Date:   Sun, 07 Nov 2021 13:51:40 -0600
From:   ebiederm@...ssion.com (Eric W. Biederman)
To:     Manfred Spraul <manfred@...orfullife.com>
Cc:     Alexander Mikhalitsyn <alexander.mikhalitsyn@...tuozzo.com>,
        linux-kernel@...r.kernel.org,
        Andrew Morton <akpm@...ux-foundation.org>,
        Davidlohr Bueso <dave@...olabs.net>,
        Greg KH <gregkh@...uxfoundation.org>,
        Andrei Vagin <avagin@...il.com>,
        Pavel Tikhomirov <ptikhomirov@...tuozzo.com>,
        Vasily Averin <vvs@...tuozzo.com>,
        Alexander Mikhalitsyn <alexander@...alicyn.com>,
        stable@...r.kernel.org
Subject: Re: [RFC] shm: extend forced shm destroy to support objects from several IPC nses (simplified)

Manfred Spraul <manfred@...orfullife.com> writes:

> Hello together,
>
> On 11/5/21 22:34, Eric W. Biederman wrote:
>>   +static inline void shm_clist_del(struct shmid_kernel *shp)
>> +{
>> +	struct task_struct *creator;
>> +
>> +	rcu_read_lock();
>> +	creator = rcu_dereference(shp->shm_creator);
>> +	if (creator) {
>> +		task_lock(creator);
>> +		list_del(&shp->shm_clist);
>> +		task_unlock(creator);
>> +	}
>> +	rcu_read_unlock();
>> +}
>> +
>
> shm_clist_del() only synchronizes against exit_shm() when shm_creator
> is not NULL.
>
>
>> +		list_del(&shp->shm_clist);
>> +		rcu_assign_pointer(shp->shm_creator, NULL);
>> +
>
> We set shm_creator to NULL -> no more synchronization.
>
> Now IPC_RMID can run in parallel - regardless if we test for
> list_empty() or shm_creator.
>
>> +
>> +		/* Guarantee shp lives after task_lock is dropped */
>> +		ipc_getref(&shp->shm_perm);
>> +
>
> task_lock() doesn't help: As soon as shm_creator is set to NULL,
> IPC_RMID won't acquire task_lock() anymore.
>
> Thus shp can disappear before we arrive at this ipc_getref.
>
> [Yes, I think I have introduced this bug. ]
>
> Corrected version attached.
>
>
> I'll reboot and retest the patch, then I would send it to akpm as
> replacement for current patch in mmotm.
>
> --
>
>     Manfred
>

> @@ -382,48 +425,94 @@ void shm_destroy_orphaned(struct ipc_namespace *ns)
>  /* Locking assumes this will only be called with task == current */
>  void exit_shm(struct task_struct *task)
>  {
> -	struct ipc_namespace *ns = task->nsproxy->ipc_ns;
> -	struct shmid_kernel *shp, *n;
> +	for (;;) {
> +		struct shmid_kernel *shp;
> +		struct ipc_namespace *ns;
>  
> -	if (list_empty(&task->sysvshm.shm_clist))
> -		return;
> +		task_lock(task);
> +
> +		if (list_empty(&task->sysvshm.shm_clist)) {
> +			task_unlock(task);
> +			break;
> +		}
> +
> +		shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
> +				shm_clist);
>  
> -	/*
> -	 * If kernel.shm_rmid_forced is not set then only keep track of
> -	 * which shmids are orphaned, so that a later set of the sysctl
> -	 * can clean them up.
> -	 */
> -	if (!ns->shm_rmid_forced) {
> -		down_read(&shm_ids(ns).rwsem);
> -		list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
> -			shp->shm_creator = NULL;
>  		/*
> -		 * Only under read lock but we are only called on current
> -		 * so no entry on the list will be shared.
> +		 * 1) get a reference to shp.
> +		 *    This must be done first: Right now, task_lock() prevents
> +		 *    any concurrent IPC_RMID calls. After the list_del_init(),
> +		 *    IPC_RMID will not acquire task_lock(->shm_creator)
> +		 *    anymore.
>  		 */
> -		list_del(&task->sysvshm.shm_clist);
> -		up_read(&shm_ids(ns).rwsem);
> -		return;
> -	}
> +		WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
>  
> -	/*
> -	 * Destroy all already created segments, that were not yet mapped,
> -	 * and mark any mapped as orphan to cover the sysctl toggling.
> -	 * Destroy is skipped if shm_may_destroy() returns false.
> -	 */
> -	down_write(&shm_ids(ns).rwsem);
> -	list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
> -		shp->shm_creator = NULL;
> +		/* 2) unlink */
> +		list_del_init(&shp->shm_clist);
> +
> +		/*
> +		 * 3) Get pointer to the ipc namespace. It is worth to say
> +		 * that this pointer is guaranteed to be valid because
> +		 * shp lifetime is always shorter than namespace lifetime
> +		 * in which shp lives.
> +		 * We taken task_lock it means that shp won't be freed.
> +		 */
> +		ns = shp->ns;
>  
> -		if (shm_may_destroy(ns, shp)) {
> -			shm_lock_by_ptr(shp);
> -			shm_destroy(ns, shp);
> +		/*
> +		 * 4) If kernel.shm_rmid_forced is not set then only keep track of
> +		 * which shmids are orphaned, so that a later set of the sysctl
> +		 * can clean them up.
> +		 */
> +		if (!ns->shm_rmid_forced) {
> +			ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
> +			task_unlock(task);
> +			continue;
>  		}
> -	}
>  
> -	/* Remove the list head from any segments still attached. */
> -	list_del(&task->sysvshm.shm_clist);
> -	up_write(&shm_ids(ns).rwsem);
> +		/*
> +		 * 5) get a reference to the namespace.
> +		 *    The refcount could be already 0. If it is 0, then
> +		 *    the shm objects will be free by free_ipc_work().
> +		 */
> +		ns = get_ipc_ns_not_zero(ns);
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Isn't this increment also too late?  Doesn't this need to move up
by ipc_rcu_getref while shp is still on the list?

Assuming the code is running in parallel with shm_exit_ns after removal
from shm_clist shm_destroy can run to completion and shm_exit_ns can
run to completion and the ipc namespace can be freed.

Eric

Powered by blists - more mailing lists