On Mar 2, 2016 12:10 PM, "Chris Metcalf" <cmetcalf@xxxxxxxxxx> wrote:
In prepare_exit_to_usermode(), call task_isolation_ready()This needs a comment indicating the intended semantics.
when we are checking the thread-info flags, and after we've handled
the other work, call task_isolation_enter() unconditionally.
In syscall_trace_enter_phase1(), we add the necessary support for
strict-mode detection of syscalls.
We add strict reporting for the kernel exception types that do
not result in signals, namely non-signalling page faults and
non-signalling MPX fixups.
Signed-off-by: Chris Metcalf <cmetcalf@xxxxxxxxxx>
---
arch/x86/entry/common.c | 18 ++++++++++++++++--
arch/x86/kernel/traps.c | 2 ++
arch/x86/mm/fault.c | 2 ++
3 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 03663740c866..27c71165416b 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -21,6 +21,7 @@
#include <linux/context_tracking.h>
#include <linux/user-return-notifier.h>
#include <linux/uprobes.h>
+#include <linux/isolation.h>
#include <asm/desc.h>
#include <asm/traps.h>
@@ -91,6 +92,10 @@ unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
*/
if (work & _TIF_NOHZ) {
enter_from_user_mode();
+ if (task_isolation_check_syscall(regs->orig_ax)) {
+ regs->orig_ax = -1;
+ return 0;
+ }
And I've still heard no explanation of why this part can't use seccomp.
On 07/13/2015 05:47 PM, Andy Lutomirski wrote:
If a user wants a syscall to kill them, use
seccomp. The kernel isn't at fault if the user does a syscall when it
didn't want to enter the kernel.
Interesting! I didn't realize how close SECCOMP_SET_MODE_STRICT
was to what I wanted here. One concern is that there doesn't seem
to be a way to "escape" from seccomp strict mode, i.e. you can't
call seccomp() again to turn it off - which makes sense for seccomp
since it's a security issue, but not so much sense with cpu_isolated.
So, do you think there's a good role for the seccomp() API to play
in achieving this goal? It's certainly not a question of "the kernel at
fault" but rather "asking the kernel to help catch user mistakes"
(typically third-party libraries in our customers' experience). You
could imagine a SECCOMP_SET_MODE_ISOLATED or something.
Alternatively, we could stick with the API proposed in my patch
series, or something similar, and just try to piggy-back on the seccomp
internals to make it happen. It would require Kconfig to ensure
that SECCOMP was enabled though, which obviously isn't currently
required to do cpu isolation.
work &= ~_TIF_NOHZ;TIF_NOHZ is already a hack and IMO this just makes it worse. At the
}
#endif
@@ -254,17 +259,26 @@ static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
if (cached_flags & _TIF_USER_RETURN_NOTIFY)
fire_user_return_notifiers();
+ task_isolation_enter();
+
/* Disable IRQs and retry */
local_irq_disable();
cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
- if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
+ if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS) &&
+ task_isolation_ready())
break;
}
}
+#ifdef CONFIG_TASK_ISOLATION
+# define EXIT_TO_USERMODE_FLAGS (EXIT_TO_USERMODE_LOOP_FLAGS | _TIF_NOHZ)
+#else
+# define EXIT_TO_USERMODE_FLAGS EXIT_TO_USERMODE_LOOP_FLAGS
+#endif
+
very least this should have a comment. It really ought to be either a
static_branch or a flag that's actually switched per cpu.
But this is also a confusing change. Don't override the definition
here -- stick it in the header where it belongs.