2 * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
16 #include <linux/if_arp.h>
17 #include <linux/if_vlan.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/kernel.h>
24 #include <linux/kthread.h>
25 #include <linux/mutex.h>
26 #include <linux/percpu.h>
27 #include <linux/rcupdate.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/version.h>
31 #include <linux/ethtool.h>
32 #include <linux/wait.h>
33 #include <asm/system.h>
34 #include <asm/div64.h>
36 #include <linux/highmem.h>
37 #include <linux/netfilter_bridge.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/inetdevice.h>
40 #include <linux/list.h>
41 #include <linux/rculist.h>
42 #include <linux/dmi.h>
43 #include <net/inet_ecn.h>
44 #include <linux/compat.h>
46 #include "openvswitch/datapath-protocol.h"
51 #include "loop_counter.h"
52 #include "odp-compat.h"
54 #include "vport-internal_dev.h"
58 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
59 EXPORT_SYMBOL(dp_ioctl_hook);
61 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
64 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
67 * It is safe to access the datapath and vport structures with just
70 static struct datapath __rcu *dps[ODP_MAX];
71 static DEFINE_MUTEX(dp_mutex);
73 static int new_vport(struct datapath *, struct odp_port *, int port_no);
75 /* Must be called with rcu_read_lock or dp_mutex. */
76 struct datapath *get_dp(int dp_idx)
78 if (dp_idx < 0 || dp_idx >= ODP_MAX)
80 return rcu_dereference(dps[dp_idx]);
82 EXPORT_SYMBOL_GPL(get_dp);
84 static struct datapath *get_dp_locked(int dp_idx)
88 mutex_lock(&dp_mutex);
91 mutex_lock(&dp->mutex);
92 mutex_unlock(&dp_mutex);
96 /* Must be called with rcu_read_lock or RTNL lock. */
97 const char *dp_name(const struct datapath *dp)
99 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
102 static inline size_t br_nlmsg_size(void)
104 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
105 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
106 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
107 + nla_total_size(4) /* IFLA_MASTER */
108 + nla_total_size(4) /* IFLA_MTU */
109 + nla_total_size(4) /* IFLA_LINK */
110 + nla_total_size(1); /* IFLA_OPERSTATE */
113 static int dp_fill_ifinfo(struct sk_buff *skb,
114 const struct vport *port,
115 int event, unsigned int flags)
117 const struct datapath *dp = port->dp;
118 int ifindex = vport_get_ifindex(port);
119 int iflink = vport_get_iflink(port);
120 struct ifinfomsg *hdr;
121 struct nlmsghdr *nlh;
129 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
133 hdr = nlmsg_data(nlh);
134 hdr->ifi_family = AF_BRIDGE;
136 hdr->ifi_type = ARPHRD_ETHER;
137 hdr->ifi_index = ifindex;
138 hdr->ifi_flags = vport_get_flags(port);
141 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
142 NLA_PUT_U32(skb, IFLA_MASTER,
143 vport_get_ifindex(rtnl_dereference(dp->ports[ODPP_LOCAL])));
144 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
145 #ifdef IFLA_OPERSTATE
146 NLA_PUT_U8(skb, IFLA_OPERSTATE,
147 vport_is_running(port)
148 ? vport_get_operstate(port)
152 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
154 if (ifindex != iflink)
155 NLA_PUT_U32(skb, IFLA_LINK,iflink);
157 return nlmsg_end(skb, nlh);
160 nlmsg_cancel(skb, nlh);
164 static void dp_ifinfo_notify(int event, struct vport *port)
169 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
173 err = dp_fill_ifinfo(skb, port, event, 0);
175 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
176 WARN_ON(err == -EMSGSIZE);
180 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
184 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
187 static void release_dp(struct kobject *kobj)
189 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
193 static struct kobj_type dp_ktype = {
194 .release = release_dp
197 static int create_dp(int dp_idx, const char __user *devnamep)
199 struct odp_port internal_dev_port;
200 char devname[IFNAMSIZ];
206 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
210 } else if (retval >= IFNAMSIZ) {
215 snprintf(devname, sizeof devname, "of%d", dp_idx);
219 mutex_lock(&dp_mutex);
221 if (!try_module_get(THIS_MODULE))
224 /* Exit early if a datapath with that number already exists.
