X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=3d32cd47ae9b6d1f4cae7ac43fcaabe76e64d72a;hb=ea7bd5973fb426f3a6ea66b0953c6000980e77ec;hp=165f0a9978df0e600f84cec5fd5c6c770bbe6fc3;hpb=1d7241c7298615ca4e2b45647bf231e22f80e667;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index 165f0a997..3d32cd47a 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks. + * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks. * Distributed under the terms of the GNU GPL version 2. * * Significant portions of this file may be copied from parts of the Linux @@ -8,6 +8,8 @@ /* Functions for managing the dp interface/device. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include #include @@ -27,32 +29,31 @@ #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 "vport-internal_dev.h" -#include "compat.h" - - int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); EXPORT_SYMBOL(dp_ioctl_hook); @@ -62,23 +63,21 @@ EXPORT_SYMBOL(dp_ioctl_hook); * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL * lock first. * - * It is safe to access the datapath and dp_port structures with just + * It is safe to access the datapath and vport structures with just * dp_mutex. */ -static struct datapath *dps[ODP_MAX]; +static struct datapath __rcu *dps[ODP_MAX]; static DEFINE_MUTEX(dp_mutex); -/* Number of milliseconds between runs of the maintenance thread. */ -#define MAINT_SLEEP_MSECS 1000 - -static int new_dp_port(struct datapath *, struct odp_port *, int port_no); +static struct vport *new_vport(const struct vport_parms *); /* Must be called with rcu_read_lock or dp_mutex. */ struct datapath *get_dp(int dp_idx) { if (dp_idx < 0 || dp_idx >= ODP_MAX) return NULL; - return rcu_dereference(dps[dp_idx]); + return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() || + lockdep_is_held(&dp_mutex)); } EXPORT_SYMBOL_GPL(get_dp); @@ -94,10 +93,22 @@ static struct datapath *get_dp_locked(int dp_idx) return dp; } +static struct tbl *get_table_protected(struct datapath *dp) +{ + return rcu_dereference_protected(dp->table, + lockdep_is_held(&dp->mutex)); +} + +static struct vport *get_vport_protected(struct datapath *dp, u16 port_no) +{ + return rcu_dereference_protected(dp->ports[port_no], + lockdep_is_held(&dp->mutex)); +} + /* Must be called with rcu_read_lock or RTNL lock. */ const char *dp_name(const struct datapath *dp) { - return vport_get_name(dp->ports[ODPP_LOCAL]->vport); + return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); } static inline size_t br_nlmsg_size(void) @@ -112,12 +123,12 @@ static inline size_t br_nlmsg_size(void) } static int dp_fill_ifinfo(struct sk_buff *skb, - const struct dp_port *port, + const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; - int ifindex = vport_get_ifindex(port->vport); - int iflink = vport_get_iflink(port->vport); + struct datapath *dp = port->dp; + int ifindex = vport_get_ifindex(port); + int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; struct nlmsghdr *nlh; @@ -136,21 +147,21 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->__ifi_pad = 0; hdr->ifi_type = ARPHRD_ETHER; hdr->ifi_index = ifindex; - hdr->ifi_flags = vport_get_flags(port->vport); + hdr->ifi_flags = vport_get_flags(port); hdr->ifi_change = 0; - NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport)); - NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport)); - NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport)); + NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port)); + NLA_PUT_U32(skb, IFLA_MASTER, + vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL))); + NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port)); #ifdef IFLA_OPERSTATE NLA_PUT_U8(skb, IFLA_OPERSTATE, - vport_is_running(port->vport) - ? vport_get_operstate(port->vport) + vport_is_running(port) + ? vport_get_operstate(port) : IF_OPER_DOWN); #endif - NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, - vport_get_addr(port->vport)); + NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port)); if (ifindex != iflink) NLA_PUT_U32(skb, IFLA_LINK,iflink); @@ -162,7 +173,7 @@ nla_put_failure: return -EMSGSIZE; } -static void dp_ifinfo_notify(int event, struct dp_port *port) +static void dp_ifinfo_notify(int event, struct vport *port) { struct sk_buff *skb; int err = -ENOBUFS; @@ -197,8 +208,9 @@ static struct kobj_type dp_ktype = { static int create_dp(int dp_idx, const char __user *devnamep) { - struct odp_port internal_dev_port; + struct vport_parms parms; char devname[IFNAMSIZ]; + struct vport *vport; struct datapath *dp; int err; int i; @@ -213,7 +225,7 @@ static int create_dp(int dp_idx, const char __user *devnamep) goto err; } } else { - snprintf(devname, sizeof devname, "of%d", dp_idx); + snprintf(devname, sizeof(devname), "of%d", dp_idx); } rtnl_lock(); @@ -230,11 +242,12 @@ static int create_dp(int dp_idx, const char __user *devnamep) goto err_put_module; err = -ENOMEM; - dp = kzalloc(sizeof *dp, GFP_KERNEL); + dp = kzalloc(sizeof(*dp), GFP_KERNEL); if (dp == NULL) goto err_put_module; INIT_LIST_HEAD(&dp->port_list); mutex_init(&dp->mutex); + mutex_lock(&dp->mutex); dp->dp_idx = dp_idx; for (i = 0; i < DP_N_QUEUES; i++) skb_queue_head_init(&dp->queues[i]); @@ -247,16 +260,19 @@ static int create_dp(int dp_idx, const char __user *devnamep) /* Allocate table. */ err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(0)); + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); if (!dp->table) goto err_free_dp; /* Set up our datapath device. */ - BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname)); - strcpy(internal_dev_port.devname, devname); - internal_dev_port.flags = ODP_PORT_INTERNAL; - err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL); - if (err) { + parms.name = devname; + parms.type = ODP_VPORT_TYPE_INTERNAL; + parms.options = NULL; + parms.dp = dp; + parms.port_no = ODPP_LOCAL; + vport = new_vport(&parms); + if (IS_ERR(vport)) { + err = PTR_ERR(vport); if (err == -EBUSY) err = -EEXIST; @@ -265,22 +281,26 @@ static int create_dp(int dp_idx, const char __user *devnamep) dp->drop_frags = 0; dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) + if (!dp->stats_percpu) { + err = -ENOMEM; goto err_destroy_local_port; + } rcu_assign_pointer(dps[dp_idx], dp); + dp_sysfs_add_dp(dp); + + mutex_unlock(&dp->mutex); mutex_unlock(&dp_mutex); rtnl_unlock(); - dp_sysfs_add_dp(dp); - return 0; err_destroy_local_port: - dp_detach_port(dp->ports[ODPP_LOCAL], 1); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); err_destroy_table: - tbl_destroy(dp->table, NULL); + tbl_destroy(get_table_protected(dp), NULL); err_free_dp: + mutex_unlock(&dp->mutex); kfree(dp); err_put_module: module_put(THIS_MODULE); @@ -291,161 +311,76 @@ err: return err; } -static void do_destroy_dp(struct datapath *dp) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct dp_port *p, *n; + struct datapath *dp = container_of(rcu, struct datapath, rcu); int i; - list_for_each_entry_safe (p, n, &dp->port_list, node) - if (p->port_no != ODPP_LOCAL) - dp_detach_port(p, 1); - - dp_sysfs_del_dp(dp); - - rcu_assign_pointer(dps[dp->dp_idx], NULL); - - dp_detach_port(dp->ports[ODPP_LOCAL], 1); - - tbl_destroy(dp->table, flow_free_tbl); - for (i = 0; i < DP_N_QUEUES; i++) skb_queue_purge(&dp->queues[i]); - for (i = 0; i < DP_MAX_GROUPS; i++) - kfree(dp->groups[i]); + + tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); - module_put(THIS_MODULE); } static int destroy_dp(int dp_idx) { struct datapath *dp; - int err; + int err = 0; + struct vport *p, *n; rtnl_lock(); mutex_lock(&dp_mutex); dp = get_dp(dp_idx); - err = -ENODEV; - if (!dp) - goto err_unlock; + if (!dp) { + err = -ENODEV; + goto out; + } - do_destroy_dp(dp); - err = 0; + mutex_lock(&dp->mutex); -err_unlock: + list_for_each_entry_safe (p, n, &dp->port_list, node) + if (p->port_no != ODPP_LOCAL) + dp_detach_port(p); + + dp_sysfs_del_dp(dp); + rcu_assign_pointer(dps[dp->dp_idx], NULL); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); + + mutex_unlock(&dp->mutex); + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); + +out: mutex_unlock(&dp_mutex); rtnl_unlock(); return err; } -static void release_dp_port(struct kobject *kobj) -{ - struct dp_port *p = container_of(kobj, struct dp_port, kobj); - kfree(p); -} - -static struct kobj_type brport_ktype = { -#ifdef CONFIG_SYSFS - .sysfs_ops = &brport_sysfs_ops, -#endif - .release = release_dp_port -}; - -/* Called