[IB] uverbs: ABI-breaking fixes for userspace verbs

Introduce new userspace verbs ABI version 3.  This eliminates some
unneeded commands, and adds support for user-created completion
channels.  This cleans up problems with file leaks on error paths, and
also makes sure that file descriptors are always installed into the
correct process.

Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/core/uverbs.h

@@ -53,14 +53,14 @@ struct ib_uverbs_device {
 	struct cdev				dev;
 	struct class_device			class_dev;
 	struct ib_device		       *ib_dev;
-	int					num_comp;
+	int					num_comp_vectors;
 };
 
 struct ib_uverbs_event_file {
 	struct kref				ref;
+	struct file			       *file;
 	struct ib_uverbs_file		       *uverbs_file;
 	spinlock_t				lock;
-	int					fd;
 	int					is_async;
 	wait_queue_head_t			poll_wait;
 	struct fasync_struct		       *async_queue;
@@ -73,8 +73,7 @@ struct ib_uverbs_file {
 	struct ib_uverbs_device		       *device;
 	struct ib_ucontext		       *ucontext;
 	struct ib_event_handler			event_handler;
-	struct ib_uverbs_event_file	        async_file;
-	struct ib_uverbs_event_file	        comp_file[1];
+	struct ib_uverbs_event_file	       *async_file;
 };
 
 struct ib_uverbs_event {
@@ -110,10 +109,17 @@ extern struct idr ib_uverbs_cq_idr;
 extern struct idr ib_uverbs_qp_idr;
 extern struct idr ib_uverbs_srq_idr;
 
+struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
+					int is_async, int *fd);
+void ib_uverbs_release_event_file(struct kref *ref);
+struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd);
+
 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context);
 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr);
 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr);
 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr);
+void ib_uverbs_event_handler(struct ib_event_handler *handler,
+			     struct ib_event *event);
 
 int ib_umem_get(struct ib_device *dev, struct ib_umem *mem,
 		void *addr, size_t size, int write);
@@ -125,16 +131,14 @@ void ib_umem_release_on_close(struct ib_device *dev, struct ib_umem *umem);
 				 const char __user *buf, int in_len,	\
 				 int out_len)
 
-IB_UVERBS_DECLARE_CMD(query_params);
 IB_UVERBS_DECLARE_CMD(get_context);
 IB_UVERBS_DECLARE_CMD(query_device);
 IB_UVERBS_DECLARE_CMD(query_port);
-IB_UVERBS_DECLARE_CMD(query_gid);
-IB_UVERBS_DECLARE_CMD(query_pkey);
 IB_UVERBS_DECLARE_CMD(alloc_pd);
 IB_UVERBS_DECLARE_CMD(dealloc_pd);
 IB_UVERBS_DECLARE_CMD(reg_mr);
 IB_UVERBS_DECLARE_CMD(dereg_mr);
+IB_UVERBS_DECLARE_CMD(create_comp_channel);
 IB_UVERBS_DECLARE_CMD(create_cq);
 IB_UVERBS_DECLARE_CMD(destroy_cq);
 IB_UVERBS_DECLARE_CMD(create_qp);

drivers/infiniband/core/uverbs_cmd.c

@@ -33,6 +33,8 @@
  * $Id: uverbs_cmd.c 2708 2005-06-24 17:27:21Z roland $
  */
 
+#include <linux/file.h>
+
 #include <asm/uaccess.h>
 
 #include "uverbs.h"
@@ -45,29 +47,6 @@
 		(udata)->outlen = (olen);				\
 	} while (0)
 
-ssize_t ib_uverbs_query_params(struct ib_uverbs_file *file,
-			       const char __user *buf,
-			       int in_len, int out_len)
-{
-	struct ib_uverbs_query_params      cmd;
-	struct ib_uverbs_query_params_resp resp;
-
-	if (out_len < sizeof resp)
-		return -ENOSPC;
-
-	if (copy_from_user(&cmd, buf, sizeof cmd))
-		return -EFAULT;
-
-	memset(&resp, 0, sizeof resp);
-
-	resp.num_cq_events = file->device->num_comp;
-
-	if (copy_to_user((void __user *) (unsigned long) cmd.response, &resp, sizeof resp))
-	    return -EFAULT;
-
-	return in_len;
-}
-
 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
 			      const char __user *buf,
 			      int in_len, int out_len)
@@ -77,7 +56,7 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
 	struct ib_udata                   udata;
 	struct ib_device                 *ibdev = file->device->ib_dev;
 	struct ib_ucontext		 *ucontext;
-	int i;
+	struct file			 *filp;
 	int ret;
 
 	if (out_len < sizeof resp)
@@ -110,26 +89,42 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
 	INIT_LIST_HEAD(&ucontext->srq_list);
 	INIT_LIST_HEAD(&ucontext->ah_list);
 