225 * (We don't use -EEXIST because that's ambiguous with 'devname'
226 * conflicting with an existing network device name.) */
232 dp = kzalloc(sizeof *dp, GFP_KERNEL);
235 INIT_LIST_HEAD(&dp->port_list);
236 mutex_init(&dp->mutex);
238 for (i = 0; i < DP_N_QUEUES; i++)
239 skb_queue_head_init(&dp->queues[i]);
240 init_waitqueue_head(&dp->waitqueue);
242 /* Initialize kobject for bridge. This will be added as
243 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
244 dp->ifobj.kset = NULL;
245 kobject_init(&dp->ifobj, &dp_ktype);
247 /* Allocate table. */
249 rcu_assign_pointer(dp->table, tbl_create(0));
253 /* Set up our datapath device. */
254 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
255 strcpy(internal_dev_port.devname, devname);
256 strcpy(internal_dev_port.type, "internal");
257 err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
262 goto err_destroy_table;
266 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
267 if (!dp->stats_percpu)
268 goto err_destroy_local_port;
270 rcu_assign_pointer(dps[dp_idx], dp);
273 mutex_unlock(&dp_mutex);
278 err_destroy_local_port:
279 dp_detach_port(dp->ports[ODPP_LOCAL]);
281 tbl_destroy(dp->table, NULL);
285 module_put(THIS_MODULE);
287 mutex_unlock(&dp_mutex);
293 static void do_destroy_dp(struct datapath *dp)
298 list_for_each_entry_safe (p, n, &dp->port_list, node)
299 if (p->port_no != ODPP_LOCAL)
304 rcu_assign_pointer(dps[dp->dp_idx], NULL);
306 dp_detach_port(dp->ports[ODPP_LOCAL]);
308 tbl_destroy(dp->table, flow_free_tbl);
310 for (i = 0; i < DP_N_QUEUES; i++)
311 skb_queue_purge(&dp->queues[i]);
312 free_percpu(dp->stats_percpu);
313 kobject_put(&dp->ifobj);
314 module_put(THIS_MODULE);
317 static int destroy_dp(int dp_idx)
323 mutex_lock(&dp_mutex);
333 mutex_unlock(&dp_mutex);
338 /* Called with RTNL lock and dp_mutex. */
339 static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
341 struct vport_parms parms;
344 parms.name = odp_port->devname;
345 parms.type = odp_port->type;
346 parms.config = odp_port->config;
348 parms.port_no = port_no;
351 vport = vport_add(&parms);
355 return PTR_ERR(vport);
357 rcu_assign_pointer(dp->ports[port_no], vport);
358 list_add_rcu(&vport->node, &dp->port_list);
361 dp_ifinfo_notify(RTM_NEWLINK, vport);
366 static int attach_port(int dp_idx, struct odp_port __user *portp)
369 struct odp_port port;
374 if (copy_from_user(&port, portp, sizeof port))
376 port.devname[IFNAMSIZ - 1] = '\0';
377 port.type[VPORT_TYPE_SIZE - 1] = '\0';
380 dp = get_dp_locked(dp_idx);
383 goto out_unlock_rtnl;
385 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
386 if (!dp->ports[port_no])
392 err = new_vport(dp, &port, port_no);
396 set_internal_devs_mtu(dp);
397 dp_sysfs_add_if(dp->ports[port_no]);
399 err = put_user(port_no, &portp->port);
402 mutex_unlock(&dp->mutex);
409 int dp_detach_port(struct vport *p)
415 if (p->port_no != ODPP_LOCAL)
417 dp_ifinfo_notify(RTM_DELLINK, p);
419 /* First drop references to device. */
421 list_del_rcu(&p->node);
422 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
424 /* Then destroy it. */
432 static int detach_port(int dp_idx, int port_no)
439 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
443 dp = get_dp_locked(dp_idx);
446 goto out_unlock_rtnl;
448 p = dp->ports[port_no];
453 err = dp_detach_port(p);
456 mutex_unlock(&dp->mutex);
463 /* Must be called with rcu_read_lock. */
464 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
466 struct datapath *dp = p->dp;
467 struct dp_stats_percpu *stats;
468 int stats_counter_off;
469 struct sw_flow_actions *acts;
470 struct loop_counter *loop;
473 OVS_CB(skb)->vport = p;
475 if (!OVS_CB(skb)->flow) {
476 struct odp_flow_key key;
477 struct tbl_node *flow_node;
480 /* Extract flow from 'skb' into 'key'. */
481 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
482 if (unlikely(error)) {
487 if (is_frag && dp->drop_frags) {
489 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
494 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
495 flow_hash(&key), flow_cmp);
496 if (unlikely(!flow_node)) {
497 dp_output_control(dp, skb, _ODPL_MISS_NR,
498 (__force u64)OVS_CB(skb)->tun_id);
499 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
503 OVS_CB(skb)->flow = flow_cast(flow_node);
506 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
507 flow_used(OVS_CB(skb)->flow, skb);
509 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
511 /* Check whether we've looped too much. */
512 loop = loop_get_counter();
513 if (unlikely(++loop->count > MAX_LOOPS))
514 loop->looping = true;
