net/mlx5: add E-switch VXLAN tunnel devices management
authorViacheslav Ovsiienko <viacheslavo@mellanox.com>
Sat, 3 Nov 2018 06:18:44 +0000 (06:18 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 5 Nov 2018 14:01:25 +0000 (15:01 +0100)
VXLAN interfaces are dynamically created for each local UDP port
of outer networks and then used as targets for TC "flower" filters
in order to perform encapsulation. These VXLAN interfaces are
system-wide, the only one device with given UDP port can exist
in the system (the attempt of creating another device with the
same UDP local port returns EEXIST), so PMD should support the
shared device instances database for PMD instances. These VXLAN
implicitly created devices are called VTEPs (Virtual Tunnel
End Points).

Creation of the VTEP occurs at the moment of rule applying. The
link is set up, root ingress qdisc is also initialized.

Encapsulation VTEPs are created on per port basis, the single
VTEP is attached to the outer interface and is shared for all
encapsulation rules on this interface. The source UDP port is
automatically selected in range 30000-60000.

For decapsulaton one VTEP is created per every unique UDP
local port to accept tunnel traffic. The name of created
VTEP consists of prefix "vmlx_" and the number of UDP port in
decimal digits without leading zeros (vmlx_4789). The VTEP
can be preliminary created in the system before the launching
application, it allows to share UDP ports between primary
and secondary processes.

Suggested-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
drivers/net/mlx5/mlx5_flow_tcf.c

index 37d9edf..252ddc0 100644 (file)
@@ -3797,6 +3797,417 @@ flow_tcf_nl_ack(struct mlx5_flow_tcf_context *tcf,
        return -err;
 }
 
+#define MNL_BUF_EXTRA_SPACE 16
+#define MNL_REQUEST_SIZE_MIN 256
+#define MNL_REQUEST_SIZE_MAX 2048
+#define MNL_REQUEST_SIZE RTE_MIN(RTE_MAX(sysconf(_SC_PAGESIZE), \
+                                MNL_REQUEST_SIZE_MIN), MNL_REQUEST_SIZE_MAX)
+
+/* VTEP device list is shared between PMD port instances. */
+static LIST_HEAD(, tcf_vtep) vtep_list_vxlan = LIST_HEAD_INITIALIZER();
+static pthread_mutex_t vtep_list_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/**
+ * Deletes VTEP network device.
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] vtep
+ *   Object represinting the network device to delete. Memory
+ *   allocated for this object is freed by routine.
+ */
+static void
+flow_tcf_vtep_delete(struct mlx5_flow_tcf_context *tcf,
+                    struct tcf_vtep *vtep)
+{
+       struct nlmsghdr *nlh;
+       struct ifinfomsg *ifm;
+       alignas(struct nlmsghdr)
+       uint8_t buf[mnl_nlmsg_size(MNL_ALIGN(sizeof(*ifm))) +
+                   MNL_BUF_EXTRA_SPACE];
+       int ret;
+
+       assert(!vtep->refcnt);
+       /* Delete only ifaces those we actually created. */
+       if (vtep->created && vtep->ifindex) {
+               DRV_LOG(INFO, "VTEP delete (%d)", vtep->ifindex);
+               nlh = mnl_nlmsg_put_header(buf);
+               nlh->nlmsg_type = RTM_DELLINK;
+               nlh->nlmsg_flags = NLM_F_REQUEST;
+               ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
+               ifm->ifi_family = AF_UNSPEC;
+               ifm->ifi_index = vtep->ifindex;
+               assert(sizeof(buf) >= nlh->nlmsg_len);
+               ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
+               if (ret)
+                       DRV_LOG(WARNING, "netlink: error deleting vxlan"
+                                        " encap/decap ifindex %u",
+                                        ifm->ifi_index);
+       }
+       rte_free(vtep);
+}
+
+/**
+ * Creates VTEP network device.
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] ifouter
+ *   Outer interface to attach new-created VXLAN device
+ *   If zero the VXLAN device will not be attached to any device.
+ *   These VTEPs are used for decapsulation and can be precreated
+ *   and shared between processes.
+ * @param[in] port
+ *   UDP port of created VTEP device.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ *
+ * @return
+ * Pointer to created device structure on success,
+ * NULL otherwise and rte_errno is set.
