Make sure that userspace does not retrieve stale asynchronous or completion events after destroying a CQ, QP or SRQ. We do this by sweeping the event lists before returning from a destroy calls, and then return the number of events already reported before the destroy call. This allows userspace wait until it has processed all events for an object returned from the kernel before it frees its context for the object.

The ABI of the destroy CQ, destroy QP and destroy SRQ commands has to
change to return the event count, so bump the ABI version from 1 to 2.
The userspace libibverbs library has already been updated to handle
both the old and new ABI versions.

Signed-off-by: Roland Dreier <[e-mail]>

drivers/infiniband/core/uverbs.h

···
 	struct ib_uverbs_event_file	        comp_file[1];
 };
 
-struct ib_uverbs_async_event {
-	struct ib_uverbs_async_event_desc	desc;
+struct ib_uverbs_event {
+	union {
+		struct ib_uverbs_async_event_desc	async;
+		struct ib_uverbs_comp_event_desc	comp;
+	}					desc;
 	struct list_head			list;
+	struct list_head			obj_list;
+	u32				       *counter;
 };
 
-struct ib_uverbs_comp_event {
-	struct ib_uverbs_comp_event_desc	desc;
-	struct list_head			list;
+struct ib_uevent_object {
+	struct ib_uobject	uobject;
+	struct list_head	event_list;
+	u32			events_reported;
 };
 
-struct ib_uobject_mr {
-	struct ib_uobject			uobj;
-	struct page			       *page_list;
-	struct scatterlist		       *sg_list;
+struct ib_ucq_object {
+	struct ib_uobject	uobject;
+	struct list_head	comp_list;
+	struct list_head	async_list;
+	u32			comp_events_reported;
+	u32			async_events_reported;
 };
 
 extern struct semaphore ib_uverbs_idr_mutex;

drivers/infiniband/core/uverbs_cmd.c

···
 	struct ib_uverbs_create_cq      cmd;
 	struct ib_uverbs_create_cq_resp resp;
 	struct ib_udata                 udata;
-	struct ib_uobject              *uobj;
+	struct ib_ucq_object           *uobj;
 	struct ib_cq                   *cq;
 	int                             ret;
 
···
 	if (!uobj)
 		return -ENOMEM;
 
-	uobj->user_handle = cmd.user_handle;
-	uobj->context     = file->ucontext;
+	uobj->uobject.user_handle   = cmd.user_handle;
+	uobj->uobject.context       = file->ucontext;
+	uobj->comp_events_reported  = 0;
+	uobj->async_events_reported = 0;
+	INIT_LIST_HEAD(&uobj->comp_list);
+	INIT_LIST_HEAD(&uobj->async_list);
 
 	cq = file->device->ib_dev->create_cq(file->device->ib_dev, cmd.cqe,
 					     file->ucontext, &udata);
···
 	}
 
 	cq->device        = file->device->ib_dev;
-	cq->uobject       = uobj;
+	cq->uobject       = &uobj->uobject;
 	cq->comp_handler  = ib_uverbs_comp_handler;
 	cq->event_handler = ib_uverbs_cq_event_handler;
 	cq->cq_context    = file;
···
 	}
 
 	down(&ib_uverbs_idr_mutex);
-	ret = idr_get_new(&ib_uverbs_cq_idr, cq, &uobj->id);
+	ret = idr_get_new(&ib_uverbs_cq_idr, cq, &uobj->uobject.id);
 	up(&ib_uverbs_idr_mutex);
 
 	if (ret == -EAGAIN)
···
 		goto err_cq;
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_add_tail(&uobj->list, &file->ucontext->cq_list);
+	list_add_tail(&uobj->uobject.list, &file->ucontext->cq_list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 	memset(&resp, 0, sizeof resp);
-	resp.cq_handle = uobj->id;
+	resp.cq_handle = uobj->uobject.id;
 	resp.cqe       = cq->cqe;
 
