[net,v2,1/3] s390/ism: Fix locking for forwarding of IRQs and events to clients

Message ID 20230707104359.3324039-2-schnelle@linux.ibm.com
State New
Headers
Series [net,v2,1/3] s390/ism: Fix locking for forwarding of IRQs and events to clients |

Commit Message

Niklas Schnelle July 7, 2023, 10:43 a.m. UTC
  The clients array references all registered clients and is protected by
the clients_lock. Besides its use as general list of clients the clients
array is accessed in ism_handle_irq() to forward ISM device events to
clients.

While the clients_lock is taken in the IRQ handler when calling
handle_event() it is however incorrectly not held during the
client->handle_irq() call and for the preceding clients[] access leaving
it unprotected against concurrent client (un-)registration.

Furthermore the accesses to ism->sba_client_arr[] in ism_register_dmb()
and ism_unregister_dmb() are not protected by any lock. This is
especially problematic as the client ID from the ism->sba_client_arr[]
is not checked against NO_CLIENT and neither is the client pointer
checked.

Instead of expanding the use of the clients_lock further add a separate
array in struct ism_dev which references clients subscribed to the
device's events and IRQs. This array is protected by ism->lock which is
already taken in ism_handle_irq() and can be taken outside the IRQ
handler when adding/removing subscribers or the accessing
ism->sba_client_arr[]. This also means that the clients_lock is no
longer taken in IRQ context.

Fixes: 89e7d2ba61b7 ("net/ism: Add new API for client registration")
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
---
 drivers/s390/net/ism_drv.c | 44 +++++++++++++++++++++++++++++++-------
 include/linux/ism.h        |  1 +
 2 files changed, 37 insertions(+), 8 deletions(-)
  

Comments

Niklas Schnelle July 7, 2023, 11:08 a.m. UTC | #1
On Fri, 2023-07-07 at 12:43 +0200, Niklas Schnelle wrote:
> The clients array references all registered clients and is protected by
> the clients_lock. Besides its use as general list of clients the clients
> array is accessed in ism_handle_irq() to forward ISM device events to
> clients.
> 
> While the clients_lock is taken in the IRQ handler when calling
> handle_event() it is however incorrectly not held during the
> client->handle_irq() call and for the preceding clients[] access leaving
> it unprotected against concurrent client (un-)registration.
> 
> Furthermore the accesses to ism->sba_client_arr[] in ism_register_dmb()
> and ism_unregister_dmb() are not protected by any lock. This is
> especially problematic as the client ID from the ism->sba_client_arr[]
> is not checked against NO_CLIENT and neither is the client pointer
> checked.
> 
> Instead of expanding the use of the clients_lock further add a separate
> array in struct ism_dev which references clients subscribed to the
> device's events and IRQs. This array is protected by ism->lock which is
> already taken in ism_handle_irq() and can be taken outside the IRQ
> handler when adding/removing subscribers or the accessing
> ism->sba_client_arr[]. This also means that the clients_lock is no
> longer taken in IRQ context.
> 
> Fixes: 89e7d2ba61b7 ("net/ism: Add new API for client registration")
> Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
> ---

Sorry for the mess. My get_maintainers.pl to-cmd setup in git send-
email stumbled over the cover letter and so the cover letter was not
sent to the right people/lists messing up the "in-reply-to" references.
So I had to resend it.

Thanks
Niklas
  
Simon Horman July 8, 2023, 1:36 p.m. UTC | #2
On Fri, Jul 07, 2023 at 12:43:57PM +0200, Niklas Schnelle wrote:
> The clients array references all registered clients and is protected by
> the clients_lock. Besides its use as general list of clients the clients
> array is accessed in ism_handle_irq() to forward ISM device events to
> clients.
> 
> While the clients_lock is taken in the IRQ handler when calling
> handle_event() it is however incorrectly not held during the
> client->handle_irq() call and for the preceding clients[] access leaving
> it unprotected against concurrent client (un-)registration.
> 
> Furthermore the accesses to ism->sba_client_arr[] in ism_register_dmb()
> and ism_unregister_dmb() are not protected by any lock. This is
> especially problematic as the client ID from the ism->sba_client_arr[]
> is not checked against NO_CLIENT and neither is the client pointer
> checked.
> 
> Instead of expanding the use of the clients_lock further add a separate
> array in struct ism_dev which references clients subscribed to the
> device's events and IRQs. This array is protected by ism->lock which is
> already taken in ism_handle_irq() and can be taken outside the IRQ
> handler when adding/removing subscribers or the accessing
> ism->sba_client_arr[]. This also means that the clients_lock is no
> longer taken in IRQ context.
> 
> Fixes: 89e7d2ba61b7 ("net/ism: Add new API for client registration")
> Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>

...