+ */
+#ifdef HAVE_IFLA_VXLAN_COLLECT_METADATA
+static struct tcf_vtep*
+flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf,
+                    unsigned int ifouter,
+                    uint16_t port, struct rte_flow_error *error)
+{
+       struct tcf_vtep *vtep;
+       struct nlmsghdr *nlh;
+       struct ifinfomsg *ifm;
+       char name[sizeof(MLX5_VXLAN_DEVICE_PFX) + 24];
+       alignas(struct nlmsghdr)
+       uint8_t buf[mnl_nlmsg_size(sizeof(*ifm)) +
+                   SZ_NLATTR_DATA_OF(sizeof(name)) +
+                   SZ_NLATTR_NEST * 2 +
+                   SZ_NLATTR_STRZ_OF("vxlan") +
+                   SZ_NLATTR_DATA_OF(sizeof(uint32_t)) +
+                   SZ_NLATTR_DATA_OF(sizeof(uint16_t)) +
+                   SZ_NLATTR_DATA_OF(sizeof(uint8_t)) * 3 +
+                   MNL_BUF_EXTRA_SPACE];
+       struct nlattr *na_info;
+       struct nlattr *na_vxlan;
+       rte_be16_t vxlan_port = rte_cpu_to_be_16(port);
+       int ret;
+
+       vtep = rte_zmalloc(__func__, sizeof(*vtep), alignof(struct tcf_vtep));
+       if (!vtep) {
+               rte_flow_error_set(error, ENOMEM,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                                  "unable to allocate memory for VTEP");
+               return NULL;
+       }
+       *vtep = (struct tcf_vtep){
+                       .port = port,
+                       .local = LIST_HEAD_INITIALIZER(),
+                       .neigh = LIST_HEAD_INITIALIZER(),
+       };
+       memset(buf, 0, sizeof(buf));
+       nlh = mnl_nlmsg_put_header(buf);
+       nlh->nlmsg_type = RTM_NEWLINK;
+       nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE  | NLM_F_EXCL;
+       ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
+       ifm->ifi_family = AF_UNSPEC;
+       ifm->ifi_type = 0;
+       ifm->ifi_index = 0;
+       ifm->ifi_flags = IFF_UP;
+       ifm->ifi_change = 0xffffffff;
+       snprintf(name, sizeof(name), "%s%u", MLX5_VXLAN_DEVICE_PFX, port);
+       mnl_attr_put_strz(nlh, IFLA_IFNAME, name);
+       na_info = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
+       assert(na_info);
+       mnl_attr_put_strz(nlh, IFLA_INFO_KIND, "vxlan");
+       na_vxlan = mnl_attr_nest_start(nlh, IFLA_INFO_DATA);
+       if (ifouter)
+               mnl_attr_put_u32(nlh, IFLA_VXLAN_LINK, ifouter);
+       assert(na_vxlan);
+       mnl_attr_put_u8(nlh, IFLA_VXLAN_COLLECT_METADATA, 1);
+       mnl_attr_put_u8(nlh, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 1);
+       mnl_attr_put_u8(nlh, IFLA_VXLAN_LEARNING, 0);
+       mnl_attr_put_u16(nlh, IFLA_VXLAN_PORT, vxlan_port);
+       mnl_attr_nest_end(nlh, na_vxlan);
+       mnl_attr_nest_end(nlh, na_info);
+       assert(sizeof(buf) >= nlh->nlmsg_len);
+       ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
+       if (ret) {
+               DRV_LOG(WARNING,
+                       "netlink: VTEP %s create failure (%d)",
+                       name, rte_errno);
+               if (rte_errno != EEXIST || ifouter)
+                       /*
+                        * Some unhandled error occurred or device is
+                        * for encapsulation and cannot be shared.
+                        */
+                       goto error;
+       } else {
+               /*
+                * Mark device we actually created.
+                * We should explicitly delete
+                * when we do not need it anymore.
