X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=dc2cfad584b793f3dc11aae8a3a90fd7e5fe9bdd;hb=51c82a49d58daebe289e045fe44009d59b1f9236;hp=cc7672058cecbd309445bddf9d629fd5f84fd00b;hpb=2d7ce2ee1af5c4925ebbc681bb72e660d87faa71;p=openvswitch diff --git a/datapath/datapath.c b/datapath/datapath.c index cc767205..dc2cfad5 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -1,25 +1,34 @@ /* - * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks. - * Distributed under the terms of the GNU GPL version 2. + * Copyright (c) 2007-2012 Nicira, Inc. * - * Significant portions of this file may be copied from parts of the Linux - * kernel, by Linus Torvalds and others. + * This program is free software; you can redistribute it and/or + * modify it under the terms of version 2 of the GNU General Public + * License as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA + * 02110-1301, USA */ -/* Functions for managing the dp interface/device. */ - #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt #include #include -#include #include #include #include #include +#include #include #include #include +#include #include #include #include @@ -30,101 +39,129 @@ #include #include #include -#include #include -#include #include #include #include #include #include +#include #include #include -#include -#include +#include +#include +#include -#include "openvswitch/datapath-protocol.h" #include "checksum.h" #include "datapath.h" -#include "actions.h" #include "flow.h" -#include "loop_counter.h" -#include "odp-compat.h" -#include "table.h" +#include "genl_exec.h" +#include "vlan.h" +#include "tunnel.h" #include "vport-internal_dev.h" -#include "compat.h" +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \ + LINUX_VERSION_CODE >= KERNEL_VERSION(3,5,0) +#error Kernels before 2.6.18 or after 3.4 are not supported by this version of Open vSwitch. +#endif + +#define REHASH_FLOW_INTERVAL (10 * 60 * HZ) +static void rehash_flow_table(struct work_struct *work); +static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table); + +int ovs_net_id __read_mostly; -int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); -EXPORT_SYMBOL(dp_ioctl_hook); +int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); +EXPORT_SYMBOL(ovs_dp_ioctl_hook); -/* Datapaths. Protected on the read side by rcu_read_lock, on the write side - * by dp_mutex. +/** + * DOC: Locking: + * + * Writes to device state (add/remove datapath, port, set operations on vports, + * etc.) are protected by RTNL. + * + * Writes to other state (flow table modifications, set miscellaneous datapath + * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside + * genl_mutex. * - * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL - * lock first. + * Reads are protected by RCU. * - * It is safe to access the datapath and vport structures with just - * dp_mutex. + * There are a few special cases (mostly stats) that have their own + * synchronization but they nest under all of above and don't interact with + * each other. */ -static struct datapath *dps[ODP_MAX]; -static DEFINE_MUTEX(dp_mutex); -static int new_vport(struct datapath *, struct odp_port *, int port_no); +static struct vport *new_vport(const struct vport_parms *); +static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *, + const struct dp_upcall_info *); +static int queue_userspace_packet(struct net *, int dp_ifindex, + struct sk_buff *, + const struct dp_upcall_info *); -/* Must be called with rcu_read_lock or dp_mutex. */ -struct datapath *get_dp(int dp_idx) +/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */ +static struct datapath *get_dp(struct net *net, int dp_ifindex) { - if (dp_idx < 0 || dp_idx >= ODP_MAX) - return NULL; - return rcu_dereference(dps[dp_idx]); -} -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(net, dp_ifindex); + if (dev) { + struct vport *vport = ovs_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; } /* Must be called with rcu_read_lock or RTNL lock. */ -const char *dp_name(const struct datapath *dp) +const char *ovs_dp_name(const struct datapath *dp) +{ + struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL); + return vport->ops->get_name(vport); +} + +static int get_dpifindex(struct datapath *dp) { - return vport_get_name(dp->ports[ODPP_LOCAL]); + struct vport *local; + int ifindex; + + rcu_read_lock(); + + local = ovs_vport_rcu(dp, OVSP_LOCAL); + if (local) + ifindex = local->ops->get_ifindex(local); + else + ifindex = 0; + + rcu_read_unlock(); + + return ifindex; } -static inline size_t br_nlmsg_size(void) +static size_t br_nlmsg_size(void) { return NLMSG_ALIGN(sizeof(struct ifinfomsg)) + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */ + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */ + nla_total_size(4) /* IFLA_MASTER */ + nla_total_size(4) /* IFLA_MTU */ - + nla_total_size(4) /* IFLA_LINK */ + nla_total_size(1); /* IFLA_OPERSTATE */ } +/* Caller must hold RTNL lock. */ static int dp_fill_ifinfo(struct sk_buff *skb, const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; - int ifindex = vport_get_ifindex(port); - int iflink = vport_get_iflink(port); + struct datapath *dp = port->dp; struct ifinfomsg *hdr; struct nlmsghdr *nlh; - if (ifindex < 0) - return ifindex; - - if (iflink < 0) - return iflink; + if (!port->ops->get_ifindex) + return -ENODEV; nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags); if (nlh == NULL) @@ -134,24 +171,21 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->ifi_family = AF_BRIDGE; hdr->__ifi_pad = 0; hdr->ifi_type = ARPHRD_ETHER; - hdr->ifi_index = ifindex; - hdr->ifi_flags = vport_get_flags(port); + hdr->ifi_index = port->ops->get_ifindex(port); + hdr->ifi_flags = port->ops->get_dev_flags(port); hdr->ifi_change = 0; - NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port)); - NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL])); - NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port)); + if (nla_put_string(skb, IFLA_IFNAME, port->ops->get_name(port)) || + nla_put_u32(skb, IFLA_MASTER, get_dpifindex(dp)) || + nla_put_u32(skb, IFLA_MTU, port->ops->get_mtu(port)) || #ifdef IFLA_OPERSTATE - NLA_PUT_U8(skb, IFLA_OPERSTATE, - vport_is_running(port) - ? vport_get_operstate(port) - : IF_OPER_DOWN); + nla_put_u8(skb, IFLA_OPERSTATE, + port->ops->is_running(port) ? + port->ops->get_operstate(port) : + IF_OPER_DOWN) || #endif - - NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port)); - - if (ifindex != iflink) - NLA_PUT_U32(skb, IFLA_LINK,iflink); + nla_put(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port))) + goto nla_put_failure; return nlmsg_end(skb, nlh); @@ -160,27 +194,36 @@ nla_put_failure: return -EMSGSIZE; } +/* Caller must hold RTNL lock. */ static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; - int err = -ENOBUFS; + int err; skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL); - if (skb == NULL) - goto errout; + if (!skb) { + err = -ENOBUFS; + goto err; + } err = dp_fill_ifinfo(skb, port, event, 0); if (err < 0) { - /* -EMSGSIZE implies BUG in br_nlmsg_size() */ - WARN_ON(err == -EMSGSIZE); - kfree_skb(skb); - goto errout; + if (err == -ENODEV) { + goto out; + } else { + /* -EMSGSIZE implies BUG in br_nlmsg_size() */ + WARN_ON(err == -EMSGSIZE); + goto err; + } } - rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL); + + rtnl_notify(skb, ovs_dp_get_net(port->dp), 0, RTNLGRP_LINK, NULL, GFP_KERNEL); + return; -errout: - if (err < 0) - rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err); +err: + rtnl_set_sk_err(ovs_dp_get_net(port->dp), RTNLGRP_LINK, err); +out: + kfree_skb(skb); } static void release_dp(struct kobject *kobj) @@ -193,585 +236,494 @@ static struct kobj_type dp_ktype = { .release = release_dp }; -static int create_dp(int dp_idx, const char __user *devnamep) -{ - struct odp_port internal_dev_port; - char devname[IFNAMSIZ]; - struct datapath *dp; - int err; - int i; - - if (devnamep) { - int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ); - if (retval < 0) { - err = -EFAULT; - goto err; - } else if (retval >= IFNAMSIZ) { - err = -ENAMETOOLONG; - goto err; - } - } else { - snprintf(devname, sizeof devname, "of%d", dp_idx); - } - - rtnl_lock(); - mutex_lock(&dp_mutex); - err = -ENODEV; - if (!try_module_get(THIS_MODULE)) - goto err_unlock; - - /* Exit early if a datapath with that number already exists. - * (We don't use -EEXIST because that's ambiguous with 'devname' - * conflicting with an existing network device name.) */ - err = -EBUSY; - if (get_dp(dp_idx)) - goto err_put_module; - - err = -ENOMEM; - dp = kzalloc(sizeof *dp, GFP_KERNEL); - if (dp == NULL) - goto err_put_module; - INIT_LIST_HEAD(&dp->port_list); - mutex_init(&dp->mutex); - dp->dp_idx = dp_idx; - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_head_init(&dp->queues[i]); - init_waitqueue_head(&dp->waitqueue); - - /* Initialize kobject for bridge. This will be added as - * /sys/class/net//brif later, if sysfs is enabled. */ - dp->ifobj.kset = NULL; - kobject_init(&dp->ifobj, &dp_ktype); - - /* Allocate table. */ - err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(0)); - if (!dp->table) - goto err_free_dp; - - /* Set up our datapath device. */ - BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname)); - strcpy(internal_dev_port.devname, devname); - strcpy(internal_dev_port.type, "internal"); - err = new_vport(dp, &internal_dev_port, ODPP_LOCAL); - if (err) { - if (err == -EBUSY) - err = -EEXIST; - - goto err_destroy_table; - } - - dp->drop_frags = 0; - dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) - goto err_destroy_local_port; - - rcu_assign_pointer(dps[dp_idx], dp); - dp_sysfs_add_dp(dp); - - mutex_unlock(&dp_mutex); - rtnl_unlock(); - - return 0; - -err_destroy_local_port: - dp_detach_port(dp->ports[ODPP_LOCAL]); -err_destroy_table: - tbl_destroy(dp->table, NULL); -err_free_dp: - kfree(dp); -err_put_module: - module_put(THIS_MODULE); -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); -err: - return err; -} - -static void do_destroy_dp(struct datapath *dp) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct vport *p, *n; - int i; - - list_for_each_entry_safe (p, n, &dp->port_list, node) - if (p->port_no != ODPP_LOCAL) - dp_detach_port(p); - - dp_sysfs_del_dp(dp); - - rcu_assign_pointer(dps[dp->dp_idx], NULL); + struct datapath *dp = container_of(rcu, struct