> @@ -71,6 +80,7 @@ int ism_register_client(struct ism_client *client)
>  		list_for_each_entry(ism, &ism_dev_list.list, list) {
>  			ism->priv[i] = NULL;
>  			client->add(ism);
> +			ism_setup_forwarding(client, ism);
>  		}
>  	}
>  	mutex_unlock(&ism_dev_list.mutex);

...

> @@ -92,6 +102,9 @@ int ism_unregister_client(struct ism_client *client)
>  		max_client--;
>  	spin_unlock_irqrestore(&clients_lock, flags);
>  	list_for_each_entry(ism, &ism_dev_list.list, list) {
> +		spin_lock_irqsave(&ism->lock, flags);

Hi Niklas,

The lock is taken here.

> +		/* Stop forwarding IRQs and events */
> +		ism->subs[client->id] = NULL;
>  		for (int i = 0; i < ISM_NR_DMBS; ++i) {
>  			if (ism->sba_client_arr[i] == client->id) {
>  				pr_err("%s: attempt to unregister client '%s'"
> @@ -101,6 +114,7 @@ int ism_unregister_client(struct ism_client *client)
>  				goto out;

But it does not appear to be released
(by the call to spin_unlock_irqrestore() below)
if goto out is called here.

>  			}
>  		}
> +		spin_unlock_irqrestore(&ism->lock, flags);
>  	}
>  out:
>  	mutex_unlock(&ism_dev_list.mutex);
  
Alexandra Winter July 10, 2023, 6:35 a.m. UTC | #3
On 08.07.23 15:36, Simon Horman wrote:
> On Fri, Jul 07, 2023 at 12:43:57PM +0200, Niklas Schnelle wrote:
[...]
>> @@ -92,6 +102,9 @@ int ism_unregister_client(struct ism_client *client)
>>  		max_client--;
>>  	spin_unlock_irqrestore(&clients_lock, flags);
>>  	list_for_each_entry(ism, &ism_dev_list.list, list) {
>> +		spin_lock_irqsave(&ism->lock, flags);
> 
> Hi Niklas,
> 
> The lock is taken here.
> 
>> +		/* Stop forwarding IRQs and events */
>> +		ism->subs[client->id] = NULL;
>>  		for (int i = 0; i < ISM_NR_DMBS; ++i) {
>>  			if (ism->sba_client_arr[i] == client->id) {
>>  				pr_err("%s: attempt to unregister client '%s'"
>> @@ -101,6 +114,7 @@ int ism_unregister_client(struct ism_client *client)
>>  				goto out;
> 
> But it does not appear to be released
> (by the call to spin_unlock_irqrestore() below)
> if goto out is called here.
> 
>>  			}
>>  		}
>> +		spin_unlock_irqrestore(&ism->lock, flags);
>>  	}
>>  out:
>>  	mutex_unlock(&ism_dev_list.mutex);
> 

Great catch, Simon.
@Niklas, the missing unlock accidentially got moved to "[PATCH net v2 3/3] s390/ism: Do not unregister clients with registered DMBs"
  