+                */
+               vtep->created = 1;
+       }
+       /* Try to get ifindex of created of pre-existing device. */
+       ret = if_nametoindex(name);
+       if (!ret) {
+               DRV_LOG(WARNING,
+                       "VTEP %s failed to get index (%d)", name, errno);
+               rte_flow_error_set
+                       (error, -errno,
+                        RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                        "netlink: failed to retrieve VTEP ifindex");
+               goto error;
+       }
+       vtep->ifindex = ret;
+       vtep->ifouter = ifouter;
+       memset(buf, 0, sizeof(buf));
+       nlh = mnl_nlmsg_put_header(buf);
+       nlh->nlmsg_type = RTM_NEWLINK;
+       nlh->nlmsg_flags = NLM_F_REQUEST;
+       ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
+       ifm->ifi_family = AF_UNSPEC;
+       ifm->ifi_type = 0;
+       ifm->ifi_index = vtep->ifindex;
+       ifm->ifi_flags = IFF_UP;
+       ifm->ifi_change = IFF_UP;
+       ret = flow_tcf_nl_ack(tcf, nlh, 0, NULL, NULL);
+       if (ret) {
+               rte_flow_error_set(error, -errno,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                                  "netlink: failed to set VTEP link up");
+               DRV_LOG(WARNING, "netlink: VTEP %s set link up failure (%d)",
+                       name, rte_errno);
+               goto clean;
+       }
+       ret = mlx5_flow_tcf_init(tcf, vtep->ifindex, error);
+       if (ret) {
+               DRV_LOG(WARNING, "VTEP %s init failure (%d)", name, rte_errno);
+               goto clean;
+       }
+       DRV_LOG(INFO, "VTEP create (%d, %d)", vtep->port, vtep->ifindex);
+       vtep->refcnt = 1;
+       return vtep;
+clean:
+       flow_tcf_vtep_delete(tcf, vtep);
+       return NULL;
+error:
+       rte_free(vtep);
+       return NULL;
+}
+#else
+static struct tcf_vtep*
+flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf __rte_unused,
+                    unsigned int ifouter __rte_unused,
+                    uint16_t port __rte_unused,
+                    struct rte_flow_error *error)
+{
+       rte_flow_error_set(error, ENOTSUP,
+                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                          "netlink: failed to create VTEP, "
+                          "vxlan metadata are not supported by kernel");
+       return NULL;
+}
+#endif /* HAVE_IFLA_VXLAN_COLLECT_METADATA */
+
+/**
+ * Acquire target interface index for VXLAN tunneling decapsulation.
+ * In order to share the UDP port within the other interfaces the
+ * VXLAN device created as not attached to any interface (if created).
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] dev_flow
+ *   Flow tcf object with tunnel structure pointer set.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ * @return
+ *   Interface descriptor pointer on success,
+ *   NULL otherwise and rte_errno is set.
+ */
+static struct tcf_vtep*
+flow_tcf_decap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
+                           struct mlx5_flow *dev_flow,
+                           struct rte_flow_error *error)
+{
+       struct tcf_vtep *vtep;
+       uint16_t port = dev_flow->tcf.vxlan_decap->udp_port;
+
+       LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
+               if (vtep->port == port)
+                       break;
+       }
+       if (vtep && vtep->ifouter) {
+               rte_flow_error_set(error, -errno,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                                  "Failed to create decap VTEP with specified"
+                                  " UDP port, atatched device exists");
+               return NULL;
+       }
+       if (vtep) {
+               /* Device exists, just increment the reference counter. */
+               vtep->refcnt++;
+               assert(vtep->ifindex);
+               return vtep;
+       }
+       /* No decapsulation device exists, try to create the new one. */
+       vtep = flow_tcf_vtep_create(tcf, 0, port, error);
+       if (vtep)
+               LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
+       return vtep;
+}
+
+/**
+ * Aqcuire target interface index for VXLAN tunneling encapsulation.
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] ifouter
+ *   Network interface index to attach VXLAN encap device to.
+ * @param[in] dev_flow
+ *   Flow tcf object with tunnel structure pointer set.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ * @return
+ *   Interface descriptor pointer on success,
+ *   NULL otherwise and rte_errno is set.
