kernel: multiple vulnerabilities

CVE-2014-9715: It was found that the netfilter connection tracking subsystem used too small a type as an offset within each connection's data structure, following a bug fix in Linux 3.2.33 and 3.6. In some configurations, this would lead to memory corruption and crashes (even without malicious traffic). This could potentially also result in violation of the netfilter policy or remote code execution.

This can be mitigated by disabling connection tracking accounting: sysctl net.netfilter.nf_conntrack_acct=0

CVE-2015-2830: Andrew Lutomirski discovered that when a 64-bit task on an amd64 kernel makes a fork(2) or clone(2) system call using int $0x80, the 32-bit compatibility flag is set (correctly) but is not cleared on return. As a result, both seccomp and audit will misinterpret the following system call by the task(s), possibly leading to a violation of security policy.

CVE-2015-3331: Stephan Mueller discovered that the optimised implementation of RFC4106 GCM for x86 processors that support AESNI miscalculated buffer addresses in some cases. If an IPsec tunnel is configured to use this mode (also known as AES-GCM-ESP) this can lead to memory corruption and crashes (even without malicious traffic). This could potentially also result in remote code execution.

CVE-2015-3332: Ben Hutchings discovered that the TCP Fast Open feature regressed in Linux 3.16.7-ckt9, resulting in a kernel BUG when it is used. This can be used as a local denial of service.

CVE-2015-3339: It was found that the execve(2) system call can race with inode attribute changes made by chown(2). Although chown(2) clears the setuid/setgid bits of a file if it changes the respective owner ID, this race condition could result in execve(2) setting effective uid/gid to the new owner ID, a privilege escalation.