 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
···
 
 err_list:
  	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 	down(&ib_uverbs_idr_mutex);
-	idr_remove(&ib_uverbs_cq_idr, uobj->id);
+	idr_remove(&ib_uverbs_cq_idr, uobj->uobject.id);
 	up(&ib_uverbs_idr_mutex);
 
 err_cq:
···
 			     const char __user *buf, int in_len,
 			     int out_len)
 {
-	struct ib_uverbs_destroy_cq cmd;
-	struct ib_cq               *cq;
-	struct ib_uobject          *uobj;
-	int                         ret = -EINVAL;
+	struct ib_uverbs_destroy_cq      cmd;
+	struct ib_uverbs_destroy_cq_resp resp;
+	struct ib_cq               	*cq;
+	struct ib_ucq_object        	*uobj;
+	struct ib_uverbs_event		*evt, *tmp;
+	u64				 user_handle;
+	int                        	 ret = -EINVAL;
 
 	if (copy_from_user(&cmd, buf, sizeof cmd))
 		return -EFAULT;
+
+	memset(&resp, 0, sizeof resp);
 
 	down(&ib_uverbs_idr_mutex);
 
···
 	if (!cq || cq->uobject->context != file->ucontext)
 		goto out;
 
-	uobj = cq->uobject;
+	user_handle = cq->uobject->user_handle;
+	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
 
 	ret = ib_destroy_cq(cq);
 	if (ret)
···
 	idr_remove(&ib_uverbs_cq_idr, cmd.cq_handle);
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
+	spin_lock_irq(&file->comp_file[0].lock);
+	list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
+		list_del(&evt->list);
+		kfree(evt);
+	}
+	spin_unlock_irq(&file->comp_file[0].lock);
+
+	spin_lock_irq(&file->async_file.lock);
+	list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
+		list_del(&evt->list);
+		kfree(evt);
+	}
+	spin_unlock_irq(&file->async_file.lock);
+
+	resp.comp_events_reported  = uobj->comp_events_reported;
+	resp.async_events_reported = uobj->async_events_reported;
+
 	kfree(uobj);
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp))
+		ret = -EFAULT;
 
 out:
 	up(&ib_uverbs_idr_mutex);
···
 	struct ib_uverbs_create_qp      cmd;
 	struct ib_uverbs_create_qp_resp resp;
 	struct ib_udata                 udata;
-	struct ib_uobject              *uobj;
+	struct ib_uevent_object        *uobj;
 	struct ib_pd                   *pd;
 	struct ib_cq                   *scq, *rcq;
 	struct ib_srq                  *srq;
···
 	attr.cap.max_recv_sge    = cmd.max_recv_sge;
 	attr.cap.max_inline_data = cmd.max_inline_data;
 
-	uobj->user_handle = cmd.user_handle;
-	uobj->context     = file->ucontext;
+	uobj->uobject.user_handle = cmd.user_handle;
+	uobj->uobject.context     = file->ucontext;
+	uobj->events_reported     = 0;
+	INIT_LIST_HEAD(&uobj->event_list);
 
 	qp = pd->device->create_qp(pd, &attr, &udata);
 	if (IS_ERR(qp)) {
···
 	qp->send_cq    	  = attr.send_cq;
 	qp->recv_cq    	  = attr.recv_cq;
 	qp->srq	       	  = attr.srq;
-	qp->uobject       = uobj;
+	qp->uobject       = &uobj->uobject;
 	qp->event_handler = attr.event_handler;
 	qp->qp_context    = attr.qp_context;
 	qp->qp_type	  = attr.qp_type;
···
 		goto err_destroy;
 	}
 
-	ret = idr_get_new(&ib_uverbs_qp_idr, qp, &uobj->id);
+	ret = idr_get_new(&ib_uverbs_qp_idr, qp, &uobj->uobject.id);
 
 	if (ret == -EAGAIN)
 		goto retry;
 	if (ret)
 		goto err_destroy;
 
-	resp.qp_handle = uobj->id;
+	resp.qp_handle = uobj->uobject.id;
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_add_tail(&uobj->list, &file->ucontext->qp_list);
+	list_add_tail(&uobj->uobject.list, &file->ucontext->qp_list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
···
 
 err_list:
 	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 err_destroy:
···
 			     const char __user *buf, int in_len,
 			     int out_len)
 {
-	struct ib_uverbs_destroy_qp cmd;
-	struct ib_qp               *qp;
-	struct ib_uobject          *uobj;
-	int                         ret = -EINVAL;
+	struct ib_uverbs_destroy_qp      cmd;
+	struct ib_uverbs_destroy_qp_resp resp;
+	struct ib_qp               	*qp;
+	struct ib_uevent_object        	*uobj;
+	struct ib_uverbs_event		*evt, *tmp;
+	int                        	 ret = -EINVAL;
 
 	if (copy_from_user(&cmd, buf, sizeof cmd))
 		return -EFAULT;
+
+	memset(&resp, 0, sizeof resp);
 
 	down(&ib_uverbs_idr_mutex);
 
