[PATCH v2] mm/mempolicy: fix lock contention on mems_allowed

[PATCH v2] mm/mempolicy: fix lock contention on mems_allowed

[Abel Wu]

The mems_allowed field can be modified by other tasks, so it isn't
safe to access it with alloc_lock unlocked even in the current
process context.

Say there are two tasks: A from cpusetA is performing set_mempolicy(2),
and B is changing cpusetA's cpuset.mems:

  A (set_mempolicy)		B (echo xx > cpuset.mems)
  -------------------------------------------------------
  pol = mpol_new();
				update_tasks_nodemask(cpusetA) {
				  foreach t in cpusetA {
				    cpuset_change_task_nodemask(t) {
  mpol_set_nodemask(pol) {
				      task_lock(t); // t could be A
    new = f(A->mems_allowed);
				      update t->mems_allowed;
    pol.create(pol, new);
				      task_unlock(t);
  }
				    }
				  }
				}
  task_lock(A);
  A->mempolicy = pol;
  task_unlock(A);

In this case A's pol->nodes is computed by old mems_allowed, and could
be inconsistent with A's new mems_allowed.

While it is different when replacing vmas' policy: the pol->nodes is
gone wild only when current_cpuset_is_being_rebound():

  A (mbind)			B (echo xx > cpuset.mems)
  -------------------------------------------------------
  pol = mpol_new();
  mmap_write_lock(A->mm);
				cpuset_being_rebound = cpusetA;
				update_tasks_nodemask(cpusetA) {
				  foreach t in cpusetA {
				    cpuset_change_task_nodemask(t) {
  mpol_set_nodemask(pol) {
				      task_lock(t); // t could be A
    mask = f(A->mems_allowed);
				      update t->mems_allowed;
    pol.create(pol, mask);
				      task_unlock(t);
  }
				    }
  foreach v in A->mm {
    if (cpuset_being_rebound == cpusetA)
      pol.rebind(pol, cpuset.mems);
    v->vma_policy = pol;
  }
  mmap_write_unlock(A->mm);
				    mmap_write_lock(t->mm);
				    mpol_rebind_mm(t->mm);
				    mmap_write_unlock(t->mm);
				  }
				}
				cpuset_being_rebound = NULL;

In this case, the cpuset.mems, which has already done updating, is
finally used for calculating pol->nodes, rather than A->mems_allowed.
So it is OK to call mpol_set_nodemask() with alloc_lock unlocked when
doing mbind(2).

Fixes: 78b132e9bae9 ("mm/mempolicy: remove or narrow the lock on current")
Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
---
 mm/mempolicy.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index d39b01fd52fe..61e4e6f5cfe8 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -855,12 +855,14 @@ static long do_set_mempolicy(unsigned short mode, unsigned short flags,
 		goto out;
 	}
 
+	task_lock(current);
 	ret = mpol_set_nodemask(new, nodes, scratch);
 	if (ret) {
+		task_unlock(current);
 		mpol_put(new);
 		goto out;
 	}
-	task_lock(current);
+
 	old = current->mempolicy;
 	current->mempolicy = new;
 	if (new && new->mode == MPOL_INTERLEAVE)
-- 
2.31.1

Re: [PATCH v2] mm/mempolicy: fix lock contention on mems_allowed

[Abel Wu]

On 8/11/22 10:06 PM, Michal Hocko Wrote:
> [Cc Wei Yang who is author of 78b132e9bae9]
> 
> On Thu 11-08-22 20:41:57, Abel Wu wrote:
>> The mems_allowed field can be modified by other tasks, so it isn't
>> safe to access it with alloc_lock unlocked even in the current
>> process context.
>>
>> Say there are two tasks: A from cpusetA is performing set_mempolicy(2),
>> and B is changing cpusetA's cpuset.mems:
>>
>>    A (set_mempolicy)		B (echo xx > cpuset.mems)
>>    -------------------------------------------------------
>>    pol = mpol_new();
>> 				update_tasks_nodemask(cpusetA) {
>> 				  foreach t in cpusetA {
>> 				    cpuset_change_task_nodemask(t) {
>>    mpol_set_nodemask(pol) {
>> 				      task_lock(t); // t could be A
>>      new = f(A->mems_allowed);
>> 				      update t->mems_allowed;
>>      pol.create(pol, new);
>> 				      task_unlock(t);
>>    }
>> 				    }
>> 				  }
>> 				}
>>    task_lock(A);
>>    A->mempolicy = pol;
>>    task_unlock(A);
>>
>> In this case A's pol->nodes is computed by old mems_allowed, and could
>> be inconsistent with A's new mems_allowed.
> 
> Just to clarify. With an unfortunate timing and those two nodemasks
> overlap the end user effect could be a premature OOM because some nodes
> wouldn't be considered, right?