with RTNL lock and dp_mutex. */ -static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no) +/* Called with RTNL lock and dp->mutex. */ +static struct vport *new_vport(const struct vport_parms *parms) { struct vport *vport; - struct dp_port *p; - int err; - - vport = vport_locate(odp_port->devname); - if (!vport) { - vport_lock(); - if (odp_port->flags & ODP_PORT_INTERNAL) - vport = __vport_add(odp_port->devname, "internal", NULL); - else - vport = __vport_add(odp_port->devname, "netdev", NULL); + vport_lock(); + vport = vport_add(parms); + if (!IS_ERR(vport)) { + struct datapath *dp = parms->dp; - vport_unlock(); + rcu_assign_pointer(dp->ports[parms->port_no], vport); + list_add_rcu(&vport->node, &dp->port_list); + dp->n_ports++; - if (IS_ERR(vport)) - return PTR_ERR(vport); + dp_ifinfo_notify(RTM_NEWLINK, vport); } + vport_unlock(); - p = kzalloc(sizeof(*p), GFP_KERNEL); - if (!p) - return -ENOMEM; - - p->port_no = port_no; - p->dp = dp; - atomic_set(&p->sflow_pool, 0); - - err = vport_attach(vport, p); - if (err) { - kfree(p); - return err; - } - - rcu_assign_pointer(dp->ports[port_no], p); - list_add_rcu(&p->node, &dp->port_list); - dp->n_ports++; - - /* Initialize kobject for bridge. This will be added as - * /sys/class/net//brport later, if sysfs is enabled. */ - p->kobj.kset = NULL; - kobject_init(&p->kobj, &brport_ktype); - - dp_ifinfo_notify(RTM_NEWLINK, p); - - return 0; -} - -static int attach_port(int dp_idx, struct odp_port __user *portp) -{ - struct datapath *dp; - struct odp_port port; - int port_no; - int err; - - err = -EFAULT; - if (copy_from_user(&port, portp, sizeof port)) - goto out; - port.devname[IFNAMSIZ - 1] = '\0'; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - for (port_no = 1; port_no < DP_MAX_PORTS; port_no++) - if (!dp->ports[port_no]) - goto got_port_no; - err = -EFBIG; - goto out_unlock_dp; - -got_port_no: - err = new_dp_port(dp, &port, port_no); - if (err) - goto out_unlock_dp; - - set_internal_devs_mtu(dp); - dp_sysfs_add_if(dp->ports[port_no]); - - err = put_user(port_no, &portp->port); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: - return err; + return vport; } -int dp_detach_port(struct dp_port *p, int may_delete) +int dp_detach_port(struct vport *p) { - struct vport *vport = p->vport; int err; ASSERT_RTNL(); @@ -459,323 +394,191 @@ int dp_detach_port(struct dp_port *p, int may_delete) list_del_rcu(&p->node); rcu_assign_pointer(p->dp->ports[p->port_no], NULL); - err = vport_detach(vport); - if (err) - return err; - - /* Then wait until no one is still using it, and destroy it. */ - synchronize_rcu(); + /* Then destroy it. */ + vport_lock(); + err = vport_del(p); + vport_unlock(); - if (may_delete) { - const char *port_type = vport_get_type(vport); - - if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) { - vport_lock(); - __vport_del(vport); - vport_unlock(); - } - } - - kobject_put(&p->kobj); - - return 0; -} - -static int detach_port(int dp_idx, int port_no) -{ - struct dp_port *p; - struct datapath *dp; - int err; - - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) - goto out; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - p = dp->ports[port_no]; - err = -ENOENT; - if (!p) - goto out_unlock_dp; - - err = dp_detach_port(p, 1); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: return err; } -/* Must be called with rcu_read_lock and with bottom-halves disabled. */ -void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb) +/* Must be called with rcu_read_lock. */ +void dp_process_received_packet(struct vport *p, struct sk_buff *skb) { struct datapath *dp = p->dp; struct dp_stats_percpu *stats; - struct odp_flow_key key; - struct tbl_node *flow_node; - - WARN_ON_ONCE(skb_shared(skb)); - skb_warn_if_lro(skb); + int stats_counter_off; + struct sw_flow_actions *acts; + struct loop_counter *loop; + int error; - OVS_CB(skb)->dp_port = p; + OVS_CB(skb)->vport = p; - /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */ - stats = percpu_ptr(dp->stats_percpu, smp_processor_id()); + if (!OVS_CB(skb)->flow) { + struct sw_flow_key key; + struct tbl_node *flow_node; + bool is_frag; - if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { - if (dp->drop_frags) { + /* Extract flow from 'skb' into 'key'. */ + error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag); + if (unlikely(error)) { kfree_skb(skb); - stats->n_frags++; return; } - } - - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); - if (flow_node) { - struct sw_flow *flow = flow_cast(flow_node); - struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts); - flow_used(flow, skb); - execute_actions(dp, skb, &key, acts->actions, acts->n_actions, - GFP_ATOMIC); - stats->n_hit++; - } else { - stats->n_missed++; - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); - } -} -#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) -/* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We - * can't call this function directly because it isn't exported in all - * versions. */ -int vswitch_skb_checksum_setup(struct sk_buff *skb) -{ - struct iphdr *iph; - unsigned char *th; - int err = -EPROTO; - __u16 csum_start, csum_offset; + if (is_frag && dp->drop_frags) { + kfree_skb(skb); + stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); + goto out; + } - if (!skb->proto_csum_blank) - return 0; + /* Look up flow. */ + flow_node = tbl_lookup(rcu_dereference(dp->table), &key, + flow_hash(&key), flow_cmp); + if (unlikely(!flow_node)) { + struct dp_upcall_info upcall; + + upcall.type = _ODPL_MISS_NR; + upcall.key = &key; + upcall.userdata = 0; + upcall.sample_pool = 0; + upcall.actions = NULL; + upcall.actions_len = 0; + dp_upcall(dp, skb, &upcall); + stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + goto out; + } - if (skb->protocol != htons(ETH_P_IP)) - goto out; + OVS_CB(skb)->flow = flow_cast(flow_node); + } - if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data)) - goto out; + stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + flow_used(OVS_CB(skb)->flow, skb); - iph = ip_hdr(skb); - th = skb_network_header(skb) + 4 * iph->ihl; + acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); - csum_start = th - skb->head; - switch (iph->protocol) { - case IPPROTO_TCP: - csum_offset = offsetof(struct tcphdr, check); - break; - case IPPROTO_UDP: - csum_offset = offsetof(struct udphdr, check); - break; - default: - if (net_ratelimit()) - printk(KERN_ERR "Attempting to checksum a non-" - "TCP/UDP packet, dropping a protocol" - " %d packet", iph->protocol); - goto out; + /* Check whether we've looped too much. */ + loop = loop_get_counter(); + if (unlikely(++loop->count > MAX_LOOPS)) + loop->looping = true; + if (unlikely(loop->looping)) { + loop_suppress(dp, acts); + kfree_skb(skb); + goto out_loop; } - if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data)) - goto out; - - skb->ip_summed = CHECKSUM_PARTIAL; - skb->proto_csum_blank = 0; + /* Execute actions. */ + execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions, + acts->actions_len); -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - skb->csum_start = csum_start; - skb->csum_offset = csum_offset; -#else - skb_set_transport_header(skb, csum_start - skb_headroom(skb)); - skb->csum = csum_offset; -#endif + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + loop_suppress(dp, acts); - err = 0; +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + loop_put_counter(); out: - return err; -} -#endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */ - - /* Types of checksums that we can receive (these all refer to L4 checksums): - * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full - * (though not verified) checksum in packet but not in skb->csum. Packets - * from the bridge local port will also have this type. - * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums, - * also the GRE module. This is the same as CHECKSUM_NONE, except it has - * a valid skb->csum. Importantly, both contain a full checksum (not - * verified) in the packet itself. The only difference is that if the - * packet gets to L4 processing on this machine (not in DomU) we won't - * have to recompute the checksum to verify. Most hardware devices do not - * produce packets with this type, even if they support receive checksum - * offloading (they produce type #5). - * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to - * be