OSDN Git Service

xen/pvcalls: implement recvmsg
authorStefano Stabellini <sstabellini@kernel.org>
Mon, 30 Oct 2017 22:41:00 +0000 (15:41 -0700)
committerBoris Ostrovsky <boris.ostrovsky@oracle.com>
Tue, 31 Oct 2017 13:05:53 +0000 (09:05 -0400)
Implement recvmsg by copying data from the "in" ring. If not enough data
is available and the recvmsg call is blocking, then wait on the
inflight_conn_req waitqueue. Take the active socket in_mutex so that
only one function can access the ring at any given time.

Signed-off-by: Stefano Stabellini <stefano@aporeto.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
CC: boris.ostrovsky@oracle.com
CC: jgross@suse.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
drivers/xen/pvcalls-front.c
drivers/xen/pvcalls-front.h

index 7672578..24ffdc9 100644 (file)
@@ -117,6 +117,20 @@ static bool pvcalls_front_write_todo(struct sock_mapping *map)
        return !!(size - pvcalls_queued(prod, cons, size));
 }
 
+static bool pvcalls_front_read_todo(struct sock_mapping *map)
+{
+       struct pvcalls_data_intf *intf = map->active.ring;
+       RING_IDX cons, prod;
+       int32_t error;
+
+       cons = intf->in_cons;
+       prod = intf->in_prod;
+       error = intf->in_error;
+       return (error != 0 ||
+               pvcalls_queued(prod, cons,
+                              XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
+}
+
 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
 {
        struct xenbus_device *dev = dev_id;
@@ -484,6 +498,103 @@ again:
        return tot_sent;
 }
 
+static int __read_ring(struct pvcalls_data_intf *intf,
+                      struct pvcalls_data *data,
+                      struct iov_iter *msg_iter,
+                      size_t len, int flags)
+{
+       RING_IDX cons, prod, size, masked_prod, masked_cons;
+       RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
+       int32_t error;
+
+       cons = intf->in_cons;
+       prod = intf->in_prod;
+       error = intf->in_error;
+       /* get pointers before reading from the ring */
+       virt_rmb();
+       if (error < 0)
+               return error;
+
+       size = pvcalls_queued(prod, cons, array_size);
+       masked_prod = pvcalls_mask(prod, array_size);
+       masked_cons = pvcalls_mask(cons, array_size);
+
+       if (size == 0)
+               return 0;
+
+       if (len > size)
+               len = size;
+
+       if (masked_prod > masked_cons) {
+               len = copy_to_iter(data->in + masked_cons, len, msg_iter);
+       } else {
+               if (len > (array_size - masked_cons)) {
+                       int ret = copy_to_iter(data->in + masked_cons,
+                                    array_size - masked_cons, msg_iter);
+                       if (ret != array_size - masked_cons) {
+                               len = ret;
+                               goto out;
+                       }
+                       len = ret + copy_to_iter(data->in, len - ret, msg_iter);
+               } else {
+                       len = copy_to_iter(data->in + masked_cons, len, msg_iter);
+               }
+       }
+out:
+       /* read data from the ring before increasing the index */
+       virt_mb();
+       if (!(flags & MSG_PEEK))
+               intf->in_cons += len;
+
+       return len;
+}
+
+int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
+                    int flags)
+{
+       struct pvcalls_bedata *bedata;
+       int ret;
+       struct sock_mapping *map;
+
+       if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
+               return -EOPNOTSUPP;
+
+       pvcalls_enter();
+       if (!pvcalls_front_dev) {
+               pvcalls_exit();
+               return -ENOTCONN;
+       }
+       bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
+
+       map = (struct sock_mapping *) sock->sk->sk_send_head;
+       if (!map) {
+               pvcalls_exit();
+               return -ENOTSOCK;
+       }
+
+       mutex_lock(&map->active.in_mutex);
+       if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
+               len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
+
+       while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
+               wait_event_interruptible(map->active.inflight_conn_req,
+                                        pvcalls_front_read_todo(map));
+       }
+       ret = __read_ring(map->active.ring, &map->active.data,
+                         &msg->msg_iter, len, flags);
+
+       if (ret > 0)
+               notify_remote_via_irq(map->active.irq);
+       if (ret == 0)
+               ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
+       if (ret == -ENOTCONN)
+               ret = 0;
+
+       mutex_unlock(&map->active.in_mutex);
+       pvcalls_exit();
+       return ret;
+}
+
 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 {
        struct pvcalls_bedata *bedata;
index d937c24..de24041 100644 (file)
@@ -16,5 +16,9 @@ int pvcalls_front_accept(struct socket *sock,
 int pvcalls_front_sendmsg(struct socket *sock,
                          struct msghdr *msg,
                          size_t len);
+int pvcalls_front_recvmsg(struct socket *sock,
+                         struct msghdr *msg,
+                         size_t len,
+                         int flags);
 
 #endif