lkml.org 
[lkml]   [2014]   [Jul]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v5 3/3] arm64: Add seccomp support
On 07/24/2014 12:52 PM, Andy Lutomirski wrote:
> On 07/22/2014 02:14 AM, AKASHI Takahiro wrote:
>> secure_computing() should always be called first in syscall_trace_enter().
>>
>> If secure_computing() returns -1, we should stop further handling. Then
>> that system call may eventually fail with a specified return value (errno),
>> be trapped or the process itself be killed depending on loaded rules.
>> In these cases, syscall_trace_enter() also returns -1, that results in
>> skiping a normal syscall handling as well as syscall_trace_exit().
>>
>> Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
>> ---
>> arch/arm64/Kconfig | 14 ++++++++++++++
>> arch/arm64/include/asm/seccomp.h | 25 +++++++++++++++++++++++++
>> arch/arm64/include/asm/unistd.h | 3 +++
>> arch/arm64/kernel/ptrace.c | 5 +++++
>> 4 files changed, 47 insertions(+)
>> create mode 100644 arch/arm64/include/asm/seccomp.h
>>
>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>> index 3a18571..eeac003 100644
>> --- a/arch/arm64/Kconfig
>> +++ b/arch/arm64/Kconfig
>> @@ -32,6 +32,7 @@ config ARM64
>> select HAVE_ARCH_AUDITSYSCALL
>> select HAVE_ARCH_JUMP_LABEL
>> select HAVE_ARCH_KGDB
>> + select HAVE_ARCH_SECCOMP_FILTER
>> select HAVE_ARCH_TRACEHOOK
>> select HAVE_C_RECORDMCOUNT
>> select HAVE_DEBUG_BUGVERBOSE
>> @@ -259,6 +260,19 @@ config ARCH_HAS_CACHE_LINE_SIZE
>>
>> source "mm/Kconfig"
>>
>> +config SECCOMP
>> + bool "Enable seccomp to safely compute untrusted bytecode"
>> + ---help---
>> + This kernel feature is useful for number crunching applications
>> + that may need to compute untrusted bytecode during their
>> + execution. By using pipes or other transports made available to
>> + the process as file descriptors supporting the read/write
>> + syscalls, it's possible to isolate those applications in
>> + their own address space using seccomp. Once seccomp is
>> + enabled via prctl(PR_SET_SECCOMP), it cannot be disabled
>> + and the task is only allowed to execute a few safe syscalls
>> + defined by each seccomp mode.
>> +
>> config XEN_DOM0
>> def_bool y
>> depends on XEN
>> diff --git a/arch/arm64/include/asm/seccomp.h b/arch/arm64/include/asm/seccomp.h
>> new file mode 100644
>> index 0000000..c76fac9
>> --- /dev/null
>> +++ b/arch/arm64/include/asm/seccomp.h
>> @@ -0,0 +1,25 @@
>> +/*
>> + * arch/arm64/include/asm/seccomp.h
>> + *
>> + * Copyright (C) 2014 Linaro Limited
>> + * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License version 2 as
>> + * published by the Free Software Foundation.
>> + */
>> +#ifndef _ASM_SECCOMP_H
>> +#define _ASM_SECCOMP_H
>> +
>> +#include <asm/unistd.h>
>> +
>> +#ifdef CONFIG_COMPAT
>> +#define __NR_seccomp_read_32 __NR_compat_read
>> +#define __NR_seccomp_write_32 __NR_compat_write
>> +#define __NR_seccomp_exit_32 __NR_compat_exit
>> +#define __NR_seccomp_sigreturn_32 __NR_compat_rt_sigreturn
>> +#endif /* CONFIG_COMPAT */
>> +
>> +#include <asm-generic/seccomp.h>
>> +
>> +#endif /* _ASM_SECCOMP_H */
>> diff --git a/arch/arm64/include/asm/unistd.h b/arch/arm64/include/asm/unistd.h
>> index c980ab7..729c155 100644
>> --- a/arch/arm64/include/asm/unistd.h
>> +++ b/arch/arm64/include/asm/unistd.h
>> @@ -31,6 +31,9 @@
>> * Compat syscall numbers used by the AArch64 kernel.
>> */
>> #define __NR_compat_restart_syscall 0
>> +#define __NR_compat_exit 1
>> +#define __NR_compat_read 3
>> +#define __NR_compat_write 4
>> #define __NR_compat_sigreturn 119
>> #define __NR_compat_rt_sigreturn 173
>>
>> diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
>> index 100d7d1..e477f6f 100644
>> --- a/arch/arm64/kernel/ptrace.c
>> +++ b/arch/arm64/kernel/ptrace.c
>> @@ -28,6 +28,7 @@
>> #include <linux/smp.h>
>> #include <linux/ptrace.h>
>> #include <linux/user.h>
>> +#include <linux/seccomp.h>
>> #include <linux/security.h>
>> #include <linux/init.h>
>> #include <linux/signal.h>
>> @@ -1115,6 +1116,10 @@ asmlinkage int syscall_trace_enter(struct pt_regs *regs)
>> saved_x0 = regs->regs[0];
>> saved_x8 = regs->regs[8];
>>
>> + if (secure_computing(regs->syscallno) == -1)
>> + /* seccomp failures shouldn't expose any additional code. */
>> + return -1;
>> +
>
> This will conflict with the fastpath stuff in Kees' tree. (Actually, it's likely to apply cleanly, but fail to
> compile.) The fix is trivial, but, given that the fastpath stuff is new, can you take a look and see if arm64 can use
> it effectively?

I will look into the code later.

> I suspect that the performance considerations are rather different on arm64 as compared to x86 (I really hope that x86
> is the only architecture with the absurd sysret vs. iret distinction), but at least the seccomp_data stuff ought to help
> anywhere. (It looks like there's a distinct fast path, too, so the two-phase thing might also be a fairly large win if
> it's supportable.)
>
> See:
>
> https://git.kernel.org/cgit/linux/kernel/git/kees/linux.git/log/?h=seccomp/fastpath
>
> Also, I'll ask the usual question? What are all of the factors other than nr and args that affect syscall execution?
> What are the audit arch values? Do they match correctly?

As far as I know,

> For example, it looks like, if arm64 adds OABI support, you'll have a problem. (Note that arm currently disables audit
> and seccomp if OABI is enabled for exactly this reason.)

I don't think that arm64 will add OABI support in the future.

> Do any syscall implementations care whether the user code is LE or BE? Are the arguments encoded the same way?

when I implemented audit for arm64, the assumptions were
* If userspace is LE, then the kernel is also LE and if BE, then the kernel is BE.
* the syscall numbers and how arguments are encoded are the same btw BE and LE.
So syscall_get_arch() always return the same value.

> An arm-specific question: will there be any confusion as a result of the fact that compat syscalls seems to stick nr in
> w7, but arm64 puts them somewhere else?

I don't know, but syscall_get_arch() returns ARCH_ARM for 32-bit tasks.

Thanks,
-Takahiro AKASHI

> --Andy


\
 
 \ /
  Last update: 2014-07-24 08:42    [W:0.096 / U:0.352 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site