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 internal_dev_port.type = ODP_VPORT_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';
398 dp = get_dp_locked(dp_idx);
401 goto out_unlock_rtnl;
403 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
404 if (!dp->ports[port_no])
410 err = new_vport(dp, &port, port_no);
414 set_internal_devs_mtu(dp);
415 dp_sysfs_add_if(get_vport_protected(dp, port_no));
417 err = put_user(port_no, &portp->port);
420 mutex_unlock(&dp->mutex);
427 int dp_detach_port(struct vport *p)
433 if (p->port_no != ODPP_LOCAL)
435 dp_ifinfo_notify(RTM_DELLINK, p);
437 /* First drop references to device. */
439 list_del_rcu(&p->node);
440 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
442 /* Then destroy it. */
450 static int detach_port(int dp_idx, int port_no)
457 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
461 dp = get_dp_locked(dp_idx);
464 goto out_unlock_rtnl;
466 p = get_vport_protected(dp, port_no);
471 err = dp_detach_port(p);
474 mutex_unlock(&dp->mutex);
481 /* Must be called with rcu_read_lock. */
482 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
484 struct datapath *dp = p->dp;
485 struct dp_stats_percpu *stats;
486 int stats_counter_off;
487 struct sw_flow_actions *acts;
488 struct loop_counter *loop;
491 OVS_CB(skb)->vport = p;
493 if (!OVS_CB(skb)->flow) {
494 struct sw_flow_key key;
495 struct tbl_node *flow_node;
498 /* Extract flow from 'skb' into 'key'. */
499 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
500 if (unlikely(error)) {
505 if (is_frag && dp->drop_frags) {
507 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
512 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
513 flow_hash(&key), flow_cmp);
514 if (unlikely(!flow_node)) {
515 struct dp_upcall_info upcall;
517 upcall.type = _ODPL_MISS_NR;
520 upcall.sample_pool = 0;
521 upcall.actions = NULL;
522 upcall.actions_len = 0;
523 dp_upcall(dp, skb, &upcall);
524 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
528 OVS_CB(skb)->flow = flow_cast(flow_node);
531 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
532 flow_used(OVS_CB(skb)->flow, skb);
534 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
536 /* Check whether we've looped too much. */
537 loop = loop_get_counter();
538 if (unlikely(++loop->count > MAX_LOOPS))
539 loop->looping = true;
540 if (unlikely(loop->looping)) {
541 loop_suppress(dp, acts);
546 /* Execute actions. */
547 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
550 /* Check whether sub-actions looped too much. */
551 if (unlikely(loop->looping))
552 loop_suppress(dp, acts);
555 /* Decrement loop counter. */
557 loop->looping = false;
561 /* Update datapath statistics. */
563 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
565 write_seqcount_begin(&stats->seqlock);
566 (*(u64 *)((u8 *)stats + stats_counter_off))++;
567 write_seqcount_end(&stats->seqlock);
572 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
574 u16 csum_start, csum_offset;
577 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
578 csum_start -= skb_headroom(skb);
579 BUG_ON(csum_start >= skb_headlen(skb));
581 skb_copy_bits(skb, 0, to, csum_start);
583 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
584 skb->len - csum_start, 0);
585 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
588 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
589 * unless we broke up a GSO packet. */
590 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
591 const struct dp_upcall_info *upcall_info)
593 struct sk_buff *nskb;
597 if (OVS_CB(skb)->vport)
598 port_no = OVS_CB(skb)->vport->port_no;
600 port_no = ODPP_LOCAL;
603 struct odp_packet *upcall;
604 struct sk_buff *user_skb; /* to be queued to userspace */
611 len = sizeof(struct odp_packet);
612 len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */
613 len += nla_total_size(skb->len);
614 len += nla_total_size(FLOW_BUFSIZE);
615 if (upcall_info->userdata)
