X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=c48dc9dfb5f42dcbe84346f99f89a8a08510199a;hb=6ce39213456b27257acbaf146398dce26d1466b9;hp=c3402849840d6436992b7a1cdd8d1503924aa24b;hpb=abfec865566e6cce961cc8660de1ddfdc85dae5f;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index c34028498..c48dc9dfb 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks. + * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks. * Distributed under the terms of the GNU GPL version 2. * * Significant portions of this file may be copied from parts of the Linux @@ -8,16 +8,19 @@ /* Functions for managing the dp interface/device. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include -#include #include #include #include #include +#include #include #include #include +#include #include #include #include @@ -27,7 +30,6 @@ #include #include #include -#include #include #include #include @@ -38,64 +40,83 @@ #include #include #include -#include #include #include -#include +#include #include "openvswitch/datapath-protocol.h" +#include "checksum.h" #include "datapath.h" #include "actions.h" #include "flow.h" -#include "odp-compat.h" +#include "loop_counter.h" #include "table.h" #include "vport-internal_dev.h" -#include "compat.h" - - int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); EXPORT_SYMBOL(dp_ioctl_hook); -/* Datapaths. Protected on the read side by rcu_read_lock, on the write side - * by dp_mutex. +/** + * DOC: Locking: + * + * Writes to device state (add/remove datapath, port, set operations on vports, + * etc.) are protected by RTNL. * - * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL - * lock first. + * Writes to other state (flow table modifications, set miscellaneous datapath + * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL + * lock nests inside genl_mutex. * - * It is safe to access the datapath and dp_port structures with just - * dp_mutex. + * Reads are protected by RCU. + * + * There are a few special cases (mostly stats) that have their own + * synchronization but they nest under all of above and don't interact with + * each other. + */ + +/* Global list of datapaths to enable dumping them all out. + * Protected by genl_mutex. */ -static struct datapath *dps[ODP_MAX]; -static DEFINE_MUTEX(dp_mutex); +static LIST_HEAD(dps); -static int new_dp_port(struct datapath *, struct odp_port *, int port_no); +static struct vport *new_vport(const struct vport_parms *); +static int queue_control_packets(struct datapath *, struct sk_buff *, + const struct dp_upcall_info *); -/* Must be called with rcu_read_lock or dp_mutex. */ -struct datapath *get_dp(int dp_idx) +/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */ +struct datapath *get_dp(int dp_ifindex) { - if (dp_idx < 0 || dp_idx >= ODP_MAX) - return NULL; - return rcu_dereference(dps[dp_idx]); + struct datapath *dp = NULL; + struct net_device *dev; + + rcu_read_lock(); + dev = dev_get_by_index_rcu(&init_net, dp_ifindex); + if (dev) { + struct vport *vport = internal_dev_get_vport(dev); + if (vport) + dp = vport->dp; + } + rcu_read_unlock(); + + return dp; } EXPORT_SYMBOL_GPL(get_dp); -static struct datapath *get_dp_locked(int dp_idx) +/* Must be called with genl_mutex. */ +static struct tbl *get_table_protected(struct datapath *dp) { - struct datapath *dp; + return rcu_dereference_protected(dp->table, lockdep_genl_is_held()); +} - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - if (dp) - mutex_lock(&dp->mutex); - mutex_unlock(&dp_mutex); - return dp; +/* Must be called with rcu_read_lock or RTNL lock. */ +static struct vport *get_vport_protected(struct datapath *dp, u16 port_no) +{ + return rcu_dereference_rtnl(dp->ports[port_no]); } /* Must be called with rcu_read_lock or RTNL lock. */ const char *dp_name(const struct datapath *dp) { - return vport_get_name(dp->ports[ODPP_LOCAL]->vport); + return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); } static inline size_t br_nlmsg_size(void) @@ -109,13 +130,14 @@ static inline size_t br_nlmsg_size(void) + nla_total_size(1); /* IFLA_OPERSTATE */ } +/* Caller must hold RTNL lock. */ static int dp_fill_ifinfo(struct sk_buff *skb, - const struct dp_port *port, + const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; - int ifindex = vport_get_ifindex(port->vport); - int iflink = vport_get_iflink(port->vport); + struct datapath *dp = port->dp; + int ifindex = vport_get_ifindex(port); + int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; struct nlmsghdr *nlh; @@ -134,21 +156,21 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->__ifi_pad = 0; hdr->ifi_type = ARPHRD_ETHER; hdr->ifi_index = ifindex; - hdr->ifi_flags = vport_get_flags(port->vport); + hdr->ifi_flags = vport_get_flags(port); hdr->ifi_change = 0; - NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport)); - NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport)); - NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport)); + NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port)); + NLA_PUT_U32(skb, IFLA_MASTER, + vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL))); + NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port)); #ifdef IFLA_OPERSTATE NLA_PUT_U8(skb, IFLA_OPERSTATE, - vport_is_running(port->vport) - ? vport_get_operstate(port->vport) + vport_is_running(port) + ? vport_get_operstate(port) : IF_OPER_DOWN); #endif - NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, - vport_get_addr(port->vport)); + NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port)); if (ifindex != iflink) NLA_PUT_U32(skb, IFLA_LINK,iflink); @@ -160,7 +182,8 @@ nla_put_failure: return -EMSGSIZE; } -static void dp_ifinfo_notify(int event, struct dp_port *port) +/* Caller must hold RTNL lock. */ +static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; int err = -ENOBUFS; @@ -193,260 +216,36 @@ static struct kobj_type dp_ktype = { .release = release_dp }; -static int create_dp(int dp_idx, const char __user *devnamep) -{ - struct odp_port internal_dev_port; - char devname[IFNAMSIZ]; - struct datapath *dp; - int err; - int i; - - if (devnamep) { - int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ); - if (retval < 0) { - err = -EFAULT; - goto err; - } else if (retval >= IFNAMSIZ) { - err = -ENAMETOOLONG; - goto err; - } - } else { - snprintf(devname, sizeof devname, "of%d", dp_idx); - } - - rtnl_lock(); - mutex_lock(&dp_mutex); - err = -ENODEV; - if (!try_module_get(THIS_MODULE)) - goto err_unlock; - - /* Exit early if a datapath with that number already exists. - * (We don't use -EEXIST because that's ambiguous with 'devname' - * conflicting with an existing network device name.) */ - err = -EBUSY; - if (get_dp(dp_idx)) - goto err_put_module; - - err = -ENOMEM; - dp = kzalloc(sizeof *dp, GFP_KERNEL); - if (dp == NULL) - goto err_put_module; - INIT_LIST_HEAD(&dp->port_list); - mutex_init(&dp->mutex); - dp->dp_idx = dp_idx; - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_head_init(&dp->queues[i]); - init_waitqueue_head(&dp->waitqueue); - - /* Initialize kobject for bridge. This will be added as - * /sys/class/net//brif later, if sysfs is enabled. */ - dp->ifobj.kset = NULL; - kobject_init(&dp->ifobj, &dp_ktype); - - /* Allocate table. */ - err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(0)); - if (!dp->table) - goto err_free_dp; - - /* Set up our datapath device. */ - BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname)); - strcpy(internal_dev_port.devname, devname); - internal_dev_port.flags = ODP_PORT_INTERNAL; - err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL); - if (err) { - if (err == -EBUSY) - err = -EEXIST; - - goto err_destroy_table; - } - - dp->drop_frags = 0; - dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) - goto err_destroy_local_port; - - rcu_assign_pointer(dps[dp_idx], dp); - mutex_unlock(&dp_mutex); - rtnl_unlock(); - - dp_sysfs_add_dp(dp); - - return 0; - -err_destroy_local_port: - dp_detach_port(dp->ports[ODPP_LOCAL], 1); -err_destroy_table: - tbl_destroy(dp->table, NULL); -err_free_dp: - kfree(dp); -err_put_module: - module_put(THIS_MODULE); -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); -err: - return err; -} - -static void do_destroy_dp(struct datapath *dp) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct dp_port *p, *n; - int i; - - list_for_each_entry_safe (p, n, &dp->port_list, node) - if (p->port_no != ODPP_LOCAL) - dp_detach_port(p, 1); - - dp_sysfs_del_dp(dp); + struct datapath *dp = container_of(rcu, struct datapath, rcu); - rcu_assign_pointer(dps[dp->dp_idx], NULL); - - dp_detach_port(dp->ports[ODPP_LOCAL], 1); - - tbl_destroy(dp->table, flow_free_tbl); - - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_purge(&dp->queues[i]); - for (i = 0; i < DP_MAX_GROUPS; i++) - kfree(dp->groups[i]); + tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); - module_put(THIS_MODULE); -} - -static int destroy_dp(int dp_idx) -{ - struct datapath *dp; - int err; - - rtnl_lock(); - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - err = -ENODEV; - if (!dp) - goto err_unlock; - - do_destroy_dp(dp); - err = 0; - -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); - return err; -} - -static void release_dp_port(struct kobject *kobj) -{ - struct dp_port *p = container_of(kobj, struct dp_port, kobj); - kfree(p); } -static struct kobj_type brport_ktype = { -#ifdef CONFIG_SYSFS - .sysfs_ops = &brport_sysfs_ops, -#endif - .release = release_dp_port -}; - -/* Called with RTNL lock and dp_mutex. */ -static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no) +/* Called with RTNL lock and genl_lock. */ +static struct vport *new_vport(const struct vport_parms *parms) { struct vport *vport; - struct dp_port *p; - int err; - - vport = vport_locate(odp_port->devname); - if (!vport) { - vport_lock(); - - if (odp_port->flags & ODP_PORT_INTERNAL) - vport = vport_add(odp_port->devname, "internal", NULL); - else - vport = vport_add(odp_port->devname, "netdev", NULL); - vport_unlock(); + vport = vport_add(parms); + if (!IS_ERR(vport)) { + struct datapath *dp = parms->dp; - if (IS_ERR(vport)) - return PTR_ERR(vport); - } - - p = kzalloc(sizeof(*p), GFP_KERNEL); - if (!p) - return -ENOMEM; - - p->port_no = port_no; - p->dp = dp; - p->vport = vport; - atomic_set(&p->sflow_pool, 0); + rcu_assign_pointer(dp->ports[parms->port_no], vport); + list_add(&vport->node, &dp->port_list); - err = vport_attach(vport, p); - if (err) { - kfree(p); - return err; + dp_ifinfo_notify(RTM_NEWLINK, vport); } - rcu_assign_pointer(dp->ports[port_no], p); - list_add_rcu(&p->node, &dp->port_list); - dp->n_ports++; - - /* Initialize kobject for bridge. This will be added as - * /sys/class/net//brport later, if sysfs is enabled. */ - p->kobj.kset = NULL; - kobject_init(&p->kobj, &brport_ktype); - - dp_ifinfo_notify(RTM_NEWLINK, p); - - return 0; -} - -static int attach_port(int dp_idx, struct odp_port __user *portp) -{ - struct datapath *dp; - struct odp_port port; - int port_no; - int err; - - err = -EFAULT; - if (copy_from_user(&port, portp, sizeof port)) - goto out; - port.devname[IFNAMSIZ - 1] = '\0'; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - for (port_no = 1; port_no < DP_MAX_PORTS; port_no++) - if (!dp->ports[port_no]) - goto got_port_no; - err = -EFBIG; - goto out_unlock_dp; - -got_port_no: - err = new_dp_port(dp, &port, port_no); - if (err) - goto out_unlock_dp; - - set_internal_devs_mtu(dp); - dp_sysfs_add_if(dp->ports[port_no]); - - err = put_user(port_no, &portp->port); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: - return err; + return vport; } -int dp_detach_port(struct