X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=529d496913ec3c418f611530d729f5e24a7aeed3;hb=94903c989826268a7fab3c730cd7d0b35ef423d5;hp=04b19894dc4ee303ae9c9df11571314cb46c7d5d;hpb=776f10ce0f008e95b457f8c6b69b9748d8d1fdbd;p=openvswitch diff --git a/datapath/datapath.c b/datapath/datapath.c index 04b19894..529d4969 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -8,6 +8,8 @@ /* Functions for managing the dp interface/device. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include #include @@ -27,30 +29,31 @@ #include #include #include -#include #include #include #include #include +#include #include #include #include #include #include -#include #include #include +#include #include "openvswitch/datapath-protocol.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); @@ -66,9 +69,6 @@ EXPORT_SYMBOL(dp_ioctl_hook); static struct datapath *dps[ODP_MAX]; static DEFINE_MUTEX(dp_mutex); -/* Number of milliseconds between runs of the maintenance thread. */ -#define MAINT_SLEEP_MSECS 1000 - static int new_dp_port(struct datapath *, struct odp_port *, int port_no); /* Must be called with rcu_read_lock or dp_mutex. */ @@ -308,8 +308,6 @@ static void do_destroy_dp(struct datapath *dp) 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]); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); module_put(THIS_MODULE); @@ -358,13 +356,14 @@ static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_ vport = vport_locate(odp_port->devname); if (!vport) { - vport_lock(); + struct vport_parms parms; - if (odp_port->flags & ODP_PORT_INTERNAL) - vport = __vport_add(odp_port->devname, "internal", NULL); - else - vport = __vport_add(odp_port->devname, "netdev", NULL); + parms.name = odp_port->devname; + parms.type = odp_port->flags & ODP_PORT_INTERNAL ? "internal" : "netdev"; + parms.config = NULL; + vport_lock(); + vport = vport_add(&parms); vport_unlock(); if (IS_ERR(vport)) @@ -377,6 +376,7 @@ static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_ p->port_no = port_no; p->dp = dp; + p->vport = vport; atomic_set(&p->sflow_pool, 0); err = vport_attach(vport, p); @@ -469,7 +469,7 @@ int dp_detach_port(struct dp_port *p, int may_delete) if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) { vport_lock(); - __vport_del(vport); + vport_del(vport); vport_unlock(); } } @@ -510,42 +510,86 @@ out: return err; } -/* Must be called with rcu_read_lock and with bottom-halves disabled. */ +/* Must be called with rcu_read_lock. */ void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb) { struct datapath *dp = p->dp; struct dp_stats_percpu *stats; - struct odp_flow_key key; - struct tbl_node *flow_node; - - WARN_ON_ONCE(skb_shared(skb)); - skb_warn_if_lro(skb); + int stats_counter_off; + struct sw_flow_actions *acts; + struct loop_counter *loop; + int error; OVS_CB(skb)->dp_port = p; - /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */ - stats = percpu_ptr(dp->stats_percpu, smp_processor_id()); + if (!OVS_CB(skb)->flow) { + struct odp_flow_key key; + struct tbl_node *flow_node; + bool is_frag; - if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { - if (dp->drop_frags) { + /* Extract flow from 'skb' into 'key'. */ + error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag); + if (unlikely(error)) { kfree_skb(skb); - stats->n_frags++; return; } + + if (is_frag && dp->drop_frags) { + kfree_skb(skb); + stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); + goto out; + } + + /* Look up flow. */ + flow_node = tbl_lookup(rcu_dereference(dp->table), &key, + flow_hash(&key), flow_cmp); + if (unlikely(!flow_node)) { + dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + goto out; + } + + OVS_CB(skb)->flow = flow_cast(flow_node); } - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); - if (flow_node) { - struct sw_flow *flow = flow_cast(flow_node); - struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts); - flow_used(flow, skb); - execute_actions(dp, skb, &key, acts->actions, acts->n_actions, - GFP_ATOMIC); - stats->n_hit++; - } else { - stats->n_missed++; - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + flow_used(OVS_CB(skb)->flow, skb); + + acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); + + /* Check whether we've looped too much. */ + loop = loop_get_counter(); + if (unlikely(++loop->count > MAX_LOOPS)) + loop->looping = true; + if (unlikely(loop->looping)) { + loop_suppress(dp, acts); + goto out_loop; } + + /* 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); + + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + loop_suppress(dp, acts); + +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + loop_put_counter(); + +out: + /* Update datapath statistics. */ + local_bh_disable(); + stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); + (*(u64 *)((u8 *)stats + stats_counter_off))++; + write_seqcount_end(&stats->seqlock); + + local_bh_enable(); } #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) @@ -581,9 +625,9 @@ int vswitch_skb_checksum_setup(struct sk_buff *skb) break; default: if (net_ratelimit()) - printk(KERN_ERR "Attempting to checksum a non-" - "TCP/UDP packet, dropping a protocol" - " %d packet", iph->protocol); + pr_err("Attempting to checksum a non-TCP/UDP packet, " + "dropping a protocol %d packet", + iph->protocol); goto out; } @@ -649,8 +693,7 @@ out: * 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) +void compute_ip_summed(struct sk_buff *skb, bool xmit) { /* For our convenience these defines change repeatedly between kernel * versions, so we can't just copy them over... */ @@ -687,11 +730,10 @@ compute_ip_summed(struct sk_buff *skb, bool xmit) break; #endif default: - printk(KERN_ERR "openvswitch: unknown checksum type %d\n", - skb->ip_summed); + pr_err("unknown checksum type %d\n", skb->ip_summed); /* None seems the safest... */ OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - } + } #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) /* Xen has a special way of representing CHECKSUM_PARTIAL on older @@ -707,8 +749,7 @@ compute_ip_summed(struct sk_buff *skb, bool xmit) * 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) +void forward_ip_summed(struct sk_buff *skb) { #ifdef CHECKSUM_HW if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) @@ -718,9 +759,8 @@ forward_ip_summed(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 sk_buff *skb, struct sk_buff_head *queue, - int queue_no, u32 arg) +static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, + int queue_no, u32 arg) { struct sk_buff *nskb; int port_no; @@ -737,31 +777,6 @@ queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, nskb = skb->next; skb->next = NULL; - /* If a checksum-deferred packet is forwarded to the - * controller, correct the pointers and checksum. - */ - err = vswitch_skb_checksum_setup(skb); - if (err) - goto err_kfree_skbs; - - if (skb->ip_summed == CHECKSUM_PARTIAL) { - -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - /* Until 2.6.22, the start of the transport header was - * also the start of data to be checksummed. Linux - * 2.6.22 introduced the csum_start field for this - * purpose, but we should point the transport header to - * it anyway for backward compatibility, as - * dev_queue_xmit() does even in 2.6.28. */ - skb_set_transport_header(skb, skb->csum_start - - skb_headroom(skb)); -#endif - - err = skb_checksum_help(skb); - if (err) - goto err_kfree_skbs; - } - err = skb_cow(skb, sizeof *header); if (err) goto err_kfree_skbs; @@ -787,9 +802,8 @@ err_kfree_skbs: return err; } -int -dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, - u32 arg) +int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, + u32 arg) { struct dp_stats_percpu *stats; struct sk_buff_head *queue; @@ -804,10 +818,14 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, forward_ip_summed(skb); + err = vswitch_skb_checksum_setup(skb); + if (err) + goto err_kfree_skb; + /* Break apart GSO packets into their component pieces. Otherwise * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ if (skb_is_gso(skb)) { - struct sk_buff *nskb = skb_gso_segment(skb, 0); + struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); if (nskb) { kfree_skb(skb); skb = nskb; @@ -828,9 +846,14 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, 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; } @@ -857,25 +880,30 @@ static int validate_actions(const struct sw_flow_actions *actions) for (i = 0; i < actions->n_actions; i++) { const union odp_action *a = &actions->actions[i]; - switch (a->type) { - case ODPAT_OUTPUT: - if (a->output.port >= DP_MAX_PORTS) - return -EINVAL; - break; - case ODPAT_OUTPUT_GROUP: - if (a->output_group.group >= DP_MAX_GROUPS) - return -EINVAL; + switch (a->type) { + case ODPAT_CONTROLLER: + case ODPAT_STRIP_VLAN: + case ODPAT_SET_DL_SRC: + case ODPAT_SET_DL_DST: + case ODPAT_SET_NW_SRC: + case ODPAT_SET_NW_DST: + case ODPAT_SET_TP_SRC: + case ODPAT_SET_TP_DST: + case ODPAT_SET_TUNNEL: + case ODPAT_SET_PRIORITY: + case ODPAT_POP_PRIORITY: + case ODPAT_DROP_SPOOFED_ARP: + /* No validation needed. */ break; - case ODPAT_SET_VLAN_VID: - if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) + case ODPAT_OUTPUT: + if (a->output.port >= DP_MAX_PORTS) return -EINVAL; break; - case ODPAT_SET_VLAN_PCP: - if (a->vlan_pcp.vlan_pcp - & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) + case ODPAT_SET_DL_TCI: + if (a->dl_tci.tci & htons(VLAN_CFI_MASK)) return -EINVAL; break; @@ -885,9 +913,7 @@ static int validate_actions(const struct sw_flow_actions *actions) break; default: - if (a->type >= ODPAT_N_ACTIONS) - return -EOPNOTSUPP; - break; + return -EOPNOTSUPP; } } @@ -922,25 +948,32 @@ 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; } @@ -968,8 +1001,6 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, struct tbl *table; int error; - memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved); - table = rcu_dereference(dp->table); flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); if (!flow_node) { @@ -989,12 +1020,12 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, } /* 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); clear_stats(flow); /* Obtain actions. */ @@ -1013,7 +1044,6 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, } else { /* We found a matching flow. */ struct sw_flow_actions *old_acts, *new_acts; - unsigned long int flags; flow = flow_cast(flow_node); @@ -1038,11 +1068,11 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, } /* Fetch stats, then clear them if necessary. */ - spin_lock_irqsave(&flow->lock, flags); + 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); } return 0; @@ -1050,7 +1080,8 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, error_free_flow_acts: kfree(flow->sf_acts); error_free_flow: - kmem_cache_free(flow_cache, flow); + flow->sf_acts = NULL; + flow_put(flow); error: return error; } @@ -1082,15 +1113,14 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, { struct sw_flow_actions *sf_acts; struct odp_flow_stats stats; - unsigned long int flags; u32 n_actions; - spin_lock_irqsave(&flow->lock, flags); + spin_lock_bh(&flow->lock); get_stats(flow, &stats); - if (query_flags & ODPFF_ZERO_TCP_FLAGS) { + if (query_flags & ODPFF_ZERO_TCP_FLAGS) flow->tcp_flags = 0; - } - spin_unlock_irqrestore(&flow->lock, flags); + + spin_unlock_bh(&flow->lock); if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || get_user(n_actions, n_actionsp)) @@ -1117,7 +1147,7 @@ static int answer_query(struct sw_flow *flow, u32 query_flags, if (get_user(actions, &ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, + return do_answer_query(flow, query_flags, &ufp->stats, actions, &ufp->n_actions); } @@ -1127,7 +1157,6 @@ static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key struct tbl_node *flow_node; int error; - memset(key->reserved, 0, sizeof key->reserved); flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp); if (!flow_node) return ERR_PTR(-ENOENT); @@ -1174,7 +1203,6 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec if (copy_from_user(&uf, ufp, sizeof uf)) return -EFAULT; - memset(uf.key.reserved, 0, sizeof uf.key.reserved); flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp); if (!flow_node) @@ -1222,6 +1250,7 @@ static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) 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; } @@ -1253,16 +1282,18 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) struct sk_buff *skb; struct sw_flow_actions *actions; struct ethhdr *eth; + bool is_frag; int err; err = -EINVAL; if (execute->length < ETH_HLEN || execute->length > 65535) goto error; - err = -ENOMEM; actions = flow_actions_alloc(execute->n_actions); - if (!actions) + if (IS_ERR(actions)) { + err = PTR_ERR(actions); goto error; + } err = -EFAULT; if (copy_from_user(actions->actions, execute->actions, @@ -1278,11 +1309,6 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) if (!skb) goto error_free_actions; - if (execute->in_port < DP_MAX_PORTS) - OVS_CB(skb)->dp_port = dp->ports[execute->in_port]; - else - OVS_CB(skb)->dp_port = NULL; - err = -EFAULT; if (copy_from_user(skb_put(skb, execute->length), execute->data, execute->length)) @@ -1299,9 +1325,14 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) else skb->protocol = htons(ETH_P_802_2); - flow_extract(skb, execute->in_port, &key); - err = execute_actions(dp, skb, &key, actions->actions, - actions->n_actions, GFP_KERNEL); + err = flow_extract(skb, -1, &key, &is_frag); + if (err) + goto error_free_skb; + + rcu_read_lock(); + err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions); + rcu_read_unlock(); + kfree(actions); return err; @@ -1334,15 +1365,23 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) stats.max_capacity = TBL_MAX_BUCKETS; stats.n_ports = dp->n_ports; stats.max_ports = DP_MAX_PORTS; - stats.max_groups = DP_MAX_GROUPS; stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; for_each_possible_cpu(i) { - const struct dp_stats_percpu *s; - s = percpu_ptr(dp->stats_percpu, i); - stats.n_frags += s->n_frags; - stats.n_hit += s->n_hit; - stats.n_missed += s->n_missed; - stats.n_lost += s->n_lost; + const struct dp_stats_percpu *percpu_stats; + struct dp_stats_percpu local_stats; + unsigned seqcount; + + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); + + stats.n_frags += local_stats.n_frags; + stats.n_hit += local_stats.n_hit; + stats.n_missed += local_stats.n_missed; + stats.n_lost += local_stats.n_lost; } stats.max_miss_queue = DP_MAX_QUEUE_LEN; stats.max_action_queue = DP_MAX_QUEUE_LEN; @@ -1390,8 +1429,7 @@ void set_internal_devs_mtu(const struct datapath *dp) } } -static int -put_port(const struct dp_port *p, struct odp_port __user *uop) +static int put_port(const struct dp_port *p, struct odp_port __user *uop) { struct odp_port op; @@ -1407,8 +1445,7 @@ put_port(const struct dp_port *p, struct odp_port __user *uop) return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; } -static int -query_port(struct datapath *dp, struct odp_port __user *uport) +static int query_port(struct datapath *dp, struct odp_port __user *uport) { struct odp_port port; @@ -1455,8 +1492,8 @@ error_unlock: return put_port(dp->ports[port.port], uport); } -static int -do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports) +static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, + int n_ports) { int idx = 0; if (n_ports) { @@ -1472,8 +1509,7 @@ do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports) return idx; } -static int -list_ports(struct datapath *dp, struct odp_portvec __user *upv) +static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) { struct odp_portvec pv; int retval; @@ -1488,88 +1524,6 @@ list_ports(struct datapath *dp, struct odp_portvec __user *upv) return put_user(retval, &upv->n_ports); } -/* RCU callback for freeing a dp_port_group */ -static void free_port_group(struct rcu_head *rcu) -{ - struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu); - kfree(g); -} - -static int -do_set_port_group(struct datapath *dp, u16 __user *ports, int n_ports, int group) -{ - struct dp_port_group *new_group, *old_group; - int error; - - error = -EINVAL; - if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS) - goto error; - - error = -ENOMEM; - new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL); - if (!new_group) - goto error; - - new_group->n_ports = n_ports; - error = -EFAULT; - if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports)) - goto error_free; - - old_group = rcu_dereference(dp->groups[group]); - rcu_assign_pointer(dp->groups[group], new_group); - if (old_group) - call_rcu(&old_group->rcu, free_port_group); - return 0; - -error_free: - kfree(new_group); -error: - return error; -} - -static int -set_port_group(struct datapath *dp, const struct odp_port_group __user *upg) -{ - struct odp_port_group pg; - - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; - - return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group); -} - -static int -do_get_port_group(struct datapath *dp, - u16 __user *ports, int n_ports, int group, - u16 __user *n_portsp) -{ - struct dp_port_group *g; - u16 n_copy; - - if (group >= DP_MAX_GROUPS) - return -EINVAL; - - g = dp->groups[group]; - n_copy = g ? min_t(int, g->n_ports, n_ports) : 0; - if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16))) - return -EFAULT; - - if (put_user(g ? g->n_ports : 0, n_portsp)) - return -EFAULT; - - return 0; -} - -static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg) -{ - struct odp_port_group pg; - - if (copy_from_user(&pg, upg, sizeof pg)) - return -EFAULT; - - return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &pg.n_ports); -} - static int get_listen_mask(const struct file *f) { return (long)f->private_data; @@ -1610,35 +1564,39 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, goto exit; case ODP_VPORT_ADD: - err = vport_add((struct odp_vport_add __user *)argp); + err = vport_user_add((struct odp_vport_add __user *)argp); goto exit; case ODP_VPORT_MOD: - err = vport_mod((struct odp_vport_mod __user *)argp); + err = vport_user_mod((struct odp_vport_mod __user *)argp); goto exit; case ODP_VPORT_DEL: - err = vport_del((char __user *)argp); + err = vport_user_del((char __user *)argp); goto exit; case ODP_VPORT_STATS_GET: - err = vport_stats_get((struct odp_vport_stats_req __user *)argp); + err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp); + goto exit; + + case ODP_VPORT_STATS_SET: + err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp); goto exit; case ODP_VPORT_ETHER_GET: - err = vport_ether_get((struct odp_vport_ether __user *)argp); + err = vport_user_ether_get((struct odp_vport_ether __user *)argp); goto exit; case ODP_VPORT_ETHER_SET: - err = vport_ether_set((struct odp_vport_ether __user *)argp); + err = vport_user_ether_set((struct odp_vport_ether __user *)argp); goto exit; case ODP_VPORT_MTU_GET: - err = vport_mtu_get((struct odp_vport_mtu __user *)argp); + err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp); goto exit; case ODP_VPORT_MTU_SET: - err = vport_mtu_set((struct odp_vport_mtu __user *)argp); + err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp); goto exit; } @@ -1700,14 +1658,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, err = list_ports(dp, (struct odp_portvec __user *)argp); break; - case ODP_PORT_GROUP_SET: - err = set_port_group(dp, (struct odp_port_group __user *)argp); - break; - - case ODP_PORT_GROUP_GET: - err = get_port_group(dp, (struct odp_port_group __user *)argp); - break; - case ODP_FLOW_FLUSH: err = flush_flows(dp); break; @@ -1751,6 +1701,376 @@ static int dp_has_packet_of_interest(struct datapath *dp, int listeners) return 0; } +#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_PORT_ATTACH: + case ODP_PORT_DETACH: + case ODP_VPORT_DEL: + case ODP_VPORT_MTU_SET: + case ODP_VPORT_MTU_GET: + case ODP_VPORT_ETHER_SET: + case ODP_VPORT_ETHER_GET: + case ODP_VPORT_STATS_SET: + case ODP_VPORT_STATS_GET: + case ODP_DP_STATS: + case ODP_GET_DROP_FRAGS: + case ODP_SET_DROP_FRAGS: + case ODP_SET_LISTEN_MASK: + case ODP_GET_LISTEN_MASK: + case ODP_SET_SFLOW_PROBABILITY: + case ODP_GET_SFLOW_PROBABILITY: + case ODP_PORT_QUERY: + /* Ioctls that just need their pointer argument extended. */ + return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); + + case ODP_VPORT_ADD32: + return compat_vport_user_add(compat_ptr(argp)); + + case ODP_VPORT_MOD32: + return compat_vport_user_mod(compat_ptr(argp)); + } + + dp = get_dp_locked(dp_idx); + err = -ENODEV; + if (!dp) + goto exit; + + switch (cmd) { + case ODP_PORT_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; + } + 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) { @@ -1759,8 +2079,7 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, int dp_idx = iminor(f->f_dentry->d_inode); struct datapath *dp = get_dp(dp_idx); struct sk_buff *skb; - struct iovec __user iov; - size_t copy_bytes; + size_t copy_bytes, tot_copy_bytes; int retval; if (!dp) @@ -1795,12 +2114,48 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, } } success: - copy_bytes = min_t(size_t, skb->len, nbytes); - iov.iov_base = buf; - iov.iov_len = copy_bytes; - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); + 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); + } + if (!retval) - retval = copy_bytes; + retval = tot_copy_bytes; + kfree_skb(skb); error: @@ -1830,6 +2185,9 @@ struct file_operations openvswitch_fops = { .read = openvswitch_read, .poll = openvswitch_poll, .unlocked_ioctl = openvswitch_ioctl, +#ifdef CONFIG_COMPAT + .compat_ioctl = openvswitch_compat_ioctl, +#endif /* XXX .fasync = openvswitch_fasync, */ };