X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=e2846f2099cea1f9c0dd5f058899f131a6b08f42;hb=6c222e55fa4222c6724094e1e7a0a69addf6b030;hp=ffa22e240500b381c15fff2fad3755e65647eef4;hpb=c19e653509deb3f1002c4ed99714e16549ec74e9;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index ffa22e240..e2846f209 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 @@ -42,67 +43,74 @@ #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 "vport-internal_dev.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: * - * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL - * lock first. + * Writes to device state (add/remove datapath, port, set operations on vports, + * etc.) are protected by RTNL. * - * It is safe to access the datapath and vport structures with just - * dp_mutex. + * 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. + * + * 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 __rcu *dps[ODP_MAX]; -static DEFINE_MUTEX(dp_mutex); +static LIST_HEAD(dps); 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_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); +/* Must be called with genl_mutex. */ static struct tbl *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. */ @@ -122,6 +130,7 @@ 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 vport *port, int event, unsigned int flags) @@ -173,6 +182,7 @@ nla_put_failure: return -EMSGSIZE; } +/* Caller must hold RTNL lock. */ static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; @@ -206,183 +216,36 @@ static struct kobj_type dp_ktype = { .release = release_dp }; -static int create_dp(int dp_idx, const char __user *devnamep) -{ - struct vport_parms parms; - char devname[IFNAMSIZ]; - struct vport *vport; - 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); - 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. */ - 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 = devname; - 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; - - goto err_destroy_table; - } - - dp->drop_frags = 0; - dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) { - err = -ENOMEM; - goto err_destroy_local_port; - } - - rcu_assign_pointer(dps[dp_idx], dp); - dp_sysfs_add_dp(dp); - - mutex_unlock(&dp->mutex); - mutex_unlock(&dp_mutex); - rtnl_unlock(); - - return 0; - -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: - mutex_unlock(&dp->mutex); - kfree(dp); -err_put_module: - module_put(THIS_MODULE); -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); -err: - return err; -} - 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); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); } -static int destroy_dp(int dp_idx) -{ - struct datapath *dp; - int err = 0; - struct vport *p, *n; - - rtnl_lock(); - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - if (!dp) { - err = -ENODEV; - goto out; - } - - mutex_lock(&dp->mutex); - - 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); - -out: - mutex_unlock(&dp_mutex); - rtnl_unlock(); - return err; -} - -/* 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); - dp->n_ports++; + list_add(&vport->node, &dp->port_list); dp_ifinfo_notify(RTM_NEWLINK, vport); } - vport_unlock(); return vport; } +/* Called with RTNL lock. */ int dp_detach_port(struct vport *p) { - int err; - ASSERT_RTNL(); if (p->port_no != ODPP_LOCAL) @@ -390,16 +253,11 @@ int dp_detach_port(struct vport *p) 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); /* Then destroy it. */ - vport_lock(); - err = vport_del(p); - vport_unlock(); - - return err; + return vport_del(p); } /* Must be called with rcu_read_lock. */ @@ -408,8 +266,6 @@ 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 sw_flow_actions *acts; - struct loop_counter *loop; int error; OVS_CB(skb)->vport = p; @@ -417,10 +273,11 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) if (!OVS_CB(skb)->flow) { struct sw_flow_key key; struct tbl_node *flow_node; + int key_len; bool is_frag; /* 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, &is_frag); if (unlikely(error)) { kfree_skb(skb); return; @@ -433,12 +290,12 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) } /* Look up flow. */ - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, - flow_hash(&key), flow_cmp); + flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len, + flow_hash(&key, key_len), flow_cmp); if (unlikely(!flow_node)) { struct dp_upcall_info upcall; - upcall.type = _ODPL_MISS_NR; + upcall.cmd = ODP_PACKET_CMD_MISS; upcall.key = &key; upcall.userdata = 0; upcall.sample_pool = 0; @@ -454,32 +311,7 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) 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(); + execute_actions(dp, skb); out: /* Update datapath statistics. */ @@ -500,7 +332,6 @@ 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); @@ -509,22 +340,113 @@ 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 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 +}; + +/* Generic Netlink multicast groups for upcalls. + * + * 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]; + +static u32 packet_mc_group(struct datapath *dp, u8 cmd) +{ + 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; +} + +static int packet_register_mc_groups(void) +{ + int i; + + for (i = 0; i < PACKET_N_MC_GROUPS; i++) { + struct genl_multicast_group *group = &packet_mc_groups[i]; + int error; + + sprintf(group->name, "packet%d", i); + error = genl_register_mc_group(&dp_packet_genl_family, group); + if (error) + return error; + } + return 0; +} + +int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info) +{ + struct dp_stats_percpu *stats; + int err; + + WARN_ON_ONCE(skb_shared(skb)); + + forward_ip_summed(skb); + + err = vswitch_skb_checksum_setup(skb); + if (err) + goto err_kfree_skb; + + /* 