Yes, indeed!

> 
>> While it is different when replacing vmas' policy: the pol->nodes is
>> gone wild only when current_cpuset_is_being_rebound():
>>
>>    A (mbind)			B (echo xx > cpuset.mems)
>>    -------------------------------------------------------
>>    pol = mpol_new();
>>    mmap_write_lock(A->mm);
>> 				cpuset_being_rebound = cpusetA;
>> 				update_tasks_nodemask(cpusetA) {
>> 				  foreach t in cpusetA {
>> 				    cpuset_change_task_nodemask(t) {
>>    mpol_set_nodemask(pol) {
>> 				      task_lock(t); // t could be A
>>      mask = f(A->mems_allowed);
>> 				      update t->mems_allowed;
>>      pol.create(pol, mask);
>> 				      task_unlock(t);
>>    }
>> 				    }
>>    foreach v in A->mm {
>>      if (cpuset_being_rebound == cpusetA)
>>        pol.rebind(pol, cpuset.mems);
>>      v->vma_policy = pol;
>>    }
>>    mmap_write_unlock(A->mm);
>> 				    mmap_write_lock(t->mm);
>> 				    mpol_rebind_mm(t->mm);
>> 				    mmap_write_unlock(t->mm);
>> 				  }
>> 				}
>> 				cpuset_being_rebound = NULL;
>>
>> In this case, the cpuset.mems, which has already done updating, is
>> finally used for calculating pol->nodes, rather than A->mems_allowed.
>> So it is OK to call mpol_set_nodemask() with alloc_lock unlocked when
>> doing mbind(2).
>>
>> Fixes: 78b132e9bae9 ("mm/mempolicy: remove or narrow the lock on current")
>> Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
> 
> The fix looks correct.
> 
>> ---
>>   mm/mempolicy.c | 4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
>> index d39b01fd52fe..61e4e6f5cfe8 100644
>> --- a/mm/mempolicy.c
>> +++ b/mm/mempolicy.c
>> @@ -855,12 +855,14 @@ static long do_set_mempolicy(unsigned short mode, unsigned short flags,
>>   		goto out;
>>   	}
>>   
>> +	task_lock(current);
>>   	ret = mpol_set_nodemask(new, nodes, scratch);
>>   	if (ret) {
>> +		task_unlock(current);
>>   		mpol_put(new);
>>   		goto out;
>>   	}
>> -	task_lock(current);
>> +
>>   	old = current->mempolicy;
>>   	current->mempolicy = new;
>>   	if (new && new->mode == MPOL_INTERLEAVE)
>> -- 
>> 2.31.1
> 

Re: [PATCH v2] mm/mempolicy: fix lock contention on mems_allowed

[Muchun Song]

> On Aug 11, 2022, at 20:41, Abel Wu <wuyun.abel@bytedance.com> wrote:
> 
> The mems_allowed field can be modified by other tasks, so it isn't
> safe to access it with alloc_lock unlocked even in the current
> process context.
> 
> Say there are two tasks: A from cpusetA is performing set_mempolicy(2),
> and B is changing cpusetA's cpuset.mems:
> 
>  A (set_mempolicy)		B (echo xx > cpuset.mems)
>  -------------------------------------------------------
>  pol = mpol_new();
> 				update_tasks_nodemask(cpusetA) {
> 				  foreach t in cpusetA {
> 				    cpuset_change_task_nodemask(t) {
>  mpol_set_nodemask(pol) {
> 				      task_lock(t); // t could be A
>    new = f(A->mems_allowed);
> 				      update t->mems_allowed;
>    pol.create(pol, new);
> 				      task_unlock(t);
>  }
> 				    }
> 				  }
> 				}
>  task_lock(A);
>  A->mempolicy = pol;
>  task_unlock(A);
> 
> In this case A's pol->nodes is computed by old mems_allowed, and could
> be inconsistent with A's new mems_allowed.
> 
> While it is different when replacing vmas' policy: the pol->nodes is
> gone wild only when current_cpuset_is_being_rebound():
> 
>  A (mbind)			B (echo xx > cpuset.mems)
>  -------------------------------------------------------
>  pol = mpol_new();
>  mmap_write_lock(A->mm);
> 				cpuset_being_rebound = cpusetA;
> 				update_tasks_nodemask(cpusetA) {
> 				  foreach t in cpusetA {
> 				    cpuset_change_task_nodemask(t) {
>  mpol_set_nodemask(pol) {
> 				      task_lock(t); // t could be A
>    mask = f(A->mems_allowed);
> 				      update t->mems_allowed;
>    pol.create(pol, mask);
> 				      task_unlock(t);
>  }
> 				    }
>  foreach v in A->mm {
>    if (cpuset_being_rebound == cpusetA)
>      pol.rebind(pol, cpuset.mems);
>    v->vma_policy = pol;
>  }
>  mmap_write_unlock(A->mm);
> 				    mmap_write_lock(t->mm);
> 				    mpol_rebind_mm(t->mm);
> 				    mmap_write_unlock(t->mm);
> 				  }
> 				}
> 				cpuset_being_rebound = NULL;
> 
> In this case, the cpuset.mems, which has already done updating, is
> finally used for calculating pol->nodes, rather than A->mems_allowed.
> So it is OK to call mpol_set_nodemask() with alloc_lock unlocked when
> doing mbind(2).
> 
> Fixes: 78b132e9bae9 ("mm/mempolicy: remove or narrow the lock on current")
> Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>

