X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=94f908a0ec5c26959a658ef7bee639d3b11ce81e;hb=82272eded1ede569bcec3ba4ab212e5e3fb632ff;hp=382a8c279376f1bb3ee6bd2a4719c23c58fb8037;hpb=e779d8d90d65297473febcf82ec44c9225cc4fe3;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index 382a8c279..94f908a0e 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 @@ -41,19 +41,18 @@ #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); @@ -66,17 +65,18 @@ EXPORT_SYMBOL(dp_ioctl_hook); * 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[256]; static DEFINE_MUTEX(dp_mutex); -static int new_vport(struct datapath *, struct odp_port *, int port_no); +static struct vport *new_vport(const struct vport_parms *); /* 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) + if (dp_idx < 0 || dp_idx >= ARRAY_SIZE(dps)) 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); @@ -92,10 +92,22 @@ 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]); + return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); } static inline size_t br_nlmsg_size(void) @@ -113,7 +125,7 @@ static int dp_fill_ifinfo(struct sk_buff *skb, const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; + struct datapath *dp = port->dp; int ifindex = vport_get_ifindex(port); int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; @@ -138,7 +150,8 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->ifi_change = 0; NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port)); - NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL])); + NLA_PUT_U32(skb, IFLA_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, @@ -192,223 +205,55 @@ static struct kobj_type dp_ktype = { .release = release_dp }; -static int create_dp(int dp_idx, const char __user *devnamep) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct odp_port internal_dev_port; - char devname[IFNAMSIZ]; - struct datapath *dp; - int err; + struct datapath *dp = container_of(rcu, struct datapath, rcu); int i; - if (devnamep) { - int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ); - if (retval < 0) { - err = -EFAULT; - goto err; - } else if (retval >= IFNAMSIZ) { - err = -ENAMETOOLONG; - goto err; - } - } else { - snprintf(devname, sizeof devname, "of%d", dp_idx); - } - - rtnl_lock(); - mutex_lock(&dp_mutex); - err = -ENODEV; - if (!try_module_get(THIS_MODULE)) - goto err_unlock; - - /* Exit early if a datapath with that number already exists. - * (We don't use -EEXIST because that's ambiguous with 'devname' - * conflicting with an existing network device name.) */ - err = -EBUSY; - if (get_dp(dp_idx)) - goto err_put_module; - - err = -ENOMEM; - dp = kzalloc(sizeof *dp, GFP_KERNEL); - if (dp == NULL) - goto err_put_module; - INIT_LIST_HEAD(&dp->port_list); - mutex_init(&dp->mutex); - dp->dp_idx = dp_idx; for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_head_init(&dp->queues[i]); - init_waitqueue_head(&dp->waitqueue); - - /* Initialize kobject for bridge. This will be added as - * /sys/class/net//brif later, if sysfs is enabled. */ - dp->ifobj.kset = NULL; - kobject_init(&dp->ifobj, &dp_ktype); - - /* Allocate table. */ - err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(0)); - if (!dp->table) - goto err_free_dp; - - /* Set up our datapath device. */ - BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname)); - strcpy(internal_dev_port.devname, devname); - strcpy(internal_dev_port.type, "internal"); - err = new_vport(dp, &internal_dev_port, ODPP_LOCAL); - if (err) { - if (err == -EBUSY) - err = -EEXIST; - - goto err_destroy_table; - } - - dp->drop_frags = 0; - dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) - goto err_destroy_local_port; - - rcu_assign_pointer(dps[dp_idx], dp); - mutex_unlock(&dp_mutex); - rtnl_unlock(); - - dp_sysfs_add_dp(dp); - - return 0; + skb_queue_purge(&dp->queues[i]); -err_destroy_local_port: - dp_detach_port(dp->ports[ODPP_LOCAL]); -err_destroy_table: - tbl_destroy(dp->table, NULL); -err_free_dp: - kfree(dp); -err_put_module: - module_put(THIS_MODULE); -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); -err: - return err; + tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl); + free_percpu(dp->stats_percpu); + kobject_put(&dp->ifobj); } -static void do_destroy_dp(struct datapath *dp) +/* Caller must hold RTNL, dp_mutex, and dp->mutex. */ +static void destroy_dp(struct datapath *dp) { struct vport *p, *n; - int i; list_for_each_entry_safe (p, n, &dp->port_list, node) if (p->port_no != ODPP_LOCAL) dp_detach_port(p); dp_sysfs_del_dp(dp); - rcu_assign_pointer(dps[dp->dp_idx], NULL); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); - dp_detach_port(dp->ports[ODPP_LOCAL]); - - tbl_destroy(dp->table, flow_free_tbl); - - for (i = 0; i < DP_N_QUEUES; i++) - skb_queue_purge(&dp->queues[i]); - free_percpu(dp->stats_percpu); - kobject_put(&dp->ifobj); + mutex_unlock(&dp->mutex); + call_rcu(&dp->rcu, destroy_dp_rcu); module_put(THIS_MODULE); } -static int destroy_dp(int dp_idx) -{ - struct datapath *dp; - int err; - - rtnl_lock(); - mutex_lock(&dp_mutex); - dp = get_dp(dp_idx); - err = -ENODEV; - if (!dp) - goto err_unlock; - - do_destroy_dp(dp); - err = 0; - -err_unlock: - mutex_unlock(&dp_mutex); - rtnl_unlock(); - return err; -} - -/* Called with RTNL lock and dp_mutex. */ -static int new_vport(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_parms parms; struct vport *vport; - int err; - - parms.name = odp_port->devname; - parms.type = odp_port->type; - parms.config = odp_port->config; - parms.dp = dp; - parms.port_no = port_no; vport_lock(); - vport = vport_add(&parms); - vport_unlock(); + vport = vport_add(parms); + if (!IS_ERR(vport)) { + struct datapath *dp = parms->dp; - if (IS_ERR(vport)) - return PTR_ERR(vport); + rcu_assign_pointer(dp->ports[parms->port_no], vport); + list_add_rcu(&vport->node, &dp->port_list); - err = vport_attach(vport); - if (err) { - vport_del(vport); - return err; + dp_ifinfo_notify(RTM_NEWLINK, vport); } + vport_unlock(); - rcu_assign_pointer(dp->ports[port_no], vport); - list_add_rcu(&vport->node, &dp->port_list); - dp->n_ports++; - - dp_ifinfo_notify(RTM_NEWLINK, vport); - - return 0; -} - -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'; - port.type[VPORT_TYPE_SIZE - 1] = '\0'; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - for (port_no = 1; port_no < DP_MAX_PORTS; port_no++) - if (!dp->ports[port_no]) - goto got_port_no; - err = -EFBIG; - goto out_unlock_dp; - -got_port_no: - err = new_vport(dp, &port, port_no); - if (err) - goto out_unlock_dp; - - set_internal_devs_mtu(dp); - dp_sysfs_add_if(dp->ports[port_no]); - - 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 vport *p) @@ -422,52 +267,14 @@ int dp_detach_port(struct vport *p) dp_ifinfo_notify(RTM_DELLINK, p); /* First drop references to device. */ - p->dp->n_ports--; list_del_rcu(&p->node); rcu_assign_pointer(p->dp->ports[p->port_no], NULL); - err = vport_detach(p); - if (err) - return err; - - /* Then wait until no one is still using it, and destroy it. */ - synchronize_rcu(); - + /* Then destroy it. */ vport_lock(); - vport_del(p); + err = vport_del(p); vport_unlock(); - return 0; -} - -static int detach_port(int dp_idx, int port_no) -{ - struct vport *p; - struct datapath *dp; - int err; - - err = -EINVAL; - if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) - goto out; - - rtnl_lock(); - dp = get_dp_locked(dp_idx); - err = -ENODEV; - if (!dp) - goto out_unlock_rtnl; - - p = dp->ports[port_no]; - err = -ENOENT; - if (!p) - goto out_unlock_dp; - - err = dp_detach_port(p); - -out_unlock_dp: - mutex_unlock(&dp->mutex); -out_unlock_rtnl: - rtnl_unlock(); -out: return err; } @@ -484,12 +291,12 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) OVS_CB(skb)->vport = p; if (!OVS_CB(skb)->flow) { - struct odp_flow_key key; + struct sw_flow_key key; struct tbl_node *flow_node; bool is_frag; /* Extract flow from 'skb' into 'key'. */ - error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag); + error = flow_extract(skb, p->port_no, &key, &is_frag); if (unlikely(error)) { kfree_skb(skb); return; @@ -505,7 +312,15 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); if (unlikely(!flow_node)) { - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + 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; } @@ -513,6 +328,7 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) OVS_CB(skb)->flow = flow_cast(flow_node); } + stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); flow_used(OVS_CB(skb)->flow, skb); acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); @@ -523,13 +339,13 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) loop->looping = true; if (unlikely(loop->looping)) { loop_suppress(dp, acts); + kfree_skb(skb); goto out_loop; } /* Execute actions. */ execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions, - acts->n_actions); - stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + acts->actions_len); /* Check whether sub-actions looped too much. */ if (unlikely(loop->looping)) @@ -553,175 +369,26 @@ out: local_bh_enable(); } -#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 (!skb->proto_csum_blank) - return 0; - - if (skb->protocol != htons(ETH_P_IP)) - goto out; - - if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data)) - goto out; - - iph = ip_hdr(skb); - th = skb_network_header(skb) + 4 * iph->ihl; - - 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()) - pr_err("Attempting to checksum a non-TCP/UDP packet, " - "dropping a protocol %d packet", - iph->protocol); - goto out; - } - - if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data)) - goto out; - - skb->ip_summed = CHECKSUM_PARTIAL; - skb->proto_csum_blank = 0; - -#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 - - err = 0; - -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) +static void copy_and_csum_skb(struct sk_buff *skb, void *to) { - /* 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; - - 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: - pr_err("unknown checksum type %d\n", skb->ip_summed); - /* None seems the safest... */ - OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - } + u16 csum_start, csum_offset; + __wsum csum; -#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; + get_skb_csum_pointers(skb, &csum_start, &csum_offset); + csum_start -= skb_headroom(skb); + BUG_ON(csum_start >= skb_headlen(skb)); - WARN_ON_ONCE(skb->proto_csum_blank && xmit); -#endif -} + skb_copy_bits(skb, 0, to, csum_start); -/* 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) -{ -#ifdef CHECKSUM_HW - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) - skb->ip_summed = CHECKSUM_NONE; -#endif + 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; @@ -733,23 +400,61 @@ static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue 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; - err = skb_cow(skb, sizeof *header); - 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; - 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 = (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); + } + + 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; @@ -763,16 +468,16 @@ 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; @@ -787,20 +492,16 @@ int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ if (skb_is_gso(skb)) { struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); - 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. */ + + 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; @@ -819,30 +520,68 @@ err: return err; } -static int flush_flows(struct datapath *dp) +static int flush_flows(int dp_idx) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table; struct tbl *new_table; + struct datapath *dp; + int err; + + dp = get_dp_locked(dp_idx); + err = -ENODEV; + if (!dp) + goto exit; - new_table = tbl_create(0); + old_table = get_table_protected(dp); + new_table = tbl_create(TBL_MIN_BUCKETS); + err = -ENOMEM; if (!new_table) - return -ENOMEM; + goto exit_unlock; rcu_assign_pointer(dp->table, new_table); tbl_deferred_destroy(old_table, flow_free_tbl); - return 0; + err = 0; + +exit_unlock: + mutex_unlock(&dp->mutex); +exit: + return err; } -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 (type) { + case ODPAT_UNSPEC: + return -EINVAL; - switch (a->type) { case ODPAT_CONTROLLER: case ODPAT_STRIP_VLAN: case ODPAT_SET_DL_SRC: @@ -859,17 +598,17 @@ static int validate_actions(const struct sw_flow_actions *actions) break; case ODPAT_OUTPUT: - if (a->output.port >= DP_MAX_PORTS) + if (nla_get_u32(a) >= DP_MAX_PORTS) return -EINVAL; break; case ODPAT_SET_DL_TCI: - if (a->dl_tci.tci & htons(VLAN_CFI_MASK)) + 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; @@ -878,57 +617,31 @@ static int validate_actions(const struct sw_flow_actions *actions) } } + if (rem > 0) + return -EINVAL; + return 0; } -static struct sw_flow_actions *get_actions(const struct odp_flow *flow) +struct dp_flowcmd { + u32 nlmsg_flags; + u32 dp_idx; + u32 total_len; + struct sw_flow_key key; + const struct nlattr *actions; + u32 actions_len; + bool clear; + u64 state; +}; + +static struct sw_flow_actions *get_actions(const struct dp_flowcmd *flowcmd) { struct sw_flow_actions *actions; - int error; - - actions = flow_actions_alloc(flow->n_actions); - 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))) - goto error_free_actions; - error = validate_actions(actions); - if (error) - goto error_free_actions; + actions = flow_actions_alloc(flowcmd->actions_len); + if (!IS_ERR(actions) && flowcmd->actions_len) + memcpy(actions->actions, flowcmd->actions, flowcmd->actions_len); return actions; - -error_free_actions: - kfree(actions); -error: - return ERR_PTR(error); -} - -static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) -{ - 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->reserved = 0; - stats->tcp_flags = flow->tcp_flags; - stats->error = 0; } static void clear_stats(struct sw_flow *flow) @@ -941,7 +654,7 @@ static void clear_stats(struct sw_flow *flow) 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); @@ -951,382 +664,102 @@ static int expand_table(struct datapath *dp) rcu_assign_pointer(dp->table, new_table); tbl_deferred_destroy(old_table, NULL); - return 0; + 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 *flow; - struct tbl *table; - int error; +static const struct nla_policy execute_policy[ODP_PACKET_ATTR_MAX + 1] = { + [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC }, + [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, +}; - table = rcu_dereference(dp->table); - flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); - if (!flow_node) { - /* No such flow. */ - struct sw_flow_actions *acts; - - 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 = rcu_dereference(dp->table); - } - - /* Allocate flow. */ - flow = flow_alloc(); - if (IS_ERR(flow)) { - error = PTR_ERR(flow); - goto error; - } - flow->key = uf->flow.key; - clear_stats(flow); - - /* Obtain actions. */ - 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. */ - error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key)); - if (error) - goto error_free_flow_acts; - - memset(stats, 0, sizeof(struct odp_flow_stats)); - } else { - /* We found a matching flow. */ - struct sw_flow_actions *old_acts, *new_acts; - - flow = flow_cast(flow_node); - - /* Bail out if we're not allowed to modify an existing flow. */ - error = -EEXIST; - if (!(uf->flags & ODPPF_MODIFY)) - goto error; - - /* Swap actions. */ - new_acts = get_actions(&uf->flow); - error = PTR_ERR(new_acts); - if (IS_ERR(new_acts)) - goto error; - old_acts = rcu_dereference(flow->sf_acts); - if (old_acts->n_actions != new_acts->n_actions || - memcmp(old_acts->actions, new_acts->actions, - sizeof(union odp_action) * old_acts->n_actions)) { - rcu_assign_pointer(flow->sf_acts, new_acts); - flow_deferred_free_acts(old_acts); - } else { - kfree(new_acts); - } - - /* Fetch stats, then clear them if necessary. */ - spin_lock_bh(&flow->lock); - get_stats(flow, stats); - if (uf->flags & ODPPF_ZERO_STATS) - clear_stats(flow); - spin_unlock_bh(&flow->lock); - } - - return 0; - -error_free_flow_acts: - kfree(flow->sf_acts); -error_free_flow: - flow->sf_acts = NULL; - flow_put(flow); -error: - return error; -} - -static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) -{ - struct odp_flow_stats stats; - struct odp_flow_put uf; - int error; - - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) - return -EFAULT; - - error = do_put_flow(dp, &uf, &stats); - if (error) - return error; - - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - return -EFAULT; - - return 0; -} - -static int do_answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow_stats __user *ustats, - union odp_action __user *actions, - u32 __user *n_actionsp) -{ - struct sw_flow_actions *sf_acts; - struct odp_flow_stats stats; - u32 n_actions; - - spin_lock_bh(&flow->lock); - get_stats(flow, &stats); - if (query_flags & ODPFF_ZERO_TCP_FLAGS) - flow->tcp_flags = 0; - - spin_unlock_bh(&flow->lock); - - if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(n_actions, n_actionsp)) - return -EFAULT; - - if (!n_actions) - return 0; - - sf_acts = rcu_dereference(flow->sf_acts); - if (put_user(sf_acts->n_actions, n_actionsp) || - (actions && copy_to_user(actions, sf_acts->actions, - sizeof(union odp_action) * - min(sf_acts->n_actions, n_actions)))) - return -EFAULT; - - return 0; -} - -static int answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow __user *ufp) -{ - union odp_action *actions; - - if (get_user(actions, &ufp->actions)) - return -EFAULT; - - return do_answer_query(flow, query_flags, - &ufp->stats, actions, &ufp->n_actions); -} - -static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key) -{ - struct tbl *table = rcu_dereference(dp->table); - struct tbl_node *flow_node; - int error; - - flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp); - if (!flow_node) - return ERR_PTR(-ENOENT); - - error = tbl_remove(table, flow_node); - if (error) - return ERR_PTR(error); - - /* XXX Returned flow_node's statistics might lose a few packets, since - * other CPUs can be using this flow. We used to synchronize_rcu() to - * make sure that we get completely accurate stats, but that blows our - * performance, badly. */ - return flow_cast(flow_node); -} - -static 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, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); - - error = answer_query(flow, 0, ufp); - flow_deferred_free(flow); - return error; -} - -static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) -{ - struct tbl *table = rcu_dereference(dp->table); - u32 i; - - for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = &flowvec->flows[i]; - struct odp_flow uf; - struct tbl_node *flow_node; - int error; - - if (copy_from_user(&uf, ufp, sizeof uf)) - return -EFAULT; - - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = answer_query(flow_cast(flow_node), uf.flags, ufp); - if (error) - return -EFAULT; - } - 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 *)) +static int execute_packet(const struct odp_packet __user *uodp_packet) { - struct odp_flowvec __user *uflowvec; - struct odp_flowvec flowvec; - int retval; + struct nlattr *a[ODP_PACKET_ATTR_MAX + 1]; + struct odp_packet *odp_packet; + struct sk_buff *skb, *packet; + unsigned int actions_len; + struct nlattr *actions; + struct sw_flow_key key; + struct datapath *dp; + struct ethhdr *eth; + bool is_frag; + u32 len; + int err; - uflowvec = (struct odp_flowvec __user *)argp; - if (copy_from_user(&flowvec, uflowvec, sizeof flowvec)) + if (get_user(len, &uodp_packet->len)) return -EFAULT; - - if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) + if (len < sizeof(struct odp_packet)) return -EINVAL; - retval = function(dp, &flowvec); - return (retval < 0 ? retval - : retval == flowvec.n_flows ? 