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 *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(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, vport_get_ifindex(dp->ports[ODPP_LOCAL]));
143 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
144 #ifdef IFLA_OPERSTATE
145 NLA_PUT_U8(skb, IFLA_OPERSTATE,
146 vport_is_running(port)
147 ? vport_get_operstate(port)
151 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
153 if (ifindex != iflink)
154 NLA_PUT_U32(skb, IFLA_LINK,iflink);
156 return nlmsg_end(skb, nlh);
159 nlmsg_cancel(skb, nlh);
163 static void dp_ifinfo_notify(int event, struct vport *port)
168 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
172 err = dp_fill_ifinfo(skb, port, event, 0);
174 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
175 WARN_ON(err == -EMSGSIZE);
179 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
183 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
186 static void release_dp(struct kobject *kobj)
188 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
192 static struct kobj_type dp_ktype = {
193 .release = release_dp
196 static int create_dp(int dp_idx, const char __user *devnamep)
198 struct odp_port internal_dev_port;
199 char devname[IFNAMSIZ];
205 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
209 } else if (retval >= IFNAMSIZ) {
214 snprintf(devname, sizeof devname, "of%d", dp_idx);
218 mutex_lock(&dp_mutex);
220 if (!try_module_get(THIS_MODULE))
223 /* Exit early if a datapath with that number already exists.
224 * (We don't use -EEXIST because that's ambiguous with 'devname'
225 * conflicting with an existing network device name.) */
231 dp = kzalloc(sizeof *dp, GFP_KERNEL);
234 INIT_LIST_HEAD(&dp->port_list);
235 mutex_init(&dp->mutex);
237 for (i = 0; i < DP_N_QUEUES; i++)
238 skb_queue_head_init(&dp->queues[i]);
239 init_waitqueue_head(&dp->waitqueue);
241 /* Initialize kobject for bridge. This will be added as
242 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
243 dp->ifobj.kset = NULL;
244 kobject_init(&dp->ifobj, &dp_ktype);
246 /* Allocate table. */
248 rcu_assign_pointer(dp->table, tbl_create(0));
252 /* Set up our datapath device. */
253 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
254 strcpy(internal_dev_port.devname, devname);
255 strcpy(internal_dev_port.type, "internal");
256 err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
261 goto err_destroy_table;
265 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
266 if (!dp->stats_percpu)
267 goto err_destroy_local_port;
269 rcu_assign_pointer(dps[dp_idx], dp);
272 mutex_unlock(&dp_mutex);
277 err_destroy_local_port:
278 dp_detach_port(dp->ports[ODPP_LOCAL]);
280 tbl_destroy(dp->table, NULL);
284 module_put(THIS_MODULE);
286 mutex_unlock(&dp_mutex);
292 static void do_destroy_dp(struct datapath *dp)
297 list_for_each_entry_safe (p, n, &dp->port_list, node)
298 if (p->port_no != ODPP_LOCAL)
303 rcu_assign_pointer(dps[dp->dp_idx], NULL);
305 dp_detach_port(dp->ports[ODPP_LOCAL]);
307 tbl_destroy(dp->table, flow_free_tbl);
309 for (i = 0; i < DP_N_QUEUES; i++)
310 skb_queue_purge(&dp->queues[i]);
311 free_percpu(dp->stats_percpu);
312 kobject_put(&dp->ifobj);
313 module_put(THIS_MODULE);
316 static int destroy_dp(int dp_idx)
322 mutex_lock(&dp_mutex);
332 mutex_unlock(&dp_mutex);
337 /* Called with RTNL lock and dp_mutex. */
338 static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
340 struct vport_parms parms;
343 parms.name = odp_port->devname;
344 parms.type = odp_port->type;
345 parms.config = odp_port->config;
347 parms.port_no = port_no;
350 vport = vport_add(&parms);
354 return PTR_ERR(vport);
356 rcu_assign_pointer(dp->ports[port_no], vport);
357 list_add_rcu(&vport->node, &dp->port_list);
360 dp_ifinfo_notify(RTM_NEWLINK, vport);
365 static int attach_port(int dp_idx, struct odp_port __user *portp)
368 struct odp_port port;
373 if (copy_from_user(&port, portp, sizeof port))
375 port.devname[IFNAMSIZ - 1] = '\0';
376 port.type[VPORT_TYPE_SIZE - 1] = '\0';
379 dp = get_dp_locked(dp_idx);
382 goto out_unlock_rtnl;
384 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