···
 	if (!qp || qp->uobject->context != file->ucontext)
 		goto out;
 
-	uobj = qp->uobject;
+	uobj = container_of(qp->uobject, struct ib_uevent_object, uobject);
 
 	ret = ib_destroy_qp(qp);
 	if (ret)
···
 	idr_remove(&ib_uverbs_qp_idr, cmd.qp_handle);
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
+	spin_lock_irq(&file->async_file.lock);
+	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
+		list_del(&evt->list);
+		kfree(evt);
+	}
+	spin_unlock_irq(&file->async_file.lock);
+
+	resp.events_reported = uobj->events_reported;
+
 	kfree(uobj);
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp))
+		ret = -EFAULT;
 
 out:
 	up(&ib_uverbs_idr_mutex);
···
 	struct ib_uverbs_create_srq      cmd;
 	struct ib_uverbs_create_srq_resp resp;
 	struct ib_udata                  udata;
-	struct ib_uobject               *uobj;
+	struct ib_uevent_object         *uobj;
 	struct ib_pd                    *pd;
 	struct ib_srq                   *srq;
 	struct ib_srq_init_attr          attr;
···
 	attr.attr.max_sge   = cmd.max_sge;
 	attr.attr.srq_limit = cmd.srq_limit;
 
-	uobj->user_handle = cmd.user_handle;
-	uobj->context     = file->ucontext;
+	uobj->uobject.user_handle = cmd.user_handle;
+	uobj->uobject.context     = file->ucontext;
+	uobj->events_reported     = 0;
+	INIT_LIST_HEAD(&uobj->event_list);
 
 	srq = pd->device->create_srq(pd, &attr, &udata);
 	if (IS_ERR(srq)) {
···
 
 	srq->device    	   = pd->device;
 	srq->pd        	   = pd;
-	srq->uobject       = uobj;
+	srq->uobject       = &uobj->uobject;
 	srq->event_handler = attr.event_handler;
 	srq->srq_context   = attr.srq_context;
 	atomic_inc(&pd->usecnt);
···
 		goto err_destroy;
 	}
 
-	ret = idr_get_new(&ib_uverbs_srq_idr, srq, &uobj->id);
+	ret = idr_get_new(&ib_uverbs_srq_idr, srq, &uobj->uobject.id);
 
 	if (ret == -EAGAIN)
 		goto retry;
 	if (ret)
 		goto err_destroy;
 
-	resp.srq_handle = uobj->id;
+	resp.srq_handle = uobj->uobject.id;
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_add_tail(&uobj->list, &file->ucontext->srq_list);
+	list_add_tail(&uobj->uobject.list, &file->ucontext->srq_list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
···
 
 err_list:
 	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
 err_destroy:
···
 			      const char __user *buf, int in_len,
 			      int out_len)
 {
-	struct ib_uverbs_destroy_srq cmd;
-	struct ib_srq               *srq;
-	struct ib_uobject           *uobj;
-	int                          ret = -EINVAL;
+	struct ib_uverbs_destroy_srq      cmd;
+	struct ib_uverbs_destroy_srq_resp resp;
+	struct ib_srq               	 *srq;
+	struct ib_uevent_object        	 *uobj;
+	struct ib_uverbs_event		 *evt, *tmp;
+	int                         	  ret = -EINVAL;
 
 	if (copy_from_user(&cmd, buf, sizeof cmd))
 		return -EFAULT;
 
 	down(&ib_uverbs_idr_mutex);
 
+	memset(&resp, 0, sizeof resp);
+
 	srq = idr_find(&ib_uverbs_srq_idr, cmd.srq_handle);
 	if (!srq || srq->uobject->context != file->ucontext)
 		goto out;
 