0 - : put_user(retval, &uflowvec->n_flows)); -} + skb = alloc_skb(len, GFP_KERNEL); + if (!skb) + return -ENOMEM; -static int do_execute(struct datapath *dp, const struct odp_execute *execute) -{ - struct odp_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(__skb_put(skb, len), uodp_packet, len)) + goto exit_free_skb; + odp_packet = (struct odp_packet *)skb->data; err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) - goto error; + if (odp_packet->len != len) + goto exit_free_skb; - actions = flow_actions_alloc(execute->n_actions); - if (IS_ERR(actions)) { - err = PTR_ERR(actions); - goto error; - } + __skb_pull(skb, sizeof(struct odp_packet)); + err = nla_parse(a, ODP_PACKET_ATTR_MAX, (struct nlattr *)skb->data, + skb->len, execute_policy); + if (err) + goto exit_free_skb; - err = -EFAULT; - if (copy_from_user(actions->actions, execute->actions, - execute->n_actions * sizeof *execute->actions)) - goto error_free_actions; + err = -EINVAL; + if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] || + nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN) + goto exit_free_skb; - err = validate_actions(actions); + actions = nla_data(a[ODP_PACKET_ATTR_ACTIONS]); + actions_len = nla_len(a[ODP_PACKET_ATTR_ACTIONS]); + err = validate_actions(actions, actions_len); if (err) - goto error_free_actions; + goto exit_free_skb; + packet = skb_clone(skb, GFP_KERNEL); err = -ENOMEM; - skb = alloc_skb(execute->length, GFP_KERNEL); - if (!skb) - goto error_free_actions; - - err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), execute->data, - execute->length)) - goto error_free_skb; + if (!packet) + goto exit_free_skb; + packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]); + packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]); - skb_reset_mac_header(skb); - eth = eth_hdr(skb); + skb_reset_mac_header(packet); + eth = eth_hdr(packet); /* Normally, setting the skb 'protocol' field would be handled by a * call to eth_type_trans(), but it assumes there's a sending * device, which we may not have. */ if (ntohs(eth->h_proto) >= 1536) - skb->protocol = eth->h_proto; + packet->protocol = eth->h_proto; else - skb->protocol = htons(ETH_P_802_2); + packet->protocol = htons(ETH_P_802_2); - err = flow_extract(skb, -1, &key, &is_frag); + err = flow_extract(packet, -1, &key, &is_frag); if (err) - goto error_free_skb; + goto exit_free_skb; rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions); + dp = get_dp(odp_packet->dp_idx); + err = -ENODEV; + if (dp) + err = execute_actions(dp, packet, &key, actions, actions_len); rcu_read_unlock(); - kfree(actions); - return err; - -error_free_skb: +exit_free_skb: kfree_skb(skb); -error_free_actions: - kfree(actions); -error: return err; } -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) +static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) { - struct tbl *table = rcu_dereference(dp->table); - struct odp_stats stats; int i; - stats.n_flows = tbl_count(table); - stats.cur_capacity = tbl_n_buckets(table); - stats.max_capacity = TBL_MAX_BUCKETS; - stats.n_ports = dp->n_ports; - stats.max_ports = DP_MAX_PORTS; - stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; + stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0; for_each_possible_cpu(i) { const struct dp_stats_percpu *percpu_stats; struct dp_stats_percpu local_stats; @@ -1339,14 +772,11 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) 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->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; } /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ @@ -1390,106 +820,1151 @@ void set_internal_devs_mtu(const struct datapath *dp) } } -static int put_port(const struct vport *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 flow_policy[ODP_FLOW_ATTR_MAX + 1] = { + [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, + [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, + [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, + [ODP_FLOW_ATTR_STATE] = { .type = NLA_U64 }, +}; + +static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp, + struct sw_flow *flow, u32 total_len, u64 state) { - struct odp_port op; + const struct sw_flow_actions *sf_acts; + struct odp_flow_stats stats; + struct odp_flow *odp_flow; + struct sk_buff *skb; + struct nlattr *nla; + unsigned long used; + u8 tcp_flags; + int err; - memset(&op, 0, sizeof op); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); + + skb = alloc_skb(128 + FLOW_BUFSIZE + sf_acts->actions_len, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; rcu_read_lock(); - strncpy(op.devname, vport_get_name(p), sizeof op.devname); - strncpy(op.type, vport_get_type(p), sizeof op.type); + odp_flow = (struct odp_flow*)__skb_put(skb, sizeof(struct odp_flow)); + odp_flow->dp_idx = dp->dp_idx; + odp_flow->total_len = total_len; + + nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY); + if (!nla) + goto nla_put_failure; + err = flow_to_nlattrs(&flow->key, skb); + if (err) + goto exit_unlock; + nla_nest_end(skb, nla); + + nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS); + if (!nla || skb_tailroom(skb) < sf_acts->actions_len) + goto nla_put_failure; + memcpy(__skb_put(skb, sf_acts->actions_len), sf_acts->actions, sf_acts->actions_len); + nla_nest_end(skb, nla); + + spin_lock_bh(&flow->lock); + used = flow->used; + stats.n_packets = flow->packet_count; + stats.n_bytes = flow->byte_count; + tcp_flags = flow->tcp_flags; + spin_unlock_bh(&flow->lock); + + if (used) + NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, used); + + if (stats.n_packets) + NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats); + + if (tcp_flags) + NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags); + + if (state) + NLA_PUT_U64(skb, ODP_FLOW_ATTR_STATE, state); + + if (skb->len > total_len) + goto nla_put_failure; + + odp_flow->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_flow_from_user(struct odp_flow __user *uodp_flow, + struct dp_flowcmd *flowcmd) +{ + struct nlattr *a[ODP_FLOW_ATTR_MAX + 1]; + struct odp_flow *odp_flow; + struct sk_buff *skb; + u32 len; + int err; + + if (get_user(len, &uodp_flow->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_flow)) + 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_flow, len)) + goto error_free_skb; + + odp_flow = (struct odp_flow *)skb->data; + err = -EINVAL; + if (odp_flow->len != len) + goto error_free_skb; + + flowcmd->nlmsg_flags = odp_flow->nlmsg_flags; + flowcmd->dp_idx = odp_flow->dp_idx; + flowcmd->total_len = odp_flow->total_len; + + err = nla_parse(a, ODP_FLOW_ATTR_MAX, + (struct nlattr *)(skb->data + sizeof(struct odp_flow)), + skb->len - sizeof(struct odp_flow), flow_policy); + if (err) + goto error_free_skb; + + /* ODP_FLOW_ATTR_KEY. */ + if (a[ODP_FLOW_ATTR_KEY]) { + err = flow_from_nlattrs(&flowcmd->key, a[ODP_FLOW_ATTR_KEY]); + if (err) + goto error_free_skb; + } else + memset(&flowcmd->key, 0, sizeof(struct sw_flow_key)); + + /* ODP_FLOW_ATTR_ACTIONS. */ + if (a[ODP_FLOW_ATTR_ACTIONS]) { + flowcmd->actions = nla_data(a[ODP_FLOW_ATTR_ACTIONS]); + flowcmd->actions_len = nla_len(a[ODP_FLOW_ATTR_ACTIONS]); + err = validate_actions(flowcmd->actions, flowcmd->actions_len); + if (err) + goto error_free_skb; + } else { + flowcmd->actions = NULL; + flowcmd->actions_len = 0; + } + + flowcmd->clear = a[ODP_FLOW_ATTR_CLEAR] != NULL; - op.port = p->port_no; + flowcmd->state = a[ODP_FLOW_ATTR_STATE] ? nla_get_u64(a[ODP_FLOW_ATTR_STATE]) : 0; - return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; + return skb; + +error_free_skb: + kfree_skb(skb); + return ERR_PTR(err); } -static int query_port(struct datapath *dp, struct odp_port __user *uport) +static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) { - struct