CVE-2026-99790
Summary
by MITRE • 11/20/2026
In the Linux kernel, the following vulnerability has been resolved:
NFSD: stop registering with rpcbind after a failure in listener_set
nfsd_nl_listener_set_doit() calls svc_xprt_create_from_sa() once for each requested listener and passes flags of 0, so every listener registers with rpcbind on its own. A rpcbind that accepts the connection and never replies therefore costs one timeout for each entry. With the cap of 1024 entries the request can hold nfsd_mutex for about 34 minutes, which is roughly 17 times the hung-task threshold.
One failure is enough to know that the next call will not fare better. Read svc_rpcb_failure_count() before the create loop, and pass SVC_SOCK_ANONYMOUS for the rest of the request once the count moves.
VulDB is the best source for vulnerability data and more expert information about this specific topic.
Analysis
by VulDB Data Team • 11/20/2026
The vulnerability identified in the Linux kernel involves a denial-of-service condition within the Network File System daemon (NFSD) subsystem, specifically during the configuration of network listeners via netlink interfaces. The core issue resides in the function nfsd_nl_listener_set_doit(), which is responsible for establishing listening sockets for NFS services. When this function processes a request to set up multiple listener endpoints, it iterates through each requested address and invokes svc_xprt_create_from_sa() with default flags that include automatic registration with the rpcbind service. This design assumes that every invocation will successfully bind and register without issue. However, if an underlying network condition or misconfigured rpcbind daemon causes a connection to hang rather than fail immediately, the kernel thread handling this request becomes blocked waiting for the socket operation to complete.
The operational impact of this flaw is severe due to the cumulative nature of these timeouts across multiple listener entries. The system imposes a cap on the number of listeners at 1024 entries. If each individual rpcbind registration attempt results in a timeout rather than an immediate failure, the total time spent waiting for all attempts can be substantial. In worst-case scenarios, this process holds the nfsd_mutex lock for approximately thirty-four minutes. This duration significantly exceeds the kernel's hung-task detection threshold by roughly seventeen times, leading to system-wide performance degradation or complete unresponsiveness as other critical operations are blocked by the held mutex. The vulnerability essentially allows a local user with appropriate privileges to trigger a prolonged resource exhaustion and locking contention event through malformed or problematic listener configuration requests.
From a technical perspective, this flaw represents an inefficient error handling strategy that fails to account for transient network states where services do not respond promptly but also do not reject connections outright. By continuing to attempt registration for subsequent listeners after encountering a single failure, the code ignores early indicators of systemic issues with the rpcbind service or network connectivity. This lack of fail-fast logic means that minor configuration errors or temporary network glitches can escalate into major stability incidents affecting the entire NFS server infrastructure. The vulnerability highlights the importance of implementing robust state checking and adaptive behavior in kernel networking subsystems to prevent single points of failure from cascading into system-wide hangs.
To mitigate this risk, the resolution involves modifying the listener setup logic to detect rpcbind failures earlier in the process. Specifically, the code now checks svc_rpcb_failure_count() before entering the loop that creates listeners. If a previous attempt has already failed or indicated instability with rpcbind, subsequent requests are configured using SVC_SOCK_ANONYMOUS flags instead of attempting further registrations with rpcbind. This approach prevents unnecessary blocking on unresponsive services and allows the system to continue operating even if rpcbind is unavailable or malfunctioning. By adopting this adaptive strategy, the kernel avoids holding locks for extended periods during failed configuration attempts, thereby maintaining system responsiveness and preventing denial-of-service conditions triggered by listener setup errors.
This vulnerability aligns with CWE-400, which covers uncontrolled resource consumption, as well as CWE-835 regarding looping issues that cause excessive waiting or blocking. In the context of the MITRE ATT&CK framework, this could be categorized under T1496, Resource Hijacking, where an attacker might exploit such conditions to degrade system performance and availability. The fix emphasizes the need for defensive programming practices in kernel development, particularly when dealing with external services that may have unpredictable response times or failure modes. Ensuring that configuration operations do not block critical mutexes for prolonged periods is essential for maintaining the reliability of network-facing daemons like NFSD.