2 * Copyright (c) 2007, 2008, 2009, 2010, 2011 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 <net/genetlink.h>
45 #include <linux/compat.h>
47 #include "openvswitch/datapath-protocol.h"
52 #include "loop_counter.h"
53 #include "odp-compat.h"
55 #include "vport-internal_dev.h"
57 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
58 EXPORT_SYMBOL(dp_ioctl_hook);
60 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
63 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
66 * It is safe to access the datapath and vport structures with just
69 static struct datapath __rcu *dps[ODP_MAX];
70 static DEFINE_MUTEX(dp_mutex);
72 static int new_vport(struct datapath *, struct odp_port *, int port_no);
74 /* Must be called with rcu_read_lock or dp_mutex. */
75 struct datapath *get_dp(int dp_idx)
77 if (dp_idx < 0 || dp_idx >= ODP_MAX)
79 return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
80 lockdep_is_held(&dp_mutex));
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 static struct tbl *get_table_protected(struct datapath *dp)
98 return rcu_dereference_protected(dp->table,
99 lockdep_is_held(&dp->mutex));
102 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
104 return rcu_dereference_protected(dp->ports[port_no],
105 lockdep_is_held(&dp->mutex));
108 /* Must be called with rcu_read_lock or RTNL lock. */
109 const char *dp_name(const struct datapath *dp)
111 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
114 static inline size_t br_nlmsg_size(void)
116 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
117 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
118 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
119 + nla_total_size(4) /* IFLA_MASTER */
120 + nla_total_size(4) /* IFLA_MTU */
121 + nla_total_size(4) /* IFLA_LINK */
122 + nla_total_size(1); /* IFLA_OPERSTATE */
125 static int dp_fill_ifinfo(struct sk_buff *skb,
126 const struct vport *port,
127 int event, unsigned int flags)
129 struct datapath *dp = port->dp;
130 int ifindex = vport_get_ifindex(port);
131 int iflink = vport_get_iflink(port);
132 struct ifinfomsg *hdr;
133 struct nlmsghdr *nlh;
141 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
145 hdr = nlmsg_data(nlh);
146 hdr->ifi_family = AF_BRIDGE;
148 hdr->ifi_type = ARPHRD_ETHER;
149 hdr->ifi_index = ifindex;
150 hdr->ifi_flags = vport_get_flags(port);
153 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
154 NLA_PUT_U32(skb, IFLA_MASTER,
155 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
156 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
157 #ifdef IFLA_OPERSTATE
158 NLA_PUT_U8(skb, IFLA_OPERSTATE,
159 vport_is_running(port)
160 ? vport_get_operstate(port)
164 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
166 if (ifindex != iflink)
167 NLA_PUT_U32(skb, IFLA_LINK,iflink);
169 return nlmsg_end(skb, nlh);
172 nlmsg_cancel(skb, nlh);
176 static void dp_ifinfo_notify(int event, struct vport *port)
181 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
185 err = dp_fill_ifinfo(skb, port, event, 0);
187 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
188 WARN_ON(err == -EMSGSIZE);
192 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
196 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
199 static void release_dp(struct kobject *kobj)
201 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
205 static struct kobj_type dp_ktype = {
206 .release = release_dp
209 static int create_dp(int dp_idx, const char __user *devnamep)
211 struct odp_port internal_dev_port;
212 char devname[IFNAMSIZ];
218 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
222 } else if (retval >= IFNAMSIZ) {
227 snprintf(devname, sizeof(devname), "of%d", dp_idx);
231 mutex_lock(&dp_mutex);
233 if (!try_module_get(THIS_MODULE))
236 /* Exit early if a datapath with that number already exists.
