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_check(dps[dp_idx], rcu_read_lock_held() ||
81 lockdep_is_held(&dp_mutex));
83 EXPORT_SYMBOL_GPL(get_dp);
85 static struct datapath *get_dp_locked(int dp_idx)
89 mutex_lock(&dp_mutex);
92 mutex_lock(&dp->mutex);
93 mutex_unlock(&dp_mutex);
97 /* Must be called with rcu_read_lock or RTNL lock. */
98 const char *dp_name(const struct datapath *dp)
100 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
103 static inline size_t br_nlmsg_size(void)
105 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
106 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
107 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
108 + nla_total_size(4) /* IFLA_MASTER */
109 + nla_total_size(4) /* IFLA_MTU */
110 + nla_total_size(4) /* IFLA_LINK */
111 + nla_total_size(1); /* IFLA_OPERSTATE */
114 static int dp_fill_ifinfo(struct sk_buff *skb,
115 const struct vport *port,
116 int event, unsigned int flags)
118 const struct datapath *dp = port->dp;
119 int ifindex = vport_get_ifindex(port);
120 int iflink = vport_get_iflink(port);
121 struct ifinfomsg *hdr;
122 struct nlmsghdr *nlh;
130 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
134 hdr = nlmsg_data(nlh);
135 hdr->ifi_family = AF_BRIDGE;
137 hdr->ifi_type = ARPHRD_ETHER;
138 hdr->ifi_index = ifindex;
139 hdr->ifi_flags = vport_get_flags(port);
142 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
143 NLA_PUT_U32(skb, IFLA_MASTER,
144 vport_get_ifindex(rtnl_dereference(dp->ports[ODPP_LOCAL])));
145 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
146 #ifdef IFLA_OPERSTATE
147 NLA_PUT_U8(skb, IFLA_OPERSTATE,
148 vport_is_running(port)
149 ? vport_get_operstate(port)
153 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
155 if (ifindex != iflink)
156 NLA_PUT_U32(skb, IFLA_LINK,iflink);
158 return nlmsg_end(skb, nlh);
161 nlmsg_cancel(skb, nlh);
165 static void dp_ifinfo_notify(int event, struct vport *port)
170 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
174 err = dp_fill_ifinfo(skb, port, event, 0);
176 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
177 WARN_ON(err == -EMSGSIZE);
181 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
185 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
188 static void release_dp(struct kobject *kobj)
190 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
194 static struct kobj_type dp_ktype = {
195 .release = release_dp
198 static int create_dp(int dp_idx, const char __user *devnamep)
200 struct odp_port internal_dev_port;
201 char devname[IFNAMSIZ];
207 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
211 } else if (retval >= IFNAMSIZ) {
216 snprintf(devname, sizeof devname, "of%d", dp_idx);
220 mutex_lock(&dp_mutex);
222 if (!try_module_get(THIS_MODULE))
225 /* Exit early if a datapath with that number already exists.
226 * (We don't use -EEXIST because that's ambiguous with 'devname'
227 * conflicting with an existing network device name.) */
233 dp = kzalloc(sizeof *dp, GFP_KERNEL);
236 INIT_LIST_HEAD(&dp->port_list);
237 mutex_init(&dp->mutex);
239 for (i = 0; i < DP_N_QUEUES; i++)
240 skb_queue_head_init(&dp->queues[i]);
241 init_waitqueue_head(&dp->waitqueue);
243 /* Initialize kobject for bridge. This will be added as
244 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
245 dp->ifobj.kset = NULL;
246 kobject_init(&dp->ifobj, &dp_ktype);
248 /* Allocate table. */
250 rcu_assign_pointer(dp->table, tbl_create(0));
254 /* Set up our datapath device. */
255 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
256 strcpy(internal_dev_port.devname, devname);
257 strcpy(internal_dev_port.type, "internal");
258 err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
263 goto err_destroy_table;
267 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
268 if (!dp->stats_percpu)
269 goto err_destroy_local_port;
271 rcu_assign_pointer(dps[dp_idx], dp);
274 mutex_unlock(&dp_mutex);
279 err_destroy_local_port:
280 dp_detach_port(dp->ports[ODPP_LOCAL]);
282 tbl_destroy(dp->table, NULL);
286 module_put(THIS_MODULE);
288 mutex_unlock(&dp_mutex);
294 static void do_destroy_dp(struct datapath *dp)
299 list_for_each_entry_safe (p, n, &dp->port_list, node)
300 if (p->port_no != ODPP_LOCAL)
305 rcu_assign_pointer(dps[dp->dp_idx], NULL);
307 dp_detach_port(dp->ports[ODPP_LOCAL]);
309 tbl_destroy(dp->table, flow_free_tbl);
