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 flow_used(OVS_CB(skb)->flow, skb);
506 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
508 /* Check whether we've looped too much. */
509 loop = loop_get_counter();
510 if (unlikely(++loop->count > MAX_LOOPS))
511 loop->looping = true;
512 if (unlikely(loop->looping)) {
513 loop_suppress(dp, acts);
517 /* Execute actions. */
518 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
520 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
522 /* Check whether sub-actions looped too much. */
523 if (unlikely(loop->looping))
524 loop_suppress(dp, acts);
527 /* Decrement loop counter. */
529 loop->looping = false;
533 /* Update datapath statistics. */
535 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
537 write_seqcount_begin(&stats->seqlock);
538 (*(u64 *)((u8 *)stats + stats_counter_off))++;
539 write_seqcount_end(&stats->seqlock);
544 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
545 * unless we broke up a GSO packet. */
546 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
547 int queue_no, u32 arg)
549 struct sk_buff *nskb;
553 if (OVS_CB(skb)->vport)
554 port_no = OVS_CB(skb)->vport->port_no;
556 port_no = ODPP_LOCAL;
559 struct odp_msg *header;
564 err = skb_cow(skb, sizeof *header);
568 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
569 header->type = queue_no;
570 header->length = skb->len;
571 header->port = port_no;
572 header->reserved = 0;
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);
616 if (unlikely(IS_ERR(skb))) {
621 /* XXX This case might not be possible. It's hard to
622 * tell from the skb_gso_segment() code and comment. */
626 err = queue_control_packets(skb, queue, queue_no, arg);
627 wake_up_interruptible(&dp->waitqueue);
634 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
636 write_seqcount_begin(&stats->seqlock);
638 write_seqcount_end(&stats->seqlock);
645 static int flush_flows(struct datapath *dp)
647 struct tbl *old_table = rcu_dereference(dp->table);
648 struct tbl *new_table;
650 new_table = tbl_create(0);
654 rcu_assign_pointer(dp->table, new_table);
656 tbl_deferred_destroy(old_table, flow_free_tbl);
661 static int validate_actions(const struct nlattr *actions, u32 actions_len)
663 const struct nlattr *a;
666 nla_for_each_attr(a, actions, actions_len, rem) {
667 static const u32 action_lens[ODPAT_MAX + 1] = {
669 [ODPAT_CONTROLLER] = 4,
670 [ODPAT_SET_DL_TCI] = 2,
671 [ODPAT_STRIP_VLAN] = 0,
672 [ODPAT_SET_DL_SRC] = ETH_ALEN,
673 [ODPAT_SET_DL_DST] = ETH_ALEN,
674 [ODPAT_SET_NW_SRC] = 4,
675 [ODPAT_SET_NW_DST] = 4,
676 [ODPAT_SET_NW_TOS] = 1,
677 [ODPAT_SET_TP_SRC] = 2,
678 [ODPAT_SET_TP_DST] = 2,
679 [ODPAT_SET_TUNNEL] = 4,
680 [ODPAT_SET_PRIORITY] = 4,
681 [ODPAT_POP_PRIORITY] = 0,
682 [ODPAT_DROP_SPOOFED_ARP] = 0,
684 int type = nla_type(a);
686 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
693 case ODPAT_CONTROLLER:
694 case ODPAT_STRIP_VLAN:
695 case ODPAT_SET_DL_SRC:
696 case ODPAT_SET_DL_DST:
697 case ODPAT_SET_NW_SRC:
698 case ODPAT_SET_NW_DST:
699 case ODPAT_SET_TP_SRC:
700 case ODPAT_SET_TP_DST:
701 case ODPAT_SET_TUNNEL:
702 case ODPAT_SET_PRIORITY:
703 case ODPAT_POP_PRIORITY:
704 case ODPAT_DROP_SPOOFED_ARP:
705 /* No validation needed. */
709 if (nla_get_u32(a) >= DP_MAX_PORTS)
712 case ODPAT_SET_DL_TCI:
713 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
717 case ODPAT_SET_NW_TOS:
718 if (nla_get_u8(a) & INET_ECN_MASK)
733 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
735 struct sw_flow_actions *actions;
738 actions = flow_actions_alloc(flow->actions_len);
739 error = PTR_ERR(actions);
744 if (copy_from_user(actions->actions, flow->actions, flow->actions_len))
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 *actions;
955 if (get_user(actions, &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 = &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 = 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, execute->actions, execute->actions_len))
1108 goto error_free_actions;