237 * (We don't use -EEXIST because that's ambiguous with 'devname'
238 * conflicting with an existing network device name.) */
244 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
247 INIT_LIST_HEAD(&dp->port_list);
248 mutex_init(&dp->mutex);
249 mutex_lock(&dp->mutex);
251 for (i = 0; i < DP_N_QUEUES; i++)
252 skb_queue_head_init(&dp->queues[i]);
253 init_waitqueue_head(&dp->waitqueue);
255 /* Initialize kobject for bridge. This will be added as
256 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
257 dp->ifobj.kset = NULL;
258 kobject_init(&dp->ifobj, &dp_ktype);
260 /* Allocate table. */
262 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
266 /* Set up our datapath device. */
267 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
268 strcpy(internal_dev_port.devname, devname);
269 strcpy(internal_dev_port.type, "internal");
270 err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
275 goto err_destroy_table;
279 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
280 if (!dp->stats_percpu) {
282 goto err_destroy_local_port;
285 rcu_assign_pointer(dps[dp_idx], dp);
288 mutex_unlock(&dp->mutex);
289 mutex_unlock(&dp_mutex);
294 err_destroy_local_port:
295 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
297 tbl_destroy(get_table_protected(dp), NULL);
299 mutex_unlock(&dp->mutex);
302 module_put(THIS_MODULE);
304 mutex_unlock(&dp_mutex);
310 static void destroy_dp_rcu(struct rcu_head *rcu)
312 struct datapath *dp = container_of(rcu, struct datapath, rcu);
315 for (i = 0; i < DP_N_QUEUES; i++)
316 skb_queue_purge(&dp->queues[i]);
318 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
319 free_percpu(dp->stats_percpu);
320 kobject_put(&dp->ifobj);
323 static int destroy_dp(int dp_idx)
330 mutex_lock(&dp_mutex);
337 mutex_lock(&dp->mutex);
339 list_for_each_entry_safe (p, n, &dp->port_list, node)
340 if (p->port_no != ODPP_LOCAL)
344 rcu_assign_pointer(dps[dp->dp_idx], NULL);
345 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
347 mutex_unlock(&dp->mutex);
348 call_rcu(&dp->rcu, destroy_dp_rcu);
349 module_put(THIS_MODULE);
352 mutex_unlock(&dp_mutex);
357 /* Called with RTNL lock and dp->mutex. */
358 static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
360 struct vport_parms parms;
363 parms.name = odp_port->devname;
364 parms.type = odp_port->type;
365 parms.config = odp_port->config;
367 parms.port_no = port_no;
370 vport = vport_add(&parms);
374 return PTR_ERR(vport);
376 rcu_assign_pointer(dp->ports[port_no], vport);
377 list_add_rcu(&vport->node, &dp->port_list);
380 dp_ifinfo_notify(RTM_NEWLINK, vport);
385 static int attach_port(int dp_idx, struct odp_port __user *portp)
388 struct odp_port port;
393 if (copy_from_user(&port, portp, sizeof(port)))
395 port.devname[IFNAMSIZ - 1] = '\0';
396 port.type[VPORT_TYPE_SIZE - 1] = '\0';
399 dp = get_dp_locked(dp_idx);
402 goto out_unlock_rtnl;
404 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
405 if (!dp->ports[port_no])
411 err = new_vport(dp, &port, port_no);
415 set_internal_devs_mtu(dp);
416 dp_sysfs_add_if(get_vport_protected(dp, port_no));
418 err = put_user(port_no, &portp->port);
421 mutex_unlock(&dp->mutex);
428 int dp_detach_port(struct vport *p)
434 if (p->port_no != ODPP_LOCAL)
436 dp_ifinfo_notify(RTM_DELLINK, p);
438 /* First drop references to device. */
440 list_del_rcu(&p->node);
441 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
443 /* Then destroy it. */
451 static int detach_port(int dp_idx, int port_no)
458 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
462 dp = get_dp_locked(dp_idx);
465 goto out_unlock_rtnl;
467 p = get_vport_protected(dp, port_no);
472 err = dp_detach_port(p);
475 mutex_unlock(&dp->mutex);
482 /* Must be called with rcu_read_lock. */
483 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
485 struct datapath *dp = p->dp;
486 struct dp_stats_percpu *stats;
487 int stats_counter_off;
488 struct sw_flow_actions *acts;
489 struct loop_counter *loop;
492 OVS_CB(skb)->vport = p;
494 if (!OVS_CB(skb)->flow) {
495 struct sw_flow_key key;
496 struct tbl_node *flow_node;
499 /* Extract flow from 'skb' into 'key'. */
500 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
501 if (unlikely(error)) {
506 if (is_frag && dp->drop_frags) {
508 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
513 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
514 flow_hash(&key), flow_cmp);
515 if (unlikely(!flow_node)) {
516 dp_output_control(dp, skb, _ODPL_MISS_NR,
517 (__force u64)OVS_CB(skb)->tun_id);
518 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
522 OVS_CB(skb)->flow = flow_cast(flow_node);
525 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
