tap: add support for persistance

Type: feature

Change-Id: I775f53531972447ebae0d69b9e2dfeee84d115e5
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>

9 files changed, 127 insertions, 45 deletions

diff --git a/src/vat/api_format.c b/src/vat/api_format.c
index e08e5469e61..ea24aef2652 100644
--- a/src/vat/api_format.c
+++ b/src/vat/api_format.c
@@ -7371,6 +7371,10 @@ api_tap_create_v2 (vat_main_t * vam)
 	tap_flags |= TAP_FLAG_GSO;
       else if (unformat (i, "csum-offload"))
 	tap_flags |= TAP_FLAG_CSUM_OFFLOAD;
+      else if (unformat (i, "persist"))
+	tap_flags |= TAP_FLAG_PERSIST;
+      else if (unformat (i, "attach"))
+	tap_flags |= TAP_FLAG_ATTACH;
       else
 	break;
     }
@@ -20626,7 +20630,7 @@ _(l2_flags,                                                             \
 _(bridge_flags,                                                         \
   "bd_id <bridge-domain-id> [learn] [forward] [uu-flood] [flood] [arp-term] [disable]\n") \
 _(tap_create_v2,                                                        \
-  "id <num> [hw-addr <mac-addr>] [host-if-name <name>] [host-ns <name>] [num-rx-queues <num>] [rx-ring-size <num>] [tx-ring-size <num>] [host-bridge <name>] [host-mac-addr <mac-addr>] [host-ip4-addr <ip4addr/mask>] [host-ip6-addr <ip6addr/mask>] [host-mtu-size <mtu>] [gso | no-gso | csum-offload]") \
+  "id <num> [hw-addr <mac-addr>] [host-if-name <name>] [host-ns <name>] [num-rx-queues <num>] [rx-ring-size <num>] [tx-ring-size <num>] [host-bridge <name>] [host-mac-addr <mac-addr>] [host-ip4-addr <ip4addr/mask>] [host-ip6-addr <ip6addr/mask>] [host-mtu-size <mtu>] [gso | no-gso | csum-offload] [persist] [attach]") \
 _(tap_delete_v2,                                                        \
   "<vpp-if-name> | sw_if_index <id>")                                   \
 _(sw_interface_tap_v2_dump, "")                                         \
diff --git a/src/vnet/devices/netlink.c b/src/vnet/devices/netlink.c
index f070602f1aa..e3e33a3b2d0 100644
--- a/src/vnet/devices/netlink.c
+++ b/src/vnet/devices/netlink.c
@@ -243,7 +243,7 @@ vnet_netlink_add_ip4_addr (int ifindex, void *addr, int pfx_len)
   ifa.ifa_index = ifindex;
 
   vnet_netlink_msg_init (&m, RTM_NEWADDR,
-			 NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL,
+			 NLM_F_REQUEST | NLM_F_REPLACE,
 			 &ifa, sizeof (struct ifaddrmsg));
 
   vnet_netlink_msg_add_rtattr (&m, IFA_LOCAL, addr, 4);
@@ -266,7 +266,7 @@ vnet_netlink_add_ip6_addr (int ifindex, void *addr, int pfx_len)
   ifa.ifa_index = ifindex;
 
   vnet_netlink_msg_init (&m, RTM_NEWADDR,
-			 NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL,
+			 NLM_F_REQUEST | NLM_F_REPLACE,
 			 &ifa, sizeof (struct ifaddrmsg));
 
   vnet_netlink_msg_add_rtattr (&m, IFA_LOCAL, addr, 16);
@@ -291,7 +291,7 @@ vnet_netlink_add_ip4_route (void *dst, u8 dst_len, void *gw)
   rtm.rtm_dst_len = dst_len;
 
   vnet_netlink_msg_init (&m, RTM_NEWROUTE,
-			 NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL,
+			 NLM_F_REQUEST | NLM_F_REPLACE,
 			 &rtm, sizeof (struct rtmsg));
 
   vnet_netlink_msg_add_rtattr (&m, RTA_GATEWAY, gw, 4);
@@ -316,7 +316,7 @@ vnet_netlink_add_ip6_route (void *dst, u8 dst_len, void *gw)
   rtm.rtm_dst_len = dst_len;
 
   vnet_netlink_msg_init (&m, RTM_NEWROUTE,
-			 NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL,
+			 NLM_F_REQUEST | NLM_F_REPLACE,
 			 &rtm, sizeof (struct rtmsg));
 
