X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=f42ead18308c6f9610a875c37137de092401019f;hb=aaff4b55a7992ac47a143abb6bffac5c931f073a;hp=95449f643294c39507a5beaaa1ee763f6acd5884;hpb=c75d4dcf957f617b40ca89546545f075e2631226;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index 95449f643..f42ead183 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -17,9 +17,11 @@ #include #include #include +#include #include #include #include +#include #include #include #include @@ -56,52 +58,50 @@ 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. */ + +/* Protected by genl_mutex. */ static struct datapath __rcu *dps[256]; -static DEFINE_MUTEX(dp_mutex); static struct vport *new_vport(const struct vport_parms *); -/* Must be called with rcu_read_lock or dp_mutex. */ +/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */ struct datapath *get_dp(int dp_idx) { 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)); + lockdep_rtnl_is_held() || + lockdep_genl_is_held()); } EXPORT_SYMBOL_GPL(get_dp); -static struct datapath *get_dp_locked(int dp_idx) -{ - struct datapath *dp; - - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - if (dp) - mutex_lock(&dp->mutex); - mutex_unlock(&dp_mutex); - return dp; -} - +/* Must be called with 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. */ @@ -121,6 +121,7 @@ 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) @@ -172,6 +173,7 @@ nla_put_failure: return -EMSGSIZE; } +/* Caller must hold RTNL lock. */ static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; @@ -208,58 +210,33 @@ 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); 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; } +/* Called with RTNL lock. */ int dp_detach_port(struct vport *p) { - int err; - ASSERT_RTNL(); if (p->port_no != ODPP_LOCAL) @@ -267,15 +244,11 @@ int dp_detach_port(struct vport *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; + return vport_del(p); } /* Must be called with rcu_read_lock. */ @@ -314,7 +287,7 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) 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; @@ -385,11 +358,23 @@ 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; +#define PACKET_N_MC_GROUPS 16 + +static int packet_mc_group(struct datapath *dp, u8 cmd) +{ + BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS); + return jhash_2words(dp->dp_idx, cmd, 0) & (PACKET_N_MC_GROUPS - 1); +} + +/* 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; @@ -400,7 +385,7 @@ static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, 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; @@ -408,7 +393,7 @@ static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, nskb = skb->next; skb->next = NULL; - len = sizeof(struct odp_packet); + len = sizeof(struct odp_header); len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */ len += nla_total_size(skb->len); len += nla_total_size(FLOW_BUFSIZE); @@ -419,15 +404,15 @@ 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 = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd); 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); @@ -451,8 +436,9 @@ 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; @@ -468,19 +454,49 @@ err_kfree_skbs: return err; } +/* 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. + */ +static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS]; + +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 +}; + +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; - 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); @@ -501,9 +517,7 @@ int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_i } } - err = queue_control_packets(dp, skb, upcall_info); - wake_up_interruptible(&dp->waitqueue); - return err; + return queue_control_packets(dp, skb, upcall_info); err_kfree_skb: kfree_skb(skb); @@ -520,34 +534,27 @@ err: return err; } +/* Called with genl_mutex. */ static int flush_flows(int dp_idx) { struct tbl *old_table; struct tbl *new_table; struct datapath *dp; - int err; - dp = get_dp_locked(dp_idx); - err = -ENODEV; + dp = get_dp(dp_idx); if (!dp) - goto exit; + return -ENODEV; old_table = get_table_protected(dp); new_table = tbl_create(TBL_MIN_BUCKETS); - err = -ENOMEM; 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; + return 0; } static int validate_actions(const struct nlattr *actions, u32 actions_len) @@ -652,6 +659,7 @@ static void clear_stats(struct sw_flow *flow) flow->byte_count = 0; } +/* Called with genl_mutex. */ static int expand_table(struct datapath *dp) { struct tbl *old_table = get_table_protected(dp); @@ -667,64 +675,34 @@ static int expand_table(struct datapath *dp) return 0; } -static const struct nla_policy execute_policy[ODP_PACKET_ATTR_MAX + 1] = { - [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, - [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, -}; - -static