Re: [PATCH] arm64: fix a concurrency issue in emulation_proc_handler()

[Mark Rutland <mark.rutland@arm.com>]

On Fri, Dec 09, 2022 at 06:55:56PM +0800, ruanjinjie wrote:
> In emulation_proc_handler(), read and write operations are performed on
> insn->current_mode. In the concurrency scenario, mutex only protects
> writing insn->current_mode, and not protects the read. Suppose there are
> two concurrent tasks, task1 updates insn->current_mode to INSN_EMULATE
> in the critical section, the prev_mode of task2 is still the old data
> INSN_UNDEF of insn->current_mode. As a result, two tasks call
> update_insn_emulation_mode twice with prev_mode = INSN_UNDEF and
> current_mode = INSN_EMULATE, then call register_emulation_hooks twice,
> resulting in a list_add double problem.
> 
> Call trace:
>  __list_add_valid+0xd8/0xe4
>  register_undef_hook+0x94/0x13c
>  update_insn_emulation_mode+0xd0/0x12c
>  emulation_proc_handler+0xd8/0xf4
>  proc_sys_call_handler+0x140/0x250
>  proc_sys_write+0x1c/0x2c
>  new_sync_write+0xec/0x18c
>  vfs_write+0x214/0x2ac
>  ksys_write+0x70/0xfc
>  __arm64_sys_write+0x24/0x30
>  el0_svc_common.constprop.0+0x7c/0x1bc
>  do_el0_svc+0x2c/0x94
>  el0_svc+0x20/0x30
>  el0_sync_handler+0xb0/0xb4
>  el0_sync+0x160/0x180

The version queued in the arm64 for-next/core branch no longer has the list
manipulation, but we do need to fix this for stable, and there is a remaining
race on reading insn->current_mode in emulation_proc_handler().

> Fixes: af483947d472 ("arm64: fix oops in concurrently setting insn_emulation sysctls")
> Signed-off-by: ruanjinjie <ruanjinjie@huawei.com>
> ---
>  arch/arm64/kernel/armv8_deprecated.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/kernel/armv8_deprecated.c b/arch/arm64/kernel/armv8_deprecated.c
> index fb0e7c7b2e20..d33e5d9e6990 100644
> --- a/arch/arm64/kernel/armv8_deprecated.c
> +++ b/arch/arm64/kernel/armv8_deprecated.c
> @@ -208,10 +208,12 @@ static int emulation_proc_handler(struct ctl_table *table, int write,
>  				  loff_t *ppos)
>  {
>  	int ret = 0;
> -	struct insn_emulation *insn = container_of(table->data, struct insn_emulation, current_mode);
> -	enum insn_emulation_mode prev_mode = insn->current_mode;
> +	struct insn_emulation *insn;
> +	enum insn_emulation_mode prev_mode;
>  
>  	mutex_lock(&insn_emulation_mutex);
> +	insn = container_of(table->data, struct insn_emulation, current_mode);
> +	prev_mode = insn->current_mode;
>  	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);

We don't strictly need to move the container_of(), but it makes no odds either
way, and this looks good to me:

Acked-by: Mark Rutland <mark.rutland@arm.com>

Mark.

>  
>  	if (ret || !write || prev_mode == insn->current_mode)
> -- 
> 2.25.1
>