computed if it is sent off box. Unfortunately on earlier kernels, - * this case is impossible to distinguish from #2, despite having opposite - * meanings. Xen adds an extra field on earlier kernels (see #4) in order - * to distinguish the different states. - * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was - * generated locally by a Xen DomU and has a partial checksum. If it is - * handled on this machine (Dom0 or DomU), then the checksum will not be - * computed. If it goes off box, the checksum in the packet needs to be - * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW - * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later - * kernels, this combination is replaced with CHECKSUM_PARTIAL. - * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct - * full checksum or using a protocol without a checksum. skb->csum is - * undefined. This is common from devices with receive checksum - * offloading. This is somewhat similar to CHECKSUM_NONE, except that - * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY. - * - * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are - * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear - * based on whether it is on the transmit or receive path. After the datapath - * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a - * checksum, we will panic. Since we can receive packets with checksums, we - * assume that all CHECKSUM_HW packets have checksums and map them to - * CHECKSUM_NONE, which has a similar meaning (the it is only different if the - * packet is processed by the local IP stack, in which case it will need to - * be reverified). If we receive a packet with CHECKSUM_HW that really means - * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there - * shouldn't be any devices that do this with bridging. */ -void -compute_ip_summed(struct sk_buff *skb, bool xmit) -{ - /* For our convenience these defines change repeatedly between kernel - * versions, so we can't just copy them over... */ - switch (skb->ip_summed) { - case CHECKSUM_NONE: - OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - break; - case CHECKSUM_UNNECESSARY: - OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY; - break; -#ifdef CHECKSUM_HW - /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE. - * However, on the receive side we should only get CHECKSUM_PARTIAL - * packets from Xen, which uses some special fields to represent this - * (see below). Since we can only make one type work, pick the one - * that actually happens in practice. - * - * On the transmit side (basically after skb_checksum_setup() - * has been run or on internal dev transmit), packets with - * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */ - case CHECKSUM_HW: - if (!xmit) - OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; - else - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; + /* Update datapath statistics. */ + local_bh_disable(); + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); - break; -#else - case CHECKSUM_COMPLETE: - OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; - break; - case CHECKSUM_PARTIAL: - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; - break; -#endif - default: - printk(KERN_ERR "openvswitch: unknown checksum type %d\n", - skb->ip_summed); - /* None seems the safest... */ - OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - } - -#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) - /* Xen has a special way of representing CHECKSUM_PARTIAL on older - * kernels. It should not be set on the transmit path though. */ - if (skb->proto_csum_blank) - OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; - - WARN_ON_ONCE(skb->proto_csum_blank && xmit); -#endif + write_seqcount_begin(&stats->seqlock); + (*(u64 *)((u8 *)stats + stats_counter_off))++; + write_seqcount_end(&stats->seqlock); + + local_bh_enable(); } -/* This function closely resembles skb_forward_csum() used by the bridge. It - * is slightly different because we are only concerned with bridging and not - * other types of forwarding and can get away with slightly more optimal - * behavior.*/ -void -forward_ip_summed(struct sk_buff *skb) +static void copy_and_csum_skb(struct sk_buff *skb, void *to) { -#ifdef CHECKSUM_HW - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) - skb->ip_summed = CHECKSUM_NONE; -#endif + u16 csum_start, csum_offset; + __wsum csum; + + get_skb_csum_pointers(skb, &csum_start, &csum_offset); + csum_start -= skb_headroom(skb); + BUG_ON(csum_start >= skb_headlen(skb)); + + skb_copy_bits(skb, 0, to, csum_start); + + csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start, + skb->len - csum_start, 0); + *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum); } /* Append each packet in 'skb' list to 'queue'. There will be only one packet * unless we broke up a GSO packet. */ -static int -queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, - int queue_no, u32 arg) +static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { struct sk_buff *nskb; int port_no; int err; - if (OVS_CB(skb)->dp_port) - port_no = OVS_CB(skb)->dp_port->port_no; + if (OVS_CB(skb)->vport) + port_no = OVS_CB(skb)->vport->port_no; else port_no = ODPP_LOCAL; do { - struct odp_msg *header; + struct odp_packet *upcall; + struct sk_buff *user_skb; /* to be queued to userspace */ + struct nlattr *nla; + unsigned int len; nskb = skb->next; skb->next = NULL; - /* If a checksum-deferred packet is forwarded to the - * controller, correct the pointers and checksum. - */ - err = vswitch_skb_checksum_setup(skb); - if (err) + len = sizeof(struct odp_packet); + len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */ + len += nla_total_size(skb->len); + len += nla_total_size(FLOW_BUFSIZE); + if (upcall_info->userdata) + len += nla_total_size(8); + if (upcall_info->sample_pool) + len += nla_total_size(4); + if (upcall_info->actions_len) + len += nla_total_size(upcall_info->actions_len); + + user_skb = alloc_skb(len, GFP_ATOMIC); + if (!user_skb) goto err_kfree_skbs; - if (skb->ip_summed == CHECKSUM_PARTIAL) { - -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - /* Until 2.6.22, the start of the transport header was - * also the start of data to be checksummed. Linux - * 2.6.22 introduced the csum_start field for this - * purpose, but we should point the transport header to - * it anyway for backward compatibility, as - * dev_queue_xmit() does even in 2.6.28. */ - skb_set_transport_header(skb, skb->csum_start - - skb_headroom(skb)); -#endif + upcall = (struct odp_packet *)__skb_put(user_skb, sizeof(*upcall)); + upcall->dp_idx = dp->dp_idx; + + nla_put_u32(user_skb, ODP_PACKET_ATTR_TYPE, upcall_info->type); + + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY); + flow_to_nlattrs(upcall_info->key, user_skb); + nla_nest_end(user_skb, nla); + + if (upcall_info->userdata) + nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata); + if (upcall_info->sample_pool) + nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool); + if (upcall_info->actions_len) { + const struct nlattr *actions = upcall_info->actions; + u32 actions_len = upcall_info->actions_len; - err = skb_checksum_help(skb); - if (err) - goto err_kfree_skbs; + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS); + memcpy(__skb_put(user_skb, actions_len), actions, actions_len); + nla_nest_end(user_skb, nla); } - err = skb_cow(skb, sizeof *header); - if (err) - goto err_kfree_skbs; + nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len); + if (skb->ip_summed == CHECKSUM_PARTIAL) + copy_and_csum_skb(skb, nla_data(nla)); + else + skb_copy_bits(skb, 0, nla_data(nla), skb->len); - header = (struct odp_msg*)__skb_push(skb, sizeof *header); - header->type = queue_no; - header->length = skb->len; - header->port = port_no; - header->reserved = 0; - header->arg = arg; - skb_queue_tail(queue, skb); + upcall->len = user_skb->len; + skb_queue_tail(&dp->queues[upcall_info->type], user_skb); + kfree_skb(skb); skb = nskb; } while (skb); return 0; @@ -789,60 +592,64 @@ err_kfree_skbs: return err; } -int -dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, - u32 arg) +int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info) { struct dp_stats_percpu *stats; struct sk_buff_head *queue; int err; WARN_ON_ONCE(skb_shared(skb)); - BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR); - queue = &dp->queues[queue_no]; + BUG_ON(upcall_info->type >= DP_N_QUEUES); + + queue = &dp->queues[upcall_info->type]; err = -ENOBUFS; if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) goto err_kfree_skb; forward_ip_summed(skb); + err = vswitch_skb_checksum_setup(skb); + if (err) + goto err_kfree_skb; + /* Break apart GSO packets into their component pieces. Otherwise * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ if (skb_is_gso(skb)) { - struct sk_buff *nskb = skb_gso_segment(skb, 0); - if (nskb) { - kfree_skb(skb); - skb = nskb; - if (unlikely(IS_ERR(skb))) { - err = PTR_ERR(skb); - goto err; - } - } else { - /* XXX This case might not be possible. It's hard to - * tell from the skb_gso_segment() code and comment. */ + struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); + + kfree_skb(skb); + skb = nskb; + if (IS_ERR(skb)) { + err = PTR_ERR(skb); + goto err; } } - err = queue_control_packets(skb, queue, queue_no, arg); + err = queue_control_packets(dp, skb, upcall_info); wake_up_interruptible(&dp->waitqueue); return err; err_kfree_skb: kfree_skb(skb); err: - stats = percpu_ptr(dp->stats_percpu, get_cpu()); + local_bh_disable(); + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); stats->n_lost++; - put_cpu(); + write_seqcount_end(&stats->seqlock); + + local_bh_enable(); return err; } static int flush_flows(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table = get_table_protected(dp); struct tbl *new_table; - new_table = tbl_create(0); + new_table = tbl_create(TBL_MIN_BUCKETS); if (!new_table) return -ENOMEM; @@ -853,46 +660,76 @@ static int flush_flows(struct datapath *dp) return 0; } -static int validate_actions(const struct sw_flow_actions *actions) -{ - unsigned int i; +static int validate_actions(const struct nlattr *actions, u32 actions_len) +{ + 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; - for (i = 0; i < actions->n_actions; i++) { - const union odp_action *a = &actions->actions[i]; - switch (a->type) { - case ODPAT_OUTPUT: - if (a->output.port >= DP_MAX_PORTS) - return -EINVAL; - break; + switch (type) { + case ODPAT_UNSPEC: + return -EINVAL; - case ODPAT_OUTPUT_GROUP: - if (a->output_group.group >= DP_MAX_GROUPS) - 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_SET_VLAN_VID: - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) + case ODPAT_OUTPUT: + if (nla_get_u32(a) >= DP_MAX_PORTS) return -EINVAL; break; - case ODPAT_SET_VLAN_PCP: - if (a->vlan_pcp.vlan_pcp - & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) + case ODPAT_SET_DL_TCI: + if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) return -EINVAL; break; case ODPAT_SET_NW_TOS: - if (a->nw_tos.nw_tos & INET_ECN_MASK) + if (nla_get_u8(a) & INET_ECN_MASK) return -EINVAL; break; default: - if (a->type >= ODPAT_N_ACTIONS) - return -EOPNOTSUPP; - break; + return -EOPNOTSUPP; } } + if (rem > 0) + return -EINVAL; + return 0; } @@ -901,16 +738,17 @@ static struct sw_flow_actions *get_actions(const struct odp_flow *flow) struct sw_flow_actions *actions; int error; - actions = flow_actions_alloc(flow->n_actions); + actions = flow_actions_alloc(flow->actions_len); error = PTR_ERR(actions); if (IS_ERR(actions)) goto error; error = -EFAULT; - if (copy_from_user(actions->actions, flow->actions, - flow->n_actions * sizeof(union odp_action))) + if (copy_from_user(actions->actions, + (struct nlattr __user __force *)flow->actions, + flow->actions_len)) goto error_free_actions; - error = validate_actions(actions); + error = validate_actions(actions->actions, actions->actions_len); if (error) goto error_free_actions; @@ -924,32 +762,39 @@ error: static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) { - if (flow->used.tv_sec) { - stats->used_sec = flow->used.tv_sec; - stats->used_nsec = flow->used.tv_nsec; + if (flow->used) { + struct timespec offset_ts, used, now_mono; + + 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); + + stats->used_sec = used.tv_sec; + stats->used_nsec = used.tv_nsec; } else { stats->used_sec = 0; stats->used_nsec = 0; } + stats->n_packets = flow->packet_count; stats->n_bytes = flow->byte_count; - stats->ip_tos = flow->ip_tos; + stats->reserved = 0; stats->tcp_flags = flow->tcp_flags; stats->error = 0; } static void clear_stats(struct sw_flow *flow) { - flow->used.tv_sec = flow->used.tv_nsec = 0; + flow->used = 0; flow->tcp_flags = 0; - flow->ip_tos = 0; flow->packet_count = 0; flow->byte_count = 0; } static int expand_table(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table = get_table_protected(dp); struct tbl *new_table; new_table = tbl_expand(old_table); @@ -966,18 +811,23 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, struct odp_flow_stats *stats) { struct tbl_node *flow_node; + struct sw_flow_key key; struct sw_flow *flow; struct tbl *table; + struct sw_flow_actions *acts = NULL; int error; + u32 hash; - memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved); + error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key, + uf->flow.key_len); + if (error) + return error; - table = rcu_dereference(dp->table); - flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); + hash = flow_hash(&key); + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, hash, flow_cmp); if (!flow_node) { /* No such flow. */ - struct sw_flow_actions *acts; - error = -ENOENT; if (!(uf->flags & ODPPF_CREATE)) goto error; @@ -987,16 +837,16 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, error = expand_table(dp); if (error) goto error; - table = rcu_dereference(dp->table); + table = get_table_protected(dp); } /* Allocate flow. */ - error = -ENOMEM; - flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); - if (flow == NULL) + flow = flow_alloc(); + if (IS_ERR(flow)) { + error = PTR_ERR(flow); goto error; - flow->key = uf->flow.key; - spin_lock_init(&flow->lock); + } + flow->key = key; clear_stats(flow); /* Obtain actions. */ @@ -1007,7 +857,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, rcu_assign_pointer(flow->sf_acts, acts); /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key)); + error = tbl_insert(table, &flow->tbl_node, hash); if (error) goto error_free_flow_acts; @@ -1028,10 +878,12 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, error = PTR_ERR(new_acts); if (IS_ERR(new_acts)) goto error; - old_acts = rcu_dereference(flow->sf_acts); - if (old_acts->n_actions != new_acts->n_actions || + + old_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); + if (old_acts->actions_len != new_acts->actions_len || memcmp(old_acts->actions, new_acts->actions, - sizeof(union odp_action) * old_acts->n_actions)) { + old_acts->actions_len)) { rcu_assign_pointer(flow->sf_acts, new_acts); flow_deferred_free_acts(old_acts); } else { @@ -1049,9 +901,10 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, return 0; error_free_flow_acts: - kfree(flow->sf_acts); + kfree(acts); error_free_flow: - kmem_cache_free(flow_cache, flow); + flow->sf_acts = NULL; + flow_put(flow); error: return error; } @@ -1076,14 +929,15 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) return 0; } -static int do_answer_query(struct sw_flow *flow, u32 query_flags, +static int do_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct odp_flow_stats __user *ustats, - union odp_action __user *actions, - u32 __user *n_actionsp) + struct nlattr __user *actions, + u32 __user *actions_lenp) { struct sw_flow_actions *sf_acts; struct odp_flow_stats stats; - u32 n_actions; + u32 actions_len; spin_lock_bh(&flow->lock); get_stats(flow, &stats); @@ -1093,42 +947,46 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, spin_unlock_bh(&flow->lock); if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(n_actions, n_actionsp)) + get_user(actions_len, actions_lenp)) return -EFAULT; - if (!n_actions) + if (!actions_len) return 0; - sf_acts = rcu_dereference(flow->sf_acts); - if (put_user(sf_acts->n_actions, n_actionsp) || + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); + if (put_user(sf_acts->actions_len, actions_lenp) || (actions && copy_to_user(actions, sf_acts->actions, - sizeof(union odp_action) * - min(sf_acts->n_actions, n_actions)))) + min(sf_acts->actions_len, actions_len)))) return -EFAULT; return 0; } -static int answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow __user *ufp) +static int answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct odp_flow __user *ufp) { - union odp_action *actions; + struct nlattr __user *actions; - if (get_user(actions, &ufp->actions)) + if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, - &ufp->stats, actions, &ufp->n_actions); + return do_answer_query(dp, flow, query_flags, + &ufp->stats, actions, &ufp->actions_len); } -static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key) +static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); struct tbl_node *flow_node; + struct sw_flow_key swkey; int error; - memset(key->reserved, 0, sizeof key->reserved); - flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp); + error = flow_copy_from_user(&swkey, key, key_len); + if (error) + return ERR_PTR(error); + + flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp); if (!flow_node) return ERR_PTR(-ENOENT); @@ -1149,83 +1007,48 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) struct odp_flow uf; int error; - if (copy_from_user(&uf, ufp, sizeof uf)) + if (copy_from_user(&uf, ufp, sizeof(uf))) return -EFAULT; - flow = do_del_flow(dp, &uf.key); + flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); if (IS_ERR(flow)) return PTR_ERR(flow); - error = answer_query(flow, 0, ufp); + error = answer_query(dp, 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); + struct tbl *table = get_table_protected(dp); u32 i; for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = &flowvec->flows[i]; + struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i]; + struct sw_flow_key key; struct odp_flow uf; struct tbl_node *flow_node; int error; - if (copy_from_user(&uf, ufp, sizeof uf)) + if (copy_from_user(&uf, ufp, sizeof(uf))) return -EFAULT; - memset(uf.key.reserved, 0, sizeof uf.key.reserved); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); + error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf.key, uf.key_len); + if (error) + return error; + + flow_node = tbl_lookup(table, &uf.key, flow_hash(&key), flow_cmp); if (!flow_node) error = put_user(ENOENT, &ufp->stats.error); else - error = answer_query(flow_cast(flow_node), uf.flags, ufp); + error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp); if (error) return -EFAULT; } return flowvec->n_flows; } -struct list_flows_cbdata { - struct odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; -}; - -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 *)) @@ -1235,7 +1058,7 @@ static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, int retval; uflowvec = (struct odp_flowvec __user *)argp; - if (copy_from_user(&flowvec, uflowvec, sizeof flowvec)) + if (copy_from_user(&flowvec, uflowvec, sizeof(flowvec))) return -EFAULT; if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) @@ -1247,29 +1070,79 @@ static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, : put_user(retval, &uflowvec->n_flows)); } +static struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state) +{ + struct tbl *table = get_table_protected(dp); + struct tbl_node *tbl_node; + u32 bucket, obj; + + if (get_user(bucket, &state[0]) || get_user(obj, &state[1])) + return ERR_PTR(-EFAULT); + + tbl_node = tbl_next(table, &bucket, &obj); + + if (put_user(bucket, &state[0]) || put_user(obj, &state[1])) + return ERR_PTR(-EFAULT); + + return tbl_node ? flow_cast(tbl_node) : NULL; +} + +static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp) +{ + struct odp_flow __user *uflowp; + struct nlattr __user *ukey; + struct sw_flow *flow; + u32 key_len; + + flow = do_dump_flow(dp, udumpp->state); + if (IS_ERR(flow)) + return PTR_ERR(flow); + + if (get_user(uflowp, (struct odp_flow __user *__user*)&udumpp->flow)) + return -EFAULT; + + if (!flow) + return put_user(ODPFF_EOF, &uflowp->flags); + + if (put_user(0, &uflowp->flags) || + get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) || + get_user(key_len, &uflowp->key_len)) + return -EFAULT; + + key_len = flow_copy_to_user(ukey, &flow->key, key_len); + if (key_len < 0) + return key_len; + if (put_user(key_len, &uflowp->key_len)) + return -EFAULT; + + return answer_query(dp, flow, 0, uflowp); +} + static int do_execute(struct datapath *dp, const struct odp_execute *execute) { - struct odp_flow_key key; + struct sw_flow_key key; struct sk_buff *skb; struct sw_flow_actions *actions; struct ethhdr *eth; + bool is_frag; int err; err = -EINVAL; if (execute->length < ETH_HLEN || execute->length > 65535) goto error; - err = -ENOMEM; - actions = flow_actions_alloc(execute->n_actions); - if (!actions) + actions = flow_actions_alloc(execute->actions_len); + if (IS_ERR(actions)) { + err = PTR_ERR(actions); goto error; + } err = -EFAULT; - if (copy_from_user(actions->actions, execute->actions, - execute->n_actions * sizeof *execute->actions)) + if (copy_from_user(actions->actions, + (struct nlattr __user __force *)execute->actions, execute->actions_len)) goto error_free_actions; - err = validate_actions(actions); + err = validate_actions(actions->actions, execute->actions_len); if (err) goto error_free_actions; @@ -1278,13 +1151,9 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) if (!skb) goto error_free_actions; - if (execute->in_port < DP_MAX_PORTS) - OVS_CB(skb)->dp_port = dp->ports[execute->in_port]; - else - OVS_CB(skb)->dp_port = NULL; - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), execute->data, + if (copy_from_user(skb_put(skb, execute->length), + (const void __user __force *)execute->data, execute->length)) goto error_free_skb; @@ -1299,9 +1168,14 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) else skb->protocol = htons(ETH_P_802_2); - flow_extract(skb, execute->in_port, &key); - err = execute_actions(dp, skb, &key, actions->actions, - actions->n_actions, GFP_KERNEL); + err = flow_extract(skb, -1, &key, &is_frag); + if (err) + goto error_free_skb; + + rcu_read_lock(); + err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); + rcu_read_unlock(); + kfree(actions); return err; @@ -1317,7 +1191,7 @@ static int execute_packet(struct datapath *dp, const struct odp_execute __user * { struct odp_execute execute; - if (copy_from_user(&execute, executep, sizeof execute)) + if (copy_from_user(&execute, executep, sizeof(execute))) return -EFAULT; return do_execute(dp, &execute); @@ -1325,34 +1199,38 @@ static int execute_packet(struct datapath *dp, const struct odp_execute __user * static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) { - struct tbl *table = rcu_dereference(dp->table); struct odp_stats stats; int i; - stats.n_flows = tbl_count(table); - stats.cur_capacity = tbl_n_buckets(table); - stats.max_capacity = TBL_MAX_BUCKETS; stats.n_ports = dp->n_ports; stats.max_ports = DP_MAX_PORTS; - stats.max_groups = DP_MAX_GROUPS; stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; for_each_possible_cpu(i) { - const struct dp_stats_percpu *s; - s = percpu_ptr(dp->stats_percpu, i); - stats.n_frags += s->n_frags; - stats.n_hit += s->n_hit; - stats.n_missed += s->n_missed; - stats.n_lost += s->n_lost; + const struct dp_stats_percpu *percpu_stats; + struct dp_stats_percpu local_stats; + unsigned seqcount; + + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); + + stats.n_frags += local_stats.n_frags; + stats.n_hit += local_stats.n_hit; + stats.n_missed += local_stats.n_missed; + stats.n_lost += local_stats.n_lost; } stats.max_miss_queue = DP_MAX_QUEUE_LEN; stats.max_action_queue = DP_MAX_QUEUE_LEN; - return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0; + return copy_to_user(statsp, &stats, sizeof(stats)) ? -EFAULT : 0; } /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ int dp_min_mtu(const struct datapath *dp) { - struct dp_port *p; + struct vport *p; int mtu = 0; ASSERT_RTNL(); @@ -1362,10 +1240,10 @@ int dp_min_mtu(const struct datapath *dp) /* Skip any internal ports, since that's what we're trying to * set. */ - if (is_internal_vport(p->vport)) + if (is_internal_vport(p)) continue; - dev_mtu = vport_get_mtu(p->vport); + dev_mtu = vport_get_mtu(p); if (!mtu || dev_mtu < mtu) mtu = dev_mtu; } @@ -1377,7 +1255,7 @@ int dp_min_mtu(const struct datapath *dp) * be called with RTNL lock. */ void set_internal_devs_mtu(const struct datapath *dp) { - struct dp_port *p; + struct vport *p; int mtu; ASSERT_RTNL(); @@ -1385,199 +1263,413 @@ void set_internal_devs_mtu(const struct datapath *dp) mtu = dp_min_mtu(dp); list_for_each_entry_rcu (p, &dp->port_list, node) { - if (is_internal_vport(p->vport)) - vport_set_mtu(p->vport, mtu); + if (is_internal_vport(p)) + vport_set_mtu(p, mtu); } } -static