CVE-2026-72464 in Linux
Summary
by MITRE • 08/15/2026
In the Linux kernel, the following vulnerability has been resolved:
xprtrdma: Repost Receive buffers for malformed replies
rpcrdma_wc_receive() decrements the transport's Receive count for every completion before it dispatches a successful Receive to rpcrdma_reply_handler(). The handler must post a replacement Receive WR before returning unless ownership of the rep has moved elsewhere, as on the backchannel path.
Commit 2ae50ad68cd7 ("xprtrdma: Close window between waking RPC senders and posting Receives") moved the Receive refill out of rpcrdma_wc_receive(), where it had run ahead of every reply, into rpcrdma_reply_handler() so that the responder's credit grant could be parsed before reposting. The bad-version and short-reply exits never reach that refill: they recycle the rep and return without calling rpcrdma_post_recvs().
A remote peer can therefore drain the client's posted Receive queue by sending a sustained stream of replies that are shorter than the fixed transport header or that carry an unrecognized RPC/RDMA version. Each such reply consumes one posted Receive without replacing it. Once the queue empties, the peer's next Send finds no posted Receive and the transport stalls until reconnect.
Route both malformed-reply exits through the shared repost tail after recycling the rep, refilling against buf->rb_credits, the most recent accepted credit grant. Neither exit updates the congestion window, so RPCs admitted under the previous grant remain in flight awaiting replies. A smaller refill target would let a stream of malformed replies ratchet the posted Receive count down to the batch floor while the congestion window still admits rb_credits RPCs; a burst of valid replies to those RPCs could then overrun the posted Receives, and because the client connects with rnr_retry_count of zero, a single RNR NAK terminates the connection. Refilling against rb_credits also restores the target that applied to malformed replies before commit 2ae50ad68cd7 ("xprtrdma: Close window between waking RPC senders and posting Receives") when rpcrdma_post_recvs() computed it from rb_credits internally. rb_credits is at least one from connection establishment onward, so the repost path always keeps Receives posted.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 08/15/2026
The vulnerability in question affects the Linux kernel's rpcrdma implementation within the xprtrdma transport layer, specifically concerning how receive buffers are managed during handling of malformed RPC/RDMA replies. The core issue stems from a change introduced in commit 2ae50ad68cd7 that restructured the logic for reposting receive work requests. Previously, receive buffer refilling occurred within rpcrdma_wc_receive() function before dispatching successful receives to rpcrdma_reply_handler(). This design ensured that every completion decremented the transport's receive count and that replacements were posted appropriately. However, the refactoring moved this refill operation into rpcrdma_reply_handler(), creating a gap in handling specific error conditions.
The flaw manifests when remote peers send malformed replies that either contain fewer bytes than the fixed transport header or carry an unrecognized RPC/RDMA version. In these cases, the rpcrdma_reply_handler() function recycles the receive buffer and returns without calling rpcrdma_post_recvs(), leaving the transport's receive queue depleted. This behavior represents a classic resource exhaustion vulnerability where malicious actors can systematically consume all available receive buffers through sustained streams of malformed replies. The vulnerability aligns with CWE-400, which covers unrestricted resource consumption, and specifically relates to improper handling of error conditions in network protocols.
The operational impact of this vulnerability is severe as it leads to complete transport stallage. Once the client's posted receive queue becomes empty, any subsequent send operations from the remote peer will find no available receive buffers to process incoming replies. This condition forces the transport layer to wait indefinitely until a reconnection occurs, effectively causing denial-of-service conditions for legitimate RPC communications. The problem is exacerbated by the fact that the client connects with rnr_retry_count set to zero, meaning that even a single RNR NAK can terminate the entire connection. This behavior creates a cascading failure scenario where malformed replies can quickly bring down the entire RPC transport infrastructure, particularly affecting high-throughput distributed systems relying on RDMA-based communications.
The fix implements a consistent reposting mechanism for all malformed reply handling paths by routing both error exits through a shared repost tail after recycling the receive buffer. This approach ensures that receive buffers are properly replenished against buf->rb_credits, which represents the most recent accepted credit grant, rather than relying on the previous implementation's internal computation of credits. The solution addresses the root cause by maintaining a minimum number of posted receives at all times, as rb_credits is guaranteed to be at least one from connection establishment. This mitigation directly aligns with ATT&CK technique T1499.004 for resource exhaustion and follows security best practices for preventing denial-of-service conditions in network protocols, particularly those involving RDMA transports where proper buffer management is critical for maintaining system availability and preventing cascading failures that could impact enterprise-level distributed computing environments.