dp_port *p, int may_delete) +/* Called with RTNL lock. */ +int dp_detach_port(struct vport *p) { - struct vport *vport = p->vport; - int err; - ASSERT_RTNL(); if (p->port_no != ODPP_LOCAL) @@ -454,326 +253,173 @@ int dp_detach_port(struct dp_port *p, int may_delete) dp_ifinfo_notify(RTM_DELLINK, p); /* First drop references to device. */ - p->dp->n_ports--; - list_del_rcu(&p->node); + list_del(&p->node); rcu_assign_pointer(p->dp->ports[p->port_no], NULL); - err = vport_detach(vport); - if (err) - return err; - - /* Then wait until no one is still using it, and destroy it. */ - synchronize_rcu(); - - if (may_delete) { - const char *port_type = vport_get_type(vport); - - if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) { - vport_lock(); - vport_del(vport); - vport_unlock(); - } - } - - kobject_put(&p->kobj); - - return 0; -} - -static int detach_port(int dp_idx, int port_no) -{ - struct dp_port *p; - struct datapath *dp; - int err; - - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) - goto out; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - p = dp->ports[port_no]; - err = -ENOENT; - if (!p) - goto out_unlock_dp; - - err = dp_detach_port(p, 1); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: - return err; + /* Then destroy it. */ + return vport_del(p); } /* Must be called with rcu_read_lock. */ -void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb) +void dp_process_received_packet(struct vport *p, struct sk_buff *skb) { struct datapath *dp = p->dp; struct dp_stats_percpu *stats; int stats_counter_off; - struct odp_flow_key key; - struct tbl_node *flow_node; + struct sw_flow_actions *acts; + struct loop_counter *loop; + int error; + + OVS_CB(skb)->vport = p; + + if (!OVS_CB(skb)->flow) { + struct sw_flow_key key; + struct tbl_node *flow_node; + bool is_frag; - OVS_CB(skb)->dp_port = p; + /* Extract flow from 'skb' into 'key'. */ + error = flow_extract(skb, p->port_no, &key, &is_frag); + if (unlikely(error)) { + kfree_skb(skb); + return; + } - if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { - if (dp->drop_frags) { + if (is_frag && dp->drop_frags) { kfree_skb(skb); stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); goto out; } + + /* Look up flow. */ + flow_node = tbl_lookup(rcu_dereference(dp->table), &key, + flow_hash(&key), flow_cmp); + if (unlikely(!flow_node)) { + struct dp_upcall_info upcall; + + upcall.cmd = ODP_PACKET_CMD_MISS; + upcall.key = &key; + upcall.userdata = 0; + upcall.sample_pool = 0; + upcall.actions = NULL; + upcall.actions_len = 0; + dp_upcall(dp, skb, &upcall); + stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + goto out; + } + + OVS_CB(skb)->flow = flow_cast(flow_node); } - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); - if (flow_node) { - struct sw_flow *flow = flow_cast(flow_node); - struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts); - flow_used(flow, skb); - execute_actions(dp, skb, &key, acts->actions, acts->n_actions, - GFP_ATOMIC); - stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); - } else { - stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + flow_used(OVS_CB(skb)->flow, skb); + + acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); + + /* Check whether we've looped too much. */ + loop = loop_get_counter(); + if (unlikely(++loop->count > MAX_LOOPS)) + loop->looping = true; + if (unlikely(loop->looping)) { + loop_suppress(dp, acts); + kfree_skb(skb); + goto out_loop; } + /* Execute actions. */ + execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions, + acts->actions_len); + + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + loop_suppress(dp, acts); + +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + loop_put_counter(); + out: + /* Update datapath statistics. */ local_bh_disable(); stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); (*(u64 *)((u8 *)stats + stats_counter_off))++; + write_seqcount_end(&stats->seqlock); + local_bh_enable(); } -#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) -/* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We - * can't call this function directly because it isn't exported in all - * versions. */ -int vswitch_skb_checksum_setup(struct sk_buff *skb) +static void copy_and_csum_skb(struct sk_buff *skb, void *to) { - struct iphdr *iph; - unsigned char *th; - int err = -EPROTO; - __u16 csum_start, csum_offset; + u16 csum_start, csum_offset; + __wsum csum; - if (!skb->proto_csum_blank) - return 0; - - if (skb->protocol != htons(ETH_P_IP)) - goto out; - - if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data)) - goto out; - - iph = ip_hdr(skb); - th = skb_network_header(skb) + 4 * iph->ihl; - - csum_start = th - skb->head; - switch (iph->protocol) { - case IPPROTO_TCP: - csum_offset = offsetof(struct tcphdr, check); - break; - case IPPROTO_UDP: - csum_offset = offsetof(struct udphdr, check); - break; - default: - if (net_ratelimit()) - printk(KERN_ERR "Attempting to checksum a non-" - "TCP/UDP packet, dropping a protocol" - " %d packet", iph->protocol); - goto out; - } - - if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data)) - goto out; + get_skb_csum_pointers(skb, &csum_start, &csum_offset); + csum_start -= skb_headroom(skb); + BUG_ON(csum_start >= skb_headlen(skb)); - skb->ip_summed = CHECKSUM_PARTIAL; - skb->proto_csum_blank = 0; + skb_copy_bits(skb, 0, to, csum_start); -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - skb->csum_start = csum_start; - skb->csum_offset = csum_offset; -#else - skb_set_transport_header(skb, csum_start - skb_headroom(skb)); - skb->csum = csum_offset; -#endif + csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start, + skb->len - csum_start, 0); + *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum); +} - err = 0; +static struct genl_family dp_packet_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct odp_header), + .name = ODP_PACKET_FAMILY, + .version = 1, + .maxattr = ODP_PACKET_ATTR_MAX +}; -out: - return err; -} -#endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */ - - /* Types of checksums that we can receive (these all refer to L4 checksums): - * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full - * (though not verified) checksum in packet but not in skb->csum. Packets - * from the bridge local port will also have this type. - * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums, - * also the GRE module. This is the same as CHECKSUM_NONE, except it has - * a valid skb->csum. Importantly, both contain a full checksum (not - * verified) in the packet itself. The only difference is that if the - * packet gets to L4 processing on this machine (not in DomU) we won't - * have to recompute the checksum to verify. Most hardware devices do not - * produce packets with this type, even if they support receive checksum - * offloading (they produce type #5). - * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to - * be computed if it is sent off box. Unfortunately on earlier kernels, - * this case is impossible to distinguish from #2, despite having opposite - * meanings. Xen adds an extra field on earlier kernels (see #4) in order - * to distinguish the different states. - * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was - * generated locally by a Xen DomU and has a partial checksum. If it is - * handled on this machine (Dom0 or DomU), then the checksum will not be - * computed. If it goes off box, the checksum in the packet needs to be - * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW - * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later - * kernels, this combination is replaced with CHECKSUM_PARTIAL. - * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct - * full checksum or using a protocol without a checksum. skb->csum is - * undefined. This is common from devices with receive checksum - * offloading. This is somewhat similar to CHECKSUM_NONE, except that - * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY. +/* Generic Netlink multicast groups for upcalls. * - * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are - * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear - * based on whether it is on the transmit or receive path. After the datapath - * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a - * checksum, we will panic. Since we can receive packets with checksums, we - * assume that all CHECKSUM_HW packets have checksums and map them to - * CHECKSUM_NONE, which has a similar meaning (the it is only different if the - * packet is processed by the local IP stack, in which case it will need to - * be reverified). If we receive a packet with CHECKSUM_HW that really means - * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there - * shouldn't be any devices that do this with bridging. */ -void compute_ip_summed(struct sk_buff *skb, bool xmit) -{ - /* For our convenience these defines change repeatedly between kernel - * versions, so we can't just copy them over... */ - switch (skb->ip_summed) { - case CHECKSUM_NONE: - OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - break; - case CHECKSUM_UNNECESSARY: - OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY; - break; -#ifdef CHECKSUM_HW - /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE. - * However, on the receive side we should only get CHECKSUM_PARTIAL - * packets from Xen, which uses some special fields to represent this - * (see below). Since we can only make one type work, pick the one - * that actually happens in practice. - * - * On the transmit side (basically after skb_checksum_setup() - * has been run or on internal dev transmit), packets with - * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */ - case CHECKSUM_HW: - if (!xmit) - OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; - else - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; - - break; -#else - case CHECKSUM_COMPLETE: - OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; - break; - case CHECKSUM_PARTIAL: - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; - break; -#endif - default: - printk(KERN_ERR "openvswitch: unknown checksum type %d\n", - skb->ip_summed); - /* None seems the safest... */ - OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - } - -#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) - /* Xen has a special way of representing CHECKSUM_PARTIAL on older - * kernels. It should not be set on the transmit path though. */ - if (skb->proto_csum_blank) - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; - - WARN_ON_ONCE(skb->proto_csum_blank && xmit); -#endif -} + * We really want three unique multicast groups per datapath, but we can't even + * get one, because genl_register_mc_group() takes genl_lock, which is also + * held during Generic Netlink message processing, so trying to acquire + * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we + * preallocate a few groups and use them round-robin for datapaths. Collision + * isn't fatal--multicast listeners should check that the family is the one + * that they want and discard others--but it wastes time and memory to receive + * unwanted messages. + */ +#define PACKET_N_MC_GROUPS 16 +static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS]; -/* This function closely resembles skb_forward_csum() used by the bridge. It - * is slightly different because we are only concerned with bridging and not - * other types of forwarding and can get away with slightly more optimal - * behavior.