Reviewed-by: Muchun Song <songmuchun@bytedance.com>

Thanks.

Re: [PATCH v2] mm/mempolicy: fix lock contention on mems_allowed

[Wei Yang]

On Thu, Aug 11, 2022 at 04:11:21PM +0200, Michal Hocko wrote:
>fix the lkml address (fat fingers, sorry)
>
>On Thu 11-08-22 16:06:37, Michal Hocko wrote:
>> [Cc Wei Yang who is author of 78b132e9bae9]
>> 
>> On Thu 11-08-22 20:41:57, Abel Wu wrote:
>> > The mems_allowed field can be modified by other tasks, so it isn't
>> > safe to access it with alloc_lock unlocked even in the current
>> > process context.
>> > 
>> > Say there are two tasks: A from cpusetA is performing set_mempolicy(2),
>> > and B is changing cpusetA's cpuset.mems:
>> > 
>> >   A (set_mempolicy)		B (echo xx > cpuset.mems)
>> >   -------------------------------------------------------
>> >   pol = mpol_new();
>> > 				update_tasks_nodemask(cpusetA) {
>> > 				  foreach t in cpusetA {
>> > 				    cpuset_change_task_nodemask(t) {
>> >   mpol_set_nodemask(pol) {
>> > 				      task_lock(t); // t could be A
>> >     new = f(A->mems_allowed);
>> > 				      update t->mems_allowed;
>> >     pol.create(pol, new);
>> > 				      task_unlock(t);
>> >   }
>> > 				    }
>> > 				  }
>> > 				}
>> >   task_lock(A);
>> >   A->mempolicy = pol;
>> >   task_unlock(A);
>> > 
>> > In this case A's pol->nodes is computed by old mems_allowed, and could
>> > be inconsistent with A's new mems_allowed.
>> 
>> Just to clarify. With an unfortunate timing and those two nodemasks
>> overlap the end user effect could be a premature OOM because some nodes
>> wouldn't be considered, right?
>> 
>> > While it is different when replacing vmas' policy: the pol->nodes is
>> > gone wild only when current_cpuset_is_being_rebound():
>> > 
>> >   A (mbind)			B (echo xx > cpuset.mems)
>> >   -------------------------------------------------------
>> >   pol = mpol_new();
>> >   mmap_write_lock(A->mm);
>> > 				cpuset_being_rebound = cpusetA;
>> > 				update_tasks_nodemask(cpusetA) {
>> > 				  foreach t in cpusetA {
>> > 				    cpuset_change_task_nodemask(t) {
>> >   mpol_set_nodemask(pol) {
>> > 				      task_lock(t); // t could be A
>> >     mask = f(A->mems_allowed);
>> > 				      update t->mems_allowed;
>> >     pol.create(pol, mask);
>> > 				      task_unlock(t);
>> >   }
>> > 				    }
>> >   foreach v in A->mm {
>> >     if (cpuset_being_rebound == cpusetA)
>> >       pol.rebind(pol, cpuset.mems);
>> >     v->vma_policy = pol;
>> >   }
>> >   mmap_write_unlock(A->mm);
>> > 				    mmap_write_lock(t->mm);
>> > 				    mpol_rebind_mm(t->mm);
>> > 				    mmap_write_unlock(t->mm);
>> > 				  }
>> > 				}
>> > 				cpuset_being_rebound = NULL;
>> > 
>> > In this case, the cpuset.mems, which has already done updating, is
>> > finally used for calculating pol->nodes, rather than A->mems_allowed.
>> > So it is OK to call mpol_set_nodemask() with alloc_lock unlocked when
>> > doing mbind(2).
>> > 
>> > Fixes: 78b132e9bae9 ("mm/mempolicy: remove or narrow the lock on current")
>> > Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
>> 

Thanks for pointing out. This looks correct.

Reviewed-by: Wei Yang <richard.weiyang@gmail.com>