[PATCH V3] Support for JIT in Seccomp BPF filters.

From: Nicolas Schichan
Date: Wed Apr 24 2013 - 13:28:16 EST


This patch serie adds support for jitted seccomp BPF filters, with the
required modifications to make it work on the ARM architecture.

- The first patch in the serie adds the required boiler plate in the
core kernel seccomp code to invoke the JIT compilation/free code.

- The second patch reworks the ARM BPF JIT code to make the generation
process less dependent on struct sk_filter.

- The last patch actually implements the ARM part in the BPF jit code.

Some benchmarks, on a 1.6Ghz 88f6282 CPU:

Each system call is tested in two way (fast/slow):

- on the fast version, the tested system call is accepted immediately
after checking the architecture (5 BPF instructions).

- on the slow version, the tested system call is accepted after
previously checking for 85 syscall (90 instructions, including the
architecture check).

The tested syscall is invoked in a loop 1000000 time, the reported
time is the time spent in the loop in seconds.

Without Seccomp JIT:

Syscall Time-Fast Time-Slow
--------------- ---------- ----------
gettimeofday 0.389 1.633
getpid 0.406 1.688
getresuid 1.003 2.266
getcwd 1.342 2.128

With Seccomp JIT:

Syscall Time-Fast Time-Slow
--------------- ----------- ---------
gettimeofday 0.348 0.428
getpid 0.365 0.480
getresuid 0.981 1.060
getcwd 1.237 1.294

For reference, the same code without any seccomp filter:

Syscall Time
--------------- -----
gettimeofday 0.119
getpid 0.137
getresuid 0.747
getcwd 1.021

The activation of the BPF JIT for seccomp is still controled with the
/proc/sys/net/core/bpf_jit_enable sysctl knob.

Those changes are based on the latest rmk-for-next branch.

V2 Changes:
- Document the @bpf_func field in struct seccomp_filter as recommended
by Kees Cook.
- Invoke seccomp_bpf_load directly from generated code without going
via a wrapper.

V3 Changes:
- add accessors giving access to the fields in struct seccomp_filter
as recommended by Andrew Morton. This avoids having to include
<linux/filter.h> in <linux/seccomp.h>, and fixes broken include
dependencies on x86_64 allmodconfig build.
- checkpatch fixes (void*) -> (void *)

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/