616 len += nla_total_size(8);
617 if (upcall_info->sample_pool)
618 len += nla_total_size(4);
619 if (upcall_info->actions_len)
620 len += nla_total_size(upcall_info->actions_len);
622 user_skb = alloc_skb(len, GFP_ATOMIC);
626 upcall = (struct odp_packet *)__skb_put(user_skb, sizeof(*upcall));
627 upcall->dp_idx = dp->dp_idx;
629 nla_put_u32(user_skb, ODP_PACKET_ATTR_TYPE, upcall_info->type);
631 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
632 flow_to_nlattrs(upcall_info->key, user_skb);
633 nla_nest_end(user_skb, nla);
635 if (upcall_info->userdata)
636 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
637 if (upcall_info->sample_pool)
638 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
639 if (upcall_info->actions_len) {
640 const struct nlattr *actions = upcall_info->actions;
641 u32 actions_len = upcall_info->actions_len;
643 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
644 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
645 nla_nest_end(user_skb, nla);
648 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
649 if (skb->ip_summed == CHECKSUM_PARTIAL)
650 copy_and_csum_skb(skb, nla_data(nla));
652 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
654 upcall->len = user_skb->len;
655 skb_queue_tail(&dp->queues[upcall_info->type], user_skb);
664 while ((skb = nskb) != NULL) {
671 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
673 struct dp_stats_percpu *stats;
674 struct sk_buff_head *queue;
677 WARN_ON_ONCE(skb_shared(skb));
678 BUG_ON(upcall_info->type >= DP_N_QUEUES);
680 queue = &dp->queues[upcall_info->type];
682 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
685 forward_ip_summed(skb);
687 err = vswitch_skb_checksum_setup(skb);
691 /* Break apart GSO packets into their component pieces. Otherwise
692 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
693 if (skb_is_gso(skb)) {
694 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
704 err = queue_control_packets(dp, skb, upcall_info);
705 wake_up_interruptible(&dp->waitqueue);
712 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
714 write_seqcount_begin(&stats->seqlock);
716 write_seqcount_end(&stats->seqlock);
723 static int flush_flows(struct datapath *dp)
725 struct tbl *old_table = get_table_protected(dp);
726 struct tbl *new_table;
728 new_table = tbl_create(TBL_MIN_BUCKETS);
732 rcu_assign_pointer(dp->table, new_table);
734 tbl_deferred_destroy(old_table, flow_free_tbl);
739 static int validate_actions(const struct nlattr *actions, u32 actions_len)
741 const struct nlattr *a;
744 nla_for_each_attr(a, actions, actions_len, rem) {
745 static const u32 action_lens[ODPAT_MAX + 1] = {
747 [ODPAT_CONTROLLER] = 8,
748 [ODPAT_SET_DL_TCI] = 2,
749 [ODPAT_STRIP_VLAN] = 0,
750 [ODPAT_SET_DL_SRC] = ETH_ALEN,
751 [ODPAT_SET_DL_DST] = ETH_ALEN,
752 [ODPAT_SET_NW_SRC] = 4,
753 [ODPAT_SET_NW_DST] = 4,
754 [ODPAT_SET_NW_TOS] = 1,
755 [ODPAT_SET_TP_SRC] = 2,
756 [ODPAT_SET_TP_DST] = 2,
757 [ODPAT_SET_TUNNEL] = 8,
758 [ODPAT_SET_PRIORITY] = 4,
759 [ODPAT_POP_PRIORITY] = 0,
760 [ODPAT_DROP_SPOOFED_ARP] = 0,
762 int type = nla_type(a);
764 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
771 case ODPAT_CONTROLLER:
772 case ODPAT_STRIP_VLAN:
773 case ODPAT_SET_DL_SRC:
774 case ODPAT_SET_DL_DST:
775 case ODPAT_SET_NW_SRC:
776 case ODPAT_SET_NW_DST:
777 case ODPAT_SET_TP_SRC:
778 case ODPAT_SET_TP_DST:
779 case ODPAT_SET_TUNNEL:
780 case ODPAT_SET_PRIORITY:
781 case ODPAT_POP_PRIORITY:
782 case ODPAT_DROP_SPOOFED_ARP:
783 /* No validation needed. */
787 if (nla_get_u32(a) >= DP_MAX_PORTS)
791 case ODPAT_SET_DL_TCI:
792 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
796 case ODPAT_SET_NW_TOS:
797 if (nla_get_u8(a) & INET_ECN_MASK)
812 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
814 struct sw_flow_actions *actions;
817 actions = flow_actions_alloc(flow->actions_len);
818 error = PTR_ERR(actions);
823 if (copy_from_user(actions->actions,
824 (struct nlattr __user __force *)flow->actions,
