Re: [PATCH] x86/alternatives: add missing mutex lock in,alternatives_text_reserved

[Rui Xiang <rui.xiang@huawei.com>]

Hi, all

In Sept 12, I reported a race between alternatives_text_reserved
and alternatives_smp_module_del. That also exists in mainline.

But adding mutex_lock(&smp_alt) in alternatives_text_reserved would
cause an obvious deadlock tested by Xiaolong <xiaolong.ye@intel.com>.
> [   18.325461]  Possible unsafe locking scenario:
> [   18.325461]
> [   18.329341]        CPU0                    CPU1
> [   18.332070]        ----                    ----
> [   18.335136]   lock(text_mutex);
> [   18.337314]                                lock(smp_alt);
> [   18.340894]                                lock(text_mutex);
> [   18.344641]   lock(smp_alt);
> [   18.346744]
> [   18.346744]  *** DEADLOCK ***
> [   18.346744]
> [   18.350563] 2 locks held by swapper/0/1:
> [   18.353217]  #0:  (kprobe_mutex){+.+...}, at: [<ffffffffa914f331>] register_kprobe+0x171/0x720
> [   18.358937]  #1:  (text_mutex){+.+...}, at: [<ffffffffa914f803>] register_kprobe+0x643/0x720
> [   18.364580]
> [   18.364580] stack backtrace:
> [   18.367479] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.8.0-rc5-00438-g7c14957 #1
> [   18.380205]  ffffab5f80073bd0 ffffffffa953120a ffffffffaa8a1c40 ffffffffaa8a1c40
> [   18.397397]  ffffffffaa89d490 ffff99f10e8be820 ffffab5f80073c10 ffffffffa90e26d1
> [   18.409510]  000000000e8be820 ffff99f10e8be858 ffff99f10e8be820 9b6c97af48d594d9
> [   18.423190] Call Trace:
> [   18.424696]  [<ffffffffa953120a>] dump_stack+0xb3/0x119
> [   18.432176]  [<ffffffffa90e26d1>] print_circular_bug+0x201/0x220
> [   18.440272]  [<ffffffffa90e5fc9>] __lock_acquire+0xcd9/0x11c0
> [   18.445740]  [<ffffffffa9149d40>] ? stop_machine_from_inactive_cpu+0x160/0x160
> [   18.450226]  [<ffffffffa90e6891>] lock_acquire+0x71/0x90
> [   18.453574]  [<ffffffffa9026e7b>] ? alternatives_text_reserved+0x1b/0x90
> [   18.470881]  [<ffffffffa98dcbb1>] mutex_lock_nested+0x71/0x370
> [   18.474546]  [<ffffffffa9026e7b>] ? alternatives_text_reserved+0x1b/0x90
> [   18.478720]  [<ffffffffa90e3be2>] ? trace_hardirqs_on_caller+0x192/0x1d0
> [   18.496043]  [<ffffffffa9026e7b>] ? alternatives_text_reserved+0x1b/0x90


It may seem like a bit complicated to fix this race. And how to fix it,
what do you think?

Any advise is welcome.

Thanks,
Rui

On 2016/9/12 15:18, Rui Xiang wrote:
> A race between alternatives_text_reserved and alternatives_smp_module_del
> occured when traversing the global list smp_alt_modules in 3.4 kernel. It
> would casue a panic while reading smp_alt_modules->mod in
> alternatives_text_reserved func.
> 
> [   14.016190] general protection fault: 0000 [#1] SMP
> [   14.016988] CPU 0
> [   14.017287] Modules linked in: mlx4_ib(O+) mlx4_en(O+) xb_sa(O)
> [   14.020005]
> [   14.020005] Pid: 1979, comm: modprobe Tainted: G           O
> [   14.020005] RIP: 0010:[<ffffffff81007eb0>]  [<ffffffff81007eb0>] alternatives_text_reserved+0x20/0x80
> [   14.020005] RSP: 0018:ffff880ea355bcb8  EFLAGS: 00010283
> [   14.020005] RAX: dead000000000100 RBX: ffffffffa02af720 RCX: dead0000000000d0
> [   14.020005] RDX: ffffffffa02f0588 RSI: ffffffffa02d2fc0 RDI: ffffffffa02d2fc0
> [   14.020005] RBP: ffff880ea355bcb8 R08: ffffffffa02f3b68 R09: 00017f4ae12d2fc0
> [   14.020005] R10: 00000000000000e8 R11: ffffffffa02bb9d7 R12: 0000000000000000
> [   14.020005] R13: ffffffffa02af720 R14: ffffffffa0307140 R15: ffffffffa02af730
> [   14.020005] FS:  00007f26c6acc700(0000) GS:ffff880fff200000(0000) knlGS:0000000000000000
> [   14.020005] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> [   14.020005] CR2: 00007fd4adc3b000 CR3: 0000000ea40ea000 CR4: 00000000001407f0
> [   14.020005] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [   14.020005] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> [   14.020005] Process modprobe (pid: 1979, threadinfo ffff880ea355a000, task ffff880e9eb9e600)
> [   14.020005] Stack:
> [   14.020005]  ffff880ea355bcd8 ffffffff8145b148 ffffffffa02af720 ffffffffa02af720
> [   14.020005]  ffff880ea355bd18 ffffffff8145ed38 0000000000000000 0000000000000000
> [   14.041015]  ffff880ea355bd90 ffffffffa02af720 0000000000000001 ffff880ea355bd90
> [   14.041015] Call Trace:
> [   14.041015]  [<ffffffff8145b148>] arch_prepare_kprobe+0x18/0x80
> [   14.042982]  [<ffffffff8145ed38>] register_kprobe+0x338/0x4c0
> [   14.042982]  [<ffffffff8145f658>] register_jprobes+0x98/0xc0
> [   14.042982]  [<ffffffff8145f69a>] register_jprobe+0x1a/0x20
> 
> It also seems to exist in mainline.
> 
> This patch adds the mutex lock smp_alt to fix it.
> 
> Cc: stable@vger.kernel.org
> Signed-off-by: Rui Xiang <rui.xiang@huawei.com>
> ---
>  arch/x86/kernel/alternative.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
> index 5cb272a..7cec0cc 100644
> --- a/arch/x86/kernel/alternative.c
> +++ b/arch/x86/kernel/alternative.c
> @@ -563,6 +563,7 @@ int alternatives_text_reserved(void *start, void *end)
>  	u8 *text_start = start;
>  	u8 *text_end = end;
> 
> +	mutex_lock(&smp_alt);
>  	list_for_each_entry(mod, &smp_alt_modules, next) {
>  		if (mod->text > text_end || mod->text_end < text_start)
>  			continue;
> @@ -573,6 +574,7 @@ int alternatives_text_reserved(void *start, void *end)
>  				return 1;
>  		}
>  	}
> +	mutex_unlock(&smp_alt);	
> 
>  	return 0;
>  }
>