X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=87056cf8cafde32b7df541fcffa91c81123e250b;hb=b3e0370a37daeecc44ddaf03bbd98c2e8ff4513d;hp=ea998edce2b14bcdc83a302e9396526e14e8a95c;hpb=d656937779f5b987ec021c21189a0deab29fd64d;p=openvswitch diff --git a/datapath/datapath.c b/datapath/datapath.c index ea998edc..87056cf8 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -12,14 +12,15 @@ #include #include -#include #include #include #include #include +#include #include #include #include +#include #include #include #include @@ -38,77 +39,109 @@ #include #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 "loop_counter.h" -#include "odp-compat.h" -#include "table.h" +#include "vlan.h" +#include "tunnel.h" #include "vport-internal_dev.h" +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \ + LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0) +#error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch. +#endif + 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. + * + * Writes to other state (flow table modifications, set miscellaneous datapath + * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside + * genl_mutex. * - * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL - * lock first. + * Reads are protected by RCU. * - * It is safe to access the datapath and vport structures with just - * dp_mutex. + * 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. */ -static struct datapath __rcu *dps[256]; -static DEFINE_MUTEX(dp_mutex); + +/* Global list of datapaths to enable dumping them all out. + * Protected by genl_mutex. + */ +static LIST_HEAD(dps); static struct vport *new_vport(const struct vport_parms *); +static int queue_gso_packets(int dp_ifindex, struct sk_buff *, + const struct dp_upcall_info *); +static int queue_userspace_packet(int dp_ifindex, 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 >= ARRAY_SIZE(dps)) - return NULL; - return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() || - lockdep_is_held(&dp_mutex)); -} -EXPORT_SYMBOL_GPL(get_dp); + struct datapath *dp = NULL; + struct net_device *dev; -static struct datapath *get_dp_locked(int dp_idx) -{ - struct datapath *dp; + 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(); - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - if (dp) - mutex_lock(&dp->mutex); - mutex_unlock(&dp_mutex); return dp; } +EXPORT_SYMBOL_GPL(get_dp); -static struct tbl *get_table_protected(struct datapath *dp) +/* Must be called with genl_mutex. */ +static struct flow_table *get_table_protected(struct datapath *dp) { - return rcu_dereference_protected(dp->table, - lockdep_is_held(&dp->mutex)); + return rcu_dereference_protected(dp->table, lockdep_genl_is_held()); } +/* 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_protected(dp->ports[port_no], - lockdep_is_held(&dp->mutex)); + 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(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); + return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL])); +} + +static int get_dpifindex(struct datapath *dp) +{ + struct vport *local; + int ifindex; + + rcu_read_lock(); + + local = get_vport_protected(dp, OVSP_LOCAL); + if (local) + ifindex = vport_get_ifindex(local); + else + ifindex = 0; + + rcu_read_unlock(); + + return ifindex; } static inline size_t br_nlmsg_size(void) @@ -118,26 +151,22 @@ static inline size_t br_nlmsg_size(void) + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */ + nla_total_size(4) /* IFLA_MASTER */ + nla_total_size(4) /* IFLA_MTU */ - + nla_total_size(4) /* IFLA_LINK */ + nla_total_size(1); /* IFLA_OPERSTATE */ } +/* Caller must hold RTNL lock. */ static int dp_fill_ifinfo(struct sk_buff *skb, const struct vport *port, int event, unsigned int flags) { struct datapath *dp = port->dp; int ifindex = vport_get_ifindex(port); - int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; struct nlmsghdr *nlh; if (ifindex < 0) return ifindex; - if (iflink < 0) - return iflink; - nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags); if (nlh == NULL) return -EMSGSIZE; @@ -151,8 +180,7 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->ifi_change = 0; 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_MASTER, get_dpifindex(dp)); NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port)); #ifdef IFLA_OPERSTATE NLA_PUT_U8(skb, IFLA_OPERSTATE, @@ -163,9 +191,6 @@ static int dp_fill_ifinfo(struct sk_buff *skb, NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port)); - if (ifindex != iflink) - NLA_PUT_U32(skb, IFLA_LINK,iflink); - return nlmsg_end(skb, nlh); nla_put_failure: @@ -173,6 +198,7 @@ nla_put_failure: return -EMSGSIZE; } +/* Caller must hold RTNL lock. */ static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; @@ -209,165 +235,98 @@ static struct kobj_type dp_ktype = { static void destroy_dp_rcu(struct rcu_head *rcu) { struct datapath *dp = container_of(rcu, struct datapath, rcu); - int i; - - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_purge(&dp->queues[i]); - tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl); + flow_tbl_destroy(dp->table); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); } -/* Caller must hold RTNL, dp_mutex, and dp->mutex. */ -static void destroy_dp(struct datapath *dp) -{ - struct vport *p, *n; - - list_for_each_entry_safe (p, n, &dp->port_list, node) - if (p->port_no != ODPP_LOCAL) - dp_detach_port(p); - - dp_sysfs_del_dp(dp); - rcu_assign_pointer(dps[dp->dp_idx], NULL); - dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); - - mutex_unlock(&dp->mutex); - call_rcu(&dp->rcu, destroy_dp_rcu); - module_put(THIS_MODULE); -} - -/* Called with RTNL lock and dp->mutex. */ +/* Called with RTNL lock and genl_lock. */ static struct vport *new_vport(const struct vport_parms *parms) { struct vport *vport; - vport_lock(); vport = vport_add(parms); if (!IS_ERR(vport)) { struct datapath *dp = parms->dp; rcu_assign_pointer(dp->ports[parms->port_no], vport); - list_add_rcu(&vport->node, &dp->port_list); + list_add(&vport->node, &dp->port_list); dp_ifinfo_notify(RTM_NEWLINK, vport); } - vport_unlock(); return vport; } -int dp_detach_port(struct vport *p) +/* Called with RTNL lock. */ +void dp_detach_port(struct vport *p) { - int err; - ASSERT_RTNL(); - if (p->port_no != ODPP_LOCAL) + if (p->port_no != OVSP_LOCAL) dp_sysfs_del_if(p); dp_ifinfo_notify(RTM_DELLINK, p); /* First drop references to device. */ - list_del_rcu(&p->node); + list_del(&p->node); rcu_assign_pointer(p->dp->ports[p->port_no], NULL); /* Then destroy it. */ - vport_lock(); - err = vport_del(p); - vport_unlock(); - - return err; + vport_del(p); } /* Must be called with rcu_read_lock. */ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) { struct datapath *dp = p->dp; + struct sw_flow *flow; struct dp_stats_percpu *stats; - int stats_counter_off; - struct sw_flow_actions *acts; - struct loop_counter *loop; + u64 *stats_counter; int error; + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); OVS_CB(skb)->vport = p; if (!OVS_CB(skb)->flow) { struct sw_flow_key key; - struct tbl_node *flow_node; - bool is_frag; + int key_len; /* Extract flow from 'skb' into 'key'. */ - error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag); + error = flow_extract(skb, p->port_no, &key, &key_len); if (unlikely(error)) { kfree_skb(skb); return; } - 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)) { + flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len); + if (unlikely(!flow)) { struct dp_upcall_info upcall; - upcall.type = _ODPL_MISS_NR; + upcall.cmd = OVS_PACKET_CMD_MISS; upcall.key = &key; - upcall.userdata = 0; - upcall.sample_pool = 0; - upcall.actions = NULL; - upcall.actions_len = 0; + upcall.userdata = NULL; + upcall.pid = p->upcall_pid; dp_upcall(dp, skb, &upcall); - stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + consume_skb(skb); + stats_counter = &stats->n_missed; goto out; } - OVS_CB(skb)->flow = flow_cast(flow_node); + OVS_CB(skb)->flow = flow; } - stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + stats_counter = &stats->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(); + execute_actions(dp, skb); 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))++; + (*stats_counter)++; write_seqcount_end(&stats->seqlock); - - local_bh_enable(); } static void copy_and_csum_skb(struct sk_buff *skb, void *to) @@ -377,7 +336,6 @@ static void copy_and_csum_skb(struct sk_buff *skb, void *to) get_skb_csum_pointers(skb, &csum_start, &csum_offset); csum_start -= skb_headroom(skb); - BUG_ON(csum_start >= skb_headlen(skb)); skb_copy_bits(skb, 0, to, csum_start); @@ -386,235 +344,360 @@ static void copy_and_csum_skb(struct sk_buff *skb, void *to) *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum); } -/* 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 datapath *dp, struct sk_buff *skb, - const struct dp_upcall_info *upcall_info) +static struct genl_family dp_packet_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_PACKET_FAMILY, + .version = OVS_PACKET_VERSION, + .maxattr = OVS_PACKET_ATTR_MAX +}; + +int dp_upcall(struct datapath *dp, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - struct sk_buff *nskb; - int port_no; + struct dp_stats_percpu *stats; + int dp_ifindex; int err; - if (OVS_CB(skb)->vport) - port_no = OVS_CB(skb)->vport->port_no; + if (upcall_info->pid == 0) { + err = -ENOTCONN; + goto err; + } + + dp_ifindex = get_dpifindex(dp); + if (!dp_ifindex) { + err = -ENODEV; + goto err; + } + + forward_ip_summed(skb, true); + + if (!skb_is_gso(skb)) + err = queue_userspace_packet(dp_ifindex, skb, upcall_info); else - port_no = ODPP_LOCAL; + err = queue_gso_packets(dp_ifindex, skb, upcall_info); + if (err) + goto err; - do { - struct odp_packet *upcall; - struct sk_buff *user_skb; /* to be queued to userspace */ - struct nlattr *nla; - unsigned int len; + return 0; - nskb = skb->next; - skb->next = NULL; - - len = sizeof(struct odp_packet); - len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */ - 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 = alloc_skb(len, GFP_ATOMIC); - if (!user_skb) - goto err_kfree_skbs; - - upcall = (struct odp_packet *)__skb_put(user_skb, sizeof(*upcall)); - upcall->dp_idx = dp->dp_idx; - - nla_put_u32(user_skb, ODP_PACKET_ATTR_TYPE, upcall_info->type); - - nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY); - flow_to_nlattrs(upcall_info->key, user_skb); - nla_nest_end(user_skb, nla); - - 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; - - 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); - } +err: + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); - 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); + write_seqcount_begin(&stats->seqlock); + stats->n_lost++; + write_seqcount_end(&stats->seqlock); - upcall->len = user_skb->len; - skb_queue_tail(&dp->queues[upcall_info->type], user_skb); + return err; +} - kfree_skb(skb); - skb = nskb; - } while (skb); - return 0; +static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) +{ + struct dp_upcall_info later_info; + struct sw_flow_key later_key; + struct sk_buff *segs, *nskb; + int err; + + segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); + if (IS_ERR(skb)) + return PTR_ERR(skb); + + /* Queue all of the segments. */ + skb = segs; + do { + err = queue_userspace_packet(dp_ifindex, skb, upcall_info); + if (err) + break; + + if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) { + /* The initial flow key extracted by flow_extract() in + * this case is for a first fragment, so we need to + * properly mark later fragments. + */ + later_key = *upcall_info->key; + later_key.ip.tos_frag &= ~OVS_FRAG_TYPE_MASK; + later_key.ip.tos_frag |= OVS_FRAG_TYPE_LATER; + + later_info = *upcall_info; + later_info.key = &later_key; + upcall_info = &later_info; + } + } while ((skb = skb->next)); -err_kfree_skbs: - kfree_skb(skb); - while ((skb = nskb) != NULL) { + /* Free all of the segments. */ + skb = segs; + do { nskb = skb->next; - kfree_skb(skb); - } + if (err) + kfree_skb(skb); + else + consume_skb(skb); + } while ((skb = nskb)); return err; } -int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info) +static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - struct dp_stats_percpu *stats; - struct sk_buff_head *queue; + struct ovs_header *upcall; + struct sk_buff *user_skb; /* to be queued to userspace */ + struct nlattr *nla; + unsigned int len; int err; - WARN_ON_ONCE(skb_shared(skb)); - BUG_ON(upcall_info->type >= DP_N_QUEUES); - - queue = &dp->queues[upcall_info->type]; - err = -ENOBUFS; - if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) - goto err_kfree_skb; + err = vlan_deaccel_tag(skb); + if (unlikely(err)) + return err; - forward_ip_summed(skb); + if (nla_attr_size(skb->len) > USHRT_MAX) + return -EFBIG; - err = vswitch_skb_checksum_setup(skb); - if (err) - goto err_kfree_skb; + len = sizeof(struct ovs_header); + len += nla_total_size(skb->len); + len += nla_total_size(FLOW_BUFSIZE); + if (upcall_info->cmd == OVS_PACKET_CMD_ACTION) + len += nla_total_size(8); - /* Break apart GSO packets into their component pieces. Otherwise - * 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); - - kfree_skb(skb); - skb = nskb; - if (IS_ERR(skb)) { - err = PTR_ERR(skb); - goto err; - } - } + user_skb = genlmsg_new(len, GFP_ATOMIC); + if (!user_skb) + return -ENOMEM; - err = queue_control_packets(dp, skb, upcall_info); - wake_up_interruptible(&dp->waitqueue); - return err; + upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, + 0, upcall_info->cmd); + upcall->dp_ifindex = dp_ifindex; -err_kfree_skb: - kfree_skb(skb); -err: - local_bh_disable(); - stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY); + flow_to_nlattrs(upcall_info->key, user_skb); + nla_nest_end(user_skb, nla); - write_seqcount_begin(&stats->seqlock); - stats->n_lost++; - write_seqcount_end(&stats->seqlock); + if (upcall_info->userdata) + nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, + nla_get_u64(upcall_info->userdata)); - local_bh_enable(); + nla = __nla_reserve(user_skb, OVS_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); - return err; + return genlmsg_unicast(&init_net, user_skb, upcall_info->pid); } -static int flush_flows(int dp_idx) +/* Called with genl_mutex. */ +static int flush_flows(int dp_ifindex) { - struct tbl *old_table; - struct tbl *new_table; + struct flow_table *old_table; + struct flow_table *new_table; struct datapath *dp; - int err; - dp = get_dp_locked(dp_idx); - err = -ENODEV; + dp = get_dp(dp_ifindex); if (!dp) - goto exit; + return -ENODEV; old_table = get_table_protected(dp); - new_table = tbl_create(TBL_MIN_BUCKETS); - err = -ENOMEM; + new_table = flow_tbl_alloc(TBL_MIN_BUCKETS); if (!new_table) - goto exit_unlock; + return -ENOMEM; rcu_assign_pointer(dp->table, new_table); - tbl_deferred_destroy(old_table, flow_free_tbl); - - err = 0; - -exit_unlock: - mutex_unlock(&dp->mutex); -exit: - return err; + flow_tbl_deferred_destroy(old_table); + return 0; } -static int validate_actions(const struct nlattr *actions, u32 actions_len) +static int validate_actions(const struct nlattr *attr, + const struct sw_flow_key *key, int depth); + +static int validate_sample(const struct nlattr *attr, + const struct sw_flow_key *key, int depth) { + const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1]; + const struct nlattr *probability, *actions; const struct nlattr *a; int rem; - nla_for_each_attr(a, actions, actions_len, rem) { - static const u32 action_lens[ODPAT_MAX + 1] = { - [ODPAT_OUTPUT] = 4, - [ODPAT_CONTROLLER] = 8, - [ODPAT_SET_DL_TCI] = 2, - [ODPAT_STRIP_VLAN] = 0, - [ODPAT_SET_DL_SRC] = ETH_ALEN, - [ODPAT_SET_DL_DST] = ETH_ALEN, - [ODPAT_SET_NW_SRC] = 4, - [ODPAT_SET_NW_DST] = 4, - [ODPAT_SET_NW_TOS] = 1, - [ODPAT_SET_TP_SRC] = 2, - [ODPAT_SET_TP_DST] = 2, - [ODPAT_SET_TUNNEL] = 8, - [ODPAT_SET_PRIORITY] = 4, - [ODPAT_POP_PRIORITY] = 0, - [ODPAT_DROP_SPOOFED_ARP] = 0, + memset(attrs, 0, sizeof(attrs)); + nla_for_each_nested (a, attr, rem) { + int type = nla_type(a); + if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type]) + return -EINVAL; + attrs[type] = a; + } + if (rem) + return -EINVAL; + + probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY]; + if (!probability || nla_len(probability) != sizeof(u32)) + return -EINVAL; + + actions = attrs[OVS_SAMPLE_ATTR_ACTIONS]; + if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) + return -EINVAL; + return validate_actions(actions, key, depth + 1); +} + +static int validate_action_key(const struct nlattr *a, + const struct sw_flow_key *flow_key) +{ + int act_type = nla_type(a); + const struct nlattr *ovs_key = nla_data(a); + int key_type = nla_type(ovs_key); + + /* There can be only one key in a action */ + if (nla_total_size(nla_len(ovs_key)) != nla_len(a)) + return -EINVAL; + + if (key_type > OVS_KEY_ATTR_MAX || + nla_len(ovs_key) != ovs_key_lens[key_type]) + return -EINVAL; + +#define ACTION(act, key) (((act) << 8) | (key)) + + switch(ACTION(act_type, key_type)) { + const struct ovs_key_ipv4 *ipv4_key; + const struct ovs_key_8021q *q_key; + + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_PRIORITY): + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID): + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET): + break; + + case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q): + q_key = nla_data(ovs_key); + if (q_key->q_tpid != htons(ETH_P_8021Q)) + return -EINVAL; + + if (q_key->q_tci & htons(VLAN_TAG_PRESENT)) + return -EINVAL; + break; + + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4): + if (flow_key->eth.type != htons(ETH_P_IP)) + return -EINVAL; + + if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst) + return -EINVAL; + + ipv4_key = nla_data(ovs_key); + if (ipv4_key->ipv4_proto != flow_key->ip.proto) + return -EINVAL; + + if (ipv4_key->ipv4_tos & INET_ECN_MASK) + return -EINVAL; + + if (ipv4_key->ipv4_frag != + (flow_key->ip.tos_frag & OVS_FRAG_TYPE_MASK)) + return -EINVAL; + + break; + + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP): + if (flow_key->ip.proto != IPPROTO_TCP) + return -EINVAL; + + if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst) + return -EINVAL; + + break; + + case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP): + if (flow_key->ip.proto != IPPROTO_UDP) + return -EINVAL; + + if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst) + return -EINVAL; + break; + + default: + return -EINVAL; + } +#undef ACTION + return 0; +} + +static int validate_userspace(const struct nlattr *attr) +{ + static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = + { + [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 }, + [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 }, + }; + struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1]; + int error; + + error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy); + if (error) + return error; + + if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID])) + return -EINVAL; + + return 0; +} + +static int validate_actions(const struct nlattr *attr, + const struct sw_flow_key *key, int depth) +{ + const struct nlattr *a; + int rem, err; + + if (depth >= SAMPLE_ACTION_DEPTH) + return -EOVERFLOW; + + nla_for_each_nested(a, attr, rem) { + /* Expected argument lengths, (u32)-1 for variable length. */ + static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = { + [OVS_ACTION_ATTR_OUTPUT] = 4, + [OVS_ACTION_ATTR_USERSPACE] = (u32)-1, + [OVS_ACTION_ATTR_PUSH] = (u32)-1, + [OVS_ACTION_ATTR_POP] = 2, + [OVS_ACTION_ATTR_SET] = (u32)-1, + [OVS_ACTION_ATTR_SAMPLE] = (u32)-1 }; int type = nla_type(a); - if (type > ODPAT_MAX || nla_len(a) != action_lens[type]) + if (type > OVS_ACTION_ATTR_MAX || + (action_lens[type] != nla_len(a) && + action_lens[type] != (u32)-1)) return -EINVAL; switch (type) { - case ODPAT_UNSPEC: + case OVS_ACTION_ATTR_UNSPEC: return -EINVAL; - case ODPAT_CONTROLLER: - case ODPAT_STRIP_VLAN: - case ODPAT_SET_DL_SRC: - case ODPAT_SET_DL_DST: - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - case ODPAT_SET_TUNNEL: - case ODPAT_SET_PRIORITY: - case ODPAT_POP_PRIORITY: - case ODPAT_DROP_SPOOFED_ARP: - /* No validation needed. */ + case OVS_ACTION_ATTR_USERSPACE: + err = validate_userspace(a); + if (err) + return err; break; - case ODPAT_OUTPUT: + case OVS_ACTION_ATTR_OUTPUT: if (nla_get_u32(a) >= DP_MAX_PORTS) return -EINVAL; break; - case ODPAT_SET_DL_TCI: - if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) + + case OVS_ACTION_ATTR_POP: + if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q) return -EINVAL; break; - case ODPAT_SET_NW_TOS: - if (nla_get_u8(a) & INET_ECN_MASK) - return -EINVAL; + case OVS_ACTION_ATTR_SET: + case OVS_ACTION_ATTR_PUSH: + err = validate_action_key(a, key); + if (err) + return err; + break; + + case OVS_ACTION_ATTR_SAMPLE: + err = validate_sample(a, key, depth); + if (err) + return err; break; default: - return -EOPNOTSUPP; + return -EINVAL; } } @@ -624,27 +707,6 @@ static int validate_actions(const struct nlattr *actions, u32 actions_len) return 0; } -struct dp_flowcmd { - u32 nlmsg_flags; - u32 dp_idx; - u32 total_len; - struct sw_flow_key key; - const struct nlattr *actions; - u32 actions_len; - bool clear; - u64 state; -}; - -static struct sw_flow_actions *get_actions(const struct dp_flowcmd *flowcmd) -{ - struct sw_flow_actions *actions; - - actions = flow_actions_alloc(flowcmd->actions_len); - if (!IS_ERR(actions) && flowcmd->actions_len) - memcpy(actions->actions, flowcmd->actions, flowcmd->actions_len); - return actions; -} - static void clear_stats(struct sw_flow *flow) { flow->used = 0; @@ -653,113 +715,127 @@ static void clear_stats(struct sw_flow *flow) flow->byte_count = 0; } -static int expand_table(struct datapath *dp) -{ - struct tbl *old_table = get_table_protected(dp); - struct tbl *new_table; - - new_table = tbl_expand(old_table); - if (IS_ERR(new_table)) - return PTR_ERR(new_table); - - rcu_assign_pointer(dp->table, new_table); - tbl_deferred_destroy(old_table, NULL); - - return 0; -} - -static int do_execute(struct datapath *dp, const struct odp_execute *execute) +static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) { - struct sw_flow_key key; - struct sk_buff *skb; - struct sw_flow_actions *actions; + struct ovs_header *ovs_header = info->userhdr; + struct nlattr **a = info->attrs; + struct sw_flow_actions *acts; + struct sk_buff *packet; + struct sw_flow *flow; + struct datapath *dp; struct ethhdr *eth; - bool is_frag; + int len; int err; + int key_len; err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) - goto error; - - actions = flow_actions_alloc(execute->actions_len); - if (IS_ERR(actions)) { - err = PTR_ERR(actions); - goto error; - } - - err = -EFAULT; - if (copy_from_user(actions->actions, - (struct nlattr __user __force *)execute->actions, execute->actions_len)) - goto error_free_actions; - - err = validate_actions(actions->actions, execute->actions_len); - if (err) - goto error_free_actions; + if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || + !a[OVS_PACKET_ATTR_ACTIONS] || + nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN) + goto err; + len = nla_len(a[OVS_PACKET_ATTR_PACKET]); + packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); err = -ENOMEM; - skb = alloc_skb(execute->length, GFP_KERNEL); - if (!skb) - goto error_free_actions; + if (!packet) + goto err; + skb_reserve(packet, NET_IP_ALIGN); - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), - (const void __user __force *)execute->data, - execute->length)) - goto error_free_skb; + memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len); - skb_reset_mac_header(skb); - eth = eth_hdr(skb); + skb_reset_mac_header(packet); + eth = eth_hdr(packet); /* 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; + packet->protocol = eth->h_proto; else - skb->protocol = htons(ETH_P_802_2); + packet->protocol = htons(ETH_P_802_2); - err = flow_extract(skb, -1, &key, &is_frag); + /* Build an sw_flow for sending this packet. */ + flow = flow_alloc(); + err = PTR_ERR(flow); + if (IS_ERR(flow)) + goto err_kfree_skb; + + err = flow_extract(packet, -1, &flow->key, &key_len); if (err) - goto error_free_skb; + goto err_flow_put; - rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); - rcu_read_unlock(); + err = flow_metadata_from_nlattrs(&flow->key.phy.priority, + &flow->key.phy.in_port, + &flow->key.phy.tun_id, + a[OVS_PACKET_ATTR_KEY]); + if (err) + goto err_flow_put; - kfree(actions); - return err; + err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0); + if (err) + goto err_flow_put; -error_free_skb: - kfree_skb(skb); -error_free_actions: - kfree(actions); -error: - return err; -} + flow->hash = flow_hash(&flow->key, key_len); -static int execute_packet(const struct odp_execute __user *executep) -{ - struct odp_execute execute; - struct datapath *dp; - int error; + acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]); + err = PTR_ERR(acts); + if (IS_ERR(acts)) + goto err_flow_put; + rcu_assign_pointer(flow->sf_acts, acts); - if (copy_from_user(&execute, executep, sizeof(execute))) - return -EFAULT; + OVS_CB(packet)->flow = flow; + packet->priority = flow->key.phy.priority; - dp = get_dp_locked(execute.dp_idx); + rcu_read_lock(); + dp = get_dp(ovs_header->dp_ifindex); + err = -ENODEV; if (!dp) - return -ENODEV; - error = do_execute(dp, &execute); - mutex_unlock(&dp->mutex); + goto err_unlock; - return error; -} + if (flow->key.phy.in_port < DP_MAX_PORTS) + OVS_CB(packet)->vport = get_vport_protected(dp, + flow->key.phy.in_port); -static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) -{ - int i; + local_bh_disable(); + err = execute_actions(dp, packet); + local_bh_enable(); + rcu_read_unlock(); + + flow_put(flow); + return err; + +err_unlock: + rcu_read_unlock(); +err_flow_put: + flow_put(flow); +err_kfree_skb: + kfree_skb(packet); +err: + return err; +} + +static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { + [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, + [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; + +static struct genl_ops dp_packet_genl_ops[] = { + { .cmd = OVS_PACKET_CMD_EXECUTE, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = packet_policy, + .doit = ovs_packet_cmd_execute + } +}; + +static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats) +{ + int i; + struct flow_table *table = get_table_protected(dp); + + stats->n_flows = flow_tbl_count(table); - stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0; + 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; @@ -772,108 +848,58 @@ static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) local_stats = *percpu_stats; } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); - 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; } } -/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ -int dp_min_mtu(const struct datapath *dp) -{ - struct vport *p; - int mtu = 0; - - ASSERT_RTNL(); - - list_for_each_entry_rcu (p, &dp->port_list, node) { - int dev_mtu; - - /* Skip any internal ports, since that's what we're trying to - * set. */ - if (is_internal_vport(p)) - continue; - - dev_mtu = vport_get_mtu(p); - if (!mtu || dev_mtu < mtu) - mtu = dev_mtu; - } - - return mtu ? mtu : ETH_DATA_LEN; -} - -/* 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) -{ - struct vport *p; - int mtu; - - ASSERT_RTNL(); - - mtu = dp_min_mtu(dp); - - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (is_internal_vport(p)) - vport_set_mtu(p, mtu); - } -} - -static int get_listen_mask(const struct file *f) -{ - return (long)f->private_data; -} +static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { + [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, + [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, + [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, +}; -static void set_listen_mask(struct file *f, int listen_mask) -{ - f->private_data = (void*)(long)listen_mask; -} +static struct genl_family dp_flow_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_FLOW_FAMILY, + .version = OVS_FLOW_VERSION, + .maxattr = OVS_FLOW_ATTR_MAX +}; -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 }, - [ODP_FLOW_ATTR_STATE] = { .type = NLA_U64 }, +static struct genl_multicast_group dp_flow_multicast_group = { + .name = OVS_FLOW_MCGROUP }; -static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, - struct sw_flow *flow, u32 total_len, u64 state) +/* Called with genl_lock. */ +static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, + struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd) { + const int skb_orig_len = skb->len; const struct sw_flow_actions *sf_acts; - struct odp_flow_stats