+ */
+static struct tcf_vtep*
+flow_tcf_encap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
+                           unsigned int ifouter,
+                           struct mlx5_flow *dev_flow __rte_unused,
+                           struct rte_flow_error *error)
+{
+       static uint16_t encap_port = MLX5_VXLAN_PORT_MIN - 1;
+       struct tcf_vtep *vtep;
+
+       assert(ifouter);
+       /* Look whether the attached VTEP for encap is created. */
+       LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
+               if (vtep->ifouter == ifouter)
+                       break;
+       }
+       if (vtep) {
+               /* VTEP already exists, just increment the reference. */
+               vtep->refcnt++;
+       } else {
+               uint16_t pcnt;
+
+               /* Not found, we should create the new attached VTEP. */
+               for (pcnt = 0; pcnt <= (MLX5_VXLAN_PORT_MAX
+                                    - MLX5_VXLAN_PORT_MIN); pcnt++) {
+                       encap_port++;
+                       /* Wraparound the UDP port index. */
+                       if (encap_port < MLX5_VXLAN_PORT_MIN ||
+                           encap_port > MLX5_VXLAN_PORT_MAX)
+                               encap_port = MLX5_VXLAN_PORT_MIN;
+                       /* Check whether UDP port is in already in use. */
+                       LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
+                               if (vtep->port == encap_port)
+                                       break;
+                       }
+                       if (vtep) {
+                               /* Port is in use, try the next one. */
+                               vtep = NULL;
+                               continue;
+                       }
+                       vtep = flow_tcf_vtep_create(tcf, ifouter,
+                                                   encap_port, error);
+                       if (vtep) {
+                               LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
+                               break;
+                       }
+                       if (rte_errno != EEXIST)
+                               break;
+               }
+               if (!vtep)
+                       return NULL;
+       }
+       assert(vtep->ifouter == ifouter);
+       assert(vtep->ifindex);
+       return vtep;
+}
+
+/**
+ * Acquires target interface index for tunneling of any type.
+ * Creates the new VTEP if needed.
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] ifouter
+ *   Network interface index to attach VXLAN encap device to.
+ * @param[in] dev_flow
+ *   Flow tcf object with tunnel structure pointer set.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ * @return
+ *   Interface descriptor pointer on success,
+ *   NULL otherwise and rte_errno is set.
+ */
+static struct tcf_vtep*
+flow_tcf_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
+                     unsigned int ifouter,
+                     struct mlx5_flow *dev_flow,
+                     struct rte_flow_error *error)
+{
+       struct tcf_vtep *vtep = NULL;
+
+       assert(dev_flow->tcf.tunnel);
+       pthread_mutex_lock(&vtep_list_mutex);
+       switch (dev_flow->tcf.tunnel->type) {
+       case FLOW_TCF_TUNACT_VXLAN_ENCAP:
+               vtep = flow_tcf_encap_vtep_acquire(tcf, ifouter,
+                                                 dev_flow, error);
+               break;
+       case FLOW_TCF_TUNACT_VXLAN_DECAP:
+               vtep = flow_tcf_decap_vtep_acquire(tcf, dev_flow, error);
+               break;
+       default:
+               rte_flow_error_set(error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                                  "unsupported tunnel type");
+               break;
+       }
+       pthread_mutex_unlock(&vtep_list_mutex);
+       return vtep;
+}
+
+/**
+ * Release tunneling interface by ifindex. Decrements reference
+ * counter and actually removes the device if counter is zero.
+ *
+ * @param[in] tcf
+ *   Context object initialized by mlx5_flow_tcf_context_create().
+ * @param[in] vtep
+ *   VTEP device descriptor structure.
+ * @param[in] dev_flow
+ *   Flow tcf object with tunnel structure pointer set.
+ */
+static void
+flow_tcf_vtep_release(struct mlx5_flow_tcf_context *tcf,
+                     struct tcf_vtep *vtep,
+                     struct mlx5_flow *dev_flow)
+{
+       assert(dev_flow->tcf.tunnel);
+       pthread_mutex_lock(&vtep_list_mutex);
+       switch (dev_flow->tcf.tunnel->type) {
+       case FLOW_TCF_TUNACT_VXLAN_DECAP:
+               break;
+       case FLOW_TCF_TUNACT_VXLAN_ENCAP:
+               break;
+       default:
+               assert(false);
+               DRV_LOG(WARNING, "Unsupported tunnel type");
+               break;
+       }
+       assert(vtep->refcnt);
+       if (--vtep->refcnt == 0) {
+               LIST_REMOVE(vtep, next);
+               flow_tcf_vtep_delete(tcf, vtep);
+       }
+       pthread_mutex_unlock(&vtep_list_mutex);
+}
+
+
 /**
  * Apply flow to E-Switch by sending Netlink message.