Niklas Schnelle July 10, 2023, 7:28 a.m. UTC | #4
On Sat, 2023-07-08 at 14:36 +0100, Simon Horman wrote:
> On Fri, Jul 07, 2023 at 12:43:57PM +0200, Niklas Schnelle wrote:
> > The clients array references all registered clients and is protected by
> > the clients_lock. Besides its use as general list of clients the clients
> > array is accessed in ism_handle_irq() to forward ISM device events to
> > clients.
> > 
> > While the clients_lock is taken in the IRQ handler when calling
> > handle_event() it is however incorrectly not held during the
> > client->handle_irq() call and for the preceding clients[] access leaving
> > it unprotected against concurrent client (un-)registration.
> > 
> > Furthermore the accesses to ism->sba_client_arr[] in ism_register_dmb()
> > and ism_unregister_dmb() are not protected by any lock. This is
> > especially problematic as the client ID from the ism->sba_client_arr[]
> > is not checked against NO_CLIENT and neither is the client pointer
> > checked.
> > 
> > Instead of expanding the use of the clients_lock further add a separate
> > array in struct ism_dev which references clients subscribed to the
> > device's events and IRQs. This array is protected by ism->lock which is
> > already taken in ism_handle_irq() and can be taken outside the IRQ
> > handler when adding/removing subscribers or the accessing
> > ism->sba_client_arr[]. This also means that the clients_lock is no
> > longer taken in IRQ context.
> > 
> > Fixes: 89e7d2ba61b7 ("net/ism: Add new API for client registration")
> > Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
> 
> ...
> 
> > @@ -71,6 +80,7 @@ int ism_register_client(struct ism_client *client)
> >  		list_for_each_entry(ism, &ism_dev_list.list, list) {
> >  			ism->priv[i] = NULL;
> >  			client->add(ism);
> > +			ism_setup_forwarding(client, ism);
> >  		}
> >  	}
> >  	mutex_unlock(&ism_dev_list.mutex);
> 
> ...
> 
> > @@ -92,6 +102,9 @@ int ism_unregister_client(struct ism_client *client)
> >  		max_client--;
> >  	spin_unlock_irqrestore(&clients_lock, flags);
> >  	list_for_each_entry(ism, &ism_dev_list.list, list) {
> > +		spin_lock_irqsave(&ism->lock, flags);
> 
> Hi Niklas,
> 
> The lock is taken here.
> 
> > +		/* Stop forwarding IRQs and events */
> > +		ism->subs[client->id] = NULL;
> >  		for (int i = 0; i < ISM_NR_DMBS; ++i) {
> >  			if (ism->sba_client_arr[i] == client->id) {
> >  				pr_err("%s: attempt to unregister client '%s'"
> > @@ -101,6 +114,7 @@ int ism_unregister_client(struct ism_client *client)
> >  				goto out;
> 
> But it does not appear to be released
> (by the call to spin_unlock_irqrestore() below)
> if goto out is called here.

Good catch. Yes I screwed this up while splitting the patch up. The
spin_unlock_irqrestore() is there after patch 3 but should have been
added in patch 1. As far as I can see all 3 patches have already been
applied, otherwise I'd send a v3. Thankfully even in the in between
state this error case can really onlt happen due to driver bugs so
maybe it's okay?

> 
> >  			}
> >  		}
> > +		spin_unlock_irqrestore(&ism->lock, flags);
> >  	}
> >  out:
> >  	mutex_unlock(&ism_dev_list.mutex);
  
Simon Horman July 10, 2023, 7:45 a.m. UTC | #5
On Mon, Jul 10, 2023 at 09:28:20AM +0200, Niklas Schnelle wrote:
> On Sat, 2023-07-08 at 14:36 +0100, Simon Horman wrote:
> > On Fri, Jul 07, 2023 at 12:43:57PM +0200, Niklas Schnelle wrote:
> > > The clients array references all registered clients and is protected by
> > > the clients_lock. Besides its use as general list of clients the clients
> > > array is accessed in ism_handle_irq() to forward ISM device events to
> > > clients.
> > > 
> > > While the clients_lock is taken in the IRQ handler when calling
> > > handle_event() it is however incorrectly not held during the
> > > client->handle_irq() call and for the preceding clients[] access leaving
> > > it unprotected against concurrent client (un-)registration.
> > > 
> > > Furthermore the accesses to ism->sba_client_arr[] in ism_register_dmb()
> > > and ism_unregister_dmb() are not protected by any lock. This is
> > > especially problematic as the client ID from the ism->sba_client_arr[]
> > > is not checked against NO_CLIENT and neither is the client pointer
> > > checked.
> > > 
> > > Instead of expanding the use of the clients_lock further add a separate
> > > array in struct ism_dev which references clients subscribed to the
> > > device's events and IRQs. This array is protected by ism->lock which is
> > > already taken in ism_handle_irq() and can be taken outside the IRQ
> > > handler when adding/removing subscribers or the accessing
> > > ism->sba_client_arr[]. This also means that the clients_lock is no
> > > longer taken in IRQ context.
> > > 
> > > Fixes: 89e7d2ba61b7 ("net/ism: Add new API for client registration")
> > > Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
> > 
> > ...
> > 
> > > @@ -71,6 +80,7 @@ int ism_register_client(struct ism_client *client)
> > >  		list_for_each_entry(ism, &ism_dev_list.list, list) {
> > >  			ism->priv[i] = NULL;
> > >  			client->add(ism);
> > > +			ism_setup_forwarding(client, ism);
> > >  		}
> > >  	}
> > >  	mutex_unlock(&ism_dev_list.mutex);
> > 
> > ...
> > 
> > > @@ -92,6 +102,9 @@ int ism_unregister_client(struct ism_client *client)
> > >  		max_client--;
> > >  	spin_unlock_irqrestore(&clients_lock, flags);
> > >  	list_for_each_entry(ism, &ism_dev_list.list, list) {
> > > +		spin_lock_irqsave(&ism->lock, flags);
> > 
> > Hi Niklas,
> > 
> > The lock is taken here.
> > 
> > > +		/* Stop forwarding IRQs and events */
> > > +		ism->subs[client->id] = NULL;
> > >  		for (int i = 0; i < ISM_NR_DMBS; ++i) {
> > >  			if (ism->sba_client_arr[i] == client->id) {
> > >  				pr_err("%s: attempt to unregister client '%s'"
> > > @@ -101,6 +114,7 @@ int ism_unregister_client(struct ism_client *client)
> > >  				goto out;
> > 
> > But it does not appear to be released
> > (by the call to spin_unlock_irqrestore() below)
> > if goto out is called here.
> 
> Good catch. Yes I screwed this up while splitting the patch up. The
> spin_unlock_irqrestore() is there after patch 3 but should have been
> added in patch 1. As far as I can see all 3 patches have already been
> applied, otherwise I'd send a v3. Thankfully even in the in between
> state this error case can really onlt happen due to driver bugs so
> maybe it's okay?