385 if (!dp->ports[port_no])
391 err = new_vport(dp, &port, port_no);
395 set_internal_devs_mtu(dp);
396 dp_sysfs_add_if(dp->ports[port_no]);
398 err = put_user(port_no, &portp->port);
401 mutex_unlock(&dp->mutex);
408 int dp_detach_port(struct vport *p)
414 if (p->port_no != ODPP_LOCAL)
416 dp_ifinfo_notify(RTM_DELLINK, p);
418 /* First drop references to device. */
420 list_del_rcu(&p->node);
421 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
423 /* Then destroy it. */
431 static int detach_port(int dp_idx, int port_no)
438 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
442 dp = get_dp_locked(dp_idx);
445 goto out_unlock_rtnl;
447 p = dp->ports[port_no];
452 err = dp_detach_port(p);
455 mutex_unlock(&dp->mutex);
462 /* Must be called with rcu_read_lock. */
463 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
465 struct datapath *dp = p->dp;
466 struct dp_stats_percpu *stats;
467 int stats_counter_off;
468 struct sw_flow_actions *acts;
469 struct loop_counter *loop;
472 OVS_CB(skb)->vport = p;
474 if (!OVS_CB(skb)->flow) {
475 struct odp_flow_key key;
476 struct tbl_node *flow_node;
479 /* Extract flow from 'skb' into 'key'. */
480 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
481 if (unlikely(error)) {
486 if (is_frag && dp->drop_frags) {
488 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
493 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
494 flow_hash(&key), flow_cmp);
495 if (unlikely(!flow_node)) {
496 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
497 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
501 OVS_CB(skb)->flow = flow_cast(flow_node);
504 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
505 flow_used(OVS_CB(skb)->flow, skb);
507 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
509 /* Check whether we've looped too much. */
510 loop = loop_get_counter();
511 if (unlikely(++loop->count > MAX_LOOPS))
512 loop->looping = true;
513 if (unlikely(loop->looping)) {
514 loop_suppress(dp, acts);
519 /* Execute actions. */
520 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
523 /* Check whether sub-actions looped too much. */
524 if (unlikely(loop->looping))
525 loop_suppress(dp, acts);
528 /* Decrement loop counter. */
530 loop->looping = false;
534 /* Update datapath statistics. */
536 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
538 write_seqcount_begin(&stats->seqlock);
539 (*(u64 *)((u8 *)stats + stats_counter_off))++;
540 write_seqcount_end(&stats->seqlock);
545 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
546 * unless we broke up a GSO packet. */
547 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
548 int queue_no, u64 arg)
550 struct sk_buff *nskb;
554 if (OVS_CB(skb)->vport)
555 port_no = OVS_CB(skb)->vport->port_no;
557 port_no = ODPP_LOCAL;
560 struct odp_msg *header;
565 err = skb_cow(skb, sizeof *header);
569 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
570 header->type = queue_no;
571 header->length = skb->len;
572 header->port = port_no;
574 skb_queue_tail(queue, skb);
582 while ((skb = nskb) != NULL) {
589 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
592 struct dp_stats_percpu *stats;
593 struct sk_buff_head *queue;
596 WARN_ON_ONCE(skb_shared(skb));
597 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
598 queue = &dp->queues[queue_no];
600 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
603 forward_ip_summed(skb);
605 err = vswitch_skb_checksum_setup(skb);
609 /* Break apart GSO packets into their component pieces. Otherwise
610 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
611 if (skb_is_gso(skb)) {
612 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
622 err = queue_control_packets(skb, queue, queue_no, arg);
623 wake_up_interruptible(&dp->waitqueue);
630 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
632 write_seqcount_begin(&stats->seqlock);
634 write_seqcount_end(&stats->seqlock);
641 static int flush_flows(struct datapath *dp)
643 struct tbl *old_table = rcu_dereference(dp->table);
644 struct tbl *new_table;
646 new_table = tbl_create(0);
650 rcu_assign_pointer(dp->table, new_table);
652 tbl_deferred_destroy(old_table, flow_free_tbl);
