On Fri, Sep 20, 2019 at 10:30:05AM +0200, Christian Brauner wrote:
- Similar precautions should be applied when stacking SECCOMP_RET_USER_NOTIF.
- For SECCOMP_RET_USER_NOTIF filters acting on the same syscall the uppermost
- filter takes precedence. This means that the uppermost
- SECCOMP_RET_USER_NOTIF filter can override any SECCOMP_IOCTL_NOTIF_SEND from
- lower filters essentially allowing all syscalls to pass by using
- SECCOMP_USER_NOTIF_FLAG_CONTINUE. Note that SECCOMP_RET_USER_NOTIF can
^^^^^^^^^^^^^^ This is meant to read RET_TRACE, yes?
- equally be overriden by SECCOMP_USER_NOTIF_FLAG_CONTINUE.
I rewrote this paragraph with that corrected and swapping some "upper/lower" to "most recently added" etc:
+ * Similar precautions should be applied when stacking SECCOMP_RET_USER_NOTIF + * or SECCOMP_RET_TRACE. For SECCOMP_RET_USER_NOTIF filters acting on the + * same syscall, the most recently added filter takes precedence. This means + * that the new SECCOMP_RET_USER_NOTIF filter can override any + * SECCOMP_IOCTL_NOTIF_SEND from earlier filters, essentially allowing all + * such filtered syscalls to be executed by sending the response + * SECCOMP_USER_NOTIF_FLAG_CONTINUE. Note that SECCOMP_RET_TRACE can equally + * be overriden by SECCOMP_USER_NOTIF_FLAG_CONTINUE.
Ultimately, I think this caveat is fine. RET_USER_NOTIF and RET_TRACE are both used from the "process manager" use-case. The benefits of "continue" semantics here outweighs the RET_USER_NOTIF and RET_TRACE "bypass". If we end up in a situation where we need to deal with some kind of nesting where this is a problem in practice, we can revisit this.
Applied to my for-next/seccomp. Thanks!