{"id":"CVE-2026-90065","published":"2026-09-17T17:16:55.327","lastModified":"2026-09-17T17:16:55.327","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nnet/smc: release the internal TCP sock on IPPROTO_SMC socket creation failure\n\nIPPROTO_SMC sockets create an internal TCP sock (\"clcsock\") from the\nproto->init hook. When socket creation fails after proto->init has\nrun - e.g. a cgroup BPF program attached to BPF_CGROUP_INET_SOCK_CREATE\ndenies the socket - sk_common_release() only invokes sk_prot->destroy\nif it is set, but neither smc_inet_prot nor smc_inet6_prot defines it,\nand smc_destruct() returns early unless sk_state is SMC_CLOSED. As a\nresult, every failing socket(AF_INET, SOCK_STREAM, IPPROTO_SMC) call\nleaks one tcp_sock, so an unprivileged task able to attach a deny-all\nBPF_CGROUP_INET_SOCK_CREATE program to its own cgroup can grow kernel\nmemory unboundedly.\n\nAdd a .destroy hook to both protos that releases the clcsock via\nsmc_clcsock_release(). smc_sk_init() hashes the sock into the smc\nhashinfo before the clcsock is created, and smc_diag dumps walk that\nhash dereferencing smc->clcsock without taking clcsock_release_lock,\nwhile sk_common_release() calls .destroy before .unhash. Unhash the\nsock before releasing the clcsock, as __smc_release() does, so a\nconcurrent dump cannot observe the release; the second unhash in\nsk_common_release() is a no-op.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/2f4ece7fed6d915ec64b7735962c0d84d6261032","tags":[]},{"url":"https://git.kernel.org/stable/c/84180988a455fd8e147c6b35c8df65d8b8cc76c2","tags":[]},{"url":"https://git.kernel.org/stable/c/94a913cd521c0f0408820bcdbd80dadf076d97b5","tags":[]},{"url":"https://git.kernel.org/stable/c/cec261b0b4c5c0b044165303198d10ffcdf3414c","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}