311 for (i = 0; i < DP_N_QUEUES; i++)
312 skb_queue_purge(&dp->queues[i]);
313 free_percpu(dp->stats_percpu);
314 kobject_put(&dp->ifobj);
315 module_put(THIS_MODULE);
318 static int destroy_dp(int dp_idx)
324 mutex_lock(&dp_mutex);
334 mutex_unlock(&dp_mutex);
339 /* Called with RTNL lock and dp_mutex. */
340 static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
342 struct vport_parms parms;
345 parms.name = odp_port->devname;
346 parms.type = odp_port->type;
347 parms.config = odp_port->config;
349 parms.port_no = port_no;
352 vport = vport_add(&parms);
356 return PTR_ERR(vport);
358 rcu_assign_pointer(dp->ports[port_no], vport);
359 list_add_rcu(&vport->node, &dp->port_list);
362 dp_ifinfo_notify(RTM_NEWLINK, vport);
367 static int attach_port(int dp_idx, struct odp_port __user *portp)
370 struct odp_port port;
375 if (copy_from_user(&port, portp, sizeof port))
377 port.devname[IFNAMSIZ - 1] = '\0';
378 port.type[VPORT_TYPE_SIZE - 1] = '\0';
381 dp = get_dp_locked(dp_idx);
384 goto out_unlock_rtnl;
386 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
387 if (!dp->ports[port_no])
393 err = new_vport(dp, &port, port_no);
397 set_internal_devs_mtu(dp);
398 dp_sysfs_add_if(dp->ports[port_no]);
400 err = put_user(port_no, &portp->port);
403 mutex_unlock(&dp->mutex);
410 int dp_detach_port(struct vport *p)
416 if (p->port_no != ODPP_LOCAL)
418 dp_ifinfo_notify(RTM_DELLINK, p);
420 /* First drop references to device. */
422 list_del_rcu(&p->node);
423 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
425 /* Then destroy it. */
433 static int detach_port(int dp_idx, int port_no)
440 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
444 dp = get_dp_locked(dp_idx);
447 goto out_unlock_rtnl;
449 p = dp->ports[port_no];
454 err = dp_detach_port(p);
457 mutex_unlock(&dp->mutex);
464 /* Must be called with rcu_read_lock. */
465 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
467 struct datapath *dp = p->dp;
468 struct dp_stats_percpu *stats;
469 int stats_counter_off;
470 struct sw_flow_actions *acts;
471 struct loop_counter *loop;
474 OVS_CB(skb)->vport = p;
476 if (!OVS_CB(skb)->flow) {
477 struct odp_flow_key key;
478 struct tbl_node *flow_node;
481 /* Extract flow from 'skb' into 'key'. */
482 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
483 if (unlikely(error)) {
488 if (is_frag && dp->drop_frags) {
490 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
495 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
496 flow_hash(&key), flow_cmp);
497 if (unlikely(!flow_node)) {
498 dp_output_control(dp, skb, _ODPL_MISS_NR,
499 (__force u64)OVS_CB(skb)->tun_id);
500 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
504 OVS_CB(skb)->flow = flow_cast(flow_node);
507 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
508 flow_used(OVS_CB(skb)->flow, skb);
510 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
512 /* Check whether we've looped too much. */
513 loop = loop_get_counter();
514 if (unlikely(++loop->count > MAX_LOOPS))
515 loop->looping = true;
516 if (unlikely(loop->looping)) {
517 loop_suppress(dp, acts);
522 /* Execute actions. */
523 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
526 /* Check whether sub-actions looped too much. */
527 if (unlikely(loop->looping))
528 loop_suppress(dp, acts);
531 /* Decrement loop counter. */
533 loop->looping = false;
537 /* Update datapath statistics. */
539 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
541 write_seqcount_begin(&stats->seqlock);
542 (*(u64 *)((u8 *)stats + stats_counter_off))++;
543 write_seqcount_end(&stats->seqlock);
548 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
549 * unless we broke up a GSO packet. */
550 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
551 int queue_no, u64 arg)
553 struct sk_buff *nskb;
557 if (OVS_CB(skb)->vport)
558 port_no = OVS_CB(skb)->vport->port_no;
560 port_no = ODPP_LOCAL;
563 struct odp_msg *header;
568 err = skb_cow(skb, sizeof *header);
572 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
573 header->type = queue_no;
574 header->length = skb->len;
575 header->port = port_no;
577 skb_queue_tail(queue, skb);
585 while ((skb = nskb) != NULL) {
592 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
595 struct dp_stats_percpu *stats;
596 struct sk_buff_head *queue;
599 WARN_ON_ONCE(skb_shared(skb));