515 if (unlikely(loop->looping)) {
516 loop_suppress(dp, acts);
521 /* Execute actions. */
522 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
525 /* Check whether sub-actions looped too much. */
526 if (unlikely(loop->looping))
527 loop_suppress(dp, acts);
530 /* Decrement loop counter. */
532 loop->looping = false;
536 /* Update datapath statistics. */
538 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
540 write_seqcount_begin(&stats->seqlock);
541 (*(u64 *)((u8 *)stats + stats_counter_off))++;
542 write_seqcount_end(&stats->seqlock);
547 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
548 * unless we broke up a GSO packet. */
549 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
550 int queue_no, u64 arg)
552 struct sk_buff *nskb;
556 if (OVS_CB(skb)->vport)
557 port_no = OVS_CB(skb)->vport->port_no;
559 port_no = ODPP_LOCAL;
562 struct odp_msg *header;
567 err = skb_cow(skb, sizeof *header);
571 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
572 header->type = queue_no;
573 header->length = skb->len;
574 header->port = port_no;
576 skb_queue_tail(queue, skb);
584 while ((skb = nskb) != NULL) {
591 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
594 struct dp_stats_percpu *stats;
595 struct sk_buff_head *queue;
598 WARN_ON_ONCE(skb_shared(skb));
599 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
600 queue = &dp->queues[queue_no];
602 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
605 forward_ip_summed(skb);
607 err = vswitch_skb_checksum_setup(skb);
611 /* Break apart GSO packets into their component pieces. Otherwise
612 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
613 if (skb_is_gso(skb)) {
614 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
624 err = queue_control_packets(skb, queue, queue_no, arg);
625 wake_up_interruptible(&dp->waitqueue);
632 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
634 write_seqcount_begin(&stats->seqlock);
636 write_seqcount_end(&stats->seqlock);
643 static int flush_flows(struct datapath *dp)
645 struct tbl *old_table = rcu_dereference(dp->table);
646 struct tbl *new_table;
648 new_table = tbl_create(0);
652 rcu_assign_pointer(dp->table, new_table);
654 tbl_deferred_destroy(old_table, flow_free_tbl);
659 static int validate_actions(const struct nlattr *actions, u32 actions_len)
661 const struct nlattr *a;
664 nla_for_each_attr(a, actions, actions_len, rem) {
665 static const u32 action_lens[ODPAT_MAX + 1] = {
667 [ODPAT_CONTROLLER] = 8,
668 [ODPAT_SET_DL_TCI] = 2,
669 [ODPAT_STRIP_VLAN] = 0,
670 [ODPAT_SET_DL_SRC] = ETH_ALEN,
671 [ODPAT_SET_DL_DST] = ETH_ALEN,
672 [ODPAT_SET_NW_SRC] = 4,
673 [ODPAT_SET_NW_DST] = 4,
674 [ODPAT_SET_NW_TOS] = 1,
675 [ODPAT_SET_TP_SRC] = 2,
676 [ODPAT_SET_TP_DST] = 2,
677 [ODPAT_SET_TUNNEL] = 8,
678 [ODPAT_SET_PRIORITY] = 4,
679 [ODPAT_POP_PRIORITY] = 0,
680 [ODPAT_DROP_SPOOFED_ARP] = 0,
682 int type = nla_type(a);
684 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
691 case ODPAT_CONTROLLER:
692 case ODPAT_STRIP_VLAN:
693 case ODPAT_SET_DL_SRC:
694 case ODPAT_SET_DL_DST:
695 case ODPAT_SET_NW_SRC:
696 case ODPAT_SET_NW_DST:
697 case ODPAT_SET_TP_SRC:
698 case ODPAT_SET_TP_DST:
699 case ODPAT_SET_TUNNEL:
700 case ODPAT_SET_PRIORITY:
701 case ODPAT_POP_PRIORITY:
702 case ODPAT_DROP_SPOOFED_ARP:
703 /* No validation needed. */
707 if (nla_get_u32(a) >= DP_MAX_PORTS)
710 case ODPAT_SET_DL_TCI:
711 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
715 case ODPAT_SET_NW_TOS:
716 if (nla_get_u8(a) & INET_ECN_MASK)
731 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
733 struct sw_flow_actions *actions;
736 actions = flow_actions_alloc(flow->actions_len);
737 error = PTR_ERR(actions);
742 if (copy_from_user(actions->actions,
743 (struct nlattr __user *)flow->actions,
745 goto error_free_actions;
746 error = validate_actions(actions->actions, actions->actions_len);
748 goto error_free_actions;
755 return ERR_PTR(error);
758 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
761 struct timespec offset_ts, used, now_mono;
763 ktime_get_ts(&now_mono);
764 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
765 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
766 now_mono.tv_nsec - offset_ts.tv_nsec);
768 stats->used_sec = used.tv_sec;
769 stats->used_nsec = used.tv_nsec;
772 stats->used_nsec = 0;
775 stats->n_packets = flow->packet_count;
776 stats->n_bytes = flow->byte_count;
778 stats->tcp_flags = flow->tcp_flags;
782 static void clear_stats(struct sw_flow *flow)
786 flow->packet_count = 0;
787 flow->byte_count = 0;
790 static int expand_table(struct datapath *dp)