Hi Niklas,

I also saw the patches have been accepted after I sent my previous email.
So, given that the problem is resolved by another patch in the series,
I think the situation is as good as it is going to get.
  

Patch

diff --git a/drivers/s390/net/ism_drv.c b/drivers/s390/net/ism_drv.c
index 9b5fccdbc7d6..b664e4a08645 100644
--- a/drivers/s390/net/ism_drv.c
+++ b/drivers/s390/net/ism_drv.c
@@ -47,6 +47,15 @@  static struct ism_dev_list ism_dev_list = {
 	.mutex = __MUTEX_INITIALIZER(ism_dev_list.mutex),
 };
 
+static void ism_setup_forwarding(struct ism_client *client, struct ism_dev *ism)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ism->lock, flags);
+	ism->subs[client->id] = client;
+	spin_unlock_irqrestore(&ism->lock, flags);
+}
+
 int ism_register_client(struct ism_client *client)
 {
 	struct ism_dev *ism;
@@ -71,6 +80,7 @@  int ism_register_client(struct ism_client *client)
 		list_for_each_entry(ism, &ism_dev_list.list, list) {
 			ism->priv[i] = NULL;
 			client->add(ism);
+			ism_setup_forwarding(client, ism);
 		}
 	}
 	mutex_unlock(&ism_dev_list.mutex);
@@ -92,6 +102,9 @@  int ism_unregister_client(struct ism_client *client)
 		max_client--;
 	spin_unlock_irqrestore(&clients_lock, flags);
 	list_for_each_entry(ism, &ism_dev_list.list, list) {
+		spin_lock_irqsave(&ism->lock, flags);
+		/* Stop forwarding IRQs and events */
+		ism->subs[client->id] = NULL;
 		for (int i = 0; i < ISM_NR_DMBS; ++i) {
 			if (ism->sba_client_arr[i] == client->id) {
 				pr_err("%s: attempt to unregister client '%s'"
@@ -101,6 +114,7 @@  int ism_unregister_client(struct ism_client *client)
 				goto out;
 			}
 		}
+		spin_unlock_irqrestore(&ism->lock, flags);
 	}
 out:
 	mutex_unlock(&ism_dev_list.mutex);
@@ -328,6 +342,7 @@  int ism_register_dmb(struct ism_dev *ism, struct ism_dmb *dmb,
 		     struct ism_client *client)
 {
 	union ism_reg_dmb cmd;
+	unsigned long flags;
 	int ret;
 
 	ret = ism_alloc_dmb(ism, dmb);
@@ -351,7 +366,9 @@  int ism_register_dmb(struct ism_dev *ism, struct ism_dmb *dmb,
 		goto out;
 	}
 	dmb->dmb_tok = cmd.response.dmb_tok;
+	spin_lock_irqsave(&ism->lock, flags);
 	ism->sba_client_arr[dmb->sba_idx - ISM_DMB_BIT_OFFSET] = client->id;
+	spin_unlock_irqrestore(&ism->lock, flags);
 out:
 	return ret;
 }
@@ -360,6 +377,7 @@  EXPORT_SYMBOL_GPL(ism_register_dmb);
 int ism_unregister_dmb(struct ism_dev *ism, struct ism_dmb *dmb)
 {
 	union ism_unreg_dmb cmd;
+	unsigned long flags;
 	int ret;
 
 	memset(&cmd, 0, sizeof(cmd));
@@ -368,7 +386,9 @@  int ism_unregister_dmb(struct ism_dev *ism, struct ism_dmb *dmb)
 
 	cmd.request.dmb_tok = dmb->dmb_tok;
 