-	resp.async_fd = file->async_file.fd;
-	for (i = 0; i < file->device->num_comp; ++i)
-		if (copy_to_user((void __user *) (unsigned long) cmd.cq_fd_tab +
-				 i * sizeof (__u32),
-				 &file->comp_file[i].fd, sizeof (__u32))) {
-			ret = -EFAULT;
-			goto err_free;
-		}
+	resp.num_comp_vectors = file->device->num_comp_vectors;
+
+	filp = ib_uverbs_alloc_event_file(file, 1, &resp.async_fd);
+	if (IS_ERR(filp)) {
+		ret = PTR_ERR(filp);
+		goto err_free;
+	}
 
 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
 			 &resp, sizeof resp)) {
 		ret = -EFAULT;
-		goto err_free;
+		goto err_file;
 	}
 
-	file->ucontext = ucontext;
+	file->async_file = filp->private_data;
+
+	INIT_IB_EVENT_HANDLER(&file->event_handler, file->device->ib_dev,
+			      ib_uverbs_event_handler);
+	ret = ib_register_event_handler(&file->event_handler);
+	if (ret)
+		goto err_file;
+
+	kref_get(&file->async_file->ref);
+	kref_get(&file->ref);
+	file->ucontext   = ucontext;
+
+	fd_install(resp.async_fd, filp);
+
 	up(&file->mutex);
 
 	return in_len;
 
+err_file:
+	put_unused_fd(resp.async_fd);
+	fput(filp);
+
 err_free:
 	ibdev->dealloc_ucontext(ucontext);
 
@@ -255,62 +250,6 @@ ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
 	return in_len;
 }
 
-ssize_t ib_uverbs_query_gid(struct ib_uverbs_file *file,
-			    const char __user *buf,
-			    int in_len, int out_len)
-{
-	struct ib_uverbs_query_gid      cmd;
-	struct ib_uverbs_query_gid_resp resp;
-	int                             ret;
-
-	if (out_len < sizeof resp)
-		return -ENOSPC;
-
-	if (copy_from_user(&cmd, buf, sizeof cmd))
-		return -EFAULT;
-
-	memset(&resp, 0, sizeof resp);
-
-	ret = ib_query_gid(file->device->ib_dev, cmd.port_num, cmd.index,
-			   (union ib_gid *) resp.gid);
-	if (ret)
-		return ret;
-
-	if (copy_to_user((void __user *) (unsigned long) cmd.response,
-			 &resp, sizeof resp))
-		return -EFAULT;
-
-	return in_len;
-}
-
-ssize_t ib_uverbs_query_pkey(struct ib_uverbs_file *file,
-			     const char __user *buf,
-			     int in_len, int out_len)
-{
-	struct ib_uverbs_query_pkey      cmd;
-	struct ib_uverbs_query_pkey_resp resp;
-	int                              ret;
-
-	if (out_len < sizeof resp)
-		return -ENOSPC;
-
-	if (copy_from_user(&cmd, buf, sizeof cmd))
-		return -EFAULT;
-
-	memset(&resp, 0, sizeof resp);
-
-	ret = ib_query_pkey(file->device->ib_dev, cmd.port_num, cmd.index,
-			    &resp.pkey);
-	if (ret)
-		return ret;
-
-	if (copy_to_user((void __user *) (unsigned long) cmd.response,
-			 &resp, sizeof resp))
-		return -EFAULT;
-
-	return in_len;
-}
-
 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
 			   const char __user *buf,
 			   int in_len, int out_len)
@@ -595,6 +534,35 @@ out:
 	return ret ? ret : in_len;
 }
 
+ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
+				      const char __user *buf, int in_len,
+				      int out_len)
+{
+	struct ib_uverbs_create_comp_channel	   cmd;
+	struct ib_uverbs_create_comp_channel_resp  resp;
+	struct file				  *filp;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	filp = ib_uverbs_alloc_event_file(file, 0, &resp.fd);
+	if (IS_ERR(filp))
+		return PTR_ERR(filp);
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp)) {
+		put_unused_fd(resp.fd);
+		fput(filp);
+		return -EFAULT;
+	}
+
+	fd_install(resp.fd, filp);
+	return in_len;
+}
+
 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
 			    const char __user *buf, int in_len,
 			    int out_len)
@@ -603,6 +571,7 @@ ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
 	struct ib_uverbs_create_cq_resp resp;
 	struct ib_udata                 udata;
 	struct ib_ucq_object           *uobj;
+	struct ib_uverbs_event_file    *ev_file = NULL;
 	struct ib_cq                   *cq;
 	int                             ret;
 
@@ -616,9 +585,12 @@ ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
 		   (unsigned long) cmd.response + sizeof resp,
 		   in_len - sizeof cmd, out_len - sizeof resp);
 
-	if (cmd.event_handler >= file->device->num_comp)
+	if (cmd.comp_vector >= file->device->num_comp_vectors)
 		return -EINVAL;
 
+	if (cmd.comp_channel >= 0)
+		ev_file = ib_uverbs_lookup_comp_file(cmd.comp_channel);
+
 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
 	if (!uobj)
 		return -ENOMEM;
@@ -641,7 +613,7 @@ ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
 	cq->uobject       = &uobj->uobject;
 	cq->comp_handler  = ib_uverbs_comp_handler;
 	cq->event_handler = ib_uverbs_cq_event_handler;
-	cq->cq_context    = file;
+	cq->cq_context    = ev_file;
 	atomic_set(&cq->usecnt, 0);
 
 retry:
@@ -700,6 +672,7 @@ ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
 	struct ib_uverbs_destroy_cq_resp resp;
 	struct ib_cq               	*cq;
 	struct ib_ucq_object        	*uobj;
+	struct ib_uverbs_event_file	*ev_file;
 	struct ib_uverbs_event		*evt, *tmp;
 	u64				 user_handle;
 	int                        	 ret = -EINVAL;
@@ -716,7 +689,8 @@ ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
 		goto out;
 
 	user_handle = cq->uobject->user_handle;
-	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
+	uobj        = container_of(cq->uobject, struct ib_ucq_object, uobject);
+	ev_file     = cq->cq_context;
 
 	ret = ib_destroy_cq(cq);
 	if (ret)
@@ -728,19 +702,23 @@ ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
 	list_del(&uobj->uobject.list);
 	up(&file->mutex);
 
-	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);
+	if (ev_file) {
+		spin_lock_irq(&ev_file->lock);
+		list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
+			list_del(&evt->list);
+			kfree(evt);
+		}
+		spin_unlock_irq(&ev_file->lock);
+
+		kref_put(&ev_file->ref, ib_uverbs_release_event_file);
 	}
-	spin_unlock_irq(&file->comp_file[0].lock);
 
-	spin_lock_irq(&file->async_file.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);
+	spin_unlock_irq(&file->async_file->lock);
 
 	resp.comp_events_reported  = uobj->comp_events_reported;
 	resp.async_events_reported = uobj->async_events_reported;
@@ -1005,12 +983,12 @@ ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
 	list_del(&uobj->uobject.list);
 	up(&file->mutex);
 
-	spin_lock_irq(&file->async_file.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);
+	spin_unlock_irq(&file->async_file->lock);
 
 	resp.events_reported = uobj->events_reported;
 
@@ -1243,12 +1221,12 @@ ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
 	list_del(&uobj->uobject.list);
 	up(&file->mutex);
 