792 struct tbl *old_table = rcu_dereference(dp->table);
793 struct tbl *new_table;
795 new_table = tbl_expand(old_table);
796 if (IS_ERR(new_table))
797 return PTR_ERR(new_table);
799 rcu_assign_pointer(dp->table, new_table);
800 tbl_deferred_destroy(old_table, NULL);
805 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
806 struct odp_flow_stats *stats)
808 struct tbl_node *flow_node;
809 struct sw_flow *flow;
813 table = rcu_dereference(dp->table);
814 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
817 struct sw_flow_actions *acts;
820 if (!(uf->flags & ODPPF_CREATE))
823 /* Expand table, if necessary, to make room. */
824 if (tbl_count(table) >= tbl_n_buckets(table)) {
825 error = expand_table(dp);
828 table = rcu_dereference(dp->table);
834 error = PTR_ERR(flow);
837 flow->key = uf->flow.key;
840 /* Obtain actions. */
841 acts = get_actions(&uf->flow);
842 error = PTR_ERR(acts);
844 goto error_free_flow;
845 rcu_assign_pointer(flow->sf_acts, acts);
847 /* Put flow in bucket. */
848 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
850 goto error_free_flow_acts;
852 memset(stats, 0, sizeof(struct odp_flow_stats));
854 /* We found a matching flow. */
855 struct sw_flow_actions *old_acts, *new_acts;
857 flow = flow_cast(flow_node);
859 /* Bail out if we're not allowed to modify an existing flow. */
861 if (!(uf->flags & ODPPF_MODIFY))
865 new_acts = get_actions(&uf->flow);
866 error = PTR_ERR(new_acts);
867 if (IS_ERR(new_acts))
869 old_acts = rcu_dereference(flow->sf_acts);
870 if (old_acts->actions_len != new_acts->actions_len ||
871 memcmp(old_acts->actions, new_acts->actions,
872 old_acts->actions_len)) {
873 rcu_assign_pointer(flow->sf_acts, new_acts);
874 flow_deferred_free_acts(old_acts);
879 /* Fetch stats, then clear them if necessary. */
880 spin_lock_bh(&flow->lock);
881 get_stats(flow, stats);
882 if (uf->flags & ODPPF_ZERO_STATS)
884 spin_unlock_bh(&flow->lock);
889 error_free_flow_acts:
890 kfree(flow->sf_acts);
892 flow->sf_acts = NULL;
898 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
900 struct odp_flow_stats stats;
901 struct odp_flow_put uf;
904 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
907 error = do_put_flow(dp, &uf, &stats);
911 if (copy_to_user(&ufp->flow.stats, &stats,
912 sizeof(struct odp_flow_stats)))
918 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
919 struct odp_flow_stats __user *ustats,
920 struct nlattr __user *actions,
921 u32 __user *actions_lenp)
923 struct sw_flow_actions *sf_acts;
924 struct odp_flow_stats stats;
927 spin_lock_bh(&flow->lock);
928 get_stats(flow, &stats);
929 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
932 spin_unlock_bh(&flow->lock);
934 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
935 get_user(actions_len, actions_lenp))
941 sf_acts = rcu_dereference(flow->sf_acts);
942 if (put_user(sf_acts->actions_len, actions_lenp) ||
943 (actions && copy_to_user(actions, sf_acts->actions,
944 min(sf_acts->actions_len, actions_len))))
950 static int answer_query(struct sw_flow *flow, u32 query_flags,
951 struct odp_flow __user *ufp)
953 struct nlattr __user *actions;
955 if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
958 return do_answer_query(flow, query_flags,
959 &ufp->stats, actions, &ufp->actions_len);
962 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
964 struct tbl *table = rcu_dereference(dp->table);
965 struct tbl_node *flow_node;
968 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
970 return ERR_PTR(-ENOENT);
972 error = tbl_remove(table, flow_node);
974 return ERR_PTR(error);
976 /* XXX Returned flow_node's statistics might lose a few packets, since
977 * other CPUs can be using this flow. We used to synchronize_rcu() to
978 * make sure that we get completely accurate stats, but that blows our
979 * performance, badly. */
980 return flow_cast(flow_node);
983 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
985 struct sw_flow *flow;
989 if (copy_from_user(&uf, ufp, sizeof uf))
992 flow = do_del_flow(dp, &uf.key);
994 return PTR_ERR(flow);
996 error = answer_query(flow, 0, ufp);
997 flow_deferred_free(flow);
1001 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1003 struct tbl *table = rcu_dereference(dp->table);
1006 for (i = 0; i < flowvec->n_flows; i++) {
1007 struct odp_flow __user *ufp = (struct odp_flow __user *)&flowvec->flows[i];
1009 struct tbl_node *flow_node;
1012 if (copy_from_user(&uf, ufp, sizeof uf))
1015 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1017 error = put_user(ENOENT, &ufp->stats.error);