657 static int validate_actions(const struct nlattr *actions, u32 actions_len)
659 const struct nlattr *a;
662 nla_for_each_attr(a, actions, actions_len, rem) {
663 static const u32 action_lens[ODPAT_MAX + 1] = {
665 [ODPAT_CONTROLLER] = 8,
666 [ODPAT_SET_DL_TCI] = 2,
667 [ODPAT_STRIP_VLAN] = 0,
668 [ODPAT_SET_DL_SRC] = ETH_ALEN,
669 [ODPAT_SET_DL_DST] = ETH_ALEN,
670 [ODPAT_SET_NW_SRC] = 4,
671 [ODPAT_SET_NW_DST] = 4,
672 [ODPAT_SET_NW_TOS] = 1,
673 [ODPAT_SET_TP_SRC] = 2,
674 [ODPAT_SET_TP_DST] = 2,
675 [ODPAT_SET_TUNNEL] = 8,
676 [ODPAT_SET_PRIORITY] = 4,
677 [ODPAT_POP_PRIORITY] = 0,
678 [ODPAT_DROP_SPOOFED_ARP] = 0,
680 int type = nla_type(a);
682 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
689 case ODPAT_CONTROLLER:
690 case ODPAT_STRIP_VLAN:
691 case ODPAT_SET_DL_SRC:
692 case ODPAT_SET_DL_DST:
693 case ODPAT_SET_NW_SRC:
694 case ODPAT_SET_NW_DST:
695 case ODPAT_SET_TP_SRC:
696 case ODPAT_SET_TP_DST:
697 case ODPAT_SET_TUNNEL:
698 case ODPAT_SET_PRIORITY:
699 case ODPAT_POP_PRIORITY:
700 case ODPAT_DROP_SPOOFED_ARP:
701 /* No validation needed. */
705 if (nla_get_u32(a) >= DP_MAX_PORTS)
708 case ODPAT_SET_DL_TCI:
709 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
713 case ODPAT_SET_NW_TOS:
714 if (nla_get_u8(a) & INET_ECN_MASK)
729 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
731 struct sw_flow_actions *actions;
734 actions = flow_actions_alloc(flow->actions_len);
735 error = PTR_ERR(actions);
740 if (copy_from_user(actions->actions, flow->actions, flow->actions_len))
741 goto error_free_actions;
742 error = validate_actions(actions->actions, actions->actions_len);
744 goto error_free_actions;
751 return ERR_PTR(error);
754 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
757 struct timespec offset_ts, used, now_mono;
759 ktime_get_ts(&now_mono);
760 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
761 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
762 now_mono.tv_nsec - offset_ts.tv_nsec);
764 stats->used_sec = used.tv_sec;
765 stats->used_nsec = used.tv_nsec;
768 stats->used_nsec = 0;
771 stats->n_packets = flow->packet_count;
772 stats->n_bytes = flow->byte_count;
774 stats->tcp_flags = flow->tcp_flags;
778 static void clear_stats(struct sw_flow *flow)
782 flow->packet_count = 0;
783 flow->byte_count = 0;
786 static int expand_table(struct datapath *dp)
788 struct tbl *old_table = rcu_dereference(dp->table);
789 struct tbl *new_table;
791 new_table = tbl_expand(old_table);
792 if (IS_ERR(new_table))
793 return PTR_ERR(new_table);
795 rcu_assign_pointer(dp->table, new_table);
796 tbl_deferred_destroy(old_table, NULL);
801 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
802 struct odp_flow_stats *stats)
804 struct tbl_node *flow_node;
805 struct sw_flow *flow;
809 table = rcu_dereference(dp->table);
810 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
813 struct sw_flow_actions *acts;
816 if (!(uf->flags & ODPPF_CREATE))
819 /* Expand table, if necessary, to make room. */
820 if (tbl_count(table) >= tbl_n_buckets(table)) {
821 error = expand_table(dp);
824 table = rcu_dereference(dp->table);
830 error = PTR_ERR(flow);
833 flow->key = uf->flow.key;
836 /* Obtain actions. */
837 acts = get_actions(&uf->flow);
838 error = PTR_ERR(acts);
840 goto error_free_flow;
841 rcu_assign_pointer(flow->sf_acts, acts);
843 /* Put flow in bucket. */
844 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
846 goto error_free_flow_acts;
848 memset(stats, 0, sizeof(struct odp_flow_stats));
850 /* We found a matching flow. */
851 struct sw_flow_actions *old_acts, *new_acts;
853 flow = flow_cast(flow_node);
855 /* Bail out if we're not allowed to modify an existing flow. */
857 if (!(uf->flags & ODPPF_MODIFY))
861 new_acts = get_actions(&uf->flow);
862 error = PTR_ERR(new_acts);
863 if (IS_ERR(new_acts))
865 old_acts = rcu_dereference(flow->sf_acts);
866 if (old_acts->actions_len != new_acts->actions_len ||
867 memcmp(old_acts->actions, new_acts->actions,
868 old_acts->actions_len)) {
869 rcu_assign_pointer(flow->sf_acts, new_acts);
870 flow_deferred_free_acts(old_acts);
875 /* Fetch stats, then clear them if necessary. */
