CVE-2026-98352: RDMA/rtrs-clt: Fix CQ pool leak when connect is interrupted

Published Oct 6, 2026
·
Updated

In the Linux kernel, the following vulnerability has been resolved:

RDMA/rtrs-clt: Fix CQ pool leak when connect is interrupted

The client borrows shared CQ credits in the ADDRRESOLVED handler via ibcqpoolget(), before the peer is connected. createcm() can return -ERESTARTSYS from waiteventinterruptibletimeout() without destroying the CM ID. The initconns() and stop-and-destroy paths then call destroyconcqqp() while cq is still NULL (no PUT) and only afterwards rdmadestroyid().

CMA serializes the handler against rdmadestroyid() with handlermutex, but that does not order the GET against destroyconcqqp(). If ADDRRESOLVED has already passed the DESTROYING check, it can take conmutex, GET credits, and then lose the con to kfree. Device unregister later hits WARNON(cq->cqeused) in ibcqpoolcleanup().

Set a per-connection flag under conmutex before CQ/QP teardown so a racing ADDRRESOLVED cannot borrow credits after teardown has begun.

Affected Software

1 affected component
Linux Linux kernel

Event History

Oct 6, 2026
CVE Published
via MITRE·08:46 AM
Data Sourced
via MITRE·08:46 AM
Description
Data Sourced
via NVD·09:18 AM
Description

Frequently Asked Questions

1

Which systems are exposed to this issue?

Systems using the Linux kernel RDMA RTRS client (rtrs-clt) and establishing RTRS connections can encounter the leak. The issue is tied to connection setup and teardown activity in the RDMA client path.

2

What condition triggers the problem?

The race requires connection setup to be interrupted while the ADDR_RESOLVED handler is borrowing shared completion-queue credits, before the peer is connected. In particular, create_cm() can return -ERESTARTSYS after an interruptible wait, allowing teardown to race with the handler.

3

How can administrators recognize that the issue has occurred?

A later RDMA device unregister may emit a WARN_ON(cq->cqe_used) in ib_cq_pool_cleanup(). This indicates completion-queue credits remained in use because they were not returned during the interrupted connection path.

4

What does the fix change?

The fix sets a per-connection flag under con_mutex before completion-queue and queue-pair teardown begins. This prevents a racing ADDR_RESOLVED handler from borrowing CQ credits after teardown has started.

Contact

SecAlerts Pty Ltd.
132 Wickham Terrace
Fortitude Valley,
QLD 4006, Australia
info@secalerts.co
By using SecAlerts services, you agree to our services end-user license agreement. This website is safeguarded by reCAPTCHA and governed by the Google Privacy Policy and Terms of Service. All names, logos, and brands of products are owned by their respective owners, and any usage of these names, logos, and brands for identification purposes only does not imply endorsement. If you possess any content that requires removal, please get in touch with us.
© 2026 SecAlerts Pty Ltd.
ABN: 70 645 966 203, ACN: 645 966 203