X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=be16044d5a44a60ab6a751273187827aedbee8cc;hb=ea7bd5973fb426f3a6ea66b0953c6000980e77ec;hp=bf8043b6054fffa428fb827321325f16cdfa6f87;hpb=0515ceb3e82ecc6160a141017c20ca26ec1a7a5f;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index bf8043b60..3d32cd47a 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007, 2008, 2009 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 @@ -8,11 +8,12 @@ /* Functions for managing the dp interface/device. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include #include #include -#include #include #include #include @@ -21,7 +22,6 @@ #include #include #include -#include #include #include #include @@ -29,61 +29,59 @@ #include #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 "dp_dev.h" #include "flow.h" - -#include "compat.h" - +#include "loop_counter.h" +#include "odp-compat.h" +#include "table.h" +#include "vport-internal_dev.h" int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); EXPORT_SYMBOL(dp_ioctl_hook); /* Datapaths. Protected on the read side by rcu_read_lock, on the write side - * by dp_mutex. dp_mutex is almost completely redundant with genl_mutex - * maintained by the Generic Netlink code, but the timeout path needs mutual - * exclusion too. + * by dp_mutex. * * 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 net_bridge_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_nbp(struct datapath *, struct net_device *, 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); -struct datapath *get_dp_locked(int dp_idx) +static struct datapath *get_dp_locked(int dp_idx) { struct datapath *dp; @@ -95,6 +93,24 @@ 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(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); +} + static inline size_t br_nlmsg_size(void) { return NLMSG_ALIGN(sizeof(struct ifinfomsg)) @@ -107,14 +123,21 @@ static inline size_t br_nlmsg_size(void) } static int dp_fill_ifinfo(struct sk_buff *skb, - const struct net_bridge_port *port, + const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; - const struct net_device *dev = port->dev; + struct datapath *dp = port->dp; + int ifindex = vport_get_ifindex(port); + int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; struct nlmsghdr *nlh; + if (ifindex < 0) + return ifindex; + + if (iflink < 0) + return iflink; + nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags); if (nlh == NULL) return -EMSGSIZE; @@ -122,24 +145,26 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr = nlmsg_data(nlh); hdr->ifi_family = AF_BRIDGE; hdr->__ifi_pad = 0; - hdr->ifi_type = dev->type; - hdr->ifi_index = dev->ifindex; - hdr->ifi_flags = dev_get_flags(dev); + hdr->ifi_type = ARPHRD_ETHER; + hdr->ifi_index = ifindex; + hdr->ifi_flags = vport_get_flags(port); hdr->ifi_change = 0; - NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name); - NLA_PUT_U32(skb, IFLA_MASTER, dp->ports[ODPP_LOCAL]->dev->ifindex); - NLA_PUT_U32(skb, IFLA_MTU, dev->mtu); + 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, - netif_running(dev) ? dev->operstate : IF_OPER_DOWN); + vport_is_running(port) + ? vport_get_operstate(port) + : IF_OPER_DOWN); #endif - if (dev->addr_len) - NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr); + NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port)); - if (dev->ifindex != dev->iflink) - NLA_PUT_U32(skb, IFLA_LINK, dev->iflink); + if (ifindex != iflink) + NLA_PUT_U32(skb, IFLA_LINK,iflink); return nlmsg_end(skb, nlh); @@ -148,9 +173,8 @@ nla_put_failure: return -EMSGSIZE; } -static void dp_ifinfo_notify(int event, struct net_bridge_port *port) +static void dp_ifinfo_notify(int event, struct vport *port) { - struct net *net = dev_net(port->dev); struct sk_buff *skb; int err = -ENOBUFS; @@ -165,10 +189,11 @@ static void dp_ifinfo_notify(int event, struct net_bridge_port *port) kfree_skb(skb); goto errout; } - err = rtnl_notify(skb, net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL); + rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL); + return; errout: if (err < 0) - rtnl_set_sk_err(net, RTNLGRP_LINK, err); + rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err); } static void release_dp(struct kobject *kobj) @@ -177,25 +202,30 @@ static void release_dp(struct kobject *kobj) kfree(dp); } -struct kobj_type dp_ktype = { +static struct kobj_type dp_ktype = { .release = release_dp }; static int create_dp(int dp_idx, const char __user *devnamep) { - struct net_device *dp_dev; + struct vport_parms parms; char devname[IFNAMSIZ]; + struct vport *vport; struct datapath *dp; int err; int i; if (devnamep) { - err = -EFAULT; - if (strncpy_from_user(devname, devnamep, IFNAMSIZ - 1) < 0) + int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ); + if (retval < 0) { + err = -EFAULT; + goto err; + } else if (retval >= IFNAMSIZ) { + err = -ENAMETOOLONG; goto err; - devname[IFNAMSIZ - 1] = '\0'; + } } else { - snprintf(devname, sizeof devname, "of%d", dp_idx); + snprintf(devname, sizeof(devname), "of%d", dp_idx); } rtnl_lock(); @@ -212,61 +242,65 @@ 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]); init_waitqueue_head(&dp->waitqueue); /* Initialize kobject for bridge. This will be added as - * /sys/class/net//bridge later, if sysfs is enabled. */ - kobject_set_name(&dp->ifobj, SYSFS_BRIDGE_PORT_SUBDIR); /* "bridge" */ + * /sys/class/net//brif later, if sysfs is enabled. */ dp->ifobj.kset = NULL; - dp->ifobj.parent = NULL; kobject_init(&dp->ifobj, &dp_ktype); /* Allocate table. */ err = -ENOMEM; - rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE)); + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); if (!dp->table) goto err_free_dp; - /* Setup our datapath device */ - dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL); - err = PTR_ERR(dp_dev); - if (IS_ERR(dp_dev)) - goto err_destroy_table; + /* Set up our datapath device. */ + parms.name = devname; + parms.type = ODP_VPORT_TYPE_INTERNAL; + parms.options = NULL; + parms.dp = dp; + parms.port_no = ODPP_LOCAL; + vport = new_vport(&parms); + if (IS_ERR(vport)) { + err = PTR_ERR(vport); + if (err == -EBUSY) + err = -EEXIST; - err = new_nbp(dp, dp_dev, ODPP_LOCAL); - if (err) { - dp_dev_destroy(dp_dev); goto err_destroy_table; } 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(); -#ifdef SUPPORT_SYSFS - dp_sysfs_add_dp(dp); -#endif - return 0; err_destroy_local_port: - dp_del_port(dp->ports[ODPP_LOCAL]); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); err_destroy_table: - dp_table_destroy(dp->table, 0); + tbl_destroy(get_table_protected(dp), NULL); err_free_dp: + mutex_unlock(&dp->mutex); kfree(dp); err_put_module: module_put(THIS_MODULE); @@ -277,523 +311,425 @@ err: return err; } -static void do_destroy_dp(struct datapath *dp) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct net_bridge_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_del_port(p); - -#ifdef SUPPORT_SYSFS - dp_sysfs_del_dp(dp); -#endif - - rcu_assign_pointer(dps[dp->dp_idx], NULL); - - dp_del_port(dp->ports[ODPP_LOCAL]); - - dp_table_destroy(dp->table, 1); - 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_nbp(struct kobject *kobj) -{ - struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj); - kfree(p); -} - -struct kobj_type brport_ktype = { -#ifdef SUPPORT_SYSFS - .sysfs_ops = &brport_sysfs_ops, -#endif - .release = release_nbp -}; - -/* Called with RTNL lock and dp_mutex. */ -static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no) +/* Called with RTNL lock and dp->mutex. */ +static struct vport *new_vport(const struct vport_parms *parms) { - struct net_bridge_port *p; + struct vport *vport; - if (dev->br_port != NULL) - return -EBUSY; + vport_lock(); + vport = vport_add(parms); + if (!IS_ERR(vport)) { + struct datapath *dp = parms->dp; - p = kzalloc(sizeof(*p), GFP_KERNEL); - if (!p) - return -ENOMEM; + rcu_assign_pointer(dp->ports[parms->port_no], vport); + list_add_rcu(&vport->node, &dp->port_list); + dp->n_ports++; - dev_set_promiscuity(dev, 1); - dev_hold(dev); - p->port_no = port_no; - p->dp = dp; - p->dev = dev; - if (!is_dp_dev(dev)) - rcu_assign_pointer(dev->br_port, p); - else { - /* It would make sense to assign dev->br_port here too, but - * that causes packets received on internal ports to get caught - * in dp_frame_hook(). In turn dp_frame_hook() can reject them - * back to network stack, but that's a waste of time. */ + dp_ifinfo_notify(RTM_NEWLINK, vport); } - 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. */ - kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR); /* "brport" */ - p->kobj.kset = NULL; - p->kobj.parent = &p->dev->NETDEV_DEV_MEMBER.kobj; - kobject_init(&p->kobj, &brport_ktype); - - dp_ifinfo_notify(RTM_NEWLINK, p); + vport_unlock(); - return 0; + return vport; } -static int add_port(int dp_idx, struct odp_port __user *portp) +int dp_detach_port(struct vport *p) { - struct net_device *dev; - 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_no = port.port; - - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS) - goto out; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - err = -EEXIST; - if (dp->ports[port_no]) - goto out_unlock_dp; - - if (!(port.flags & ODP_PORT_INTERNAL)) { - err = -ENODEV; - dev = dev_get_by_name(&init_net, port.devname); - if (!dev) - goto out_unlock_dp; + ASSERT_RTNL(); - err = -EINVAL; - if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER || - is_dp_dev(dev)) - goto out_put; - } else { - dev = dp_dev_create(dp, port.devname, port_no); - err = PTR_ERR(dev); - if (IS_ERR(dev)) - goto out_unlock_dp; - dev_hold(dev); - } + if (p->port_no != ODPP_LOCAL) + dp_sysfs_del_if(p); + dp_ifinfo_notify(RTM_DELLINK, p); - err = new_nbp(dp, dev, port_no); - if (err) - goto out_put; + /* First drop references to device. */ + p->dp->n_ports--; + list_del_rcu(&p->node); + rcu_assign_pointer(p->dp->ports[p->port_no], NULL); -#ifdef SUPPORT_SYSFS - dp_sysfs_add_if(dp->ports[port_no]); -#endif + /* Then destroy it. */ + vport_lock(); + err = vport_del(p); + vport_unlock(); -out_put: - dev_put(dev); -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: return err; } -int dp_del_port(struct net_bridge_port *p) +/* Must be called with rcu_read_lock. */ +void dp_process_received_packet(struct vport *p, struct sk_buff *skb) { - ASSERT_RTNL(); + struct datapath *dp = p->dp; + struct dp_stats_percpu *stats; + int stats_counter_off; + struct sw_flow_actions *acts; + struct loop_counter *loop; + int error; -#ifdef SUPPORT_SYSFS - if (p->port_no != ODPP_LOCAL) - dp_sysfs_del_if(p); -#endif - dp_ifinfo_notify(RTM_DELLINK, p); + OVS_CB(skb)->vport = p; - p->dp->n_ports--; + if (!OVS_CB(skb)->flow) { + struct sw_flow_key key; + struct tbl_node *flow_node; + bool is_frag; - if (is_dp_dev(p->dev)) { - /* Make sure that no packets arrive from now on, since - * dp_dev_xmit() will try to find itself through - * p->dp->ports[], and we're about to set that to null. */ - netif_tx_disable(p->dev); - } + /* 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); + return; + } - /* First drop references to device. */ - dev_set_promiscuity(p->dev, -1); - list_del_rcu(&p->node); - rcu_assign_pointer(p->dp->ports[p->port_no], NULL); - rcu_assign_pointer(p->dev->br_port, NULL); + if (is_frag && dp->drop_frags) { + kfree_skb(skb); + stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); + goto out; + } - /* Then wait until no one is still using it, and destroy it. */ - synchronize_rcu(); + /* 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 (is_dp_dev(p->dev)) - dp_dev_destroy(p->dev); - dev_put(p->dev); - kobject_put(&p->kobj); + OVS_CB(skb)->flow = flow_cast(flow_node); + } - return 0; -} + stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + flow_used(OVS_CB(skb)->flow, skb); -static int del_port(int dp_idx, int port_no) -{ - struct net_bridge_port *p; - struct datapath *dp; - LIST_HEAD(dp_devs); - int err; + acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) - 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; + } - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; + /* Execute actions. */ + execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions, + acts->actions_len); - p = dp->ports[port_no]; - err = -ENOENT; - if (!p) - goto out_unlock_dp; + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + loop_suppress(dp, acts); - err = dp_del_port(p); +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + loop_put_counter(); -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); out: - return err; -} + /* Update datapath statistics. */ + local_bh_disable(); + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); -/* Must be called with rcu_read_lock. */ -static void -do_port_input(struct net_bridge_port *p, struct sk_buff *skb) -{ - /* Make our own copy of the packet. Otherwise we will mangle the - * packet for anyone who came before us (e.g. tcpdump via AF_PACKET). - * (No one comes after us, since we tell handle_bridge() that we took - * the packet.) */ - skb = skb_share_check(skb, GFP_ATOMIC); - if (!skb) - return; + write_seqcount_begin(&stats->seqlock); + (*(u64 *)((u8 *)stats + stats_counter_off))++; + write_seqcount_end(&stats->seqlock); - /* Push the Ethernet header back on. */ - skb_push(skb, ETH_HLEN); - skb_reset_mac_header(skb); - dp_process_received_packet(skb, p); + local_bh_enable(); } -/* Must be called with rcu_read_lock and with bottom-halves disabled. */ -void dp_process_received_packet(struct sk_buff *skb, struct net_bridge_port *p) +static void copy_and_csum_skb(struct sk_buff *skb, void *to) { - struct datapath *dp = p->dp; - struct dp_stats_percpu *stats; - struct odp_flow_key key; - struct sw_flow *flow; - - WARN_ON_ONCE(skb_shared(skb)); + u16 csum_start, csum_offset; + __wsum csum; - /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */ - stats = percpu_ptr(dp->stats_percpu, smp_processor_id()); + get_skb_csum_pointers(skb, &csum_start, &csum_offset); + csum_start -= skb_headroom(skb); + BUG_ON(csum_start >= skb_headlen(skb)); - if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { - if (dp->drop_frags) { - kfree_skb(skb); - stats->n_frags++; - return; - } - } + skb_copy_bits(skb, 0, to, csum_start); - flow = dp_table_lookup(rcu_dereference(dp->table), &key); - if (flow) { - 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, 0); - } + 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); } -/* - * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on - * different set of devices!) - */ -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) -/* Called with rcu_read_lock and bottom-halves disabled. */ -static struct sk_buff *dp_frame_hook(struct net_bridge_port *p, - struct sk_buff *skb) +/* Append each packet in 'skb' list to 'queue'. There will be only one packet + * unless we broke up a GSO packet. */ +static int queue_control_packets(struct datapath *dp, struct sk_buff *skb, + const struct dp_upcall_info *upcall_info) { - do_port_input(p, skb); - return NULL; -} -#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) -/* Called with rcu_read_lock and bottom-halves disabled. */ -static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb) -{ - do_port_input(p, *pskb); - return 1; -} -#else -#error -#endif + struct sk_buff *nskb; + int port_no; + int err; -#ifdef CONFIG_XEN -/* This code is copied verbatim from net/dev/core.c in Xen's - * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644. We can't call those functions - * directly because they aren't exported. */ -static int skb_pull_up_to(struct sk_buff *skb, void *ptr) -{ - if (ptr < (void *)skb->tail) - return 1; - if (__pskb_pull_tail(skb, - ptr - (void *)skb->data - skb_headlen(skb))) { - return 1; - } else { - return 0; - } -} + if (OVS_CB(skb)->vport) + port_no = OVS_CB(skb)->vport->port_no; + else + port_no = ODPP_LOCAL; -int skb_checksum_setup(struct sk_buff *skb) -{ - if (skb->proto_csum_blank) { - if (skb->protocol != htons(ETH_P_IP)) - goto out; - if (!skb_pull_up_to(skb, skb->nh.iph + 1)) - goto out; - skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl; - switch (skb->nh.iph->protocol) { - case IPPROTO_TCP: - skb->csum = offsetof(struct tcphdr, check); - break; - case IPPROTO_UDP: - skb->csum = 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", skb->nh.iph->protocol); - goto out; + do { + 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; + + len = sizeof(struct odp_packet); + len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */ + len += nla_total_size(skb->len); + len += nla_total_size(FLOW_BUFSIZE); + if (upcall_info->userdata) + len += nla_total_size(8); + if (upcall_info->sample_pool) + len += nla_total_size(4); + if (upcall_info->actions_len) + len += nla_total_size(upcall_info->actions_len); + + user_skb = alloc_skb(len, GFP_ATOMIC); + if (!user_skb) + goto err_kfree_skbs; + + upcall = (struct odp_packet *)__skb_put(user_skb, sizeof(*upcall)); + upcall->dp_idx = dp->dp_idx; + + nla_put_u32(user_skb, ODP_PACKET_ATTR_TYPE, upcall_info->type); + + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY); + flow_to_nlattrs(upcall_info->key, user_skb); + nla_nest_end(user_skb, nla); + + if (upcall_info->userdata) + nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata); + if (upcall_info->sample_pool) + nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool); + if (upcall_info->actions_len) { + const struct nlattr *actions = upcall_info->actions; + u32 actions_len = upcall_info->actions_len; + + nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS); + memcpy(__skb_put(user_skb, actions_len), actions, actions_len); + nla_nest_end(user_skb, nla); } - if (!skb_pull_up_to(skb, skb->h.raw + skb->csum + 2)) - goto out; - skb->ip_summed = CHECKSUM_HW; - skb->proto_csum_blank = 0; - } + + 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); + + upcall->len = user_skb->len; + skb_queue_tail(&dp->queues[upcall_info->type], user_skb); + + kfree_skb(skb); + skb = nskb; + } while (skb); return 0; -out: - return -EPROTO; + +err_kfree_skbs: + kfree_skb(skb); + while ((skb = nskb) != NULL) { + nskb = skb->next; + kfree_skb(skb); + } + return err; } -#endif -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 port_no; int err; WARN_ON_ONCE(skb_shared(skb)); - BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR); + BUG_ON(upcall_info->type >= DP_N_QUEUES); - queue = &dp->queues[queue_no]; + queue = &dp->queues[upcall_info->type]; err = -ENOBUFS; if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) goto err_kfree_skb; - /* If a checksum-deferred packet is forwarded to the controller, - * correct the pointers and checksum. This happens on a regular basis - * only on Xen (the CHECKSUM_HW case), on which VMs can pass up packets - * that do not have their checksum computed. We also implement it for - * the non-Xen case, but it is difficult to trigger or test this case - * there, hence the WARN_ON_ONCE(). - */ - err = skb_checksum_setup(skb); - if (err) + forward_ip_summed(skb); + + err = vswitch_skb_checksum_setup(skb); + if (err) goto err_kfree_skb; -#ifndef CHECKSUM_HW - if (skb->ip_summed == CHECKSUM_PARTIAL) { - WARN_ON_ONCE(1); -#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 - err = skb_checksum_help(skb); - if (err) - goto err_kfree_skb; - } -#else - if (skb->ip_summed == CHECKSUM_HW) { - err = skb_checksum_help(skb, 0); - if (err) - goto err_kfree_skb; - } -#endif /* 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; } } - /* Figure out port number. */ - port_no = ODPP_LOCAL; - if (skb->dev) { - if (skb->dev->br_port) - port_no = skb->dev->br_port->port_no; - else if (is_dp_dev(skb->dev)) - port_no = dp_dev_priv(skb->dev)->port_no; - } - - /* Append each packet to queue. There will be only one packet unless - * we broke up a GSO packet above. */ - do { - struct odp_msg *header; - struct sk_buff *nskb = skb->next; - skb->next = NULL; - - err = skb_cow(skb, sizeof *header); - if (err) { - while (nskb) { - kfree_skb(skb); - skb = nskb; - nskb = skb->next; - } - goto err_kfree_skb; - } - - 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); - - skb = nskb; - } while (skb); - + err = queue_control_packets(dp, skb, upcall_info); wake_up_interruptible(&dp->waitqueue); - return 0; + 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) { - dp->n_flows = 0; - return dp_table_flush(dp); + struct tbl *old_table = get_table_protected(dp); + struct tbl *new_table; + + new_table = tbl_create(TBL_MIN_BUCKETS); + if (!new_table) + return -ENOMEM; + + rcu_assign_pointer(dp->table, new_table); + + tbl_deferred_destroy(old_table, flow_free_tbl); + + return 0; } -static int validate_actions(const struct sw_flow_actions *actions) +static int validate_actions(const struct nlattr *actions, u32 actions_len) { - unsigned int i; + 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; + switch (type) { + case ODPAT_UNSPEC: + return -EINVAL; + + case