int execute_packet(const struct odp_upcall __user *uodp_upcall) +static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_PACKET_ATTR_MAX + 1]; - struct odp_upcall *odp_upcall; - struct sk_buff *skb, *packet; + struct odp_header *odp_header = info->userhdr; + struct nlattr **a = info->attrs; + struct sk_buff *packet; unsigned int actions_len; struct nlattr *actions; struct sw_flow_key key; struct datapath *dp; struct ethhdr *eth; bool is_frag; - u32 len; int err; - if (get_user(len, &uodp_upcall->len)) - return -EFAULT; - if (len < sizeof(struct odp_upcall)) - return -EINVAL; - - skb = alloc_skb(len, GFP_KERNEL); - if (!skb) - return -ENOMEM; - - err = -EFAULT; - if (copy_from_user(__skb_put(skb, len), uodp_upcall, len)) - goto exit_free_skb; - - odp_upcall = (struct odp_upcall *)skb->data; - err = -EINVAL; - if (odp_upcall->len != len) - goto exit_free_skb; - - __skb_pull(skb, sizeof(struct odp_upcall)); - err = nla_parse(a, ODP_PACKET_ATTR_MAX, (struct nlattr *)skb->data, - skb->len, execute_policy); - if (err) - goto exit_free_skb; - err = -EINVAL; if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] || nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN) - goto exit_free_skb; + goto exit; actions = nla_data(a[ODP_PACKET_ATTR_ACTIONS]); actions_len = nla_len(a[ODP_PACKET_ATTR_ACTIONS]); err = validate_actions(actions, actions_len); if (err) - goto exit_free_skb; + goto exit; packet = skb_clone(skb, GFP_KERNEL); err = -ENOMEM; if (!packet) - goto exit_free_skb; + goto exit; packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]); packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]); @@ -741,20 +719,32 @@ static int execute_packet(const struct odp_upcall __user *uodp_upcall) err = flow_extract(packet, -1, &key, &is_frag); if (err) - goto exit_free_skb; + goto exit; rcu_read_lock(); - dp = get_dp(odp_upcall->dp_idx); + dp = get_dp(odp_header->dp_idx); err = -ENODEV; if (dp) err = execute_actions(dp, packet, &key, actions, actions_len); rcu_read_unlock(); -exit_free_skb: - kfree_skb(skb); +exit: return err; } +static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = { + [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; + +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; @@ -779,7 +769,9 @@ static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) } } -/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ +/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports. + * Called with RTNL lock. + */ int dp_min_mtu(const struct datapath *dp) { struct vport *p; @@ -787,7 +779,7 @@ int dp_min_mtu(const struct datapath *dp) ASSERT_RTNL(); - list_for_each_entry_rcu (p, &dp->port_list, node) { + list_for_each_entry (p, &dp->port_list, node) { int dev_mtu; /* Skip any internal ports, since that's what we're trying to @@ -803,8 +795,9 @@ int dp_min_mtu(const struct datapath *dp) 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. */ +/* 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 vport *p; @@ -814,22 +807,12 @@ void set_internal_devs_mtu(const struct datapath *dp) mtu = dp_min_mtu(dp); - list_for_each_entry_rcu (p, &dp->port_list, node) { + list_for_each_entry (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 void set_listen_mask(struct file *f, int listen_mask) -{ - f->private_data = (void*)(long)listen_mask; -} - 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 }, @@ -837,6 +820,7 @@ static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = { [ODP_FLOW_ATTR_STATE] = { .type = NLA_U64 }, }; + static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, struct sw_flow *flow, u32 total_len, u64 state) { @@ -850,14 +834,13 @@ static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, 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; - 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; @@ -867,7 +850,7 @@ static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, goto nla_put_failure; err = flow_to_nlattrs(&flow->key, skb); if (err) - goto exit_unlock; + goto exit_free; nla_nest_end(skb, nla); nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS); @@ -900,17 +883,17 @@ static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, odp_flow->len = skb->len; err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_unlock; + goto exit_free; nla_put_failure: err = -EMSGSIZE; -exit_unlock: - rcu_read_unlock(); +exit_free: kfree_skb(skb); exit: return err; } +/* Called with genl_mutex. */ static struct sk_buff *copy_flow_from_user(struct odp_flow __user *uodp_flow, struct dp_flowcmd *flowcmd) { @@ -995,10 +978,10 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) if (IS_ERR(skb)) goto exit; - dp = get_dp_locked(flowcmd.dp_idx); + dp = get_dp(flowcmd.dp_idx); error = -ENODEV; if (!dp) - goto error_kfree_skb; + goto exit; hash = flow_hash(&flowcmd.key); table = get_table_protected(dp); @@ -1009,13 +992,13 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) /* Bail out if we're not allowed to create a new flow. */ error = -ENOENT; if (cmd == ODP_FLOW_SET) - goto error_unlock_dp; + goto exit; /* Expand table, if necessary, to make room. */ if (tbl_count(table) >= tbl_n_buckets(table)) { error = expand_table(dp); if (error) - goto error_unlock_dp; + goto exit; table = get_table_protected(dp); } @@ -1023,7 +1006,7 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) flow = flow_alloc(); if (IS_ERR(flow)) { error = PTR_ERR(flow); - goto error_unlock_dp; + goto exit; } flow->key = flowcmd.key; clear_stats(flow); @@ -1060,7 +1043,7 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) /* Update actions. */ flow = flow_cast(flow_node); old_acts = rcu_dereference_protected(flow->sf_acts, - lockdep_is_held(&dp->mutex)); + lockdep_genl_is_held()); if (flowcmd.actions && (old_acts->actions_len != flowcmd.actions_len || memcmp(old_acts->actions, flowcmd.actions, @@ -1088,13 +1071,10 @@ static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) } } kfree_skb(skb); - mutex_unlock(&dp->mutex); return 0; error_free_flow: flow_put(flow); -error_unlock_dp: - mutex_unlock(&dp->mutex); error_kfree_skb: kfree_skb(skb); exit: @@ -1112,25 +1092,22 @@ static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) int err; skb = copy_flow_from_user(uodp_flow, &flowcmd); - err = PTR_ERR(skb); if (IS_ERR(skb)) - goto exit; + return PTR_ERR(skb); - dp = get_dp_locked(flowcmd.dp_idx); - err = -ENODEV; + dp = get_dp(flowcmd.dp_idx); 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; + return -ENOENT; if (cmd == ODP_FLOW_DEL) { err = tbl_remove(table, flow_node); if (err) - goto exit_unlock_dp; + return err; } flow = flow_cast(flow_node); @@ -1138,11 +1115,6 @@ static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) if (!err && cmd == ODP_FLOW_DEL) flow_deferred_free(flow); -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_kfree_skb: - kfree_skb(skb); -exit: return err; } @@ -1161,51 +1133,61 @@ static int dump_flow(struct odp_flow __user *uodp_flow) if (IS_ERR(skb)) goto exit; - dp = get_dp_locked(flowcmd.dp_idx); + dp = get_dp(flowcmd.dp_idx); err = -ENODEV; if (!dp) - goto exit_free; + goto exit_kfree_skb; bucket = flowcmd.state >> 32; obj = flowcmd.state; - flow_node = tbl_next(dp->table, &bucket, &obj); + flow_node = tbl_next(get_table_protected(dp), &bucket, &obj); err = -ENODEV; if (!flow_node) - goto exit_unlock_dp; + goto exit_kfree_skb; flow = flow_cast(flow_node); err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, ((u64)bucket << 32) | obj); -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_free: +exit_kfree_skb: kfree_skb(skb); exit: return err; } static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, +#endif [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 }, [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 }, }; -static int copy_datapath_to_user(void __user *dst, struct datapath *dp, uint32_t total_len) +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 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 odp_datapath *odp_datapath; - struct sk_buff *skb; + struct odp_header *odp_header; struct nlattr *nla; int err; - skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); - err = -ENOMEM; - if (!skb) - goto exit; + odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, + flags, cmd); + if (!odp_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; + odp_header->dp_idx = dp->dp_idx; rcu_read_lock(); err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp)); @@ -1224,105 +1206,76 @@ static int copy_datapath_to_user(void __user *dst, struct datapath *dp, uint32_t if (dp->sflow_probability) NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability); - if (skb->len > total_len) + 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); - odp_datapath->len = skb->len; - err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_free_skb; + return genlmsg_end(skb, odp_header); nla_put_failure: - err = -EMSGSIZE; -exit_free_skb: - kfree_skb(skb); -exit: - return err; + genlmsg_cancel(skb, odp_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 *odp_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; + retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } + return skb; +} +static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ 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; + return -EINVAL; } - 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); + return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1); } -/* 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]) +/* 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]) { - WARN_ON_ONCE(!mutex_is_locked(&dp_mutex)); - if (!a[ODP_DP_ATTR_NAME]) { - struct datapath *dp; - - dp = get_dp(odp_datapath->dp_idx); + struct datapath *dp = get_dp(odp_header->dp_idx); if (!dp) return ERR_PTR(-ENODEV); - mutex_lock(&dp->mutex); return dp; } else { - struct datapath *dp; struct vport *vport; int dp_idx; - vport_lock(); + rcu_read_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(); + rcu_read_unlock(); if (dp_idx < 0) return ERR_PTR(-ENODEV); - - dp = get_dp(dp_idx); - mutex_lock(&dp->mutex); - return dp; + return vport->dp; } } +/* 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]) @@ -1331,35 +1284,31 @@ static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MA dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]); } -static int new_datapath(struct odp_datapath __user *uodp_datapath) +static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct odp_datapath *odp_datapath; + struct nlattr **a = info->attrs; + struct odp_header *odp_header = info->userhdr; 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)) - goto err; - odp_datapath = (struct odp_datapath *)skb->data; err = -EINVAL; if (!a[ODP_DP_ATTR_NAME]) - goto err_free_skb; + goto err; + + err = odp_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; + goto err_unlock_rtnl; - dp_idx = odp_datapath->dp_idx; + dp_idx = odp_header->dp_idx; if (dp_idx < 0) { err = -EFBIG; for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) { @@ -1380,12 +1329,7 @@ static int new_datapath(struct odp_datapath __user *uodp_datapath) 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. */ @@ -1422,13 +1366,18 @@ static int new_datapath(struct odp_datapath __user *uodp_datapath) change_datapath(dp, a); + 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; + rcu_assign_pointer(dps[dp_idx], dp); 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: @@ -1436,143 +1385,150 @@ err_destroy_local_port: 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_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 odp_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 = odp_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; + + 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; + + list_for_each_entry_safe (vport, next_vport, &dp->port_list, node) + if (vport->port_no != ODPP_LOCAL) + dp_detach_port(vport); + + dp_sysfs_del_dp(dp); + rcu_assign_pointer(dps[dp->dp_idx], NULL); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); + + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); - destroy_dp(dp); + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); err = 0; -exit_free: - kfree_skb(skb); - mutex_unlock(&dp_mutex); +exit_unlock: rtnl_unlock(); exit: return err; } -static int set_datapath(struct odp_datapath __user *uodp_datapath) +static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + 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; + err = odp_dp_cmd_validate(info->attrs); + if (err) + return err; - mutex_lock(&dp_mutex); - dp = lookup_datapath((struct odp_datapath *)skb->data, a); - err = PTR_ERR(dp); + dp = lookup_datapath(info->userhdr, info->attrs); if (IS_ERR(dp)) - goto exit_free; + return PTR_ERR(dp); - change_datapath(dp, a); - mutex_unlock(&dp->mutex); - err = 0; + change_datapath(dp, info->attrs); -exit_free: - kfree_skb(skb); - mutex_unlock(&dp_mutex); -exit: - return err; + 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; + } + + genl_notify(reply, genl_info_net(info), info->snd_pid, + dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -static int get_datapath(struct odp_datapath __user *uodp_datapath) +static int odp_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 *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 = odp_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 = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); + + return genlmsg_reply(reply, info); } -static int dump_datapath(struct odp_datapath __user *uodp_datapath) +static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - struct nlattr *a[ODP_DP_ATTR_MAX + 1]; - struct odp_datapath *odp_datapath; - struct sk_buff *skb; u32 dp_idx; - 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); - for (dp_idx = odp_datapath->dp_idx; dp_idx < ARRAY_SIZE(dps); dp_idx++) { + for (dp_idx = cb->args[0]; dp_idx < ARRAY_SIZE(dps); dp_idx++) { struct datapath *dp = get_dp(dp_idx); if (!dp) continue; - - 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; + 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; } - mutex_unlock(&dp_mutex); - err = -ENODEV; -exit_free: - kfree_skb(skb); -exit: - return err; + cb->args[0] = dp_idx; + 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] = { [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, @@ -1583,7 +1539,8 @@ static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = { [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, }; -static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t