1019 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1023 return flowvec->n_flows;
1026 struct list_flows_cbdata {
1027 struct odp_flow __user *uflows;
1032 static int list_flow(struct tbl_node *node, void *cbdata_)
1034 struct sw_flow *flow = flow_cast(node);
1035 struct list_flows_cbdata *cbdata = cbdata_;
1036 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1039 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1041 error = answer_query(flow, 0, ufp);
1045 if (cbdata->listed_flows >= cbdata->n_flows)
1046 return cbdata->listed_flows;
1050 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1052 struct list_flows_cbdata cbdata;
1055 if (!flowvec->n_flows)
1058 cbdata.uflows = (struct odp_flow __user *)flowvec->flows;
1059 cbdata.n_flows = flowvec->n_flows;
1060 cbdata.listed_flows = 0;
1062 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1063 return error ? error : cbdata.listed_flows;
1066 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1067 int (*function)(struct datapath *,
1068 const struct odp_flowvec *))
1070 struct odp_flowvec __user *uflowvec;
1071 struct odp_flowvec flowvec;
1074 uflowvec = (struct odp_flowvec __user *)argp;
1075 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1078 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1081 retval = function(dp, &flowvec);
1082 return (retval < 0 ? retval
1083 : retval == flowvec.n_flows ? 0
1084 : put_user(retval, &uflowvec->n_flows));
1087 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1089 struct odp_flow_key key;
1090 struct sk_buff *skb;
1091 struct sw_flow_actions *actions;
1097 if (execute->length < ETH_HLEN || execute->length > 65535)
1100 actions = flow_actions_alloc(execute->actions_len);
1101 if (IS_ERR(actions)) {
1102 err = PTR_ERR(actions);
1107 if (copy_from_user(actions->actions,
1108 (struct nlattr __user *)execute->actions, execute->actions_len))
1109 goto error_free_actions;
1111 err = validate_actions(actions->actions, execute->actions_len);
1113 goto error_free_actions;
1116 skb = alloc_skb(execute->length, GFP_KERNEL);
1118 goto error_free_actions;
1121 if (copy_from_user(skb_put(skb, execute->length),
1122 (const void __user *)execute->data,
1124 goto error_free_skb;
1126 skb_reset_mac_header(skb);
1129 /* Normally, setting the skb 'protocol' field would be handled by a
1130 * call to eth_type_trans(), but it assumes there's a sending
1131 * device, which we may not have. */
1132 if (ntohs(eth->h_proto) >= 1536)
1133 skb->protocol = eth->h_proto;
1135 skb->protocol = htons(ETH_P_802_2);
1137 err = flow_extract(skb, -1, &key, &is_frag);
1139 goto error_free_skb;
1142 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1156 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1158 struct odp_execute execute;
1160 if (copy_from_user(&execute, executep, sizeof execute))
1163 return do_execute(dp, &execute);
1166 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1168 struct tbl *table = rcu_dereference(dp->table);
1169 struct odp_stats stats;
1172 stats.n_flows = tbl_count(table);
1173 stats.cur_capacity = tbl_n_buckets(table);
1174 stats.max_capacity = TBL_MAX_BUCKETS;
1175 stats.n_ports = dp->n_ports;
1176 stats.max_ports = DP_MAX_PORTS;
1177 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1178 for_each_possible_cpu(i) {
1179 const struct dp_stats_percpu *percpu_stats;
1180 struct dp_stats_percpu local_stats;
1183 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1186 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1187 local_stats = *percpu_stats;
1188 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1190 stats.n_frags += local_stats.n_frags;
1191 stats.n_hit += local_stats.n_hit;
1192 stats.n_missed += local_stats.n_missed;
1193 stats.n_lost += local_stats.n_lost;
1195 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1196 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1197 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1200 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1201 int dp_min_mtu(const struct datapath *dp)
1208 list_for_each_entry_rcu (p, &dp->port_list, node) {
1211 /* Skip any internal ports, since that's what we're trying to
1213 if (is_internal_vport(p))
1216 dev_mtu = vport_get_mtu(p);
1217 if (!mtu || dev_mtu < mtu)
1221 return mtu ? mtu : ETH_DATA_LEN;
1224 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1225 * be called with RTNL lock. */
1226 void set_internal_devs_mtu(const struct datapath *dp)
1233 mtu = dp_min_mtu(dp);
1235 list_for_each_entry_rcu (p, &dp->port_list, node) {
1236 if (is_internal_vport(p))