600 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
601 queue = &dp->queues[queue_no];
603 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
606 forward_ip_summed(skb);
608 err = vswitch_skb_checksum_setup(skb);
612 /* Break apart GSO packets into their component pieces. Otherwise
613 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
614 if (skb_is_gso(skb)) {
615 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
625 err = queue_control_packets(skb, queue, queue_no, arg);
626 wake_up_interruptible(&dp->waitqueue);
633 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
635 write_seqcount_begin(&stats->seqlock);
637 write_seqcount_end(&stats->seqlock);
644 static int flush_flows(struct datapath *dp)
646 struct tbl *old_table = rcu_dereference(dp->table);
647 struct tbl *new_table;
649 new_table = tbl_create(0);
653 rcu_assign_pointer(dp->table, new_table);
655 tbl_deferred_destroy(old_table, flow_free_tbl);
660 static int validate_actions(const struct nlattr *actions, u32 actions_len)
662 const struct nlattr *a;
665 nla_for_each_attr(a, actions, actions_len, rem) {
666 static const u32 action_lens[ODPAT_MAX + 1] = {
668 [ODPAT_CONTROLLER] = 8,
669 [ODPAT_SET_DL_TCI] = 2,
670 [ODPAT_STRIP_VLAN] = 0,
671 [ODPAT_SET_DL_SRC] = ETH_ALEN,
672 [ODPAT_SET_DL_DST] = ETH_ALEN,
673 [ODPAT_SET_NW_SRC] = 4,
674 [ODPAT_SET_NW_DST] = 4,
675 [ODPAT_SET_NW_TOS] = 1,
676 [ODPAT_SET_TP_SRC] = 2,
677 [ODPAT_SET_TP_DST] = 2,
678 [ODPAT_SET_TUNNEL] = 8,
679 [ODPAT_SET_PRIORITY] = 4,
680 [ODPAT_POP_PRIORITY] = 0,
681 [ODPAT_DROP_SPOOFED_ARP] = 0,
683 int type = nla_type(a);
685 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
692 case ODPAT_CONTROLLER:
693 case ODPAT_STRIP_VLAN:
694 case ODPAT_SET_DL_SRC:
695 case ODPAT_SET_DL_DST:
696 case ODPAT_SET_NW_SRC:
697 case ODPAT_SET_NW_DST:
698 case ODPAT_SET_TP_SRC:
699 case ODPAT_SET_TP_DST:
700 case ODPAT_SET_TUNNEL:
701 case ODPAT_SET_PRIORITY:
702 case ODPAT_POP_PRIORITY:
703 case ODPAT_DROP_SPOOFED_ARP:
704 /* No validation needed. */
708 if (nla_get_u32(a) >= DP_MAX_PORTS)
711 case ODPAT_SET_DL_TCI:
712 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
716 case ODPAT_SET_NW_TOS:
717 if (nla_get_u8(a) & INET_ECN_MASK)
732 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
734 struct sw_flow_actions *actions;
737 actions = flow_actions_alloc(flow->actions_len);
738 error = PTR_ERR(actions);
743 if (copy_from_user(actions->actions,
744 (struct nlattr __user *)flow->actions,
746 goto error_free_actions;
747 error = validate_actions(actions->actions, actions->actions_len);
749 goto error_free_actions;
756 return ERR_PTR(error);
759 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
762 struct timespec offset_ts, used, now_mono;
764 ktime_get_ts(&now_mono);
765 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
766 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
767 now_mono.tv_nsec - offset_ts.tv_nsec);
769 stats->used_sec = used.tv_sec;
770 stats->used_nsec = used.tv_nsec;
773 stats->used_nsec = 0;
776 stats->n_packets = flow->packet_count;
777 stats->n_bytes = flow->byte_count;
779 stats->tcp_flags = flow->tcp_flags;
783 static void clear_stats(struct sw_flow *flow)
787 flow->packet_count = 0;
788 flow->byte_count = 0;
791 static int expand_table(struct datapath *dp)
793 struct tbl *old_table = rcu_dereference(dp->table);
794 struct tbl *new_table;
796 new_table = tbl_expand(old_table);
797 if (IS_ERR(new_table))
798 return PTR_ERR(new_table);
800 rcu_assign_pointer(dp->table, new_table);
801 tbl_deferred_destroy(old_table, NULL);
806 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
807 struct odp_flow_stats *stats)
809 struct tbl_node *flow_node;
810 struct sw_flow *flow;
814 table = rcu_dereference(dp->table);
815 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
818 struct sw_flow_actions *acts;
821 if (!(uf->flags & ODPPF_CREATE))
824 /* Expand table, if necessary, to make room. */
825 if (tbl_count(table) >= tbl_n_buckets(table)) {
826 error = expand_table(dp);
829 table = rcu_dereference(dp->table);
835 error = PTR_ERR(flow);
838 flow->key = uf->flow.key;
841 /* Obtain actions. */
842 acts = get_actions(&uf->flow);
843 error = PTR_ERR(acts);
845 goto error_free_flow;
846 rcu_assign_pointer(flow->sf_acts, acts);
848 /* Put flow in bucket. */
849 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