1110 err = validate_actions(actions->actions, execute->actions_len);
1112 goto error_free_actions;
1115 skb = alloc_skb(execute->length, GFP_KERNEL);
1117 goto error_free_actions;
1120 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1122 goto error_free_skb;
1124 skb_reset_mac_header(skb);
1127 /* Normally, setting the skb 'protocol' field would be handled by a
1128 * call to eth_type_trans(), but it assumes there's a sending
1129 * device, which we may not have. */
1130 if (ntohs(eth->h_proto) >= 1536)
1131 skb->protocol = eth->h_proto;
1133 skb->protocol = htons(ETH_P_802_2);
1135 err = flow_extract(skb, -1, &key, &is_frag);
1137 goto error_free_skb;
1140 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1154 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1156 struct odp_execute execute;
1158 if (copy_from_user(&execute, executep, sizeof execute))
1161 return do_execute(dp, &execute);
1164 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1166 struct tbl *table = rcu_dereference(dp->table);
1167 struct odp_stats stats;
1170 stats.n_flows = tbl_count(table);
1171 stats.cur_capacity = tbl_n_buckets(table);
1172 stats.max_capacity = TBL_MAX_BUCKETS;
1173 stats.n_ports = dp->n_ports;
1174 stats.max_ports = DP_MAX_PORTS;
1175 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1176 for_each_possible_cpu(i) {
1177 const struct dp_stats_percpu *percpu_stats;
1178 struct dp_stats_percpu local_stats;
1181 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1184 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1185 local_stats = *percpu_stats;
1186 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1188 stats.n_frags += local_stats.n_frags;
1189 stats.n_hit += local_stats.n_hit;
1190 stats.n_missed += local_stats.n_missed;
1191 stats.n_lost += local_stats.n_lost;
1193 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1194 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1195 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1198 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1199 int dp_min_mtu(const struct datapath *dp)
1206 list_for_each_entry_rcu (p, &dp->port_list, node) {
1209 /* Skip any internal ports, since that's what we're trying to
1211 if (is_internal_vport(p))
1214 dev_mtu = vport_get_mtu(p);
1215 if (!mtu || dev_mtu < mtu)
1219 return mtu ? mtu : ETH_DATA_LEN;
1222 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1223 * be called with RTNL lock. */
1224 void set_internal_devs_mtu(const struct datapath *dp)
1231 mtu = dp_min_mtu(dp);
1233 list_for_each_entry_rcu (p, &dp->port_list, node) {
1234 if (is_internal_vport(p))
1235 vport_set_mtu(p, mtu);
1239 static int put_port(const struct vport *p, struct odp_port __user *uop)
1243 memset(&op, 0, sizeof op);
1246 strncpy(op.devname, vport_get_name(p), sizeof op.devname);
1247 strncpy(op.type, vport_get_type(p), sizeof op.type);
1250 op.port = p->port_no;
1252 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1255 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1257 struct odp_port port;
1259 if (copy_from_user(&port, uport, sizeof port))
1262 if (port.devname[0]) {
1263 struct vport *vport;
1266 port.devname[IFNAMSIZ - 1] = '\0';
1271 vport = vport_locate(port.devname);
1276 if (vport->dp != dp) {
1281 port.port = vport->port_no;
1290 if (port.port >= DP_MAX_PORTS)
1292 if (!dp->ports[port.port])
1296 return put_port(dp->ports[port.port], uport);
1299 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1306 list_for_each_entry_rcu (p, &dp->port_list, node) {
1307 if (put_port(p, &uports[idx]))
1309 if (idx++ >= n_ports)
1316 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1318 struct odp_portvec pv;
1321 if (copy_from_user(&pv, upv, sizeof pv))
1324 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1328 return put_user(retval, &upv->n_ports);
1331 static int get_listen_mask(const struct file *f)
1333 return (long)f->private_data;
1336 static void set_listen_mask(struct file *f, int listen_mask)
1338 f->private_data = (void*)(long)listen_mask;
