Re: [PATCH v2 1/2] sctp: rcu-ify addr_waitq

[Marcelo Ricardo Leitner <mleitner@redhat.com>]

On Thu, Jun 04, 2015 at 10:27:10AM -0400, Neil Horman wrote:
> On Wed, Jun 03, 2015 at 01:54:01PM -0300, mleitner@redhat.com wrote:
> > From: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
> > 
> > That's needed for the next patch, so we break the lock inversion between
> > netns_sctp->addr_wq_lock and socket lock on
> > sctp_addr_wq_timeout_handler(). With this, we can traverse addr_waitq
> > without taking addr_wq_lock, taking it just for the write operations.
> > 
> > Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
> > ---
> > 
> > Notes:
> >     v1->v2:
> >      As asked by Neil, this now reuses addr_wq_lock. And for that, also
> >      rcu-ifyies addr_waitq.
> > 
> >  include/net/netns/sctp.h |  2 +-
> >  net/sctp/protocol.c      | 81 +++++++++++++++++++++++++++++-------------------
> >  2 files changed, 50 insertions(+), 33 deletions(-)
> > 
> > diff --git a/include/net/netns/sctp.h b/include/net/netns/sctp.h
> > index 3573a81815ad9e0efb6ceb721eb066d3726419f0..9e53412c4ed829e8e45777a6d95406d490dbaa75 100644
> > --- a/include/net/netns/sctp.h
> > +++ b/include/net/netns/sctp.h
> > @@ -28,7 +28,7 @@ struct netns_sctp {
> >  	 * It is a list of sctp_sockaddr_entry.
> >  	 */
> >  	struct list_head local_addr_list;
> > -	struct list_head addr_waitq;
> > +	struct list_head __rcu addr_waitq;
> >  	struct timer_list addr_wq_timer;
> >  	struct list_head auto_asconf_splist;
> >  	spinlock_t addr_wq_lock;
> > diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c
> > index 53b7acde9aa37bf3d4029c459421564d5270f4c0..a5089883b28195f3aef69ef35b5397322a01126f 100644
> > --- a/net/sctp/protocol.c
> > +++ b/net/sctp/protocol.c
> > @@ -593,15 +593,46 @@ static void sctp_v4_ecn_capable(struct sock *sk)
> >  	INET_ECN_xmit(sk);
> >  }
> >  
> > +static void sctp_free_addr_wq(struct net *net)
> > +{
> > +	struct sctp_sockaddr_entry *addrw;
> > +
> > +	spin_lock_bh(&net->sctp.addr_wq_lock);
> > +	del_timer(&net->sctp.addr_wq_timer);
> > +	list_for_each_entry_rcu(addrw, &net->sctp.addr_waitq, list) {
> > +		list_del_rcu(&addrw->list);
> > +		kfree_rcu(addrw, rcu);
> > +	}
> > +	spin_unlock_bh(&net->sctp.addr_wq_lock);
> > +}
> > +
> > +/* As there is no refcnt on sctp_sockaddr_entry, we must check inside
> > + * the lock if it wasn't removed from addr_waitq already, otherwise we
> > + * could double-free it.
> > + */
> > +static void sctp_free_addr_wq_entry(struct net *net,
> > +				    struct sctp_sockaddr_entry *addrw)
> > +{
> > +	struct sctp_sockaddr_entry *temp;
> > +
> > +	spin_lock_bh(&net->sctp.addr_wq_lock);
> > +	list_for_each_entry_rcu(temp, &net->sctp.addr_waitq, list) {
> > +		if (temp == addrw) {
> > +			list_del_rcu(&addrw->list);
> > +			kfree_rcu(addrw, rcu);
> > +		}
> > +	}
> > +	spin_unlock_bh(&net->sctp.addr_wq_lock);
> > +}
> > +
> >  static void sctp_addr_wq_timeout_handler(unsigned long arg)
> >  {
> >  	struct net *net = (struct net *)arg;
> > -	struct sctp_sockaddr_entry *addrw, *temp;
> > +	struct sctp_sockaddr_entry *addrw;
> >  	struct sctp_sock *sp;
> >  
> > -	spin_lock_bh(&net->sctp.addr_wq_lock);
> > -
> > -	list_for_each_entry_safe(addrw, temp, &net->sctp.addr_waitq, list) {
> > +	rcu_read_lock_bh();
> > +	list_for_each_entry_rcu(addrw, &net->sctp.addr_waitq, list) {
> >  		pr_debug("%s: the first ent in wq:%p is addr:%pISc for cmd:%d at "
> >  			 "entry:%p\n", __func__, &net->sctp.addr_waitq, &addrw->a.sa,
> >  			 addrw->state, addrw);
> > @@ -627,7 +658,9 @@ static void sctp_addr_wq_timeout_handler(unsigned long arg)
> >  
> >  				timeo_val = jiffies;
> >  				timeo_val += msecs_to_jiffies(SCTP_ADDRESS_TICK_DELAY);
> > +				spin_lock_bh(&net->sctp.addr_wq_lock);
> >  				mod_timer(&net->sctp.addr_wq_timer, timeo_val);
> > +				spin_unlock_bh(&net->sctp.addr_wq_lock);
> Do we actually need to lock the addr_wq_lock here?  mod_timer has its own
> internal locking.