ODPAT_CONTROLLER: + case ODPAT_STRIP_VLAN: + case ODPAT_SET_DL_SRC: + case ODPAT_SET_DL_DST: + case ODPAT_SET_NW_SRC: + case ODPAT_SET_NW_DST: + case ODPAT_SET_TP_SRC: + case ODPAT_SET_TP_DST: + case ODPAT_SET_TUNNEL: + case ODPAT_SET_PRIORITY: + case ODPAT_POP_PRIORITY: + case ODPAT_DROP_SPOOFED_ARP: + /* No validation needed. */ break; - case ODPAT_OUTPUT_GROUP: - if (a->output_group.group >= DP_MAX_GROUPS) + case ODPAT_OUTPUT: + if (nla_get_u32(a) >= DP_MAX_PORTS) return -EINVAL; break; - case ODPAT_SET_VLAN_VID: - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) + case ODPAT_SET_DL_TCI: + if (nla_get_be16(a) & htons(VLAN_CFI_MASK)) return -EINVAL; break; - case ODPAT_SET_VLAN_PCP: - if (a->vlan_pcp.vlan_pcp & ~VLAN_PCP_MASK) + case ODPAT_SET_NW_TOS: + 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; } @@ -802,16 +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; @@ -825,107 +762,128 @@ 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 put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) +static int expand_table(struct datapath *dp) { - struct odp_flow_put uf; - struct sw_flow *flow, **bucket; - struct dp_table *table; - struct odp_flow_stats stats; - int error; - - error = -EFAULT; - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) - goto error; - uf.flow.key.reserved = 0; - -retry: - table = rcu_dereference(dp->table); - bucket = dp_table_lookup_for_insert(table, &uf.flow.key); - if (!bucket) { - /* No such flow, and the slots where it could go are full. */ - error = uf.flags & ODPPF_CREATE ? -EXFULL : -ENOENT; - goto error; - } else if (!*bucket) { - /* No such flow, but we found an available slot for it. */ - struct sw_flow_actions *acts; + struct tbl *old_table = get_table_protected(dp); + struct tbl *new_table; - error = -ENOENT; - if (!(uf.flags & ODPPF_CREATE)) - goto error; + new_table = tbl_expand(old_table); + if (IS_ERR(new_table)) + return PTR_ERR(new_table); - /* Expand table, if necessary, to make room. */ - if (dp->n_flows * 4 >= table->n_buckets && - table->n_buckets < DP_MAX_BUCKETS) { - error = dp_table_expand(dp); - if (error) - goto error; + rcu_assign_pointer(dp->table, new_table); + tbl_deferred_destroy(old_table, NULL); - /* The bucket's location has changed. Try again. */ - goto retry; + return 0; +} + +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; + + error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key, + uf->flow.key_len); + if (error) + return error; + + hash = flow_hash(&key); + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &key, hash, flow_cmp); + if (!flow_node) { + /* No such flow. */ + error = -ENOENT; + if (!(uf->flags & ODPPF_CREATE)) + 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 = 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. */ - acts = get_actions(&uf.flow); + acts = get_actions(&uf->flow); error = PTR_ERR(acts); if (IS_ERR(acts)) goto error_free_flow; rcu_assign_pointer(flow->sf_acts, acts); /* Put flow in bucket. */ - rcu_assign_pointer(*bucket, flow); - dp->n_flows++; - memset(&stats, 0, sizeof(struct odp_flow_stats)); + error = tbl_insert(table, &flow->tbl_node, hash); + if (error) + goto error_free_flow_acts; + + memset(stats, 0, sizeof(struct odp_flow_stats)); } else { /* We found a matching flow. */ - struct sw_flow *flow = *rcu_dereference(bucket); struct sw_flow_actions *old_acts, *new_acts; - unsigned long int flags; + + flow = flow_cast(flow_node); /* Bail out if we're not allowed to modify an existing flow. */ error = -EEXIST; - if (!(uf.flags & ODPPF_MODIFY)) + if (!(uf->flags & ODPPF_MODIFY)) goto error; /* Swap actions. */ - new_acts = get_actions(&uf.flow); + 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->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 { @@ -933,243 +891,291 @@ retry: } /* Fetch stats, then clear them if necessary. */ - spin_lock_irqsave(&flow->lock, flags); - get_stats(flow, &stats); - if (uf.flags & ODPPF_ZERO_STATS) + spin_lock_bh(&flow->lock); + get_stats(flow, stats); + if (uf->flags & ODPPF_ZERO_STATS) clear_stats(flow); - spin_unlock_irqrestore(&flow->lock, flags); + spin_unlock_bh(&flow->lock); } - /* Copy stats to userspace. */ - if (__copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; return 0; +error_free_flow_acts: + kfree(acts); error_free_flow: - kmem_cache_free(flow_cache, flow); + flow->sf_acts = NULL; + flow_put(flow); error: return error; } -static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp) +static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) { - union odp_action __user *actions; - struct sw_flow_actions *sf_acts; - u32 n_actions; + struct odp_flow_stats stats; + struct odp_flow_put uf; + int error; - if (__get_user(actions, &ufp->actions) || - __get_user(n_actions, &ufp->n_actions)) + if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) return -EFAULT; - if (!n_actions) - return 0; - if (ufp->n_actions > INT_MAX / sizeof(union odp_action)) - return -EINVAL; + error = do_put_flow(dp, &uf, &stats); + if (error) + return error; - sf_acts = rcu_dereference(flow->sf_acts); - if (__put_user(sf_acts->n_actions, &ufp->n_actions) || - (actions && copy_to_user(actions, sf_acts->actions, - sizeof(union odp_action) * - min(sf_acts->n_actions, n_actions)))) + if (copy_to_user(&ufp->flow.stats, &stats, + sizeof(struct odp_flow_stats))) return -EFAULT; return 0; } -static int answer_query(struct sw_flow *flow, struct odp_flow __user *ufp) +static int do_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, + struct odp_flow_stats __user *ustats, + struct nlattr __user *actions, + u32 __user *actions_lenp) { + struct sw_flow_actions *sf_acts; struct odp_flow_stats stats; - unsigned long int flags; + u32 actions_len; - spin_lock_irqsave(&flow->lock, flags); + spin_lock_bh(&flow->lock); get_stats(flow, &stats); - spin_unlock_irqrestore(&flow->lock, flags); + if (query_flags & ODPFF_ZERO_TCP_FLAGS) + flow->tcp_flags = 0; - if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats))) + spin_unlock_bh(&flow->lock); + + if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || + get_user(actions_len, actions_lenp)) return -EFAULT; - return put_actions(flow, ufp); + + if (!actions_len) + return 0; + + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); + if (put_user(sf_acts->actions_len, actions_lenp) || + (actions && copy_to_user(actions, sf_acts->actions, + min(sf_acts->actions_len, actions_len)))) + return -EFAULT; + + return 0; } -static int del_or_query_flow(struct datapath *dp, - struct odp_flow __user *ufp, - unsigned int cmd) +static int answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct odp_flow __user *ufp) { - struct dp_table *table = rcu_dereference(dp->table); - struct odp_flow uf; - struct sw_flow *flow; + struct nlattr __user *actions; + + if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions)) + return -EFAULT; + + return