On Tue, Feb 28, 2012 at 11:04 AM, Will Drewry <wad@xxxxxxxxxxxx> wrote: > On Tue, Feb 28, 2012 at 10:43 AM, Oleg Nesterov <oleg@xxxxxxxxxx> wrote: >> On 02/27, Will Drewry wrote: >>> On Mon, Feb 27, 2012 at 11:54 AM, Oleg Nesterov <oleg@xxxxxxxxxx> wrote: >>> > On 02/24, Will Drewry wrote: >>> >> >>> >> arch/Kconfig | 1 + >>> >> include/linux/ptrace.h | 7 +++++-- >>> >> include/linux/seccomp.h | 4 +++- >>> >> include/linux/tracehook.h | 6 ++++++ >>> >> kernel/ptrace.c | 4 ++++ >>> >> kernel/seccomp.c | 18 ++++++++++++++++++ >>> > >>> > FYI, this conflicts with the changes -mm tree. >>> > >>> > The changes in ptrace.* confict with Denys's >>> > "ptrace: simplify PTRACE_foo constants and PTRACE_SETOPTIONS code" >>> > >>> > The change in tracehook.h conflicts with >>> > "ptrace: the killed tracee should not enter the syscall" >>> >>> What's the best way to reconcile this in this day and age? >> >> Of course I'd prefer if you make this change on top of Denys's patch ;) >> >> Besides, if you agree with PTRACE_EVENT_SECCOMP/PTRACE_O_SECCOMP you >> need only one trivial change in ptrace.h. > > I think that works quite well :) > >>> I don't see >>> these in kernel-next yet and I can't tell if there is a public -mm >>> anywhere anymore. >> >> Strange... I didn't check, but every patch in >> http://marc.info/?l=linux-mm-commits has this note: >> >> The -mm tree is included into linux-next and is updated >> there every 3-4 working days > > It appears to have been pulled in ~8 hours ago. I'm rebasing to next now. > >>> >> --- a/kernel/seccomp.c >>> >> +++ b/kernel/seccomp.c >>> >> @@ -354,6 +354,24 @@ int __secure_computing_int(int this_syscall) >>> >> seccomp_send_sigsys(this_syscall, reason_code); >>> >> return -1; >>> >> } >>> >> + case SECCOMP_RET_TRACE: { >>> >> + int ret; >>> >> + struct pt_regs *regs = task_pt_regs(current); >>> >> + if (!(test_tsk_thread_flag(current, TIF_SYSCALL_TRACE)) || >>> >> + !(current->ptrace & PT_TRACE_SECCOMP)) >>> >> + return -1; >>> >> + /* >>> >> + * PT_TRACE_SECCOMP and seccomp.trace indicate whether >>> >> + * tracehook_report_syscall_entry needs to signal the >>> >> + * tracer. This avoids race conditions in hand off and >>> >> + * the requirement for TIF_SYSCALL_TRACE ensures that >>> >> + * we are in the syscall slow path. >>> >> + */ >>> >> + current->seccomp.trace = 1; >>> >> + ret = tracehook_report_syscall_entry(regs); >>> >> + current->seccomp.trace = 0; >>> >> + return ret; >>> > >>> > To be honest, this interface looks a bit strange to me... >>> > >>> > Once again, sorry if this was already discussed. But perhaps it would >>> > be better to introduce PTRACE_EVENT_SECCOMP/PTRACE_O_SECCOMP instead? >>> > >>> > SECCOMP_RET_TRACE: could simply do ptrace_event(PTRACE_EVENT_SECCOMP) >>> > unconditionaly. The tracer can set the option and do PTRACE_CONT if it >>> > doesn't want the system call notifications. >>> >>> Works for me - this also gets rid of the extra int for brief state >>> tracking. I'll switch over to that in the next rev. >> >> Great. In this case this patch becomes really trivial. Just 2 defines >> in ptrace.h and the unconditional ptrace_event() under SECCOMP_RET_TRACE. hrm the only snag is that I can't then rely on TIF_SYSCALL_TRACE to ensure seccomp is in the slow-path. Right now, on x86, seccomp is slow-path, but it doesn't have to be to have the syscall and args. However, for ptrace to behavior properly, I believed it did need to be in the slow path. If SECCOMP_RET_TRACE doesn't rely on PTRACE_SYSCALL, then it introduces a need for seccomp to always be in the slow path or to flag (somehow) when it needs slow path. Any suggestions there? Thanks! will -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