526 flow_used(OVS_CB(skb)->flow, skb);
528 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
530 /* Check whether we've looped too much. */
531 loop = loop_get_counter();
532 if (unlikely(++loop->count > MAX_LOOPS))
533 loop->looping = true;
534 if (unlikely(loop->looping)) {
535 loop_suppress(dp, acts);
540 /* Execute actions. */
541 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
544 /* Check whether sub-actions looped too much. */
545 if (unlikely(loop->looping))
546 loop_suppress(dp, acts);
549 /* Decrement loop counter. */
551 loop->looping = false;
555 /* Update datapath statistics. */
557 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
559 write_seqcount_begin(&stats->seqlock);
560 (*(u64 *)((u8 *)stats + stats_counter_off))++;
561 write_seqcount_end(&stats->seqlock);
566 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
567 * unless we broke up a GSO packet. */
568 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
569 int queue_no, u64 arg)
571 struct sk_buff *nskb;
575 if (OVS_CB(skb)->vport)
576 port_no = OVS_CB(skb)->vport->port_no;
578 port_no = ODPP_LOCAL;
581 struct odp_msg *header;
586 err = skb_cow(skb, sizeof(*header));
590 header = (struct odp_msg*)__skb_push(skb, sizeof(*header));
591 header->type = queue_no;
592 header->length = skb->len;
593 header->port = port_no;
595 skb_queue_tail(queue, skb);
603 while ((skb = nskb) != NULL) {
610 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
613 struct dp_stats_percpu *stats;
614 struct sk_buff_head *queue;
617 WARN_ON_ONCE(skb_shared(skb));
618 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
619 queue = &dp->queues[queue_no];
621 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
624 forward_ip_summed(skb);
626 err = vswitch_skb_checksum_setup(skb);
630 /* Break apart GSO packets into their component pieces. Otherwise
631 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
632 if (skb_is_gso(skb)) {
633 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
643 err = queue_control_packets(skb, queue, queue_no, arg);
644 wake_up_interruptible(&dp->waitqueue);
651 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
653 write_seqcount_begin(&stats->seqlock);
655 write_seqcount_end(&stats->seqlock);
662 static int flush_flows(struct datapath *dp)
664 struct tbl *old_table = get_table_protected(dp);
665 struct tbl *new_table;
667 new_table = tbl_create(TBL_MIN_BUCKETS);
671 rcu_assign_pointer(dp->table, new_table);
673 tbl_deferred_destroy(old_table, flow_free_tbl);
678 static int validate_actions(const struct nlattr *actions, u32 actions_len)
680 const struct nlattr *a;
683 nla_for_each_attr(a, actions, actions_len, rem) {
684 static const u32 action_lens[ODPAT_MAX + 1] = {
686 [ODPAT_CONTROLLER] = 8,
687 [ODPAT_SET_DL_TCI] = 2,
688 [ODPAT_STRIP_VLAN] = 0,
689 [ODPAT_SET_DL_SRC] = ETH_ALEN,
690 [ODPAT_SET_DL_DST] = ETH_ALEN,
691 [ODPAT_SET_NW_SRC] = 4,
692 [ODPAT_SET_NW_DST] = 4,
693 [ODPAT_SET_NW_TOS] = 1,
694 [ODPAT_SET_TP_SRC] = 2,
695 [ODPAT_SET_TP_DST] = 2,
696 [ODPAT_SET_TUNNEL] = 8,
697 [ODPAT_SET_PRIORITY] = 4,
698 [ODPAT_POP_PRIORITY] = 0,
699 [ODPAT_DROP_SPOOFED_ARP] = 0,
701 int type = nla_type(a);
703 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
710 case ODPAT_CONTROLLER:
711 case ODPAT_STRIP_VLAN:
712 case ODPAT_SET_DL_SRC:
713 case ODPAT_SET_DL_DST:
714 case ODPAT_SET_NW_SRC:
715 case ODPAT_SET_NW_DST:
716 case ODPAT_SET_TP_SRC:
717 case ODPAT_SET_TP_DST:
718 case ODPAT_SET_TUNNEL:
719 case ODPAT_SET_PRIORITY:
720 case ODPAT_POP_PRIORITY:
721 case ODPAT_DROP_SPOOFED_ARP:
722 /* No validation needed. */
726 if (nla_get_u32(a) >= DP_MAX_PORTS)
730 case ODPAT_SET_DL_TCI:
731 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
735 case ODPAT_SET_NW_TOS:
736 if (nla_get_u8(a) & INET_ECN_MASK)
751 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
753 struct sw_flow_actions *actions;
756 actions = flow_actions_alloc(flow->actions_len);
757 error = PTR_ERR(actions);
762 if (copy_from_user(actions->actions,
763 (struct nlattr __user __force *)flow->actions,
765 goto error_free_actions;
766 error = validate_actions(actions->actions, actions->actions_len);
768 goto error_free_actions;
775 return ERR_PTR(error);
778 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
781 struct timespec offset_ts, used, now_mono;
783 ktime_get_ts(&now_mono);
784 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
785 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
786 now_mono.tv_nsec - offset_ts.tv_nsec);
788 stats->used_sec = used.tv_sec;
789 stats->used_nsec = used.tv_nsec;
792 stats->used_nsec = 0;
795 stats->n_packets = flow->packet_count;
796 stats->n_bytes = flow->byte_count;
798 stats->tcp_flags = flow->tcp_flags;