876 spin_lock_bh(&flow->lock);
877 get_stats(flow, stats);
878 if (uf->flags & ODPPF_ZERO_STATS)
880 spin_unlock_bh(&flow->lock);
885 error_free_flow_acts:
886 kfree(flow->sf_acts);
888 flow->sf_acts = NULL;
894 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
896 struct odp_flow_stats stats;
897 struct odp_flow_put uf;
900 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
903 error = do_put_flow(dp, &uf, &stats);
907 if (copy_to_user(&ufp->flow.stats, &stats,
908 sizeof(struct odp_flow_stats)))
914 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
915 struct odp_flow_stats __user *ustats,
916 struct nlattr __user *actions,
917 u32 __user *actions_lenp)
919 struct sw_flow_actions *sf_acts;
920 struct odp_flow_stats stats;
923 spin_lock_bh(&flow->lock);
924 get_stats(flow, &stats);
925 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
928 spin_unlock_bh(&flow->lock);
930 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
931 get_user(actions_len, actions_lenp))
937 sf_acts = rcu_dereference(flow->sf_acts);
938 if (put_user(sf_acts->actions_len, actions_lenp) ||
939 (actions && copy_to_user(actions, sf_acts->actions,
940 min(sf_acts->actions_len, actions_len))))
946 static int answer_query(struct sw_flow *flow, u32 query_flags,
947 struct odp_flow __user *ufp)
949 struct nlattr *actions;
951 if (get_user(actions, &ufp->actions))
954 return do_answer_query(flow, query_flags,
955 &ufp->stats, actions, &ufp->actions_len);
958 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
960 struct tbl *table = rcu_dereference(dp->table);
961 struct tbl_node *flow_node;
964 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
966 return ERR_PTR(-ENOENT);
968 error = tbl_remove(table, flow_node);
970 return ERR_PTR(error);
972 /* XXX Returned flow_node's statistics might lose a few packets, since
973 * other CPUs can be using this flow. We used to synchronize_rcu() to
974 * make sure that we get completely accurate stats, but that blows our
975 * performance, badly. */
976 return flow_cast(flow_node);
979 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
981 struct sw_flow *flow;
985 if (copy_from_user(&uf, ufp, sizeof uf))
988 flow = do_del_flow(dp, &uf.key);
990 return PTR_ERR(flow);
992 error = answer_query(flow, 0, ufp);
993 flow_deferred_free(flow);
997 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
999 struct tbl *table = rcu_dereference(dp->table);
1002 for (i = 0; i < flowvec->n_flows; i++) {
1003 struct odp_flow __user *ufp = &flowvec->flows[i];
1005 struct tbl_node *flow_node;
1008 if (copy_from_user(&uf, ufp, sizeof uf))
1011 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1013 error = put_user(ENOENT, &ufp->stats.error);
1015 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1019 return flowvec->n_flows;
1022 struct list_flows_cbdata {
1023 struct odp_flow __user *uflows;
1028 static int list_flow(struct tbl_node *node, void *cbdata_)
1030 struct sw_flow *flow = flow_cast(node);
1031 struct list_flows_cbdata *cbdata = cbdata_;
1032 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1035 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1037 error = answer_query(flow, 0, ufp);
1041 if (cbdata->listed_flows >= cbdata->n_flows)
1042 return cbdata->listed_flows;
1046 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1048 struct list_flows_cbdata cbdata;
1051 if (!flowvec->n_flows)
1054 cbdata.uflows = flowvec->flows;
1055 cbdata.n_flows = flowvec->n_flows;
1056 cbdata.listed_flows = 0;
1058 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1059 return error ? error : cbdata.listed_flows;
1062 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1063 int (*function)(struct datapath *,
1064 const struct odp_flowvec *))
1066 struct odp_flowvec __user *uflowvec;
1067 struct odp_flowvec flowvec;
1070 uflowvec = (struct odp_flowvec __user *)argp;
1071 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1074 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1077 retval = function(dp, &flowvec);
1078 return (retval < 0 ? retval
1079 : retval == flowvec.n_flows ? 0
1080 : put_user(retval, &uflowvec->n_flows));
