Re: [patch] cpuset: fix crash when adding CPUs to an empty set

[Zefan Li <lizefan@huawei.com>]

Cc: Ingo
Cc: Juri

On 2015/1/28 11:53, Mike Galbraith wrote:
> On Wed, 2015-01-28 at 10:41 +0800, Zefan Li wrote:
> 
>>> ---
>>>  kernel/cpuset.c |    2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> --- a/kernel/cpuset.c
>>> +++ b/kernel/cpuset.c
>>> @@ -511,7 +511,7 @@ static int validate_change(struct cpuset
>>>  	 * tasks.
>>>  	 */
>>>  	ret = -EBUSY;
>>> -	if (is_cpu_exclusive(cur) &&
>>> +	if (is_cpu_exclusive(cur) && !cpumask_empty(cur->cpus_allowed) &&
>>>  	    !cpuset_cpumask_can_shrink(cur->cpus_allowed,
>>>  				       trial->cpus_allowed))
>>>  		goto out;
>>>
>>
>> We should instead fix cpuset_cpumask_can_shrink() to check if the return value of
>> cpumask_any() >= nr_cpu_ids.
> 
> I like 0 lines better, but ok.
> 
> sched: Fix crash if cpuset_cpumask_can_shrink() is passed an empty cpumask
> 
> While creating an exclusive cpuset, we passed cpuset_cpumask_can_shrink()
> an empty cpumask (cur), and dl_bw_of(cpumask_any(cur)) made boom with it.
> 
> [  514.513841] CPU: 0 PID: 6942 Comm: shield.sh Not tainted 3.19.0-master #19
> [  514.513841] Hardware name: MEDIONPC MS-7502/MS-7502, BIOS 6.00 PG 12/26/2007
> [  514.513841] task: ffff880224552450 ti: ffff8800caab8000 task.ti: ffff8800caab8000
> [  514.513841] RIP: 0010:[<ffffffff81073846>]  [<ffffffff81073846>] cpuset_cpumask_can_shrink+0x56/0xb0
> [  514.513841] RSP: 0018:ffff8800caabbda8  EFLAGS: 00010202
> [  514.513841] RAX: 0000ffff00000000 RBX: ffff880037ae2ab8 RCX: 0000000000000000
> [  514.513841] RDX: 0000000000012900 RSI: 0000000000000000 RDI: 000000000000000c
> [  514.513841] RBP: ffff8800caabbdc8 R08: ffff880037ae2ab8 R09: 0000000000000034
> [  514.513841] R10: 00000077e6fd3d1a R11: ffff88022e9b301c R12: ffff880223c98ab8
> [  514.513841] R13: ffff880037ae2a00 R14: 0000000000000004 R15: ffff8802254b3c60
> [  514.513841] FS:  00007fdbab533700(0000) GS:ffff88022fc00000(0000) knlGS:0000000000000000
> [  514.513841] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  514.513841] CR2: 00007ffff8ab8000 CR3: 00000000bef11000 CR4: 00000000000007f0
> [  514.513841] Stack:
> [  514.513841]  ffff880223c98ab8 ffffffff81a46c20 ffff880223c98a00 ffff880037ae2a00
> [  514.513841]  ffff8800caabbdf8 ffffffff810cb82a ffff8800caabbdf8 ffff880037ae2a00
> [  514.513841]  ffff8800ca9e5240 0000000000000000 ffff8800caabbe48 ffffffff810cc877
> [  514.513841] Call Trace:
> [  514.513841]  [<ffffffff810cb82a>] validate_change+0x18a/0x200
> [  514.513841]  [<ffffffff810cc877>] cpuset_write_resmask+0x3b7/0x720
> [  514.513841]  [<ffffffff810c4d58>] cgroup_file_write+0x38/0x100
> [  514.513841]  [<ffffffff811d953a>] kernfs_fop_write+0x12a/0x180
> [  514.513841]  [<ffffffff8116e1a3>] vfs_write+0xb3/0x1d0
> [  514.513841]  [<ffffffff8116ed06>] SyS_write+0x46/0xb0
> [  514.513841]  [<ffffffff8159ced6>] system_call_fastpath+0x16/0x1b
> 
> Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
> 

Acked-by: Zefan Li <lizefan@huawei.com>

This fixes f82f80426f7afcf55953924e71555984a4bd6ce6 
("sched/deadline: Ensure that updates to exclusive cpusets don't break AC").

> ---
>  kernel/sched/core.c |    3 +++
>  1 file changed, 3 insertions(+)
> 
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -4739,6 +4739,9 @@ int cpuset_cpumask_can_shrink(const stru
>  	struct dl_bw *cur_dl_b;
>  	unsigned long flags;
>  
> +	if (!cpumask_weight(cur))
> +		return ret;
> +
>  	rcu_read_lock_sched();
>  	cur_dl_b = dl_bw_of(cpumask_any(cur));
>  	trial_cpus = cpumask_weight(trial);
> 
> 
> .
>