-	uobj = srq->uobject;
+	uobj = container_of(srq->uobject, struct ib_uevent_object, uobject);
 
 	ret = ib_destroy_srq(srq);
 	if (ret)
···
 	idr_remove(&ib_uverbs_srq_idr, cmd.srq_handle);
 
 	spin_lock_irq(&file->ucontext->lock);
-	list_del(&uobj->list);
+	list_del(&uobj->uobject.list);
 	spin_unlock_irq(&file->ucontext->lock);
 
+	spin_lock_irq(&file->async_file.lock);
+	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
+		list_del(&evt->list);
+		kfree(evt);
+	}
+	spin_unlock_irq(&file->async_file.lock);
+
+	resp.events_reported = uobj->events_reported;
+
 	kfree(uobj);
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp))
+		ret = -EFAULT;
 
 out:
 	up(&ib_uverbs_idr_mutex);

drivers/infiniband/core/uverbs_main.c

···
 		idr_remove(&ib_uverbs_qp_idr, uobj->id);
 		ib_destroy_qp(qp);
 		list_del(&uobj->list);
-		kfree(uobj);
+		kfree(container_of(uobj, struct ib_uevent_object, uobject));
 	}
 
 	list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
···
 		idr_remove(&ib_uverbs_cq_idr, uobj->id);
 		ib_destroy_cq(cq);
 		list_del(&uobj->list);
-		kfree(uobj);
+		kfree(container_of(uobj, struct ib_ucq_object, uobject));
 	}
 
 	list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
···
 		idr_remove(&ib_uverbs_srq_idr, uobj->id);
 		ib_destroy_srq(srq);
 		list_del(&uobj->list);
-		kfree(uobj);
+		kfree(container_of(uobj, struct ib_uevent_object, uobject));
 	}
 
 	/* XXX Free MWs */
···
 				    size_t count, loff_t *pos)
 {
 	struct ib_uverbs_event_file *file = filp->private_data;
-	void *event;
+	struct ib_uverbs_event *event;
 	int eventsz;
 	int ret = 0;
 
···
 		return -ENODEV;
 	}
 
-	if (file->is_async) {
-		event   = list_entry(file->event_list.next,
-				     struct ib_uverbs_async_event, list);
+	event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
+
+	if (file->is_async)
 		eventsz = sizeof (struct ib_uverbs_async_event_desc);
-	} else {
-		event   = list_entry(file->event_list.next,
-				     struct ib_uverbs_comp_event, list);
+	else
 		eventsz = sizeof (struct ib_uverbs_comp_event_desc);
-	}
 
 	if (eventsz > count) {
 		ret   = -EINVAL;
 		event = NULL;
-	} else
+	} else {
 		list_del(file->event_list.next);
+		if (event->counter) {
+			++(*event->counter);
+			list_del(&event->obj_list);
+		}
+	}
 
 	spin_unlock_irq(&file->lock);
 
···
 
 static void ib_uverbs_event_release(struct ib_uverbs_event_file *file)
 {
-	struct list_head *entry, *tmp;
+	struct ib_uverbs_event *entry, *tmp;
 
 	spin_lock_irq(&file->lock);
 	if (file->fd != -1) {
 		file->fd = -1;
-		list_for_each_safe(entry, tmp, &file->event_list)
-			if (file->is_async)
-				kfree(list_entry(entry, struct ib_uverbs_async_event, list));
-			else
-				kfree(list_entry(entry, struct ib_uverbs_comp_event, list));
+		list_for_each_entry_safe(entry, tmp, &file->event_list, list)
+			kfree(entry);
 	}
 	spin_unlock_irq(&file->lock);
 }
···
 
 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
 {
-	struct ib_uverbs_file       *file = cq_context;
-	struct ib_uverbs_comp_event *entry;
-	unsigned long                flags;
+	struct ib_uverbs_file  *file = cq_context;
+	struct ib_ucq_object   *uobj;
+	struct ib_uverbs_event *entry;
+	unsigned long           flags;
 
 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
 	if (!entry)
 		return;
 