1083 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1085 struct odp_flow_key key;
1086 struct sk_buff *skb;
1087 struct sw_flow_actions *actions;
1093 if (execute->length < ETH_HLEN || execute->length > 65535)
1096 actions = flow_actions_alloc(execute->actions_len);
1097 if (IS_ERR(actions)) {
1098 err = PTR_ERR(actions);
1103 if (copy_from_user(actions->actions, execute->actions, execute->actions_len))
1104 goto error_free_actions;
1106 err = validate_actions(actions->actions, execute->actions_len);
1108 goto error_free_actions;
1111 skb = alloc_skb(execute->length, GFP_KERNEL);
1113 goto error_free_actions;
1116 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1118 goto error_free_skb;
1120 skb_reset_mac_header(skb);
1123 /* Normally, setting the skb 'protocol' field would be handled by a
1124 * call to eth_type_trans(), but it assumes there's a sending
1125 * device, which we may not have. */
1126 if (ntohs(eth->h_proto) >= 1536)
1127 skb->protocol = eth->h_proto;
1129 skb->protocol = htons(ETH_P_802_2);
1131 err = flow_extract(skb, -1, &key, &is_frag);
1133 goto error_free_skb;
1136 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1150 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1152 struct odp_execute execute;
1154 if (copy_from_user(&execute, executep, sizeof execute))
1157 return do_execute(dp, &execute);
1160 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1162 struct tbl *table = rcu_dereference(dp->table);
1163 struct odp_stats stats;
1166 stats.n_flows = tbl_count(table);
1167 stats.cur_capacity = tbl_n_buckets(table);
1168 stats.max_capacity = TBL_MAX_BUCKETS;
1169 stats.n_ports = dp->n_ports;
1170 stats.max_ports = DP_MAX_PORTS;
1171 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1172 for_each_possible_cpu(i) {
1173 const struct dp_stats_percpu *percpu_stats;
1174 struct dp_stats_percpu local_stats;
1177 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1180 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1181 local_stats = *percpu_stats;
1182 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1184 stats.n_frags += local_stats.n_frags;
1185 stats.n_hit += local_stats.n_hit;
1186 stats.n_missed += local_stats.n_missed;
1187 stats.n_lost += local_stats.n_lost;
1189 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1190 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1191 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1194 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1195 int dp_min_mtu(const struct datapath *dp)
1202 list_for_each_entry_rcu (p, &dp->port_list, node) {
1205 /* Skip any internal ports, since that's what we're trying to
1207 if (is_internal_vport(p))
1210 dev_mtu = vport_get_mtu(p);
1211 if (!mtu || dev_mtu < mtu)
1215 return mtu ? mtu : ETH_DATA_LEN;
1218 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1219 * be called with RTNL lock. */
1220 void set_internal_devs_mtu(const struct datapath *dp)
1227 mtu = dp_min_mtu(dp);
1229 list_for_each_entry_rcu (p, &dp->port_list, node) {
1230 if (is_internal_vport(p))
1231 vport_set_mtu(p, mtu);
1235 static int put_port(const struct vport *p, struct odp_port __user *uop)
1239 memset(&op, 0, sizeof op);
1242 strncpy(op.devname, vport_get_name(p), sizeof op.devname);
1243 strncpy(op.type, vport_get_type(p), sizeof op.type);
1246 op.port = p->port_no;
1248 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1251 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1253 struct odp_port port;
1255 if (copy_from_user(&port, uport, sizeof port))
1258 if (port.devname[0]) {
1259 struct vport *vport;
1262 port.devname[IFNAMSIZ - 1] = '\0';
1267 vport = vport_locate(port.devname);
1272 if (vport->dp != dp) {
1277 port.port = vport->port_no;
1286 if (port.port >= DP_MAX_PORTS)
1288 if (!dp->ports[port.port])
1292 return put_port(dp->ports[port.port], uport);
1295 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1302 list_for_each_entry_rcu (p, &dp->port_list, node) {
1303 if (put_port(p, &uports[idx]))
1305 if (idx++ >= n_ports)
1312 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1314 struct odp_portvec pv;
1317 if (copy_from_user(&pv, upv, sizeof pv))