826 goto error_free_actions;
827 error = validate_actions(actions->actions, actions->actions_len);
829 goto error_free_actions;
836 return ERR_PTR(error);
839 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
842 struct timespec offset_ts, used, now_mono;
844 ktime_get_ts(&now_mono);
845 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
846 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
847 now_mono.tv_nsec - offset_ts.tv_nsec);
849 stats->used_sec = used.tv_sec;
850 stats->used_nsec = used.tv_nsec;
853 stats->used_nsec = 0;
856 stats->n_packets = flow->packet_count;
857 stats->n_bytes = flow->byte_count;
859 stats->tcp_flags = flow->tcp_flags;
863 static void clear_stats(struct sw_flow *flow)
867 flow->packet_count = 0;
868 flow->byte_count = 0;
871 static int expand_table(struct datapath *dp)
873 struct tbl *old_table = get_table_protected(dp);
874 struct tbl *new_table;
876 new_table = tbl_expand(old_table);
877 if (IS_ERR(new_table))
878 return PTR_ERR(new_table);
880 rcu_assign_pointer(dp->table, new_table);
881 tbl_deferred_destroy(old_table, NULL);
886 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
887 struct odp_flow_stats *stats)
889 struct tbl_node *flow_node;
890 struct sw_flow_key key;
891 struct sw_flow *flow;
893 struct sw_flow_actions *acts = NULL;
897 error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key,
902 hash = flow_hash(&key);
903 table = get_table_protected(dp);
904 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
908 if (!(uf->flags & ODPPF_CREATE))
911 /* Expand table, if necessary, to make room. */
912 if (tbl_count(table) >= tbl_n_buckets(table)) {
913 error = expand_table(dp);
916 table = get_table_protected(dp);
922 error = PTR_ERR(flow);
928 /* Obtain actions. */
929 acts = get_actions(&uf->flow);
930 error = PTR_ERR(acts);
932 goto error_free_flow;
933 rcu_assign_pointer(flow->sf_acts, acts);
935 /* Put flow in bucket. */
936 error = tbl_insert(table, &flow->tbl_node, hash);
938 goto error_free_flow_acts;
940 memset(stats, 0, sizeof(struct odp_flow_stats));
942 /* We found a matching flow. */
943 struct sw_flow_actions *old_acts, *new_acts;
945 flow = flow_cast(flow_node);
947 /* Bail out if we're not allowed to modify an existing flow. */
949 if (!(uf->flags & ODPPF_MODIFY))
953 new_acts = get_actions(&uf->flow);
954 error = PTR_ERR(new_acts);
955 if (IS_ERR(new_acts))
958 old_acts = rcu_dereference_protected(flow->sf_acts,
959 lockdep_is_held(&dp->mutex));
960 if (old_acts->actions_len != new_acts->actions_len ||
961 memcmp(old_acts->actions, new_acts->actions,
962 old_acts->actions_len)) {
963 rcu_assign_pointer(flow->sf_acts, new_acts);
964 flow_deferred_free_acts(old_acts);
969 /* Fetch stats, then clear them if necessary. */
970 spin_lock_bh(&flow->lock);
971 get_stats(flow, stats);
972 if (uf->flags & ODPPF_ZERO_STATS)
974 spin_unlock_bh(&flow->lock);
979 error_free_flow_acts:
982 flow->sf_acts = NULL;
988 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
990 struct odp_flow_stats stats;
991 struct odp_flow_put uf;
994 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
997 error = do_put_flow(dp, &uf, &stats);
1001 if (copy_to_user(&ufp->flow.stats, &stats,
1002 sizeof(struct odp_flow_stats)))
1008 static int do_answer_query(struct datapath *dp, struct sw_flow *flow,
1010 struct odp_flow_stats __user *ustats,
1011 struct nlattr __user *actions,
1012 u32 __user *actions_lenp)
1014 struct sw_flow_actions *sf_acts;
1015 struct odp_flow_stats stats;
1018 spin_lock_bh(&flow->lock);
1019 get_stats(flow, &stats);
1020 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1021 flow->tcp_flags = 0;
1023 spin_unlock_bh(&flow->lock);
1025 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1026 get_user(actions_len, actions_lenp))
1032 sf_acts = rcu_dereference_protected(flow->sf_acts,
1033 lockdep_is_held(&dp->mutex));
1034 if (put_user(sf_acts->actions_len, actions_lenp) ||