datapath, rcu); - dp_detach_port(dp->ports[ODPP_LOCAL]); - - tbl_destroy(dp->table, flow_free_tbl); - - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_purge(&dp->queues[i]); + ovs_flow_tbl_destroy((__force struct flow_table *)dp->table); free_percpu(dp->stats_percpu); + release_net(ovs_dp_get_net(dp)); + kfree(dp->ports); kobject_put(&dp->ifobj); - module_put(THIS_MODULE); } -static int destroy_dp(int dp_idx) +static struct hlist_head *vport_hash_bucket(const struct datapath *dp, + u16 port_no) { - struct datapath *dp; - int err; - - rtnl_lock(); - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - err = -ENODEV; - if (!dp) - goto err_unlock; - - do_destroy_dp(dp); - err = 0; - -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); - return err; + return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; } -/* Called with RTNL lock and dp_mutex. */ -static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no) +struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no) { - struct vport_parms parms; struct vport *vport; + struct hlist_node *n; + struct hlist_head *head; - parms.name = odp_port->devname; - parms.type = odp_port->type; - parms.config = odp_port->config; - parms.dp = dp; - parms.port_no = port_no; - - vport_lock(); - vport = vport_add(&parms); - vport_unlock(); - - if (IS_ERR(vport)) - return PTR_ERR(vport); - - rcu_assign_pointer(dp->ports[port_no], vport); - list_add_rcu(&vport->node, &dp->port_list); - dp->n_ports++; - - dp_ifinfo_notify(RTM_NEWLINK, vport); - - return 0; + head = vport_hash_bucket(dp, port_no); + hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) { + if (vport->port_no == port_no) + return vport; + } + return NULL; } -static int attach_port(int dp_idx, struct odp_port __user *portp) +/* Called with RTNL lock and genl_lock. */ +static struct vport *new_vport(const struct vport_parms *parms) { - struct datapath *dp; - struct odp_port port; - int port_no; - int err; - - err = -EFAULT; - if (copy_from_user(&port, portp, sizeof port)) - goto out; - port.devname[IFNAMSIZ - 1] = '\0'; - port.type[VPORT_TYPE_SIZE - 1] = '\0'; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - for (port_no = 1; port_no < DP_MAX_PORTS; port_no++) - if (!dp->ports[port_no]) - goto got_port_no; - err = -EFBIG; - goto out_unlock_dp; - -got_port_no: - err = new_vport(dp, &port, port_no); - if (err) - goto out_unlock_dp; - - set_internal_devs_mtu(dp); - dp_sysfs_add_if(dp->ports[port_no]); + struct vport *vport; - err = put_user(port_no, &portp->port); + vport = ovs_vport_add(parms); + if (!IS_ERR(vport)) { + struct datapath *dp = parms->dp; + struct hlist_head *head = vport_hash_bucket(dp, vport->port_no); -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: - return err; + hlist_add_head_rcu(&vport->dp_hash_node, head); + dp_ifinfo_notify(RTM_NEWLINK, vport); + } + return vport; } -int dp_detach_port(struct vport *p) +/* Called with RTNL lock. */ +void ovs_dp_detach_port(struct vport *p) { - int err; - ASSERT_RTNL(); - if (p->port_no != ODPP_LOCAL) - dp_sysfs_del_if(p); + if (p->port_no != OVSP_LOCAL) + ovs_dp_sysfs_del_if(p); + dp_ifinfo_notify(RTM_DELLINK, p); /* First drop references to device. */ - p->dp->n_ports--; - list_del_rcu(&p->node); - rcu_assign_pointer(p->dp->ports[p->port_no], NULL); + hlist_del_rcu(&p->dp_hash_node); /* Then destroy it. */ - vport_lock(); - err = vport_del(p); - vport_unlock(); - - return err; -} - -static int detach_port(int dp_idx, int port_no) -{ - struct vport *p; - struct datapath *dp; - int err; - - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) - goto out; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - p = dp->ports[port_no]; - err = -ENOENT; - if (!p) - goto out_unlock_dp; - - err = dp_detach_port(p); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: - return err; + ovs_vport_del(p); } /* Must be called with rcu_read_lock. */ -void dp_process_received_packet(struct vport *p, struct sk_buff *skb) +void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb) { struct datapath *dp = p->dp; + struct sw_flow *flow; struct dp_stats_percpu *stats; - int stats_counter_off; - struct sw_flow_actions *acts; - struct loop_counter *loop; + u64 *stats_counter; int error; - OVS_CB(skb)->vport = p; + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); if (!OVS_CB(skb)->flow) { - struct odp_flow_key key; - struct tbl_node *flow_node; - bool is_frag; + struct sw_flow_key key; + int key_len; /* Extract flow from 'skb' into 'key'. */ - error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag); + error = ovs_flow_extract(skb, p->port_no, &key, &key_len); if (unlikely(error)) { kfree_skb(skb); return; } - if (is_frag && dp->drop_frags) { - kfree_skb(skb); - stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); - goto out; - } - /* Look up flow. */ - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, - flow_hash(&key), flow_cmp); - if (unlikely(!flow_node)) { - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); - stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), + &key, key_len); + if (unlikely(!flow)) { + struct dp_upcall_info upcall; + + upcall.cmd = OVS_PACKET_CMD_MISS; + upcall.key = &key; + upcall.userdata = NULL; + upcall.pid = p->upcall_pid; + ovs_dp_upcall(dp, skb, &upcall); + consume_skb(skb); + stats_counter = &stats->n_missed; goto out; } - OVS_CB(skb)->flow = flow_cast(flow_node); + OVS_CB(skb)->flow = flow; } - flow_used(OVS_CB(skb)->flow, skb); + stats_counter = &stats->n_hit; + ovs_flow_used(OVS_CB(skb)->flow, skb); + ovs_execute_actions(dp, skb); + +out: + /* Update datapath statistics. */ + u64_stats_update_begin(&stats->sync); + (*stats_counter)++; + u64_stats_update_end(&stats->sync); +} + +static struct genl_family dp_packet_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_PACKET_FAMILY, + .version = OVS_PACKET_VERSION, + .maxattr = OVS_PACKET_ATTR_MAX, + SET_NETNSOK +}; + +int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) +{ + struct dp_stats_percpu *stats; + int dp_ifindex; + int err; - acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); + if (upcall_info->pid == 0) { + err = -ENOTCONN; + goto err; + } - /* 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); - goto out_loop; + dp_ifindex = get_dpifindex(dp); + if (!dp_ifindex) { + err = -ENODEV; + goto err; } - /* Execute actions. */ - execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions, - acts->actions_len); - stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + forward_ip_summed(skb, true); - /* Check whether sub-actions looped too much. */ - if (unlikely(loop->looping)) - loop_suppress(dp, acts); + if (!skb_is_gso(skb)) + err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info); + else + err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info); + if (err) + goto err; -out_loop: - /* Decrement loop counter. */ - if (!--loop->count) - loop->looping = false; - loop_put_counter(); + return 0; -out: - /* Update datapath statistics. */ - local_bh_disable(); +err: stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); - write_seqcount_begin(&stats->seqlock); - (*(u64 *)((u8 *)stats + stats_counter_off))++; - write_seqcount_end(&stats->seqlock); + u64_stats_update_begin(&stats->sync); + stats->n_lost++; + u64_stats_update_end(&stats->sync); - local_bh_enable(); + return err; } -/* Append each packet in 'skb' list to 'queue'. There will be only one packet - * unless we broke up a GSO packet. */ -static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, - int queue_no, u64 arg) +static int queue_gso_packets(struct net *net, int dp_ifindex, + struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - struct sk_buff *nskb; - int port_no; + unsigned short gso_type = skb_shinfo(skb)->gso_type; + struct dp_upcall_info later_info; + struct sw_flow_key later_key; + struct sk_buff *segs, *nskb; int err; - if (OVS_CB(skb)->vport) - port_no = OVS_CB(skb)->vport->port_no; - else - port_no = ODPP_LOCAL; + segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); + if (IS_ERR(segs)) + return PTR_ERR(segs); + /* Queue all of the segments. */ + skb = segs; do { - struct odp_msg *header; - - nskb = skb->next; - skb->next = NULL; - - err = skb_cow(skb, sizeof *header); + err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info); if (err) - goto err_kfree_skbs; - - header = (struct odp_msg*)__skb_push(skb, sizeof *header); - header->type = queue_no; - header->length = skb->len; - header->port = port_no; - header->arg = arg; - skb_queue_tail(queue, skb); + break; - skb = nskb; - } while (skb); - return 0; + if (skb == segs && gso_type & SKB_GSO_UDP) { + /* The initial flow key extracted by ovs_flow_extract() + * in this case is for a first fragment, so we need to + * properly mark later fragments. + */ + later_key = *upcall_info->key; + later_key.ip.frag = OVS_FRAG_TYPE_LATER; + + later_info = *upcall_info; + later_info.key = &later_key; + upcall_info = &later_info; + } + } while ((skb = skb->next)); -err_kfree_skbs: - kfree_skb(skb); - while ((skb = nskb) != NULL) { + /* Free all of the segments. */ + skb = segs; + do { nskb = skb->next; - kfree_skb(skb); - } + if (err) + kfree_skb(skb); + else + consume_skb(skb); + } while ((skb = nskb)); return err; } -int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, - u64 arg) +static int queue_userspace_packet(struct net *net, int dp_ifindex, + struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - struct dp_stats_percpu *stats; - struct sk_buff_head *queue; + struct ovs_header *upcall; + struct sk_buff *nskb = NULL; + struct sk_buff *user_skb; /* to be queued to userspace */ + struct nlattr *nla; + unsigned int len; int err; - WARN_ON_ONCE(skb_shared(skb)); - BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR); - queue = &dp->queues[queue_no]; - err = -ENOBUFS; - if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) - goto err_kfree_skb; + if (vlan_tx_tag_present(skb)) { + nskb = skb_clone(skb, GFP_ATOMIC); + if (!nskb) + return -ENOMEM; + + err = vlan_deaccel_tag(nskb); + if (err) + return err; - forward_ip_summed(skb); + skb = nskb; + } - err = vswitch_skb_checksum_setup(skb); - if (err) - goto err_kfree_skb; + if (nla_attr_size(skb->len) > USHRT_MAX) { + err = -EFBIG; + goto out; + } - /* 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 (unlikely(IS_ERR(skb))) { - err = PTR_ERR(skb); - goto err; - } + len = sizeof(struct ovs_header); + len += nla_total_size(skb->len); + len += nla_total_size(FLOW_BUFSIZE); + if (upcall_info->cmd == OVS_PACKET_CMD_ACTION) + len += nla_total_size(8); + + user_skb = genlmsg_new(len, GFP_ATOMIC); + if (!user_skb) { + err = -ENOMEM; + goto out; } - err = queue_control_packets(skb, queue, queue_no, arg); - wake_up_interruptible(&dp->waitqueue); - return err; + upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, + 0, upcall_info->cmd); + upcall->dp_ifindex = dp_ifindex; -err_kfree_skb: - kfree_skb(skb); -err: - local_bh_disable(); - stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY); + ovs_flow_to_nlattrs(upcall_info->key, user_skb); + nla_nest_end(user_skb, nla); - write_seqcount_begin(&stats->seqlock); - stats->n_lost++; - write_seqcount_end(&stats->seqlock); + if (upcall_info->userdata) + nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, + nla_get_u64(upcall_info->userdata)); - local_bh_enable(); + nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len); + skb_copy_and_csum_dev(skb, nla_data(nla)); + + err = genlmsg_unicast(net, user_skb, upcall_info->pid); + +out: + kfree_skb(nskb); return err; } +/* Called with genl_mutex. */ static int flush_flows(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); - struct tbl *new_table; + struct flow_table *old_table; + struct flow_table *new_table; - new_table = tbl_create(0); + old_table = genl_dereference(dp->table); + new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS); if (!new_table) return -ENOMEM; rcu_assign_pointer(dp->table, new_table); - tbl_deferred_destroy(old_table, flow_free_tbl); - + ovs_flow_tbl_deferred_destroy(old_table); return 0; } -static int validate_actions(const struct nlattr *actions, u32 actions_len) +static int validate_actions(const struct nlattr *attr, + const struct sw_flow_key *key, int depth); + +static int validate_sample(const struct nlattr *attr, + const struct sw_flow_key *key, int depth) { - const struct nlattr *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, - }; - int type = nla_type(a); - - if (type > ODPAT_MAX || nla_len(a) != action_lens[type]) - return -EINVAL; - - switch (type) { - case ODPAT_UNSPEC: + const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1]; + const struct nlattr *probability, *actions; + const struct nlattr *a; + int rem; + + memset(attrs, 0, sizeof(attrs)); + nla_for_each_nested(a, attr, rem) { + int type = nla_type(a); + if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type]) return -EINVAL; + attrs[type] = a; + } + if (rem) + return -EINVAL; - case ODPAT_CONTROLLER: - case ODPAT_STRIP_VLAN: - case ODPAT_SET_DL_SRC: - case ODPAT_SET_DL_DST: - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - case ODPAT_SET_TUNNEL: - case ODPAT_SET_PRIORITY: - case ODPAT_POP_PRIORITY: - case ODPAT_DROP_SPOOFED_ARP: - /* No validation needed. */ - break; - - case ODPAT_OUTPUT: - if (nla_get_u32(a) >= DP_MAX_PORTS) - return -EINVAL; - - case ODPAT_SET_DL_TCI: - if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) - return -EINVAL; - break; - - case ODPAT_SET_NW_TOS: - if (nla_get_u8(a) & INET_ECN_MASK) - return -EINVAL; - break; + probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY]; + if (!probability || nla_len(probability) != sizeof(u32)) + return -EINVAL; - default: - return -EOPNOTSUPP; - } - } + actions = attrs[OVS_SAMPLE_ATTR_ACTIONS]; + if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN)) + return -EINVAL; + return validate_actions(actions, key, depth + 1); +} - if (rem > 0) - return -EINVAL; +static int validate_tp_port(const struct sw_flow_key *flow_key) +{ + if (flow_key->eth.type == htons(ETH_P_IP)) { + if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst) + return 0; + } else if (flow_key->eth.type == htons(ETH_P_IPV6)) { + if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst) + return 0; + } - return 0; + return -EINVAL; } -static struct sw_flow_actions *get_actions(const struct odp_flow *flow) +static int validate_set(const struct nlattr *a, + const struct sw_flow_key *flow_key) { - struct sw_flow_actions *actions; - int error; + const struct nlattr *ovs_key = nla_data(a); + int key_type = nla_type(ovs_key); - actions = flow_actions_alloc(flow->actions_len); - error = PTR_ERR(actions); - if (IS_ERR(actions)) - goto error; + /* There can be only one key in a action */ + if (nla_total_size(nla_len(ovs_key)) != nla_len(a)) + return -EINVAL; - error = -EFAULT; - if (copy_from_user(actions->actions, flow->actions, flow->actions_len)) - goto error_free_actions; - error = validate_actions(actions->actions, actions->actions_len); - if (error) - goto error_free_actions; + if (key_type > OVS_KEY_ATTR_MAX || + nla_len(ovs_key) != ovs_key_lens[key_type]) + return -EINVAL; - return actions; + switch (key_type) { + const struct ovs_key_ipv4 *ipv4_key; -error_free_actions: - kfree(actions); -error: - return ERR_PTR(error); + case OVS_KEY_ATTR_PRIORITY: + case OVS_KEY_ATTR_TUN_ID: + case OVS_KEY_ATTR_ETHERNET: + break; + + case OVS_KEY_ATTR_IPV4: + if (flow_key->eth.type != htons(ETH_P_IP)) + return -EINVAL; + + if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst) + return -EINVAL; + + ipv4_key = nla_data(ovs_key); + if (ipv4_key->ipv4_proto != flow_key->ip.proto) + return -EINVAL; + + if (ipv4_key->ipv4_frag != flow_key->ip.frag) + return -EINVAL; + + break; + + case OVS_KEY_ATTR_TCP: + if (flow_key->ip.proto != IPPROTO_TCP) + return -EINVAL; + + return validate_tp_port(flow_key); + + case OVS_KEY_ATTR_UDP: + if (flow_key->ip.proto != IPPROTO_UDP) + return -EINVAL; + + return validate_tp_port(flow_key); + + default: + return -EINVAL; + } + + return 0; } -static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) +static int validate_userspace(const struct nlattr *attr) { - if (flow->used) { - struct timespec offset_ts, used, now_mono; + static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = { + [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 }, + [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 }, + }; + struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1]; + int error; - 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); + error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, + attr, userspace_policy); + if (error) + return error; - stats->used_sec = used.tv_sec; - stats->used_nsec = used.tv_nsec; - } else { - stats->used_sec = 0; - stats->used_nsec = 0; + if (!a[OVS_USERSPACE_ATTR_PID] || + !nla_get_u32(a[OVS_USERSPACE_ATTR_PID])) + return -EINVAL; + + return 0; +} + +static int validate_actions(const struct nlattr *attr, + const struct sw_flow_key *key, int depth) +{ + const struct nlattr *a; + int rem, err; + + if (depth >= SAMPLE_ACTION_DEPTH) + return -EOVERFLOW; + + nla_for_each_nested(a, attr, rem) { + /* Expected argument lengths, (u32)-1 for variable length. */ + static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = { + [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32), + [OVS_ACTION_ATTR_USERSPACE] = (u32)-1, + [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan), + [OVS_ACTION_ATTR_POP_VLAN] = 0, + [OVS_ACTION_ATTR_SET] = (u32)-1, + [OVS_ACTION_ATTR_SAMPLE] = (u32)-1 + }; + const struct ovs_action_push_vlan *vlan; + int type = nla_type(a); + + if (type > OVS_ACTION_ATTR_MAX || + (action_lens[type] != nla_len(a) && + action_lens[type] != (u32)-1)) + return -EINVAL; + + switch (type) { + case OVS_ACTION_ATTR_UNSPEC: + return -EINVAL; + + case OVS_ACTION_ATTR_USERSPACE: + err = validate_userspace(a); + if (err) + return err; + break; + + case OVS_ACTION_ATTR_OUTPUT: + if (nla_get_u32(a) >= DP_MAX_PORTS) + return -EINVAL; + break; + + + case OVS_ACTION_ATTR_POP_VLAN: + break; + + case OVS_ACTION_ATTR_PUSH_VLAN: + vlan = nla_data(a); + if (vlan->vlan_tpid != htons(ETH_P_8021Q)) + return -EINVAL; + if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT))) + return -EINVAL; + break; + + case OVS_ACTION_ATTR_SET: + err = validate_set(a, key); + if (err) + return err; + break; + + case OVS_ACTION_ATTR_SAMPLE: + err = validate_sample(a, key, depth); + if (err) + return err; + break; + + default: + return -EINVAL; + } } - stats->n_packets = flow->packet_count; - stats->n_bytes = flow->byte_count; - stats->reserved = 0; - stats->tcp_flags = flow->tcp_flags; - stats->error = 0; + if (rem > 0) + return -EINVAL; + + return 0; } static void clear_stats(struct sw_flow *flow) @@ -782,1197 +734,1444 @@ static void clear_stats(struct sw_flow *flow) flow->byte_count = 0; } -static int expand_table(struct datapath *dp) +static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) { - struct tbl *old_table = rcu_dereference(dp->table); - struct tbl *new_table; + struct ovs_header *ovs_header = info->userhdr; + struct nlattr **a = info->attrs; + struct sw_flow_actions *acts; + struct sk_buff *packet; + struct sw_flow *flow; + struct datapath *dp; + struct ethhdr *eth; + int len; + int err; + int key_len; - new_table = tbl_expand(old_table); - if (IS_ERR(new_table)) - return PTR_ERR(new_table); + err = -EINVAL; + if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || + !a[OVS_PACKET_ATTR_ACTIONS] || + nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN) + goto err; - rcu_assign_pointer(dp->table, new_table); - tbl_deferred_destroy(old_table, NULL); + len = nla_len(a[OVS_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); - return 0; + memcpy(__skb_put(packet, len), nla_data(a[OVS_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 = ovs_flow_alloc(); + err = PTR_ERR(flow); + if (IS_ERR(flow)) + goto err_kfree_skb; + + err = ovs_flow_extract(packet, -1, &flow->key, &key_len); + if (err) + goto err_flow_put; + + err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority, + &flow->key.phy.in_port, + &flow->key.phy.tun_id, + a[OVS_PACKET_ATTR_KEY]); + if (err) + goto err_flow_put; + + err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0); + if (err) + goto err_flow_put; + + flow->hash = ovs_flow_hash(&flow->key, key_len); + + acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]); + err = PTR_ERR(acts); + if (IS_ERR(acts)) + goto err_flow_put; + rcu_assign_pointer(flow->sf_acts, acts); + + OVS_CB(packet)->flow = flow; + packet->priority = flow->key.phy.priority; + + rcu_read_lock(); + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + err = -ENODEV; + if (!dp) + goto err_unlock; + + local_bh_disable(); + err = ovs_execute_actions(dp, packet); + local_bh_enable(); + rcu_read_unlock(); + + ovs_flow_put(flow); + return