802 static void clear_stats(struct sw_flow *flow)
806 flow->packet_count = 0;
807 flow->byte_count = 0;
810 static int expand_table(struct datapath *dp)
812 struct tbl *old_table = get_table_protected(dp);
813 struct tbl *new_table;
815 new_table = tbl_expand(old_table);
816 if (IS_ERR(new_table))
817 return PTR_ERR(new_table);
819 rcu_assign_pointer(dp->table, new_table);
820 tbl_deferred_destroy(old_table, NULL);
825 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
826 struct odp_flow_stats *stats)
828 struct tbl_node *flow_node;
829 struct sw_flow_key key;
830 struct sw_flow *flow;
832 struct sw_flow_actions *acts = NULL;
836 error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key,
841 hash = flow_hash(&key);
842 table = get_table_protected(dp);
843 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
847 if (!(uf->flags & ODPPF_CREATE))
850 /* Expand table, if necessary, to make room. */
851 if (tbl_count(table) >= tbl_n_buckets(table)) {
852 error = expand_table(dp);
855 table = get_table_protected(dp);
861 error = PTR_ERR(flow);
867 /* Obtain actions. */
868 acts = get_actions(&uf->flow);
869 error = PTR_ERR(acts);
871 goto error_free_flow;
872 rcu_assign_pointer(flow->sf_acts, acts);
874 /* Put flow in bucket. */
875 error = tbl_insert(table, &flow->tbl_node, hash);
877 goto error_free_flow_acts;
879 memset(stats, 0, sizeof(struct odp_flow_stats));
881 /* We found a matching flow. */
882 struct sw_flow_actions *old_acts, *new_acts;
884 flow = flow_cast(flow_node);
886 /* Bail out if we're not allowed to modify an existing flow. */
888 if (!(uf->flags & ODPPF_MODIFY))
892 new_acts = get_actions(&uf->flow);
893 error = PTR_ERR(new_acts);
894 if (IS_ERR(new_acts))
897 old_acts = rcu_dereference_protected(flow->sf_acts,
898 lockdep_is_held(&dp->mutex));
899 if (old_acts->actions_len != new_acts->actions_len ||
900 memcmp(old_acts->actions, new_acts->actions,
901 old_acts->actions_len)) {
902 rcu_assign_pointer(flow->sf_acts, new_acts);
903 flow_deferred_free_acts(old_acts);
908 /* Fetch stats, then clear them if necessary. */
909 spin_lock_bh(&flow->lock);
910 get_stats(flow, stats);
911 if (uf->flags & ODPPF_ZERO_STATS)
913 spin_unlock_bh(&flow->lock);
918 error_free_flow_acts:
921 flow->sf_acts = NULL;
927 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
929 struct odp_flow_stats stats;
930 struct odp_flow_put uf;
933 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
936 error = do_put_flow(dp, &uf, &stats);
940 if (copy_to_user(&ufp->flow.stats, &stats,
941 sizeof(struct odp_flow_stats)))
947 static int do_answer_query(struct datapath *dp, struct sw_flow *flow,
949 struct odp_flow_stats __user *ustats,
950 struct nlattr __user *actions,
951 u32 __user *actions_lenp)
953 struct sw_flow_actions *sf_acts;
954 struct odp_flow_stats stats;
957 spin_lock_bh(&flow->lock);
958 get_stats(flow, &stats);
959 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
962 spin_unlock_bh(&flow->lock);
964 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
965 get_user(actions_len, actions_lenp))
971 sf_acts = rcu_dereference_protected(flow->sf_acts,
972 lockdep_is_held(&dp->mutex));
973 if (put_user(sf_acts->actions_len, actions_lenp) ||
974 (actions && copy_to_user(actions, sf_acts->actions,
975 min(sf_acts->actions_len, actions_len))))
981 static int answer_query(struct datapath *dp, struct sw_flow *flow,
982 u32 query_flags, struct odp_flow __user *ufp)
984 struct nlattr __user *actions;
986 if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
989 return do_answer_query(dp, flow, query_flags,
990 &ufp->stats, actions, &ufp->actions_len);
993 static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len)
995 struct tbl *table = get_table_protected(dp);
996 struct tbl_node *flow_node;
997 struct sw_flow_key swkey;
1000 error = flow_copy_from_user(&swkey, key, key_len);
1002 return ERR_PTR(error);
1004 flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp);
1006 return ERR_PTR(-ENOENT);
1008 error = tbl_remove(table, flow_node);
1010 return ERR_PTR(error);
1012 /* XXX Returned flow_node's statistics might lose a few packets, since
1013 * other CPUs can be using this flow. We used to synchronize_rcu() to
1014 * make sure that we get completely accurate stats, but that blows our
1015 * performance, badly. */
1016 return flow_cast(flow_node);
1019 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1021 struct sw_flow *flow;
1025 if (copy_from_user(&uf, ufp, sizeof(uf)))
1028 flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
1030 return PTR_ERR(flow);
1032 error = answer_query(dp, flow, 0, ufp);
1033 flow_deferred_free(flow);