1035 (actions && copy_to_user(actions, sf_acts->actions,
1036 min(sf_acts->actions_len, actions_len))))
1042 static int answer_query(struct datapath *dp, struct sw_flow *flow,
1043 u32 query_flags, struct odp_flow __user *ufp)
1045 struct nlattr __user *actions;
1047 if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
1050 return do_answer_query(dp, flow, query_flags,
1051 &ufp->stats, actions, &ufp->actions_len);
1054 static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len)
1056 struct tbl *table = get_table_protected(dp);
1057 struct tbl_node *flow_node;
1058 struct sw_flow_key swkey;
1061 error = flow_copy_from_user(&swkey, key, key_len);
1063 return ERR_PTR(error);
1065 flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp);
1067 return ERR_PTR(-ENOENT);
1069 error = tbl_remove(table, flow_node);
1071 return ERR_PTR(error);
1073 /* XXX Returned flow_node's statistics might lose a few packets, since
1074 * other CPUs can be using this flow. We used to synchronize_rcu() to
1075 * make sure that we get completely accurate stats, but that blows our
1076 * performance, badly. */
1077 return flow_cast(flow_node);
1080 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1082 struct sw_flow *flow;
1086 if (copy_from_user(&uf, ufp, sizeof(uf)))
1089 flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
1091 return PTR_ERR(flow);
1093 error = answer_query(dp, flow, 0, ufp);
1094 flow_deferred_free(flow);
1098 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1100 struct tbl *table = get_table_protected(dp);
1103 for (i = 0; i < flowvec->n_flows; i++) {
1104 struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i];
1105 struct sw_flow_key key;
1107 struct tbl_node *flow_node;
1110 if (copy_from_user(&uf, ufp, sizeof(uf)))
1113 error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf.key, uf.key_len);
1117 flow_node = tbl_lookup(table, &uf.key, flow_hash(&key), flow_cmp);
1119 error = put_user(ENOENT, &ufp->stats.error);
1121 error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp);
1125 return flowvec->n_flows;
1128 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1129 int (*function)(struct datapath *,
1130 const struct odp_flowvec *))
1132 struct odp_flowvec __user *uflowvec;
1133 struct odp_flowvec flowvec;
1136 uflowvec = (struct odp_flowvec __user *)argp;
1137 if (copy_from_user(&flowvec, uflowvec, sizeof(flowvec)))
1140 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1143 retval = function(dp, &flowvec);
1144 return (retval < 0 ? retval
1145 : retval == flowvec.n_flows ? 0
1146 : put_user(retval, &uflowvec->n_flows));
1149 static struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state)
1151 struct tbl *table = get_table_protected(dp);
1152 struct tbl_node *tbl_node;
1155 if (get_user(bucket, &state[0]) || get_user(obj, &state[1]))
1156 return ERR_PTR(-EFAULT);
1158 tbl_node = tbl_next(table, &bucket, &obj);
1160 if (put_user(bucket, &state[0]) || put_user(obj, &state[1]))
1161 return ERR_PTR(-EFAULT);
1163 return tbl_node ? flow_cast(tbl_node) : NULL;
1166 static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp)
1168 struct odp_flow __user *uflowp;
1169 struct nlattr __user *ukey;
1170 struct sw_flow *flow;
1173 flow = do_dump_flow(dp, udumpp->state);
1175 return PTR_ERR(flow);
1177 if (get_user(uflowp, (struct odp_flow __user *__user*)&udumpp->flow))
1181 return put_user(ODPFF_EOF, &uflowp->flags);
1183 if (put_user(0, &uflowp->flags) ||
1184 get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) ||
1185 get_user(key_len, &uflowp->key_len))
1188 key_len = flow_copy_to_user(ukey, &flow->key, key_len);
1191 if (put_user(key_len, &uflowp->key_len))
1194 return answer_query(dp, flow, 0, uflowp);
1197 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1199 struct sw_flow_key key;
1200 struct sk_buff *skb;
1201 struct sw_flow_actions *actions;
1207 if (execute->length < ETH_HLEN || execute->length > 65535)
1210 actions = flow_actions_alloc(execute->actions_len);