err; + +err_unlock: + rcu_read_unlock(); +err_flow_put: + ovs_flow_put(flow); +err_kfree_skb: + kfree_skb(packet); +err: + return err; +} + +static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { + [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, + [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; + +static struct genl_ops dp_packet_genl_ops[] = { + { .cmd = OVS_PACKET_CMD_EXECUTE, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = packet_policy, + .doit = ovs_packet_cmd_execute + } +}; + +static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats) +{ + int i; + struct flow_table *table = genl_dereference(dp->table); + + stats->n_flows = ovs_flow_tbl_count(table); + + 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 int start; + + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + + do { + start = u64_stats_fetch_begin_bh(&percpu_stats->sync); + local_stats = *percpu_stats; + } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start)); + + stats->n_hit += local_stats.n_hit; + stats->n_missed += local_stats.n_missed; + stats->n_lost += local_stats.n_lost; + } +} + +static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { + [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, + [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, + [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, +}; + +static struct genl_family dp_flow_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_FLOW_FAMILY, + .version = OVS_FLOW_VERSION, + .maxattr = OVS_FLOW_ATTR_MAX, + SET_NETNSOK +}; + +static struct genl_multicast_group ovs_dp_flow_multicast_group = { + .name = OVS_FLOW_MCGROUP +}; + +/* Called with genl_lock. */ +static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, + struct sk_buff *skb, u32 pid, + u32 seq, u32 flags, u8 cmd) +{ + const int skb_orig_len = skb->len; + const struct sw_flow_actions *sf_acts; + struct ovs_flow_stats stats; + struct ovs_header *ovs_header; + struct nlattr *nla; + unsigned long used; + u8 tcp_flags; + int err; + + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + + ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd); + if (!ovs_header) + return -EMSGSIZE; + + ovs_header->dp_ifindex = get_dpifindex(dp); + + nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY); + if (!nla) + goto nla_put_failure; + err = ovs_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, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used))) + goto nla_put_failure; + + if (stats.n_packets && + nla_put(skb, OVS_FLOW_ATTR_STATS, + sizeof(struct ovs_flow_stats), &stats)) + goto nla_put_failure; + + if (tcp_flags && + nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags)) + goto nla_put_failure; + + /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if + * this is the first flow to be dumped into 'skb'. This is unusual for + * Netlink but individual action lists can be longer than + * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. + * The userspace caller can always fetch the actions separately if it + * really wants them. (Most userspace callers in fact don't care.) + * + * This can only fail for dump operations because the skb is always + * properly sized for single flows. + */ + err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len, + sf_acts->actions); + if (err < 0 && skb_orig_len) + goto error; + + return genlmsg_end(skb, ovs_header); + +nla_put_failure: + err = -EMSGSIZE; +error: + genlmsg_cancel(skb, ovs_header); + return err; } -static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, - struct odp_flow_stats *stats) +static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow) { - struct tbl_node *flow_node; + const struct sw_flow_actions *sf_acts; + int len; + + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + + /* OVS_FLOW_ATTR_KEY */ + len = nla_total_size(FLOW_BUFSIZE); + /* OVS_FLOW_ATTR_ACTIONS */ + len += nla_total_size(sf_acts->actions_len); + /* OVS_FLOW_ATTR_STATS */ + len += nla_total_size(sizeof(struct ovs_flow_stats)); + /* OVS_FLOW_ATTR_TCP_FLAGS */ + len += nla_total_size(1); + /* OVS_FLOW_ATTR_USED */ + len += nla_total_size(8); + + len += NLMSG_ALIGN(sizeof(struct ovs_header)); + + return genlmsg_new(len, GFP_KERNEL); +} + +static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, + struct datapath *dp, + u32 pid, u32 seq, u8 cmd) +{ + struct sk_buff *skb; + int retval; + + skb = ovs_flow_cmd_alloc_info(flow); + if (!skb) + return ERR_PTR(-ENOMEM); + + retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd); + BUG_ON(retval < 0); + return skb; +} + +static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) +{ + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; struct sw_flow *flow; - struct tbl *table; + struct sk_buff *reply; + struct datapath *dp; + struct flow_table *table; int error; + int key_len; + + /* Extract key. */ + error = -EINVAL; + if (!a[OVS_FLOW_ATTR_KEY]) + goto error; + error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (error) + goto error; + + /* Validate actions. */ + if (a[OVS_FLOW_ATTR_ACTIONS]) { + error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0); + if (error) + goto error; + } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) { + error = -EINVAL; + goto error; + } + + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + error = -ENODEV; + if (!dp) + goto error; - table = rcu_dereference(dp->table); - flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); - if (!flow_node) { - /* No such flow. */ + table = genl_dereference(dp->table); + flow = ovs_flow_tbl_lookup(table, &key, key_len); + if (!flow) { struct sw_flow_actions *acts; + /* Bail out if we're not allowed to create a new flow. */ error = -ENOENT; - if (!(uf->flags & ODPPF_CREATE)) + if (info->genlhdr->cmd == OVS_FLOW_CMD_SET) goto error; /* Expand table, if necessary, to make room. */ - if (tbl_count(table) >= tbl_n_buckets(table)) { - error = expand_table(dp); - if (error) - goto error; - table = rcu_dereference(dp->table); + if (ovs_flow_tbl_need_to_expand(table)) { + struct flow_table *new_table; + + new_table = ovs_flow_tbl_expand(table); + if (!IS_ERR(new_table)) { + rcu_assign_pointer(dp->table, new_table); + ovs_flow_tbl_deferred_destroy(table); + table = genl_dereference(dp->table); + } } /* Allocate flow. */ - flow = flow_alloc(); + flow = ovs_flow_alloc(); if (IS_ERR(flow)) { error = PTR_ERR(flow); goto error; } - flow->key = uf->flow.key; + flow->key = key; clear_stats(flow); /* Obtain actions. */ - acts = get_actions(&uf->flow); + acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]); error = PTR_ERR(acts); if (IS_ERR(acts)) goto error_free_flow; rcu_assign_pointer(flow->sf_acts, acts); /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key)); - if (error) - goto error_free_flow_acts; + flow->hash = ovs_flow_hash(&key, key_len); + ovs_flow_tbl_insert(table, flow); - memset(stats, 0, sizeof(struct odp_flow_stats)); + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, + OVS_FLOW_CMD_NEW); } else { /* We found a matching flow. */ - struct sw_flow_actions *old_acts, *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; + struct nlattr *acts_attrs; + + /* 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)) + if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW && + info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) goto error; - /* Swap actions. */ - new_acts = get_actions(&uf->flow); - error = PTR_ERR(new_acts); - if (IS_ERR(new_acts)) - goto error; - old_acts = rcu_dereference(flow->sf_acts); - if (old_acts->actions_len != new_acts->actions_len || - memcmp(old_acts->actions, new_acts->actions, - old_acts->actions_len)) { + /* Update actions. */ + old_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_genl_is_held()); + acts_attrs = a[OVS_FLOW_ATTR_ACTIONS]; + if (acts_attrs && + (old_acts->actions_len != nla_len(acts_attrs) || + memcmp(old_acts->actions, nla_data(acts_attrs), + old_acts->actions_len))) { + struct sw_flow_actions *new_acts; + + new_acts = ovs_flow_actions_alloc(acts_attrs); + 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); + ovs_flow_deferred_free_acts(old_acts); } - /* Fetch stats, then clear them if necessary. */ - spin_lock_bh(&flow->lock); - get_stats(flow, stats); - if (uf->flags & ODPPF_ZERO_STATS) + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, OVS_FLOW_CMD_NEW); + + /* Clear stats. */ + if (a[OVS_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, + ovs_dp_flow_multicast_group.id, info->nlhdr, + GFP_KERNEL); + else + netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0, + ovs_dp_flow_multicast_group.id, PTR_ERR(reply)); return 0; -error_free_flow_acts: - kfree(flow->sf_acts); error_free_flow: - flow->sf_acts = NULL; - flow_put(flow); + ovs_flow_put(flow); error: return error; } -static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) +static int ovs_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 ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; + struct sk_buff *reply; + struct sw_flow *flow; + struct datapath *dp; + struct flow_table *table; + int err; + int key_len; - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) - return -EFAULT; + if (!a[OVS_FLOW_ATTR_KEY]) + return -EINVAL; + err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (err) + return err; - error = do_put_flow(dp, &uf, &stats); - if (error) - return error; + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + if (!dp) + return -ENODEV; - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; + table = genl_dereference(dp->table); + flow = ovs_flow_tbl_lookup(table, &key, key_len); + if (!flow) + return -ENOENT; - return 0; + reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, + info->snd_seq, OVS_FLOW_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); + + return genlmsg_reply(reply, info); } -static int do_answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow_stats __user *ustats, - struct nlattr __user *actions, - u32 __user *actions_lenp) +static int ovs_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; - - spin_lock_bh(&flow->lock); - get_stats(flow, &stats); - if (query_flags & ODPFF_ZERO_TCP_FLAGS) - flow->tcp_flags = 0; - - spin_unlock_bh(&flow->lock); - - if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(actions_len, actions_lenp)) - return -EFAULT; - - if (!actions_len) - return 0; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sw_flow_key