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; + } + } + + err = queue_control_packets(dp, skb, upcall_info); + if (err) + goto err; + + return 0; + +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; +} + +/* 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) { + u32 group = packet_mc_group(dp, upcall_info->cmd); struct sk_buff *nskb; - int port_no; int err; - if (OVS_CB(skb)->vport) - port_no = OVS_CB(skb)->vport->port_no; - else - port_no = ODPP_LOCAL; - do { - struct odp_packet *upcall; + struct odp_header *upcall; struct sk_buff *user_skb; /* to be queued to userspace */ struct nlattr *nla; unsigned int len; @@ -532,8 +454,14 @@ static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, nskb = skb->next; skb->next = NULL; - len = sizeof(struct odp_packet); - len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */ + err = vlan_deaccel_tag(skb); + if (unlikely(err)) + goto err_kfree_skbs; + + if (nla_attr_size(skb->len) > USHRT_MAX) + goto err_kfree_skbs; + + len = sizeof(struct odp_header); len += nla_total_size(skb->len); len += nla_total_size(FLOW_BUFSIZE); if (upcall_info->userdata) @@ -543,14 +471,14 @@ static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, if (upcall_info->actions_len) len += nla_total_size(upcall_info->actions_len); - user_skb = alloc_skb(len, GFP_ATOMIC); - if (!user_skb) + user_skb = genlmsg_new(len, GFP_ATOMIC); + if (!user_skb) { + netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS); 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); + upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd); + upcall->dp_ifindex = dp->dp_ifindex; nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY); flow_to_nlattrs(upcall_info->key, user_skb); @@ -575,8 +503,9 @@ static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, else skb_copy_bits(skb, 0, nla_data(nla), skb->len); - upcall->len = user_skb->len; - skb_queue_tail(&dp->queues[upcall_info->type], user_skb); + err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC); + if (err) + goto err_kfree_skbs; kfree_skb(skb); skb = nskb; @@ -592,63 +521,18 @@ err_kfree_skbs: return err; } -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(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; - - forward_ip_summed(skb); - - err = vswitch_skb_checksum_setup(skb); - if (err) - goto err_kfree_skb; - - /* 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; - } - } - - err = queue_control_packets(dp, skb, upcall_info); - wake_up_interruptible(&dp->waitqueue); - return err; - -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) +/* Called with genl_mutex. */ +static int flush_flows(int dp_ifindex) { - struct tbl *old_table = get_table_protected(dp); + struct tbl *old_table; struct tbl *new_table; + struct datapath *dp; + + dp = get_dp(dp_ifindex); + if (!dp) + return -ENODEV; + old_table = get_table_protected(dp); new_table = tbl_create(TBL_MIN_BUCKETS); if (!new_table) return -ENOMEM; @@ -660,64 +544,62 @@ static int flush_flows(struct datapath *dp) return 0; } -static int validate_actions(const struct nlattr *actions, u32 actions_len) +static int validate_actions(const struct nlattr *attr) { 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, + 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, }; int type = nla_type(a); - if (type > ODPAT_MAX || nla_len(a) != action_lens[type]) + if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type]) return -EINVAL; switch (type) { - case ODPAT_UNSPEC: + case ODP_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: + 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: /* No validation needed. */ break; - case ODPAT_OUTPUT: + case ODP_ACTION_ATTR_OUTPUT: if (nla_get_u32(a) >= DP_MAX_PORTS) return -EINVAL; break; - case ODPAT_SET_DL_TCI: + case ODP_ACTION_ATTR_SET_DL_TCI: if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) return -EINVAL; break; - case ODPAT_SET_NW_TOS: + case ODP_ACTION_ATTR_SET_NW_TOS: if (nla_get_u8(a) & INET_ECN_MASK) return -EINVAL; break; @@ -732,104 +614,369 @@ static int validate_actions(const struct nlattr *actions, u32 actions_len) return 0; } - -static struct sw_flow_actions *get_actions(const struct odp_flow *flow) +static void clear_stats(struct sw_flow *flow) { - struct sw_flow_actions *actions; - int error; + flow->used = 0; + flow->tcp_flags = 0; + flow->packet_count = 0; + flow->byte_count = 0; +} - actions = flow_actions_alloc(flow->actions_len); - error = PTR_ERR(actions); - if (IS_ERR(actions)) - goto error; +/* Called with genl_mutex. */ +static int expand_table(struct datapath *dp) +{ + struct tbl *old_table = get_table_protected(dp); + struct tbl *new_table; - error = -EFAULT; - if (copy_from_user(actions->actions, - (struct nlattr __user __force *)flow->actions, - flow->actions_len)) - goto error_free_actions; - error = validate_actions(actions->actions, actions->actions_len); - if (error) - goto error_free_actions; + new_table = tbl_expand(old_table); + if (IS_ERR(new_table)) + return PTR_ERR(new_table); - return actions; + rcu_assign_pointer(dp->table, new_table); + tbl_deferred_destroy(old_table, NULL); -error_free_actions: - kfree(actions); -error: - return ERR_PTR(error); + return 0; } -static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) +static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) { - if (flow->used) { - struct timespec offset_ts, used, now_mono; + struct odp_header *odp_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; - ktime_get_ts(&now_mono); - jiffies_to_timespec(jiffies - flow->used, &offset_ts); - set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec, - now_mono.tv_nsec - offset_ts.tv_nsec); + err = -EINVAL; + if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_KEY] || + !a[ODP_PACKET_ATTR_ACTIONS] || + nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN) + goto