-	spin_lock_irq(&file->async_file.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);
+	spin_unlock_irq(&file->async_file->lock);
 
 	resp.events_reported = uobj->events_reported;
 

drivers/infiniband/core/uverbs_main.c

@@ -77,26 +77,24 @@ static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
 				     const char __user *buf, int in_len,
 				     int out_len) = {
-	[IB_USER_VERBS_CMD_QUERY_PARAMS]  = ib_uverbs_query_params,
-	[IB_USER_VERBS_CMD_GET_CONTEXT]   = ib_uverbs_get_context,
-	[IB_USER_VERBS_CMD_QUERY_DEVICE]  = ib_uverbs_query_device,
-	[IB_USER_VERBS_CMD_QUERY_PORT]    = ib_uverbs_query_port,
-	[IB_USER_VERBS_CMD_QUERY_GID]     = ib_uverbs_query_gid,
-	[IB_USER_VERBS_CMD_QUERY_PKEY]    = ib_uverbs_query_pkey,
-	[IB_USER_VERBS_CMD_ALLOC_PD]      = ib_uverbs_alloc_pd,
-	[IB_USER_VERBS_CMD_DEALLOC_PD]    = ib_uverbs_dealloc_pd,
-	[IB_USER_VERBS_CMD_REG_MR]        = ib_uverbs_reg_mr,
-	[IB_USER_VERBS_CMD_DEREG_MR]      = ib_uverbs_dereg_mr,
-	[IB_USER_VERBS_CMD_CREATE_CQ]     = ib_uverbs_create_cq,
-	[IB_USER_VERBS_CMD_DESTROY_CQ]    = ib_uverbs_destroy_cq,
-	[IB_USER_VERBS_CMD_CREATE_QP]     = ib_uverbs_create_qp,
-	[IB_USER_VERBS_CMD_MODIFY_QP]     = ib_uverbs_modify_qp,
-	[IB_USER_VERBS_CMD_DESTROY_QP]    = ib_uverbs_destroy_qp,
-	[IB_USER_VERBS_CMD_ATTACH_MCAST]  = ib_uverbs_attach_mcast,
-	[IB_USER_VERBS_CMD_DETACH_MCAST]  = ib_uverbs_detach_mcast,
-	[IB_USER_VERBS_CMD_CREATE_SRQ]    = ib_uverbs_create_srq,
-	[IB_USER_VERBS_CMD_MODIFY_SRQ]    = ib_uverbs_modify_srq,
-	[IB_USER_VERBS_CMD_DESTROY_SRQ]   = ib_uverbs_destroy_srq,
+	[IB_USER_VERBS_CMD_GET_CONTEXT]   	= ib_uverbs_get_context,
+	[IB_USER_VERBS_CMD_QUERY_DEVICE]  	= ib_uverbs_query_device,
+	[IB_USER_VERBS_CMD_QUERY_PORT]    	= ib_uverbs_query_port,
+	[IB_USER_VERBS_CMD_ALLOC_PD]      	= ib_uverbs_alloc_pd,
+	[IB_USER_VERBS_CMD_DEALLOC_PD]    	= ib_uverbs_dealloc_pd,
+	[IB_USER_VERBS_CMD_REG_MR]        	= ib_uverbs_reg_mr,
+	[IB_USER_VERBS_CMD_DEREG_MR]      	= ib_uverbs_dereg_mr,
+	[IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
+	[IB_USER_VERBS_CMD_CREATE_CQ]     	= ib_uverbs_create_cq,
+	[IB_USER_VERBS_CMD_DESTROY_CQ]    	= ib_uverbs_destroy_cq,
+	[IB_USER_VERBS_CMD_CREATE_QP]     	= ib_uverbs_create_qp,
+	[IB_USER_VERBS_CMD_MODIFY_QP]     	= ib_uverbs_modify_qp,
+	[IB_USER_VERBS_CMD_DESTROY_QP]    	= ib_uverbs_destroy_qp,
+	[IB_USER_VERBS_CMD_ATTACH_MCAST]  	= ib_uverbs_attach_mcast,
+	[IB_USER_VERBS_CMD_DETACH_MCAST]  	= ib_uverbs_detach_mcast,
+	[IB_USER_VERBS_CMD_CREATE_SRQ]    	= ib_uverbs_create_srq,
+	[IB_USER_VERBS_CMD_MODIFY_SRQ]    	= ib_uverbs_modify_srq,
+	[IB_USER_VERBS_CMD_DESTROY_SRQ]   	= ib_uverbs_destroy_srq,
 };
 
 static struct vfsmount *uverbs_event_mnt;
@@ -188,25 +186,19 @@ static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
 
 	spin_lock_irq(&file->lock);
 