*/ -void forward_ip_summed(struct sk_buff *skb) +static u32 packet_mc_group(struct datapath *dp, u8 cmd) { -#ifdef CHECKSUM_HW - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) - skb->ip_summed = CHECKSUM_NONE; -#endif + u32 idx; + BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS); + + idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1); + return packet_mc_groups[idx].id; } -/* Append each packet in 'skb' list to 'queue'. There will be only one packet - * unless we broke up a GSO packet. */ -static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, - int queue_no, u32 arg) +static int packet_register_mc_groups(void) { - struct sk_buff *nskb; - int port_no; - int err; - - if (OVS_CB(skb)->dp_port) - port_no = OVS_CB(skb)->dp_port->port_no; - else - port_no = ODPP_LOCAL; - - do { - struct odp_msg *header; - - nskb = skb->next; - skb->next = NULL; - - err = skb_cow(skb, sizeof *header); - if (err) - goto err_kfree_skbs; - - header = (struct odp_msg*)__skb_push(skb, sizeof *header); - header->type = queue_no; - header->length = skb->len; - header->port = port_no; - header->reserved = 0; - header->arg = arg; - skb_queue_tail(queue, skb); + int i; - skb = nskb; - } while (skb); - return 0; + for (i = 0; i < PACKET_N_MC_GROUPS; i++) { + struct genl_multicast_group *group = &packet_mc_groups[i]; + int error; -err_kfree_skbs: - kfree_skb(skb); - while ((skb = nskb) != NULL) { - nskb = skb->next; - kfree_skb(skb); + sprintf(group->name, "packet%d", i); + error = genl_register_mc_group(&dp_packet_genl_family, group); + if (error) + return error; } - return err; + return 0; } -int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, - u32 arg) +int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info) { struct dp_stats_percpu *stats; - struct sk_buff_head *queue; int err; WARN_ON_ONCE(skb_shared(skb)); - BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR); - queue = &dp->queues[queue_no]; - err = -ENOBUFS; - if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) - goto err_kfree_skb; forward_ip_summed(skb); @@ -785,152 +431,221 @@ int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ if (skb_is_gso(skb)) { struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); - if (nskb) { - kfree_skb(skb); - skb = nskb; - if (unlikely(IS_ERR(skb))) { - err = PTR_ERR(skb); - goto err; - } - } else { - /* XXX This case might not be possible. It's hard to - * tell from the skb_gso_segment() code and comment. */ + + kfree_skb(skb); + skb = nskb; + if (IS_ERR(skb)) { + err = PTR_ERR(skb); + goto err; } } - err = queue_control_packets(skb, queue, queue_no, arg); - wake_up_interruptible(&dp->waitqueue); - return err; + return queue_control_packets(dp, skb, upcall_info); err_kfree_skb: kfree_skb(skb); err: local_bh_disable(); stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); stats->n_lost++; + write_seqcount_end(&stats->seqlock); + local_bh_enable(); return err; } -static int flush_flows(struct datapath *dp) +/* Send each packet in the 'skb' list to userspace for 'dp' as directed by + * 'upcall_info'. There will be only one packet unless we broke up a GSO + * packet. + */ +static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - struct tbl *old_table = rcu_dereference(dp->table); - struct tbl *new_table; - - new_table = tbl_create(0); - if (!new_table) - return -ENOMEM; - - rcu_assign_pointer(dp->table, new_table); + u32 group = packet_mc_group(dp, upcall_info->cmd); + struct sk_buff *nskb; + int port_no; + int err; - tbl_deferred_destroy(old_table, flow_free_tbl); + if (OVS_CB(skb)->vport) + port_no = OVS_CB(skb)->vport->port_no; + else + port_no = ODPP_LOCAL; - return 0; -} + do { + struct odp_header *upcall; + struct sk_buff *user_skb; /* to be queued to userspace */ + struct nlattr *nla; + unsigned int len; -static int validate_actions(const struct sw_flow_actions *actions) -{ - unsigned int i; + nskb = skb->next; + skb->next = NULL; - for (i = 0; i < actions->n_actions; i++) { - const union odp_action *a = &actions->actions[i]; - switch (a->type) { - case ODPAT_OUTPUT: - if (a->output.port >= DP_MAX_PORTS) - return -EINVAL; - break; + if (vlan_tx_tag_present(skb)) { + skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); + if (unlikely(!skb)) { + err = -ENOMEM; + goto err_kfree_skbs; + } + } - case ODPAT_OUTPUT_GROUP: - if (a->output_group.group >= DP_MAX_GROUPS) - return -EINVAL; - break; + len = sizeof(struct odp_header); + len += nla_total_size(skb->len); + len += nla_total_size(FLOW_BUFSIZE); + if (upcall_info->userdata) + len += nla_total_size(8); + if (upcall_info->sample_pool) + len += nla_total_size(4); + if (upcall_info->actions_len) + len += nla_total_size(upcall_info->actions_len); + + user_skb = genlmsg_new(len, GFP_ATOMIC); + if (!user_skb) { + netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS); + goto err_kfree_skbs; + } - case ODPAT_SET_VLAN_VID: - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) - return -EINVAL; - break; + upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd); + upcall->dp_ifindex = dp->dp_ifindex; - case ODPAT_SET_VLAN_PCP: - if (a->vlan_pcp.vlan_pcp - & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) - return -EINVAL; - break; + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY); + flow_to_nlattrs(upcall_info->key, user_skb); + nla_nest_end(user_skb, nla); - case ODPAT_SET_NW_TOS: - if (a->nw_tos.nw_tos & INET_ECN_MASK) - return -EINVAL; - break; + if (upcall_info->userdata) + nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata); + if (upcall_info->sample_pool) + nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool); + if (upcall_info->actions_len) { + const struct nlattr *actions = upcall_info->actions; + u32 actions_len = upcall_info->actions_len; - default: - if (a->type >= ODPAT_N_ACTIONS) - return -EOPNOTSUPP; - break; + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS); + memcpy(__skb_put(user_skb, actions_len), actions, actions_len); + nla_nest_end(user_skb, nla); } - } + nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len); + if (skb->ip_summed == CHECKSUM_PARTIAL) + copy_and_csum_skb(skb, nla_data(nla)); + else + skb_copy_bits(skb, 0, nla_data(nla), skb->len); + + err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC); + if (err) + goto err_kfree_skbs; + + kfree_skb(skb); + skb = nskb; + } while (skb); return 0; + +err_kfree_skbs: + kfree_skb(skb); + while ((skb = nskb) != NULL) { + nskb = skb->next; + kfree_skb(skb); + } + return err; } -static struct sw_flow_actions *get_actions(const struct odp_flow *flow) +/* Called with genl_mutex. */ +static int flush_flows(int dp_ifindex) { - struct sw_flow_actions *actions; - int error; + struct tbl *old_table; + struct tbl *new_table; + struct datapath *dp; - actions = flow_actions_alloc(flow->n_actions); - error = PTR_ERR(actions); - if (IS_ERR(actions)) - goto error; + dp = get_dp(dp_ifindex); + if (!dp) + return -ENODEV; - error = -EFAULT; - if (copy_from_user(actions->actions, flow->actions, - flow->n_actions * sizeof(union odp_action))) - goto error_free_actions; - error = validate_actions(actions); - if (error) - goto error_free_actions; + old_table = get_table_protected(dp); + new_table = tbl_create(TBL_MIN_BUCKETS); + if (!new_table) + return -ENOMEM; - return actions; + rcu_assign_pointer(dp->table, new_table); -error_free_actions: - kfree(actions); -error: - return ERR_PTR(error); + tbl_deferred_destroy(old_table, flow_free_tbl); + + return 0; } -static struct timespec get_time_offset(void) -{ - struct timespec now_mono, now_jiffies; +static int validate_actions(const struct nlattr *attr) +{ + const struct nlattr *a; + int rem; + + nla_for_each_nested(a, attr, rem) { + static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = { + [ODP_ACTION_ATTR_OUTPUT] = 4, + [ODP_ACTION_ATTR_CONTROLLER] = 8, + [ODP_ACTION_ATTR_SET_DL_TCI] = 2, + [ODP_ACTION_ATTR_STRIP_VLAN] = 0, + [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN, + [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN, + [ODP_ACTION_ATTR_SET_NW_SRC] = 4, + [ODP_ACTION_ATTR_SET_NW_DST] = 4, + [ODP_ACTION_ATTR_SET_NW_TOS] = 1, + [ODP_ACTION_ATTR_SET_TP_SRC] = 2, + [ODP_ACTION_ATTR_SET_TP_DST] = 2, + [ODP_ACTION_ATTR_SET_TUNNEL] = 8, + [ODP_ACTION_ATTR_SET_PRIORITY] = 4, + [ODP_ACTION_ATTR_POP_PRIORITY] = 0, + [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0, + }; + int type = nla_type(a); + + if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type]) + return -EINVAL; - ktime_get_ts(&now_mono); - jiffies_to_timespec(jiffies, &now_jiffies); - return timespec_sub(now_mono, now_jiffies); -} + switch (type) { + case ODP_ACTION_ATTR_UNSPEC: + return -EINVAL; -static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats, - struct timespec time_offset) -{ - if (flow->used) { - struct timespec flow_ts, used; + case ODP_ACTION_ATTR_CONTROLLER: + case ODP_ACTION_ATTR_STRIP_VLAN: + case ODP_ACTION_ATTR_SET_DL_SRC: + case ODP_ACTION_ATTR_SET_DL_DST: + case ODP_ACTION_ATTR_SET_NW_SRC: + case ODP_ACTION_ATTR_SET_NW_DST: + case ODP_ACTION_ATTR_SET_TP_SRC: + case ODP_ACTION_ATTR_SET_TP_DST: + case ODP_ACTION_ATTR_SET_TUNNEL: + case ODP_ACTION_ATTR_SET_PRIORITY: + case ODP_ACTION_ATTR_POP_PRIORITY: + case ODP_ACTION_ATTR_DROP_SPOOFED_ARP: + /* No validation needed. */ + break; - jiffies_to_timespec(flow->used, &flow_ts); - set_normalized_timespec(&used, flow_ts.tv_sec + time_offset.tv_sec, - flow_ts.tv_nsec + time_offset.tv_nsec); + case ODP_ACTION_ATTR_OUTPUT: + if (nla_get_u32(a) >= DP_MAX_PORTS) + return -EINVAL; + break; - stats->used_sec = used.tv_sec; - stats->used_nsec = used.tv_nsec; - } else { - stats->used_sec = 0; - stats->used_nsec = 0; + case ODP_ACTION_ATTR_SET_DL_TCI: + if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) + return -EINVAL; + break; + + case ODP_ACTION_ATTR_SET_NW_TOS: + if (nla_get_u8(a) & INET_ECN_MASK) + return -EINVAL; + break; + + default: + return -EOPNOTSUPP; + } } - stats->n_packets = flow->packet_count; - stats->n_bytes = flow->byte_count; - stats->reserved = 0; - stats->tcp_flags = flow->tcp_flags; - stats->error = 0; -} + if (rem > 0) + return -EINVAL; + return 0; +} static void clear_stats(struct sw_flow *flow) { flow->used = 0; @@ -939,9 +654,10 @@ static void clear_stats(struct sw_flow *flow) flow->byte_count = 0; } +/* Called with genl_mutex. */ static int expand_table(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table = get_table_protected(dp); struct tbl *new_table; new_table = tbl_expand(old_table); @@ -951,1338 +667,1359 @@ static int expand_table(struct datapath *dp) rcu_assign_pointer(dp->table, new_table); tbl_deferred_destroy(old_table, NULL); - return 0; + return 0; } -static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, - struct odp_flow_stats *stats) +static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) { - struct tbl_node *flow_node; - struct sw_flow *flow; - struct tbl *table; - int error; - - memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved); - - table = rcu_dereference(dp->table); - flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); - if (!flow_node) { - /* No such flow. */ - struct sw_flow_actions *acts; + struct odp_header *odp_header = info->userhdr; + struct nlattr **a = info->attrs; + struct sk_buff *packet; + struct sw_flow_key key; + struct datapath *dp; + struct ethhdr *eth; + bool is_frag; + int err; - error = -ENOENT; - if (!