1211 if (IS_ERR(actions)) {
1212 err = PTR_ERR(actions);
1217 if (copy_from_user(actions->actions,
1218 (struct nlattr __user __force *)execute->actions, execute->actions_len))
1219 goto error_free_actions;
1221 err = validate_actions(actions->actions, execute->actions_len);
1223 goto error_free_actions;
1226 skb = alloc_skb(execute->length, GFP_KERNEL);
1228 goto error_free_actions;
1231 if (copy_from_user(skb_put(skb, execute->length),
1232 (const void __user __force *)execute->data,
1234 goto error_free_skb;
1236 skb_reset_mac_header(skb);
1239 /* Normally, setting the skb 'protocol' field would be handled by a
1240 * call to eth_type_trans(), but it assumes there's a sending
1241 * device, which we may not have. */
1242 if (ntohs(eth->h_proto) >= 1536)
1243 skb->protocol = eth->h_proto;
1245 skb->protocol = htons(ETH_P_802_2);
1247 err = flow_extract(skb, -1, &key, &is_frag);
1249 goto error_free_skb;
1252 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1266 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1268 struct odp_execute execute;
1270 if (copy_from_user(&execute, executep, sizeof(execute)))
1273 return do_execute(dp, &execute);
1276 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1278 struct tbl *table = get_table_protected(dp);
1279 struct odp_stats stats;
1282 stats.n_flows = tbl_count(table);
1283 stats.cur_capacity = tbl_n_buckets(table);
1284 stats.max_capacity = TBL_MAX_BUCKETS;
1285 stats.n_ports = dp->n_ports;
1286 stats.max_ports = DP_MAX_PORTS;
1287 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1288 for_each_possible_cpu(i) {
1289 const struct dp_stats_percpu *percpu_stats;
1290 struct dp_stats_percpu local_stats;
1293 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1296 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1297 local_stats = *percpu_stats;
1298 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1300 stats.n_frags += local_stats.n_frags;
1301 stats.n_hit += local_stats.n_hit;
1302 stats.n_missed += local_stats.n_missed;
1303 stats.n_lost += local_stats.n_lost;
1305 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1306 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1307 return copy_to_user(statsp, &stats, sizeof(stats)) ? -EFAULT : 0;
1310 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1311 int dp_min_mtu(const struct datapath *dp)
1318 list_for_each_entry_rcu (p, &dp->port_list, node) {
1321 /* Skip any internal ports, since that's what we're trying to
1323 if (is_internal_vport(p))
1326 dev_mtu = vport_get_mtu(p);
1327 if (!mtu || dev_mtu < mtu)
1331 return mtu ? mtu : ETH_DATA_LEN;
1334 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1335 * be called with RTNL lock. */
1336 void set_internal_devs_mtu(const struct datapath *dp)
1343 mtu = dp_min_mtu(dp);
1345 list_for_each_entry_rcu (p, &dp->port_list, node) {
1346 if (is_internal_vport(p))
1347 vport_set_mtu(p, mtu);
1351 static void compose_odp_port(const struct vport *vport, struct odp_port *odp_port)
1354 strncpy(odp_port->devname, vport_get_name(vport), sizeof(odp_port->devname));
1355 odp_port->type = vport_get_type(vport);
1356 vport_get_config(vport, odp_port->config);
1357 odp_port->port = vport->port_no;
1358 odp_port->dp_idx = vport->dp->dp_idx;
1362 static int query_port(int dp_idx, struct odp_port __user *uport)
1364 struct odp_port port;
1366 if (copy_from_user(&port, uport, sizeof(port)))
1369 if (port.devname[0]) {
1370 struct vport *vport;
1372 port.devname[IFNAMSIZ - 1] = '\0';
1375 vport = vport_locate(port.devname);
1377 compose_odp_port(vport, &port);
1383 struct vport *vport;
1384 struct datapath *dp;
1386 if (port.port >= DP_MAX_PORTS)
1389 dp = get_dp_locked(dp_idx);
1393 vport = get_vport_protected(dp, port.port);
1395 compose_odp_port(vport, &port);
1396 mutex_unlock(&dp->mutex);
1402 return copy_to_user(uport, &port, sizeof(struct odp_port));
1405 static int do_dump_port(struct datapath *dp, struct odp_vport_dump *dump)