1237 vport_set_mtu(p, mtu);
1241 static int put_port(const struct vport *p, struct odp_port __user *uop)
1245 memset(&op, 0, sizeof op);
1248 strncpy(op.devname, vport_get_name(p), sizeof op.devname);
1249 strncpy(op.type, vport_get_type(p), sizeof op.type);
1252 op.port = p->port_no;
1254 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1257 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1259 struct odp_port port;
1261 if (copy_from_user(&port, uport, sizeof port))
1264 if (port.devname[0]) {
1265 struct vport *vport;
1268 port.devname[IFNAMSIZ - 1] = '\0';
1273 vport = vport_locate(port.devname);
1278 if (vport->dp != dp) {
1283 port.port = vport->port_no;
1292 if (port.port >= DP_MAX_PORTS)
1294 if (!dp->ports[port.port])
1298 return put_port(dp->ports[port.port], uport);
1301 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1308 list_for_each_entry_rcu (p, &dp->port_list, node) {
1309 if (put_port(p, &uports[idx]))
1311 if (idx++ >= n_ports)
1318 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1320 struct odp_portvec pv;
1323 if (copy_from_user(&pv, upv, sizeof pv))
1326 retval = do_list_ports(dp, (struct odp_port __user *)pv.ports,
1331 return put_user(retval, &upv->n_ports);
1334 static int get_listen_mask(const struct file *f)
1336 return (long)f->private_data;
1339 static void set_listen_mask(struct file *f, int listen_mask)
1341 f->private_data = (void*)(long)listen_mask;
1344 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1347 int dp_idx = iminor(f->f_dentry->d_inode);
1348 struct datapath *dp;
1349 int drop_frags, listeners, port_no;
1350 unsigned int sflow_probability;
1353 /* Handle commands with special locking requirements up front. */
1356 err = create_dp(dp_idx, (char __user *)argp);
1359 case ODP_DP_DESTROY:
1360 err = destroy_dp(dp_idx);
1363 case ODP_VPORT_ATTACH:
1364 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1367 case ODP_VPORT_DETACH:
1368 err = get_user(port_no, (int __user *)argp);
1370 err = detach_port(dp_idx, port_no);
1374 err = vport_user_mod((struct odp_port __user *)argp);
1377 case ODP_VPORT_STATS_GET:
1378 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1381 case ODP_VPORT_STATS_SET:
1382 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1385 case ODP_VPORT_ETHER_GET:
1386 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1389 case ODP_VPORT_ETHER_SET:
1390 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1393 case ODP_VPORT_MTU_GET:
1394 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1397 case ODP_VPORT_MTU_SET:
1398 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1402 dp = get_dp_locked(dp_idx);
1409 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1412 case ODP_GET_DROP_FRAGS:
1413 err = put_user(dp->drop_frags, (int __user *)argp);
1416 case ODP_SET_DROP_FRAGS:
1417 err = get_user(drop_frags, (int __user *)argp);
1421 if (drop_frags != 0 && drop_frags != 1)
1423 dp->drop_frags = drop_frags;
1427 case ODP_GET_LISTEN_MASK:
1428 err = put_user(get_listen_mask(f), (int __user *)argp);
1431 case ODP_SET_LISTEN_MASK:
1432 err = get_user(listeners, (int __user *)argp);
1436 if (listeners & ~ODPL_ALL)
1439 set_listen_mask(f, listeners);
1442 case ODP_GET_SFLOW_PROBABILITY:
1443 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1446 case ODP_SET_SFLOW_PROBABILITY:
1447 err = get_user(sflow_probability, (unsigned int __user *)argp);
1449 dp->sflow_probability = sflow_probability;
1452 case ODP_VPORT_QUERY:
1453 err = query_port(dp, (struct odp_port __user *)argp);
1456 case ODP_VPORT_LIST:
1457 err = list_ports(dp, (struct odp_portvec __user *)argp);
1460 case ODP_FLOW_FLUSH:
1461 err = flush_flows(dp);
1465 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1469 err = del_flow(dp, (struct odp_flow __user *)argp);
1473 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1477 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1481 err = execute_packet(dp, (struct odp_execute __user *)argp);
1488 mutex_unlock(&dp->mutex);
1493 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1496 for (i = 0; i < DP_N_QUEUES; i++) {
1497 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1503 #ifdef CONFIG_COMPAT
1504 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1506 struct compat_odp_portvec pv;
1509 if (copy_from_user(&pv, upv, sizeof pv))
1512 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1516 return put_user(retval, &upv->n_ports);
1519 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1521 compat_uptr_t actions;