odp_port port; + struct tbl_node *flow_node; + struct dp_flowcmd flowcmd; + struct sw_flow *flow; + struct sk_buff *skb; + struct datapath *dp; + struct tbl *table; + u32 hash; + int error; - if (copy_from_user(&port, uport, sizeof port)) - return -EFAULT; + skb = copy_flow_from_user(uodp_flow, &flowcmd); + error = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - if (port.devname[0]) { - struct vport *vport; - int err = 0; + dp = get_dp_locked(flowcmd.dp_idx); + error = -ENODEV; + if (!dp) + goto error_kfree_skb; - port.devname[IFNAMSIZ - 1] = '\0'; + hash = flow_hash(&flowcmd.key); + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &flowcmd.key, hash, flow_cmp); + if (!flow_node) { + struct sw_flow_actions *acts; - vport_lock(); - rcu_read_lock(); + /* Bail out if we're not allowed to create a new flow. */ + error = -ENOENT; + if (cmd == ODP_FLOW_SET) + goto error_unlock_dp; - vport = vport_locate(port.devname); - if (!vport) { - err = -ENODEV; - goto error_unlock; - } - if (vport->dp != dp) { - err = -ENOENT; - goto error_unlock; + /* Expand table, if necessary, to make room. */ + if (tbl_count(table) >= tbl_n_buckets(table)) { + error = expand_table(dp); + if (error) + goto error_unlock_dp; + table = get_table_protected(dp); } - port.port = vport->port_no; + /* Allocate flow. */ + flow = flow_alloc(); + if (IS_ERR(flow)) { + error = PTR_ERR(flow); + goto error_unlock_dp; + } + flow->key = flowcmd.key; + clear_stats(flow); + + /* Obtain actions. */ + acts = get_actions(&flowcmd); + error = PTR_ERR(acts); + if (IS_ERR(acts)) + goto error_free_flow; + rcu_assign_pointer(flow->sf_acts, acts); + + error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); + if (error) + goto error_free_flow; + + /* Put flow in bucket. */ + error = tbl_insert(table, &flow->tbl_node, hash); + if (error) + goto error_free_flow; + } else { + /* We found a matching flow. */ + struct sw_flow_actions *old_acts; + + /* Bail out if we're not allowed to modify an existing flow. + * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL + * because Generic Netlink treats the latter as a dump + * request. We also accept NLM_F_EXCL in case that bug ever + * gets fixed. + */ + error = -EEXIST; + if (flowcmd.nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) + goto error_kfree_skb; + + /* Update actions. */ + flow = flow_cast(flow_node); + old_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); + if (flowcmd.actions && + (old_acts->actions_len != flowcmd.actions_len || + memcmp(old_acts->actions, flowcmd.actions, + flowcmd.actions_len))) { + struct sw_flow_actions *new_acts; + + new_acts = get_actions(&flowcmd); + error = PTR_ERR(new_acts); + if (IS_ERR(new_acts)) + goto error_kfree_skb; + + rcu_assign_pointer(flow->sf_acts, new_acts); + flow_deferred_free_acts(old_acts); + } + + error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); + if (error) + goto error_kfree_skb; + + /* Clear stats. */ + if (flowcmd.clear) { + spin_lock_bh(&flow->lock); + clear_stats(flow); + spin_unlock_bh(&flow->lock); + } + } + kfree_skb(skb); + mutex_unlock(&dp->mutex); + return 0; + +error_free_flow: + flow_put(flow); +error_unlock_dp: + mutex_unlock(&dp->mutex); +error_kfree_skb: + kfree_skb(skb); +exit: + return error; +} + +static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) +{ + struct tbl_node *flow_node; + struct dp_flowcmd flowcmd; + struct sw_flow *flow; + struct sk_buff *skb; + struct datapath *dp; + struct tbl *table; + int err; + + skb = copy_flow_from_user(uodp_flow, &flowcmd); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + + dp = get_dp_locked(flowcmd.dp_idx); + err = -ENODEV; + if (!dp) + goto exit_kfree_skb; + + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &flowcmd.key, flow_hash(&flowcmd.key), flow_cmp); + err = -ENOENT; + if (!flow_node) + goto exit_unlock_dp; + + if (cmd == ODP_FLOW_DEL) { + err = tbl_remove(table, flow_node); + if (err) + goto exit_unlock_dp; + } + + flow = flow_cast(flow_node); + err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0); + if (!err && cmd == ODP_FLOW_DEL) + flow_deferred_free(flow); + +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_kfree_skb: + kfree_skb(skb); +exit: + return err; +} + +static int dump_flow(struct odp_flow __user *uodp_flow) +{ + struct tbl_node *flow_node; + struct dp_flowcmd flowcmd; + struct sw_flow *flow; + struct sk_buff *skb; + struct datapath *dp; + u32 bucket, obj; + int err; + + skb = copy_flow_from_user(uodp_flow, &flowcmd); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + + dp = get_dp_locked(flowcmd.dp_idx); + err = -ENODEV; + if (!dp) + goto exit_free; + + bucket = flowcmd.state >> 32; + obj = flowcmd.state; + flow_node = tbl_next(dp->table, &bucket, &obj); + err = -ENODEV; + if (!flow_node) + goto exit_unlock_dp; + + flow = flow_cast(flow_node); + err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, + ((u64)bucket << 32) | obj); + +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); +exit: + return err; +} + +static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = { + [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, + [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 }, + [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 }, +}; + +static int copy_datapath_to_user(void __user *dst, struct datapath *dp, uint32_t total_len) +{ + struct odp_datapath *odp_datapath; + struct sk_buff *skb; + struct nlattr *nla; + int err; + + skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; + + odp_datapath = (struct odp_datapath*)__skb_put(skb, sizeof(struct odp_datapath)); + odp_datapath->dp_idx = dp->dp_idx; + odp_datapath->total_len = total_len; + + rcu_read_lock(); + err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp)); + rcu_read_unlock(); + if (err) + goto nla_put_failure; + + nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats)); + if (!nla) + goto nla_put_failure; + get_dp_stats(dp, nla_data(nla)); + + NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS, + dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO); + + if (dp->sflow_probability) + NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability); + + if (skb->len > total_len) + goto nla_put_failure; + + odp_datapath->len = skb->len; + err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0; + goto exit_free_skb; + +nla_put_failure: + err = -EMSGSIZE; +exit_free_skb: + kfree_skb(skb); +exit: + return err; +} + +static struct sk_buff *copy_datapath_from_user(struct odp_datapath __user *uodp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ + struct odp_datapath *odp_datapath; + struct sk_buff *skb; + u32 len; + int err; + + if (get_user(len, &uodp_datapath->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_datapath)) + return ERR_PTR(-EINVAL); + + skb = alloc_skb(len, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); + + err = -EFAULT; + if (copy_from_user(__skb_put(skb, len), uodp_datapath, len)) + goto error_free_skb; + + odp_datapath = (struct odp_datapath *)skb->data; + err = -EINVAL; + if (odp_datapath->len != len) + goto error_free_skb; + + err = nla_parse(a, ODP_DP_ATTR_MAX, + (struct nlattr *)(skb->data + sizeof(struct odp_datapath)), + skb->len - sizeof(struct odp_datapath), datapath_policy); + if (err) + goto error_free_skb; + + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); + + err = -EINVAL; + if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP) + goto error_free_skb; + } + + err = VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1); + if (err) + goto error_free_skb; + + return skb; + +error_free_skb: + kfree_skb(skb); + return ERR_PTR(err); +} + +/* Called with dp_mutex and optionally with RTNL lock also. + * Holds the returned datapath's mutex on return. + */ +static struct datapath *lookup_datapath(struct odp_datapath *odp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ + WARN_ON_ONCE(!mutex_is_locked(&dp_mutex)); + + if (!a[ODP_DP_ATTR_NAME]) { + struct datapath *dp; + + dp = get_dp(odp_datapath->dp_idx); + if (!dp) + return ERR_PTR(-ENODEV); + mutex_lock(&dp->mutex); + return dp; + } else { + struct datapath *dp; + struct vport *vport; + int dp_idx; + + vport_lock(); + vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME])); + dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1; + vport_unlock(); + + if (dp_idx < 0) + return ERR_PTR(-ENODEV); + + dp = get_dp(dp_idx); + mutex_lock(&dp->mutex); + return dp; + } +} + +static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ + if (a[ODP_DP_ATTR_IPV4_FRAGS]) + dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP; + if (a[ODP_DP_ATTR_SAMPLING]) + dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]); +} + +static int new_datapath(struct odp_datapath __user *uodp_datapath) +{ + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct odp_datapath *odp_datapath; + struct vport_parms parms; + struct sk_buff *skb; + struct datapath *dp; + struct vport *vport; + int dp_idx; + int err; + int i; + + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto err; + odp_datapath = (struct odp_datapath *)skb->data; + + err = -EINVAL; + if (!a[ODP_DP_ATTR_NAME]) + goto err_free_skb; + + rtnl_lock(); + mutex_lock(&dp_mutex); + err = -ENODEV; + if (!try_module_get(THIS_MODULE)) + goto err_unlock_dp_mutex; + + dp_idx = odp_datapath->dp_idx; + if (dp_idx < 0) { + err = -EFBIG; + for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) { + if (get_dp(dp_idx)) + continue; + err = 0; + break; + } + } else if (dp_idx < ARRAY_SIZE(dps)) + err = get_dp(dp_idx) ? -EBUSY : 0; + else + err = -EINVAL; + if (err) + goto err_put_module; + + err = -ENOMEM; + dp = kzalloc(sizeof(*dp), GFP_KERNEL); + if (dp == NULL) + goto err_put_module; + INIT_LIST_HEAD(&dp->port_list); + mutex_init(&dp->mutex); + mutex_lock(&dp->mutex); + dp->dp_idx = dp_idx; + for (i = 0; i < DP_N_QUEUES; i++) + skb_queue_head_init(&dp->queues[i]); + init_waitqueue_head(&dp->waitqueue); + + /* Initialize kobject for bridge. This will be added as + * /sys/class/net//brif later, if sysfs is enabled. */ + dp->ifobj.kset = NULL; + kobject_init(&dp->ifobj, &dp_ktype); + + /* Allocate table. */ + err = -ENOMEM; + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); + if (!dp->table) + goto err_free_dp; + + /* Set up our datapath device. */ + parms.name = nla_data(a[ODP_DP_ATTR_NAME]); + parms.type = ODP_VPORT_TYPE_INTERNAL; + parms.options = NULL; + parms.dp = dp; + parms.port_no = ODPP_LOCAL; + vport = new_vport(&parms); + if (IS_ERR(vport)) { + err = PTR_ERR(vport); + if (err == -EBUSY) + err = -EEXIST; + + goto err_destroy_table; + } + + dp->drop_frags = 0; + dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); + if (!dp->stats_percpu) { + err = -ENOMEM; + goto err_destroy_local_port; + } + + change_datapath(dp, a); + + rcu_assign_pointer(dps[dp_idx], dp); + dp_sysfs_add_dp(dp); + + mutex_unlock(&dp->mutex); + mutex_unlock(&dp_mutex); + rtnl_unlock(); + + return 0; + +err_destroy_local_port: + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); +err_destroy_table: + tbl_destroy(get_table_protected(dp), NULL); +err_free_dp: + mutex_unlock(&dp->mutex); + kfree(dp); +err_put_module: + module_put(THIS_MODULE); +err_unlock_dp_mutex: + mutex_unlock(&dp_mutex); + rtnl_unlock(); +err_free_skb: + kfree_skb(skb); +err: + return err; +} + +static int del_datapath(struct odp_datapath __user *uodp_datapath) +{ + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct datapath *dp; + struct sk_buff *skb; + int err; + + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + + rtnl_lock(); + mutex_lock(&dp_mutex); + dp = lookup_datapath((struct odp_datapath *)skb->data, a); + err = PTR_ERR(dp); + if (IS_ERR(dp)) + goto exit_free; + + destroy_dp(dp); + err = 0; + +exit_free: + kfree_skb(skb); + mutex_unlock(&dp_mutex); + rtnl_unlock(); +exit: + return err; +} + +static int set_datapath(struct odp_datapath __user *uodp_datapath) +{ + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct datapath *dp; + struct sk_buff *skb; + int err; + + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + + mutex_lock(&dp_mutex); + dp = lookup_datapath((struct odp_datapath *)skb->data, a); + err = PTR_ERR(dp); + if (IS_ERR(dp)) + goto exit_free; + + change_datapath(dp, a); + mutex_unlock(&dp->mutex); + err = 0; + +exit_free: + kfree_skb(skb); + mutex_unlock(&dp_mutex); +exit: + return err; +} + +static int get_datapath(struct odp_datapath __user *uodp_datapath) +{ + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct odp_datapath *odp_datapath; + struct datapath *dp; + struct sk_buff *skb; + int err; + + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_datapath = (struct odp_datapath *)skb->data; + + mutex_lock(&dp_mutex); + dp = lookup_datapath(odp_datapath, a); + mutex_unlock(&dp_mutex); + + err = PTR_ERR(dp); + if (IS_ERR(dp)) + goto exit_free; + + err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len); + mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); +exit: + return err; +} + +static int dump_datapath(struct odp_datapath __user *uodp_datapath) +{ + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; + struct odp_datapath *odp_datapath; + struct sk_buff *skb; + u32 dp_idx; + int err; + + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; + odp_datapath = (struct odp_datapath *)skb->data; + + mutex_lock(&dp_mutex); + for (dp_idx = odp_datapath->dp_idx; dp_idx < ARRAY_SIZE(dps); dp_idx++) { + struct datapath *dp = get_dp(dp_idx); + if (!dp) + continue; + + mutex_lock(&dp->mutex); + mutex_unlock(&dp_mutex); + err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len); + mutex_unlock(&dp->mutex); + goto exit_free; + } + mutex_unlock(&dp_mutex); + err = -ENODEV; + +exit_free: + kfree_skb(skb); +exit: + return err; +} + +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); + } + dp_idx = vport->dp->dp_idx; + port_no = vport->port_no; + vport_unlock(); + + dp = get_dp_locked(dp_idx); + if (!dp) + goto retry; + + vport = get_vport_protected(dp, port_no); + if (!vport || + strcmp(vport_get_name(vport), nla_data(a[ODP_VPORT_ATTR_NAME]))) { + mutex_unlock(&dp->mutex); + goto retry; + } + + return vport; + } else if (a[ODP_VPORT_ATTR_PORT_NO]) { + u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); + + if (port_no >= DP_MAX_PORTS) + return ERR_PTR(-EINVAL); + + 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 change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]) +{ + 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 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]); + + 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; + } + } + + 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; +} + +static int set_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) +{ + 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; -error_unlock: - rcu_read_unlock(); - vport_unlock(); + rtnl_lock(); + vport = lookup_vport((struct odp_vport *)skb->data, a); + err = PTR_ERR(vport); + if (IS_ERR(vport)) + goto exit_free; - if (err) - return err; - } else { - if (port.port >= DP_MAX_PORTS) - return -EINVAL; - if (!dp->ports[port.port]) - return -ENOENT; - } + 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 put_port(dp->ports[port.port], uport); + mutex_unlock(&vport->dp->mutex); +exit_free: + kfree_skb(skb); + rtnl_unlock(); +exit: + return err; } -static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, - int n_ports) +static int del_vport(unsigned int cmd, struct odp_vport __user *uodp_vport) { - int idx = 0; - if (n_ports) { - struct vport *p; - - list_for_each_entry_rcu (p, &dp->port_list, node) { - if (put_port(p, &uports[idx])) - return -EFAULT; - if (idx++ >= n_ports) - break; - } - } - return idx; -} + struct nlattr *a[ODP_VPORT_ATTR_MAX + 1]; + struct datapath *dp; + struct vport *vport; + struct sk_buff *skb; + int err; -static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) -{ - struct odp_portvec pv; - int retval; + skb = copy_vport_from_user(uodp_vport, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - if (copy_from_user(&pv, upv, sizeof pv)) - return -EFAULT; + 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; - retval = do_list_ports(dp, pv.ports, pv.n_ports); - if (retval < 0) - return retval; + err = -EINVAL; + if (vport->port_no == ODPP_LOCAL) + goto exit_free; - return put_user(retval, &upv->n_ports); + 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, @@ -1497,56 +1972,71 @@ 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; - unsigned int sflow_probability; + int listeners; int err; /* Handle commands with special locking requirements up front. */ switch (cmd) { - case ODP_DP_CREATE: - err = create_dp(dp_idx, (char __user *)argp); + case ODP_DP_NEW: + err = new_datapath((struct odp_datapath __user *)argp); + goto exit; + + case ODP_DP_GET: + err = get_datapath((struct odp_datapath __user *)argp); + goto exit; + + case ODP_DP_DEL: + err = del_datapath((struct odp_datapath __user *)argp); + goto exit; + + case ODP_DP_SET: + err = set_datapath((struct odp_datapath __user *)argp); goto exit; - case ODP_DP_DESTROY: - err = destroy_dp(dp_idx); + case ODP_DP_DUMP: + err = dump_datapath((struct odp_datapath __user *)argp); goto exit; - case ODP_VPORT_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_VPORT_DETACH: - err = get_user(port_no, (int __user *)argp); - if (!err) - err = detach_port(dp_idx, port_no); + case ODP_VPORT_GET: + err = get_vport((struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_MOD: - err = vport_user_mod((struct odp_port __user *)argp); + case ODP_VPORT_DEL: + err = del_vport(cmd, (struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_STATS_GET: - err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp); + case ODP_VPORT_SET: + err = set_vport(cmd, (struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_STATS_SET: - err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp); + case ODP_VPORT_DUMP: + err = dump_vport((struct odp_vport __user *)argp); goto exit; - case ODP_VPORT_ETHER_GET: - err = vport_user_ether_get((struct odp_vport_ether __user *)argp); + case ODP_FLOW_FLUSH: + err = flush_flows(argp); + goto exit; + + case ODP_FLOW_NEW: + case ODP_FLOW_SET: + err = new_flow(cmd, (struct odp_flow __user *)argp); goto exit; - case ODP_VPORT_ETHER_SET: - err = vport_user_ether_set((struct odp_vport_ether __user *)argp); + case ODP_FLOW_GET: + case ODP_FLOW_DEL: + err = get_or_del_flow(cmd, (struct odp_flow __user *)argp); goto exit; - case ODP_VPORT_MTU_GET: - err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp); + case ODP_FLOW_DUMP: + err = dump_flow((struct odp_flow __user *)argp); goto exit; - case ODP_VPORT_MTU_SET: - err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp); + case ODP_EXECUTE: + err = execute_packet((struct odp_packet __user *)argp); goto exit; } @@ -1556,25 +2046,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, goto exit; switch (cmd) { - case ODP_DP_STATS: - err = get_dp_stats(dp, (struct odp_stats __user *)argp); - break; - - case ODP_GET_DROP_FRAGS: - err = put_user(dp->drop_frags, (int __user *)argp); - break; - - case ODP_SET_DROP_FRAGS: - err = get_user(drop_frags, (int __user *)argp); - if (err) - break; - err = -EINVAL; - if (drop_frags != 0 && drop_frags != 1) - break; - dp->drop_frags = drop_frags; - err = 0; - break; - case ODP_GET_LISTEN_MASK: err = put_user(get_listen_mask(f), (int __user *)argp); break; @@ -1590,48 +2061,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, set_listen_mask(f, listeners); break; - case ODP_GET_SFLOW_PROBABILITY: - err = put_user(dp->sflow_probability, (unsigned int __user *)argp); - break; - - case ODP_SET_SFLOW_PROBABILITY: - err = get_user(sflow_probability, (unsigned int __user *)argp); - if (!err) - dp->sflow_probability = sflow_probability; - break; - - case ODP_VPORT_QUERY: - err = query_port(dp, (struct odp_port __user *)argp); - break; - - case ODP_VPORT_LIST: - err = list_ports(dp, (struct odp_portvec __user *)argp); - break; - - case ODP_FLOW_FLUSH: - err = flush_flows(dp); - break; - - case ODP_FLOW_PUT: - err = put_flow(dp, (struct odp_flow_put __user *)argp); - break; - - case ODP_FLOW_DEL: - err = del_flow(dp, (struct odp_flow __user *)argp); - break; - - case ODP_FLOW_GET: - err = do_flowvec_ioctl(dp, argp, do_query_flows); - break; - - case ODP_FLOW_LIST: - err = do_flowvec_ioctl(dp, argp, do_list_flows); - break; - - case ODP_EXECUTE: - err = execute_packet(dp, (struct odp_execute __user *)argp); - break; - default: err = -ENOIOCTLCMD; break; @@ -1652,378 +2081,48 @@ 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_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat) -{ - compat_uptr_t 