   vnet_netlink_msg_add_rtattr (&m, RTA_GATEWAY, gw, 16);
diff --git a/src/vnet/devices/tap/FEATURE.yaml b/src/vnet/devices/tap/FEATURE.yaml
index ce2c510cc9b..a143d333a48 100644
--- a/src/vnet/devices/tap/FEATURE.yaml
+++ b/src/vnet/devices/tap/FEATURE.yaml
@@ -3,6 +3,8 @@ name: Tap Device
 maintainer: damarion@cisco.com sluong@cisco.com sykazmi@cisco.com
 features:
   - Virtio
+  - persistence
+  - attach to existing tap at host
 description: "Create a tap v2 device interface, which connects to a
               tap interface on the host system."
 missing:
diff --git a/src/vnet/devices/tap/cli.c b/src/vnet/devices/tap/cli.c
index 8d59210c293..68b8cef2a21 100644
--- a/src/vnet/devices/tap/cli.c
+++ b/src/vnet/devices/tap/cli.c
@@ -92,6 +92,10 @@ tap_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
 	    args.tap_flags |= TAP_FLAG_GSO;
 	  else if (unformat (line_input, "csum-offload"))
 	    args.tap_flags |= TAP_FLAG_CSUM_OFFLOAD;
+	  else if (unformat (line_input, "persist"))
+	    args.tap_flags |= TAP_FLAG_PERSIST;
+	  else if (unformat (line_input, "attach"))
+	    args.tap_flags |= TAP_FLAG_ATTACH;
 	  else if (unformat (line_input, "hw-addr %U",
 			     unformat_ethernet_address, args.mac_addr.bytes))
 	    args.mac_addr_set = 1;
@@ -131,7 +135,8 @@ VLIB_CLI_COMMAND (tap_create_command, static) = {
     "[host-bridge <bridge-name>] [host-ip4-addr <ip4addr/mask>] "
     "[host-ip6-addr <ip6-addr>] [host-ip4-gw <ip4-addr>] "
     "[host-ip6-gw <ip6-addr>] [host-mac-addr <host-mac-address>] "
-    "[host-if-name <name>] [host-mtu-size <size>] [no-gso|gso|csum-offload]",
+    "[host-if-name <name>] [host-mtu-size <size>] [no-gso|gso|csum-offload] "
+    "[persist] [attach]",
   .function = tap_create_command_fn,
 };
 /* *INDENT-ON* */
diff --git a/src/vnet/devices/tap/tap.c b/src/vnet/devices/tap/tap.c
index 6f16a4cd71d..e17d01d88b1 100644
--- a/src/vnet/devices/tap/tap.c
+++ b/src/vnet/devices/tap/tap.c
@@ -91,6 +91,7 @@ tap_free (vlib_main_t * vm, virtio_if_t * vif)
 {
   virtio_main_t *mm = &virtio_main;
   tap_main_t *tm = &tap_main;
+  clib_error_t *err = 0;
   int i;
 
   /* *INDENT-OFF* */
@@ -102,6 +103,9 @@ tap_free (vlib_main_t * vm, virtio_if_t * vif)
     virtio_vring_free_tx (vm, vif, TX_QUEUE (i));
   /* *INDENT-ON* */
 
+  _IOCTL (vif->tap_fd, TUNSETPERSIST, (void *) (uintptr_t) 0);
+  tap_log_dbg (vif, "TUNSETPERSIST: unset");
+error:
   if (vif->tap_fd != -1)
     close (vif->tap_fd);
 
@@ -131,6 +135,7 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
   int i;
   int old_netns_fd = -1;
   struct ifreq ifr = {.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR };
+  struct ifreq get_ifr = {.ifr_flags = 0 };
   size_t hdrsz;
   struct vhost_memory *vhost_mem = 0;
   virtio_if_t *vif = 0;
@@ -170,6 +175,39 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
   vif->num_rxqs = args->num_rx_queues;
   num_q_pairs = clib_max (vif->num_rxqs, vif->num_txqs);
 