+	spin_lock_irqsave(&ism->lock, flags);
 	ism->sba_client_arr[dmb->sba_idx - ISM_DMB_BIT_OFFSET] = NO_CLIENT;
+	spin_unlock_irqrestore(&ism->lock, flags);
 
 	ret = ism_cmd(ism, &cmd);
 	if (ret && ret != ISM_ERROR)
@@ -491,6 +511,7 @@  static u16 ism_get_chid(struct ism_dev *ism)
 static void ism_handle_event(struct ism_dev *ism)
 {
 	struct ism_event *entry;
+	struct ism_client *clt;
 	int i;
 
 	while ((ism->ieq_idx + 1) != READ_ONCE(ism->ieq->header.idx)) {
@@ -499,21 +520,21 @@  static void ism_handle_event(struct ism_dev *ism)
 
 		entry = &ism->ieq->entry[ism->ieq_idx];
 		debug_event(ism_debug_info, 2, entry, sizeof(*entry));
-		spin_lock(&clients_lock);
-		for (i = 0; i < max_client; ++i)
-			if (clients[i])
-				clients[i]->handle_event(ism, entry);
-		spin_unlock(&clients_lock);
+		for (i = 0; i < max_client; ++i) {
+			clt = ism->subs[i];
+			if (clt)
+				clt->handle_event(ism, entry);
+		}
 	}
 }
 
 static irqreturn_t ism_handle_irq(int irq, void *data)
 {
 	struct ism_dev *ism = data;
-	struct ism_client *clt;
 	unsigned long bit, end;
 	unsigned long *bv;
 	u16 dmbemask;
+	u8 client_id;
 
 	bv = (void *) &ism->sba->dmb_bits[ISM_DMB_WORD_OFFSET];
 	end = sizeof(ism->sba->dmb_bits) * BITS_PER_BYTE - ISM_DMB_BIT_OFFSET;
@@ -530,8 +551,10 @@  static irqreturn_t ism_handle_irq(int irq, void *data)
 		dmbemask = ism->sba->dmbe_mask[bit + ISM_DMB_BIT_OFFSET];
 		ism->sba->dmbe_mask[bit + ISM_DMB_BIT_OFFSET] = 0;
 		barrier();
-		clt = clients[ism->sba_client_arr[bit]];
-		clt->handle_irq(ism, bit + ISM_DMB_BIT_OFFSET, dmbemask);
+		client_id = ism->sba_client_arr[bit];
+		if (unlikely(client_id == NO_CLIENT || !ism->subs[client_id]))
+			continue;
+		ism->subs[client_id]->handle_irq(ism, bit + ISM_DMB_BIT_OFFSET, dmbemask);
 	}
 
 	if (ism->sba->e) {
@@ -554,6 +577,7 @@  static void ism_dev_add_work_func(struct work_struct *work)
 						 add_work);
 
 	client->add(client->tgt_ism);
+	ism_setup_forwarding(client, client->tgt_ism);
 	atomic_dec(&client->tgt_ism->add_dev_cnt);
 	wake_up(&client->tgt_ism->waitq);
 }
@@ -691,7 +715,11 @@  static void ism_dev_remove_work_func(struct work_struct *work)
 {
 	struct ism_client *client = container_of(work, struct ism_client,
 						 remove_work);
+	unsigned long flags;
 
+	spin_lock_irqsave(&client->tgt_ism->lock, flags);
+	client->tgt_ism->subs[client->id] = NULL;
+	spin_unlock_irqrestore(&client->tgt_ism->lock, flags);
 	client->remove(client->tgt_ism);
 	atomic_dec(&client->tgt_ism->free_clients_cnt);
 	wake_up(&client->tgt_ism->waitq);
diff --git a/include/linux/ism.h b/include/linux/ism.h
index ea2bcdae7401..5160d47e5ea9 100644
--- a/include/linux/ism.h
+++ b/include/linux/ism.h
@@ -44,6 +44,7 @@  struct ism_dev {
 	u64 local_gid;
 	int ieq_idx;
 
+	struct ism_client *subs[MAX_CLIENTS];
 	atomic_t free_clients_cnt;
 	atomic_t add_dev_cnt;
 	wait_queue_head_t waitq;