851 goto error_free_flow_acts;
853 memset(stats, 0, sizeof(struct odp_flow_stats));
855 /* We found a matching flow. */
856 struct sw_flow_actions *old_acts, *new_acts;
858 flow = flow_cast(flow_node);
860 /* Bail out if we're not allowed to modify an existing flow. */
862 if (!(uf->flags & ODPPF_MODIFY))
866 new_acts = get_actions(&uf->flow);
867 error = PTR_ERR(new_acts);
868 if (IS_ERR(new_acts))
870 old_acts = rcu_dereference(flow->sf_acts);
871 if (old_acts->actions_len != new_acts->actions_len ||
872 memcmp(old_acts->actions, new_acts->actions,
873 old_acts->actions_len)) {
874 rcu_assign_pointer(flow->sf_acts, new_acts);
875 flow_deferred_free_acts(old_acts);
880 /* Fetch stats, then clear them if necessary. */
881 spin_lock_bh(&flow->lock);
882 get_stats(flow, stats);
883 if (uf->flags & ODPPF_ZERO_STATS)
885 spin_unlock_bh(&flow->lock);
890 error_free_flow_acts:
891 kfree(flow->sf_acts);
893 flow->sf_acts = NULL;
899 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
901 struct odp_flow_stats stats;
902 struct odp_flow_put uf;
905 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
908 error = do_put_flow(dp, &uf, &stats);
912 if (copy_to_user(&ufp->flow.stats, &stats,
913 sizeof(struct odp_flow_stats)))
919 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
920 struct odp_flow_stats __user *ustats,
921 struct nlattr __user *actions,
922 u32 __user *actions_lenp)
924 struct sw_flow_actions *sf_acts;
925 struct odp_flow_stats stats;
928 spin_lock_bh(&flow->lock);
929 get_stats(flow, &stats);
930 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
933 spin_unlock_bh(&flow->lock);
935 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
936 get_user(actions_len, actions_lenp))
942 sf_acts = rcu_dereference(flow->sf_acts);
943 if (put_user(sf_acts->actions_len, actions_lenp) ||
944 (actions && copy_to_user(actions, sf_acts->actions,
945 min(sf_acts->actions_len, actions_len))))
951 static int answer_query(struct sw_flow *flow, u32 query_flags,
952 struct odp_flow __user *ufp)
954 struct nlattr __user *actions;
956 if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
959 return do_answer_query(flow, query_flags,
960 &ufp->stats, actions, &ufp->actions_len);
963 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
965 struct tbl *table = rcu_dereference(dp->table);
966 struct tbl_node *flow_node;
969 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
971 return ERR_PTR(-ENOENT);
973 error = tbl_remove(table, flow_node);
975 return ERR_PTR(error);
977 /* XXX Returned flow_node's statistics might lose a few packets, since
978 * other CPUs can be using this flow. We used to synchronize_rcu() to
979 * make sure that we get completely accurate stats, but that blows our
980 * performance, badly. */
981 return flow_cast(flow_node);
984 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
986 struct sw_flow *flow;
990 if (copy_from_user(&uf, ufp, sizeof uf))
993 flow = do_del_flow(dp, &uf.key);
995 return PTR_ERR(flow);
997 error = answer_query(flow, 0, ufp);
998 flow_deferred_free(flow);
1002 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1004 struct tbl *table = rcu_dereference(dp->table);
1007 for (i = 0; i < flowvec->n_flows; i++) {
1008 struct odp_flow __user *ufp = (struct odp_flow __user *)&flowvec->flows[i];
1010 struct tbl_node *flow_node;
1013 if (copy_from_user(&uf, ufp, sizeof uf))
1016 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1018 error = put_user(ENOENT, &ufp->stats.error);
1020 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1024 return flowvec->n_flows;
1027 struct list_flows_cbdata {
1028 struct odp_flow __user *uflows;
1033 static int list_flow(struct tbl_node *node, void *cbdata_)
1035 struct sw_flow *flow = flow_cast(node);
1036 struct list_flows_cbdata *cbdata = cbdata_;
1037 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1040 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1042 error = answer_query(flow, 0, ufp);
1046 if (cbdata->listed_flows >= cbdata->n_flows)
1047 return cbdata->listed_flows;
1051 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1053 struct list_flows_cbdata cbdata;
1056 if (!flowvec->n_flows)
1059 cbdata.uflows = (struct odp_flow __user *)flowvec->flows;
1060 cbdata.n_flows = flowvec->n_flows;
1061 cbdata.listed_flows = 0;