-	while (list_empty(&file->event_list) && file->fd >= 0) {
+	while (list_empty(&file->event_list)) {
 		spin_unlock_irq(&file->lock);
 
 		if (filp->f_flags & O_NONBLOCK)
 			return -EAGAIN;
 
 		if (wait_event_interruptible(file->poll_wait,
-					     !list_empty(&file->event_list) ||
-					     file->fd < 0))
+					     !list_empty(&file->event_list)))
 			return -ERESTARTSYS;
 
 		spin_lock_irq(&file->lock);
 	}
 
-	if (file->fd < 0) {
-		spin_unlock_irq(&file->lock);
-		return -ENODEV;
-	}
-
 	event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
 
 	if (file->is_async)
@@ -248,26 +240,19 @@ static unsigned int ib_uverbs_event_poll(struct file *filp,
 	poll_wait(filp, &file->poll_wait, wait);
 
 	spin_lock_irq(&file->lock);
-	if (file->fd < 0)
-		pollflags = POLLERR;
-	else if (!list_empty(&file->event_list))
+	if (!list_empty(&file->event_list))
 		pollflags = POLLIN | POLLRDNORM;
 	spin_unlock_irq(&file->lock);
 
 	return pollflags;
 }
 
-static void ib_uverbs_event_release(struct ib_uverbs_event_file *file)
+void ib_uverbs_release_event_file(struct kref *ref)
 {
-	struct ib_uverbs_event *entry, *tmp;
+	struct ib_uverbs_event_file *file =
+		container_of(ref, struct ib_uverbs_event_file, ref);
 
-	spin_lock_irq(&file->lock);
-	if (file->fd != -1) {
-		file->fd = -1;
-		list_for_each_entry_safe(entry, tmp, &file->event_list, list)
-			kfree(entry);
-	}
-	spin_unlock_irq(&file->lock);
+	kfree(file);
 }
 
 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
@@ -280,21 +265,30 @@ static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
 {
 	struct ib_uverbs_event_file *file = filp->private_data;
+	struct ib_uverbs_event *entry, *tmp;
+
+	spin_lock_irq(&file->lock);
+	file->file = NULL;
+	list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
+		if (entry->counter)
+			list_del(&entry->obj_list);
+		kfree(entry);
+	}
+	spin_unlock_irq(&file->lock);
 
-	ib_uverbs_event_release(file);
 	ib_uverbs_event_fasync(-1, filp, 0);
-	kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
+
+	if (file->is_async) {
+		ib_unregister_event_handler(&file->uverbs_file->event_handler);
+		kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
+	}
+	kref_put(&file->ref, ib_uverbs_release_event_file);
 
 	return 0;
 }
 
 static struct file_operations uverbs_event_fops = {
-	/*
-	 * No .owner field since we artificially create event files,
-	 * so there is no increment to the module reference count in
-	 * the open path.  All event files come from a uverbs command
-	 * file, which already takes a module reference, so this is OK.
-	 */
+	.owner	 = THIS_MODULE,
 	.read 	 = ib_uverbs_event_read,
 	.poll    = ib_uverbs_event_poll,
 	.release = ib_uverbs_event_close,
@@ -303,10 +297,19 @@ static struct file_operations uverbs_event_fops = {
 
 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
 {
-	struct ib_uverbs_file  *file = cq_context;
-	struct ib_ucq_object   *uobj;
-	struct ib_uverbs_event *entry;
-	unsigned long           flags;
+	struct ib_uverbs_event_file    *file = cq_context;
+	struct ib_ucq_object	       *uobj;
+	struct ib_uverbs_event	       *entry;
+	unsigned long			flags;
+
+	if (!file)
+		return;
+
+	spin_lock_irqsave(&file->lock, flags);
+	if (!file->file) {
+		spin_unlock_irqrestore(&file->lock, flags);
+		return;
+	}
 
 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
 	if (!entry)
@@ -317,13 +320,12 @@ void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
 	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->list, &file->event_list);
 	list_add_tail(&entry->obj_list, &uobj->comp_list);
-	spin_unlock_irqrestore(&file->comp_file[0].lock, flags);
+	spin_unlock_irqrestore(&file->lock, flags);
 