1037 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1039 struct tbl *table = get_table_protected(dp);
1042 for (i = 0; i < flowvec->n_flows; i++) {
1043 struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i];
1044 struct sw_flow_key key;
1046 struct tbl_node *flow_node;
1049 if (copy_from_user(&uf, ufp, sizeof(uf)))
1052 error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf.key, uf.key_len);
1056 flow_node = tbl_lookup(table, &uf.key, flow_hash(&key), flow_cmp);
1058 error = put_user(ENOENT, &ufp->stats.error);
1060 error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp);
1064 return flowvec->n_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 struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state)
1090 struct tbl *table = get_table_protected(dp);
1091 struct tbl_node *tbl_node;
1094 if (get_user(bucket, &state[0]) || get_user(obj, &state[1]))
1095 return ERR_PTR(-EFAULT);
1097 tbl_node = tbl_next(table, &bucket, &obj);
1099 if (put_user(bucket, &state[0]) || put_user(obj, &state[1]))
1100 return ERR_PTR(-EFAULT);
1102 return tbl_node ? flow_cast(tbl_node) : NULL;
1105 static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp)
1107 struct odp_flow __user *uflowp;
1108 struct nlattr __user *ukey;
1109 struct sw_flow *flow;
1112 flow = do_dump_flow(dp, udumpp->state);
1114 return PTR_ERR(flow);
1116 if (get_user(uflowp, (struct odp_flow __user *__user*)&udumpp->flow))
1120 return put_user(ODPFF_EOF, &uflowp->flags);
1122 if (put_user(0, &uflowp->flags) ||
1123 get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) ||
1124 get_user(key_len, &uflowp->key_len))
1127 key_len = flow_copy_to_user(ukey, &flow->key, key_len);
1130 if (put_user(key_len, &uflowp->key_len))
1133 return answer_query(dp, flow, 0, uflowp);
1136 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1138 struct sw_flow_key key;
1139 struct sk_buff *skb;
1140 struct sw_flow_actions *actions;
1146 if (execute->length < ETH_HLEN || execute->length > 65535)
1149 actions = flow_actions_alloc(execute->actions_len);
1150 if (IS_ERR(actions)) {
1151 err = PTR_ERR(actions);
1156 if (copy_from_user(actions->actions,
1157 (struct nlattr __user __force *)execute->actions, execute->actions_len))
1158 goto error_free_actions;
1160 err = validate_actions(actions->actions, execute->actions_len);
1162 goto error_free_actions;
1165 skb = alloc_skb(execute->length, GFP_KERNEL);
1167 goto error_free_actions;
1170 if (copy_from_user(skb_put(skb, execute->length),
1171 (const void __user __force *)execute->data,
1173 goto error_free_skb;
1175 skb_reset_mac_header(skb);
1178 /* Normally, setting the skb 'protocol' field would be handled by a
1179 * call to eth_type_trans(), but it assumes there's a sending
1180 * device, which we may not have. */
1181 if (ntohs(eth->h_proto) >= 1536)
1182 skb->protocol = eth->h_proto;
1184 skb->protocol = htons(ETH_P_802_2);
1186 err = flow_extract(skb, -1, &key, &is_frag);
1188 goto error_free_skb;
1191 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1205 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1207 struct odp_execute execute;
1209 if (copy_from_user(&execute, executep, sizeof(execute)))
1212 return do_execute(dp, &execute);
1215 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1217 struct tbl *table = get_table_protected(dp);
1218 struct odp_stats stats;
1221 stats.n_flows = tbl_count(table);
1222 stats.cur_capacity = tbl_n_buckets(table);
1223 stats.max_capacity = TBL_MAX_BUCKETS;
1224 stats.n_ports = dp->n_ports;
1225 stats.max_ports = DP_MAX_PORTS;
1226 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1227 for_each_possible_cpu(i) {
1228 const struct dp_stats_percpu *percpu_stats;
1229 struct dp_stats_percpu local_stats;
1232 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1235 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1236 local_stats = *percpu_stats;
1237 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1239 stats.n_frags += local_stats.n_frags;
1240 stats.n_hit += local_stats.n_hit;
1241 stats.n_missed += local_stats.n_missed;
1242 stats.n_lost += local_stats.n_lost;
1244 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1245 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1246 return copy_to_user(statsp, &stats, sizeof(stats)) ? -EFAULT : 0;
1249 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1250 int dp_min_mtu(const struct datapath *dp)
1257 list_for_each_entry_rcu (p, &dp->port_list, node) {
1260 /* Skip any internal ports, since that's what we're trying to
1262 if (is_internal_vport(p))
1265 dev_mtu = vport_get_mtu(p);
1266 if (!mtu || dev_mtu < mtu)
1270 return mtu ? mtu : ETH_DATA_LEN;
1273 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1274 * be called with RTNL lock. */
