Re: [PATCH 1/2] memcg: cgroup fix rmdir hang

[Daisuke Nishimura]

On Tue, 23 Jun 2009 16:08:54 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> 
> Now, cgroup has a logic to wait until ready-to-rmdir for avoiding
> frequent -EBUSY at rmdir.
>  (See Commit ec64f51545fffbc4cb968f0cea56341a4b07e85a
>   cgroup: fix frequent -EBUSY at rmdir.
> 
> Nishimura-san reported bad case for waiting and This is a fix to
> make it reliable. A thread waiting for thread cannot be waken up
> when a refcnt gotten by css_tryget() isn't put immediately.
> (Original code assumed css_put() will be called soon.)
> 
> memcg has this case and this is a fix for the problem. This adds
> retry_rmdir() callback to subsys and check we can sleep or not.
> 
> Note: another solution will be adding "rmdir state" to subsys.
> But it will be much complicated than this do-enough-check solution.
> 
> Changelog v1 -> v2:
>  - splitted into 2 patches. This just includes retry_rmdir() modification.
> 
> Reported-by: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>

Looks good to me.

	Reviewed-by: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>

I've been testing with [2/2] applied (because both of these patches are
necessary to fix this problem completely), and it works fine so far.
But I want to test in more test cases.


Thanks,
Daisuke Nishimura.

> ---
>  Documentation/cgroups/cgroups.txt |   11 +++++++++++
>  include/linux/cgroup.h            |    1 +
>  kernel/cgroup.c                   |   20 +++++++++++++++++++-
>  mm/memcontrol.c                   |   14 ++++++++++++--
>  4 files changed, 43 insertions(+), 3 deletions(-)
> 
> Index: fix-rmdir-cgroup/include/linux/cgroup.h
> ===================================================================
> --- fix-rmdir-cgroup.orig/include/linux/cgroup.h
> +++ fix-rmdir-cgroup/include/linux/cgroup.h
> @@ -374,6 +374,7 @@ struct cgroup_subsys {
>  	struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
>  						  struct cgroup *cgrp);
>  	int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
> +	int (*retry_rmdir)(struct cgroup_subsys *ss, struct cgroup *cgrp);
>  	void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
>  	int (*can_attach)(struct cgroup_subsys *ss,
>  			  struct cgroup *cgrp, struct task_struct *tsk);
> Index: fix-rmdir-cgroup/kernel/cgroup.c
> ===================================================================
> --- fix-rmdir-cgroup.orig/kernel/cgroup.c
> +++ fix-rmdir-cgroup/kernel/cgroup.c
> @@ -636,6 +636,23 @@ static int cgroup_call_pre_destroy(struc
>  		}
>  	return ret;
>  }
> +/*
> + * Call subsys's retry_rmdir() handler. If this returns non-Zero, we retry
> + * rmdir immediately and call pre_destroy again.
> + */
> +static int cgroup_check_retry_rmdir(struct cgroup *cgrp)
> +{
> +	struct cgroup_subsys *ss;
> +	int ret = 0;
> +
> +	for_each_subsys(cgrp->root, ss)
> +		if (ss->pre_destroy) {
> +			ret = ss->retry_rmdir(ss, cgrp);
> +			if (ret)
> +				break;
> +		}
> +	return ret;
> +}
>  
>  static void free_cgroup_rcu(struct rcu_head *obj)
>  {
> @@ -2722,7 +2739,8 @@ again:
>  
>  	if (!cgroup_clear_css_refs(cgrp)) {
>  		mutex_unlock(&cgroup_mutex);
> -		schedule();
> +		if (!cgroup_check_retry_rmdir(cgrp))
> +			schedule();
>  		finish_wait(&cgroup_rmdir_waitq, &wait);
>  		clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
>  		if (signal_pending(current))
> Index: fix-rmdir-cgroup/mm/memcontrol.c
> ===================================================================
> --- fix-rmdir-cgroup.orig/mm/memcontrol.c
> +++ fix-rmdir-cgroup/mm/memcontrol.c
> @@ -1457,8 +1457,6 @@ __mem_cgroup_commit_charge_swapin(struct
>  		}
>  		rcu_read_unlock();
>  	}
> -	/* add this page(page_cgroup) to the LRU we want. */
> -
>  }
>  
>  void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
> @@ -2571,6 +2569,17 @@ static int mem_cgroup_pre_destroy(struct
>  	return mem_cgroup_force_empty(mem, false);
>  }
>  
> +static int mem_cgroup_retry_rmdir(struct cgroup_subsys *ss,
> +				  struct cgroup *cont)
> +{
> +	struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
> +
> +	if (res_counter_read_u64(&mem->res, RES_USAGE))
> +		return 1;
> +	return 0;
> +}
> +
> +
>  static void mem_cgroup_destroy(struct cgroup_subsys *ss,
>  				struct cgroup *cont)
>  {
> @@ -2610,6 +2619,7 @@ struct cgroup_subsys mem_cgroup_subsys =
>  	.subsys_id = mem_cgroup_subsys_id,
>  	.create = mem_cgroup_create,
>  	.pre_destroy = mem_cgroup_pre_destroy,
> +	.retry_rmdir = mem_cgroup_retry_rmdir,
>  	.destroy = mem_cgroup_destroy,
>  	.populate = mem_cgroup_populate,
>  	.attach = mem_cgroup_move_task,
> Index: fix-rmdir-cgroup/Documentation/cgroups/cgroups.txt
> ===================================================================
> --- fix-rmdir-cgroup.orig/Documentation/cgroups/cgroups.txt
> +++ fix-rmdir-cgroup/Documentation/cgroups/cgroups.txt
> @@ -500,6 +500,17 @@ there are not tasks in the cgroup. If pr
>  rmdir() will fail with it. From this behavior, pre_destroy() can be
>  called multiple times against a cgroup.
>  
> +int retry_rmdir(struct cgroup_subsys *ss, struct cgroup *cgrp);
> +
> +Called at rmdir right after the kernel finds there are remaining refcnt on
> +subsystems after pre_destroy(). When retry_rmdir() returns 0, the caller enter
> +sleep and wakes up when css's refcnt goes down to 0 by css_put().
> +When this returns 1, the caller doesn't sleep and retry rmdir immediately.
> +This is useful when the subsys knows remaining css's refcnt is not temporal
> +and to calling pre_destroy() again is proper way to remove that.
> +(or proper way to retrun -EBUSY.)
> +
> +
>  int can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
>  	       struct task_struct *task)
>  (cgroup_mutex held by caller)
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
--
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/