Re: [PATCH v2] arm: prevent BUG_ON in audit_syscall_entry()

From: Kees Cook
Date: Wed Oct 01 2014 - 15:02:09 EST


On Wed, Oct 1, 2014 at 3:45 AM, AKASHI Takahiro
<takahiro.akashi@xxxxxxxxxx> wrote:
> BUG_ON() in audit_syscall_entry() is hit under the following conditions:
> 1) syscall auditing is enabled (that is, by running auditd).
> 2) user process issues syscall(-1)
> In fact, syscall(-1) just fails (not raising SIGILL, that is an expected
> behavior if not audited), and the succeeding syscall will hit this BUG_ON.

Thanks for digging in to this!

>
> ------------[ cut here ]------------
> kernel BUG at (...)/kernel/auditsc.c:1534!
> Internal error: Oops - BUG: 0 [#1] SMP ARM
> Modules linked in:
> CPU: 0 PID: 61 Comm: syscall_arm Not tainted 3.17.0-rc4 #60
> task: ea0c4380 ti: ea48e000 task.ti: ea48e000
> PC is at __audit_syscall_entry+0xe4/0x110
> LR is at 0xea0c4380
> pc : [<c02b2ae8>] lr : [<ea0c4380>] psr: 20000013
> sp : ea48ff68 ip : 00000001 fp : 0000869c
> r10: 00000200 r9 : ea48e000 r8 : c020f4e4
> r7 : 000000c5 r6 : ea48e000 r5 : ea48ffb0 r4 : ea490800
> r3 : bef1e718 r2 : 00000001 r1 : 000000c5 r0 : 40000028
> Flags: nzCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
> Control: 10c5387d Table: 8a4ac059 DAC: 00000015
> Process syscall_arm (pid: 61, stack limit = 0xea48e250)
> Stack: (0xea48ff68 to 0xea490000)
> ff60: 540e7a69 3ab5e840 00000200 000000c5 ea48ffb0 ea48e000
> ff80: 000000c5 c02114bc bef1e718 00000001 b6efdb58 ffffffff 08400000 000000c5
> ffa0: c020f4e4 c020f49c b6efdb58 ffffffff 00000001 bef1e718 bef1e718 00000001
> ffc0: b6efdb58 ffffffff 08400000 000000c5 00000000 b6f22850 00000008 0000869c
> ffe0: 000000c5 bef1e704 b6ea315f b6e318e6 20000030 00000001 00000000 00000000
> [<c02b2ae8>] (__audit_syscall_entry) from [<c02114bc>] (syscall_trace_enter+0xf0/0x120)
> [<c02114bc>] (syscall_trace_enter) from [<c020f49c>] (__sys_trace+0xc/0x38)
> Code: e584500c e5842004 e28dd00c e8bd80f0 (e7f001f2)
> ---[ end trace 0c4441660aba5692 ]---
>
> Under the current implementation, audit_syscall_entry() is called even
> against syscall(-1), but audit_syscall_exit() is not. Because 'in_syscall'
> flag in a current thread's audit context is kept on, next called
> audit_syscall_entry() will see BUG_ON(in_syscall).
>
> This patch fixes this bug by
> 1) preventing syscall enter/exit tracing, ftrace tracepoint and audit,
> from being executed if an invalid (including -1) or pseudo syscall
> number (starting from __NR_ARM_BASE + 1).
> Please note that tracehook_report_syscall(PTRACE_SYSCALL_EXIT) is
> still executed as it should be paired with
> tracehook_report_syscall(PTRACE_SYSCALL_ENTER).
>
> 2) handling user-issued syscall(-1) with arm_syscall() for compatibility
> if tracing is on.
> Since UL(-1) > __NR_ARM_BASE - __NR_SYSCALL_BASE, arm_syscall() is
> called at local_restart in entry-common.S if tracing is off.
>
> Signed-off-by: AKASHI Takahiro <takahiro.akashi@xxxxxxxxxx>
> ---
> arch/arm/include/asm/traps.h | 1 +
> arch/arm/kernel/entry-common.S | 3 ++-
> arch/arm/kernel/ptrace.c | 49 +++++++++++++++++++++++++---------------
> 3 files changed, 34 insertions(+), 19 deletions(-)
>
> diff --git a/arch/arm/include/asm/traps.h b/arch/arm/include/asm/traps.h
> index f555bb3..de01145 100644
> --- a/arch/arm/include/asm/traps.h
> +++ b/arch/arm/include/asm/traps.h
> @@ -49,6 +49,7 @@ static inline int in_exception_text(unsigned long ptr)
> extern void __init early_trap_init(void *);
> extern void dump_backtrace_entry(unsigned long where, unsigned long from, unsigned long frame);
> extern void ptrace_break(struct task_struct *tsk, struct pt_regs *regs);
> +extern int arm_syscall(int no, struct pt_regs *regs);
>
> extern void *vectors_page;
>
> diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
> index e52fe5a..4c62324 100644
> --- a/arch/arm/kernel/entry-common.S
> +++ b/arch/arm/kernel/entry-common.S
> @@ -476,10 +476,11 @@ __sys_trace:
> cmp scno, #-1 @ skip the syscall?
> bne 2b
> add sp, sp, #S_OFF @ restore stack
> - b ret_slow_syscall
> + b __sys_trace_return_skipped
>
> __sys_trace_return:
> str r0, [sp, #S_R0 + S_OFF]! @ save returned r0
> +__sys_trace_return_skipped:
> mov r0, sp
> bl syscall_trace_exit
> b ret_slow_syscall
> diff --git a/arch/arm/kernel/ptrace.c b/arch/arm/kernel/ptrace.c
> index 0c27ed6..d458367 100644
> --- a/arch/arm/kernel/ptrace.c
> +++ b/arch/arm/kernel/ptrace.c
> @@ -930,7 +930,9 @@ static void tracehook_report_syscall(struct pt_regs *regs,
>
> asmlinkage int syscall_trace_enter(struct pt_regs *regs, int scno)
> {
> - current_thread_info()->syscall = scno;
> + int orig_scno;
> +
> + current_thread_info()->syscall = orig_scno = scno;
>
> /* Do the secure computing check first; failures should be fast. */
> if (secure_computing(scno) == -1)
> @@ -941,31 +943,42 @@ asmlinkage int syscall_trace_enter(struct pt_regs *regs, int scno)
>
> scno = current_thread_info()->syscall;
>
> - if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
> - trace_sys_enter(regs, scno);
> + if (scno >= 0 && scno < NR_syscalls) {
> + if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
> + trace_sys_enter(regs, scno);
>
> - audit_syscall_entry(AUDIT_ARCH_ARM, scno, regs->ARM_r0, regs->ARM_r1,
> - regs->ARM_r2, regs->ARM_r3);
> + audit_syscall_entry(AUDIT_ARCH_ARM, scno,
> + regs->ARM_r0, regs->ARM_r1,
> + regs->ARM_r2, regs->ARM_r3);
> + } else if (scno == -1 && orig_scno == -1) {
> + /*
> + * For compatibility, we handles user-issued syscall(-1)
> + * here with arm_syscall().
> + */
> + arm_syscall(scno, regs);