(uf->flags & ODPPF_CREATE)) - goto error; + err = -EINVAL; + if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] || + nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN) + goto exit; - /* Expand table, if necessary, to make room. */ - if (tbl_count(table) >= tbl_n_buckets(table)) { - error = expand_table(dp); - if (error) - goto error; - table = rcu_dereference(dp->table); - } + err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]); + if (err) + goto exit; - /* Allocate flow. */ - error = -ENOMEM; - flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); - if (flow == NULL) - goto error; - flow->key = uf->flow.key; - spin_lock_init(&flow->lock); - clear_stats(flow); + packet = skb_clone(skb, GFP_KERNEL); + err = -ENOMEM; + if (!packet) + goto exit; + packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]); + packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]); - /* Obtain actions. */ - acts = get_actions(&uf->flow); - error = PTR_ERR(acts); - if (IS_ERR(acts)) - goto error_free_flow; - rcu_assign_pointer(flow->sf_acts, acts); + skb_reset_mac_header(packet); + eth = eth_hdr(packet); - /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key)); - if (error) - goto error_free_flow_acts; + /* Normally, setting the skb 'protocol' field would be handled by a + * call to eth_type_trans(), but it assumes there's a sending + * device, which we may not have. */ + if (ntohs(eth->h_proto) >= 1536) + packet->protocol = eth->h_proto; + else + packet->protocol = htons(ETH_P_802_2); - memset(stats, 0, sizeof(struct odp_flow_stats)); - } else { - /* We found a matching flow. */ - struct sw_flow_actions *old_acts, *new_acts; + err = flow_extract(packet, -1, &key, &is_frag); + if (err) + goto exit; - flow = flow_cast(flow_node); + rcu_read_lock(); + dp = get_dp(odp_header->dp_ifindex); + err = -ENODEV; + if (dp) + err = execute_actions(dp, packet, &key, + nla_data(a[ODP_PACKET_ATTR_ACTIONS]), + nla_len(a[ODP_PACKET_ATTR_ACTIONS])); + rcu_read_unlock(); - /* Bail out if we're not allowed to modify an existing flow. */ - error = -EEXIST; - if (!(uf->flags & ODPPF_MODIFY)) - goto error; +exit: + return err; +} - /* Swap actions. */ - new_acts = get_actions(&uf->flow); - error = PTR_ERR(new_acts); - if (IS_ERR(new_acts)) - goto error; - old_acts = rcu_dereference(flow->sf_acts); - if (old_acts->n_actions != new_acts->n_actions || - memcmp(old_acts->actions, new_acts->actions, - sizeof(union odp_action) * old_acts->n_actions)) { - rcu_assign_pointer(flow->sf_acts, new_acts); - flow_deferred_free_acts(old_acts); - } else { - kfree(new_acts); - } +static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = { + [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; - /* Fetch stats, then clear them if necessary. */ - spin_lock_bh(&flow->lock); - get_stats(flow, stats, get_time_offset()); - if (uf->flags & ODPPF_ZERO_STATS) - clear_stats(flow); - spin_unlock_bh(&flow->lock); +static struct genl_ops dp_packet_genl_ops[] = { + { .cmd = ODP_PACKET_CMD_EXECUTE, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = packet_policy, + .doit = odp_packet_cmd_execute } +}; - return 0; - -error_free_flow_acts: - kfree(flow->sf_acts); -error_free_flow: - kmem_cache_free(flow_cache, flow); -error: - return error; -} - -static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) +static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) { - struct odp_flow_stats stats; - struct odp_flow_put uf; - int error; + int i; - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) - return -EFAULT; + stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0; + for_each_possible_cpu(i) { + const struct dp_stats_percpu *percpu_stats; + struct dp_stats_percpu local_stats; + unsigned seqcount; - error = do_put_flow(dp, &uf, &stats); - if (error) - return error; + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); - return 0; + stats->n_frags += local_stats.n_frags; + stats->n_hit += local_stats.n_hit; + stats->n_missed += local_stats.n_missed; + stats->n_lost += local_stats.n_lost; + } } -static int do_answer_query(struct sw_flow *flow, u32 query_flags, - struct timespec time_offset, - struct odp_flow_stats __user *ustats, - union odp_action __user *actions, - u32 __user *n_actionsp) +/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports. + * Called with RTNL lock. + */ +int dp_min_mtu(const struct datapath *dp) { - struct sw_flow_actions *sf_acts; - struct odp_flow_stats stats; - u32 n_actions; - - spin_lock_bh(&flow->lock); - get_stats(flow, &stats, time_offset); - if (query_flags & ODPFF_ZERO_TCP_FLAGS) - flow->tcp_flags = 0; + struct vport *p; + int mtu = 0; - spin_unlock_bh(&flow->lock); + ASSERT_RTNL(); - if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(n_actions, n_actionsp)) - return -EFAULT; + list_for_each_entry (p, &dp->port_list, node) { + int dev_mtu; - if (!n_actions) - return 0; + /* Skip any internal ports, since that's what we're trying to + * set. */ + if (is_internal_vport(p)) + continue; - sf_acts = rcu_dereference(flow->sf_acts); - if (put_user(sf_acts->n_actions, n_actionsp) || - (actions && copy_to_user(actions, sf_acts->actions, - sizeof(union odp_action) * - min(sf_acts->n_actions, n_actions)))) - return -EFAULT; + dev_mtu = vport_get_mtu(p); + if (!dev_mtu) + continue; + if (!mtu || dev_mtu < mtu) + mtu = dev_mtu; + } - return 0; + return mtu ? mtu : ETH_DATA_LEN; } -static int answer_query(struct sw_flow *flow, u32 query_flags, - struct timespec time_offset, - struct odp_flow __user *ufp) +/* Sets the MTU of all datapath devices to the minimum of the ports + * Called with RTNL lock. + */ +void set_internal_devs_mtu(const struct datapath *dp) { - union odp_action *actions; + struct vport *p; + int mtu; - if (get_user(actions, &ufp->actions)) - return -EFAULT; + ASSERT_RTNL(); - return do_answer_query(flow, query_flags, time_offset, - &ufp->stats, actions, &ufp->n_actions); -} + mtu = dp_min_mtu(dp); -static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key) -{ - struct tbl *table = rcu_dereference(dp->table); - struct tbl_node *flow_node; - int error; + list_for_each_entry (p, &dp->port_list, node) { + if (is_internal_vport(p)) + vport_set_mtu(p, mtu); + } +} - memset(key->reserved, 0, sizeof key->reserved); - flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp); - if (!flow_node) - return ERR_PTR(-ENOENT); +static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = { + [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, + [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, + [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, +}; - error = tbl_remove(table, flow_node); - if (error) - return ERR_PTR(error); +static struct genl_family dp_flow_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct odp_header), + .name = ODP_FLOW_FAMILY, + .version = 1, + .maxattr = ODP_FLOW_ATTR_MAX +}; - /* XXX Returned flow_node's statistics might lose a few packets, since - * other CPUs can be using this flow. We used to synchronize_rcu() to - * make sure that we get completely accurate stats, but that blows our - * performance, badly. */ - return flow_cast(flow_node); -} +static struct genl_multicast_group dp_flow_multicast_group = { + .name = ODP_FLOW_MCGROUP +}; -static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) +/* Called with genl_lock. */ +static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, + struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd) { - struct sw_flow *flow; - struct odp_flow uf; - int error; - - if (copy_from_user(&uf, ufp, sizeof uf)) - return -EFAULT; + const int skb_orig_len = skb->len; + const struct sw_flow_actions *sf_acts; + struct odp_flow_stats stats; + struct odp_header *odp_header; + struct nlattr *nla; + unsigned long used; + u8 tcp_flags; + int err; - flow = do_del_flow(dp, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); - error = answer_query(flow, 0, get_time_offset(), ufp); - flow_deferred_free(flow); - return error; -} + odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd); + if (!odp_header) + return -EMSGSIZE; -static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) -{ - struct tbl *table = rcu_dereference(dp->table); - struct timespec time_offset; - u32 i; + odp_header->dp_ifindex = dp->dp_ifindex; - time_offset = get_time_offset(); + nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY); + if (!nla) + goto nla_put_failure; + err = flow_to_nlattrs(&flow->key, skb); + if (err) + goto error; + nla_nest_end(skb, nla); - for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = &flowvec->flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; + spin_lock_bh(&flow->lock); + used = flow->used; + stats.n_packets = flow->packet_count; + stats.n_bytes = flow->byte_count; + tcp_flags = flow->tcp_flags; + spin_unlock_bh(&flow->lock); - if (copy_from_user(&uf, ufp, sizeof uf)) - return -EFAULT; - memset(uf.key.reserved, 0, sizeof uf.key.reserved); + if (used) + NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used)); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp); - if (error) - return -EFAULT; - } - return flowvec->n_flows; -} + if (stats.n_packets) + NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats); -struct list_flows_cbdata { - struct odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; - struct timespec time_offset; -}; + if (tcp_flags) + NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags); -static int list_flow(struct tbl_node *node, void *cbdata_) -{ - struct sw_flow *flow = flow_cast(node); - struct list_flows_cbdata *cbdata = cbdata_; - struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; - int error; + /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if + * this is the first flow to be dumped into 'skb'. This is unusual for + * Netlink but individual action lists can be longer than + * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. + * The userspace caller can always fetch the actions separately if it + * really wants them. (Most userspace callers in fact don't care.) + * + * This can only fail for dump operations because the skb is always + * properly sized for single flows. + */ + err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len, + sf_acts->actions); + if (err < 0 && skb_orig_len) + goto error; - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) - return -EFAULT; - error = answer_query(flow, 0, cbdata->time_offset, ufp); - if (error) - return error; + return genlmsg_end(skb, odp_header); - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; +nla_put_failure: + err = -EMSGSIZE; +error: + genlmsg_cancel(skb, odp_header); + return err; } -static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) +static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow) { - struct list_flows_cbdata cbdata; - int error; - - if (!flowvec->n_flows) - return 0; + const struct sw_flow_actions *sf_acts; + int len; - cbdata.uflows = flowvec->flows; - cbdata.n_flows = flowvec->n_flows; - cbdata.listed_flows = 0; - cbdata.time_offset = get_time_offset(); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); - error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata); - return error ? error : cbdata.listed_flows; + len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */ + len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */ + len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */ + len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */ + len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */ + return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL); } -static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - const struct odp_flowvec *)) +static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp, + u32 pid, u32 seq, u8 cmd) { - struct odp_flowvec __user *uflowvec; - struct odp_flowvec flowvec; + struct sk_buff *skb; int retval; - uflowvec = (struct odp_flowvec __user *)argp; - if (copy_from_user(&flowvec, uflowvec, sizeof flowvec)) - return -EFAULT; - - if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) - return -EINVAL; + skb = odp_flow_cmd_alloc_info(flow); + if (!skb) + return ERR_PTR(-ENOMEM); - retval = function(dp, &flowvec); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : put_user(retval, &uflowvec->n_flows)); + retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); + BUG_ON(retval < 0); + return skb; } -static int do_execute(struct datapath *dp, const struct odp_execute *execute) +static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) { - struct odp_flow_key key; - struct sk_buff *skb; - struct sw_flow_actions *actions; - struct ethhdr *eth; - int err; + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; + struct tbl_node *flow_node; + struct sw_flow_key key; + struct sw_flow *flow; + struct sk_buff *reply; + struct datapath *dp; + struct tbl *table; + u32 hash; + int error; - err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) + /* Extract key. */ + error = -EINVAL; + if (!a[ODP_FLOW_ATTR_KEY]) goto error; - - err = -ENOMEM; - actions = flow_actions_alloc(execute->n_actions); - if (!actions) + error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]); + if (error) goto error; - err = -EFAULT; - if (copy_from_user(actions->actions, execute->actions, - execute->n_actions * sizeof *execute->actions)) - goto error_free_actions; - - err = validate_actions(actions); - if (err) - goto error_free_actions; + /* Validate actions. */ + if (a[ODP_FLOW_ATTR_ACTIONS]) { + error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]); + if (error) + goto error; + } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) { + error = -EINVAL; + goto error; + } - err = -ENOMEM; - skb = alloc_skb(execute->length, GFP_KERNEL); - if (!skb) - goto error_free_actions; + dp = get_dp(odp_header->dp_ifindex); + error = -ENODEV; + if (!dp) + goto error; - if (execute->in_port < DP_MAX_PORTS) - OVS_CB(skb)->dp_port = dp->ports[execute->in_port]; - else - OVS_CB(skb)->dp_port = NULL; + hash = flow_hash(&key); + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, hash, flow_cmp); + if (!flow_node) { + struct sw_flow_actions *acts; - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), execute->data, - execute->length)) - goto error_free_skb; + /* Bail out if we're not allowed to create a new flow. */ + error = -ENOENT; + if (info->genlhdr->cmd == ODP_FLOW_CMD_SET) + goto error; - skb_reset_mac_header(skb); - eth = eth_hdr(skb); + /* Expand table, if necessary, to make room. */ + if (tbl_count(table) >= tbl_n_buckets(table)) { + error = expand_table(dp); + if (error) + goto error; + table = get_table_protected(dp); + } - /* Normally, setting the skb 'protocol' field would be handled by a - * call to eth_type_trans(), but it assumes there's a sending - * device, which we may not have. */ - if (ntohs(eth->h_proto) >= 1536) - skb->protocol = eth->h_proto; - else - skb->protocol = htons(ETH_P_802_2); + /* Allocate flow. */ + flow = flow_alloc(); + if (IS_ERR(flow)) { + error = PTR_ERR(flow); + goto error; + } + flow->key = key; + clear_stats(flow); - flow_extract(skb, execute->in_port, &key); + /* Obtain actions. */ + acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]); + error = PTR_ERR(acts); + if (IS_ERR(acts)) + goto error_free_flow; + rcu_assign_pointer(flow->sf_acts, acts); - rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, - actions->n_actions, GFP_KERNEL); - rcu_read_unlock(); + /* Put flow in bucket. */ + error = tbl_insert(table, &flow->tbl_node, hash); + if (error) + goto error_free_flow; - kfree(actions); - return err; + reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, ODP_FLOW_CMD_NEW); + } else { + /* We found a matching flow. */ + struct sw_flow_actions *old_acts; + + /* Bail out if we're not allowed to modify an existing flow. + * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL + * because Generic Netlink treats the latter as a dump + * request. We also accept NLM_F_EXCL in case that bug ever + * gets fixed. + */ + error = -EEXIST; + if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW && + info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) + goto error; -error_free_skb: - kfree_skb(skb); -error_free_actions: - kfree(actions); -error: - return err; -} + /* Update actions. */ + flow = flow_cast(flow_node); + old_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + if (a[ODP_FLOW_ATTR_ACTIONS] && + (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) || + memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]), + old_acts->actions_len))) { + struct sw_flow_actions *new_acts; + + new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]); + error = PTR_ERR(new_acts); + if (IS_ERR(new_acts)) + goto error; -static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep) -{ - struct odp_execute execute; + rcu_assign_pointer(flow->sf_acts, new_acts); + flow_deferred_free_acts(old_acts); + } - if (copy_from_user(&execute, executep, sizeof execute)) - return -EFAULT; + reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, ODP_FLOW_CMD_NEW); - return do_execute(dp, &execute); -} + /* Clear stats. */ + if (a[ODP_FLOW_ATTR_CLEAR]) { + spin_lock_bh(&flow->lock); + clear_stats(flow); + spin_unlock_bh(&flow->lock); + } + } -static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) -{ - struct tbl *table = rcu_dereference(dp->table); - struct odp_stats stats; - int i; + if (!IS_ERR(reply)) + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); + else + netlink_set_err(INIT_NET_GENL_SOCK, 0, + dp_flow_multicast_group.id, PTR_ERR(reply)); + return 0; - stats.n_flows = tbl_count(table); - stats.cur_capacity = tbl_n_buckets(table); - stats.max_capacity = TBL_MAX_BUCKETS; - stats.n_ports = dp->n_ports; - stats.max_ports = DP_MAX_PORTS; - stats.max_groups = DP_MAX_GROUPS; - stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; - for_each_possible_cpu(i) { - const struct dp_stats_percpu *s; - s = per_cpu_ptr(dp->stats_percpu, i); - stats.n_frags += s->n_frags; - stats.n_hit += s->n_hit; - stats.n_missed += s->n_missed; - stats.n_lost += s->n_lost; - } - stats.max_miss_queue = DP_MAX_QUEUE_LEN; - stats.max_action_queue = DP_MAX_QUEUE_LEN; - return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0; +error_free_flow: + flow_put(flow); +error: + return error; } -/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ -int dp_min_mtu(const struct datapath *dp) +static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct dp_port *p; - int mtu = 0; + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; + struct sw_flow_key key; + struct tbl_node *flow_node; + struct sk_buff *reply; + struct sw_flow *flow; + struct datapath *dp; + struct tbl *table; + int err; - ASSERT_RTNL(); + if (!a[ODP_FLOW_ATTR_KEY]) + return -EINVAL; + err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]); + if (err) + return err; - list_for_each_entry_rcu (p, &dp->port_list, node) { - int dev_mtu; + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - /* Skip any internal ports, since that's what we're trying to - * set. */ - if (is_internal_vport(p->vport)) - continue; + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp); + if (!flow_node) + return -ENOENT; - dev_mtu = vport_get_mtu(p->vport); - if (!mtu || dev_mtu < mtu) - mtu = dev_mtu; - } + flow = flow_cast(flow_node); + reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); - return mtu ? mtu : ETH_DATA_LEN; + return genlmsg_reply(reply, info); } -/* Sets the MTU of all datapath devices to the minimum of the ports. Must - * be called with RTNL lock. */ -void set_internal_devs_mtu(const struct datapath *dp) +static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct dp_port *p; - int mtu; + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; + struct sw_flow_key key; + struct tbl_node *flow_node; + struct sk_buff *reply; + struct sw_flow *flow; + struct datapath *dp; + struct tbl *table; + int err; - ASSERT_RTNL(); + if (!a[ODP_FLOW_ATTR_KEY]) + return flush_flows(odp_header->dp_ifindex); + err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]); + if (err) + return err; - mtu = dp_min_mtu(dp); + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (is_internal_vport(p->vport)) - vport_set_mtu(p->vport, mtu); - } -} + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp); + if (!flow_node) + return -ENOENT; + flow = flow_cast(flow_node); -static int put_port(const struct dp_port *p, struct odp_port __user *uop) -{ - struct odp_port op; + reply = odp_flow_cmd_alloc_info(flow); + if (!reply) + return -ENOMEM; - memset(&op, 0, sizeof op); + err = tbl_remove(table, flow_node); + if (err) { + kfree_skb(reply); + return err; + } - rcu_read_lock(); - strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname); - rcu_read_unlock(); + err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, + info->snd_seq, 0, ODP_FLOW_CMD_DEL); + BUG_ON(err < 0); - op.port = p->port_no; - op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0; + flow_deferred_free(flow); - return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -static int query_port(struct datapath *dp, struct odp_port __user *uport) +static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - struct odp_port port; - - if (copy_from_user(&port, uport, sizeof port)) - return -EFAULT; - - if (port.devname[0]) { - struct vport *vport; - struct dp_port *dp_port; - int err = 0; - - port.devname[IFNAMSIZ - 1] = '\0'; - - vport_lock(); - rcu_read_lock(); + struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; - vport = vport_locate(port.devname); - if (!vport) { - err = -ENODEV; - goto error_unlock; - } + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - dp_port = vport_get_dp_port(vport); - if (!dp_port || dp_port->dp != dp) { - err = -ENOENT; - goto error_unlock; - } + for (;;) { + struct tbl_node *flow_node; + struct sw_flow *flow; + u32 bucket, obj; - port.port = dp_port->port_no; + bucket = cb->args[0]; + obj = cb->args[1]; + flow_node = tbl_next(get_table_protected(dp), &bucket, &obj); + if (!flow_node) + break; -error_unlock: - rcu_read_unlock(); - vport_unlock(); + flow = flow_cast(flow_node); + if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + ODP_FLOW_CMD_NEW) < 0) + break; - if (err) - return err; - } else { - if (port.port >= DP_MAX_PORTS) - return -EINVAL; - if (!dp->ports[port.port]) - return -ENOENT; + cb->args[0] = bucket; + cb->args[1] = obj; } + return skb->len; +} + +static struct genl_ops dp_flow_genl_ops[] = { + { .cmd = ODP_FLOW_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = odp_flow_cmd_new_or_set + }, + { .cmd = ODP_FLOW_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = odp_flow_cmd_del + }, + { .cmd = ODP_FLOW_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = flow_policy, + .doit = odp_flow_cmd_get, + .dumpit = odp_flow_cmd_dump + }, + { .cmd = ODP_FLOW_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = odp_flow_cmd_new_or_set, + }, +}; - return put_port(dp->ports[port.port], uport); -} +static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, +#endif + [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 }, + [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 }, +}; -static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, - int n_ports) -{ - int idx = 0; - if (n_ports) { - struct dp_port *p; - - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (put_port(p, &uports[idx])) - return -EFAULT; - if (idx++ >= n_ports) - break; - } - } - return idx; -} +static struct genl_family dp_datapath_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct odp_header), + .name = ODP_DATAPATH_FAMILY, + .version = 1, + .maxattr = ODP_DP_ATTR_MAX +}; -static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) -{ - struct odp_portvec pv; - int retval; +static struct genl_multicast_group dp_datapath_multicast_group = { + .name = ODP_DATAPATH_MCGROUP +}; - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; +static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) +{ + struct odp_header *odp_header; + struct nlattr *nla; + int err; - retval = do_list_ports(dp, pv.ports, pv.n_ports); - if (retval < 0) - return retval; + odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, + flags, cmd); + if (!odp_header) + goto error; - return put_user(retval, &upv->n_ports); -} + odp_header->dp_ifindex = dp->dp_ifindex; -/* RCU callback for freeing a dp_port_group */ -static void free_port_group(struct rcu_head *rcu) -{ - struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu); - kfree(g); -} + rcu_read_lock(); + err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp)); + rcu_read_unlock(); + if (err) + goto nla_put_failure; -static int do_set_port_group(struct datapath *dp, u16 __user *ports, - int n_ports, int group) -{ - struct dp_port_group *new_group, *old_group; - int error; + nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats)); + if (!nla) + goto nla_put_failure; + get_dp_stats(dp, nla_data(nla)); - error = -EINVAL; - if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS) - goto error; + NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS, + dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO); - error = -ENOMEM; - new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL); - if (!new_group) - goto error; + if (dp->sflow_probability) + NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability); - new_group->n_ports = n_ports; - error = -EFAULT; - if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports)) - goto error_free; + nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS); + if (!nla) + goto nla_put_failure; + NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS)); + NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION)); + NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE)); + nla_nest_end(skb, nla); - old_group = rcu_dereference(dp->groups[group]); - rcu_assign_pointer(dp->groups[group], new_group); - if (old_group) - call_rcu(&old_group->rcu, free_port_group); - return 0; + return genlmsg_end(skb, odp_header); -error_free: - kfree(new_group); +nla_put_failure: + genlmsg_cancel(skb, odp_header); error: - return error; + return -EMSGSIZE; } -static int set_port_group(struct datapath *dp, - const struct odp_port_group __user *upg) +static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid, + u32 seq, u8 cmd) { - struct odp_port_group pg; + struct sk_buff *skb; + int retval; - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); - return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group); + retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } + return skb; } -static int do_get_port_group(struct datapath *dp, - u16 __user *ports, int n_ports, int group, - u16 __user *n_portsp) +static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1]) { - struct dp_port_group *g; - u16 n_copy; + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - if (group >= DP_MAX_GROUPS) - return -EINVAL; - - g = dp->groups[group]; - n_copy = g ? min_t(int, g->n_ports, n_ports) : 0; - if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16))) - return -EFAULT; - - if (put_user(g ? g->n_ports : 0, n_portsp)) - return -EFAULT; + if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP) + return -EINVAL; + } - return 0; + return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1); } -static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg) +/* Called with genl_mutex and optionally with RTNL lock also. */ +static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) { - struct odp_port_group pg; - - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + struct datapath *dp; - return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &upg->n_ports); -} + if (!a[ODP_DP_ATTR_NAME]) + dp = get_dp(odp_header->dp_ifindex); + else { + struct vport *vport; -static int get_listen_mask(const struct file *f) -{ - return (long)f->private_data; + rcu_read_lock(); + vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME])); + dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL; + rcu_read_unlock(); + } + return dp ? dp : ERR_PTR(-ENODEV); } -static void set_listen_mask(struct file *f, int listen_mask) +/* Called with genl_mutex. */ +static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) { - f->private_data = (void*)(long)listen_mask; + if (a[ODP_DP_ATTR_IPV4_FRAGS]) + dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP; + if (a[ODP_DP_ATTR_SAMPLING]) + dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]); } -static long openvswitch_ioctl(struct file *f, unsigned int cmd, - unsigned long argp) +static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) { - int dp_idx = iminor(f->f_dentry->d_inode); + struct nlattr **a = info->attrs; + struct vport_parms parms; + struct sk_buff *reply; struct datapath *dp; - int drop_frags, listeners, port_no; - unsigned int sflow_probability; + struct vport *vport; int err; - /* Handle commands with special locking requirements up front. */ - switch (cmd) { - case ODP_DP_CREATE: - err = create_dp(dp_idx, (char __user *)argp); - goto exit; + err = -EINVAL; + if (!a[ODP_DP_ATTR_NAME]) + goto err; - case ODP_DP_DESTROY: - err = destroy_dp(dp_idx); - goto exit; + err = odp_dp_cmd_validate(a); + if (err) + goto err; + + rtnl_lock(); + err = -ENODEV; + if (!try_module_get(THIS_MODULE)) + goto err_unlock_rtnl; + + err = -ENOMEM; + dp = kzalloc(sizeof(*dp), GFP_KERNEL); + if (dp == NULL) + goto err_put_module; + INIT_LIST_HEAD(&dp->port_list); + + /* Initialize kobject for bridge. This will be added as + * /sys/class/net//brif later, if sysfs is enabled. */ + dp->ifobj.kset = NULL; + kobject_init(&dp->ifobj, &dp_ktype); + + /* Allocate table. */ + err = -ENOMEM; + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); + if (!dp->table) + goto err_free_dp; + + /* Set up our datapath device. */ + parms.name = nla_data(a[ODP_DP_ATTR_NAME]); + parms.type = ODP_VPORT_TYPE_INTERNAL; + parms.options = NULL; + parms.dp = dp; + parms.port_no = ODPP_LOCAL; + vport = new_vport(&parms); + if (IS_ERR(vport)) { + err = PTR_ERR(vport); + if (err == -EBUSY) + err = -EEXIST; - case ODP_PORT_ATTACH: - err = attach_port(dp_idx, (struct odp_port __user *)argp); - goto exit; + goto err_destroy_table; + } + dp->dp_ifindex = vport_get_ifindex(vport); - case ODP_PORT_DETACH: - err = get_user(port_no, (int __user *)argp); - if (!err) - err = detach_port(dp_idx, port_no); - goto exit; + dp->drop_frags = 0; + dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); + if (!dp->stats_percpu) { + err = -ENOMEM; + goto err_destroy_local_port; + } - case ODP_VPORT_ADD: - err = vport_user_add((struct odp_vport_add __user *)argp); - goto exit; + change_datapath(dp, a); - case ODP_VPORT_MOD: - err = vport_user_mod((struct odp_vport_mod __user *)argp); - goto exit; + reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto err_destroy_local_port; - case ODP_VPORT_DEL: - err = vport_user_del((char __user *)argp); - goto exit; + list_add_tail(&dp->list_node, &dps); + dp_sysfs_add_dp(dp); - case ODP_VPORT_STATS_GET: - err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp); - goto exit; + rtnl_unlock(); - case ODP_VPORT_STATS_SET: - err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp); - goto exit; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; - case ODP_VPORT_ETHER_GET: - err = vport_user_ether_get((struct odp_vport_ether __user *)argp); - goto exit; +err_destroy_local_port: + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); +err_destroy_table: + tbl_destroy(get_table_protected(dp), NULL); +err_free_dp: + kfree(dp); +err_put_module: + module_put(THIS_MODULE); +err_unlock_rtnl: + rtnl_unlock(); +err: + return err; +} - case ODP_VPORT_ETHER_SET: - err = vport_user_ether_set((struct odp_vport_ether __user *)argp); - goto exit; +static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) +{ + struct vport *vport, *next_vport; + struct sk_buff *reply; + struct datapath *dp; + int err; - case ODP_VPORT_MTU_GET: - err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp); + err = odp_dp_cmd_validate(info->attrs); + if (err) goto exit; - case ODP_VPORT_MTU_SET: - err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp); - goto exit; - } + rtnl_lock(); + dp = lookup_datapath(info->userhdr, info->attrs); + err = PTR_ERR(dp); + if (IS_ERR(dp)) + goto exit_unlock; - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto exit; + reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; - switch (cmd) { - case ODP_DP_STATS: - err = get_dp_stats(dp, (struct odp_stats __user *)argp); - break; + list_for_each_entry_safe (vport, next_vport, &dp->port_list, node) + if (vport->port_no != ODPP_LOCAL) + dp_detach_port(vport); - case ODP_GET_DROP_FRAGS: - err = put_user(dp->drop_frags, (int __user *)argp); - break; + dp_sysfs_del_dp(dp); + list_del(&dp->list_node); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); - case ODP_SET_DROP_FRAGS: - err = get_user(drop_frags, (int __user *)argp); - if (err) - break; - err = -EINVAL; - if (drop_frags != 0 && drop_frags != 1) - break; - dp->drop_frags = drop_frags; - err = 0; - break; + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); - case ODP_GET_LISTEN_MASK: - err = put_user(get_listen_mask(f), (int __user *)argp); - break; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + err = 0; - case ODP_SET_LISTEN_MASK: - err = get_user(listeners, (int __user *)argp); - if (err) - break; - err = -EINVAL; - if (listeners & ~ODPL_ALL) - break; - err = 0; - set_listen_mask(f, listeners); - break; - - case ODP_GET_SFLOW_PROBABILITY: - err = put_user(dp->sflow_probability, (unsigned int __user *)argp); - break; - - case ODP_SET_SFLOW_PROBABILITY: - err = get_user(sflow_probability, (unsigned int __user *)argp); - if (!err) - dp->sflow_probability = sflow_probability; - break; - - case ODP_PORT_QUERY: - err = query_port(dp, (struct odp_port __user *)argp); - break; - - case ODP_PORT_LIST: - err = list_ports(dp, (struct odp_portvec __user *)argp); - break; - - case ODP_PORT_GROUP_SET: - err = set_port_group(dp, (struct odp_port_group __user *)argp); - break; - - case ODP_PORT_GROUP_GET: - err = get_port_group(dp, (struct odp_port_group __user *)argp); - break; - - case ODP_FLOW_FLUSH: - err = flush_flows(dp); - break; - - case ODP_FLOW_PUT: - err = put_flow(dp, (struct odp_flow_put __user *)argp); - break; - - case ODP_FLOW_DEL: - err = del_flow(dp, (struct odp_flow __user *)argp); - break; - - case ODP_FLOW_GET: - err = do_flowvec_ioctl(dp, argp, do_query_flows); - break; - - case ODP_FLOW_LIST: - err = do_flowvec_ioctl(dp, argp, do_list_flows); - break; - - case ODP_EXECUTE: - err = execute_packet(dp, (struct odp_execute __user *)argp); - break; - - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); +exit_unlock: + rtnl_unlock(); exit: return err; } -static int dp_has_packet_of_interest(struct datapath *dp, int listeners) +static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) { - int i; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i])) - return 1; - } - return 0; -} + struct sk_buff *reply; + struct datapath *dp; + int err; -#ifdef CONFIG_COMPAT -static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv) -{ - struct compat_odp_portvec pv; - int retval; + err = odp_dp_cmd_validate(info->attrs); + if (err) + return err; + + dp = lookup_datapath(info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; + change_datapath(dp, info->attrs); - retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports); - if (retval < 0) - return retval; + reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW); + if (IS_ERR(reply)) { + err = PTR_ERR(reply); + netlink_set_err(INIT_NET_GENL_SOCK, 0, + dp_datapath_multicast_group.id, err); + return 0; + } - return put_user(retval, &upv->n_ports); + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg) +static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct compat_odp_port_group pg; + struct sk_buff *reply; + struct datapath *dp; + int err; + + err = odp_dp_cmd_validate(info->attrs); + if (err) + return err; - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + dp = lookup_datapath(info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group); + reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); + + return genlmsg_reply(reply, info); } -static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg) +static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - struct compat_odp_port_group pg; + struct datapath *dp; + int skip = cb->args[0]; + int i = 0; - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + list_for_each_entry (dp, &dps, list_node) { + if (i < skip) + continue; + if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + ODP_DP_CMD_NEW) < 0) + break; + i++; + } - return do_get_port_group(dp, compat_ptr(pg.ports), pg.n_ports, - pg.group, &upg->n_ports); -} + cb->args[0] = i; + + return skb->len; +} + +static struct genl_ops dp_datapath_genl_ops[] = { + { .cmd = ODP_DP_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = odp_dp_cmd_new + }, + { .cmd = ODP_DP_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = odp_dp_cmd_del + }, + { .cmd = ODP_DP_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = datapath_policy, + .doit = odp_dp_cmd_get, + .dumpit = odp_dp_cmd_dump + }, + { .cmd = ODP_DP_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = odp_dp_cmd_set, + }, +}; -static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) -{ - compat_uptr_t actions; +static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, + [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) }, + [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN }, +#else + [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) }, + [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN }, +#endif + [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, +}; - if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) || - __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) || - __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) || - __get_user(actions, &compat->actions) || - __get_user(flow->n_actions, &compat->n_actions) || - __get_user(flow->flags, &compat->flags)) - return -EFAULT; +static struct genl_family dp_vport_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct odp_header), + .name = ODP_VPORT_FAMILY, + .version = 1, + .maxattr = ODP_VPORT_ATTR_MAX +}; - flow->actions = compat_ptr(actions); - return 0; -} +static struct genl_multicast_group dp_vport_multicast_group = { + .name = ODP_VPORT_MCGROUP +}; -static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp) +/* Called with RTNL lock or RCU read lock. */ +static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) { - struct odp_flow_stats stats; - struct odp_flow_put fp; - int error; + struct odp_header *odp_header; + struct nlattr *nla; + int ifindex, iflink; + int mtu; + int err; - if (compat_get_flow(&fp.flow, &ufp->flow) || - get_user(fp.flags, &ufp->flags)) - return -EFAULT; + odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, + flags, cmd); + if (!odp_header) + return -EMSGSIZE; - error = do_put_flow(dp, &fp, &stats); - if (error) - return error; + odp_header->dp_ifindex = vport->dp->dp_ifindex; - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; + NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no); + NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport)); + NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport)); - return 0; -} + nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64)); + if (!nla) + goto nla_put_failure; + if (vport_get_stats(vport, nla_data(nla))) + __skb_trim(skb, skb->len - nla->nla_len); -static int compat_answer_query(struct sw_flow *flow, u32 query_flags, - struct timespec time_offset, - struct compat_odp_flow __user *ufp) -{ - compat_uptr_t actions; + NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); - if (get_user(actions, &ufp->actions)) - return -EFAULT; + mtu = vport_get_mtu(vport); + if (mtu) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu); - return do_answer_query(flow, query_flags, time_offset, &ufp->stats, - compat_ptr(actions), &ufp->n_actions); -} + err = vport_get_options(vport, skb); + if (err == -EMSGSIZE) + goto error; -static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp) -{ - struct sw_flow *flow; - struct odp_flow uf; - int error; + ifindex = vport_get_ifindex(vport); + if (ifindex > 0) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex); - if (compat_get_flow(&uf, ufp)) - return -EFAULT; + iflink = vport_get_iflink(vport); + if (iflink > 0) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink); - flow = do_del_flow(dp, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); + return genlmsg_end(skb, odp_header); - error = compat_answer_query(flow, 0, get_time_offset(), ufp); - flow_deferred_free(flow); - return error; +nla_put_failure: + err = -EMSGSIZE; +error: + genlmsg_cancel(skb, odp_header); + return err; } -static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) +/* Called with RTNL lock or RCU read lock. */ +static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid, + u32 seq, u8 cmd) { - struct tbl *table = rcu_dereference(dp->table); - struct timespec time_offset; - u32 i; - - time_offset = get_time_offset(); - - for (i = 0; i < n_flows; i++) { - struct compat_odp_flow __user *ufp = &flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; + struct sk_buff *skb; + int retval; - if (compat_get_flow(&uf, ufp)) - return -EFAULT; - memset(uf.key.reserved, 0, sizeof uf.key.reserved); + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); + if (!skb) + return ERR_PTR(-ENOMEM); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = compat_answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp); - if (error) - return -EFAULT; + retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); } - return n_flows; + return skb; } -struct compat_list_flows_cbdata { - struct compat_odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; - struct timespec time_offset; -}; - -static int compat_list_flow(struct tbl_node *node, void *cbdata_) +static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { - struct sw_flow *flow = flow_cast(node); - struct compat_list_flows_cbdata *cbdata = cbdata_; - struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; - int error; + return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1); +} - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) - return -EFAULT; - error = compat_answer_query(flow, 0, cbdata->time_offset, ufp); - if (error) - return error; +/* Called with RTNL lock or RCU read lock. */ +static struct vport *lookup_vport(struct odp_header *odp_header, + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + struct datapath *dp; + struct vport *vport; - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; + if (a[ODP_VPORT_ATTR_NAME]) { + vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME])); + if (!vport) + return ERR_PTR(-ENODEV); + return vport; + } else if (a[ODP_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + + if (port_no >= DP_MAX_PORTS) + return ERR_PTR(-EFBIG); + + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return ERR_PTR(-ENODEV); + + vport = get_vport_protected(dp, port_no); + if (!vport) + return ERR_PTR(-ENOENT); + return vport; + } else + return ERR_PTR(-EINVAL); +} + +/* Called with RTNL lock. */ +static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + int err = 0; + if (a[ODP_VPORT_ATTR_STATS]) + err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS])); + if (!err && a[ODP_VPORT_ATTR_ADDRESS]) + err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS])); + if (!err && a[ODP_VPORT_ATTR_MTU]) + err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU])); + return err; } -static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) +static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) { - struct compat_list_flows_cbdata cbdata; - int error; + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; + struct vport_parms parms; + struct sk_buff *reply; + struct vport *vport; + struct datapath *dp; + u32 port_no; + int err; - if (!n_flows) - return 0; + err = -EINVAL; + if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE]) + goto exit; - cbdata.uflows = flows; - cbdata.n_flows = n_flows; - cbdata.listed_flows = 0; - cbdata.time_offset = get_time_offset(); + err = odp_vport_cmd_validate(a); + if (err) + goto exit; - error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata); - return error ? error : cbdata.listed_flows; -} + rtnl_lock(); + dp = get_dp(odp_header->dp_ifindex); + err = -ENODEV; + if (!dp) + goto exit_unlock; -static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - struct compat_odp_flow *, - u32 n_flows)) -{ - struct compat_odp_flowvec __user *uflowvec; - struct compat_odp_flow __user *flows; - struct compat_odp_flowvec flowvec; - int retval; + if (a[ODP_VPORT_ATTR_PORT_NO]) { + port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - uflowvec = compat_ptr(argp); - if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) || - copy_from_user(&flowvec, uflowvec, sizeof flowvec)) - return -EFAULT; + err = -EFBIG; + if (port_no >= DP_MAX_PORTS) + goto exit_unlock; - if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow)) - return -EINVAL; + vport = get_vport_protected(dp, port_no); + err = -EBUSY; + if (vport) + goto exit_unlock; + } else { + for (port_no = 1; ; port_no++) { + if (port_no >= DP_MAX_PORTS) { + err = -EFBIG; + goto exit_unlock; + } + vport = get_vport_protected(dp, port_no); + if (!vport) + break; + } + } - flows = compat_ptr(flowvec.flows); - if (!access_ok(VERIFY_WRITE, flows, - flowvec.n_flows * sizeof(struct compat_odp_flow))) - return -EFAULT; + parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]); + parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]); + parms.options = a[ODP_VPORT_ATTR_OPTIONS]; + parms.dp = dp; + parms.port_no = port_no; + + vport = new_vport(&parms); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; + + set_internal_devs_mtu(dp); + dp_sysfs_add_if(vport); + + err = change_vport(vport, a); + if (!err) { + reply = odp_vport_cmd_build_info(vport, info->snd_pid, + info->snd_seq, ODP_VPORT_CMD_NEW); + if (IS_ERR(reply)) + err = PTR_ERR(reply); + } + if (err) { + dp_detach_port(vport); + goto exit_unlock; + } + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - retval = function(dp, flows, flowvec.n_flows); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : put_user(retval, &uflowvec->n_flows)); -} -static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute) -{ - struct odp_execute execute; - compat_uptr_t actions; - compat_uptr_t