1320 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1324 return put_user(retval, &upv->n_ports);
1327 static int get_listen_mask(const struct file *f)
1329 return (long)f->private_data;
1332 static void set_listen_mask(struct file *f, int listen_mask)
1334 f->private_data = (void*)(long)listen_mask;
1337 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1340 int dp_idx = iminor(f->f_dentry->d_inode);
1341 struct datapath *dp;
1342 int drop_frags, listeners, port_no;
1343 unsigned int sflow_probability;
1346 /* Handle commands with special locking requirements up front. */
1349 err = create_dp(dp_idx, (char __user *)argp);
1352 case ODP_DP_DESTROY:
1353 err = destroy_dp(dp_idx);
1356 case ODP_VPORT_ATTACH:
1357 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1360 case ODP_VPORT_DETACH:
1361 err = get_user(port_no, (int __user *)argp);
1363 err = detach_port(dp_idx, port_no);
1367 err = vport_user_mod((struct odp_port __user *)argp);
1370 case ODP_VPORT_STATS_GET:
1371 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1374 case ODP_VPORT_STATS_SET:
1375 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1378 case ODP_VPORT_ETHER_GET:
1379 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1382 case ODP_VPORT_ETHER_SET:
1383 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1386 case ODP_VPORT_MTU_GET:
1387 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1390 case ODP_VPORT_MTU_SET:
1391 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1395 dp = get_dp_locked(dp_idx);
1402 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1405 case ODP_GET_DROP_FRAGS:
1406 err = put_user(dp->drop_frags, (int __user *)argp);
1409 case ODP_SET_DROP_FRAGS:
1410 err = get_user(drop_frags, (int __user *)argp);
1414 if (drop_frags != 0 && drop_frags != 1)
1416 dp->drop_frags = drop_frags;
1420 case ODP_GET_LISTEN_MASK:
1421 err = put_user(get_listen_mask(f), (int __user *)argp);
1424 case ODP_SET_LISTEN_MASK:
1425 err = get_user(listeners, (int __user *)argp);
1429 if (listeners & ~ODPL_ALL)
1432 set_listen_mask(f, listeners);
1435 case ODP_GET_SFLOW_PROBABILITY:
1436 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1439 case ODP_SET_SFLOW_PROBABILITY:
1440 err = get_user(sflow_probability, (unsigned int __user *)argp);
1442 dp->sflow_probability = sflow_probability;
1445 case ODP_VPORT_QUERY:
1446 err = query_port(dp, (struct odp_port __user *)argp);
1449 case ODP_VPORT_LIST:
1450 err = list_ports(dp, (struct odp_portvec __user *)argp);
1453 case ODP_FLOW_FLUSH:
1454 err = flush_flows(dp);
1458 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1462 err = del_flow(dp, (struct odp_flow __user *)argp);
1466 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1470 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1474 err = execute_packet(dp, (struct odp_execute __user *)argp);
1481 mutex_unlock(&dp->mutex);
1486 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1489 for (i = 0; i < DP_N_QUEUES; i++) {
1490 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1496 #ifdef CONFIG_COMPAT
1497 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1499 struct compat_odp_portvec pv;
1502 if (copy_from_user(&pv, upv, sizeof pv))
1505 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1509 return put_user(retval, &upv->n_ports);
1512 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1514 compat_uptr_t actions;
1516 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1517 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1518 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1519 __get_user(actions, &compat->actions) ||
1520 __get_user(flow->actions_len, &compat->actions_len) ||
1521 __get_user(flow->flags, &compat->flags))
1524 flow->actions = compat_ptr(actions);
1528 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1530 struct odp_flow_stats stats;
1531 struct odp_flow_put fp;
1534 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1535 get_user(fp.flags, &ufp->flags))
1538 error = do_put_flow(dp, &fp, &stats);
1542 if (copy_to_user(&ufp->flow.stats, &stats,
1543 sizeof(struct odp_flow_stats)))
1549 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1550 struct compat_odp_flow __user *ufp)