stats; - struct odp_flow *odp_flow; - struct sk_buff *skb; + struct ovs_flow_stats stats; + struct ovs_header *ovs_header; struct nlattr *nla; unsigned long used; u8 tcp_flags; int err; sf_acts = rcu_dereference_protected(flow->sf_acts, - lockdep_is_held(&dp->mutex)); + lockdep_genl_is_held()); - skb = alloc_skb(128 + FLOW_BUFSIZE + sf_acts->actions_len, GFP_KERNEL); - err = -ENOMEM; - if (!skb) - goto exit; + ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd); + if (!ovs_header) + return -EMSGSIZE; - rcu_read_lock(); - odp_flow = (struct odp_flow*)__skb_put(skb, sizeof(struct odp_flow)); - odp_flow->dp_idx = dp->dp_idx; - odp_flow->total_len = total_len; + ovs_header->dp_ifindex = get_dpifindex(dp); - nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY); + nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY); if (!nla) goto nla_put_failure; err = flow_to_nlattrs(&flow->key, skb); if (err) - goto exit_unlock; - nla_nest_end(skb, nla); - - nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS); - if (!nla || skb_tailroom(skb) < sf_acts->actions_len) - goto nla_put_failure; - memcpy(__skb_put(skb, sf_acts->actions_len), sf_acts->actions, sf_acts->actions_len); + goto error; nla_nest_end(skb, nla); spin_lock_bh(&flow->lock); @@ -884,165 +910,148 @@ static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, spin_unlock_bh(&flow->lock); if (used) - NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, used); + NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used)); if (stats.n_packets) - NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats); + NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats); if (tcp_flags) - NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags); - - if (state) - NLA_PUT_U64(skb, ODP_FLOW_ATTR_STATE, state); - - if (skb->len > total_len) - goto nla_put_failure; + NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags); + + /* If OVS_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, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len, + sf_acts->actions); + if (err < 0 && skb_orig_len) + goto error; - odp_flow->len = skb->len; - err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_unlock; + return genlmsg_end(skb, ovs_header); nla_put_failure: err = -EMSGSIZE; -exit_unlock: - rcu_read_unlock(); - kfree_skb(skb); -exit: +error: + genlmsg_cancel(skb, ovs_header); return err; } -static struct sk_buff *copy_flow_from_user(struct odp_flow __user *uodp_flow, - struct dp_flowcmd *flowcmd) +static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow) { - struct nlattr *a[ODP_FLOW_ATTR_MAX + 1]; - struct odp_flow *odp_flow; - struct sk_buff *skb; - u32 len; - int err; + const struct sw_flow_actions *sf_acts; + int len; - if (get_user(len, &uodp_flow->len)) - return ERR_PTR(-EFAULT); - if (len < sizeof(struct odp_flow)) - return ERR_PTR(-EINVAL); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + + len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */ + len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */ + len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */ + len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */ + len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */ + return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL); +} - skb = alloc_skb(len, GFP_KERNEL); +static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp, + u32 pid, u32 seq, u8 cmd) +{ + struct sk_buff *skb; + int retval; + + skb = ovs_flow_cmd_alloc_info(flow); if (!skb) return ERR_PTR(-ENOMEM); - err = -EFAULT; - if (copy_from_user(__skb_put(skb, len), uodp_flow, len)) - goto error_free_skb; - - odp_flow = (struct odp_flow *)skb->data; - err = -EINVAL; - if (odp_flow->len != len) - goto error_free_skb; - - flowcmd->nlmsg_flags = odp_flow->nlmsg_flags; - flowcmd->dp_idx = odp_flow->dp_idx; - flowcmd->total_len = odp_flow->total_len; - - err = nla_parse(a, ODP_FLOW_ATTR_MAX, - (struct nlattr *)(skb->data + sizeof(struct odp_flow)), - skb->len - sizeof(struct odp_flow), flow_policy); - if (err) - goto error_free_skb; - - /* ODP_FLOW_ATTR_KEY. */ - if (a[ODP_FLOW_ATTR_KEY]) { - err = flow_from_nlattrs(&flowcmd->key, a[ODP_FLOW_ATTR_KEY]); - if (err) - goto error_free_skb; - } else - memset(&flowcmd->key, 0, sizeof(struct sw_flow_key)); - - /* ODP_FLOW_ATTR_ACTIONS. */ - if (a[ODP_FLOW_ATTR_ACTIONS]) { - flowcmd->actions = nla_data(a[ODP_FLOW_ATTR_ACTIONS]); - flowcmd->actions_len = nla_len(a[ODP_FLOW_ATTR_ACTIONS]); - err = validate_actions(flowcmd->actions, flowcmd->actions_len); - if (err) - goto error_free_skb; - } else { - flowcmd->actions = NULL; - flowcmd->actions_len = 0; - } - - flowcmd->clear = a[ODP_FLOW_ATTR_CLEAR] != NULL; - - flowcmd->state = a[ODP_FLOW_ATTR_STATE] ? nla_get_u64(a[ODP_FLOW_ATTR_STATE]) : 0; - + retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); + BUG_ON(retval < 0); return skb; - -error_free_skb: - kfree_skb(skb); - return ERR_PTR(err); } -static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) +static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) { - struct tbl_node *flow_node; - struct dp_flowcmd flowcmd; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; struct sw_flow *flow; - struct sk_buff *skb; + struct sk_buff *reply; struct datapath *dp; - struct tbl *table; - u32 hash; + struct flow_table *table; int error; + int key_len; - skb = copy_flow_from_user(uodp_flow, &flowcmd); - error = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; + /* Extract key. */ + error = -EINVAL; + if (!a[OVS_FLOW_ATTR_KEY]) + goto error; + error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (error) + goto error; + + /* Validate actions. */ + if (a[OVS_FLOW_ATTR_ACTIONS]) { + error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0); + if (error) + goto error; + } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) { + error = -EINVAL; + goto error; + } - dp = get_dp_locked(flowcmd.dp_idx); + dp = get_dp(ovs_header->dp_ifindex); error = -ENODEV; if (!dp) - goto error_kfree_skb; + goto error; - hash = flow_hash(&flowcmd.key); table = get_table_protected(dp); - flow_node = tbl_lookup(table, &flowcmd.key, hash, flow_cmp); - if (!flow_node) { + flow = flow_tbl_lookup(table, &key, key_len); + if (!flow) { struct sw_flow_actions *acts; /* Bail out if we're not allowed to create a new flow. */ error = -ENOENT; - if (cmd == ODP_FLOW_SET) - goto error_unlock_dp; + if (info->genlhdr->cmd == OVS_FLOW_CMD_SET) + goto error; /* Expand table, if necessary, to make room. */ - if (tbl_count(table) >= tbl_n_buckets(table)) { - error = expand_table(dp); - if (error) - goto error_unlock_dp; - table = get_table_protected(dp); + if (flow_tbl_need_to_expand(table)) { + struct flow_table *new_table; + + new_table = flow_tbl_expand(table); + if (!IS_ERR(new_table)) { + rcu_assign_pointer(dp->table, new_table); + flow_tbl_deferred_destroy(table); + table = get_table_protected(dp); + } } /* Allocate flow. */ flow = flow_alloc(); if (IS_ERR(flow)) { error = PTR_ERR(flow); - goto error_unlock_dp; + goto error; } - flow->key = flowcmd.key; + flow->key = key; clear_stats(flow); /* Obtain actions. */ - acts = get_actions(&flowcmd); + acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]); error = PTR_ERR(acts); if (IS_ERR(acts)) goto error_free_flow; rcu_assign_pointer(flow->sf_acts, acts); - error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); - if (error) - goto error_free_flow; - /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, hash); - if (error) - goto error_free_flow; + flow->hash = flow_hash(&key, key_len); + flow_tbl_insert(table, flow); + + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, OVS_FLOW_CMD_NEW); } else { /* We found a matching flow. */ struct sw_flow_actions *old_acts; @@ -1054,338 +1063,305 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) * gets fixed. */ error = -EEXIST; - if (flowcmd.nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) - goto error_kfree_skb; + if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW && + info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) + goto error; /* Update actions. */ - flow = flow_cast(flow_node); old_acts = rcu_dereference_protected(flow->sf_acts, - lockdep_is_held(&dp->mutex)); - if (flowcmd.actions && - (old_acts->actions_len != flowcmd.actions_len || - memcmp(old_acts->actions, flowcmd.actions, - flowcmd.actions_len))) { + lockdep_genl_is_held()); + if (a[OVS_FLOW_ATTR_ACTIONS] && + (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) || + memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]), + old_acts->actions_len))) { struct sw_flow_actions *new_acts; - new_acts = get_actions(&flowcmd); + new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]); error = PTR_ERR(new_acts); if (IS_ERR(new_acts)) - goto error_kfree_skb; + goto error; rcu_assign_pointer(flow->sf_acts, new_acts); flow_deferred_free_acts(old_acts); } - error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); - if (error) - goto error_kfree_skb; + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, OVS_FLOW_CMD_NEW); /* Clear stats. */ - if (flowcmd.clear) { + if (a[OVS_FLOW_ATTR_CLEAR]) { spin_lock_bh(&flow->lock); clear_stats(flow); spin_unlock_bh(&flow->lock); } } - kfree_skb(skb); - mutex_unlock(&dp->mutex); + + 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; error_free_flow: flow_put(flow); -error_unlock_dp: - mutex_unlock(&dp->mutex); -error_kfree_skb: - kfree_skb(skb); -exit: +error: return error; } -static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) +static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct tbl_node *flow_node; - struct dp_flowcmd flowcmd; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; + struct sk_buff *reply; struct sw_flow *flow; - struct sk_buff *skb; struct datapath *dp; - struct tbl *table; + struct flow_table *table; int err; + int key_len; - skb = copy_flow_from_user(uodp_flow, &flowcmd); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; + if (!a[OVS_FLOW_ATTR_KEY]) + return -EINVAL; + err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (err) + return err; - dp = get_dp_locked(flowcmd.dp_idx); - err = -ENODEV; + dp = get_dp(ovs_header->dp_ifindex); if (!dp) - goto exit_kfree_skb; + return -ENODEV; table = get_table_protected(dp); - flow_node = tbl_lookup(table, &flowcmd.key, flow_hash(&flowcmd.key), flow_cmp); - err = -ENOENT; - if (!flow_node) - goto exit_unlock_dp; - - if (cmd == ODP_FLOW_DEL) { - err = tbl_remove(table, flow_node); - if (err) - goto exit_unlock_dp; - } + flow = flow_tbl_lookup(table, &key, key_len); + if (!flow) + return -ENOENT; - flow = flow_cast(flow_node); - err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); - if (!err && cmd == ODP_FLOW_DEL) - flow_deferred_free(flow); + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_kfree_skb: - kfree_skb(skb); -exit: - return err; + return genlmsg_reply(reply, info); } -static int dump_flow(struct odp_flow __user *uodp_flow) +static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct tbl_node *flow_node; - struct dp_flowcmd flowcmd; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; + struct sk_buff *reply; struct sw_flow *flow; - struct sk_buff *skb; struct datapath *dp; - u32 bucket, obj; + struct flow_table *table; int err; + int key_len; - skb = copy_flow_from_user(uodp_flow, &flowcmd); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; + if (!a[OVS_FLOW_ATTR_KEY]) + return flush_flows(ovs_header->dp_ifindex); + err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (err) + return err; - dp = get_dp_locked(flowcmd.dp_idx); - err = -ENODEV; + dp = get_dp(ovs_header->dp_ifindex); if (!dp) - goto exit_free; + return -ENODEV; - bucket = flowcmd.state >> 32; - obj = flowcmd.state; - flow_node = tbl_next(dp->table, &bucket, &obj); - err = -ENODEV; - if (!flow_node) - goto exit_unlock_dp; + table = get_table_protected(dp); + flow = flow_tbl_lookup(table, &key, key_len); + if (!flow) + return -ENOENT; - flow = flow_cast(flow_node); - err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, - ((u64)bucket << 32) | obj); + reply = ovs_flow_cmd_alloc_info(flow); + if (!reply) + return -ENOMEM; -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_free: - kfree_skb(skb); -exit: - return err; + flow_tbl_remove(table, flow); + + err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, + info->snd_seq, 0, OVS_FLOW_CMD_DEL); + BUG_ON(err < 0); + + flow_deferred_free(flow); + + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; +} + +static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) +{ + struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; + + dp = get_dp(ovs_header->dp_ifindex); + if (!dp) + return -ENODEV; + + for (;;) { + struct sw_flow *flow; + u32 bucket, obj; + + bucket = cb->args[0]; + obj = cb->args[1]; + flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj); + if (!flow) + break; + + if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + OVS_FLOW_CMD_NEW) < 0) + break; + + cb->args[0] = bucket; + cb->args[1] = obj; + } + return skb->len; } -static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = { - [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, - [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 }, - [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 }, +static struct genl_ops dp_flow_genl_ops[] = { + { .cmd = OVS_FLOW_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_new_or_set + }, + { .cmd = OVS_FLOW_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_del + }, + { .cmd = OVS_FLOW_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_get, + .dumpit = ovs_flow_cmd_dump + }, + { .cmd = OVS_FLOW_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_new_or_set, + }, +}; + +static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, +#endif + [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, +}; + +static struct genl_family dp_datapath_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_DATAPATH_FAMILY, + .version = OVS_DATAPATH_VERSION, + .maxattr = OVS_DP_ATTR_MAX +}; + +static struct genl_multicast_group dp_datapath_multicast_group = { + .name = OVS_DATAPATH_MCGROUP }; -static int copy_datapath_to_user(void __user *dst, struct datapath *dp, uint32_t total_len) +static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) { - struct odp_datapath *odp_datapath; - struct sk_buff *skb; + struct ovs_header *ovs_header; struct nlattr *nla; int err; - skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); - err = -ENOMEM; - if (!skb) - goto exit; + ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, + flags, cmd); + if (!ovs_header) + goto error; - odp_datapath = (struct odp_datapath*)__skb_put(skb, sizeof(struct odp_datapath)); - odp_datapath->dp_idx = dp->dp_idx; - odp_datapath->total_len = total_len; + ovs_header->dp_ifindex = get_dpifindex(dp); rcu_read_lock(); - err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp)); + err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp)); rcu_read_unlock(); if (err) goto nla_put_failure; - nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats)); + nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats)); if (!nla) goto nla_put_failure; get_dp_stats(dp, nla_data(nla)); - NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS, - dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO); - - if (dp->sflow_probability) - NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability); - - if (skb->len > total_len) - goto nla_put_failure; - - odp_datapath->len = skb->len; - err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_free_skb; + return genlmsg_end(skb, ovs_header); nla_put_failure: - err = -EMSGSIZE; -exit_free_skb: - kfree_skb(skb); -exit: - return err; + genlmsg_cancel(skb, ovs_header); +error: + return -EMSGSIZE; } -static struct sk_buff *copy_datapath_from_user(struct odp_datapath __user *uodp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid, + u32 seq, u8 cmd) { - struct odp_datapath *odp_datapath; struct sk_buff *skb; - u32 len; - int err; - - if (get_user(len, &uodp_datapath->len)) - return ERR_PTR(-EFAULT); - if (len < sizeof(struct odp_datapath)) - return ERR_PTR(-EINVAL); + int retval; - skb = alloc_skb(len, GFP_KERNEL); + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); if (!skb) return ERR_PTR(-ENOMEM); - err = -EFAULT; - if (copy_from_user(__skb_put(skb, len), uodp_datapath, len)) - goto error_free_skb; - - odp_datapath = (struct odp_datapath *)skb->data; - err = -EINVAL; - if (odp_datapath->len != len) - goto error_free_skb; - - err = nla_parse(a, ODP_DP_ATTR_MAX, - (struct nlattr *)(skb->data + sizeof(struct odp_datapath)), - skb->len - sizeof(struct odp_datapath), datapath_policy); - if (err) - goto error_free_skb; - - if (a[ODP_DP_ATTR_IPV4_FRAGS]) { - u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - - err = -EINVAL; - if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP) - goto error_free_skb; + retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); } - - err = VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1); - if (err) - goto error_free_skb; - return skb; - -error_free_skb: - kfree_skb(skb); - return ERR_PTR(err); } -/* Called with dp_mutex and optionally with RTNL lock also. - * Holds the returned datapath's mutex on return. - */ -static struct datapath *lookup_datapath(struct odp_datapath *odp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1]) { - WARN_ON_ONCE(!mutex_is_locked(&dp_mutex)); + return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1); +} - if (!a[ODP_DP_ATTR_NAME]) { - struct datapath *dp; +/* Called with genl_mutex and optionally with RTNL lock also. */ +static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1]) +{ + struct datapath *dp; - dp = get_dp(odp_datapath->dp_idx); - if (!dp) - return ERR_PTR(-ENODEV); - mutex_lock(&dp->mutex); - return dp; - } else { - struct datapath *dp; + if (!a[OVS_DP_ATTR_NAME]) + dp = get_dp(ovs_header->dp_ifindex); + else { struct vport *vport; - int dp_idx; - - vport_lock(); - vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME])); - dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1; - vport_unlock(); - - if (dp_idx < 0) - return ERR_PTR(-ENODEV); - dp = get_dp(dp_idx); - mutex_lock(&dp->mutex); - return dp; + rcu_read_lock(); + vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME])); + dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; + rcu_read_unlock(); } + return dp ? dp : ERR_PTR(-ENODEV); } -static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) { - 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 int new_datapath(struct odp_datapath __user *uodp_datapath) -{ - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct odp_datapath *odp_datapath; + struct nlattr **a = info->attrs; struct vport_parms parms; - struct sk_buff *skb; + struct sk_buff *reply; struct datapath *dp; struct vport *vport; - int dp_idx; int err; - int i; - skb = copy_datapath_from_user(uodp_datapath, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = -EINVAL; + if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) goto err; - odp_datapath = (struct