Won't this always land in bad_syscall(-1, regs) during arm_syscall?
Should syscall_trace-enter just call bad_syscall directly instead, or
are you using arm_syscall since it's the common entry point for this
kind of logic on "non-standard" syscalls?

Either way, this look good! Thanks!

Reviewed-by: Kees Cook <keescook@xxxxxxxxxxxx>

-Kees

> + }
>
> return scno;
> }
>
> asmlinkage void syscall_trace_exit(struct pt_regs *regs)
> {
> - /*
> - * Audit the syscall before anything else, as a debugger may
> - * come in and change the current registers.
> - */
> - audit_syscall_exit(regs);
> + if (current_thread_info()->syscall < NR_syscalls) {
> + /*
> + * Audit the syscall before anything else, as a debugger may
> + * come in and change the current registers.
> + */
> + audit_syscall_exit(regs);
>
> - /*
> - * Note that we haven't updated the ->syscall field for the
> - * current thread. This isn't a problem because it will have
> - * been set on syscall entry and there hasn't been an opportunity
> - * for a PTRACE_SET_SYSCALL since then.
> - */
> - if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
> - trace_sys_exit(regs, regs_return_value(regs));
> + /*
> + * Note that we haven't updated the ->syscall field for the
> + * current thread. This isn't a problem because it will have
> + * been set on syscall entry and there hasn't been
> + * an opportunity for a PTRACE_SET_SYSCALL since then.
> + */
> + if (test_thread_flag(TIF_SYSCALL_TRACEPOINT))
> + trace_sys_exit(regs, regs_return_value(regs));
> + }
>
> if (test_thread_flag(TIF_SYSCALL_TRACE))
> tracehook_report_syscall(regs, PTRACE_SYSCALL_EXIT);
> --
> 1.7.9.5
>



--
Kees Cook
Chrome OS Security
--
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/