  *
@@ -3822,11 +4233,35 @@ flow_tcf_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
        dev_flow = LIST_FIRST(&flow->dev_flows);
        /* E-Switch flow can't be expanded. */
        assert(!LIST_NEXT(dev_flow, next));
+       if (dev_flow->tcf.applied)
+               return 0;
        nlh = dev_flow->tcf.nlh;
        nlh->nlmsg_type = RTM_NEWTFILTER;
        nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
-       if (!flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL))
+       if (dev_flow->tcf.tunnel) {
+               /*
+                * Replace the interface index, target for
+                * encapsulation, source for decapsulation.
+                */
+               assert(!dev_flow->tcf.tunnel->vtep);
+               assert(dev_flow->tcf.tunnel->ifindex_ptr);
+               /* Acquire actual VTEP device when rule is being applied. */
+               dev_flow->tcf.tunnel->vtep =
+                       flow_tcf_vtep_acquire(ctx,
+                                       dev_flow->tcf.tunnel->ifindex_org,
+                                       dev_flow, error);
+               if (!dev_flow->tcf.tunnel->vtep)
+                       return -rte_errno;
+               DRV_LOG(INFO, "Replace ifindex: %d->%d",
+                               dev_flow->tcf.tunnel->vtep->ifindex,
+                               dev_flow->tcf.tunnel->ifindex_org);
+               *dev_flow->tcf.tunnel->ifindex_ptr =
+                       dev_flow->tcf.tunnel->vtep->ifindex;
+       }
+       if (!flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL)) {
+               dev_flow->tcf.applied = 1;
                return 0;
+       }
        return rte_flow_error_set(error, rte_errno,
                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
                                  "netlink: failed to create TC flow rule");
@@ -3855,10 +4290,20 @@ flow_tcf_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
                return;
        /* E-Switch flow can't be expanded. */
        assert(!LIST_NEXT(dev_flow, next));
-       nlh = dev_flow->tcf.nlh;
-       nlh->nlmsg_type = RTM_DELTFILTER;
-       nlh->nlmsg_flags = NLM_F_REQUEST;
-       flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL);
+       if (dev_flow->tcf.applied) {
+               nlh = dev_flow->tcf.nlh;
+               nlh->nlmsg_type = RTM_DELTFILTER;
+               nlh->nlmsg_flags = NLM_F_REQUEST;
+               flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL);
+               if (dev_flow->tcf.tunnel) {
+                       assert(dev_flow->tcf.tunnel->vtep);
+                       flow_tcf_vtep_release(ctx,
+                               dev_flow->tcf.tunnel->vtep,
+                               dev_flow);
+                       dev_flow->tcf.tunnel->vtep = NULL;
+               }
+               dev_flow->tcf.applied = 0;
+       }
 }
 
 /**
@@ -4385,7 +4830,9 @@ mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
        struct nlmsghdr *nlh;
        struct tcmsg *tcm;
        alignas(struct nlmsghdr)
-       uint8_t buf[mnl_nlmsg_size(sizeof(*tcm) + 128)];
+       uint8_t buf[mnl_nlmsg_size(sizeof(*tcm)) +
+                   SZ_NLATTR_STRZ_OF("ingress") +
+                   MNL_BUF_EXTRA_SPACE];
 
        /* Destroy existing ingress qdisc and everything attached to it. */
        nlh = mnl_nlmsg_put_header(buf);
@@ -4396,6 +4843,7 @@ mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
        tcm->tcm_ifindex = ifindex;
        tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
        tcm->tcm_parent = TC_H_INGRESS;
+       assert(sizeof(buf) >= nlh->nlmsg_len);
        /* Ignore errors when qdisc is already absent. */
        if (flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL) &&
            rte_errno != EINVAL && rte_errno != ENOENT)
@@ -4413,6 +4861,7 @@ mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
        tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
        tcm->tcm_parent = TC_H_INGRESS;
        mnl_attr_put_strz_check(nlh, sizeof(buf), TCA_KIND, "ingress");
+       assert(sizeof(buf) >= nlh->nlmsg_len);
        if (flow_tcf_nl_ack(ctx, nlh, 0, NULL, NULL))
                return rte_flow_error_set(error, rte_errno,
                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,