1523 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1524 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1525 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1526 __get_user(actions, &compat->actions) ||
1527 __get_user(flow->actions_len, &compat->actions_len) ||
1528 __get_user(flow->flags, &compat->flags))
1531 flow->actions = (struct nlattr __force *)compat_ptr(actions);
1535 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1537 struct odp_flow_stats stats;
1538 struct odp_flow_put fp;
1541 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1542 get_user(fp.flags, &ufp->flags))
1545 error = do_put_flow(dp, &fp, &stats);
1549 if (copy_to_user(&ufp->flow.stats, &stats,
1550 sizeof(struct odp_flow_stats)))
1556 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1557 struct compat_odp_flow __user *ufp)
1559 compat_uptr_t actions;
1561 if (get_user(actions, &ufp->actions))
1564 return do_answer_query(flow, query_flags, &ufp->stats,
1565 compat_ptr(actions), &ufp->actions_len);
1568 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1570 struct sw_flow *flow;
1574 if (compat_get_flow(&uf, ufp))
1577 flow = do_del_flow(dp, &uf.key);
1579 return PTR_ERR(flow);
1581 error = compat_answer_query(flow, 0, ufp);
1582 flow_deferred_free(flow);
1586 static int compat_query_flows(struct datapath *dp,
1587 struct compat_odp_flow __user *flows,
1590 struct tbl *table = rcu_dereference(dp->table);
1593 for (i = 0; i < n_flows; i++) {
1594 struct compat_odp_flow __user *ufp = &flows[i];
1596 struct tbl_node *flow_node;
1599 if (compat_get_flow(&uf, ufp))
1602 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1604 error = put_user(ENOENT, &ufp->stats.error);
1606 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1613 struct compat_list_flows_cbdata {
1614 struct compat_odp_flow __user *uflows;
1619 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1621 struct sw_flow *flow = flow_cast(node);
1622 struct compat_list_flows_cbdata *cbdata = cbdata_;
1623 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1626 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1628 error = compat_answer_query(flow, 0, ufp);
1632 if (cbdata->listed_flows >= cbdata->n_flows)
1633 return cbdata->listed_flows;
1637 static int compat_list_flows(struct datapath *dp,
1638 struct compat_odp_flow __user *flows, u32 n_flows)
1640 struct compat_list_flows_cbdata cbdata;
1646 cbdata.uflows = flows;
1647 cbdata.n_flows = n_flows;
1648 cbdata.listed_flows = 0;
1650 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1651 return error ? error : cbdata.listed_flows;
1654 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1655 int (*function)(struct datapath *,
1656 struct compat_odp_flow __user *,
1659 struct compat_odp_flowvec __user *uflowvec;
1660 struct compat_odp_flow __user *flows;
1661 struct compat_odp_flowvec flowvec;
1664 uflowvec = compat_ptr(argp);
1665 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1666 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1669 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1672 flows = compat_ptr(flowvec.flows);
1673 if (!access_ok(VERIFY_WRITE, flows,
1674 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1677 retval = function(dp, flows, flowvec.n_flows);
1678 return (retval < 0 ? retval
1679 : retval == flowvec.n_flows ? 0
1680 : put_user(retval, &uflowvec->n_flows));
1683 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1685 struct odp_execute execute;
1686 compat_uptr_t actions;
1689 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1690 __get_user(actions, &uexecute->actions) ||
1691 __get_user(execute.actions_len, &uexecute->actions_len) ||
1692 __get_user(data, &uexecute->data) ||
1693 __get_user(execute.length, &uexecute->length))
1696 execute.actions = (struct nlattr __force *)compat_ptr(actions);
1697 execute.data = (const void __force *)compat_ptr(data);
1699 return do_execute(dp, &execute);
1702 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1704 int dp_idx = iminor(f->f_dentry->d_inode);
1705 struct datapath *dp;
1709 case ODP_DP_DESTROY:
1710 case ODP_FLOW_FLUSH:
1711 /* Ioctls that don't need any translation at all. */
1712 return openvswitch_ioctl(f, cmd, argp);
1715 case ODP_VPORT_ATTACH:
1716 case ODP_VPORT_DETACH:
1718 case ODP_VPORT_MTU_SET:
1719 case ODP_VPORT_MTU_GET:
1720 case ODP_VPORT_ETHER_SET:
1721 case ODP_VPORT_ETHER_GET:
1722 case ODP_VPORT_STATS_SET:
1723 case ODP_VPORT_STATS_GET:
1725 case ODP_GET_DROP_FRAGS:
1726 case ODP_SET_DROP_FRAGS:
1727 case ODP_SET_LISTEN_MASK:
1728 case ODP_GET_LISTEN_MASK:
1729 case ODP_SET_SFLOW_PROBABILITY:
1730 case ODP_GET_SFLOW_PROBABILITY:
