(Internal git) Commit c31d1f3f363a5c3b6dbd4d6daa36fa0f4273e948 "a64: fix deadlock in ia64 sys_ptrace" moved ptrace_check_attach() from find_thread_for_addr() to tasklist-is-not-held area. However it introduced other problems. In the likely case, when rbs_child == child, ptrace_check_attach() is never called. Otherwise we do ptrace_check_attach(rbs_child) twice. This means that any peek/poke request can silently fail and fool the tracer. With this patch the logic is ptrace_check_attach(child); if (rbs_child != child) { // we should use rbs_child ... if (!ptrace_check_attach(rbs_child)) child = rbs_child; // but if check_attach() fails we continue to // use child, we are doing as well as we can. } To simplify the error-handling and the logic, this patch changes the code to detect the likely rbs_child == child case earlier, right after find_thread_for_addr(). This only affects Red Hat Enterprise Linux 4.
From Oleg, the problem is, ptrace_check_attach() also verifies the caller is the tracer and thus it have the rights to do "anything it wants" with the tracee. But since ptrace_check_attach() can be missed, any user can do ptrace() on any target even without PTRACE_ATTACH.
This issue has been addressed in following products: Red Hat Enterprise Linux 4 Via RHSA-2010:0394 https://rhn.redhat.com/errata/RHSA-2010-0394.html