Date: Wed, 1 Apr 2015 12:02:58 -0400
From: Tejun Heo <tj@kernel.org>
To: Aleksa Sarai <cyphar@cyphar.com>
Cc: lizefan@huawei.com, mingo@redhat.com, peterz@infradead.org, richard@nod.at,
        fweisbec@gmail.com, linux-kernel@vger.kernel.org,
        cgroups@vger.kernel.org
Subject: Re: [PATCH v8 3/4] cgroups: allow a cgroup subsystem to reject a fork
Message-ID: <20150401160258.GP9974@htj.duckdns.org>
References: <1427878641-5273-1-git-send-email-cyphar@cyphar.com>
 <1427878641-5273-4-git-send-email-cyphar@cyphar.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1427878641-5273-4-git-send-email-cyphar@cyphar.com>
User-Agent: Mutt/1.5.23 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3066
Lines: 108

Hello, Aleksa.

On Wed, Apr 01, 2015 at 07:57:20PM +1100, Aleksa Sarai wrote:
> +struct cgroup_fork_state {
> +	void *ss_state[CGROUP_SUBSYS_COUNT];
> +};

Can we collect the subsystems which require pre/post fork callbacks to
the front in group_subsys.h and do do CGROUP_SUBSYS_FORK_COUNT (or
whatever) instead?  Then, we don't need all these subsys bitmasks
either we can just test the index against that and be done with it.

> +
> +/**
> + * cgroup_cfs_alloc - allocates an empty cgroup_fork_state
> + */
> +struct cgroup_fork_state *cgroup_cfs_alloc(void)
> +{
> +	struct cgroup_fork_state *cfs;
> +
> +	cfs = kzalloc(sizeof(struct cgroup_fork_state), GFP_KERNEL);
> +	if (!cfs)
> +		return ERR_PTR(-ENOMEM);
> +
> +	return cfs;
> +}

Just make it a void * array and put it on stack.  Abstraction at this
level doesn't serve any purpose.  No controller code is gonna see this
anyway.

> +int cgroup_can_fork(struct task_struct *child, struct cgroup_fork_state *cfs)
> +{
> +	struct cgroup_subsys *ss;
> +	int i, j, retval;
> +
> +	for_each_subsys_which(need_canfork_callback, ss, i) {
> +		retval = ss->can_fork(child, &cfs->ss_state[i]);
> +		if (retval)
> +			goto out_revert;
> +	}
> +
> +	return 0;
> +
> +out_revert:
> +	for_each_subsys_which(need_cancelfork_callback, ss, j) {
> +		if (j >= i)
> +			break;
> +		ss->cancel_fork(child, &cfs->ss_state[i]);

cancel_fork() has no reason to update the opaque pointer.  No reason
to pass pointer of it.

> +void cgroup_cancel_fork(struct task_struct *child, struct cgroup_fork_state *cfs)
> +{
> +	struct cgroup_subsys *ss;
> +	int i;
> +
> +	for_each_subsys_which(need_cancelfork_callback, ss, i) {
> +		void **state = NULL;
> +
> +		/*
> +		 * Only if %ss has a can_fork() callback is %cfs->ss_state[i] meaningful

I don't think we do %var, do we?  % is used for macros and consts.

> +		 * -- otherwise just pass a NULL.
> +		 */
> +		if (need_canfork_callback & (1 << i))
> +			state = &cfs->ss_state[i];
> +
> +		ss->cancel_fork(child, &cfs->ss_state[i]);

Ditto, just pass the pointer itself.

> @@ -5241,8 +5346,18 @@ void cgroup_post_fork(struct task_struct *child)
>  	 * css_set; otherwise, @child might change state between ->fork()
>  	 * and addition to css_set.
>  	 */
> -	for_each_subsys_which(need_fork_callback, ss, i)
> -		ss->fork(child);
> +	for_each_subsys_which(need_fork_callback, ss, i) {
> +		void **state = NULL;
> +
> +		/*
> +		 * Only if %ss has a can_fork() callback is %old_cfs->ss_state[i]
> +		 * meaningful -- otherwise just pass a NULL.
> +		 */

Again, if you just passed the pointer, you wouldn't need the above.
Just clear the array on init and pass in whatever value is in there.

> +		if (need_canfork_callback & (1 << i))
> +			state = &old_cfs->ss_state[i];
> +
> +		ss->fork(child, state);
> +	}
>  }

Thanks.

-- 
tejun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/