1731 case ODP_VPORT_QUERY:
1732 /* Ioctls that just need their pointer argument extended. */
1733 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1736 dp = get_dp_locked(dp_idx);
1742 case ODP_VPORT_LIST32:
1743 err = compat_list_ports(dp, compat_ptr(argp));
1746 case ODP_FLOW_PUT32:
1747 err = compat_put_flow(dp, compat_ptr(argp));
1750 case ODP_FLOW_DEL32:
1751 err = compat_del_flow(dp, compat_ptr(argp));
1754 case ODP_FLOW_GET32:
1755 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1758 case ODP_FLOW_LIST32:
1759 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1763 err = compat_execute(dp, compat_ptr(argp));
1770 mutex_unlock(&dp->mutex);
1776 /* Unfortunately this function is not exported so this is a verbatim copy
1777 * from net/core/datagram.c in 2.6.30. */
1778 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1779 u8 __user *to, int len,
1782 int start = skb_headlen(skb);
1784 int i, copy = start - offset;
1791 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1795 if ((len -= copy) == 0)
1802 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1805 WARN_ON(start > offset + len);
1807 end = start + skb_shinfo(skb)->frags[i].size;
1808 if ((copy = end - offset) > 0) {
1812 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1813 struct page *page = frag->page;
1818 csum2 = csum_and_copy_to_user(vaddr +
1825 *csump = csum_block_add(*csump, csum2, pos);
1835 if (skb_shinfo(skb)->frag_list) {
1836 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1838 for (; list; list=list->next) {
1841 WARN_ON(start > offset + len);
1843 end = start + list->len;
1844 if ((copy = end - offset) > 0) {
1848 if (skb_copy_and_csum_datagram(list,
1853 *csump = csum_block_add(*csump, csum2, pos);
1854 if ((len -= copy) == 0)
1870 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1871 size_t nbytes, loff_t *ppos)
1873 int listeners = get_listen_mask(f);
1874 int dp_idx = iminor(f->f_dentry->d_inode);
1875 struct datapath *dp = get_dp_locked(dp_idx);
1876 struct sk_buff *skb;
1877 size_t copy_bytes, tot_copy_bytes;
1883 if (nbytes == 0 || !listeners)
1889 for (i = 0; i < DP_N_QUEUES; i++) {
1890 if (listeners & (1 << i)) {
1891 skb = skb_dequeue(&dp->queues[i]);
1897 if (f->f_flags & O_NONBLOCK) {
1902 wait_event_interruptible(dp->waitqueue,
1903 dp_has_packet_of_interest(dp,
1906 if (signal_pending(current)) {
1907 retval = -ERESTARTSYS;
1912 mutex_unlock(&dp->mutex);
1914 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1917 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1918 if (copy_bytes == skb->len) {
1920 u16 csum_start, csum_offset;
1922 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1923 BUG_ON(csum_start >= skb_headlen(skb));
1924 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1925 copy_bytes - csum_start, &csum);
1927 __sum16 __user *csump;
1929 copy_bytes = csum_start;
1930 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1932 BUG_ON((char __user *)csump + sizeof(__sum16) >
1934 put_user(csum_fold(csum), csump);
1937 retval = skb_checksum_help(skb);
1944 iov.iov_len = copy_bytes;
1945 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1949 retval = tot_copy_bytes;
1955 mutex_unlock(&dp->mutex);
1959 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1961 int dp_idx = iminor(file->f_dentry->d_inode);
1962 struct datapath *dp = get_dp_locked(dp_idx);
1967 poll_wait(file, &dp->waitqueue, wait);
1968 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1969 mask |= POLLIN | POLLRDNORM;
1970 mutex_unlock(&dp->mutex);
1972 mask = POLLIN | POLLRDNORM | POLLHUP;
1977 static struct file_operations openvswitch_fops = {
1978 .read = openvswitch_read,
1979 .poll = openvswitch_poll,
1980 .unlocked_ioctl = openvswitch_ioctl,
1981 #ifdef CONFIG_COMPAT
1982 .compat_ioctl = openvswitch_compat_ioctl,
1988 static int __init dp_init(void)
1990 struct sk_buff *dummy_skb;
1993 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1995 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2003 goto error_flow_exit;
2005 err = register_netdevice_notifier(&dp_device_notifier);
2007 goto error_vport_exit;
2009 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2011 goto error_unreg_notifier;
2015 error_unreg_notifier:
2016 unregister_netdevice_notifier(&dp_device_notifier);
2025 static void dp_cleanup(void)
2028 unregister_chrdev(major, "openvswitch");
2029 unregister_netdevice_notifier(&dp_device_notifier);
2034 module_init(dp_init);
2035 module_exit(dp_cleanup);
2037 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2038 MODULE_LICENSE("GPL");