err; - stats->used_sec = used.tv_sec; - stats->used_nsec = used.tv_nsec; - } else { - stats->used_sec = 0; - stats->used_nsec = 0; + err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]); + if (err) + goto err; + + len = nla_len(a[ODP_PACKET_ATTR_PACKET]); + packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); + err = -ENOMEM; + if (!packet) + goto err; + skb_reserve(packet, NET_IP_ALIGN); + + memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len); + + 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) + packet->protocol = eth->h_proto; + else + packet->protocol = htons(ETH_P_802_2); + + /* 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, &is_frag); + if (err) + goto err_flow_put; + flow->tbl_node.hash = flow_hash(&flow->key, key_len); + + err = flow_metadata_from_nlattrs(&flow->key.eth.in_port, + &flow->key.eth.tun_id, + a[ODP_PACKET_ATTR_KEY]); + if (err) + goto err_flow_put; + + acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]); + err = PTR_ERR(acts); + if (IS_ERR(acts)) + goto err_flow_put; + rcu_assign_pointer(flow->sf_acts, acts); + + OVS_CB(packet)->flow = flow; + + rcu_read_lock(); + dp = get_dp(odp_header->dp_ifindex); + err = -ENODEV; + if (!dp) + goto err_unlock; + err = execute_actions(dp, packet); + 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[ODP_PACKET_ATTR_MAX + 1] = { + [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, + [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; + +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 } +}; + +static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) +{ + int i; + + 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; + + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); - stats->n_packets = flow->packet_count; - stats->n_bytes = flow->byte_count; - stats->reserved = 0; - stats->tcp_flags = flow->tcp_flags; - stats->error = 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 void clear_stats(struct sw_flow *flow) +/* 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) { - flow->used = 0; - flow->tcp_flags = 0; - flow->packet_count = 0; - flow->byte_count = 0; + struct vport *p; + int mtu = 0; + + ASSERT_RTNL(); + + list_for_each_entry (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 (!dev_mtu) + continue; + if (!mtu || dev_mtu < mtu) + mtu = dev_mtu; + } + + return mtu ? mtu : ETH_DATA_LEN; } -static int expand_table(struct datapath *dp) +/* 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) { - struct tbl *old_table = get_table_protected(dp); - struct tbl *new_table; + struct vport *p; + int mtu; - new_table = tbl_expand(old_table); - if (IS_ERR(new_table)) - return PTR_ERR(new_table); + ASSERT_RTNL(); - rcu_assign_pointer(dp->table, new_table); - tbl_deferred_destroy(old_table, NULL); + mtu = dp_min_mtu(dp); - return 0; + list_for_each_entry (p, &dp->port_list, node) { + if (is_internal_vport(p)) + vport_set_mtu(p, mtu); + } } -static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, - struct odp_flow_stats *stats) +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 }, +}; + +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 +}; + +static struct genl_multicast_group dp_flow_multicast_group = { + .name = ODP_FLOW_MCGROUP +}; + +/* 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) { + 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; + + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + + odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd); + if (!odp_header) + return -EMSGSIZE; + + odp_header->dp_ifindex = dp->dp_ifindex; + + 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); + + 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 (used) + NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used)); + + if (stats.n_packets) + NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats); + + if (tcp_flags) + NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags); + + /* 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; + + return genlmsg_end(skb, odp_header); + +nla_put_failure: + err = -EMSGSIZE; +error: + genlmsg_cancel(skb, odp_header); + return err; +} + +static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow) +{ + const struct sw_flow_actions *sf_acts; + int len; + + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + + 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 struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp, + u32 pid, u32 seq, u8 cmd) +{ + struct sk_buff *skb; + int retval; + + skb = odp_flow_cmd_alloc_info(flow); + if (!skb) + return ERR_PTR(-ENOMEM); + + retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); + BUG_ON(retval < 0); + return skb; +} + +static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) +{ + 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; - struct sw_flow_actions *acts = NULL; - int error; u32 hash; + int error; + int key_len; - error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key, - uf->flow.key_len); + /* Extract key. */ + error = -EINVAL; + if (!a[ODP_FLOW_ATTR_KEY]) + goto error; + error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]); if (error) - return error; + goto error; - hash = flow_hash(&key); + /* 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; + } + + dp = get_dp(odp_header->dp_ifindex); + error = -ENODEV; + if (!dp) + goto error; + + hash = flow_hash(&key, key_len); table = get_table_protected(dp); - flow_node = tbl_lookup(table, &key, hash, flow_cmp); + flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp); if (!flow_node) { - /* No such flow. */ + struct sw_flow_actions *acts; + + /* Bail out if we're not allowed to create a new flow. */ error = -ENOENT; - if (!