1275 void set_internal_devs_mtu(const struct datapath *dp)
1282 mtu = dp_min_mtu(dp);
1284 list_for_each_entry_rcu (p, &dp->port_list, node) {
1285 if (is_internal_vport(p))
1286 vport_set_mtu(p, mtu);
1290 static int put_port(const struct vport *p, struct odp_port __user *uop)
1294 memset(&op, 0, sizeof(op));
1297 strncpy(op.devname, vport_get_name(p), sizeof(op.devname));
1298 strncpy(op.type, vport_get_type(p), sizeof(op.type));
1299 vport_get_config(p, op.config);
1302 op.port = p->port_no;
1304 return copy_to_user(uop, &op, sizeof(op)) ? -EFAULT : 0;
1307 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1309 struct odp_port port;
1310 struct vport *vport;
1312 if (copy_from_user(&port, uport, sizeof(port)))
1315 if (port.devname[0]) {
1316 port.devname[IFNAMSIZ - 1] = '\0';
1319 vport = vport_locate(port.devname);
1324 if (vport->dp != dp)
1327 if (port.port >= DP_MAX_PORTS)
1330 vport = get_vport_protected(dp, port.port);
1335 return put_port(vport, uport);
1338 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1345 list_for_each_entry_rcu (p, &dp->port_list, node) {
1346 if (put_port(p, &uports[idx]))
1348 if (idx++ >= n_ports)
1355 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1357 struct odp_portvec pv;
1360 if (copy_from_user(&pv, upv, sizeof(pv)))
1363 retval = do_list_ports(dp, (struct odp_port __user __force *)pv.ports,
1368 return put_user(retval, &upv->n_ports);
1371 static int get_listen_mask(const struct file *f)
1373 return (long)f->private_data;
1376 static void set_listen_mask(struct file *f, int listen_mask)
1378 f->private_data = (void*)(long)listen_mask;
1381 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1384 int dp_idx = iminor(f->f_dentry->d_inode);
1385 struct datapath *dp;
1386 int drop_frags, listeners, port_no;
1387 unsigned int sflow_probability;
1390 /* Handle commands with special locking requirements up front. */
1393 err = create_dp(dp_idx, (char __user *)argp);
1396 case ODP_DP_DESTROY:
1397 err = destroy_dp(dp_idx);
1400 case ODP_VPORT_ATTACH:
1401 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1404 case ODP_VPORT_DETACH:
1405 err = get_user(port_no, (int __user *)argp);
1407 err = detach_port(dp_idx, port_no);
1411 err = vport_user_mod((struct odp_port __user *)argp);
1414 case ODP_VPORT_STATS_GET:
1415 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1418 case ODP_VPORT_STATS_SET:
1419 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1422 case ODP_VPORT_ETHER_GET:
1423 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1426 case ODP_VPORT_ETHER_SET:
1427 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1430 case ODP_VPORT_MTU_GET:
1431 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1434 case ODP_VPORT_MTU_SET:
1435 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1439 dp = get_dp_locked(dp_idx);
1446 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1449 case ODP_GET_DROP_FRAGS:
1450 err = put_user(dp->drop_frags, (int __user *)argp);
1453 case ODP_SET_DROP_FRAGS:
1454 err = get_user(drop_frags, (int __user *)argp);
1458 if (drop_frags != 0 && drop_frags != 1)
1460 dp->drop_frags = drop_frags;
1464 case ODP_GET_LISTEN_MASK:
1465 err = put_user(get_listen_mask(f), (int __user *)argp);
1468 case ODP_SET_LISTEN_MASK:
1469 err = get_user(listeners, (int __user *)argp);
1473 if (listeners & ~ODPL_ALL)
1476 set_listen_mask(f, listeners);
1479 case ODP_GET_SFLOW_PROBABILITY:
1480 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1483 case ODP_SET_SFLOW_PROBABILITY:
1484 err = get_user(sflow_probability, (unsigned int __user *)argp);
1486 dp->sflow_probability = sflow_probability;
1489 case ODP_VPORT_QUERY:
1490 err = query_port(dp, (struct odp_port __user *)argp);
1493 case ODP_VPORT_LIST:
1494 err = list_ports(dp, (struct odp_portvec __user *)argp);
1497 case ODP_FLOW_FLUSH:
1498 err = flush_flows(dp);
1502 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1506 err = del_flow(dp, (struct odp_flow __user *)argp);
1510 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1514 err = dump_flow(dp, (struct odp_flow_dump __user *)argp);
1518 err = execute_packet(dp, (struct odp_execute __user *)argp);
1525 mutex_unlock(&dp->mutex);
1530 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1533 for (i = 0; i < DP_N_QUEUES; i++) {
1534 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1540 #ifdef CONFIG_COMPAT
1541 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1543 struct compat_odp_portvec pv;
1546 if (copy_from_user(&pv, upv, sizeof(pv)))
1549 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1553 return put_user(retval, &upv->n_ports);
1556 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1558 compat_uptr_t key, actions;