1341 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1344 int dp_idx = iminor(f->f_dentry->d_inode);
1345 struct datapath *dp;
1346 int drop_frags, listeners, port_no;
1347 unsigned int sflow_probability;
1350 /* Handle commands with special locking requirements up front. */
1353 err = create_dp(dp_idx, (char __user *)argp);
1356 case ODP_DP_DESTROY:
1357 err = destroy_dp(dp_idx);
1360 case ODP_VPORT_ATTACH:
1361 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1364 case ODP_VPORT_DETACH:
1365 err = get_user(port_no, (int __user *)argp);
1367 err = detach_port(dp_idx, port_no);
1371 err = vport_user_mod((struct odp_port __user *)argp);
1374 case ODP_VPORT_STATS_GET:
1375 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1378 case ODP_VPORT_STATS_SET:
1379 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1382 case ODP_VPORT_ETHER_GET:
1383 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1386 case ODP_VPORT_ETHER_SET:
1387 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1390 case ODP_VPORT_MTU_GET:
1391 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1394 case ODP_VPORT_MTU_SET:
1395 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1399 dp = get_dp_locked(dp_idx);
1406 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1409 case ODP_GET_DROP_FRAGS:
1410 err = put_user(dp->drop_frags, (int __user *)argp);
1413 case ODP_SET_DROP_FRAGS:
1414 err = get_user(drop_frags, (int __user *)argp);
1418 if (drop_frags != 0 && drop_frags != 1)
1420 dp->drop_frags = drop_frags;
1424 case ODP_GET_LISTEN_MASK:
1425 err = put_user(get_listen_mask(f), (int __user *)argp);
1428 case ODP_SET_LISTEN_MASK:
1429 err = get_user(listeners, (int __user *)argp);
1433 if (listeners & ~ODPL_ALL)
1436 set_listen_mask(f, listeners);
1439 case ODP_GET_SFLOW_PROBABILITY:
1440 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1443 case ODP_SET_SFLOW_PROBABILITY:
1444 err = get_user(sflow_probability, (unsigned int __user *)argp);
1446 dp->sflow_probability = sflow_probability;
1449 case ODP_VPORT_QUERY:
1450 err = query_port(dp, (struct odp_port __user *)argp);
1453 case ODP_VPORT_LIST:
1454 err = list_ports(dp, (struct odp_portvec __user *)argp);
1457 case ODP_FLOW_FLUSH:
1458 err = flush_flows(dp);
1462 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1466 err = del_flow(dp, (struct odp_flow __user *)argp);
1470 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1474 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1478 err = execute_packet(dp, (struct odp_execute __user *)argp);
1485 mutex_unlock(&dp->mutex);
1490 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1493 for (i = 0; i < DP_N_QUEUES; i++) {
1494 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1500 #ifdef CONFIG_COMPAT
1501 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1503 struct compat_odp_portvec pv;
1506 if (copy_from_user(&pv, upv, sizeof pv))
1509 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1513 return put_user(retval, &upv->n_ports);
1516 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1518 compat_uptr_t actions;
1520 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1521 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1522 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1523 __get_user(actions, &compat->actions) ||
1524 __get_user(flow->actions_len, &compat->actions_len) ||
1525 __get_user(flow->flags, &compat->flags))
1528 flow->actions = compat_ptr(actions);
1532 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1534 struct odp_flow_stats stats;
1535 struct odp_flow_put fp;
1538 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1539 get_user(fp.flags, &ufp->flags))
1542 error = do_put_flow(dp, &fp, &stats);
1546 if (copy_to_user(&ufp->flow.stats, &stats,
1547 sizeof(struct odp_flow_stats)))
1553 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1554 struct compat_odp_flow __user *ufp)
1556 compat_uptr_t actions;
1558 if (get_user(actions, &ufp->actions))
1561 return do_answer_query(flow, query_flags, &ufp->stats,
1562 compat_ptr(actions), &ufp->actions_len);