(uf->flags & ODPPF_CREATE)) + if (info->genlhdr->cmd == ODP_FLOW_CMD_SET) goto error; /* Expand table, if necessary, to make room. */ @@ -850,7 +997,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, clear_stats(flow); /* Obtain actions. */ - acts = get_actions(&uf->flow); + acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]); error = PTR_ERR(acts); if (IS_ERR(acts)) goto error_free_flow; @@ -859,456 +1006,618 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, /* Put flow in bucket. */ error = tbl_insert(table, &flow->tbl_node, hash); if (error) - goto error_free_flow_acts; + goto error_free_flow; - memset(stats, 0, sizeof(struct odp_flow_stats)); + 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, *new_acts; - - flow = flow_cast(flow_node); - - /* Bail out if we're not allowed to modify an existing 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 (!(uf->flags & ODPPF_MODIFY)) - goto error; - - /* Swap actions. */ - new_acts = get_actions(&uf->flow); - error = PTR_ERR(new_acts); - if (IS_ERR(new_acts)) + if (info->genlhdr->cmd == ODP_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 (old_acts->actions_len != new_acts->actions_len || - memcmp(old_acts->actions, new_acts->actions, - old_acts->actions_len)) { + 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; + rcu_assign_pointer(flow->sf_acts, new_acts); flow_deferred_free_acts(old_acts); - } else { - kfree(new_acts); } - /* Fetch stats, then clear them if necessary. */ - spin_lock_bh(&flow->lock); - get_stats(flow, stats); - if (uf->flags & ODPPF_ZERO_STATS) + reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, ODP_FLOW_CMD_NEW); + + /* Clear stats. */ + if (a[ODP_FLOW_ATTR_CLEAR]) { + spin_lock_bh(&flow->lock); clear_stats(flow); - spin_unlock_bh(&flow->lock); + spin_unlock_bh(&flow->lock); + } } + 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_acts: - kfree(acts); error_free_flow: - flow->sf_acts = NULL; flow_put(flow); error: return error; } -static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) +static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct odp_flow_stats stats; - struct odp_flow_put uf; - int error; + 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; + int key_len; + + if (!a[ODP_FLOW_ATTR_KEY]) + return -EINVAL; + err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]); + if (err) + return err; - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) - return -EFAULT; + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - error = do_put_flow(dp, &uf, &stats); - if (error) - return error; + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len), + flow_cmp); + if (!flow_node) + return -ENOENT; - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; + 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 0; + return genlmsg_reply(reply, info); } -static int do_answer_query(struct datapath *dp, struct sw_flow *flow, - u32 query_flags, - struct odp_flow_stats __user *ustats, - struct nlattr __user *actions, - u32 __user *actions_lenp) +static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct sw_flow_actions *sf_acts; - struct odp_flow_stats stats; - u32 actions_len; + 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; + int key_len; - spin_lock_bh(&flow->lock); - get_stats(flow, &stats); - if (query_flags & ODPFF_ZERO_TCP_FLAGS) - flow->tcp_flags = 0; + if (!a[ODP_FLOW_ATTR_KEY]) + return flush_flows(odp_header->dp_ifindex); + err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]); + if (err) + return err; - spin_unlock_bh(&flow->lock); + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(actions_len, actions_lenp)) - return -EFAULT; + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len), + flow_cmp); + if (!flow_node) + return -ENOENT; + flow = flow_cast(flow_node); - if (!actions_len) - return 0; + reply = odp_flow_cmd_alloc_info(flow); + if (!reply) + return -ENOMEM; - sf_acts = rcu_dereference_protected(flow->sf_acts, - lockdep_is_held(&dp->mutex)); - if (put_user(sf_acts->actions_len, actions_lenp) || - (actions && copy_to_user(actions, sf_acts->actions, - min(sf_acts->actions_len, actions_len)))) - return -EFAULT; + err = tbl_remove(table, flow_node); + if (err) { + kfree_skb(reply); + return err; + } + err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, + info->snd_seq, 0, ODP_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 answer_query(struct datapath *dp, struct sw_flow *flow, - u32 query_flags, struct odp_flow __user *ufp) +static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - struct nlattr __user *actions; + struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; - if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions)) - return -EFAULT; + dp = get_dp(odp_header->dp_ifindex); + if (!dp) + return -ENODEV; - return do_answer_query(dp, flow, query_flags, - &ufp->stats, actions, &ufp->actions_len); -} + for (;;) { + struct tbl_node *flow_node; + struct sw_flow *flow; + u32 bucket, obj; -static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len) -{ - struct tbl *table = get_table_protected(dp); - struct tbl_node *flow_node; - struct sw_flow_key swkey; - int error; + bucket = cb->args[0]; + obj = cb->args[1]; + flow_node = tbl_next(get_table_protected(dp), &bucket, &obj); + if (!flow_node) + break; - error = flow_copy_from_user(&swkey, key, key_len); - if (error) - return ERR_PTR(error); + 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; - flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp); - if (!flow_node) - return ERR_PTR(-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, + }, +}; - error = tbl_remove(table, flow_node); - if (error) - return ERR_PTR(error); +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 }, +}; - /* 