-	wake_up_interruptible(&file->comp_file[0].poll_wait);
-	kill_fasync(&file->comp_file[0].async_queue, SIGIO, POLL_IN);
+	wake_up_interruptible(&file->poll_wait);
+	kill_fasync(&file->async_queue, SIGIO, POLL_IN);
 }
 
 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
@@ -334,6 +336,12 @@ static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
 	struct ib_uverbs_event *entry;
 	unsigned long flags;
 
+	spin_lock_irqsave(&file->async_file->lock, flags);
+	if (!file->async_file->file) {
+		spin_unlock_irqrestore(&file->async_file->lock, flags);
+		return;
+	}
+
 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
 	if (!entry)
 		return;
@@ -342,24 +350,24 @@ static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
 	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);
+	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);
+	spin_unlock_irqrestore(&file->async_file->lock, flags);
 
-	wake_up_interruptible(&file->async_file.poll_wait);
-	kill_fasync(&file->async_file.async_queue, SIGIO, POLL_IN);
+	wake_up_interruptible(&file->async_file->poll_wait);
+	kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
 }
 
 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
 {
+	struct ib_uverbs_event_file *ev_file = context_ptr;
 	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,
+	ib_uverbs_async_handler(ev_file->uverbs_file, uobj->uobject.user_handle,
 				event->event, &uobj->async_list,
 				&uobj->async_events_reported);
 				
@@ -389,8 +397,8 @@ void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
 				&uobj->events_reported);
 }
 
-static void ib_uverbs_event_handler(struct ib_event_handler *handler,
-				    struct ib_event *event)
+void ib_uverbs_event_handler(struct ib_event_handler *handler,
+			     struct ib_event *event)
 {
 	struct ib_uverbs_file *file =
 		container_of(handler, struct ib_uverbs_file, event_handler);
@@ -399,38 +407,90 @@ static void ib_uverbs_event_handler(struct ib_event_handler *handler,
 				NULL, NULL);
 }
 
-static int ib_uverbs_event_init(struct ib_uverbs_event_file *file,
-				struct ib_uverbs_file *uverbs_file)
+struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
+					int is_async, int *fd)
 {
+	struct ib_uverbs_event_file *ev_file;
 	struct file *filp;
+	int ret;
 
-	spin_lock_init(&file->lock);
-	INIT_LIST_HEAD(&file->event_list);
-	init_waitqueue_head(&file->poll_wait);
-	file->uverbs_file = uverbs_file;
-	file->async_queue = NULL;
-
-	file->fd = get_unused_fd();
-	if (file->fd < 0)
-		return file->fd;
+	ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
+	if (!ev_file)
+		return ERR_PTR(-ENOMEM);
+
+	kref_init(&ev_file->ref);
+	spin_lock_init(&ev_file->lock);
+	INIT_LIST_HEAD(&ev_file->event_list);
+	init_waitqueue_head(&ev_file->poll_wait);
+	ev_file->uverbs_file = uverbs_file;
+	ev_file->async_queue = NULL;
+	ev_file->is_async    = is_async;
+
+	*fd = get_unused_fd();
+	if (*fd < 0) {
+		ret = *fd;
+		goto err;
+	}
 
 	filp = get_empty_filp();
 	if (!filp) {
-		put_unused_fd(file->fd);
-		return -ENFILE;
+		ret = -ENFILE;
+		goto err_fd;
 	}
 
-	filp->f_op 	   = &uverbs_event_fops;
+	ev_file->file      = filp;
+
+	/*
+	 * fops_get() can't fail here, because we're coming from a
+	 * system call on a uverbs file, which will already have a
+	 * module reference.
+	 */
+	filp->f_op 	   = fops_get(&uverbs_event_fops);
 	filp->f_vfsmnt 	   = mntget(uverbs_event_mnt);
 	filp->f_dentry 	   = dget(uverbs_event_mnt->mnt_root);
 	filp->f_mapping    = filp->f_dentry->d_inode->i_mapping;
 	filp->f_flags      = O_RDONLY;
 	filp->f_mode       = FMODE_READ;
-	filp->private_data = file;
+	filp->private_data = ev_file;
 