do_answer_query(dp, flow, query_flags, + &ufp->stats, actions, &ufp->actions_len); +} + +static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len) +{ + struct tbl *table = get_table_protected(dp); + struct tbl_node *flow_node; + struct sw_flow_key swkey; int error; - error = -EFAULT; - if (copy_from_user(&uf, ufp, sizeof uf)) - goto error; - uf.key.reserved = 0; + error = flow_copy_from_user(&swkey, key, key_len); + if (error) + return ERR_PTR(error); - flow = dp_table_lookup(table, &uf.key); - error = -ENOENT; - if (!flow) - goto error; + flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp); + if (!flow_node) + return ERR_PTR(-ENOENT); - if (cmd == ODP_FLOW_DEL) { - /* XXX redundant lookup */ - error = dp_table_delete(table, flow); - if (error) - goto error; + error = tbl_remove(table, flow_node); + if (error) + return ERR_PTR(error); - /* XXX These 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. */ - dp->n_flows--; - error = answer_query(flow, ufp); - flow_deferred_free(flow); - } else { - error = answer_query(flow, ufp); - } + /* 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); +} -error: +static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) +{ + struct sw_flow *flow; + struct odp_flow uf; + int error; + + if (copy_from_user(&uf, ufp, sizeof(uf))) + return -EFAULT; + + 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(dp, flow, 0, ufp); + flow_deferred_free(flow); return error; } -static int query_multiple_flows(struct datapath *dp, - const struct odp_flowvec *flowvec) +static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) { - struct dp_table *table = rcu_dereference(dp->table); - int i; + struct tbl *table = get_table_protected(dp); + u32 i; + for (i = 0; i < flowvec->n_flows; i++) { - struct __user odp_flow *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 sw_flow *flow; + 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; - uf.key.reserved = 0; - flow = dp_table_lookup(table, &uf.key); - if (!flow) - error = __clear_user(&ufp->stats, sizeof ufp->stats); + 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, 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; - int n_flows; - int listed_flows; -}; - -static int list_flow(struct sw_flow *flow, void *cbdata_) +static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, + int (*function)(struct datapath *, + const struct odp_flowvec *)) { - struct list_flows_cbdata *cbdata = cbdata_; - struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; - int error; + struct odp_flowvec __user *uflowvec; + struct odp_flowvec flowvec; + int retval; - if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) + uflowvec = (struct odp_flowvec __user *)argp; + if (copy_from_user(&flowvec, uflowvec, sizeof(flowvec))) return -EFAULT; - error = answer_query(flow, ufp); - if (error) - return error; - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; + if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) + return -EINVAL; + + retval = function(dp, &flowvec); + return (retval < 0 ? retval + : retval == flowvec.n_flows ? 0 + : put_user(retval, &uflowvec->n_flows)); } -static int list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) +static struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state) { - struct list_flows_cbdata cbdata; - int error; + struct tbl *table = get_table_protected(dp); + struct tbl_node *tbl_node; + u32 bucket, obj; - if (!flowvec->n_flows) - return 0; + 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); - cbdata.uflows = flowvec->flows; - cbdata.n_flows = flowvec->n_flows; - cbdata.listed_flows = 0; - error = dp_table_foreach(rcu_dereference(dp->table), - list_flow, &cbdata); - return error ? error : cbdata.listed_flows; + return tbl_node ? flow_cast(tbl_node) : NULL; } -static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - const struct odp_flowvec *)) +static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp) { - struct odp_flowvec __user *uflowvec; - struct odp_flowvec flowvec; - int retval; + struct odp_flow __user *uflowp; + struct nlattr __user *ukey; + struct sw_flow *flow; + u32 key_len; - uflowvec = (struct odp_flowvec __user *)argp; - if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) || - copy_from_user(&flowvec, uflowvec, sizeof flowvec)) + 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 (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) - return -EINVAL; + if (!flow) + return put_user(ODPFF_EOF, &uflowp->flags); - if (!access_ok(VERIFY_WRITE, flowvec.flows, - flowvec.n_flows * sizeof(struct odp_flow))) + if (put_user(0, &uflowp->flags) || + get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) || + get_user(key_len, &uflowp->key_len)) return -EFAULT; - retval = function(dp, &flowvec); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : __put_user(retval, &uflowvec->n_flows)); + 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 *executep) +static int do_execute(struct datapath *dp, const struct odp_execute *execute) { - 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 = -EFAULT; - if (copy_from_user(&execute, executep, sizeof execute)) - goto error; - err = -EINVAL; - if (execute.length < ETH_HLEN || execute.length > 65535) + 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; err = -ENOMEM; - skb = alloc_skb(execute.length, GFP_KERNEL); + skb = alloc_skb(execute->length, GFP_KERNEL); if (!skb) goto error_free_actions; - if (execute.in_port < DP_MAX_PORTS) { - struct net_bridge_port *p = dp->ports[execute.in_port]; - if (p) - skb->dev = p->dev; - } err = -EFAULT; - if (copy_from_user(skb_put(skb, execute.length), execute.data, - execute.length)) + if (copy_from_user(skb_put(skb, execute->length), + (const void __user __force *)execute->data, + execute->length)) goto error_free_skb; - flow_extract(skb, execute.in_port, &key); - err = execute_actions(dp, skb, &key, actions->actions, - actions->n_actions, GFP_KERNEL); + 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; + + rcu_read_lock(); + err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); + rcu_read_unlock(); + kfree(actions); return err; @@ -1181,193 +1187,489 @@ error: return err; } -static int -get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) +static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep) +{ + struct odp_execute execute; + + if (copy_from_user(&execute, executep, sizeof(execute))) + return -EFAULT; + + return do_execute(dp, &execute); +} + +static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) { struct odp_stats stats; int i; - stats.n_flows = dp->n_flows; - stats.cur_capacity = rcu_dereference(dp->table)->n_buckets * 2; - stats.max_capacity = DP_MAX_BUCKETS * 2; 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 net_bridge_port *p; + struct vport *p; int mtu = 0; ASSERT_RTNL(); list_for_each_entry_rcu (p, &dp->port_list, node) { - struct net_device *dev = p->dev; + int dev_mtu; /* Skip any internal ports, since that's what we're trying to * set. */ - if (is_dp_dev(dev)) + if (is_internal_vport(p)) continue; - if (!mtu || dev->mtu < mtu) - mtu = dev->mtu; + dev_mtu = vport_get_mtu(p); + if (!mtu || dev_mtu < mtu) + mtu = dev_mtu; } return mtu ? mtu : ETH_DATA_LEN; } -static int -put_port(const struct net_bridge_port *p, struct odp_port __user *uop) +/* Sets the MTU of all datapath devices to the minimum of the ports. Must + * be called with RTNL lock. */ +void set_internal_devs_mtu(const struct datapath *dp) +{ + struct vport *p; + int mtu; + + ASSERT_RTNL(); + + mtu = dp_min_mtu(dp); + + list_for_each_entry_rcu (p, &dp->port_list, node) { + if (is_internal_vport(p)) + vport_set_mtu(p, mtu); + } +} + +static int get_listen_mask(const struct file *f) { - struct odp_port op; - memset(&op, 0, sizeof op); - strncpy(op.devname, p->dev->name, sizeof op.devname); - op.port = p->port_no; - op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0; - return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; + return (long)f->private_data; } -static int -query_port(struct datapath *dp, struct odp_port __user *uport) +static void set_listen_mask(struct file *f, int listen_mask) { - struct odp_port port; + f->private_data = (void*)(long)listen_mask; +} - if (copy_from_user(&port, uport, sizeof port)) - return -EFAULT; - if (port.devname[0]) { - struct net_bridge_port *p; - struct net_device *dev; - int err; - - port.devname[IFNAMSIZ - 1] = '\0'; - - dev = dev_get_by_name(&init_net, port.devname); - if (!dev) - return -ENODEV; - - p = dev->br_port; - if (!p && is_dp_dev(dev)) { - struct dp_dev *dp_dev = dp_dev_priv(dev); - if (dp_dev->dp == dp) - p = dp->ports[dp_dev->port_no]; +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_vport *odp_vport; + struct sk_buff *skb; + struct nlattr *nla; + int ifindex, iflink; + int err; + + skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; + + rcu_read_lock(); + odp_vport = (struct odp_vport*)__skb_put(skb, sizeof(struct odp_vport)); + odp_vport->dp_idx = vport->dp->dp_idx; + odp_vport->total_len = total_len; + + 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); + + NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport)); + + NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport)); + + 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 struct sk_buff *copy_vport_from_user(struct odp_vport __user *uodp_vport, + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + struct odp_vport *odp_vport; + struct sk_buff *skb; + u32 len; + int err; + + if (get_user(len, &uodp_vport->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_vport)) + return ERR_PTR(-EINVAL); + + skb = alloc_skb(len, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); + + err = -EFAULT; + if (copy_from_user(__skb_put(skb, len), uodp_vport, len)) + goto error_free_skb; + + odp_vport = (struct odp_vport *)skb->data; + err = -EINVAL; + if (odp_vport->len != len) + goto error_free_skb; + + err = nla_parse(a, ODP_VPORT_ATTR_MAX, (struct nlattr *)(skb->data + sizeof(struct odp_vport)), + skb->len - sizeof(struct odp_vport), vport_policy); + if (err) + goto error_free_skb; + + err = VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1); + if (err) + goto error_free_skb; + + return skb; + +error_free_skb: + kfree_skb(skb); + return ERR_PTR(err); +} + + +/* 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) { + vport_unlock(); + return ERR_PTR(-ENODEV); } - err = p && p->dp == dp ? put_port(p, uport) : -ENOENT; - dev_put(dev); + dp_idx = vport->dp->dp_idx; + port_no = vport->port_no; + vport_unlock(); - return err; - } else { - if (port.port >= DP_MAX_PORTS) - return -EINVAL; - if (!dp->ports[port.port]) - return -ENOENT; - return put_port(dp->ports[port.port], uport); - } + dp = get_dp_locked(dp_idx); + if (!dp) + goto retry; + + vport = get_vport_protected(dp, port_no); + if (!vport || + strcmp(vport_get_name(vport), nla_data(a[ODP_VPORT_ATTR_NAME]))) { + mutex_unlock(&dp->mutex); + goto retry; + } + + return vport; + } else if (a[ODP_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + + if (port_no >= DP_MAX_PORTS) + return ERR_PTR(-EINVAL); + + dp = get_dp_locked(odp_vport->dp_idx); + if (!dp) + return ERR_PTR(-ENODEV); + + 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 -list_ports(struct datapath *dp, struct odp_portvec __user *pvp) +static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) { - struct odp_portvec pv; - struct net_bridge_port *p; - int 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; +} - if (copy_from_user(&pv, pvp, sizeof pv)) - return -EFAULT; +static int attach_vport(struct odp_vport __user *uodp_vport) +{ + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + struct vport_parms parms; + struct vport *vport; + struct sk_buff *skb; + struct datapath *dp; + u32 port_no; + int err; + + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_vport = (struct odp_vport *)skb->data; + + err = -EINVAL; + if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE]) + goto exit_kfree_skb; + + rtnl_lock(); + + dp = get_dp_locked(odp_vport->dp_idx); + err = -ENODEV; + if (!dp) + goto exit_unlock_rtnl; + + if (a[ODP_VPORT_ATTR_PORT_NO]) { + port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); - idx = 0; - if (pv.n_ports) { - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (put_port(p, &pv.ports[idx])) - return -EFAULT; - if (idx++ >= pv.n_ports) + err = -EFBIG; + if (port_no >= DP_MAX_PORTS) + goto exit_unlock_dp; + + 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; } } - return put_user(idx, &pvp->n_ports); + + 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; + + vport = new_vport(&parms); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_unlock_dp; + + set_internal_devs_mtu(dp); + dp_sysfs_add_if(vport); + + err = change_vport(vport, a); + if (err) { + dp_detach_port(vport); + goto exit_unlock_dp; + } + + err = copy_vport_to_user(uodp_vport, vport, odp_vport->total_len); + +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_unlock_rtnl: + rtnl_unlock(); +exit_kfree_skb: + kfree_skb(skb); +exit: + return err; } -/* RCU callback for freeing a dp_port_group */ -static void free_port_group(struct rcu_head *rcu) +static int set_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) { - struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu); - kfree(g); + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + 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; + + rtnl_lock(); + vport = lookup_vport((struct odp_vport *)skb->data, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; + + 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); + + mutex_unlock(&vport->dp->mutex); +exit_free: + kfree_skb(skb); + rtnl_unlock(); +exit: + return err; } -static int -set_port_group(struct datapath *dp, const 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 dp_port_group *new_group, *old_group; - int error; - - error = -EFAULT; - if (copy_from_user(&pg, upg, sizeof pg)) - goto error; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct datapath *dp; + struct vport *vport; + struct sk_buff *skb; + int err; - error = -EINVAL; - if (pg.n_ports > DP_MAX_PORTS || pg.group >= DP_MAX_GROUPS) - goto error; + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - error = -ENOMEM; - new_group = kmalloc(sizeof *new_group + sizeof(u16) * pg.n_ports, - GFP_KERNEL); - if (!new_group) - goto error; + 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; - new_group->n_ports = pg.n_ports; - error = -EFAULT; - if (copy_from_user(new_group->ports, pg.ports, - sizeof(u16) * pg.n_ports)) - goto error_free; - - old_group = rcu_dereference(dp->groups[pg.group]); - rcu_assign_pointer(dp->groups[pg.group], new_group); - if (old_group) - call_rcu(&old_group->rcu, free_port_group); - return 0; + err = -EINVAL; + if (vport->port_no == ODPP_LOCAL) + goto exit_free; -error_free: - kfree(new_group); -error: - return error; + err = dp_detach_port(vport); + mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); + rtnl_unlock(); +exit: + return err; } -static int -get_port_group(struct datapath *dp, struct odp_port_group *upg) +static int get_vport(struct odp_vport __user *uodp_vport) { - struct odp_port_group pg; - struct dp_port_group *g; - u16 n_copy; + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + 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; + odp_vport = (struct odp_vport *)skb->data; - if (pg.group >= DP_MAX_GROUPS) - return -EINVAL; + vport = lookup_vport(odp_vport, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; - g = dp->groups[pg.group]; - n_copy = g ? min_t(int, g->n_ports, pg.n_ports) : 0; - if (n_copy && copy_to_user(pg.ports, g->ports, n_copy * sizeof(u16))) - return -EFAULT; + 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; +} - if (put_user(g ? g->n_ports : 0, &upg->n_ports)) - return -EFAULT; +static int dump_vport(struct odp_vport __user *uodp_vport) +{ + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct odp_vport *odp_vport; + struct sk_buff *skb; + struct datapath *dp; + u32 port_no; + int err; - return 0; + 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, @@ -1375,30 +1677,45 @@ 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; /* Handle commands with special locking requirements up front. */ switch (cmd) { case ODP_DP_CREATE: - return create_dp(dp_idx, (char __user *)argp); + err = create_dp(dp_idx, (char __user *)argp); + goto exit; case ODP_DP_DESTROY: - return destroy_dp(dp_idx); + err = destroy_dp(dp_idx); + goto exit; - case ODP_PORT_ADD: - return add_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_DEL: - err = get_user(port_no, (int __user *)argp); - if (err) - break; - return del_port(dp_idx, port_no); + case ODP_VPORT_GET: + err = get_vport((struct odp_vport __user *)argp); + goto exit; + + case ODP_VPORT_DEL: + err = del_vport(cmd, (struct odp_vport __user *)argp); + goto exit; + + case ODP_VPORT_SET: + err = set_vport(cmd, (struct odp_vport __user *)argp); + goto exit; + + case ODP_VPORT_DUMP: + err = dump_vport((struct odp_vport __user *)argp); + goto exit; } dp = get_dp_locked(dp_idx); + err = -ENODEV; if (!dp) - return -ENODEV; + goto exit; switch (cmd) { case ODP_DP_STATS: @@ -1421,7 +1738,7 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, break; case ODP_GET_LISTEN_MASK: - err = put_user((int)f->private_data, (int __user *)argp); + err = put_user(get_listen_mask(f), (int __user *)argp); break; case ODP_SET_LISTEN_MASK: @@ -1432,23 +1749,17 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, if (listeners & ~ODPL_ALL) break; err = 0; - f->private_data = (void*)listeners; - break; - - case ODP_PORT_QUERY: - err = query_port(dp, (struct odp_port __user *)argp); + set_listen_mask(f, listeners); break; - case ODP_PORT_LIST: - err = list_ports(dp, (struct odp_portvec __user *)argp); + case ODP_GET_SFLOW_PROBABILITY: + err = put_user(dp->sflow_probability, (unsigned int __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); + case ODP_SET_SFLOW_PROBABILITY: + err = get_user(sflow_probability, (unsigned int __user *)argp); + if (!err) + dp->sflow_probability = sflow_probability; break; case ODP_FLOW_FLUSH: @@ -1460,21 +1771,19 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, break; case ODP_FLOW_DEL: - case ODP_FLOW_GET: - err = del_or_query_flow(dp, (struct odp_flow __user *)argp, - cmd); + err = del_flow(dp, (struct odp_flow __user *)argp); break; - case ODP_FLOW_GET_MULTIPLE: - err = do_flowvec_ioctl(dp, argp, query_multiple_flows); + case ODP_FLOW_GET: + err = do_flowvec_ioctl(dp, argp, do_query_flows); break; - case ODP_FLOW_LIST: - err = do_flowvec_ioctl(dp, argp, list_flows); + case ODP_FLOW_DUMP: + err = dump_flow(dp, (struct odp_flow_dump __user *)argp); break; case ODP_EXECUTE: - err = do_execute(dp, (struct odp_execute __user *)argp); + err = execute_packet(dp, (struct odp_execute __user *)argp); break; default: @@ -1482,6 +1791,7 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, break; } mutex_unlock(&dp->mutex); +exit: return err; } @@ -1495,16 +1805,264 @@ static int dp_has_packet_of_interest(struct datapath *dp, int listeners) return 0; } -ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, - loff_t *ppos) +#ifdef CONFIG_COMPAT +static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) +{ + compat_uptr_t key, actions; + + if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) || + __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) || + __get_user(key, &compat->key) || + __get_user(flow->key_len, &compat->key_len) || + __get_user(actions, &compat->actions) || + __get_user(flow->actions_len, &compat->actions_len) || + __get_user(flow->flags, &compat->flags)) + return -EFAULT; + + flow->key = (struct nlattr __force *)compat_ptr(key); + flow->actions = (struct nlattr __force *)compat_ptr(actions); + return 0; +} + +static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp) +{ + struct odp_flow_stats stats; + struct odp_flow_put fp; + int error; + + if (compat_get_flow(&fp.flow, &ufp->flow) || + get_user(fp.flags, &ufp->flags)) + return -EFAULT; + + error = do_put_flow(dp, &fp, &stats); + if (error) + return error; + + if (copy_to_user(&ufp->flow.stats, &stats, + sizeof(struct odp_flow_stats))) + return -EFAULT; + + return 0; +} + +static int compat_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, + struct compat_odp_flow __user *ufp) +{ + compat_uptr_t actions; + + if (get_user(actions, &ufp->actions)) + return -EFAULT; + + 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) +{ + struct sw_flow *flow; + struct odp_flow uf; + int error; + + if (compat_get_flow(&uf, ufp)) + return -EFAULT; + + 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(dp, flow, 0, ufp); + flow_deferred_free(flow); + return error; +} + +static int compat_query_flows(struct datapath *dp, + struct compat_odp_flow __user *flows, + u32 n_flows) +{ + struct tbl *table = get_table_protected(dp); + u32 i; + + for (i = 0; i < n_flows; i++) { + struct compat_odp_flow __user *ufp = &flows[i]; + struct odp_flow uf; + struct tbl_node *flow_node; + struct sw_flow_key key; + int error; + + if (compat_get_flow(&uf, ufp)) + return -EFAULT; + + error = flow_copy_from_user(&key, (const struct nlattr __force __user *) uf.key, uf.key_len); + if (error) + return error; + + flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp); + if (!flow_node) + error = put_user(ENOENT, &ufp->stats.error); + else + error = compat_answer_query(dp, flow_cast(flow_node), + uf.flags, ufp); + if (error) + return -EFAULT; + } + return n_flows; +} + +static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp) +{ + struct compat_odp_flow __user *uflowp; + compat_uptr_t compat_ufp; + struct sw_flow *flow; + compat_uptr_t ukey; + u32 key_len; + + flow = do_dump_flow(dp, udumpp->state); + if (IS_ERR(flow)) + return PTR_ERR(flow); + + if (get_user(compat_ufp, &udumpp->flow)) + return -EFAULT; + uflowp = compat_ptr(compat_ufp); + + if (!flow) + return put_user(ODPFF_EOF, &uflowp->flags); + + if (put_user(0, &uflowp->flags) || + get_user(ukey, &uflowp->key) || + get_user(key_len, &uflowp->key_len)) + return -EFAULT; + + key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len); + if (key_len < 0) + return key_len; + if (put_user(key_len, &uflowp->key_len)) + return -EFAULT; + + return compat_answer_query(dp, flow, 0, uflowp); +} + +static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, + int (*function)(struct datapath *, + struct compat_odp_flow __user *, + u32 n_flows)) +{ + struct compat_odp_flowvec __user *uflowvec; + struct compat_odp_flow __user *flows; + struct compat_odp_flowvec flowvec; + int retval; + + uflowvec = compat_ptr(argp); + if (!access_ok(VERIFY_WRITE, uflowvec, sizeof(*uflowvec)) || + copy_from_user(&flowvec, uflowvec, sizeof(flowvec))) + return -EFAULT; + + if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow)) + return -EINVAL; + + flows = compat_ptr(flowvec.flows); + if (!access_ok(VERIFY_WRITE, flows, + flowvec.n_flows * sizeof(struct compat_odp_flow))) + return -EFAULT; + + retval = function(dp, flows, flowvec.n_flows); + return (retval < 0 ? retval + : retval == flowvec.n_flows ? 0 + : put_user(retval, &uflowvec->n_flows)); +} + +static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute) +{ + struct odp_execute execute; + compat_uptr_t actions; + compat_uptr_t data; + + if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) || + __get_user(actions, &uexecute->actions) || + __get_user(execute.actions_len, &uexecute->actions_len) || + __get_user(data, &uexecute->data) || + __get_user(execute.length, &uexecute->length)) + return -EFAULT; + + execute.actions = (struct nlattr __force *)compat_ptr(actions); + execute.data = (const void __force *)compat_ptr(data); + + return do_execute(dp, &execute); +} + +static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) { - /* XXX is there sufficient synchronization here? */ - int listeners = (int) f->private_data; int dp_idx = iminor(f->f_dentry->d_inode); - struct datapath *dp = get_dp(dp_idx); + struct datapath *dp; + 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_NEW: + case ODP_VPORT_DEL: + case ODP_VPORT_GET: + case ODP_VPORT_SET: + case ODP_VPORT_DUMP: + case ODP_DP_STATS: + case ODP_GET_DROP_FRAGS: + case ODP_SET_DROP_FRAGS: + case ODP_SET_LISTEN_MASK: + case ODP_GET_LISTEN_MASK: + case ODP_SET_SFLOW_PROBABILITY: + case ODP_GET_SFLOW_PROBABILITY: + /* Ioctls that just need their pointer argument extended. */ + return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); + } + + dp = get_dp_locked(dp_idx); + err = -ENODEV; + if (!dp) + goto exit; + + switch (cmd) { + case ODP_FLOW_PUT32: + err = compat_put_flow(dp, compat_ptr(argp)); + break; + + case ODP_FLOW_DEL32: + err = compat_del_flow(dp, compat_ptr(argp)); + break; + + case ODP_FLOW_GET32: + err = compat_flowvec_ioctl(dp, argp, compat_query_flows); + break; + + case ODP_FLOW_DUMP32: + err = compat_dump_flow(dp, compat_ptr(argp)); + break; + + case ODP_EXECUTE32: + err = compat_execute(dp, compat_ptr(argp)); + break; + + default: + err = -ENOIOCTLCMD; + break; + } + mutex_unlock(&dp->mutex); +exit: + return err; +} +#endif + +static ssize_t openvswitch_read(struct file *f, char __user *buf, + size_t nbytes, loff_t *ppos) +{ + int listeners = get_listen_mask(f); + int dp_idx = iminor(f->f_dentry->d_inode); + struct datapath *dp = get_dp_locked(dp_idx); struct sk_buff *skb; - struct iovec __user iov; - size_t copy_bytes; + struct iovec iov; int retval; if (!dp) @@ -1539,94 +2097,83 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, } } success: - copy_bytes = min(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) { mask = 0; poll_wait(file, &dp->waitqueue, wait); - if (dp_has_packet_of_interest(dp, (int)file->private_data)) + 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, - /* XXX .fasync = openvswitch_fasync, */ +#ifdef CONFIG_COMPAT + .compat_ioctl = openvswitch_compat_ioctl, +#endif }; static int major; -static struct llc_sap *dp_stp_sap; - -static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev, - struct packet_type *pt, struct net_device *orig_dev) -{ - /* We don't really care about STP packets, we just listen for them for - * mutual exclusion with the bridge module, so this just discards - * them. */ - kfree_skb(skb); - return 0; -} static int __init dp_init(void) { + struct sk_buff *dummy_skb; int err; - printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR); + BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb)); - /* Register to receive STP packets because the bridge module also - * attempts to do so. Since there can only be a single listener for a - * given protocol, this provides mutual exclusion against the bridge - * module, preventing both of them from being loaded at the same - * time. */ - dp_stp_sap = llc_sap_open(LLC_SAP_BSPAN, dp_stp_rcv); - if (!dp_stp_sap) { - printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n"); - return -EADDRINUSE; - } + printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR); err = flow_init(); if (err) goto error; - err = register_netdevice_notifier(&dp_device_notifier); + err = vport_init(); if (err) goto error_flow_exit; + err = register_netdevice_notifier(&dp_device_notifier); + if (err) + goto error_vport_exit; + major = register_chrdev(0, "openvswitch", &openvswitch_fops); if (err < 0) goto error_unreg_notifier; - /* Hook into callback used by the bridge to intercept packets. - * Parasites we are. */ - br_handle_frame_hook = dp_frame_hook; - return 0; error_unreg_notifier: unregister_netdevice_notifier(&dp_device_notifier); +error_vport_exit: + vport_exit(); error_flow_exit: flow_exit(); error: @@ -1638,9 +2185,8 @@ static void dp_cleanup(void) rcu_barrier(); unregister_chrdev(major, "openvswitch"); unregister_netdevice_notifier(&dp_device_notifier); + vport_exit(); flow_exit(); - br_handle_frame_hook = NULL; - llc_sap_put(dp_stp_sap); } module_init(dp_init);