-	entry->desc.cq_handle = cq->uobject->user_handle;
+	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
+
+	entry->desc.comp.cq_handle = cq->uobject->user_handle;
+	entry->counter		   = &uobj->comp_events_reported;
 
 	spin_lock_irqsave(&file->comp_file[0].lock, flags);
 	list_add_tail(&entry->list, &file->comp_file[0].event_list);
+	list_add_tail(&entry->obj_list, &uobj->comp_list);
 	spin_unlock_irqrestore(&file->comp_file[0].lock, flags);
 
 	wake_up_interruptible(&file->comp_file[0].poll_wait);
···
 }
 
 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
-				    __u64 element, __u64 event)
+				    __u64 element, __u64 event,
+				    struct list_head *obj_list,
+				    u32 *counter)
 {
-	struct ib_uverbs_async_event *entry;
+	struct ib_uverbs_event *entry;
 	unsigned long flags;
 
 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
 	if (!entry)
 		return;
 
-	entry->desc.element    = element;
-	entry->desc.event_type = event;
+	entry->desc.async.element    = element;
+	entry->desc.async.event_type = event;
+	entry->counter               = counter;
 
 	spin_lock_irqsave(&file->async_file.lock, flags);
 	list_add_tail(&entry->list, &file->async_file.event_list);
+	if (obj_list)
+		list_add_tail(&entry->obj_list, obj_list);
 	spin_unlock_irqrestore(&file->async_file.lock, flags);
 
 	wake_up_interruptible(&file->async_file.poll_wait);
···
 
 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
 {
-	ib_uverbs_async_handler(context_ptr,
-				event->element.cq->uobject->user_handle,
-				event->event);
+	struct ib_ucq_object *uobj;
+
+	uobj = container_of(event->element.cq->uobject,
+			    struct ib_ucq_object, uobject);
+
+	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
+				event->event, &uobj->async_list,
+				&uobj->async_events_reported);
+				
 }
 
 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
 {
-	ib_uverbs_async_handler(context_ptr,
-				event->element.qp->uobject->user_handle,
-				event->event);
+	struct ib_uevent_object *uobj;
+
+	uobj = container_of(event->element.qp->uobject,
+			    struct ib_uevent_object, uobject);
+
+	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
+				event->event, &uobj->event_list,
+				&uobj->events_reported);
 }
 
 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
 {
-	ib_uverbs_async_handler(context_ptr,
-				event->element.srq->uobject->user_handle,
-				event->event);
+	struct ib_uevent_object *uobj;
+
+	uobj = container_of(event->element.srq->uobject,
+			    struct ib_uevent_object, uobject);
+
+	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
+				event->event, &uobj->event_list,
+				&uobj->events_reported);
 }
 
 static void ib_uverbs_event_handler(struct ib_event_handler *handler,
···
 	struct ib_uverbs_file *file =
 		container_of(handler, struct ib_uverbs_file, event_handler);
 
-	ib_uverbs_async_handler(file, event->element.port_num, event->event);
+	ib_uverbs_async_handler(file, event->element.port_num, event->event,
+				NULL, NULL);
 }
 
 static int ib_uverbs_event_init(struct ib_uverbs_event_file *file,

include/rdma/ib_user_verbs.h

···
  * Increment this value if any changes that break userspace ABI
  * compatibility are made.
  */
-#define IB_USER_VERBS_ABI_VERSION	1
+#define IB_USER_VERBS_ABI_VERSION	2
 
 enum {
 	IB_USER_VERBS_CMD_QUERY_PARAMS,
···
 };
 
 struct ib_uverbs_destroy_cq {
+	__u64 response;
 	__u32 cq_handle;
+	__u32 reserved;
+};
+
+struct ib_uverbs_destroy_cq_resp {
+	__u32 comp_events_reported;
+	__u32 async_events_reported;
 };
 
 struct ib_uverbs_create_qp {
···
 };
 
 struct ib_uverbs_destroy_qp {
+	__u64 response;
 	__u32 qp_handle;
+	__u32 reserved;
+};
+
+struct ib_uverbs_destroy_qp_resp {
+	__u32 events_reported;
 };
 
 struct ib_uverbs_attach_mcast {
···
 };
 
 struct ib_uverbs_destroy_srq {
+	__u64 response;
 	__u32 srq_handle;
+	__u32 reserved;
+};
+
+struct ib_uverbs_destroy_srq_resp {
+	__u32 events_reported;
 };
 
 #endif /* IB_USER_VERBS_H */