1565 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1567 struct sw_flow *flow;
1571 if (compat_get_flow(&uf, ufp))
1574 flow = do_del_flow(dp, &uf.key);
1576 return PTR_ERR(flow);
1578 error = compat_answer_query(flow, 0, ufp);
1579 flow_deferred_free(flow);
1583 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1585 struct tbl *table = rcu_dereference(dp->table);
1588 for (i = 0; i < n_flows; i++) {
1589 struct compat_odp_flow __user *ufp = &flows[i];
1591 struct tbl_node *flow_node;
1594 if (compat_get_flow(&uf, ufp))
1597 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1599 error = put_user(ENOENT, &ufp->stats.error);
1601 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1608 struct compat_list_flows_cbdata {
1609 struct compat_odp_flow __user *uflows;
1614 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1616 struct sw_flow *flow = flow_cast(node);
1617 struct compat_list_flows_cbdata *cbdata = cbdata_;
1618 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1621 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1623 error = compat_answer_query(flow, 0, ufp);
1627 if (cbdata->listed_flows >= cbdata->n_flows)
1628 return cbdata->listed_flows;
1632 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1634 struct compat_list_flows_cbdata cbdata;
1640 cbdata.uflows = flows;
1641 cbdata.n_flows = n_flows;
1642 cbdata.listed_flows = 0;
1644 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1645 return error ? error : cbdata.listed_flows;
1648 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1649 int (*function)(struct datapath *,
1650 struct compat_odp_flow *,
1653 struct compat_odp_flowvec __user *uflowvec;
1654 struct compat_odp_flow __user *flows;
1655 struct compat_odp_flowvec flowvec;
1658 uflowvec = compat_ptr(argp);
1659 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1660 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1663 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1666 flows = compat_ptr(flowvec.flows);
1667 if (!access_ok(VERIFY_WRITE, flows,
1668 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1671 retval = function(dp, flows, flowvec.n_flows);
1672 return (retval < 0 ? retval
1673 : retval == flowvec.n_flows ? 0
1674 : put_user(retval, &uflowvec->n_flows));
1677 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1679 struct odp_execute execute;
1680 compat_uptr_t actions;
1683 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1684 __get_user(actions, &uexecute->actions) ||
1685 __get_user(execute.actions_len, &uexecute->actions_len) ||
1686 __get_user(data, &uexecute->data) ||
1687 __get_user(execute.length, &uexecute->length))
1690 execute.actions = compat_ptr(actions);
1691 execute.data = compat_ptr(data);
1693 return do_execute(dp, &execute);
1696 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1698 int dp_idx = iminor(f->f_dentry->d_inode);
1699 struct datapath *dp;
1703 case ODP_DP_DESTROY:
1704 case ODP_FLOW_FLUSH:
1705 /* Ioctls that don't need any translation at all. */
1706 return openvswitch_ioctl(f, cmd, argp);
1709 case ODP_VPORT_ATTACH:
1710 case ODP_VPORT_DETACH:
1712 case ODP_VPORT_MTU_SET:
1713 case ODP_VPORT_MTU_GET:
1714 case ODP_VPORT_ETHER_SET:
1715 case ODP_VPORT_ETHER_GET:
1716 case ODP_VPORT_STATS_SET:
1717 case ODP_VPORT_STATS_GET:
1719 case ODP_GET_DROP_FRAGS:
1720 case ODP_SET_DROP_FRAGS:
1721 case ODP_SET_LISTEN_MASK:
1722 case ODP_GET_LISTEN_MASK:
1723 case ODP_SET_SFLOW_PROBABILITY:
1724 case ODP_GET_SFLOW_PROBABILITY:
1725 case ODP_VPORT_QUERY:
1726 /* Ioctls that just need their pointer argument extended. */
1727 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1730 dp = get_dp_locked(dp_idx);
1736 case ODP_VPORT_LIST32:
1737 err = compat_list_ports(dp, compat_ptr(argp));
1740 case ODP_FLOW_PUT32:
1741 err = compat_put_flow(dp, compat_ptr(argp));
1744 case ODP_FLOW_DEL32:
1745 err = compat_del_flow(dp, compat_ptr(argp));
1748 case ODP_FLOW_GET32:
1749 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1752 case ODP_FLOW_LIST32:
1753 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1757 err = compat_execute(dp, compat_ptr(argp));
1764 mutex_unlock(&dp->mutex);
1770 /* Unfortunately this function is not exported so this is a verbatim copy
1771 * from net/core/datagram.c in 2.6.30. */
1772 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1773 u8 __user *to, int len,
1776 int start = skb_headlen(skb);
1778 int i, copy = start - offset;
1785 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1789 if ((len -= copy) == 0)
1796 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1799 WARN_ON(start > offset + len);
1801 end = start + skb_shinfo(skb)->frags[i].size;
1802 if ((copy = end - offset) > 0) {
1806 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1807 struct page *page = frag->page;
1812 csum2 = csum_and_copy_to_user(vaddr +
1819 *csump = csum_block_add(*csump, csum2, pos);
1829 if (skb_shinfo(skb)->frag_list) {
1830 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1832 for (; list; list=list->next) {
1835 WARN_ON(start > offset + len);
1837 end = start + list->len;
1838 if ((copy = end - offset) > 0) {
1842 if (skb_copy_and_csum_datagram(list,
1847 *csump = csum_block_add(*csump, csum2, pos);
1848 if ((len -= copy) == 0)
1864 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1865 size_t nbytes, loff_t *ppos)
1867 int listeners = get_listen_mask(f);
1868 int dp_idx = iminor(f->f_dentry->d_inode);
1869 struct datapath *dp = get_dp_locked(dp_idx);
1870 struct sk_buff *skb;
1871 size_t copy_bytes, tot_copy_bytes;
1877 if (nbytes == 0 || !listeners)
1883 for (i = 0; i < DP_N_QUEUES; i++) {
1884 if (listeners & (1 << i)) {
1885 skb = skb_dequeue(&dp->queues[i]);
1891 if (f->f_flags & O_NONBLOCK) {
1896 wait_event_interruptible(dp->waitqueue,
1897 dp_has_packet_of_interest(dp,
1900 if (signal_pending(current)) {
1901 retval = -ERESTARTSYS;
1906 mutex_unlock(&dp->mutex);
1908 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1911 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1912 if (copy_bytes == skb->len) {
1914 u16 csum_start, csum_offset;
1916 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1917 BUG_ON(csum_start >= skb_headlen(skb));
1918 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1919 copy_bytes - csum_start, &csum);
1921 __sum16 __user *csump;
1923 copy_bytes = csum_start;
1924 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1926 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
1927 put_user(csum_fold(csum), csump);
1930 retval = skb_checksum_help(skb);
1934 struct iovec __user iov;
1937 iov.iov_len = copy_bytes;
1938 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1942 retval = tot_copy_bytes;
1948 mutex_unlock(&dp->mutex);
1952 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1954 int dp_idx = iminor(file->f_dentry->d_inode);
1955 struct datapath *dp = get_dp_locked(dp_idx);
1960 poll_wait(file, &dp->waitqueue, wait);
1961 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1962 mask |= POLLIN | POLLRDNORM;
1963 mutex_unlock(&dp->mutex);
1965 mask = POLLIN | POLLRDNORM | POLLHUP;
1970 static struct file_operations openvswitch_fops = {
1971 .read = openvswitch_read,
1972 .poll = openvswitch_poll,
1973 .unlocked_ioctl = openvswitch_ioctl,
1974 #ifdef CONFIG_COMPAT
1975 .compat_ioctl = openvswitch_compat_ioctl,
1981 static int __init dp_init(void)
1983 struct sk_buff *dummy_skb;
1986 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1988 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1996 goto error_flow_exit;
1998 err = register_netdevice_notifier(&dp_device_notifier);
2000 goto error_vport_exit;
2002 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2004 goto error_unreg_notifier;
2008 error_unreg_notifier:
2009 unregister_netdevice_notifier(&dp_device_notifier);
2018 static void dp_cleanup(void)
2021 unregister_chrdev(major, "openvswitch");
2022 unregister_netdevice_notifier(&dp_device_notifier);
2027 module_init(dp_init);
2028 module_exit(dp_cleanup);
2030 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2031 MODULE_LICENSE("GPL");