data; - - if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) || - __get_user(execute.in_port, &uexecute->in_port) || - __get_user(actions, &uexecute->actions) || - __get_user(execute.n_actions, &uexecute->n_actions) || - __get_user(data, &uexecute->data) || - __get_user(execute.length, &uexecute->length)) - return -EFAULT; - - execute.actions = compat_ptr(actions); - execute.data = compat_ptr(data); - - return do_execute(dp, &execute); +exit_unlock: + rtnl_unlock(); +exit: + return err; } -static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) +static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) { - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp; + struct nlattr **a = info->attrs; + struct sk_buff *reply; + struct vport *vport; int err; - switch (cmd) { - case ODP_DP_DESTROY: - case ODP_FLOW_FLUSH: - /* Ioctls that don't need any translation at all. */ - return openvswitch_ioctl(f, cmd, argp); - - case ODP_DP_CREATE: - case ODP_PORT_ATTACH: - case ODP_PORT_DETACH: - case ODP_VPORT_DEL: - case ODP_VPORT_MTU_SET: - case ODP_VPORT_MTU_GET: - case ODP_VPORT_ETHER_SET: - case ODP_VPORT_ETHER_GET: - case ODP_VPORT_STATS_SET: - case ODP_VPORT_STATS_GET: - case ODP_DP_STATS: - case ODP_GET_DROP_FRAGS: - case ODP_SET_DROP_FRAGS: - case ODP_SET_LISTEN_MASK: - case ODP_GET_LISTEN_MASK: - case ODP_SET_SFLOW_PROBABILITY: - case ODP_GET_SFLOW_PROBABILITY: - case ODP_PORT_QUERY: - /* Ioctls that just need their pointer argument extended. */ - return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); - - case ODP_VPORT_ADD32: - return compat_vport_user_add(compat_ptr(argp)); - - case ODP_VPORT_MOD32: - return compat_vport_user_mod(compat_ptr(argp)); + err = odp_vport_cmd_validate(a); + if (err) + goto exit; + + rtnl_lock(); + vport = lookup_vport(info->userhdr, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; + + err = 0; + if (a[ODP_VPORT_ATTR_OPTIONS]) + err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]); + if (!err) + err = change_vport(vport, a); + + reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + ODP_VPORT_CMD_NEW); + if (IS_ERR(reply)) { + err = PTR_ERR(reply); + netlink_set_err(INIT_NET_GENL_SOCK, 0, + dp_vport_multicast_group.id, err); + return 0; } - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto exit; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - switch (cmd) { - case ODP_PORT_LIST32: - err = compat_list_ports(dp, compat_ptr(argp)); - break; +exit_unlock: + rtnl_unlock(); +exit: + return err; +} - case ODP_PORT_GROUP_SET32: - err = compat_set_port_group(dp, compat_ptr(argp)); - break; +static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) +{ + struct nlattr **a = info->attrs; + struct sk_buff *reply; + struct vport *vport; + int err; - case ODP_PORT_GROUP_GET32: - err = compat_get_port_group(dp, compat_ptr(argp)); - break; + err = odp_vport_cmd_validate(a); + if (err) + goto exit; - case ODP_FLOW_PUT32: - err = compat_put_flow(dp, compat_ptr(argp)); - break; + rtnl_lock(); + vport = lookup_vport(info->userhdr, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; - case ODP_FLOW_DEL32: - err = compat_del_flow(dp, compat_ptr(argp)); - break; + if (vport->port_no == ODPP_LOCAL) { + err = -EINVAL; + goto exit_unlock; + } - case ODP_FLOW_GET32: - err = compat_flowvec_ioctl(dp, argp, compat_query_flows); - break; + reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + ODP_VPORT_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; - case ODP_FLOW_LIST32: - err = compat_flowvec_ioctl(dp, argp, compat_list_flows); - break; + err = dp_detach_port(vport); - case ODP_EXECUTE32: - err = compat_execute(dp, compat_ptr(argp)); - break; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); +exit_unlock: + rtnl_unlock(); exit: return err; } -#endif -/* Unfortunately this function is not exported so this is a verbatim copy - * from net/core/datagram.c in 2.6.30. */ -static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, - u8 __user *to, int len, - __wsum *csump) +static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) { - int start = skb_headlen(skb); - int pos = 0; - int i, copy = start - offset; - - /* Copy header. */ - if (copy > 0) { - int err = 0; - if (copy > len) - copy = len; - *csump = csum_and_copy_to_user(skb->data + offset, to, copy, - *csump, &err); - if (err) - goto fault; - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos = copy; - } + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; + struct sk_buff *reply; + struct vport *vport; + int err; - for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { - int end; - - WARN_ON(start > offset + len); - - end = start + skb_shinfo(skb)->frags[i].size; - if ((copy = end - offset) > 0) { - __wsum csum2; - int err = 0; - u8 *vaddr; - skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; - struct page *page = frag->page; - - if (copy > len) - copy = len; - vaddr = kmap(page); - csum2 = csum_and_copy_to_user(vaddr + - frag->page_offset + - offset - start, - to, copy, 0, &err); - kunmap(page); - if (err) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if (!(len -= copy)) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; - } + err = odp_vport_cmd_validate(a); + if (err) + goto exit; - if (skb_shinfo(skb)->frag_list) { - struct sk_buff *list = skb_shinfo(skb)->frag_list; - - for (; list; list=list->next) { - int end; - - WARN_ON(start > offset + len); - - end = start + list->len; - if ((copy = end - offset) > 0) { - __wsum csum2 = 0; - if (copy > len) - copy = len; - if (skb_copy_and_csum_datagram(list, - offset - start, - to, copy, - &csum2)) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; - } - } - if (!len) - return 0; + rcu_read_lock(); + vport = lookup_vport(odp_header, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; + + reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + ODP_VPORT_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; -fault: - return -EFAULT; + err = genlmsg_reply(reply, info); + +exit_unlock: + rcu_read_unlock(); +exit: + return err; } -ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, - loff_t *ppos) +static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - /* XXX is there sufficient synchronization here? */ - int listeners = get_listen_mask(f); - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp = get_dp(dp_idx); - struct sk_buff *skb; - size_t copy_bytes, tot_copy_bytes; + struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; + u32 port_no; int retval; + dp = get_dp(odp_header->dp_ifindex); if (!dp) return -ENODEV; - if (nbytes == 0 || !listeners) - return 0; + rcu_read_lock(); + for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) { + struct vport *vport; - for (;;) { - int i; + vport = get_vport_protected(dp, port_no); + if (!vport) + continue; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i)) { - skb = skb_dequeue(&dp->queues[i]); - if (skb) - goto success; - } - } + if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + ODP_VPORT_CMD_NEW) < 0) + break; + } + rcu_read_unlock(); - if (f->f_flags & O_NONBLOCK) { - retval = -EAGAIN; - goto error; - } + cb->args[0] = port_no; + retval = skb->len; - wait_event_interruptible(dp->waitqueue, - dp_has_packet_of_interest(dp, - listeners)); + return retval; +} - if (signal_pending(current)) { - retval = -ERESTARTSYS; - goto error; - } - } -success: - copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes); - - retval = 0; - if (skb->ip_summed == CHECKSUM_PARTIAL) { - if (copy_bytes == skb->len) { - __wsum csum = 0; - unsigned int csum_start, csum_offset; - -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - csum_start = skb->csum_start - skb_headroom(skb); - csum_offset = skb->csum_offset; -#else - csum_start = skb_transport_header(skb) - skb->data; - csum_offset = skb->csum; -#endif - BUG_ON(csum_start >= skb_headlen(skb)); - retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start, - copy_bytes - csum_start, &csum); - if (!retval) { - __sum16 __user *csump; +static struct genl_ops dp_vport_genl_ops[] = { + { .cmd = ODP_VPORT_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = odp_vport_cmd_new + }, + { .cmd = ODP_VPORT_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = odp_vport_cmd_del + }, + { .cmd = ODP_VPORT_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = vport_policy, + .doit = odp_vport_cmd_get, + .dumpit = odp_vport_cmd_dump + }, + { .cmd = ODP_VPORT_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = odp_vport_cmd_set, + }, +}; - copy_bytes = csum_start; - csump = (__sum16 __user *)(buf + csum_start + csum_offset); +struct genl_family_and_ops { + struct genl_family *family; + struct genl_ops *ops; + int n_ops; + struct genl_multicast_group *group; +}; - BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes); - put_user(csum_fold(csum), csump); - } - } else - retval = skb_checksum_help(skb); - } +static const struct genl_family_and_ops dp_genl_families[] = { + { &dp_datapath_genl_family, + dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops), + &dp_datapath_multicast_group }, + { &dp_vport_genl_family, + dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops), + &dp_vport_multicast_group }, + { &dp_flow_genl_family, + dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops), + &dp_flow_multicast_group }, + { &dp_packet_genl_family, + dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), + NULL }, +}; - if (!retval) { - struct iovec __user iov; +static void dp_unregister_genl(int n_families) +{ + int i; - iov.iov_base = buf; - iov.iov_len = copy_bytes; - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); + for (i = 0; i < n_families; i++) { + genl_unregister_family(dp_genl_families[i].family); } +} - if (!retval) - retval = tot_copy_bytes; +static int dp_register_genl(void) +{ + int n_registered; + int err; + int i; - kfree_skb(skb); + n_registered = 0; + for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { + const struct genl_family_and_ops *f = &dp_genl_families[i]; -error: - return retval; -} + err = genl_register_family_with_ops(f->family, f->ops, + f->n_ops); + if (err) + goto error; + n_registered++; -static unsigned int openvswitch_poll(struct file *file, poll_table *wait) -{ - /* XXX is there sufficient synchronization here? */ - int dp_idx = iminor(file->f_dentry->d_inode); - struct datapath *dp = get_dp(dp_idx); - unsigned int mask; - - if (dp) { - mask = 0; - poll_wait(file, &dp->waitqueue, wait); - if (dp_has_packet_of_interest(dp, get_listen_mask(file))) - mask |= POLLIN | POLLRDNORM; - } else { - mask = POLLIN | POLLRDNORM | POLLHUP; + if (f->group) { + err = genl_register_mc_group(f->family, f->group); + if (err) + goto error; + } } - return mask; -} -struct file_operations openvswitch_fops = { - /* XXX .aio_read = openvswitch_aio_read, */ - .read = openvswitch_read, - .poll = openvswitch_poll, - .unlocked_ioctl = openvswitch_ioctl, -#ifdef CONFIG_COMPAT - .compat_ioctl = openvswitch_compat_ioctl, -#endif - /* XXX .fasync = openvswitch_fasync, */ -}; + err = packet_register_mc_groups(); + if (err) + goto error; + return 0; -static int major; +error: + dp_unregister_genl(n_registered); + return err; +} static int __init dp_init(void) { @@ -2305,7 +2042,7 @@ static int __init dp_init(void) if (err) goto error_vport_exit; - major = register_chrdev(0, "openvswitch", &openvswitch_fops); + err = dp_register_genl(); if (err < 0) goto error_unreg_notifier; @@ -2324,7 +2061,7 @@ error: static void dp_cleanup(void) { rcu_barrier(); - unregister_chrdev(major, "openvswitch"); + dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); unregister_netdevice_notifier(&dp_device_notifier); vport_exit(); flow_exit();