No, we don't. I'll remove these.

> >  				break;
> >  			}
> >  		}
> > @@ -647,35 +680,20 @@ static void sctp_addr_wq_timeout_handler(unsigned long arg)
> >  #if IS_ENABLED(CONFIG_IPV6)
> >  free_next:
> >  #endif
> > -		list_del(&addrw->list);
> > -		kfree(addrw);
> > -	}
> > -	spin_unlock_bh(&net->sctp.addr_wq_lock);
> > -}
> > -
> > -static void sctp_free_addr_wq(struct net *net)
> > -{
> > -	struct sctp_sockaddr_entry *addrw;
> > -	struct sctp_sockaddr_entry *temp;
> > -
> > -	spin_lock_bh(&net->sctp.addr_wq_lock);
> > -	del_timer(&net->sctp.addr_wq_timer);
> > -	list_for_each_entry_safe(addrw, temp, &net->sctp.addr_waitq, list) {
> > -		list_del(&addrw->list);
> > -		kfree(addrw);
> > +		sctp_free_addr_wq_entry(net, addrw);
> >  	}
> > -	spin_unlock_bh(&net->sctp.addr_wq_lock);
> > +	rcu_read_unlock_bh();
> >  }
> >  
> >  /* lookup the entry for the same address in the addr_waitq
> > - * sctp_addr_wq MUST be locked
> > + * rcu read MUST be locked
> >   */
> >  static struct sctp_sockaddr_entry *sctp_addr_wq_lookup(struct net *net,
> >  					struct sctp_sockaddr_entry *addr)
> >  {
> >  	struct sctp_sockaddr_entry *addrw;
> >  
> > -	list_for_each_entry(addrw, &net->sctp.addr_waitq, list) {
> > +	list_for_each_entry_rcu(addrw, &net->sctp.addr_waitq, list) {
> >  		if (addrw->a.sa.sa_family != addr->a.sa.sa_family)
> >  			continue;
> >  		if (addrw->a.sa.sa_family == AF_INET) {
> > @@ -702,7 +720,7 @@ void sctp_addr_wq_mgmt(struct net *net, struct sctp_sockaddr_entry *addr, int cm
> >  	 * new address after a couple of addition and deletion of that address
> >  	 */
> >  
> > -	spin_lock_bh(&net->sctp.addr_wq_lock);
> > +	rcu_read_lock_bh();
> >  	/* Offsets existing events in addr_wq */
> >  	addrw = sctp_addr_wq_lookup(net, addr);
> >  	if (addrw) {
> > @@ -710,22 +728,21 @@ void sctp_addr_wq_mgmt(struct net *net, struct sctp_sockaddr_entry *addr, int cm
> >  			pr_debug("%s: offsets existing entry for %d, addr:%pISc "
> >  				 "in wq:%p\n", __func__, addrw->state, &addrw->a.sa,
> >  				 &net->sctp.addr_waitq);
> > -
> > -			list_del(&addrw->list);
> > -			kfree(addrw);
> > +			sctp_free_addr_wq_entry(net, addrw);
> >  		}
> > -		spin_unlock_bh(&net->sctp.addr_wq_lock);
> > +		rcu_read_unlock_bh();
> >  		return;
> >  	}
> > +	rcu_read_unlock_bh();
> >  
> >  	/* OK, we have to add the new address to the wait queue */
> >  	addrw = kmemdup(addr, sizeof(struct sctp_sockaddr_entry), GFP_ATOMIC);
> > -	if (addrw == NULL) {
> > -		spin_unlock_bh(&net->sctp.addr_wq_lock);
> > +	if (!addrw)
> >  		return;
> > -	}
> >  	addrw->state = cmd;
> > -	list_add_tail(&addrw->list, &net->sctp.addr_waitq);
> > +
> > +	spin_lock_bh(&net->sctp.addr_wq_lock);
> > +	list_add_tail_rcu(&addrw->list, &net->sctp.addr_waitq);
> >  
> >  	pr_debug("%s: add new entry for cmd:%d, addr:%pISc in wq:%p\n",
> >  		 __func__, addrw->state, &addrw->a.sa, &net->sctp.addr_waitq);
> 
> Other than the comment above, and the break you need to insert, I think this
> looks good, thanks for taking the extra time on it!
> Best
> Neil

Cool, thx! v3 in a few.

  Marcelo