total_len) +/* Called with RCU read lock. */ +static struct sk_buff *odp_vport_build_info(struct vport *vport, uint32_t total_len) { struct odp_vport *odp_vport; struct sk_buff *skb; @@ -1591,12 +1548,11 @@ static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t to int ifindex, iflink; int err; - skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); + skb = alloc_skb(NLMSG_GOODSIZE, GFP_ATOMIC); err = -ENOMEM; if (!skb) - goto exit; + goto err; - 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; @@ -1627,19 +1583,17 @@ static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t to err = -EMSGSIZE; if (skb->len > total_len) - goto exit_unlock; + goto err_free; odp_vport->len = skb->len; - err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; - goto exit_unlock; + return skb; nla_put_failure: err = -EMSGSIZE; -exit_unlock: - rcu_read_unlock(); +err_free: kfree_skb(skb); -exit: - return err; +err: + return ERR_PTR(err); } static struct sk_buff *copy_vport_from_user(struct odp_vport __user *uodp_vport, @@ -1684,10 +1638,7 @@ error_free_skb: return ERR_PTR(err); } - -/* Called without any locks (or with RTNL lock). - * Returns holding vport->dp->mutex. - */ +/* Called with RTNL lock or RCU read lock. */ static struct vport *lookup_vport(struct odp_vport *odp_vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { @@ -1695,30 +1646,9 @@ static struct vport *lookup_vport(struct odp_vport *odp_vport, 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]); @@ -1726,20 +1656,19 @@ static struct vport *lookup_vport(struct odp_vport *odp_vport, if (port_no >= DP_MAX_PORTS) return ERR_PTR(-EINVAL); - dp = get_dp_locked(odp_vport->dp_idx); + dp = get_dp(odp_vport->dp_idx); 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; @@ -1757,6 +1686,7 @@ static int attach_vport(struct odp_vport __user *uodp_vport) struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; struct odp_vport *odp_vport; struct vport_parms parms; + struct sk_buff *reply; struct vport *vport; struct sk_buff *skb; struct datapath *dp; @@ -1774,28 +1704,27 @@ static int attach_vport(struct odp_vport __user *uodp_vport) goto exit_kfree_skb; rtnl_lock(); - - dp = get_dp_locked(odp_vport->dp_idx); + dp = get_dp(odp_vport->dp_idx); 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) @@ -1812,7 +1741,7 @@ 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); @@ -1820,14 +1749,18 @@ static int attach_vport(struct odp_vport __user *uodp_vport) err = change_vport(vport, a); if (err) { dp_detach_port(vport); - goto exit_unlock_dp; + goto exit_unlock; } - err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); + reply = odp_vport_build_info(vport, odp_vport->total_len); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; + + err = copy_to_user(uodp_vport, reply->data, reply->len) ? -EFAULT : 0; + kfree_skb(reply); -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_unlock_rtnl: +exit_unlock: rtnl_unlock(); exit_kfree_skb: kfree_skb(skb); @@ -1859,7 +1792,6 @@ 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); rtnl_unlock(); @@ -1870,7 +1802,6 @@ exit: static int del_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) { struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; - struct datapath *dp; struct vport *vport; struct sk_buff *skb; int err; @@ -1883,17 +1814,9 @@ static int del_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) rtnl_lock(); vport = lookup_vport((struct odp_vport *)skb->data, a); err = PTR_ERR(vport); - if (IS_ERR(vport)) - goto exit_free; - dp = vport->dp; + if (!IS_ERR(vport)) + err = dp_detach_port(vport); - err = -EINVAL; - if (vport->port_no == ODPP_LOCAL) - goto exit_free; - - err = dp_detach_port(vport); - mutex_unlock(&dp->mutex); -exit_free: kfree_skb(skb); rtnl_unlock(); exit: @@ -1904,6 +1827,7 @@ static int get_vport(struct odp_vport __user *uodp_vport) { struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; struct odp_vport *odp_vport; + struct sk_buff *reply; struct vport *vport; struct sk_buff *skb; int err; @@ -1911,19 +1835,32 @@ static int get_vport(struct odp_vport __user *uodp_vport) skb = copy_vport_from_user(uodp_vport, a); err = PTR_ERR(skb); if (IS_ERR(skb)) - goto exit; + goto err; odp_vport = (struct odp_vport *)skb->data; + rcu_read_lock(); vport = lookup_vport(odp_vport, a); err = PTR_ERR(vport); if (IS_ERR(vport)) - goto exit_free; + goto err_unlock_rcu; + reply = odp_vport_build_info(vport, odp_vport->total_len); + rcu_read_unlock(); - err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); - mutex_unlock(&vport->dp->mutex); -exit_free: + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto err_kfree_skb; + + err = copy_to_user(uodp_vport, reply->data, reply->len) ? -EFAULT : 0; + kfree_skb(reply); kfree_skb(skb); -exit: + + return err; + +err_unlock_rcu: + rcu_read_unlock(); +err_kfree_skb: + kfree_skb(skb); +err: return err; } @@ -1939,64 +1876,57 @@ static int dump_vport(struct odp_vport __user *uodp_vport) skb = copy_vport_from_user(uodp_vport, a); err = PTR_ERR(skb); if (IS_ERR(skb)) - goto exit; + goto err; odp_vport = (struct odp_vport *)skb->data; - dp = get_dp_locked(odp_vport->dp_idx); + dp = get_dp(odp_vport->dp_idx); err = -ENODEV; if (!dp) - goto exit_free; + goto err_kfree_skb; port_no = 0; if (a[ODP_VPORT_ATTR_PORT_NO]) port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + + rcu_read_lock(); 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; - } + struct sk_buff *skb_out; + struct vport *vport; + int retval; + + vport = get_vport_protected(dp, port_no); + if (!vport) + continue; + + skb_out = odp_vport_build_info(vport, odp_vport->total_len); + rcu_read_unlock(); + + err = PTR_ERR(skb_out); + if (IS_ERR(skb_out)) + goto err_kfree_skb; + + retval = copy_to_user(uodp_vport, skb_out->data, skb_out->len); + kfree_skb(skb_out); + kfree_skb(skb); + + return retval ? -EFAULT : 0; } + rcu_read_unlock(); err = -ENODEV; -exit_unlock_dp: - mutex_unlock(&dp->mutex); -exit_free: +err_kfree_skb: kfree_skb(skb); -exit: +err: return err; } static long openvswitch_ioctl(struct file *f, unsigned int cmd, unsigned long argp) { - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp; - int listeners; int err; - /* Handle commands with special locking requirements up front. */ + genl_lock(); 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; @@ -2035,51 +1965,15 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, err = dump_flow((struct odp_flow __user *)argp); goto exit; - case ODP_EXECUTE: - err = execute_packet((struct odp_upcall __user *)argp); - goto exit; - } - - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto exit; - - switch (cmd) { - case ODP_GET_LISTEN_MASK: - err = put_user(get_listen_mask(f), (int __user *)argp); - break; - - 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; - default: err = -ENOIOCTLCMD; break; } - mutex_unlock(&dp->mutex); exit: + genl_unlock(); 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 long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) { @@ -2088,11 +1982,6 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned /* 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: @@ -2103,9 +1992,6 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned case ODP_FLOW_GET: case ODP_FLOW_SET: case ODP_FLOW_DUMP: - case ODP_SET_LISTEN_MASK: - case ODP_GET_LISTEN_MASK: - case ODP_EXECUTE: /* Ioctls that just need their pointer argument extended. */ return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); @@ -2115,94 +2001,74 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned } #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; +static struct file_operations openvswitch_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = openvswitch_ioctl, +#ifdef CONFIG_COMPAT + .compat_ioctl = openvswitch_compat_ioctl, +#endif +}; - if (!dp) - return -ENODEV; +static int major; - if (nbytes == 0 || !listeners) - return 0; +struct genl_family_and_ops { + struct genl_family *family; + struct genl_ops *ops; + int n_ops; + struct genl_multicast_group *group; +}; - for (;;) { - int i; +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_packet_genl_family, + dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), + NULL }, +}; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i)) { - skb = skb_dequeue(&dp->queues[i]); - if (skb) - goto success; - } - } +static void dp_unregister_genl(int n_families) +{ + int i; - if (f->f_flags & O_NONBLOCK) { - retval = -EAGAIN; - goto error; - } + for (i = 0; i < n_families; i++) { + genl_unregister_family(dp_genl_families[i].family); + } +} + +static int dp_register_genl(void) +{ + int n_registered; + int err; + int i; - wait_event_interruptible(dp->waitqueue, - dp_has_packet_of_interest(dp, - listeners)); + 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 (signal_pending(current)) { - retval = -ERESTARTSYS; + err = genl_register_family_with_ops(f->family, f->ops, + f->n_ops); + if (err) goto error; + n_registered++; + + 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; @@ -2228,8 +2094,14 @@ static int __init dp_init(void) if (err < 0) goto error_unreg_notifier; + err = dp_register_genl(); + if (err < 0) + goto error_unreg_chrdev; + return 0; +error_unreg_chrdev: + unregister_chrdev(major, "openvswitch"); error_unreg_notifier: unregister_netdevice_notifier(&dp_device_notifier); error_vport_exit: @@ -2243,6 +2115,7 @@ error: static void dp_cleanup(void) { rcu_barrier(); + dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); unregister_chrdev(major, "openvswitch"); unregister_netdevice_notifier(&dp_device_notifier); vport_exit();