1560 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1561 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1562 __get_user(key, &compat->key) ||
1563 __get_user(flow->key_len, &compat->key_len) ||
1564 __get_user(actions, &compat->actions) ||
1565 __get_user(flow->actions_len, &compat->actions_len) ||
1566 __get_user(flow->flags, &compat->flags))
1569 flow->key = (struct nlattr __force *)compat_ptr(key);
1570 flow->actions = (struct nlattr __force *)compat_ptr(actions);
1574 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1576 struct odp_flow_stats stats;
1577 struct odp_flow_put fp;
1580 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1581 get_user(fp.flags, &ufp->flags))
1584 error = do_put_flow(dp, &fp, &stats);
1588 if (copy_to_user(&ufp->flow.stats, &stats,
1589 sizeof(struct odp_flow_stats)))
1595 static int compat_answer_query(struct datapath *dp, struct sw_flow *flow,
1597 struct compat_odp_flow __user *ufp)
1599 compat_uptr_t actions;
1601 if (get_user(actions, &ufp->actions))
1604 return do_answer_query(dp, flow, query_flags, &ufp->stats,
1605 compat_ptr(actions), &ufp->actions_len);
1608 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1610 struct sw_flow *flow;
1614 if (compat_get_flow(&uf, ufp))
1617 flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
1619 return PTR_ERR(flow);
1621 error = compat_answer_query(dp, flow, 0, ufp);
1622 flow_deferred_free(flow);
1626 static int compat_query_flows(struct datapath *dp,
1627 struct compat_odp_flow __user *flows,
1630 struct tbl *table = get_table_protected(dp);
1633 for (i = 0; i < n_flows; i++) {
1634 struct compat_odp_flow __user *ufp = &flows[i];
1636 struct tbl_node *flow_node;
1637 struct sw_flow_key key;
1640 if (compat_get_flow(&uf, ufp))
1643 error = flow_copy_from_user(&key, (const struct nlattr __force __user *) uf.key, uf.key_len);
1647 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1649 error = put_user(ENOENT, &ufp->stats.error);
1651 error = compat_answer_query(dp, flow_cast(flow_node),
1659 static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp)
1661 struct compat_odp_flow __user *uflowp;
1662 compat_uptr_t compat_ufp;
1663 struct sw_flow *flow;
1667 flow = do_dump_flow(dp, udumpp->state);
1669 return PTR_ERR(flow);
1671 if (get_user(compat_ufp, &udumpp->flow))
1673 uflowp = compat_ptr(compat_ufp);
1676 return put_user(ODPFF_EOF, &uflowp->flags);
1678 if (put_user(0, &uflowp->flags) ||
1679 get_user(ukey, &uflowp->key) ||
1680 get_user(key_len, &uflowp->key_len))
1683 key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len);
1686 if (put_user(key_len, &uflowp->key_len))
1689 return compat_answer_query(dp, flow, 0, uflowp);
1692 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1693 int (*function)(struct datapath *,
1694 struct compat_odp_flow __user *,
1697 struct compat_odp_flowvec __user *uflowvec;
1698 struct compat_odp_flow __user *flows;
1699 struct compat_odp_flowvec flowvec;
1702 uflowvec = compat_ptr(argp);
1703 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof(*uflowvec)) ||
1704 copy_from_user(&flowvec, uflowvec, sizeof(flowvec)))
1707 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1710 flows = compat_ptr(flowvec.flows);
1711 if (!access_ok(VERIFY_WRITE, flows,
1712 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1715 retval = function(dp, flows, flowvec.n_flows);
1716 return (retval < 0 ? retval
1717 : retval == flowvec.n_flows ? 0
1718 : put_user(retval, &uflowvec->n_flows));
1721 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1723 struct odp_execute execute;
1724 compat_uptr_t actions;
1727 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1728 __get_user(actions, &uexecute->actions) ||
1729 __get_user(execute.actions_len, &uexecute->actions_len) ||
1730 __get_user(data, &uexecute->data) ||
1731 __get_user(execute.length, &uexecute->length))
1734 execute.actions = (struct nlattr __force *)compat_ptr(actions);
1735 execute.data = (const void __force *)compat_ptr(data);
1737 return do_execute(dp, &execute);
1740 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1742 int dp_idx = iminor(f->f_dentry->d_inode);
1743 struct datapath *dp;
1747 case ODP_DP_DESTROY:
1748 case ODP_FLOW_FLUSH:
1749 /* Ioctls that don't need any translation at all. */
1750 return openvswitch_ioctl(f, cmd, argp);
1753 case ODP_VPORT_ATTACH:
1754 case ODP_VPORT_DETACH:
1756 case ODP_VPORT_MTU_SET:
1757 case ODP_VPORT_MTU_GET:
1758 case ODP_VPORT_ETHER_SET:
1759 case ODP_VPORT_ETHER_GET:
1760 case ODP_VPORT_STATS_SET:
1761 case ODP_VPORT_STATS_GET:
1763 case ODP_GET_DROP_FRAGS:
1764 case ODP_SET_DROP_FRAGS:
1765 case ODP_SET_LISTEN_MASK:
1766 case ODP_GET_LISTEN_MASK:
1767 case ODP_SET_SFLOW_PROBABILITY:
1768 case ODP_GET_SFLOW_PROBABILITY:
1769 case ODP_VPORT_QUERY:
1770 /* Ioctls that just need their pointer argument extended. */
1771 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1774 dp = get_dp_locked(dp_idx);
1780 case ODP_VPORT_LIST32:
1781 err = compat_list_ports(dp, compat_ptr(argp));
1784 case ODP_FLOW_PUT32:
1785 err = compat_put_flow(dp, compat_ptr(argp));
1788 case ODP_FLOW_DEL32:
1789 err = compat_del_flow(dp, compat_ptr(argp));
1792 case ODP_FLOW_GET32:
1793 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1796 case ODP_FLOW_DUMP32:
1797 err = compat_dump_flow(dp, compat_ptr(argp));
1801 err = compat_execute(dp, compat_ptr(argp));
1808 mutex_unlock(&dp->mutex);
1814 /* Unfortunately this function is not exported so this is a verbatim copy
1815 * from net/core/datagram.c in 2.6.30. */
1816 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1817 u8 __user *to, int len,
1820 int start = skb_headlen(skb);
1822 int i, copy = start - offset;
1829 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1833 if ((len -= copy) == 0)
1840 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1843 WARN_ON(start > offset + len);
1845 end = start + skb_shinfo(skb)->frags[i].size;
1846 if ((copy = end - offset) > 0) {
1850 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1851 struct page *page = frag->page;
1856 csum2 = csum_and_copy_to_user(vaddr +
1863 *csump = csum_block_add(*csump, csum2, pos);
1873 if (skb_shinfo(skb)->frag_list) {
1874 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1876 for (; list; list=list->next) {
1879 WARN_ON(start > offset + len);
1881 end = start + list->len;
1882 if ((copy = end - offset) > 0) {
1886 if (skb_copy_and_csum_datagram(list,
1891 *csump = csum_block_add(*csump, csum2, pos);
1892 if ((len -= copy) == 0)
1908 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1909 size_t nbytes, loff_t *ppos)
1911 int listeners = get_listen_mask(f);
1912 int dp_idx = iminor(f->f_dentry->d_inode);
1913 struct datapath *dp = get_dp_locked(dp_idx);
1914 struct sk_buff *skb;
1915 size_t copy_bytes, tot_copy_bytes;
1921 if (nbytes == 0 || !listeners)
1927 for (i = 0; i < DP_N_QUEUES; i++) {
1928 if (listeners & (1 << i)) {
1929 skb = skb_dequeue(&dp->queues[i]);
1935 if (f->f_flags & O_NONBLOCK) {
1940 wait_event_interruptible(dp->waitqueue,
1941 dp_has_packet_of_interest(dp,
1944 if (signal_pending(current)) {
1945 retval = -ERESTARTSYS;
1950 mutex_unlock(&dp->mutex);
1952 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1955 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1956 if (copy_bytes == skb->len) {
1958 u16 csum_start, csum_offset;
1960 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1961 csum_start -= skb_headroom(skb);
1963 BUG_ON(csum_start >= skb_headlen(skb));
1964 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1965 copy_bytes - csum_start, &csum);
1967 __sum16 __user *csump;
1969 copy_bytes = csum_start;
1970 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1972 BUG_ON((char __user *)csump + sizeof(__sum16) >
1974 put_user(csum_fold(csum), csump);
1977 retval = skb_checksum_help(skb);
1984 iov.iov_len = copy_bytes;
1985 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1989 retval = tot_copy_bytes;
1995 mutex_unlock(&dp->mutex);
1999 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2001 int dp_idx = iminor(file->f_dentry->d_inode);
2002 struct datapath *dp = get_dp_locked(dp_idx);
2007 poll_wait(file, &dp->waitqueue, wait);
2008 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2009 mask |= POLLIN | POLLRDNORM;
2010 mutex_unlock(&dp->mutex);
2012 mask = POLLIN | POLLRDNORM | POLLHUP;
2017 static struct file_operations openvswitch_fops = {
2018 .owner = THIS_MODULE,
2019 .read = openvswitch_read,
2020 .poll = openvswitch_poll,
2021 .unlocked_ioctl = openvswitch_ioctl,
2022 #ifdef CONFIG_COMPAT
2023 .compat_ioctl = openvswitch_compat_ioctl,
2029 static int __init dp_init(void)
2031 struct sk_buff *dummy_skb;
2034 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2036 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2044 goto error_flow_exit;
2046 err = register_netdevice_notifier(&dp_device_notifier);
2048 goto error_vport_exit;
2050 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2052 goto error_unreg_notifier;
2056 error_unreg_notifier:
2057 unregister_netdevice_notifier(&dp_device_notifier);
2066 static void dp_cleanup(void)
2069 unregister_chrdev(major, "openvswitch");
2070 unregister_netdevice_notifier(&dp_device_notifier);
2075 module_init(dp_init);
2076 module_exit(dp_cleanup);
2078 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2079 MODULE_LICENSE("GPL");