int -put_port(const struct dp_port *p, struct odp_port __user *uop) +static int get_listen_mask(const struct file *f) +{ + return (long)f->private_data; +} + +static void set_listen_mask(struct file *f, int listen_mask) +{ + f->private_data = (void*)(long)listen_mask; +} + +static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = { + [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, + [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) }, + [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN }, + [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 }, + [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, +}; + +static int copy_vport_to_user(void __user *dst, struct vport *vport, uint32_t total_len) { - struct odp_port op; + struct odp_vport *odp_vport; + struct sk_buff *skb; + struct nlattr *nla; + int ifindex, iflink; + int err; - memset(&op, 0, sizeof op); + skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; rcu_read_lock(); - strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname); - rcu_read_unlock(); + odp_vport = (struct odp_vport*)__skb_put(skb, sizeof(struct odp_vport)); + odp_vport->dp_idx = vport->dp->dp_idx; + odp_vport->total_len = total_len; + + NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no); + NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport)); + NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport)); + + nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64)); + if (!nla) + goto nla_put_failure; + if (vport_get_stats(vport, nla_data(nla))) + __skb_trim(skb, skb->len - nla->nla_len); - op.port = p->port_no; - op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0; + NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); - return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; + NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport)); + + err = vport_get_options(vport, skb); + + ifindex = vport_get_ifindex(vport); + if (ifindex > 0) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex); + + iflink = vport_get_iflink(vport); + if (iflink > 0) + NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink); + + err = -EMSGSIZE; + if (skb->len > total_len) + goto exit_unlock; + + odp_vport->len = skb->len; + err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; + goto exit_unlock; + +nla_put_failure: + err = -EMSGSIZE; +exit_unlock: + rcu_read_unlock(); + kfree_skb(skb); +exit: + return err; } -static int -query_port(struct datapath *dp, struct odp_port __user *uport) +static struct sk_buff *copy_vport_from_user(struct odp_vport __user *uodp_vport, + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { - struct odp_port port; + struct odp_vport *odp_vport; + struct sk_buff *skb; + u32 len; + int err; - if (copy_from_user(&port, uport, sizeof port)) - return -EFAULT; + if (get_user(len, &uodp_vport->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_vport)) + return ERR_PTR(-EINVAL); + + skb = alloc_skb(len, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); - if (port.devname[0]) { - struct vport *vport; - struct dp_port *dp_port; - int err = 0; + err = -EFAULT; + if (copy_from_user(__skb_put(skb, len), uodp_vport, len)) + goto error_free_skb; + + odp_vport = (struct odp_vport *)skb->data; + err = -EINVAL; + if (odp_vport->len != len) + goto error_free_skb; - port.devname[IFNAMSIZ - 1] = '\0'; + err = nla_parse(a, ODP_VPORT_ATTR_MAX, (struct nlattr *)(skb->data + sizeof(struct odp_vport)), + skb->len - sizeof(struct odp_vport), vport_policy); + if (err) + goto error_free_skb; + + err = VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1); + if (err) + goto error_free_skb; + + return skb; + +error_free_skb: + kfree_skb(skb); + return ERR_PTR(err); +} - vport_lock(); - rcu_read_lock(); - vport = vport_locate(port.devname); +/* Called without any locks (or with RTNL lock). + * Returns holding vport->dp->mutex. + */ +static struct vport *lookup_vport(struct odp_vport *odp_vport, + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + struct datapath *dp; + struct vport *vport; + + if (a[ODP_VPORT_ATTR_NAME]) { + int dp_idx, port_no; + + retry: + vport_lock(); + vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME])); if (!vport) { - err = -ENODEV; - goto error_unlock; + vport_unlock(); + return ERR_PTR(-ENODEV); } + dp_idx = vport->dp->dp_idx; + port_no = vport->port_no; + vport_unlock(); + + dp = get_dp_locked(dp_idx); + if (!dp) + goto retry; - dp_port = vport_get_dp_port(vport); - if (!dp_port || dp_port->dp != dp) { - err = -ENOENT; - goto error_unlock; + vport = get_vport_protected(dp, port_no); + if (!vport || + strcmp(vport_get_name(vport), nla_data(a[ODP_VPORT_ATTR_NAME]))) { + mutex_unlock(&dp->mutex); + goto retry; } - port.port = dp_port->port_no; + return vport; + } else if (a[ODP_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); -error_unlock: - rcu_read_unlock(); - vport_unlock(); + if (port_no >= DP_MAX_PORTS) + return ERR_PTR(-EINVAL); - if (err) - return err; - } else { - if (port.port >= DP_MAX_PORTS) - return -EINVAL; - if (!dp->ports[port.port]) - return -ENOENT; - } + dp = get_dp_locked(odp_vport->dp_idx); + if (!dp) + return ERR_PTR(-ENODEV); - return put_port(dp->ports[port.port], uport); + vport = get_vport_protected(dp, port_no); + if (!vport) { + mutex_unlock(&dp->mutex); + return ERR_PTR(-ENOENT); + } + return vport; + } else + return ERR_PTR(-EINVAL); } -static int -do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports) +static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { - int idx = 0; - if (n_ports) { - struct dp_port *p; - - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (put_port(p, &uports[idx])) - return -EFAULT; - if (idx++ >= n_ports) - break; - } - } - return idx; + int err = 0; + if (a[ODP_VPORT_ATTR_STATS]) + err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS])); + if (!err && a[ODP_VPORT_ATTR_ADDRESS]) + err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS])); + if (!err && a[ODP_VPORT_ATTR_MTU]) + err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU])); + return err; } -static int -list_ports(struct datapath *dp, struct odp_portvec __user *upv) +static int attach_vport(struct odp_vport __user *uodp_vport) { - struct odp_portvec pv; - int retval; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + struct vport_parms parms; + struct vport *vport; + struct sk_buff *skb; + struct datapath *dp; + u32 port_no; + int err; - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_vport = (struct odp_vport *)skb->data; - retval = do_list_ports(dp, pv.ports, pv.n_ports); - if (retval < 0) - return retval; + err = -EINVAL; + if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE]) + goto exit_kfree_skb; - return put_user(retval, &upv->n_ports); -} + rtnl_lock(); -/* RCU callback for freeing a dp_port_group */ -static void free_port_group(struct rcu_head *rcu) -{ - struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu); - kfree(g); -} + dp = get_dp_locked(odp_vport->dp_idx); + err = -ENODEV; + if (!dp) + goto exit_unlock_rtnl; -static int -do_set_port_group(struct datapath *dp, u16 __user *ports, int n_ports, int group) -{ - struct dp_port_group *new_group, *old_group; - int error; + if (a[ODP_VPORT_ATTR_PORT_NO]) { + port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - error = -EINVAL; - if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS) - goto error; + err = -EFBIG; + if (port_no >= DP_MAX_PORTS) + goto exit_unlock_dp; - error = -ENOMEM; - new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL); - if (!new_group) - goto error; + vport = get_vport_protected(dp, port_no); + err = -EBUSY; + if (vport) + goto exit_unlock_dp; + } else { + for (port_no = 1; ; port_no++) { + if (port_no >= DP_MAX_PORTS) { + err = -EFBIG; + goto exit_unlock_dp; + } + vport = get_vport_protected(dp, port_no); + if (!vport) + break; + } + } - new_group->n_ports = n_ports; - error = -EFAULT; - if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports)) - goto error_free; + parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]); + parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]); + parms.options = a[ODP_VPORT_ATTR_OPTIONS]; + parms.dp = dp; + parms.port_no = port_no; - old_group = rcu_dereference(dp->groups[group]); - rcu_assign_pointer(dp->groups[group], new_group); - if (old_group) - call_rcu(&old_group->rcu, free_port_group); - return 