odp_datapath *)skb->data; - err = -EINVAL; - if (!a[ODP_DP_ATTR_NAME]) - goto err_free_skb; + err = ovs_dp_cmd_validate(a); + if (err) + goto err; rtnl_lock(); - mutex_lock(&dp_mutex); err = -ENODEV; if (!try_module_get(THIS_MODULE)) - goto err_unlock_dp_mutex; - - dp_idx = odp_datapath->dp_idx; - if (dp_idx < 0) { - err = -EFBIG; - for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) { - if (get_dp(dp_idx)) - continue; - err = 0; - break; - } - } else if (dp_idx < ARRAY_SIZE(dps)) - err = get_dp(dp_idx) ? -EBUSY : 0; - else - err = -EINVAL; - if (err) - goto err_put_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); - mutex_init(&dp->mutex); - mutex_lock(&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. */ @@ -1394,408 +1370,385 @@ static int new_datapath(struct odp_datapath __user *uodp_datapath) /* Allocate table. */ err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); + rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS)); if (!dp->table) goto err_free_dp; + dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); + if (!dp->stats_percpu) { + err = -ENOMEM; + goto err_destroy_table; + } + /* Set up our datapath device. */ - parms.name = nla_data(a[ODP_DP_ATTR_NAME]); - parms.type = ODP_VPORT_TYPE_INTERNAL; + parms.name = nla_data(a[OVS_DP_ATTR_NAME]); + parms.type = OVS_VPORT_TYPE_INTERNAL; parms.options = NULL; parms.dp = dp; - parms.port_no = ODPP_LOCAL; + parms.port_no = OVSP_LOCAL; + parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]); + vport = new_vport(&parms); if (IS_ERR(vport)) { err = PTR_ERR(vport); if (err == -EBUSY) err = -EEXIST; - goto err_destroy_table; + goto err_destroy_percpu; } - dp->drop_frags = 0; - dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) { - err = -ENOMEM; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) goto err_destroy_local_port; - } - change_datapath(dp, a); - - rcu_assign_pointer(dps[dp_idx], dp); + list_add_tail(&dp->list_node, &dps); dp_sysfs_add_dp(dp); - mutex_unlock(&dp->mutex); - mutex_unlock(&dp_mutex); rtnl_unlock(); + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); return 0; err_destroy_local_port: - dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); + dp_detach_port(get_vport_protected(dp, OVSP_LOCAL)); +err_destroy_percpu: + free_percpu(dp->stats_percpu); err_destroy_table: - tbl_destroy(get_table_protected(dp), NULL); + flow_tbl_destroy(get_table_protected(dp)); err_free_dp: - mutex_unlock(&dp->mutex); kfree(dp); err_put_module: module_put(THIS_MODULE); -err_unlock_dp_mutex: - mutex_unlock(&dp_mutex); +err_unlock_rtnl: rtnl_unlock(); -err_free_skb: - kfree_skb(skb); err: return err; } -static int del_datapath(struct odp_datapath __user *uodp_datapath) +static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct vport *vport, *next_vport; + struct sk_buff *reply; struct datapath *dp; - struct sk_buff *skb; int err; - skb = copy_datapath_from_user(uodp_datapath, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = ovs_dp_cmd_validate(info->attrs); + if (err) goto exit; rtnl_lock(); - mutex_lock(&dp_mutex); - dp = lookup_datapath((struct odp_datapath *)skb->data, a); + dp = lookup_datapath(info->userhdr, info->attrs); err = PTR_ERR(dp); if (IS_ERR(dp)) - goto exit_free; + goto exit_unlock; - destroy_dp(dp); - err = 0; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; -exit_free: - kfree_skb(skb); - mutex_unlock(&dp_mutex); - rtnl_unlock(); -exit: - return err; -} + list_for_each_entry_safe (vport, next_vport, &dp->port_list, node) + if (vport->port_no != OVSP_LOCAL) + dp_detach_port(vport); -static int set_datapath(struct odp_datapath __user *uodp_datapath) -{ - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct datapath *dp; - struct sk_buff *skb; - int err; + dp_sysfs_del_dp(dp); + list_del(&dp->list_node); + dp_detach_port(get_vport_protected(dp, OVSP_LOCAL)); + + /* rtnl_unlock() will wait until all the references to devices that + * are pending unregistration have been dropped. We do it here to + * ensure that any internal devices (which contain DP pointers) are + * fully destroyed before freeing the datapath. + */ + rtnl_unlock(); - skb = copy_datapath_from_user(uodp_datapath, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); - mutex_lock(&dp_mutex); - dp = lookup_datapath((struct odp_datapath *)skb->data, a); - err = PTR_ERR(dp); - if (IS_ERR(dp)) - goto exit_free; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); - change_datapath(dp, a); - mutex_unlock(&dp->mutex); - err = 0; + return 0; -exit_free: - kfree_skb(skb); - mutex_unlock(&dp_mutex); +exit_unlock: + rtnl_unlock(); exit: return err; } -static int get_datapath(struct odp_datapath __user *uodp_datapath) +static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct odp_datapath *odp_datapath; + struct sk_buff *reply; struct datapath *dp; - struct sk_buff *skb; int err; - skb = copy_datapath_from_user(uodp_datapath, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; - odp_datapath = (struct odp_datapath *)skb->data; - - mutex_lock(&dp_mutex); - dp = lookup_datapath(odp_datapath, a); - mutex_unlock(&dp_mutex); + err = ovs_dp_cmd_validate(info->attrs); + if (err) + return err; - err = PTR_ERR(dp); + dp = lookup_datapath(info->userhdr, info->attrs); if (IS_ERR(dp)) - goto exit_free; + return PTR_ERR(dp); - err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len); - mutex_unlock(&dp->mutex); -exit_free: - kfree_skb(skb); -exit: - return err; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_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; + } + + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -static int dump_datapath(struct odp_datapath __user *uodp_datapath) +static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct odp_datapath *odp_datapath; - struct sk_buff *skb; - u32 dp_idx; + struct sk_buff *reply; + struct datapath *dp; int err; - skb = copy_datapath_from_user(uodp_datapath, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; - odp_datapath = (struct odp_datapath *)skb->data; + err = ovs_dp_cmd_validate(info->attrs); + if (err) + return err; - mutex_lock(&dp_mutex); - for (dp_idx = odp_datapath->dp_idx; dp_idx < ARRAY_SIZE(dps); dp_idx++) { - struct datapath *dp = get_dp(dp_idx); - if (!dp) - continue; + dp = lookup_datapath(info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - mutex_lock(&dp->mutex); - mutex_unlock(&dp_mutex); - err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len); - mutex_unlock(&dp->mutex); - goto exit_free; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); + + return genlmsg_reply(reply, info); +} + +static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) +{ + struct datapath *dp; + int skip = cb->args[0]; + int i = 0; + + list_for_each_entry (dp, &dps, list_node) { + if (i < skip) + continue; + if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + OVS_DP_CMD_NEW) < 0) + break; + i++; } - mutex_unlock(&dp_mutex); - err = -ENODEV; -exit_free: - kfree_skb(skb); -exit: - return err; + cb->args[0] = i; + + return skb->len; } -static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = { - [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 }, - [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 }, - [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, +static struct genl_ops dp_datapath_genl_ops[] = { + { .cmd = OVS_DP_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_new + }, + { .cmd = OVS_DP_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_del + }, + { .cmd = OVS_DP_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_get, + .dumpit = ovs_dp_cmd_dump + }, + { .cmd = OVS_DP_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_set, + }, }; -static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t total_len) +static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, + [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, + [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN }, +#else + [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) }, + [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN }, +#endif + [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, +}; + +static struct genl_family dp_vport_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_VPORT_FAMILY, + .version = OVS_VPORT_VERSION, + .maxattr = OVS_VPORT_ATTR_MAX +}; + +struct genl_multicast_group dp_vport_multicast_group = { + .name = OVS_VPORT_MCGROUP +}; + +/* Called with RTNL lock or RCU read lock. */ +static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) { - struct odp_vport *odp_vport; - struct sk_buff *skb; + struct ovs_header *ovs_header; struct nlattr *nla; - int ifindex, iflink; int err; - skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); - err = -ENOMEM; - if (!skb) - goto exit; + ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, + flags, cmd); + if (!ovs_header) + return -EMSGSIZE; - rcu_read_lock(); - odp_vport = (struct odp_vport*)__skb_put(skb, sizeof(struct odp_vport)); - odp_vport->dp_idx = vport->dp->dp_idx; - odp_vport->total_len = total_len; + ovs_header->dp_ifindex = get_dpifindex(vport->dp); - 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)); + NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no); + NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport)); + NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport)); + NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid); - nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64)); + nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats)); if (!nla) goto nla_put_failure; - if (vport_get_stats(vport, nla_data(nla))) - __skb_trim(skb, skb->len - nla->nla_len); - NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); + vport_get_stats(vport, nla_data(nla)); - NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport)); + NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); err = vport_get_options(vport, skb); + if (err == -EMSGSIZE) + goto error; - ifindex = vport_get_ifindex(vport); - if (ifindex > 0) - NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex); - - iflink = vport_get_iflink(vport); - if (iflink > 0) - NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink); - - err = -EMSGSIZE; - if (skb->len > total_len) - goto exit_unlock; - - odp_vport->len = skb->len; - err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_unlock; + return genlmsg_end(skb, ovs_header); nla_put_failure: err = -EMSGSIZE; -exit_unlock: - rcu_read_unlock(); - kfree_skb(skb); -exit: +error: + genlmsg_cancel(skb, ovs_header); return err; } -static struct sk_buff *copy_vport_from_user(struct odp_vport __user *uodp_vport, - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +/* Called with RTNL lock or RCU read lock. */ +struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid, + u32 seq, u8 cmd) { - struct odp_vport *odp_vport; struct sk_buff *skb; - u32 len; - int err; - - if (get_user(len, &uodp_vport->len)) - return ERR_PTR(-EFAULT); - if (len < sizeof(struct odp_vport)) - return ERR_PTR(-EINVAL); + int retval; - skb = alloc_skb(len, GFP_KERNEL); + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); if (!skb) return ERR_PTR(-ENOMEM); - err = -EFAULT; - if (copy_from_user(__skb_put(skb, len), uodp_vport, len)) - goto error_free_skb; - - odp_vport = (struct odp_vport *)skb->data; - err = -EINVAL; - if (odp_vport->len != len) - goto error_free_skb; - - err = nla_parse(a, ODP_VPORT_ATTR_MAX, (struct nlattr *)(skb->data + sizeof(struct odp_vport)), - skb->len - sizeof(struct odp_vport), vport_policy); - if (err) - goto error_free_skb; - - err = VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1); - if (err) - goto error_free_skb; - + retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } return skb; - -error_free_skb: - kfree_skb(skb); - return ERR_PTR(err); } +static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) +{ + return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1); +} -/* Called without any locks (or with RTNL lock). - * Returns holding vport->dp->mutex. - */ -static struct vport *lookup_vport(struct odp_vport *odp_vport, - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +/* Called with RTNL lock or RCU read lock. */ +static struct vport *lookup_vport(struct ovs_header *ovs_header, + struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) { struct datapath *dp; struct vport *vport; - if (a[ODP_VPORT_ATTR_NAME]) { - int dp_idx, port_no; - - retry: - vport_lock(); - vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME])); - if (!vport) { - vport_unlock(); + if (a[OVS_VPORT_ATTR_NAME]) { + vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME])); + if (!vport) return ERR_PTR(-ENODEV); - } - dp_idx = vport->dp->dp_idx; - port_no = vport->port_no; - vport_unlock(); - - dp = get_dp_locked(dp_idx); - if (!dp) - goto retry; - - vport = get_vport_protected(dp, port_no); - if (!vport || - strcmp(vport_get_name(vport), nla_data(a[ODP_VPORT_ATTR_NAME]))) { - mutex_unlock(&dp->mutex); - goto retry; - } - return vport; - } else if (a[ODP_VPORT_ATTR_PORT_NO]) { - u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + } else if (a[OVS_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); if (port_no >= DP_MAX_PORTS) - return ERR_PTR(-EINVAL); + return ERR_PTR(-EFBIG); - dp = get_dp_locked(odp_vport->dp_idx); + dp = get_dp(ovs_header->dp_ifindex); if (!dp) return ERR_PTR(-ENODEV); vport = get_vport_protected(dp, port_no); - if (!vport) { - mutex_unlock(&dp->mutex); + if (!vport) return ERR_PTR(-ENOENT); - } return vport; } else return ERR_PTR(-EINVAL); } -static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +/* Called with RTNL lock. */ +static int change_vport(struct vport *vport, struct nlattr *a[OVS_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])); + + if (a[OVS_VPORT_ATTR_STATS]) + vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS])); + + if (a[OVS_VPORT_ATTR_ADDRESS]) + err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS])); + return err; } -static int attach_vport(struct odp_vport __user *uodp_vport) +static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; - struct odp_vport *odp_vport; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; struct vport_parms parms; + struct sk_buff *reply; struct vport *vport; - struct sk_buff *skb; struct datapath *dp; u32 port_no; int err; - skb = copy_vport_from_user(uodp_vport, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = -EINVAL; + if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || + !a[OVS_VPORT_ATTR_UPCALL_PID]) goto exit; - odp_vport = (struct odp_vport *)skb->data; - err = -EINVAL; - if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE]) - goto exit_kfree_skb; + err = ovs_vport_cmd_validate(a); + if (err) + goto exit; rtnl_lock(); - - dp = get_dp_locked(odp_vport->dp_idx); + dp = get_dp(ovs_header->dp_ifindex); err = -ENODEV; if (!dp) - goto exit_unlock_rtnl; + goto exit_unlock; - if (a[ODP_VPORT_ATTR_PORT_NO]) { - port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + if (a[OVS_VPORT_ATTR_PORT_NO]) { + port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); err = -EFBIG; if (port_no >= DP_MAX_PORTS) - goto exit_unlock_dp; + goto exit_unlock; vport = get_vport_protected(dp, port_no); err = -EBUSY; if (vport) - goto exit_unlock_dp; + goto exit_unlock; } else { for (port_no = 1; ; port_no++) { if (port_no >= DP_MAX_PORTS) { err = -EFBIG; - goto exit_unlock_dp; + goto exit_unlock; } vport = get_vport_protected(dp, port_no); if (!vport) @@ -1803,436 +1756,275 @@ static int attach_vport(struct odp_vport __user *uodp_vport) } } - 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.name = nla_data(a[OVS_VPORT_ATTR_NAME]); + parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); + parms.options = a[OVS_VPORT_ATTR_OPTIONS]; parms.dp = dp; parms.port_no = port_no; + parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); vport = new_vport(&parms); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_unlock_dp; + goto exit_unlock; - set_internal_devs_mtu(dp); dp_sysfs_add_if(vport); err = change_vport(vport, a); + if (!err) { + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, + info->snd_seq, OVS_VPORT_CMD_NEW); + if (IS_ERR(reply)) + err = PTR_ERR(reply); + } if (err) { dp_detach_port(vport); - goto exit_unlock_dp; + goto exit_unlock; } + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_unlock_rtnl: +exit_unlock: rtnl_unlock(); -exit_kfree_skb: - kfree_skb(skb); exit: return err; } -static int set_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) +static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct nlattr **a = info->attrs; + struct sk_buff *reply; struct vport *vport; - struct sk_buff *skb; int err; - skb = copy_vport_from_user(uodp_vport, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = ovs_vport_cmd_validate(a); + if (err) goto exit; rtnl_lock(); - vport = lookup_vport((struct odp_vport *)skb->data, a); + vport = lookup_vport(info->userhdr, a); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_free; + goto exit_unlock; err = 0; - if (a[ODP_VPORT_ATTR_OPTIONS]) - err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]); + if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport)) + err = -EINVAL; + if (!err && a[OVS_VPORT_ATTR_OPTIONS]) + err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); if (!err) err = change_vport(vport, a); + if (!err && a[OVS_VPORT_ATTR_UPCALL_PID]) + vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); + + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_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; + } - mutex_unlock(&vport->dp->mutex); -exit_free: - kfree_skb(skb); + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); + +exit_unlock: rtnl_unlock(); exit: return err; } -static int del_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) +static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; - struct datapath *dp; + struct nlattr **a = info->attrs; + struct sk_buff *reply; struct vport *vport; - struct sk_buff *skb; int err; - skb = copy_vport_from_user(uodp_vport, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = ovs_vport_cmd_validate(a); + if (err) goto exit; rtnl_lock(); - vport = lookup_vport((struct odp_vport *)skb->data, a); + vport = lookup_vport(info->userhdr, a); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_free; - dp = vport->dp; + goto exit_unlock; - err = -EINVAL; - if (vport->port_no == ODPP_LOCAL) - goto exit_free; + if (vport->port_no == OVSP_LOCAL) { + err = -EINVAL; + goto exit_unlock; + } + + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_VPORT_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; + + dp_detach_port(vport); - err = dp_detach_port(vport); - mutex_unlock(&dp->mutex); -exit_free: - kfree_skb(skb); + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); + +exit_unlock: rtnl_unlock(); exit: return err; } -static int get_vport(struct odp_vport __user *uodp_vport) +static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; - struct odp_vport *odp_vport; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sk_buff *reply; struct vport *vport; - struct sk_buff *skb; int err; - skb = copy_vport_from_user(uodp_vport, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) + err = ovs_vport_cmd_validate(a); + if (err) goto exit; - odp_vport = (struct odp_vport *)skb->data; - vport = lookup_vport(odp_vport, a); + rcu_read_lock(); + vport = lookup_vport(ovs_header, a); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_free; + goto exit_unlock; - err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); - mutex_unlock(&vport->dp->mutex); -exit_free: - kfree_skb(skb); -exit: - return err; -} + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_VPORT_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; -static int dump_vport(struct odp_vport __user *uodp_vport) -{ - struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; - struct odp_vport *odp_vport; - struct sk_buff *skb; - struct datapath *dp; - u32 port_no; - int err; + rcu_read_unlock(); - skb = copy_vport_from_user(uodp_vport, a); - err = PTR_ERR(skb); - if (IS_ERR(skb)) - goto exit; - odp_vport = (struct odp_vport *)skb->data; + return genlmsg_reply(reply, info); - dp = get_dp_locked(odp_vport->dp_idx); - err = -ENODEV; - if (!dp) - goto exit_free; - - port_no = 0; - if (a[ODP_VPORT_ATTR_PORT_NO]) - port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - for (; port_no < DP_MAX_PORTS; port_no++) { - struct vport *vport = get_vport_protected(dp, port_no); - if (vport) { - err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); - goto exit_unlock_dp; - } - } - err = -ENODEV; - -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_free: - kfree_skb(skb); +exit_unlock: + rcu_read_unlock(); exit: return err; } -static long openvswitch_ioctl(struct file *f, unsigned int cmd, - unsigned long argp) +static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - int dp_idx = iminor(f->f_dentry->d_inode); + struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); struct datapath *dp; - int listeners; - int err; - - /* Handle commands with special locking requirements up front. */ - switch (cmd) { - case ODP_DP_NEW: - err = new_datapath((struct odp_datapath __user *)argp); - goto exit; - - case ODP_DP_GET: - err = get_datapath((struct odp_datapath __user *)argp); - goto exit; - - case ODP_DP_DEL: - err = del_datapath((struct odp_datapath __user *)argp); - goto exit; - - case ODP_DP_SET: - err = set_datapath((struct odp_datapath __user *)argp); - goto exit; - - case ODP_DP_DUMP: - err = dump_datapath((struct odp_datapath __user *)argp); - goto exit; - - case ODP_VPORT_NEW: - err = attach_vport((struct odp_vport __user *)argp); - goto exit; - - case ODP_VPORT_GET: - err = get_vport((struct odp_vport __user *)argp); - goto exit; - - case ODP_VPORT_DEL: - err = del_vport(cmd, (struct odp_vport __user *)argp); - goto exit; - - case ODP_VPORT_SET: - err = set_vport(cmd, (struct odp_vport __user *)argp); - goto exit; - - case ODP_VPORT_DUMP: - err = dump_vport((struct odp_vport __user *)argp); - goto exit; - - case ODP_FLOW_FLUSH: - err = flush_flows(argp); - goto exit; - - case ODP_FLOW_NEW: - case ODP_FLOW_SET: - err = new_flow(cmd, (struct odp_flow __user *)argp); - goto exit; - - case ODP_FLOW_GET: - case ODP_FLOW_DEL: - err = get_or_del_flow(cmd, (struct odp_flow __user *)argp); - goto exit; - - case ODP_FLOW_DUMP: - err = dump_flow((struct odp_flow __user *)argp); - goto exit; - - case ODP_EXECUTE: - err = execute_packet((struct odp_execute __user *)argp); - goto exit; - } + u32 port_no; + int retval; - dp = get_dp_locked(dp_idx); - err = -ENODEV; + dp = get_dp(ovs_header->dp_ifindex); if (!dp) - goto exit; - - switch (cmd) { - case ODP_GET_LISTEN_MASK: - err = put_user(get_listen_mask(f), (int __user *)argp); - break; + return -ENODEV; - 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; + rcu_read_lock(); + for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) { + struct vport *vport; - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); -exit: - return err; -} + vport = get_vport_protected(dp, port_no); + if (!vport) + continue; -static int dp_has_packet_of_interest(struct datapath *dp, int listeners) -{ - int i; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i])) - return 1; + if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + OVS_VPORT_CMD_NEW) < 0) + break; } - return 0; -} + rcu_read_unlock(); -#ifdef CONFIG_COMPAT -static int compat_execute(const struct compat_odp_execute __user *uexecute) -{ - struct odp_execute execute; - compat_uptr_t actions; - compat_uptr_t data; - struct datapath *dp; - int error; + cb->args[0] = port_no; + retval = skb->len; - if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) || - __get_user(execute.dp_idx, &uexecute->dp_idx) || - __get_user(actions, &uexecute->actions) || - __get_user(execute.actions_len, &uexecute->actions_len) || - __get_user(data, &uexecute->data) || - __get_user(execute.length, &uexecute->length)) - return -EFAULT; + return retval; +} - execute.actions = (struct nlattr __force *)compat_ptr(actions); - execute.data = (const void __force *)compat_ptr(data); +static struct genl_ops dp_vport_genl_ops[] = { + { .cmd = OVS_VPORT_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_new + }, + { .cmd = OVS_VPORT_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_del + }, + { .cmd = OVS_VPORT_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_get, + .dumpit = ovs_vport_cmd_dump + }, + { .cmd = OVS_VPORT_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_set, + }, +}; - dp = get_dp_locked(execute.dp_idx); - if (!dp) - return -ENODEV; - error = do_execute(dp, &execute); - mutex_unlock(&dp->mutex); +struct genl_family_and_ops { + struct genl_family *family; + struct genl_ops *ops; + int n_ops; + struct genl_multicast_group *group; +}; - return error; -} +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 }, +}; -static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) +static void dp_unregister_genl(int n_families) { - switch (cmd) { - case ODP_FLOW_FLUSH: - /* Ioctls that don't need any translation at all. */ - return openvswitch_ioctl(f, cmd, argp); - - case ODP_DP_NEW: - case ODP_DP_GET: - case ODP_DP_DEL: - case ODP_DP_SET: - case ODP_DP_DUMP: - case ODP_VPORT_NEW: - case ODP_VPORT_DEL: - case ODP_VPORT_GET: - case ODP_VPORT_SET: - case ODP_VPORT_DUMP: - case ODP_FLOW_NEW: - case ODP_FLOW_DEL: - case ODP_FLOW_GET: - case ODP_FLOW_SET: - case ODP_FLOW_DUMP: - case ODP_SET_LISTEN_MASK: - case ODP_GET_LISTEN_MASK: - /* Ioctls that just need their pointer argument extended. */ - return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); - - case ODP_EXECUTE32: - return compat_execute(compat_ptr(argp)); + int i; - default: - return -ENOIOCTLCMD; - } + for (i = 0; i < n_families; i++) + genl_unregister_family(dp_genl_families[i].family); } -#endif -static ssize_t openvswitch_read(struct file *f, char __user *buf, - size_t nbytes, loff_t *ppos) +static int dp_register_genl(void) { - int listeners = get_listen_mask(f); - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp = get_dp_locked(dp_idx); - struct sk_buff *skb; - struct iovec iov; - int retval; - - if (!dp) - return -ENODEV; - - if (nbytes == 0 || !listeners) - return 0; + int n_registered; + int err; + int i; - for (;;) { - int i; + n_registered = 0; + for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { + const struct genl_family_and_ops *f = &dp_genl_families[i]; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i)) { - skb = skb_dequeue(&dp->queues[i]); - if (skb) - goto success; - } - } - - if (f->f_flags & O_NONBLOCK) { - retval = -EAGAIN; + err = genl_register_family_with_ops(f->family, f->ops, + f->n_ops); + if (err) goto error; - } + n_registered++; - wait_event_interruptible(dp->waitqueue, - dp_has_packet_of_interest(dp, - listeners)); - - if (signal_pending(current)) { - retval = -ERESTARTSYS; - goto error; + if (f->group) { + err = genl_register_mc_group(f->family, f->group); + if (err) + goto error; } } -success: - mutex_unlock(&dp->mutex); - iov.iov_base = buf; - iov.iov_len = min_t(size_t, skb->len, nbytes); - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); - if (!retval) - retval = skb->len; - - kfree_skb(skb); - return retval; + return 0; error: - mutex_unlock(&dp->mutex); - return retval; -} - -static unsigned int openvswitch_poll(struct file *file, poll_table *wait) -{ - int dp_idx = iminor(file->f_dentry->d_inode); - struct datapath *dp = get_dp_locked(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; - mutex_unlock(&dp->mutex); - } else { - mask = POLLIN | POLLRDNORM | POLLHUP; - } - return mask; + dp_unregister_genl(n_registered); + return err; } -static struct file_operations openvswitch_fops = { - .owner = THIS_MODULE, - .read = openvswitch_read, - .poll = openvswitch_poll, - .unlocked_ioctl = openvswitch_ioctl, -#ifdef CONFIG_COMPAT - .compat_ioctl = openvswitch_compat_ioctl, -#endif -}; - -static int major; - static int __init dp_init(void) { struct sk_buff *dummy_skb; @@ -2242,10 +2034,14 @@ static int __init dp_init(void) printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR); - err = flow_init(); + err = tnl_init(); if (err) goto error; + err = flow_init(); + if (err) + goto error_tnl_exit; + err = vport_init(); if (err) goto error_flow_exit; @@ -2254,7 +2050,7 @@ 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; @@ -2266,6 +2062,8 @@ error_vport_exit: vport_exit(); error_flow_exit: flow_exit(); +error_tnl_exit: + tnl_exit(); error: return err; } @@ -2273,10 +2071,11 @@ 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(); + tnl_exit(); } module_init(dp_init);