1063 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1064 return error ? error : cbdata.listed_flows;
1067 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1068 int (*function)(struct datapath *,
1069 const struct odp_flowvec *))
1071 struct odp_flowvec __user *uflowvec;
1072 struct odp_flowvec flowvec;
1075 uflowvec = (struct odp_flowvec __user *)argp;
1076 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1079 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1082 retval = function(dp, &flowvec);
1083 return (retval < 0 ? retval
1084 : retval == flowvec.n_flows ? 0
1085 : put_user(retval, &uflowvec->n_flows));
1088 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1090 struct odp_flow_key key;
1091 struct sk_buff *skb;
1092 struct sw_flow_actions *actions;
1098 if (execute->length < ETH_HLEN || execute->length > 65535)
1101 actions = flow_actions_alloc(execute->actions_len);
1102 if (IS_ERR(actions)) {
1103 err = PTR_ERR(actions);
1108 if (copy_from_user(actions->actions,
1109 (struct nlattr __user *)execute->actions, execute->actions_len))
1110 goto error_free_actions;
1112 err = validate_actions(actions->actions, execute->actions_len);
1114 goto error_free_actions;
1117 skb = alloc_skb(execute->length, GFP_KERNEL);
1119 goto error_free_actions;
1122 if (copy_from_user(skb_put(skb, execute->length),
1123 (const void __user *)execute->data,
1125 goto error_free_skb;
1127 skb_reset_mac_header(skb);
1130 /* Normally, setting the skb 'protocol' field would be handled by a
1131 * call to eth_type_trans(), but it assumes there's a sending
1132 * device, which we may not have. */
1133 if (ntohs(eth->h_proto) >= 1536)
1134 skb->protocol = eth->h_proto;
1136 skb->protocol = htons(ETH_P_802_2);
1138 err = flow_extract(skb, -1, &key, &is_frag);
1140 goto error_free_skb;
1143 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1157 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1159 struct odp_execute execute;
1161 if (copy_from_user(&execute, executep, sizeof execute))
1164 return do_execute(dp, &execute);
1167 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1169 struct tbl *table = rcu_dereference(dp->table);
1170 struct odp_stats stats;
1173 stats.n_flows = tbl_count(table);
1174 stats.cur_capacity = tbl_n_buckets(table);
1175 stats.max_capacity = TBL_MAX_BUCKETS;
1176 stats.n_ports = dp->n_ports;
1177 stats.max_ports = DP_MAX_PORTS;
1178 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1179 for_each_possible_cpu(i) {
1180 const struct dp_stats_percpu *percpu_stats;
1181 struct dp_stats_percpu local_stats;
1184 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1187 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1188 local_stats = *percpu_stats;
1189 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1191 stats.n_frags += local_stats.n_frags;
1192 stats.n_hit += local_stats.n_hit;
1193 stats.n_missed += local_stats.n_missed;
1194 stats.n_lost += local_stats.n_lost;
1196 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1197 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1198 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1201 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1202 int dp_min_mtu(const struct datapath *dp)
1209 list_for_each_entry_rcu (p, &dp->port_list, node) {
1212 /* Skip any internal ports, since that's what we're trying to
1214 if (is_internal_vport(p))
1217 dev_mtu = vport_get_mtu(p);
1218 if (!mtu || dev_mtu < mtu)
1222 return mtu ? mtu : ETH_DATA_LEN;
1225 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1226 * be called with RTNL lock. */
1227 void set_internal_devs_mtu(const struct datapath *dp)
1234 mtu = dp_min_mtu(dp);
1236 list_for_each_entry_rcu (p, &dp->port_list, node) {
1237 if (is_internal_vport(p))
1238 vport_set_mtu(p, mtu);
1242 static int put_port(const struct vport *p, struct odp_port __user *uop)
1246 memset(&op, 0, sizeof op);
1249 strncpy(op.devname, vport_get_name(p), sizeof op.devname);
1250 strncpy(op.type, vport_get_type(p), sizeof op.type);
1253 op.port = p->port_no;
1255 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1258 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1260 struct odp_port port;
1262 if (copy_from_user(&port, uport, sizeof port))
1265 if (port.devname[0]) {
1266 struct vport *vport;
1269 port.devname[IFNAMSIZ - 1] = '\0';
1274 vport = vport_locate(port.devname);
1279 if (vport->dp != dp) {