+  if (args->tap_flags & TAP_FLAG_ATTACH)
+    {
+      if (args->host_if_name != NULL)
+	{
+	  host_if_name = (char *) args->host_if_name;
+	  clib_memcpy (ifr.ifr_name, host_if_name,
+		       clib_min (IFNAMSIZ, strlen (host_if_name)));
+	}
+      else
+	{
+	  args->rv = VNET_API_ERROR_NO_MATCHING_INTERFACE;
+	  err = clib_error_return (0, "host_if_name is not provided");
+	  goto error;
+	}
+      if (args->host_namespace)
+	{
+	  old_netns_fd = open ("/proc/self/ns/net", O_RDONLY);
+	  if ((nfd = open_netns_fd ((char *) args->host_namespace)) == -1)
+	    {
+	      args->rv = VNET_API_ERROR_SYSCALL_ERROR_2;
+	      args->error = clib_error_return_unix (0, "open_netns_fd '%s'",
+						    args->host_namespace);
+	      goto error;
+	    }
+	  if (setns (nfd, CLONE_NEWNET) == -1)
+	    {
+	      args->rv = VNET_API_ERROR_SYSCALL_ERROR_3;
+	      args->error = clib_error_return_unix (0, "setns '%s'",
+						    args->host_namespace);
+	      goto error;
+	    }
+	}
+    }
   if ((vif->tap_fd = tfd = open ("/dev/net/tun", O_RDWR | O_NONBLOCK)) < 0)
     {
       args->rv = VNET_API_ERROR_SYSCALL_ERROR_2;
@@ -231,6 +269,33 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
       goto error;
     }
 
+  /*
+   * unset the persistence when attaching to existing
+   * interface
+   */
+  if (args->tap_flags & TAP_FLAG_ATTACH)
+    {
+      _IOCTL (tfd, TUNSETPERSIST, (void *) (uintptr_t) 0);
+      tap_log_dbg (vif, "TUNSETPERSIST: unset");
+    }
+
+  /* set the persistence */
+  if (args->tap_flags & TAP_FLAG_PERSIST)
+    {
+      _IOCTL (tfd, TUNSETPERSIST, (void *) (uintptr_t) 1);
+      tap_log_dbg (vif, "TUNSETPERSIST: set");
+
+      /* verify persistence is set, read the flags */
+      _IOCTL (tfd, TUNGETIFF, (void *) &get_ifr);
+      tap_log_dbg (vif, "TUNGETIFF: flags 0x%lx", get_ifr.ifr_flags);
+      if ((get_ifr.ifr_flags & IFF_PERSIST) == 0)
+	{
+	  args->rv = VNET_API_ERROR_SYSCALL_ERROR_2;
+	  args->error = clib_error_return (0, "persistence not supported");
+	  goto error;
+	}
+    }
+
   tap_log_dbg (vif, "TUNSETVNETHDRSZ: fd %d vnet_hdr_sz %u", tfd, hdrsz);
   _IOCTL (tfd, TUNSETVNETHDRSZ, &hdrsz);
 
@@ -291,46 +356,46 @@ tap_create_if (vlib_main_t * vm, tap_create_if_args_t * args)
 
   virtio_set_net_hdr_size (vif);
 
