\000\000\000\000\000\000", __align = 14}



The RC of the deadlock is a 3 threads deadlock scenario:

[08/Mar/2021:18:09:15.255947668 +0000] conn=4 op=561 ADD dn="cn=FleetCommander Desktop Profile Administrators,cn=roles,cn=accounts,dc=ipa,dc=test"
[08/Mar/2021:18:09:15.261133390 +0000] conn=4 op=562 SRCH base="cn=FleetCommander Desktop Profile Administrators,cn=privileges,cn=pbac,dc=ipa,dc=test" scope=0 filter="(objectClass=*)" attrs="objectClasses aci * attributeTypes"
[08/Mar/2021:18:09:15.263940289 +0000] conn=4 op=563 ADD dn="cn=FleetCommander Desktop Profile Administrators,cn=privileges,cn=pbac,dc=ipa,dc=test"
[08/Mar/2021:18:09:15.264024493 +0000] conn=4 op=562 RESULT err=32 tag=101 nentries=0 wtime=0.000045722 optime=0.002898152 etime=0.002941253
[08/Mar/2021:18:09:15.261304370 +0000] conn=4 op=561 RESULT err=0 tag=105 nentries=0 wtime=0.000103907 optime=0.005360651 etime=0.005394639
Thread 14
conn=4 op=561 ADD "cn=FleetCommander Desktop Profile Administrators,cn=roles,cn=accounts,dc=ipa,dc=test"
        Hold vattr lock in read and wait for DB page (WAIT userRoot/objectclass.db) (hold by Thread 20)
        -> SIDGEN (post-op)
                -> internal SRCH -b "dc=ipa,dc=test" "(objectclass=ipantdomainattrs)"
                   op_shared_search => hold vattr lock in read
                        -> index read => DB page

Thread 20
conn=4 op=563 ADD "cn=FleetCommander Desktop Profile Administrators,cn=privileges,cn=pbac,dc=ipa,dc=test"
        Hold DB page (HOLD userRoot/objectclass.db) waiting for vattr lock in read
        -> ADD -> memberof modify (txnbe post)
                        -> DNA -> internal_search
                                        -> vattr_map_lookup => wait for vattr in read

Thread 6
On backend state change, it rebuild the cos cache
        Try to acquire vattr in write blocking new readers
        Internal search SRCH -b "dc=ipa,dc=test" "(&(|(objectclass=cosSuperDefinition)(objectclass=cosDefinition))(objectclass=ldapsubentry))"
        -> cos_dn_defs_cb
                -> vattr_map_insert : wait vattr in write

        Thread  6 is blocked by Thread 14
        Thread 14 is blocked by Thread 20
        Thread 20 is blocked by Thread  6
Comment 4 thierry bordaz 2021-04-29 07:32:04 UTC 
Fix pushed upstream => POST
Comment 9 bsmejkal 2021-05-27 17:42:36 UTC 
Build tested:
389-ds-base-1.4.3.23-1.module+el8.5.0+11016+7e7e9011.x86_64

I had freeipa installation running 141x times in a loop without hang.
The fix is in the build. Marking as Verified:Tested, SanityOnly.
Comment 12 errata-xmlrpc 2021-11-09 18:11:20 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 (389-ds-base 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-2021:4203
Note You need to log in before you can comment on or make changes to this bug.