1284 port.port = vport->port_no;
1293 if (port.port >= DP_MAX_PORTS)
1295 if (!dp->ports[port.port])
1299 return put_port(dp->ports[port.port], uport);
1302 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1309 list_for_each_entry_rcu (p, &dp->port_list, node) {
1310 if (put_port(p, &uports[idx]))
1312 if (idx++ >= n_ports)
1319 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1321 struct odp_portvec pv;
1324 if (copy_from_user(&pv, upv, sizeof pv))
1327 retval = do_list_ports(dp, (struct odp_port __user *)pv.ports,
1332 return put_user(retval, &upv->n_ports);
1335 static int get_listen_mask(const struct file *f)
1337 return (long)f->private_data;
1340 static void set_listen_mask(struct file *f, int listen_mask)
1342 f->private_data = (void*)(long)listen_mask;
1345 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1348 int dp_idx = iminor(f->f_dentry->d_inode);
1349 struct datapath *dp;
1350 int drop_frags, listeners, port_no;
1351 unsigned int sflow_probability;
1354 /* Handle commands with special locking requirements up front. */
1357 err = create_dp(dp_idx, (char __user *)argp);
1360 case ODP_DP_DESTROY:
1361 err = destroy_dp(dp_idx);
1364 case ODP_VPORT_ATTACH:
1365 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1368 case ODP_VPORT_DETACH:
1369 err = get_user(port_no, (int __user *)argp);
1371 err = detach_port(dp_idx, port_no);
1375 err = vport_user_mod((struct odp_port __user *)argp);
1378 case ODP_VPORT_STATS_GET:
1379 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1382 case ODP_VPORT_STATS_SET:
1383 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1386 case ODP_VPORT_ETHER_GET:
1387 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1390 case ODP_VPORT_ETHER_SET:
1391 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1394 case ODP_VPORT_MTU_GET:
1395 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1398 case ODP_VPORT_MTU_SET:
1399 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1403 dp = get_dp_locked(dp_idx);
1410 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1413 case ODP_GET_DROP_FRAGS:
1414 err = put_user(dp->drop_frags, (int __user *)argp);
1417 case ODP_SET_DROP_FRAGS:
1418 err = get_user(drop_frags, (int __user *)argp);
1422 if (drop_frags != 0 && drop_frags != 1)
1424 dp->drop_frags = drop_frags;
1428 case ODP_GET_LISTEN_MASK:
1429 err = put_user(get_listen_mask(f), (int __user *)argp);
1432 case ODP_SET_LISTEN_MASK:
1433 err = get_user(listeners, (int __user *)argp);
1437 if (listeners & ~ODPL_ALL)
1440 set_listen_mask(f, listeners);
1443 case ODP_GET_SFLOW_PROBABILITY:
1444 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1447 case ODP_SET_SFLOW_PROBABILITY:
1448 err = get_user(sflow_probability, (unsigned int __user *)argp);
1450 dp->sflow_probability = sflow_probability;
1453 case ODP_VPORT_QUERY:
1454 err = query_port(dp, (struct odp_port __user *)argp);
1457 case ODP_VPORT_LIST:
1458 err = list_ports(dp, (struct odp_portvec __user *)argp);
1461 case ODP_FLOW_FLUSH:
1462 err = flush_flows(dp);
1466 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1470 err = del_flow(dp, (struct odp_flow __user *)argp);
1474 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1478 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1482 err = execute_packet(dp, (struct odp_execute __user *)argp);
1489 mutex_unlock(&dp->mutex);
1494 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1497 for (i = 0; i < DP_N_QUEUES; i++) {
1498 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1504 #ifdef CONFIG_COMPAT
1505 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1507 struct compat_odp_portvec pv;
1510 if (copy_from_user(&pv, upv, sizeof pv))
1513 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1517 return put_user(retval, &upv->n_ports);
1520 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1522 compat_uptr_t actions;
1524 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1525 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1526 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1527 __get_user(actions, &compat->actions) ||
1528 __get_user(flow->actions_len, &compat->actions_len) ||
1529 __get_user(flow->flags, &compat->flags))
1532 flow->actions = (struct nlattr __force *)compat_ptr(actions);
1536 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1538 struct odp_flow_stats stats;