key; + struct sk_buff *reply; + struct sw_flow *flow; + struct datapath *dp; + struct flow_table *table; + int err; + int key_len; - sf_acts = rcu_dereference(flow->sf_acts); - 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; + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + if (!dp) + return -ENODEV; - return 0; -} + if (!a[OVS_FLOW_ATTR_KEY]) + return flush_flows(dp); -static int answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow __user *ufp) -{ - struct nlattr *actions; + err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]); + if (err) + return err; - if (get_user(actions, &ufp->actions)) - return -EFAULT; + table = genl_dereference(dp->table); + flow = ovs_flow_tbl_lookup(table, &key, key_len); + if (!flow) + return -ENOENT; - return do_answer_query(flow, query_flags, - &ufp->stats, actions, &ufp->actions_len); -} + reply = ovs_flow_cmd_alloc_info(flow); + if (!reply) + return -ENOMEM; -static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key) -{ - struct tbl *table = rcu_dereference(dp->table); - struct tbl_node *flow_node; - int error; + ovs_flow_tbl_remove(table, flow); - flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp); - if (!flow_node) - return ERR_PTR(-ENOENT); + err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid, + info->snd_seq, 0, OVS_FLOW_CMD_DEL); + BUG_ON(err < 0); - error = tbl_remove(table, flow_node); - if (error) - return ERR_PTR(error); + ovs_flow_deferred_free(flow); - /* 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); + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL); + return 0; } -static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) +static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - struct sw_flow *flow; - struct odp_flow uf; - int error; + struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; + struct flow_table *table; - if (copy_from_user(&uf, ufp, sizeof uf)) - return -EFAULT; + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + if (!dp) + return -ENODEV; - flow = do_del_flow(dp, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); + table = genl_dereference(dp->table); - error = answer_query(flow, 0, ufp); - flow_deferred_free(flow); - return error; -} - -static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) -{ - struct tbl *table = rcu_dereference(dp->table); - u32 i; + for (;;) { + struct sw_flow *flow; + u32 bucket, obj; - for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = &flowvec->flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; + bucket = cb->args[0]; + obj = cb->args[1]; + flow = ovs_flow_tbl_next(table, &bucket, &obj); + if (!flow) + break; - if (copy_from_user(&uf, ufp, sizeof uf)) - return -EFAULT; + if (ovs_flow_cmd_fill_info(flow, dp, skb, + NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + OVS_FLOW_CMD_NEW) < 0) + break; - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = answer_query(flow_cast(flow_node), uf.flags, ufp); - if (error) - return -EFAULT; + cb->args[0] = bucket; + cb->args[1] = obj; } - return flowvec->n_flows; + return skb->len; } -struct list_flows_cbdata { - struct odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; +static struct genl_ops dp_flow_genl_ops[] = { + { .cmd = OVS_FLOW_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_new_or_set + }, + { .cmd = OVS_FLOW_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_del + }, + { .cmd = OVS_FLOW_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_get, + .dumpit = ovs_flow_cmd_dump + }, + { .cmd = OVS_FLOW_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = flow_policy, + .doit = ovs_flow_cmd_new_or_set, + }, }; -static int list_flow(struct tbl_node *node, void *cbdata_) -{ - struct sw_flow *flow = flow_cast(node); - struct list_flows_cbdata *cbdata = cbdata_; - struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; - int error; - - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) - return -EFAULT; - error = answer_query(flow, 0, ufp); - if (error) - return error; - - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; -} - -static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) -{ - struct list_flows_cbdata cbdata; - int error; - - if (!flowvec->n_flows) - return 0; - - cbdata.uflows = flowvec->flows; - cbdata.n_flows = flowvec->n_flows; - cbdata.listed_flows = 0; - - error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata); - return error ? error : cbdata.listed_flows; -} - -static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - const struct odp_flowvec *)) -{ - struct odp_flowvec __user *uflowvec; - struct odp_flowvec flowvec; - int retval; - - uflowvec = (struct odp_flowvec __user *)argp; - if (copy_from_user(&flowvec, uflowvec, sizeof flowvec)) - return -EFAULT; +static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, +#endif + [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, +}; - if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) - return -EINVAL; +static struct genl_family dp_datapath_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_DATAPATH_FAMILY, + .version = OVS_DATAPATH_VERSION, + .maxattr = OVS_DP_ATTR_MAX, + SET_NETNSOK +}; - retval = function(dp, &flowvec); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : put_user(retval, &uflowvec->n_flows)); -} +static struct genl_multicast_group ovs_dp_datapath_multicast_group = { + .name = OVS_DATAPATH_MCGROUP +}; -static int do_execute(struct datapath *dp, const struct odp_execute *execute) +static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) { - struct odp_flow_key key; - struct sk_buff *skb; - struct sw_flow_actions *actions; - struct ethhdr *eth; - bool is_frag; + struct ovs_header *ovs_header; + struct ovs_dp_stats dp_stats; int err; - err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) - goto error; - - actions = flow_actions_alloc(execute->actions_len); - if (IS_ERR(actions)) { - err = PTR_ERR(actions); + ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family, + flags, cmd); + if (!ovs_header) goto error; - } - - err = -EFAULT; - if (copy_from_user(actions->actions, execute->actions, execute->actions_len)) - goto error_free_actions; - - err = validate_actions(actions->actions, execute->actions_len); - if (err) - goto error_free_actions; - err = -ENOMEM; - skb = alloc_skb(execute->length, GFP_KERNEL); - if (!skb) - goto error_free_actions; - - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), execute->data, - execute->length)) - goto error_free_skb; - - skb_reset_mac_header(skb); - eth = eth_hdr(skb); - - /* 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); - - err = flow_extract(skb, -1, &key, &is_frag); - if (err) - goto error_free_skb; + ovs_header->dp_ifindex = get_dpifindex(dp); rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); + err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); rcu_read_unlock(); + if (err) + goto nla_put_failure; - kfree(actions); - return err; + get_dp_stats(dp, &dp_stats); + if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats)) + goto nla_put_failure; -error_free_skb: - kfree_skb(skb); -error_free_actions: - kfree(actions); + return genlmsg_end(skb, ovs_header); + +nla_put_failure: + genlmsg_cancel(skb, ovs_header); error: - return err; + return -EMSGSIZE; } -static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep) +static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid, + u32 seq, u8 cmd) { - struct odp_execute execute; + struct sk_buff *skb; + int retval; - if (copy_from_user(&execute, executep, sizeof execute)) - return -EFAULT; + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); - return do_execute(dp, &execute); + retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } + return skb; } -static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) +static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1]) { - struct tbl *table = rcu_dereference(dp->table); - struct odp_stats stats; - int i; - - 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; + return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1); +} - percpu_stats = per_cpu_ptr(dp->stats_percpu, i); +/* Called with genl_mutex and optionally with RTNL lock also. */ +static struct datapath *lookup_datapath(struct net *net, + struct ovs_header *ovs_header, + struct nlattr *a[OVS_DP_ATTR_MAX + 1]) +{ + struct datapath *dp; - do { - seqcount = read_seqcount_begin(&percpu_stats->seqlock); - local_stats = *percpu_stats; - } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); + if (!a[OVS_DP_ATTR_NAME]) + dp = get_dp(net, ovs_header->dp_ifindex); + else { + struct vport *vport; - 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; + rcu_read_lock(); + vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); + dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; + rcu_read_unlock(); } - 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; + return dp ? dp : ERR_PTR(-ENODEV); } -/* 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 ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) { - struct vport *p; - int mtu = 0; + struct nlattr **a = info->attrs; + struct vport_parms parms; + struct sk_buff *reply; + struct datapath *dp; + struct vport *vport; + struct ovs_net *ovs_net; + int err, i; - ASSERT_RTNL(); + err = -EINVAL; + if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) + goto err; + + err = ovs_dp_cmd_validate(a); + if (err) + goto err; + + rtnl_lock(); + + err = -ENOMEM; + dp = kzalloc(sizeof(*dp), GFP_KERNEL); + if (dp == NULL) + goto err_unlock_rtnl; - list_for_each_entry_rcu (p, &dp->port_list, node) { - int dev_mtu; + /* 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); - /* Skip any internal ports, since that's what we're trying to - * set. */ - if (is_internal_vport(p)) - continue; + /* Allocate table. */ + err = -ENOMEM; + rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS)); + if (!dp->table) + goto err_free_dp; - dev_mtu = vport_get_mtu(p); - if (!mtu || dev_mtu < mtu) - mtu = dev_mtu; + dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); + if (!dp->stats_percpu) { + err = -ENOMEM; + goto err_destroy_table; } + ovs_dp_set_net(dp, hold_net(sock_net(skb->sk))); - return mtu ? mtu : ETH_DATA_LEN; -} + dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head), + GFP_KERNEL); + if (!dp->ports) { + err = -ENOMEM; + goto err_destroy_percpu; + } -/* Sets the MTU of all datapath devices to the minimum