1409 for (port_no = dump->port_no; port_no < DP_MAX_PORTS; port_no++) {
1410 struct vport *vport = get_vport_protected(dp, port_no);
1412 struct odp_port odp_port;
1414 compose_odp_port(vport, &odp_port);
1415 return copy_to_user((struct odp_port __force __user*)dump->port, &odp_port, sizeof(struct odp_port));
1419 return put_user('\0', (char __force __user*)&dump->port->devname[0]);
1422 static int dump_port(struct datapath *dp, struct odp_vport_dump __user *udump)
1424 struct odp_vport_dump dump;
1426 if (copy_from_user(&dump, udump, sizeof(dump)))
1429 return do_dump_port(dp, &dump);
1432 static int get_listen_mask(const struct file *f)
1434 return (long)f->private_data;
1437 static void set_listen_mask(struct file *f, int listen_mask)
1439 f->private_data = (void*)(long)listen_mask;
1442 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1445 int dp_idx = iminor(f->f_dentry->d_inode);
1446 struct datapath *dp;
1447 int drop_frags, listeners, port_no;
1448 unsigned int sflow_probability;
1451 /* Handle commands with special locking requirements up front. */
1454 err = create_dp(dp_idx, (char __user *)argp);
1457 case ODP_DP_DESTROY:
1458 err = destroy_dp(dp_idx);
1461 case ODP_VPORT_ATTACH:
1462 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1465 case ODP_VPORT_DETACH:
1466 err = get_user(port_no, (int __user *)argp);
1468 err = detach_port(dp_idx, port_no);
1471 case ODP_VPORT_QUERY:
1472 err = query_port(dp_idx, (struct odp_port __user *)argp);
1476 err = vport_user_mod((struct odp_port __user *)argp);
1479 case ODP_VPORT_STATS_GET:
1480 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1483 case ODP_VPORT_STATS_SET:
1484 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1487 case ODP_VPORT_ETHER_GET:
1488 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1491 case ODP_VPORT_ETHER_SET:
1492 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1495 case ODP_VPORT_MTU_GET:
1496 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1499 case ODP_VPORT_MTU_SET:
1500 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1504 dp = get_dp_locked(dp_idx);
1511 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1514 case ODP_GET_DROP_FRAGS:
1515 err = put_user(dp->drop_frags, (int __user *)argp);
1518 case ODP_SET_DROP_FRAGS:
1519 err = get_user(drop_frags, (int __user *)argp);
1523 if (drop_frags != 0 && drop_frags != 1)
1525 dp->drop_frags = drop_frags;
1529 case ODP_GET_LISTEN_MASK:
1530 err = put_user(get_listen_mask(f), (int __user *)argp);
1533 case ODP_SET_LISTEN_MASK:
1534 err = get_user(listeners, (int __user *)argp);
1538 if (listeners & ~ODPL_ALL)
1541 set_listen_mask(f, listeners);
1544 case ODP_GET_SFLOW_PROBABILITY:
1545 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1548 case ODP_SET_SFLOW_PROBABILITY:
1549 err = get_user(sflow_probability, (unsigned int __user *)argp);
1551 dp->sflow_probability = sflow_probability;
1554 case ODP_VPORT_DUMP:
1555 err = dump_port(dp, (struct odp_vport_dump __user *)argp);
1558 case ODP_FLOW_FLUSH:
1559 err = flush_flows(dp);
1563 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1567 err = del_flow(dp, (struct odp_flow __user *)argp);
1571 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1575 err = dump_flow(dp, (struct odp_flow_dump __user *)argp);
1579 err = execute_packet(dp, (struct odp_execute __user *)argp);
1586 mutex_unlock(&dp->mutex);
1591 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1594 for (i = 0; i < DP_N_QUEUES; i++) {
1595 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1601 #ifdef CONFIG_COMPAT
1602 static int compat_dump_port(struct datapath *dp, struct compat_odp_vport_dump __user *compat)
1604 struct odp_vport_dump dump;
1607 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_vport_dump)) ||
1608 __get_user(port, &compat->port) ||
1609 __get_user(dump.port_no, &compat->port_no))
1612 dump.port = (struct odp_port __force *)compat_ptr(port);
1613 return do_dump_port(dp, &dump);