1539 struct odp_flow_put fp;
1542 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1543 get_user(fp.flags, &ufp->flags))
1546 error = do_put_flow(dp, &fp, &stats);
1550 if (copy_to_user(&ufp->flow.stats, &stats,
1551 sizeof(struct odp_flow_stats)))
1557 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1558 struct compat_odp_flow __user *ufp)
1560 compat_uptr_t actions;
1562 if (get_user(actions, &ufp->actions))
1565 return do_answer_query(flow, query_flags, &ufp->stats,
1566 compat_ptr(actions), &ufp->actions_len);
1569 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1571 struct sw_flow *flow;
1575 if (compat_get_flow(&uf, ufp))
1578 flow = do_del_flow(dp, &uf.key);
1580 return PTR_ERR(flow);
1582 error = compat_answer_query(flow, 0, ufp);
1583 flow_deferred_free(flow);
1587 static int compat_query_flows(struct datapath *dp,
1588 struct compat_odp_flow __user *flows,
1591 struct tbl *table = rcu_dereference(dp->table);
1594 for (i = 0; i < n_flows; i++) {
1595 struct compat_odp_flow __user *ufp = &flows[i];
1597 struct tbl_node *flow_node;
1600 if (compat_get_flow(&uf, ufp))
1603 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1605 error = put_user(ENOENT, &ufp->stats.error);
1607 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1614 struct compat_list_flows_cbdata {
1615 struct compat_odp_flow __user *uflows;
1620 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1622 struct sw_flow *flow = flow_cast(node);
1623 struct compat_list_flows_cbdata *cbdata = cbdata_;
1624 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1627 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1629 error = compat_answer_query(flow, 0, ufp);
1633 if (cbdata->listed_flows >= cbdata->n_flows)
1634 return cbdata->listed_flows;
1638 static int compat_list_flows(struct datapath *dp,
1639 struct compat_odp_flow __user *flows, u32 n_flows)
1641 struct compat_list_flows_cbdata cbdata;
1647 cbdata.uflows = flows;
1648 cbdata.n_flows = n_flows;
1649 cbdata.listed_flows = 0;
1651 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1652 return error ? error : cbdata.listed_flows;
1655 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1656 int (*function)(struct datapath *,
1657 struct compat_odp_flow __user *,
1660 struct compat_odp_flowvec __user *uflowvec;
1661 struct compat_odp_flow __user *flows;
1662 struct compat_odp_flowvec flowvec;
1665 uflowvec = compat_ptr(argp);
1666 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1667 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1670 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1673 flows = compat_ptr(flowvec.flows);
1674 if (!access_ok(VERIFY_WRITE, flows,
1675 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1678 retval = function(dp, flows, flowvec.n_flows);
1679 return (retval < 0 ? retval
1680 : retval == flowvec.n_flows ? 0
1681 : put_user(retval, &uflowvec->n_flows));
1684 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1686 struct odp_execute execute;
1687 compat_uptr_t actions;
1690 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1691 __get_user(actions, &uexecute->actions) ||
1692 __get_user(execute.actions_len, &uexecute->actions_len) ||
1693 __get_user(data, &uexecute->data) ||
1694 __get_user(execute.length, &uexecute->length))
1697 execute.actions = (struct nlattr __force *)compat_ptr(actions);
1698 execute.data = (const void __force *)compat_ptr(data);
1700 return do_execute(dp, &execute);
1703 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1705 int dp_idx = iminor(f->f_dentry->d_inode);
1706 struct datapath *dp;
1710 case ODP_DP_DESTROY:
1711 case ODP_FLOW_FLUSH:
1712 /* Ioctls that don't need any translation at all. */
1713 return openvswitch_ioctl(f, cmd, argp);
1716 case ODP_VPORT_ATTACH:
1717 case ODP_VPORT_DETACH:
1719 case ODP_VPORT_MTU_SET:
1720 case ODP_VPORT_MTU_GET:
1721 case ODP_VPORT_ETHER_SET:
1722 case ODP_VPORT_ETHER_GET:
1723 case ODP_VPORT_STATS_SET:
1724 case ODP_VPORT_STATS_GET:
1726 case ODP_GET_DROP_FRAGS:
1727 case ODP_SET_DROP_FRAGS:
1728 case ODP_SET_LISTEN_MASK:
1729 case ODP_GET_LISTEN_MASK:
1730 case ODP_SET_SFLOW_PROBABILITY:
1731 case ODP_GET_SFLOW_PROBABILITY:
1732 case ODP_VPORT_QUERY:
1733 /* Ioctls that just need their pointer argument extended. */
1734 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1737 dp = get_dp_locked(dp_idx);
1743 case ODP_VPORT_LIST32:
1744 err = compat_list_ports(dp, compat_ptr(argp));
1747 case ODP_FLOW_PUT32:
1748 err = compat_put_flow(dp, compat_ptr(argp));
1751 case ODP_FLOW_DEL32:
1752 err = compat_del_flow(dp, compat_ptr(argp));
1755 case ODP_FLOW_GET32:
1756 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1759 case ODP_FLOW_LIST32:
1760 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1764 err = compat_execute(dp, compat_ptr(argp));
1771 mutex_unlock(&dp->mutex);
1777 /* Unfortunately this function is not exported so this is a verbatim copy
1778 * from net/core/datagram.c in 2.6.30. */
1779 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1780 u8 __user *to, int len,
1783 int start = skb_headlen(skb);
1785 int i, copy = start - offset;
1792 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1796 if ((len -= copy) == 0)
1803 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1806 WARN_ON(start > offset + len);
1808 end = start + skb_shinfo(skb)->frags[i].size;
1809 if ((copy = end - offset) > 0) {
1813 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1814 struct page *page = frag->page;
1819 csum2 = csum_and_copy_to_user(vaddr +
1826 *csump = csum_block_add(*csump, csum2, pos);
1836 if (skb_shinfo(skb)->frag_list) {
1837 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1839 for (; list; list=list->next) {
1842 WARN_ON(start > offset + len);
1844 end = start + list->len;
1845 if ((copy = end - offset) > 0) {
1849 if (skb_copy_and_csum_datagram(list,
1854 *csump = csum_block_add(*csump, csum2, pos);
1855 if ((len -= copy) == 0)
1871 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1872 size_t nbytes, loff_t *ppos)
1874 int listeners = get_listen_mask(f);
1875 int dp_idx = iminor(f->f_dentry->d_inode);
1876 struct datapath *dp = get_dp_locked(dp_idx);
1877 struct sk_buff *skb;
1878 size_t copy_bytes, tot_copy_bytes;
1884 if (nbytes == 0 || !listeners)
1890 for (i = 0; i < DP_N_QUEUES; i++) {
1891 if (listeners & (1 << i)) {
1892 skb = skb_dequeue(&dp->queues[i]);
1898 if (f->f_flags & O_NONBLOCK) {
1903 wait_event_interruptible(dp->waitqueue,
1904 dp_has_packet_of_interest(dp,
1907 if (signal_pending(current)) {
1908 retval = -ERESTARTSYS;
1913 mutex_unlock(&dp->mutex);
1915 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1918 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1919 if (copy_bytes == skb->len) {
1921 u16 csum_start, csum_offset;
1923 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1924 BUG_ON(csum_start >= skb_headlen(skb));
1925 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1926 copy_bytes - csum_start, &csum);
1928 __sum16 __user *csump;
1930 copy_bytes = csum_start;
1931 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1933 BUG_ON((char __user *)csump + sizeof(__sum16) >
1935 put_user(csum_fold(csum), csump);
1938 retval = skb_checksum_help(skb);
1945 iov.iov_len = copy_bytes;
1946 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1950 retval = tot_copy_bytes;
1956 mutex_unlock(&dp->mutex);
1960 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1962 int dp_idx = iminor(file->f_dentry->d_inode);
1963 struct datapath *dp = get_dp_locked(dp_idx);
1968 poll_wait(file, &dp->waitqueue, wait);
1969 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1970 mask |= POLLIN | POLLRDNORM;
1971 mutex_unlock(&dp->mutex);
1973 mask = POLLIN | POLLRDNORM | POLLHUP;
1978 static struct file_operations openvswitch_fops = {
1979 .read = openvswitch_read,
1980 .poll = openvswitch_poll,
1981 .unlocked_ioctl = openvswitch_ioctl,
1982 #ifdef CONFIG_COMPAT
1983 .compat_ioctl = openvswitch_compat_ioctl,
1989 static int __init dp_init(void)
1991 struct sk_buff *dummy_skb;
1994 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1996 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2004 goto error_flow_exit;
2006 err = register_netdevice_notifier(&dp_device_notifier);
2008 goto error_vport_exit;
2010 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2012 goto error_unreg_notifier;
2016 error_unreg_notifier:
2017 unregister_netdevice_notifier(&dp_device_notifier);
2026 static void dp_cleanup(void)
2029 unregister_chrdev(major, "openvswitch");
2030 unregister_netdevice_notifier(&dp_device_notifier);
2035 module_init(dp_init);
2036 module_exit(dp_cleanup);
2038 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2039 MODULE_LICENSE("GPL");