0; + vport = new_vport(&parms); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock_dp; -error_free: - kfree(new_group); -error: - return error; -} + set_internal_devs_mtu(dp); + dp_sysfs_add_if(vport); -static int -set_port_group(struct datapath *dp, const struct odp_port_group __user *upg) -{ - struct odp_port_group pg; + err = change_vport(vport, a); + if (err) { + dp_detach_port(vport); + goto exit_unlock_dp; + } - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); - return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group); +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_unlock_rtnl: + rtnl_unlock(); +exit_kfree_skb: + kfree_skb(skb); +exit: + return err; } -static int -do_get_port_group(struct datapath *dp, - u16 __user *ports, int n_ports, int group, - u16 __user *n_portsp) +static int set_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) { - struct dp_port_group *g; - u16 n_copy; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct vport *vport; + struct sk_buff *skb; + int err; - if (group >= DP_MAX_GROUPS) - return -EINVAL; + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - g = dp->groups[group]; - n_copy = g ? min_t(int, g->n_ports, n_ports) : 0; - if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16))) - return -EFAULT; + rtnl_lock(); + vport = lookup_vport((struct odp_vport *)skb->data, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; - if (put_user(g ? g->n_ports : 0, n_portsp)) - return -EFAULT; + err = 0; + if (a[ODP_VPORT_ATTR_OPTIONS]) + err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]); + if (!err) + err = change_vport(vport, a); - return 0; + mutex_unlock(&vport->dp->mutex); +exit_free: + kfree_skb(skb); + rtnl_unlock(); +exit: + return err; } -static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg) +static int del_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) { - struct odp_port_group pg; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct datapath *dp; + struct vport *vport; + struct sk_buff *skb; + int err; - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &pg.n_ports); + rtnl_lock(); + vport = lookup_vport((struct odp_vport *)skb->data, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; + dp = vport->dp; + + err = -EINVAL; + if (vport->port_no == ODPP_LOCAL) + goto exit_free; + + err = dp_detach_port(vport); + mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); + rtnl_unlock(); +exit: + return err; } -static int get_listen_mask(const struct file *f) +static int get_vport(struct odp_vport __user *uodp_vport) { - return (long)f->private_data; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + struct vport *vport; + struct sk_buff *skb; + int err; + + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_vport = (struct odp_vport *)skb->data; + + vport = lookup_vport(odp_vport, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; + + err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); + mutex_unlock(&vport->dp->mutex); +exit_free: + kfree_skb(skb); +exit: + return err; } -static void set_listen_mask(struct file *f, int listen_mask) +static int dump_vport(struct odp_vport __user *uodp_vport) { - f->private_data = (void*)(long)listen_mask; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + struct sk_buff *skb; + struct datapath *dp; + u32 port_no; + int err; + + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_vport = (struct odp_vport *)skb->data; + + dp = get_dp_locked(odp_vport->dp_idx); + err = -ENODEV; + if (!dp) + goto exit_free; + + port_no = 0; + if (a[ODP_VPORT_ATTR_PORT_NO]) + port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + for (; port_no < DP_MAX_PORTS; port_no++) { + struct vport *vport = get_vport_protected(dp, port_no); + if (vport) { + err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); + goto exit_unlock_dp; + } + } + err = -ENODEV; + +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); +exit: + return err; } static long openvswitch_ioctl(struct file *f, unsigned int cmd, @@ -1585,7 +1677,7 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, { int dp_idx = iminor(f->f_dentry->d_inode); struct datapath *dp; - int drop_frags, listeners, port_no; + int drop_frags, listeners; unsigned int sflow_probability; int err; @@ -1599,46 +1691,24 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, err = destroy_dp(dp_idx); goto exit; - case ODP_PORT_ATTACH: - err = attach_port(dp_idx, (struct odp_port __user *)argp); + case ODP_VPORT_NEW: + err = attach_vport((struct odp_vport __user *)argp); goto exit; - case ODP_PORT_DETACH: - err = get_user(port_no, (int __user *)argp); - if (!err) - err = detach_port(dp_idx, port_no); - goto exit; - - case ODP_VPORT_ADD: - err = vport_add((struct odp_vport_add __user *)argp); - goto exit; - - case ODP_VPORT_MOD: - err = vport_mod((struct odp_vport_mod __user *)argp); + case ODP_VPORT_GET: + err = get_vport((struct odp_vport __user *)argp); goto exit; case ODP_VPORT_DEL: - err = vport_del((char __user *)argp); - goto exit; - - case ODP_VPORT_STATS_GET: - err = vport_stats_get((struct odp_vport_stats_req __user *)argp); - goto exit; - - case ODP_VPORT_ETHER_GET: - err = vport_ether_get((struct odp_vport_ether __user *)argp); - goto exit; - - case ODP_VPORT_ETHER_SET: - err = vport_ether_set((struct odp_vport_ether __user *)argp); + err = del_vport(cmd, (struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_MTU_GET: - err = vport_mtu_get((struct odp_vport_mtu __user *)argp); + case ODP_VPORT_SET: + err = set_vport(cmd, (struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_MTU_SET: - err = vport_mtu_set((struct odp_vport_mtu __user *)argp); + case ODP_VPORT_DUMP: + err = dump_vport((struct odp_vport __user *)argp); goto exit; } @@ -1692,22 +1762,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, dp->sflow_probability = sflow_probability; break; - case ODP_PORT_QUERY: - err = query_port(dp, (struct odp_port __user *)argp); - break; - - case ODP_PORT_LIST: - err = list_ports(dp, (struct odp_portvec __user *)argp); - break; - - case ODP_PORT_GROUP_SET: - err = set_port_group(dp, (struct odp_port_group __user *)argp); - break; - - case ODP_PORT_GROUP_GET: - err = get_port_group(dp, (struct odp_port_group __user *)argp); - break; - case ODP_FLOW_FLUSH: err = flush_flows(dp); break; @@ -1724,8 +1778,8 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, err = do_flowvec_ioctl(dp, argp, do_query_flows); break; - case ODP_FLOW_LIST: - err = do_flowvec_ioctl(dp, argp, do_list_flows); + case ODP_FLOW_DUMP: + err = dump_flow(dp, (struct odp_flow_dump __user *)argp); break; case ODP_EXECUTE: @@ -1752,55 +1806,21 @@ static int dp_has_packet_of_interest(struct datapath *dp, int listeners) } #ifdef CONFIG_COMPAT -static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv) -{ - struct compat_odp_portvec pv; - 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; - - return put_user(retval, &upv->n_ports); -} - -static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg) -{ - struct compat_odp_port_group pg; - - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; - - return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group); -} - -static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg) -{ - struct compat_odp_port_group pg; - - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; - - return do_get_port_group(dp, compat_ptr(pg.ports), pg.n_ports, - pg.group, &pg.n_ports); -} - static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) { - compat_uptr_t actions; + compat_uptr_t key, actions; if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) || __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) || - __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) || + __get_user(key, &compat->key) || + __get_user(flow->key_len, &compat->key_len) || __get_user(actions, &compat->actions) || - __get_user(flow->n_actions, &compat->n_actions) || + __get_user(flow->actions_len, &compat->actions_len) || __get_user(flow->flags, &compat->flags)) return -EFAULT; - flow->actions = compat_ptr(actions); + flow->key = (struct nlattr __force *)compat_ptr(key); + flow->actions = (struct nlattr __force *)compat_ptr(actions); return 0; } @@ -1825,7 +1845,8 @@ static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __use return 0; } -static int compat_answer_query(struct sw_flow *flow, u32 query_flags, +static int