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_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 del_flow(struct datapath *dp, struct odp_flow __user *ufp) +static struct genl_multicast_group dp_datapath_multicast_group = { + .name = ODP_DATAPATH_MCGROUP +}; + +static int odp_dp_cmd_fill_info(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; + struct odp_header *odp_header; + struct nlattr *nla; + int err; - if (copy_from_user(&uf, ufp, sizeof(uf))) - return -EFAULT; + odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, + flags, cmd); + if (!odp_header) + goto error; - flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); - if (IS_ERR(flow)) - return PTR_ERR(flow); + odp_header->dp_ifindex = dp->dp_ifindex; - error = answer_query(dp, flow, 0, ufp); - flow_deferred_free(flow); - return error; -} + 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_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) -{ - struct tbl *table = get_table_protected(dp); - u32 i; + 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)); - for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i]; - struct sw_flow_key key; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; + NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS, + dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO); - if (copy_from_user(&uf, ufp, sizeof(uf))) - return -EFAULT; + if (dp->sflow_probability) + NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability); - error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf.key, uf.key_len); - if (error) - return error; + 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); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp); - if (error) - return -EFAULT; - } - return flowvec->n_flows; + return genlmsg_end(skb, odp_header); + +nla_put_failure: + genlmsg_cancel(skb, odp_header); +error: + return -EMSGSIZE; } -static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - const struct odp_flowvec *)) +static struct sk_buff *odp_dp_cmd_build_info(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 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); + 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_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } + return skb; } -static struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state) +static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1]) { - struct tbl *table = get_table_protected(dp); - struct tbl_node *tbl_node; - u32 bucket, obj; + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - if (get_user(bucket, &state[0]) || get_user(obj, &state[1])) - return ERR_PTR(-EFAULT); - - tbl_node = tbl_next(table, &bucket, &obj); - - if (put_user(bucket, &state[0]) || put_user(obj, &state[1])) - return ERR_PTR(-EFAULT); + if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP) + return -EINVAL; + } - return tbl_node ? flow_cast(tbl_node) : NULL; + return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1); } -static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp) +/* 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_flow __user *uflowp; - struct nlattr __user *ukey; - struct sw_flow *flow; - u32 key_len; - - flow = do_dump_flow(dp, udumpp->state); - if (IS_ERR(flow)) - return PTR_ERR(flow); - - if (get_user(uflowp, (struct odp_flow __user *__user*)&udumpp->flow)) - return -EFAULT; - - if (!flow) - return put_user(ODPFF_EOF, &uflowp->flags); + struct datapath *dp; - if (put_user(0, &uflowp->flags) || - get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) || - get_user(key_len, &uflowp->key_len)) - return -EFAULT; + if (!a[ODP_DP_ATTR_NAME]) + dp = get_dp(odp_header->dp_ifindex); + else { + struct vport *vport; - key_len = flow_copy_to_user(ukey, &flow->key, key_len); - if (key_len < 0) - return key_len; - if (put_user(key_len, &uflowp->key_len)) - return -EFAULT; + 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); +} - return answer_query(dp, flow, 0, uflowp); +/* Called with genl_mutex. */ +static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ + 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 do_execute(struct datapath *dp, const struct odp_execute *execute) +static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) { - struct sw_flow_key key; - struct sk_buff *skb; - struct sw_flow_actions *actions; - struct ethhdr *eth; - bool is_frag; + struct nlattr **a = info->attrs; + struct vport_parms parms; + struct sk_buff *reply; + struct datapath *dp; + struct vport *vport; int err; err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) - goto error; + if (!a[ODP_DP_ATTR_NAME]) + goto err; - actions = flow_actions_alloc(execute->actions_len); - if (IS_ERR(actions)) { - err = PTR_ERR(actions); - goto error; - } + err = odp_dp_cmd_validate(a); + if (err) + goto err; - err = -EFAULT; - if (copy_from_user(actions->actions, - (struct nlattr __user __force *)execute->actions, execute->actions_len)) - goto error_free_actions; + rtnl_lock(); + err = -ENODEV; + if (!try_module_get(THIS_MODULE)) + goto err_unlock_rtnl; - err = validate_actions(actions->actions, execute->actions_len); - if (err) - goto error_free_actions; + 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; - skb = alloc_skb(execute->length, GFP_KERNEL); - if (!skb) - goto error_free_actions; + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); + if (!dp->table) + goto err_free_dp; - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), - (const void __user __force *)execute->data, - execute->length)) - goto error_free_skb; + /* 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; - skb_reset_mac_header(skb); - eth = eth_hdr(skb); + goto err_destroy_table; + } + dp->dp_ifindex = vport_get_ifindex(vport); - /* 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); + dp->drop_frags = 0; + dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); + if (!dp->stats_percpu) { + err = -ENOMEM; + goto err_destroy_local_port; + } - err = flow_extract(skb, -1, &key, &is_frag); - if (err) - goto error_free_skb; + change_datapath(dp, a); - rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); - rcu_read_unlock(); + 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; - kfree(actions); - return err; + list_add_tail(&dp->list_node, &dps); + dp_sysfs_add_dp(dp); -error_free_skb: - kfree_skb(skb); -error_free_actions: - kfree(actions); -error: + 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)); +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; } -static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep) +static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) { - struct odp_execute execute; + struct vport *vport, *next_vport; + struct sk_buff *reply; + struct datapath *dp; + int err; - if (copy_from_user(&execute, executep, sizeof(execute))) - return -EFAULT; + err = odp_dp_cmd_validate(info->attrs); + if (err) + goto exit; - return do_execute(dp, &execute); -} + rtnl_lock(); + dp = lookup_datapath(info->userhdr, info->attrs); + err = PTR_ERR(dp); + if (IS_ERR(dp)) + goto exit_unlock; -static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) -{ - struct tbl *table = get_table_protected(dp); - struct odp_stats stats; - int i; + 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; - 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.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; + list_for_each_entry_safe (vport, next_vport, &dp->port_list, node) + if (vport->port_no != ODPP_LOCAL) + dp_detach_port(vport); - percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + dp_sysfs_del_dp(dp); + list_del(&dp->list_node); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); - do { - seqcount = read_seqcount_begin(&percpu_stats->seqlock); - local_stats = *percpu_stats; - } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); + /* 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(); - 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; - } - 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; + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); + + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + + return 0; + +exit_unlock: + rtnl_unlock(); +exit: + return err; } -/* 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_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) { - struct vport *p; - int mtu = 0; + struct sk_buff *reply; + struct datapath *dp; + int err; - ASSERT_RTNL(); + err = odp_dp_cmd_validate(info->attrs); + if (err) + return err; - list_for_each_entry_rcu (p, &dp->port_list, node) { - int dev_mtu; + dp = lookup_datapath(info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - /* Skip any internal ports, since that's what we're trying to - * set. */ - if (is_internal_vport(p)) - continue; + change_datapath(dp, info->attrs); - dev_mtu = vport_get_mtu(p); - if (!mtu || dev_mtu < mtu) - mtu = dev_mtu; + 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 mtu ? mtu : ETH_DATA_LEN; + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -/* 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_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) { - struct vport *p; - int mtu; + struct sk_buff *reply; + struct datapath *dp; + int err; - ASSERT_RTNL(); + err = odp_dp_cmd_validate(info->attrs); + if (err) + return err; - mtu = dp_min_mtu(dp); + dp = lookup_datapath(info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (is_internal_vport(p)) - vport_set_mtu(p, mtu); - } -} + 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); -static int get_listen_mask(const struct file *f) -{ - return (long)f->private_data; + return genlmsg_reply(reply, info); } -static void set_listen_mask(struct file *f, int listen_mask) +static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - f->private_data = (void*)(long)listen_mask; -} + struct datapath *dp; + int skip = cb->args[0]; + int i = 0; + + 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++; + } + + 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 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 }, }; -static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t total_len) +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 +}; + +static struct genl_multicast_group dp_vport_multicast_group = { + .name = ODP_VPORT_MCGROUP +}; + +/* 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_vport *odp_vport; - struct sk_buff *skb; + struct odp_header *odp_header; struct nlattr *nla; int ifindex, iflink; + int mtu; int err; - skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); - err = -ENOMEM; - if (!skb) - goto exit; + odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, + flags, cmd); + if (!odp_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; + odp_header->dp_ifindex = vport->dp->dp_ifindex; NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no); NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport)); @@ -1322,9 +1631,13 @@ static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t to NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); - NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport)); + mtu = vport_get_mtu(vport); + if (mtu) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu); err = vport_get_options(vport, skb); + if (err == -EMSGSIZE) + goto error; ifindex = vport_get_ifindex(vport); if (ifindex > 0) @@ -1334,121 +1647,70 @@ static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t to 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, odp_header); nla_put_failure: err = -EMSGSIZE; -exit_unlock: - rcu_read_unlock(); - kfree_skb(skb); -exit: +error: + genlmsg_cancel(skb, odp_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. */ +static struct sk_buff *odp_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 = odp_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 odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + return CHECK_NUL_STRING(a[ODP_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, +/* 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 (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 (!