1552 compat_uptr_t actions;
1554 if (get_user(actions, &ufp->actions))
1557 return do_answer_query(flow, query_flags, &ufp->stats,
1558 compat_ptr(actions), &ufp->actions_len);
1561 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1563 struct sw_flow *flow;
1567 if (compat_get_flow(&uf, ufp))
1570 flow = do_del_flow(dp, &uf.key);
1572 return PTR_ERR(flow);
1574 error = compat_answer_query(flow, 0, ufp);
1575 flow_deferred_free(flow);
1579 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1581 struct tbl *table = rcu_dereference(dp->table);
1584 for (i = 0; i < n_flows; i++) {
1585 struct compat_odp_flow __user *ufp = &flows[i];
1587 struct tbl_node *flow_node;
1590 if (compat_get_flow(&uf, ufp))
1593 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1595 error = put_user(ENOENT, &ufp->stats.error);
1597 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1604 struct compat_list_flows_cbdata {
1605 struct compat_odp_flow __user *uflows;
1610 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1612 struct sw_flow *flow = flow_cast(node);
1613 struct compat_list_flows_cbdata *cbdata = cbdata_;
1614 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1617 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1619 error = compat_answer_query(flow, 0, ufp);
1623 if (cbdata->listed_flows >= cbdata->n_flows)
1624 return cbdata->listed_flows;
1628 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1630 struct compat_list_flows_cbdata cbdata;
1636 cbdata.uflows = flows;
1637 cbdata.n_flows = n_flows;
1638 cbdata.listed_flows = 0;
1640 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1641 return error ? error : cbdata.listed_flows;
1644 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1645 int (*function)(struct datapath *,
1646 struct compat_odp_flow *,
1649 struct compat_odp_flowvec __user *uflowvec;
1650 struct compat_odp_flow __user *flows;
1651 struct compat_odp_flowvec flowvec;
1654 uflowvec = compat_ptr(argp);
1655 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1656 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1659 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1662 flows = compat_ptr(flowvec.flows);
1663 if (!access_ok(VERIFY_WRITE, flows,
1664 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1667 retval = function(dp, flows, flowvec.n_flows);
1668 return (retval < 0 ? retval
1669 : retval == flowvec.n_flows ? 0
1670 : put_user(retval, &uflowvec->n_flows));
1673 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1675 struct odp_execute execute;
1676 compat_uptr_t actions;
1679 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1680 __get_user(actions, &uexecute->actions) ||
1681 __get_user(execute.actions_len, &uexecute->actions_len) ||
1682 __get_user(data, &uexecute->data) ||
1683 __get_user(execute.length, &uexecute->length))
1686 execute.actions = compat_ptr(actions);
1687 execute.data = compat_ptr(data);
1689 return do_execute(dp, &execute);
1692 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1694 int dp_idx = iminor(f->f_dentry->d_inode);
1695 struct datapath *dp;
1699 case ODP_DP_DESTROY:
1700 case ODP_FLOW_FLUSH:
1701 /* Ioctls that don't need any translation at all. */
1702 return openvswitch_ioctl(f, cmd, argp);
1705 case ODP_VPORT_ATTACH:
1706 case ODP_VPORT_DETACH:
1708 case ODP_VPORT_MTU_SET:
1709 case ODP_VPORT_MTU_GET:
1710 case ODP_VPORT_ETHER_SET:
1711 case ODP_VPORT_ETHER_GET:
1712 case ODP_VPORT_STATS_SET:
1713 case ODP_VPORT_STATS_GET:
1715 case ODP_GET_DROP_FRAGS:
1716 case ODP_SET_DROP_FRAGS:
1717 case ODP_SET_LISTEN_MASK:
1718 case ODP_GET_LISTEN_MASK:
1719 case ODP_SET_SFLOW_PROBABILITY:
1720 case ODP_GET_SFLOW_PROBABILITY:
1721 case ODP_VPORT_QUERY:
1722 /* Ioctls that just need their pointer argument extended. */
1723 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1726 dp = get_dp_locked(dp_idx);
1732 case ODP_VPORT_LIST32:
1733 err = compat_list_ports(dp, compat_ptr(argp));
1736 case ODP_FLOW_PUT32:
1737 err = compat_put_flow(dp, compat_ptr(argp));
1740 case ODP_FLOW_DEL32:
1741 err = compat_del_flow(dp, compat_ptr(argp));
1744 case ODP_FLOW_GET32:
1745 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1748 case ODP_FLOW_LIST32:
1749 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1753 err = compat_execute(dp, compat_ptr(argp));
1760 mutex_unlock(&dp->mutex);
1766 /* Unfortunately this function is not exported so this is a verbatim copy
1767 * from net/core/datagram.c in 2.6.30. */
1768 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1769 u8 __user *to, int len,
1772 int start = skb_headlen(skb);
1774 int i, copy = start - offset;
1781 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1785 if ((len -= copy) == 0)
1792 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1795 WARN_ON(start > offset + len);
1797 end = start + skb_shinfo(skb)->frags[i].size;
1798 if ((copy = end - offset) > 0) {
1802 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1803 struct page *page = frag->page;
1808 csum2 = csum_and_copy_to_user(vaddr +
1815 *csump = csum_block_add(*csump, csum2, pos);
1825 if (skb_shinfo(skb)->frag_list) {
1826 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1828 for (; list; list=list->next) {
1831 WARN_ON(start > offset + len);
1833 end = start + list->len;
1834 if ((copy = end - offset) > 0) {
1838 if (skb_copy_and_csum_datagram(list,
1843 *csump = csum_block_add(*csump, csum2, pos);
1844 if ((len -= copy) == 0)
1860 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1861 size_t nbytes, loff_t *ppos)
1863 int listeners = get_listen_mask(f);
1864 int dp_idx = iminor(f->f_dentry->d_inode);
1865 struct datapath *dp = get_dp_locked(dp_idx);
1866 struct sk_buff *skb;
1867 size_t copy_bytes, tot_copy_bytes;
1873 if (nbytes == 0 || !listeners)
1879 for (i = 0; i < DP_N_QUEUES; i++) {
1880 if (listeners & (1 << i)) {
1881 skb = skb_dequeue(&dp->queues[i]);
1887 if (f->f_flags & O_NONBLOCK) {
1892 wait_event_interruptible(dp->waitqueue,
1893 dp_has_packet_of_interest(dp,
1896 if (signal_pending(current)) {
1897 retval = -ERESTARTSYS;
1902 mutex_unlock(&dp->mutex);
1904 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1907 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1908 if (copy_bytes == skb->len) {
1910 u16 csum_start, csum_offset;
1912 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1913 BUG_ON(csum_start >= skb_headlen(skb));
1914 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1915 copy_bytes - csum_start, &csum);
1917 __sum16 __user *csump;
1919 copy_bytes = csum_start;
1920 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1922 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
1923 put_user(csum_fold(csum), csump);
1926 retval = skb_checksum_help(skb);
1930 struct iovec __user iov;
1933 iov.iov_len = copy_bytes;
1934 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1938 retval = tot_copy_bytes;
1944 mutex_unlock(&dp->mutex);
1948 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1950 int dp_idx = iminor(file->f_dentry->d_inode);
1951 struct datapath *dp = get_dp_locked(dp_idx);
1956 poll_wait(file, &dp->waitqueue, wait);
1957 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1958 mask |= POLLIN | POLLRDNORM;
1959 mutex_unlock(&dp->mutex);
1961 mask = POLLIN | POLLRDNORM | POLLHUP;
1966 static struct file_operations openvswitch_fops = {
1967 .read = openvswitch_read,
1968 .poll = openvswitch_poll,
1969 .unlocked_ioctl = openvswitch_ioctl,
1970 #ifdef CONFIG_COMPAT
1971 .compat_ioctl = openvswitch_compat_ioctl,
1977 static int __init dp_init(void)
1979 struct sk_buff *dummy_skb;
1982 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1984 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1992 goto error_flow_exit;
1994 err = register_netdevice_notifier(&dp_device_notifier);
1996 goto error_vport_exit;
1998 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2000 goto error_unreg_notifier;
2004 error_unreg_notifier:
2005 unregister_netdevice_notifier(&dp_device_notifier);
2014 static void dp_cleanup(void)
2017 unregister_chrdev(major, "openvswitch");
2018 unregister_netdevice_notifier(&dp_device_notifier);
2023 module_init(dp_init);
2024 module_exit(dp_cleanup);
2026 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2027 MODULE_LICENSE("GPL");