compat_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct compat_odp_flow __user *ufp) { compat_uptr_t actions; @@ -1833,8 +1854,8 @@ static int compat_answer_query(struct sw_flow *flow, u32 query_flags, if (get_user(actions, &ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, &ufp->stats, - compat_ptr(actions), &ufp->n_actions); + return do_answer_query(dp, flow, query_flags, &ufp->stats, + compat_ptr(actions), &ufp->actions_len); } static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp) @@ -1846,83 +1867,84 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u if (compat_get_flow(&uf, ufp)) return -EFAULT; - flow = do_del_flow(dp, &uf.key); + flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); if (IS_ERR(flow)) return PTR_ERR(flow); - error = compat_answer_query(flow, 0, ufp); + error = compat_answer_query(dp, flow, 0, ufp); flow_deferred_free(flow); return error; } -static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) +static int compat_query_flows(struct datapath *dp, + struct compat_odp_flow __user *flows, + u32 n_flows) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); u32 i; for (i = 0; i < n_flows; i++) { struct compat_odp_flow __user *ufp = &flows[i]; struct odp_flow uf; struct tbl_node *flow_node; + struct sw_flow_key key; int error; if (compat_get_flow(&uf, ufp)) return -EFAULT; - memset(uf.key.reserved, 0, sizeof uf.key.reserved); - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); + error = flow_copy_from_user(&key, (const struct nlattr __force __user *) uf.key, uf.key_len); + if (error) + return error; + + flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp); if (!flow_node) error = put_user(ENOENT, &ufp->stats.error); else - error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp); + error = compat_answer_query(dp, flow_cast(flow_node), + uf.flags, ufp); if (error) return -EFAULT; } return n_flows; } -struct compat_list_flows_cbdata { - struct compat_odp_flow __user *uflows; - u32 n_flows; - u32 listed_flows; -}; - -static int compat_list_flow(struct tbl_node *node, void *cbdata_) +static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp) { - 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 compat_odp_flow __user *uflowp; + compat_uptr_t compat_ufp; + struct sw_flow *flow; + compat_uptr_t ukey; + u32 key_len; + + flow = do_dump_flow(dp, udumpp->state); + if (IS_ERR(flow)) + return PTR_ERR(flow); - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) + if (get_user(compat_ufp, &udumpp->flow)) return -EFAULT; - error = compat_answer_query(flow, 0, ufp); - if (error) - return error; + uflowp = compat_ptr(compat_ufp); - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; -} + if (!flow) + return put_user(ODPFF_EOF, &uflowp->flags); -static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) -{ - struct compat_list_flows_cbdata cbdata; - int error; + if (put_user(0, &uflowp->flags) || + get_user(ukey, &uflowp->key) || + get_user(key_len, &uflowp->key_len)) + return -EFAULT; - if (!n_flows) - return 0; + key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len); + if (key_len < 0) + return key_len; + if (put_user(key_len, &uflowp->key_len)) + return -EFAULT; - cbdata.uflows = flows; - cbdata.n_flows = n_flows; - cbdata.listed_flows = 0; - error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata); - return error ? error : cbdata.listed_flows; + return compat_answer_query(dp, flow, 0, uflowp); } static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, int (*function)(struct datapath *, - struct compat_odp_flow *, + struct compat_odp_flow __user *, u32 n_flows)) { struct compat_odp_flowvec __user *uflowvec; @@ -1931,8 +1953,8 @@ static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, int retval; uflowvec = compat_ptr(argp); - if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) || - copy_from_user(&flowvec, uflowvec, sizeof flowvec)) + if (!access_ok(VERIFY_WRITE, uflowvec, sizeof(*uflowvec)) || + copy_from_user(&flowvec, uflowvec, sizeof(flowvec))) return -EFAULT; if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow)) @@ -1956,15 +1978,14 @@ static int compat_execute(struct datapath *dp, const struct compat_odp_execute _ compat_uptr_t data; if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) || - __get_user(execute.in_port, &uexecute->in_port) || __get_user(actions, &uexecute->actions) || - __get_user(execute.n_actions, &uexecute->n_actions) || + __get_user(execute.actions_len, &uexecute->actions_len) || __get_user(data, &uexecute->data) || __get_user(execute.length, &uexecute->length)) return -EFAULT; - execute.actions = compat_ptr(actions); - execute.data = compat_ptr(data); + execute.actions = (struct nlattr __force *)compat_ptr(actions); + execute.data = (const void __force *)compat_ptr(data); return do_execute(dp, &execute); } @@ -1982,14 +2003,11 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned return openvswitch_ioctl(f, cmd, argp); case ODP_DP_CREATE: - case ODP_PORT_ATTACH: - case ODP_PORT_DETACH: + case ODP_VPORT_NEW: case ODP_VPORT_DEL: - case ODP_VPORT_MTU_SET: - case ODP_VPORT_MTU_GET: - case ODP_VPORT_ETHER_SET: - case ODP_VPORT_ETHER_GET: - case ODP_VPORT_STATS_GET: + case ODP_VPORT_GET: + case ODP_VPORT_SET: + case ODP_VPORT_DUMP: case ODP_DP_STATS: case ODP_GET_DROP_FRAGS: case ODP_SET_DROP_FRAGS: @@ -1997,15 +2015,8 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned case ODP_GET_LISTEN_MASK: case ODP_SET_SFLOW_PROBABILITY: case ODP_GET_SFLOW_PROBABILITY: - case ODP_PORT_QUERY: /* Ioctls that just need their pointer argument extended. */ return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); - - case ODP_VPORT_ADD32: - return compat_vport_add(compat_ptr(argp)); - - case ODP_VPORT_MOD32: - return compat_vport_mod(compat_ptr(argp)); } dp = get_dp_locked(dp_idx); @@ -2014,18 +2025,6 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned goto exit; switch (cmd) { - case ODP_PORT_LIST32: - err = compat_list_ports(dp, compat_ptr(argp)); - break; - - case ODP_PORT_GROUP_SET32: - err = compat_set_port_group(dp, compat_ptr(argp)); - break; - - case ODP_PORT_GROUP_GET32: - err = compat_get_port_group(dp, compat_ptr(argp)); - break; - case ODP_FLOW_PUT32: err = compat_put_flow(dp, compat_ptr(argp)); break; @@ -2038,8 +2037,8 @@ static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned err = compat_flowvec_ioctl(dp, argp, compat_query_flows); break; - case ODP_FLOW_LIST32: - err = compat_flowvec_ioctl(dp, argp, compat_list_flows); + case ODP_FLOW_DUMP32: + err = compat_dump_flow(dp, compat_ptr(argp)); break; case ODP_EXECUTE32: @@ -2056,16 +2055,14 @@ exit: } #endif -ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, - loff_t *ppos) +static ssize_t openvswitch_read(struct file *f, char __user *buf, + size_t nbytes, loff_t *ppos) { - /* XXX is there sufficient synchronization here? */ int listeners = get_listen_mask(f); int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp = get_dp(dp_idx); + struct datapath *dp = get_dp_locked(dp_idx); struct sk_buff *skb; - struct iovec __user iov; - size_t copy_bytes; + struct iovec iov; int retval; if (!dp) @@ -2100,23 +2097,26 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, } } success: - copy_bytes = min_t(size_t, skb->len, nbytes); + mutex_unlock(&dp->mutex); + iov.iov_base = buf; - iov.iov_len = copy_bytes; + iov.iov_len = min_t(size_t, skb->len, nbytes); retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); if (!retval) - retval = copy_bytes; + retval = skb->len; + kfree_skb(skb); + return retval; error: + mutex_unlock(&dp->mutex); return retval; } static unsigned int openvswitch_poll(struct file *file, poll_table *wait) { - /* XXX is there sufficient synchronization here? */ int dp_idx = iminor(file->f_dentry->d_inode); - struct datapath *dp = get_dp(dp_idx); + struct datapath *dp = get_dp_locked(dp_idx); unsigned int mask; if (dp) { @@ -2124,21 +2124,21 @@ static unsigned int openvswitch_poll(struct file *file, poll_table *wait) 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 file_operations openvswitch_fops = { - /* XXX .aio_read = openvswitch_aio_read, */ +static struct file_operations openvswitch_fops = { + .owner = THIS_MODULE, .read = openvswitch_read, .poll = openvswitch_poll, .unlocked_ioctl = openvswitch_ioctl, #ifdef CONFIG_COMPAT .compat_ioctl = openvswitch_compat_ioctl, #endif - /* XXX .fasync = openvswitch_fasync, */ }; static int major;