-	fd_install(file->fd, filp);
+	return filp;
 
-	return 0;
+err_fd:
+	put_unused_fd(*fd);
+
+err:
+	kfree(ev_file);
+	return ERR_PTR(ret);
+}
+
+/*
+ * Look up a completion event file by FD.  If lookup is successful,
+ * takes a ref to the event file struct that it returns; if
+ * unsuccessful, returns NULL.
+ */
+struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
+{
+	struct ib_uverbs_event_file *ev_file = NULL;
+	struct file *filp;
+
+	filp = fget(fd);
+	if (!filp)
+		return NULL;
+
+	if (filp->f_op != &uverbs_event_fops)
+		goto out;
+
+	ev_file = filp->private_data;
+	if (ev_file->is_async) {
+		ev_file = NULL;
+		goto out;
+	}
+
+	kref_get(&ev_file->ref);
+
+out:
+	fput(filp);
+	return ev_file;
 }
 
 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
@@ -453,8 +513,7 @@ static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
 	    !uverbs_cmd_table[hdr.command])
 		return -EINVAL;
 
-	if (!file->ucontext                               &&
-	    hdr.command != IB_USER_VERBS_CMD_QUERY_PARAMS &&
+	if (!file->ucontext &&
 	    hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
 		return -EINVAL;
 
@@ -477,82 +536,33 @@ static int ib_uverbs_open(struct inode *inode, struct file *filp)
 	struct ib_uverbs_device *dev =
 		container_of(inode->i_cdev, struct ib_uverbs_device, dev);
 	struct ib_uverbs_file *file;
-	int i = 0;
-	int ret;
 
 	if (!try_module_get(dev->ib_dev->owner))
 		return -ENODEV;
 
-	file = kmalloc(sizeof *file +
-		       (dev->num_comp - 1) * sizeof (struct ib_uverbs_event_file),
-		       GFP_KERNEL);
+	file = kmalloc(sizeof *file, GFP_KERNEL);
 	if (!file) {
-		ret = -ENOMEM;
-		goto err;
+		module_put(dev->ib_dev->owner);
+		return -ENOMEM;
 	}
 
-	file->device = dev;
+	file->device   = dev;
+	file->ucontext = NULL;
 	kref_init(&file->ref);
 	init_MUTEX(&file->mutex);
 
-	file->ucontext = NULL;
-
-	kref_get(&file->ref);
-	ret = ib_uverbs_event_init(&file->async_file, file);
-	if (ret)
-		goto err_kref;
-
-	file->async_file.is_async = 1;
-
-	for (i = 0; i < dev->num_comp; ++i) {
-		kref_get(&file->ref);
-		ret = ib_uverbs_event_init(&file->comp_file[i], file);
-		if (ret)
-			goto err_async;
-		file->comp_file[i].is_async = 0;
-	}
-
-
 	filp->private_data = file;
 
-	INIT_IB_EVENT_HANDLER(&file->event_handler, dev->ib_dev,
-			      ib_uverbs_event_handler);
-	if (ib_register_event_handler(&file->event_handler))
-		goto err_async;
-
 	return 0;
-
-err_async:
-	while (i--)
-		ib_uverbs_event_release(&file->comp_file[i]);
-
-	ib_uverbs_event_release(&file->async_file);
-
-err_kref:
-	/*
-	 * One extra kref_put() because we took a reference before the
-	 * event file creation that failed and got us here.
-	 */
-	kref_put(&file->ref, ib_uverbs_release_file);
-	kref_put(&file->ref, ib_uverbs_release_file);
-
-err:
-	module_put(dev->ib_dev->owner);
-	return ret;
 }
 
 static int ib_uverbs_close(struct inode *inode, struct file *filp)
 {
 	struct ib_uverbs_file *file = filp->private_data;
-	int i;
 
-	ib_unregister_event_handler(&file->event_handler);
-	ib_uverbs_event_release(&file->async_file);
 	ib_dealloc_ucontext(file->ucontext);
 