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(actions, &compat->actions) || - __get_user(flow->n_actions, &compat->n_actions) || - __get_user(flow->flags, &compat->flags)) - return -EFAULT; - - flow->actions = 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 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(flow, query_flags, &ufp->stats, - compat_ptr(actions), &ufp->n_actions); -} - -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, &uf.key); - if (IS_ERR(flow)) - return PTR_ERR(flow); - - error = compat_answer_query(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) -{ - struct tbl *table = rcu_dereference(dp->table); - 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; - int error; - - if (compat_get_flow(&uf, ufp)) - return -EFAULT; - - flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); - if (!flow_node) - error = put_user(ENOENT, &ufp->stats.error); - else - error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp); - if (error) - return -EFAULT; - } - 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_) -{ - 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; - - if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) - return -EFAULT; - error = compat_answer_query(flow, 0, ufp); - if (error) - return error; - - if (cbdata->listed_flows >= cbdata->n_flows) - return cbdata->listed_flows; - return 0; -} - -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 (!n_flows) - return 0; - - 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; -} - -static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp, - int (*function)(struct datapath *, - struct compat_odp_flow *, - 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.n_actions, &uexecute->n_actions) || - __get_user(data, &uexecute->data) || - __get_user(execute.length, &uexecute->length)) - return -EFAULT; - - execute.actions = compat_ptr(actions); - execute.data = compat_ptr(data); - - return do_execute(dp, &execute); -} - static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) { - int dp_idx = iminor(f->f_dentry->d_inode); - 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_ATTACH: - case ODP_VPORT_DETACH: - case ODP_VPORT_MOD: - case ODP_VPORT_MTU_SET: - case ODP_VPORT_MTU_GET: - case ODP_VPORT_ETHER_SET: - case ODP_VPORT_ETHER_GET: - case ODP_VPORT_STATS_SET: - case ODP_VPORT_STATS_GET: - case ODP_DP_STATS: - case ODP_GET_DROP_FRAGS: - case ODP_SET_DROP_FRAGS: + case ODP_DP_NEW: + case ODP_DP_GET: + case ODP_DP_DEL: + case ODP_DP_SET: + case ODP_DP_DUMP: + case ODP_VPORT_NEW: + case ODP_VPORT_DEL: + case ODP_VPORT_GET: + case ODP_VPORT_SET: + case ODP_VPORT_DUMP: + case ODP_FLOW_NEW: + case ODP_FLOW_DEL: + case ODP_FLOW_GET: + case ODP_FLOW_SET: + case ODP_FLOW_DUMP: case ODP_SET_LISTEN_MASK: case ODP_GET_LISTEN_MASK: - case ODP_SET_SFLOW_PROBABILITY: - case ODP_GET_SFLOW_PROBABILITY: - case ODP_VPORT_QUERY: + case ODP_EXECUTE: /* 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_VPORT_LIST32: - err = compat_list_ports(dp, compat_ptr(argp)); - break; - - case ODP_FLOW_PUT32: - err = compat_put_flow(dp, compat_ptr(argp)); - break; - - case ODP_FLOW_DEL32: - err = compat_del_flow(dp, compat_ptr(argp)); - break; - - case ODP_FLOW_GET32: - err = compat_flowvec_ioctl(dp, argp, compat_query_flows); - break; - - case ODP_FLOW_LIST32: - err = compat_flowvec_ioctl(dp, argp, compat_list_flows); - break; - - case ODP_EXECUTE32: - err = compat_execute(dp, compat_ptr(argp)); - break; default: - err = -ENOIOCTLCMD; - break; + return -ENOIOCTLCMD; } - mutex_unlock(&dp->mutex); -exit: - return err; } #endif -/* Unfortunately this function is not exported so this is a verbatim copy - * from net/core/datagram.c in 2.6.30. */ -static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, - u8 __user *to, int len, - __wsum *csump) -{ - int start = skb_headlen(skb); - int pos = 0; - int i, copy = start - offset; - - /* Copy header. */ - if (copy > 0) { - int err = 0; - if (copy > len) - copy = len; - *csump = csum_and_copy_to_user(skb->data + offset, to, copy, - *csump, &err); - if (err) - goto fault; - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos = copy; - } - - for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { - int end; - - WARN_ON(start > offset + len); - - end = start + skb_shinfo(skb)->frags[i].size; - if ((copy = end - offset) > 0) { - __wsum csum2; - int err = 0; - u8 *vaddr; - skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; - struct page *page = frag->page; - - if (copy > len) - copy = len; - vaddr = kmap(page); - csum2 = csum_and_copy_to_user(vaddr + - frag->page_offset + - offset - start, - to, copy, 0, &err); - kunmap(page); - if (err) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if (!(len -= copy)) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; - } - - if (skb_shinfo(skb)->frag_list) { - struct sk_buff *list = skb_shinfo(skb)->frag_list; - - for (; list; list=list->next) { - int end; - - WARN_ON(start > offset + len); - - end = start + list->len; - if ((copy = end - offset) > 0) { - __wsum csum2 = 0; - if (copy > len) - copy = len; - if (skb_copy_and_csum_datagram(list, - offset - start, - to, copy, - &csum2)) - goto fault; - *csump = csum_block_add(*csump, csum2, pos); - if ((len -= copy) == 0) - return 0; - offset += copy; - to += copy; - pos += copy; - } - start = end; - } - } - if (!len) - return 0; - -fault: - return -EFAULT; -} - -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; - size_t copy_bytes, tot_copy_bytes; + struct iovec iov; int retval; if (!dp) @@ -2058,59 +2157,26 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, } } success: - copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes); - - retval = 0; - if (skb->ip_summed == CHECKSUM_PARTIAL) { - if (copy_bytes == skb->len) { - __wsum csum = 0; - unsigned int csum_start, csum_offset; - -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - csum_start = skb->csum_start - skb_headroom(skb); - csum_offset = skb->csum_offset; -#else - csum_start = skb_transport_header(skb) - skb->data; - csum_offset = skb->csum; -#endif - BUG_ON(csum_start >= skb_headlen(skb)); - retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start, - copy_bytes - csum_start, &csum); - if (!retval) { - __sum16 __user *csump; - - copy_bytes = csum_start; - csump = (__sum16 __user *)(buf + csum_start + csum_offset); - - BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes); - put_user(csum_fold(csum), csump); - } - } else - retval = skb_checksum_help(skb); - } - - if (!retval) { - struct iovec __user iov; - - iov.iov_base = buf; - iov.iov_len = copy_bytes; - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); - } + mutex_unlock(&dp->mutex); + iov.iov_base = buf; + iov.iov_len = min_t(size_t, skb->len, nbytes); + retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); if (!retval) - retval = tot_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) { @@ -2118,21 +2184,21 @@ 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;