-  /* if namespace is specified, all further netlink messages should be executed
-     after we change our net namespace */
-  if (args->host_namespace)
+  if (!(args->tap_flags & TAP_FLAG_ATTACH))
     {
-      old_netns_fd = open ("/proc/self/ns/net", O_RDONLY);
-      if ((nfd = open_netns_fd ((char *) args->host_namespace)) == -1)
+      /* if namespace is specified, all further netlink messages should be executed
+         after we change our net namespace */
+      if (args->host_namespace)
 	{
-	  args->rv = VNET_API_ERROR_SYSCALL_ERROR_2;
-	  args->error = clib_error_return_unix (0, "open_netns_fd '%s'",
-						args->host_namespace);
-	  goto error;
-	}
-      args->error = vnet_netlink_set_link_netns (vif->ifindex, nfd,
-						 host_if_name);
-      if (args->error)
-	{
-	  args->rv = VNET_API_ERROR_NETLINK_ERROR;
-	  goto error;
-	}
-      if (setns (nfd, CLONE_NEWNET) == -1)
-	{
-	  args->rv = VNET_API_ERROR_SYSCALL_ERROR_3;
-	  args->error = clib_error_return_unix (0, "setns '%s'",
-						args->host_namespace);
-	  goto error;
-	}
-      if ((vif->ifindex = if_nametoindex (host_if_name)) == 0)
-	{
-	  args->rv = VNET_API_ERROR_SYSCALL_ERROR_3;
-	  args->error = clib_error_return_unix (0, "if_nametoindex '%s'",
-						host_if_name);
-	  goto error;
+	  old_netns_fd = open ("/proc/self/ns/net", O_RDONLY);
+	  if ((nfd = open_netns_fd ((char *) args->host_namespace)) == -1)
+	    {
+	      args->rv = VNET_API_ERROR_SYSCALL_ERROR_2;
+	      args->error = clib_error_return_unix (0, "open_netns_fd '%s'",
+						    args->host_namespace);
+	      goto error;
+	    }
+	  args->error = vnet_netlink_set_link_netns (vif->ifindex, nfd,
+						     host_if_name);
+	  if (args->error)
+	    {
+	      args->rv = VNET_API_ERROR_NETLINK_ERROR;
+	      goto error;
+	    }
+	  if (setns (nfd, CLONE_NEWNET) == -1)
+	    {
+	      args->rv = VNET_API_ERROR_SYSCALL_ERROR_3;
+	      args->error = clib_error_return_unix (0, "setns '%s'",
+						    args->host_namespace);
+	      goto error;
+	    }
+	  if ((vif->ifindex = if_nametoindex (host_if_name)) == 0)
+	    {
+	      args->rv = VNET_API_ERROR_SYSCALL_ERROR_3;
+	      args->error = clib_error_return_unix (0, "if_nametoindex '%s'",
+						    host_if_name);
+	      goto error;
+	    }
 	}
-    }
-  else
-    {
-      if (host_if_name)
+      else if (host_if_name)
 	{
-	  args->error = vnet_netlink_set_link_name (vif->ifindex,
-						    host_if_name);
+	  args->error =
+	    vnet_netlink_set_link_name (vif->ifindex, host_if_name);
 	  if (args->error)
 	    {
 	      args->rv = VNET_API_ERROR_NETLINK_ERROR;
diff --git a/src/vnet/devices/tap/tap.h b/src/vnet/devices/tap/tap.h
index fd4006b5d52..5d087ad9da5 100644
--- a/src/vnet/devices/tap/tap.h
+++ b/src/vnet/devices/tap/tap.h
@@ -33,6 +33,8 @@ typedef struct
   u32 tap_flags;
 #define TAP_FLAG_GSO (1 << 0)
 #define TAP_FLAG_CSUM_OFFLOAD (1 << 1)
+#define TAP_FLAG_PERSIST (1 << 2)
+#define TAP_FLAG_ATTACH (1 << 3)
   u8 *host_namespace;
   u8 *host_if_name;
   mac_address_t host_mac_addr;
diff --git a/src/vnet/devices/tap/tapv2.api b/src/vnet/devices/tap/tapv2.api
index 861fe170959..ccbbe158160 100644
--- a/src/vnet/devices/tap/tapv2.api
+++ b/src/vnet/devices/tap/tapv2.api
@@ -28,6 +28,8 @@ import "vnet/ip/ip_types.api";
 enum tap_flags {
         TAP_FLAG_GSO = 1,
         TAP_FLAG_CSUM_OFFLOAD = 2,
+	TAP_FLAG_PERSIST = 4,
+	TAP_FLAG_ATTACH = 8,
 };
 
 /** \brief Initialize a new tap interface with the given parameters
diff --git a/src/vnet/devices/tap/tapv2_api.c b/src/vnet/devices/tap/tapv2_api.c
index 3b66bf0d6ec..9e09459c9ab 100644
--- a/src/vnet/devices/tap/tapv2_api.c
+++ b/src/vnet/devices/tap/tapv2_api.c
@@ -164,8 +164,6 @@ vl_api_tap_delete_v2_t_handler (vl_api_tap_delete_v2_t * mp)
 
   rv = tap_delete_if (vm, sw_if_index);
 
-
-
   rmp = vl_msg_api_alloc (sizeof (*rmp));
   rmp->_vl_msg_id = ntohs (VL_API_TAP_DELETE_V2_REPLY);
   rmp->context = mp->context;
diff --git a/src/vpp/api/custom_dump.c b/src/vpp/api/custom_dump.c
index a9d85568b33..854af20afe9 100644
--- a/src/vpp/api/custom_dump.c
+++ b/src/vpp/api/custom_dump.c
@@ -582,7 +582,11 @@ static void *vl_api_tap_create_v2_t_print
   if ((mp->tap_flags) & 0x1)
     s = format (s, "gso-enabled ");
   if ((mp->tap_flags) & 0x2)
-    s = format (s, "csum-offload-enabled");
+    s = format (s, "csum-offload-enabled ");
+  if ((mp->tap_flags) & 0x4)
+    s = format (s, "persist ");
+  if ((mp->tap_flags) & 0x8)
+    s = format (s, "attach ");
   FINISH;
 }