-	for (i = 0; i < file->device->num_comp; ++i)
-		ib_uverbs_event_release(&file->comp_file[i]);
-
+	kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
 	kref_put(&file->ref, ib_uverbs_release_file);
 
 	return 0;
@@ -631,8 +641,8 @@ static void ib_uverbs_add_one(struct ib_device *device)
 	set_bit(uverbs_dev->devnum, dev_map);
 	spin_unlock(&map_lock);
 
-	uverbs_dev->ib_dev   = device;
-	uverbs_dev->num_comp = 1;
+	uverbs_dev->ib_dev           = device;
+	uverbs_dev->num_comp_vectors = 1;
 
 	if (device->mmap)
 		cdev_init(&uverbs_dev->dev, &uverbs_mmap_fops);

include/rdma/ib_user_verbs.h

@@ -42,15 +42,12 @@
  * Increment this value if any changes that break userspace ABI
  * compatibility are made.
  */
-#define IB_USER_VERBS_ABI_VERSION	2
+#define IB_USER_VERBS_ABI_VERSION	3
 
 enum {
-	IB_USER_VERBS_CMD_QUERY_PARAMS,
 	IB_USER_VERBS_CMD_GET_CONTEXT,
 	IB_USER_VERBS_CMD_QUERY_DEVICE,
 	IB_USER_VERBS_CMD_QUERY_PORT,
-	IB_USER_VERBS_CMD_QUERY_GID,
-	IB_USER_VERBS_CMD_QUERY_PKEY,
 	IB_USER_VERBS_CMD_ALLOC_PD,
 	IB_USER_VERBS_CMD_DEALLOC_PD,
 	IB_USER_VERBS_CMD_CREATE_AH,
@@ -65,6 +62,7 @@ enum {
 	IB_USER_VERBS_CMD_ALLOC_MW,
 	IB_USER_VERBS_CMD_BIND_MW,
 	IB_USER_VERBS_CMD_DEALLOC_MW,
+	IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL,
 	IB_USER_VERBS_CMD_CREATE_CQ,
 	IB_USER_VERBS_CMD_RESIZE_CQ,
 	IB_USER_VERBS_CMD_DESTROY_CQ,
@@ -118,27 +116,14 @@ struct ib_uverbs_cmd_hdr {
 	__u16 out_words;
 };
 
-/*
- * No driver_data for "query params" command, since this is intended
- * to be a core function with no possible device dependence.
- */
-struct ib_uverbs_query_params {
-	__u64 response;
-};
-
-struct ib_uverbs_query_params_resp {
-	__u32 num_cq_events;
-};
-
 struct ib_uverbs_get_context {
 	__u64 response;
-	__u64 cq_fd_tab;
 	__u64 driver_data[0];
 };
 
 struct ib_uverbs_get_context_resp {
 	__u32 async_fd;
-	__u32 reserved;
+	__u32 num_comp_vectors;
 };
 
 struct ib_uverbs_query_device {
@@ -220,31 +205,6 @@ struct ib_uverbs_query_port_resp {
 	__u8  reserved[3];
 };
 
-struct ib_uverbs_query_gid {
-	__u64 response;
-	__u8  port_num;
-	__u8  index;
-	__u8  reserved[6];
-	__u64 driver_data[0];
-};
-
-struct ib_uverbs_query_gid_resp {
-	__u8  gid[16];
-};
-
-struct ib_uverbs_query_pkey {
-	__u64 response;
-	__u8  port_num;
-	__u8  index;
-	__u8  reserved[6];
-	__u64 driver_data[0];
-};
-
-struct ib_uverbs_query_pkey_resp {
-	__u16 pkey;
-	__u16 reserved;
-};
-
 struct ib_uverbs_alloc_pd {
 	__u64 response;
 	__u64 driver_data[0];
@@ -278,11 +238,21 @@ struct ib_uverbs_dereg_mr {
 	__u32 mr_handle;
 };
 
+struct ib_uverbs_create_comp_channel {
+	__u64 response;
+};
+
+struct ib_uverbs_create_comp_channel_resp {
+	__u32 fd;
+};
+
 struct ib_uverbs_create_cq {
 	__u64 response;
 	__u64 user_handle;
 	__u32 cqe;
-	__u32 event_handler;
+	__u32 comp_vector;
+	__s32 comp_channel;
+	__u32 reserved;
 	__u64 driver_data[0];
 };