of the ports. Must - * be called with RTNL lock. */ -void set_internal_devs_mtu(const struct datapath *dp) -{ - struct vport *p; - int mtu; + for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) + INIT_HLIST_HEAD(&dp->ports[i]); - ASSERT_RTNL(); + /* Set up our datapath device. */ + parms.name = nla_data(a[OVS_DP_ATTR_NAME]); + parms.type = OVS_VPORT_TYPE_INTERNAL; + parms.options = NULL; + parms.dp = dp; + parms.port_no = OVSP_LOCAL; + parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]); - mtu = dp_min_mtu(dp); + vport = new_vport(&parms); + if (IS_ERR(vport)) { + err = PTR_ERR(vport); + if (err == -EBUSY) + err = -EEXIST; - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (is_internal_vport(p)) - vport_set_mtu(p, mtu); + goto err_destroy_ports_array; } -} -static int put_port(const struct vport *p, struct odp_port __user *uop) -{ - struct odp_port op; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, + info->snd_seq, OVS_DP_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto err_destroy_local_port; - memset(&op, 0, sizeof op); + ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); + list_add_tail(&dp->list_node, &ovs_net->dps); + ovs_dp_sysfs_add_dp(dp); - rcu_read_lock(); - strncpy(op.devname, vport_get_name(p), sizeof op.devname); - strncpy(op.type, vport_get_type(p), sizeof op.type); - rcu_read_unlock(); + rtnl_unlock(); - op.port = p->port_no; + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_datapath_multicast_group.id, info->nlhdr, + GFP_KERNEL); + return 0; - return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; +err_destroy_local_port: + ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL)); +err_destroy_ports_array: + kfree(dp->ports); +err_destroy_percpu: + free_percpu(dp->stats_percpu); +err_destroy_table: + ovs_flow_tbl_destroy(genl_dereference(dp->table)); +err_free_dp: + kfree(dp); +err_unlock_rtnl: + rtnl_unlock(); +err: + return err; } -static int query_port(struct datapath *dp, struct odp_port __user *uport) +/* Called with genl_mutex. */ +static void __dp_destroy(struct datapath *dp) { - struct odp_port port; + int i; - if (copy_from_user(&port, uport, sizeof port)) - return -EFAULT; + rtnl_lock(); - if (port.devname[0]) { + for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { struct vport *vport; - int err = 0; - - port.devname[IFNAMSIZ - 1] = '\0'; - - vport_lock(); - rcu_read_lock(); + struct hlist_node *node, *n; - vport = vport_locate(port.devname); - if (!vport) { - err = -ENODEV; - goto error_unlock; - } - if (vport->dp != dp) { - err = -ENOENT; - goto error_unlock; - } - - port.port = vport->port_no; + hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node) + if (vport->port_no != OVSP_LOCAL) + ovs_dp_detach_port(vport); + } -error_unlock: - rcu_read_unlock(); - vport_unlock(); + ovs_dp_sysfs_del_dp(dp); + list_del(&dp->list_node); + ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL)); - if (err) - return err; - } else { - if (port.port >= DP_MAX_PORTS) - return -EINVAL; - if (!dp->ports[port.port]) - return -ENOENT; - } + /* 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(); - return put_port(dp->ports[port.port], uport); + call_rcu(&dp->rcu, destroy_dp_rcu); } -static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, - int n_ports) +static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) { - int idx = 0; - if (n_ports) { - struct vport *p; - - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (put_port(p, &uports[idx])) - return -EFAULT; - if (idx++ >= n_ports) - break; - } - } - return idx; -} + struct sk_buff *reply; + struct datapath *dp; + int err; -static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) -{ - struct odp_portvec pv; - int retval; + err = ovs_dp_cmd_validate(info->attrs); + if (err) + return err; - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; + dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); + err = PTR_ERR(dp); + if (IS_ERR(dp)) + return err; - retval = do_list_ports(dp, pv.ports, pv.n_ports); - if (retval < 0) - return retval; + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, + info->snd_seq, OVS_DP_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + return err; - return put_user(retval, &upv->n_ports); -} + __dp_destroy(dp); -static int get_listen_mask(const struct file *f) -{ - return (long)f->private_data; -} + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_datapath_multicast_group.id, info->nlhdr, + GFP_KERNEL); -static void set_listen_mask(struct file *f, int listen_mask) -{ - f->private_data = (void*)(long)listen_mask; + return 0; } -static long openvswitch_ioctl(struct file *f, unsigned int cmd, - unsigned long argp) +static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) { - int dp_idx = iminor(f->f_dentry->d_inode); + struct sk_buff *reply; struct datapath *dp; - int drop_frags, listeners, port_no; - 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_ATTACH: - err = attach_port(dp_idx, (struct odp_port __user *)argp); - goto exit; - - case ODP_VPORT_DETACH: - err = get_user(port_no, (int __user *)argp); - if (!err) - err = detach_port(dp_idx, port_no); - goto exit; - - case ODP_VPORT_MOD: - err = vport_user_mod((struct odp_port __user *)argp); - goto exit; + err = ovs_dp_cmd_validate(info->attrs); + if (err) + return err; + + dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); + + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, + info->snd_seq, OVS_DP_CMD_NEW); + if (IS_ERR(reply)) { + err = PTR_ERR(reply); + netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0, + ovs_dp_datapath_multicast_group.id, err); + return 0; + } - case ODP_VPORT_STATS_GET: - err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp); - goto exit; + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_datapath_multicast_group.id, info->nlhdr, + GFP_KERNEL); - case ODP_VPORT_STATS_SET: - err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp); - goto exit; + return 0; +} - case ODP_VPORT_ETHER_GET: - err = vport_user_ether_get((struct odp_vport_ether __user *)argp); - goto exit; +static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) +{ + struct sk_buff *reply; + struct datapath *dp; + int err; - case ODP_VPORT_ETHER_SET: - err = vport_user_ether_set((struct odp_vport_ether __user *)argp); - goto exit; + err = ovs_dp_cmd_validate(info->attrs); + if (err) + return err; - case ODP_VPORT_MTU_GET: - err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp); - goto exit; + dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); + if (IS_ERR(dp)) + return PTR_ERR(dp); - case ODP_VPORT_MTU_SET: - err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp); - goto exit; - } + reply = ovs_dp_cmd_build_info(dp, info->snd_pid, + info->snd_seq, OVS_DP_CMD_NEW); + if (IS_ERR(reply)) + return PTR_ERR(reply); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto exit; + return genlmsg_reply(reply, info); +} - switch (cmd) { - case ODP_DP_STATS: - err = get_dp_stats(dp, (struct odp_stats __user *)argp); - break; +static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) +{ + struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); + struct datapath *dp; + int skip = cb->args[0]; + int i = 0; + + list_for_each_entry(dp, &ovs_net->dps, list_node) { + if (i >= skip && + ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, NLM_F_MULTI, + OVS_DP_CMD_NEW) < 0) + break; + i++; + } - case ODP_GET_DROP_FRAGS: - err = put_user(dp->drop_frags, (int __user *)argp); - break; + cb->args[0] = i; - 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; + return skb->len; +} - case ODP_GET_LISTEN_MASK: - err = put_user(get_listen_mask(f), (int __user *)argp); - break; +static struct genl_ops dp_datapath_genl_ops[] = { + { .cmd = OVS_DP_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_new + }, + { .cmd = OVS_DP_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_del + }, + { .cmd = OVS_DP_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_get, + .dumpit = ovs_dp_cmd_dump + }, + { .cmd = OVS_DP_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = datapath_policy, + .doit = ovs_dp_cmd_set, + }, +}; - 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; +static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { +#ifdef HAVE_NLA_NUL_STRING + [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, + [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, + [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN }, +#else + [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) }, + [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN }, +#endif + [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, + [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, +}; - case ODP_GET_SFLOW_PROBABILITY: - err = put_user(dp->sflow_probability, (unsigned int __user *)argp); - break; +static struct genl_family dp_vport_genl_family = { + .id = GENL_ID_GENERATE, + .hdrsize = sizeof(struct ovs_header), + .name = OVS_VPORT_FAMILY, + .version = OVS_VPORT_VERSION, + .maxattr = OVS_VPORT_ATTR_MAX, + SET_NETNSOK +}; - case ODP_SET_SFLOW_PROBABILITY: - err = get_user(sflow_probability, (unsigned int __user *)argp); - if (!err) - dp->sflow_probability = sflow_probability; - break; +struct genl_multicast_group ovs_dp_vport_multicast_group = { + .name = OVS_VPORT_MCGROUP +}; - case ODP_VPORT_QUERY: - err = query_port(dp, (struct odp_port __user *)argp); - break; +/* Called with RTNL lock or RCU read lock. */ +static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, + u32 pid, u32 seq, u32 flags, u8 cmd) +{ + struct ovs_header *ovs_header; + struct ovs_vport_stats vport_stats; + int err; - case ODP_VPORT_LIST: - err = list_ports(dp, (struct odp_portvec __user *)argp); - break; + ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family, + flags, cmd); + if (!ovs_header) + return -EMSGSIZE; - case ODP_FLOW_FLUSH: - err = flush_flows(dp); - break; + ovs_header->dp_ifindex = get_dpifindex(vport->dp); - case ODP_FLOW_PUT: - err = put_flow(dp, (struct odp_flow_put __user *)argp); - break; + if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || + nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || + nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) || + nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid)) + goto nla_put_failure; - case ODP_FLOW_DEL: - err = del_flow(dp, (struct odp_flow __user *)argp); - break; + ovs_vport_get_stats(vport, &vport_stats); + if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats), + &vport_stats)) + goto nla_put_failure; - case ODP_FLOW_GET: - err = do_flowvec_ioctl(dp, argp, do_query_flows); - break; + if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, + vport->ops->get_addr(vport))) + goto nla_put_failure; - case ODP_FLOW_LIST: - err = do_flowvec_ioctl(dp, argp, do_list_flows); - break; + err = ovs_vport_get_options(vport, skb); + if (err == -EMSGSIZE) + goto error; - case ODP_EXECUTE: - err = execute_packet(dp, (struct odp_execute __user *)argp); - break; + return genlmsg_end(skb, ovs_header); - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); -exit: +nla_put_failure: + err = -EMSGSIZE; +error: + genlmsg_cancel(skb, ovs_header); 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_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv) +/* Called with RTNL lock or RCU read lock. */ +struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid, + u32 seq, u8 cmd) { - struct compat_odp_portvec pv; + struct sk_buff *skb; int retval; - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; - - retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports); - if (retval < 0) - return retval; + skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); + if (!skb) + return ERR_PTR(-ENOMEM); - return put_user(retval, &upv->n_ports); + retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd); + if (retval < 0) { + kfree_skb(skb); + return ERR_PTR(retval); + } + return skb; } -static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) +static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) { - compat_uptr_t actions; + return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1); +} - if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) || - __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) || - __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) || - __get_user(actions, &compat->actions) || - __get_user(flow->actions_len, &compat->actions_len) || - __get_user(flow->flags, &compat->flags)) - return -EFAULT; +/* Called with RTNL lock or RCU read lock. */ +static struct vport *lookup_vport(struct net *net, + struct ovs_header *ovs_header, + struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) +{ + struct datapath *dp; + struct vport *vport; - flow->actions = compat_ptr(actions); - return 0; + if (a[OVS_VPORT_ATTR_NAME]) { + vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); + if (!vport) + return ERR_PTR(-ENODEV); + if (ovs_header->dp_ifindex && + ovs_header->dp_ifindex != get_dpifindex(vport->dp)) + return ERR_PTR(-ENODEV); + return vport; + } else if (a[OVS_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); + + if (port_no >= DP_MAX_PORTS) + return ERR_PTR(-EFBIG); + + dp = get_dp(net, ovs_header->dp_ifindex); + if (!dp) + return ERR_PTR(-ENODEV); + + vport = ovs_vport_rtnl_rcu(dp, port_no); + if (!vport) + return ERR_PTR(-ENOENT); + return vport; + } else + return ERR_PTR(-EINVAL); } -static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp) +/* Called with RTNL lock. */ +static int change_vport(struct vport *vport, + struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) { - 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; + int err = 0; - error = do_put_flow(dp, &fp, &stats); - if (error) - return error; + if (a[OVS_VPORT_ATTR_STATS]) + ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS])); - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; + if (a[OVS_VPORT_ATTR_ADDRESS]) + err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS])); - return 0; + return err; } -static int compat_answer_query(struct sw_flow *flow, u32 query_flags, - struct compat_odp_flow __user *ufp) +static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) { - compat_uptr_t actions; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct vport_parms parms; + struct sk_buff *reply; + struct vport *vport; + struct datapath *dp; + u32 port_no; + int err; - if (get_user(actions, &ufp->actions)) - return -EFAULT; + err = -EINVAL; + if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || + !a[OVS_VPORT_ATTR_UPCALL_PID]) + goto exit; - return do_answer_query(flow, query_flags, &ufp->stats, - compat_ptr(actions), &ufp->actions_len); -} + err = ovs_vport_cmd_validate(a); + if (err) + goto exit; -static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp) -{ - struct sw_flow *flow; - struct odp_flow uf; - int error; + rtnl_lock(); + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + err = -ENODEV; + if (!dp) + goto exit_unlock; - if (compat_get_flow(&uf, ufp)) - return -EFAULT; + if (a[OVS_VPORT_ATTR_PORT_NO]) { + port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); - flow = do_del_flow(dp, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); + err = -EFBIG; + if (port_no >= DP_MAX_PORTS) + goto exit_unlock; - error = compat_answer_query(flow, 0, ufp); - flow_deferred_free(flow); - return error; -} + vport = ovs_vport_rtnl(dp, port_no); + err = -EBUSY; + if (vport) + goto exit_unlock; + } else { + for (port_no = 1; ; port_no++) { + if (port_no >= DP_MAX_PORTS) { + err = -EFBIG; + goto exit_unlock; + } + vport = ovs_vport_rtnl(dp, port_no); + if (!vport) + break; + } + } -static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) -{ - struct tbl *table = rcu_dereference(dp->table); - u32 i; + parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); + parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); + parms.options = a[OVS_VPORT_ATTR_OPTIONS]; + parms.dp = dp; + parms.port_no = port_no; + parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); - for (i = 0; i < n_flows; i++) { - struct compat_odp_flow __user *ufp = &flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; + vport = new_vport(&parms); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; - if (compat_get_flow(&uf, ufp)) - return -EFAULT; + ovs_dp_sysfs_add_if(vport); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp); - if (error) - return -EFAULT; + err = change_vport(vport, a); + if (!err) { + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, + info->snd_seq, + OVS_VPORT_CMD_NEW); + if (IS_ERR(reply)) + err = PTR_ERR(reply); } - return n_flows; -} + if (err) { + ovs_dp_detach_port(vport); + goto exit_unlock; + } + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); -struct compat_list_flows_cbdata { - struct compat_odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; -}; +exit_unlock: + rtnl_unlock(); +exit: + return err; +} -static int compat_list_flow(struct tbl_node *node, void *cbdata_) +static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) { - struct sw_flow *flow = flow_cast(node); - struct compat_list_flows_cbdata *cbdata = cbdata_; - struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; - int error; + struct nlattr **a = info->attrs; + struct sk_buff *reply; + struct vport *vport; + int err; - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) - return -EFAULT; - error = compat_answer_query(flow, 0, ufp); - if (error) - return error; + err = ovs_vport_cmd_validate(a); + if (err) + goto exit; - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; -} + rtnl_lock(); + vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; -static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) -{ - struct compat_list_flows_cbdata cbdata; - int error; + err = 0; + if (a[OVS_VPORT_ATTR_TYPE] && + nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) + err = -EINVAL; - if (!n_flows) - return 0; + if (!err && a[OVS_VPORT_ATTR_OPTIONS]) + err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); + if (!err) + err = change_vport(vport, a); + else + goto exit_unlock; + if (!err && a[OVS_VPORT_ATTR_UPCALL_PID]) + vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); + + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_VPORT_CMD_NEW); + if (IS_ERR(reply)) { + netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0, + ovs_dp_vport_multicast_group.id, PTR_ERR(reply)); + goto exit_unlock; + } - cbdata.uflows = flows; - cbdata.n_flows = n_flows; - cbdata.listed_flows = 0; + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata); - return error ? error : cbdata.listed_flows; +exit_unlock: + rtnl_unlock(); +exit: + return err; } -static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - struct compat_odp_flow *, - u32 n_flows)) +static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) { - 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; + struct nlattr **a = info->attrs; + struct sk_buff *reply; + struct vport *vport; + int err; - if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow)) - return -EINVAL; + err = ovs_vport_cmd_validate(a); + if (err) + goto exit; - flows = compat_ptr(flowvec.flows); - if (!access_ok(VERIFY_WRITE, flows, - flowvec.n_flows * sizeof(struct compat_odp_flow))) - return -EFAULT; + rtnl_lock(); + vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; - retval = function(dp, flows, flowvec.n_flows); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : put_user(retval, &uflowvec->n_flows)); -} + if (vport->port_no == OVSP_LOCAL) { + err = -EINVAL; + goto exit_unlock; + } -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; + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_VPORT_CMD_DEL); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; - 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; + ovs_dp_detach_port(vport); - execute.actions = compat_ptr(actions); - execute.data = compat_ptr(data); + genl_notify(reply, genl_info_net(info), info->snd_pid, + ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL); - return do_execute(dp, &execute); +exit_unlock: + rtnl_unlock(); +exit: + return err; } -static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) +static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) { - int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp; + struct nlattr **a = info->attrs; + struct ovs_header *ovs_header = info->userhdr; + struct sk_buff *reply; + struct vport *vport; int err; - switch (cmd) { - case ODP_DP_DESTROY: - case ODP_FLOW_FLUSH: - /* Ioctls that don't need any translation at all. */ - return openvswitch_ioctl(f, cmd, argp); - - case ODP_DP_CREATE: - case ODP_VPORT_ATTACH: - case ODP_VPORT_DETACH: - case ODP_VPORT_MOD: - case ODP_VPORT_MTU_SET: - case ODP_VPORT_MTU_GET: - case ODP_VPORT_ETHER_SET: - case ODP_VPORT_ETHER_GET: - case ODP_VPORT_STATS_SET: - case ODP_VPORT_STATS_GET: - case ODP_DP_STATS: - case ODP_GET_DROP_FRAGS: - case ODP_SET_DROP_FRAGS: - case ODP_SET_LISTEN_MASK: - case ODP_GET_LISTEN_MASK: - case ODP_SET_SFLOW_PROBABILITY: - case ODP_GET_SFLOW_PROBABILITY: - case ODP_VPORT_QUERY: - /* 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) + err = ovs_vport_cmd_validate(a); + if (err) goto exit; - switch (cmd) { - case ODP_VPORT_LIST32: - err = compat_list_ports(dp, compat_ptr(argp)); - break; - - 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; + rcu_read_lock(); + vport = lookup_vport(sock_net(skb->sk), ovs_header, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock; - case ODP_FLOW_GET32: - err = compat_flowvec_ioctl(dp, argp, compat_query_flows); - break; + reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq, + OVS_VPORT_CMD_NEW); + err = PTR_ERR(reply); + if (IS_ERR(reply)) + goto exit_unlock; - case ODP_FLOW_LIST32: - err = compat_flowvec_ioctl(dp, argp, compat_list_flows); - break; + rcu_read_unlock(); - case ODP_EXECUTE32: - err = compat_execute(dp, compat_ptr(argp)); - break; + return genlmsg_reply(reply, info); - default: - err = -ENOIOCTLCMD; - break; - } - mutex_unlock(&dp->mutex); +exit_unlock: + rcu_read_unlock(); exit: return err; } -#endif -/* Unfortunately this function is not exported so this is a verbatim copy - * from net/core/datagram.c in 2.6.30. */ -static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, - u8 __user *to, int len, - __wsum *csump) +static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) { - int start = skb_headlen(skb); - int pos = 0; - int i, copy = start - offset; - - /* Copy header. */ - if (copy > 0) { - int err = 0; - if (copy > len) - copy = len; - *csump = csum_and_copy_to_user(skb->data + offset, to, copy, - *csump, &err); - if (err) - goto fault; - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos = copy; - } + struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); + struct datapath *dp; + int bucket = cb->args[0], skip = cb->args[1]; + int i, j = 0; - for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { - int end; - - WARN_ON(start > offset + len); - - end = start + skb_shinfo(skb)->frags[i].size; - if ((copy = end - offset) > 0) { - __wsum csum2; - int err = 0; - u8 *vaddr; - skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; - struct page *page = frag->page; - - if (copy > len) - copy = len; - vaddr = kmap(page); - csum2 = csum_and_copy_to_user(vaddr + - frag->page_offset + - offset - start, - to, copy, 0, &err); - kunmap(page); - if (err) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if (!(len -= copy)) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; - } + dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); + if (!dp) + return -ENODEV; - if (skb_shinfo(skb)->frag_list) { - struct sk_buff *list = skb_shinfo(skb)->frag_list; - - for (; list; list=list->next) { - int end; - - WARN_ON(start > offset + len); - - end = start + list->len; - if ((copy = end - offset) > 0) { - __wsum csum2 = 0; - if (copy > len) - copy = len; - if (skb_copy_and_csum_datagram(list, - offset - start, - to, copy, - &csum2)) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; + rcu_read_lock(); + for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { + struct vport *vport; + struct hlist_node *n; + + j = 0; + hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) { + if (j >= skip && + ovs_vport_cmd_fill_info(vport, skb, + NETLINK_CB(cb->skb).pid, + cb->nlh->nlmsg_seq, + NLM_F_MULTI, + OVS_VPORT_CMD_NEW) < 0) + goto out; + + j++; } + skip = 0; } - if (!len) - return 0; +out: + rcu_read_unlock(); -fault: - return -EFAULT; + cb->args[0] = i; + cb->args[1] = j; + + return skb->len; } -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; - size_t copy_bytes, tot_copy_bytes; - int retval; +static struct genl_ops dp_vport_genl_ops[] = { + { .cmd = OVS_VPORT_CMD_NEW, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_new + }, + { .cmd = OVS_VPORT_CMD_DEL, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_del + }, + { .cmd = OVS_VPORT_CMD_GET, + .flags = 0, /* OK for unprivileged users. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_get, + .dumpit = ovs_vport_cmd_dump + }, + { .cmd = OVS_VPORT_CMD_SET, + .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ + .policy = vport_policy, + .doit = ovs_vport_cmd_set, + }, +}; - if (!dp) - return -ENODEV; +struct genl_family_and_ops { + struct genl_family *family; + struct genl_ops *ops; + int n_ops; + struct genl_multicast_group *group; +}; - if (nbytes == 0 || !listeners) - return 0; +static const struct genl_family_and_ops dp_genl_families[] = { + { &dp_datapath_genl_family, + dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops), + &ovs_dp_datapath_multicast_group }, + { &dp_vport_genl_family, + dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops), + &ovs_dp_vport_multicast_group }, + { &dp_flow_genl_family, + dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops), + &ovs_dp_flow_multicast_group }, + { &dp_packet_genl_family, + dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), + NULL }, +}; - for (;;) { - int i; +static void dp_unregister_genl(int n_families) +{ + int i; - for (i = 0; i < DP_N_QUEUES; i++) { - if (listeners & (1 << i)) { - skb = skb_dequeue(&dp->queues[i]); - if (skb) - goto success; - } - } + for (i = 0; i < n_families; i++) + genl_unregister_family(dp_genl_families[i].family); +} - if (f->f_flags & O_NONBLOCK) { - retval = -EAGAIN; - goto error; - } +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); - copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes); + return 0; + +error: + dp_unregister_genl(n_registered); + return err; +} - retval = 0; - if (skb->ip_summed == CHECKSUM_PARTIAL) { - if (copy_bytes == skb->len) { - __wsum csum = 0; - u16 csum_start, csum_offset; +static int __rehash_flow_table(void *dummy) +{ + struct datapath *dp; + struct net *net; - get_skb_csum_pointers(skb, &csum_start, &csum_offset); - BUG_ON(csum_start >= skb_headlen(skb)); - retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start, - copy_bytes - csum_start, &csum); - if (!retval) { - __sum16 __user *csump; + rtnl_lock(); + for_each_net(net) { + struct ovs_net *ovs_net = net_generic(net, ovs_net_id); - copy_bytes = csum_start; - csump = (__sum16 __user *)(buf + csum_start + csum_offset); + list_for_each_entry(dp, &ovs_net->dps, list_node) { + struct flow_table *old_table = genl_dereference(dp->table); + struct flow_table *new_table; - BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes); - put_user(csum_fold(csum), csump); + new_table = ovs_flow_tbl_rehash(old_table); + if (!IS_ERR(new_table)) { + rcu_assign_pointer(dp->table, new_table); + ovs_flow_tbl_deferred_destroy(old_table); } - } else - retval = skb_checksum_help(skb); + } } + rtnl_unlock(); + return 0; +} - if (!retval) { - struct iovec __user iov; +static void rehash_flow_table(struct work_struct *work) +{ + genl_exec(__rehash_flow_table, NULL); + schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); +} - iov.iov_base = buf; - iov.iov_len = copy_bytes; - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); - } +static int dp_destroy_all(void *data) +{ + struct datapath *dp, *dp_next; + struct ovs_net *ovs_net = data; - if (!retval) - retval = tot_copy_bytes; + list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) + __dp_destroy(dp); - kfree_skb(skb); - return retval; + return 0; +} -error: - mutex_unlock(&dp->mutex); - return retval; +static int __net_init ovs_init_net(struct net *net) +{ + struct ovs_net *ovs_net = net_generic(net, ovs_net_id); + + INIT_LIST_HEAD(&ovs_net->dps); + return 0; } -static unsigned int openvswitch_poll(struct file *file, poll_table *wait) +static void __net_exit ovs_exit_net(struct net *net) { - 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; + struct ovs_net *ovs_net = net_generic(net, ovs_net_id); + + genl_exec(dp_destroy_all, ovs_net); } -static struct file_operations openvswitch_fops = { - .read = openvswitch_read, - .poll = openvswitch_poll, - .unlocked_ioctl = openvswitch_ioctl, -#ifdef CONFIG_COMPAT - .compat_ioctl = openvswitch_compat_ioctl, -#endif +static struct pernet_operations ovs_net_ops = { + .init = ovs_init_net, + .exit = ovs_exit_net, + .id = &ovs_net_id, + .size = sizeof(struct ovs_net), }; -static int major; - static int __init dp_init(void) { struct sk_buff *dummy_skb; @@ -1980,43 +2179,75 @@ static int __init dp_init(void) BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb)); - printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR); + pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n", + VERSION); - err = flow_init(); + err = genl_exec_init(); if (err) goto error; - err = vport_init(); + err = ovs_workqueues_init(); + if (err) + goto error_genl_exec; + + err = ovs_tnl_init(); + if (err) + goto error_wq; + + err = ovs_flow_init(); + if (err) + goto error_tnl_exit; + + err = ovs_vport_init(); if (err) goto error_flow_exit; - err = register_netdevice_notifier(&dp_device_notifier); + err = register_pernet_device(&ovs_net_ops); if (err) goto error_vport_exit; - major = register_chrdev(0, "openvswitch", &openvswitch_fops); + err = register_netdevice_notifier(&ovs_dp_device_notifier); + if (err) + goto error_netns_exit; + + err = dp_register_genl(); if (err < 0) goto error_unreg_notifier; + schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL); + return 0; error_unreg_notifier: - unregister_netdevice_notifier(&dp_device_notifier); + unregister_netdevice_notifier(&ovs_dp_device_notifier); +error_netns_exit: + unregister_pernet_device(&ovs_net_ops); error_vport_exit: - vport_exit(); + ovs_vport_exit(); error_flow_exit: - flow_exit(); + ovs_flow_exit(); +error_tnl_exit: + ovs_tnl_exit(); +error_wq: + ovs_workqueues_exit(); +error_genl_exec: + genl_exec_exit(); error: return err; } static void dp_cleanup(void) { + cancel_delayed_work_sync(&rehash_flow_wq); + dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); + unregister_netdevice_notifier(&ovs_dp_device_notifier); + unregister_pernet_device(&ovs_net_ops); rcu_barrier(); - unregister_chrdev(major, "openvswitch"); - unregister_netdevice_notifier(&dp_device_notifier); - vport_exit(); - flow_exit(); + ovs_vport_exit(); + ovs_flow_exit(); + ovs_tnl_exit(); + ovs_workqueues_exit(); + genl_exec_exit(); } module_init(dp_init);