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]); 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(odp_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); } +/* Called with RTNL lock. */ static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { int err = 0; @@ -1461,50 +1723,47 @@ static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX return err; } -static int attach_vport(struct odp_vport __user *uodp_vport) +static int odp_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 odp_header *odp_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)) - 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; + goto exit; - rtnl_lock(); + err = odp_vport_cmd_validate(a); + if (err) + goto exit; - dp = get_dp_locked(odp_vport->dp_idx); + rtnl_lock(); + dp = get_dp(odp_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]); 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) @@ -1521,46 +1780,48 @@ static int attach_vport(struct odp_vport __user *uodp_vport) 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 = 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_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 odp_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 = odp_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]) @@ -1568,585 +1829,217 @@ static int set_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) if (!err) err = change_vport(vport, a); - mutex_unlock(&vport->dp->mutex); -exit_free: - kfree_skb(skb); + 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; + } + + 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 odp_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 = odp_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 == ODPP_LOCAL) { + err = -EINVAL; + goto exit_unlock; + } + + 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; 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 odp_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 odp_header *odp_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 = odp_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(odp_header, a); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_free; - - 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; -} - -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; + goto exit_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; + 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; - dp = get_dp_locked(odp_vport->dp_idx); - err = -ENODEV; - if (!dp) - goto exit_free; + rcu_read_unlock(); - 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; + return genlmsg_reply(reply, info); -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 odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - int dp_idx = iminor(f->f_dentry->d_inode); + struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh)); struct datapath *dp; - int drop_frags, listeners; - unsigned int sflow_probability; - 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; - - case ODP_DP_DESTROY: - err = destroy_dp(dp_idx); - 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; - } + u32 port_no; + int retval; - dp = get_dp_locked(dp_idx); - err = -ENODEV; + dp = get_dp(odp_header->dp_ifindex); if (!dp) - goto exit; - - switch (cmd) { - case ODP_DP_STATS: - err = get_dp_stats(dp, (struct odp_stats __user *)argp); - break; - - case ODP_GET_DROP_FRAGS: - err = put_user(dp->drop_frags, (int __user *)argp); - break; + return -ENODEV; - 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; + rcu_read_lock(); + for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) { + struct vport *vport; - case ODP_GET_LISTEN_MASK: - err = put_user(get_listen_mask(f), (int __user *)argp); - break; + vport = get_vport_protected(dp, port_no); + if (!vport) + continue; - case ODP_SET_LISTEN_MASK: - err = get_user(listeners, (int __user *)argp); - if (err) - break; - err = -EINVAL; - if (listeners & ~ODPL_ALL) + 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; - 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_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_DUMP: - err = dump_flow(dp, (struct odp_flow_dump __user *)argp); - break; - - case ODP_EXECUTE: - err = execute_packet(dp, (struct odp_execute __user *)argp); - break; - - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); -exit: - return err; -} - -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; } - return 0; -} - -#ifdef CONFIG_COMPAT -static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) -{ - compat_uptr_t key, actions; - - if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) || - __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) || - __get_user(key, &compat->key) || - __get_user(flow->key_len, &compat->key_len) || - __get_user(actions, &compat->actions) || - __get_user(flow->actions_len, &compat->actions_len) || - __get_user(flow->flags, &compat->flags)) - return -EFAULT; - - flow->key = (struct nlattr __force *)compat_ptr(key); - flow->actions = (struct nlattr __force *)compat_ptr(actions); - return 0; -} - -static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp) -{ - struct odp_flow_stats stats; - struct odp_flow_put fp; - int error; - - if (compat_get_flow(&fp.flow, &ufp->flow) || - get_user(fp.flags, &ufp->flags)) - return -EFAULT; - - error = do_put_flow(dp, &fp, &stats); - if (error) - return error; - - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; - - return 0; -} - -static int compat_answer_query(struct datapath *dp, struct sw_flow *flow, - u32 query_flags, - struct compat_odp_flow __user *ufp) -{ - compat_uptr_t actions; + rcu_read_unlock(); - if (get_user(actions, &ufp->actions)) - return -EFAULT; + cb->args[0] = port_no; + retval = skb->len; - return do_answer_query(dp, flow, query_flags, &ufp->stats, - compat_ptr(actions), &ufp->actions_len); + return retval; } -static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp) -{ - struct sw_flow *flow; - struct odp_flow uf; - int error; - - if (compat_get_flow(&uf, ufp)) - return -EFAULT; +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, + }, +}; - flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); - if (IS_ERR(flow)) - return PTR_ERR(flow); +struct genl_family_and_ops { + struct genl_family *family; + struct genl_ops *ops; + int n_ops; + struct genl_multicast_group *group; +}; - error = compat_answer_query(dp, flow, 0, ufp); - flow_deferred_free(flow); - 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 int compat_query_flows(struct datapath *dp, - struct compat_odp_flow __user *flows, - u32 n_flows) +static void dp_unregister_genl(int n_families) { - struct tbl *table = get_table_protected(dp); - u32 i; - - for (i = 0; i < n_flows; i++) { - struct compat_odp_flow __user *ufp = &flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - struct sw_flow_key key; - int error; - - if (compat_get_flow(&uf, ufp)) - return -EFAULT; - - error = flow_copy_from_user(&key, (const struct nlattr __force __user *) uf.key, uf.key_len); - if (error) - return error; + int i; - flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = compat_answer_query(dp, flow_cast(flow_node), - uf.flags, ufp); - if (error) - return -EFAULT; + for (i = 0; i < n_families; i++) { + genl_unregister_family(dp_genl_families[i].family); } - return n_flows; -} - -static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp) -{ - struct compat_odp_flow __user *uflowp; - compat_uptr_t compat_ufp; - struct sw_flow *flow; - compat_uptr_t ukey; - u32 key_len; - - flow = do_dump_flow(dp, udumpp->state); - if (IS_ERR(flow)) - return PTR_ERR(flow); - - if (get_user(compat_ufp, &udumpp->flow)) - return -EFAULT; - uflowp = compat_ptr(compat_ufp); - - if (!flow) - return put_user(ODPFF_EOF, &uflowp->flags); - - if (put_user(0, &uflowp->flags) || - get_user(ukey, &uflowp->key) || - get_user(key_len, &uflowp->key_len)) - return -EFAULT; - - key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len); - if (key_len < 0) - return key_len; - if (put_user(key_len, &uflowp->key_len)) - return -EFAULT; - - return compat_answer_query(dp, flow, 0, uflowp); -} - -static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - struct compat_odp_flow __user *, - u32 n_flows)) -{ - struct compat_odp_flowvec __user *uflowvec; - struct compat_odp_flow __user *flows; - struct compat_odp_flowvec flowvec; - int retval; - - uflowvec = compat_ptr(argp); - if (!access_ok(VERIFY_WRITE, uflowvec, sizeof(*uflowvec)) || - copy_from_user(&flowvec, uflowvec, sizeof(flowvec))) - return -EFAULT; - - if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow)) - return -EINVAL; - - flows = compat_ptr(flowvec.flows); - if (!access_ok(VERIFY_WRITE, flows, - flowvec.n_flows * sizeof(struct compat_odp_flow))) - return -EFAULT; - - 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(actions, &uexecute->actions) || - __get_user(execute.actions_len, &uexecute->actions_len) || - __get_user(data, &uexecute->data) || - __get_user(execute.length, &uexecute->length)) - return -EFAULT; - - execute.actions = (struct nlattr __force *)compat_ptr(actions); - execute.data = (const void __force *)compat_ptr(data); - - return do_execute(dp, &execute); } -static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) +static int dp_register_genl(void) { - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp; + int n_registered; int err; + int i; - 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_VPORT_NEW: - case ODP_VPORT_DEL: - case ODP_VPORT_GET: - case ODP_VPORT_SET: - case ODP_VPORT_DUMP: - 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: - /* Ioctls that just need their pointer argument extended. */ - return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); - } - - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto exit; - - switch (cmd) { - case ODP_FLOW_PUT32: - err = compat_put_flow(dp, compat_ptr(argp)); - break; - - case ODP_FLOW_DEL32: - err = compat_del_flow(dp, compat_ptr(argp)); - break; - - case ODP_FLOW_GET32: - err = compat_flowvec_ioctl(dp, argp, compat_query_flows); - break; - - case ODP_FLOW_DUMP32: - err = compat_dump_flow(dp, compat_ptr(argp)); - break; - - case ODP_EXECUTE32: - err = compat_execute(dp, compat_ptr(argp)); - break; - - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); -exit: - return err; -} -#endif - -static ssize_t openvswitch_read(struct file *f, char __user *buf, - size_t nbytes, loff_t *ppos) -{ - 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; - - for (;;) { - int i; - - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i)) { - skb = skb_dequeue(&dp->queues[i]); - if (skb) - goto success; - } - } + n_registered = 0; + for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { + const struct genl_family_and_ops *f = &dp_genl_families[i]; - 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; + err = packet_register_mc_groups(); + if (err) + goto error; + 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; @@ -2168,7 +2061,7 @@ 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; @@ -2187,7 +2080,7 @@ 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();