1616 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1618 compat_uptr_t key, actions;
1620 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1621 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1622 __get_user(key, &compat->key) ||
1623 __get_user(flow->key_len, &compat->key_len) ||
1624 __get_user(actions, &compat->actions) ||
1625 __get_user(flow->actions_len, &compat->actions_len) ||
1626 __get_user(flow->flags, &compat->flags))
1629 flow->key = (struct nlattr __force *)compat_ptr(key);
1630 flow->actions = (struct nlattr __force *)compat_ptr(actions);
1634 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1636 struct odp_flow_stats stats;
1637 struct odp_flow_put fp;
1640 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1641 get_user(fp.flags, &ufp->flags))
1644 error = do_put_flow(dp, &fp, &stats);
1648 if (copy_to_user(&ufp->flow.stats, &stats,
1649 sizeof(struct odp_flow_stats)))
1655 static int compat_answer_query(struct datapath *dp, struct sw_flow *flow,
1657 struct compat_odp_flow __user *ufp)
1659 compat_uptr_t actions;
1661 if (get_user(actions, &ufp->actions))
1664 return do_answer_query(dp, flow, query_flags, &ufp->stats,
1665 compat_ptr(actions), &ufp->actions_len);
1668 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1670 struct sw_flow *flow;
1674 if (compat_get_flow(&uf, ufp))
1677 flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
1679 return PTR_ERR(flow);
1681 error = compat_answer_query(dp, flow, 0, ufp);
1682 flow_deferred_free(flow);
1686 static int compat_query_flows(struct datapath *dp,
1687 struct compat_odp_flow __user *flows,
1690 struct tbl *table = get_table_protected(dp);
1693 for (i = 0; i < n_flows; i++) {
1694 struct compat_odp_flow __user *ufp = &flows[i];
1696 struct tbl_node *flow_node;
1697 struct sw_flow_key key;
1700 if (compat_get_flow(&uf, ufp))
1703 error = flow_copy_from_user(&key, (const struct nlattr __force __user *) uf.key, uf.key_len);
1707 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1709 error = put_user(ENOENT, &ufp->stats.error);
1711 error = compat_answer_query(dp, flow_cast(flow_node),
1719 static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp)
1721 struct compat_odp_flow __user *uflowp;
1722 compat_uptr_t compat_ufp;
1723 struct sw_flow *flow;
1727 flow = do_dump_flow(dp, udumpp->state);
1729 return PTR_ERR(flow);
1731 if (get_user(compat_ufp, &udumpp->flow))
1733 uflowp = compat_ptr(compat_ufp);
1736 return put_user(ODPFF_EOF, &uflowp->flags);
1738 if (put_user(0, &uflowp->flags) ||
1739 get_user(ukey, &uflowp->key) ||
1740 get_user(key_len, &uflowp->key_len))
1743 key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len);
1746 if (put_user(key_len, &uflowp->key_len))
1749 return compat_answer_query(dp, flow, 0, uflowp);
1752 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1753 int (*function)(struct datapath *,
1754 struct compat_odp_flow __user *,
1757 struct compat_odp_flowvec __user *uflowvec;
1758 struct compat_odp_flow __user *flows;
1759 struct compat_odp_flowvec flowvec;
1762 uflowvec = compat_ptr(argp);
1763 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof(*uflowvec)) ||
1764 copy_from_user(&flowvec, uflowvec, sizeof(flowvec)))
1767 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1770 flows = compat_ptr(flowvec.flows);
1771 if (!access_ok(VERIFY_WRITE, flows,
1772 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1775 retval = function(dp, flows, flowvec.n_flows);
1776 return (retval < 0 ? retval
1777 : retval == flowvec.n_flows ? 0
1778 : put_user(retval, &uflowvec->n_flows));
1781 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1783 struct odp_execute execute;
1784 compat_uptr_t actions;
1787 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1788 __get_user(actions, &uexecute->actions) ||
1789 __get_user(execute.actions_len, &uexecute->actions_len) ||
1790 __get_user(data, &uexecute->data) ||
1791 __get_user(execute.length, &uexecute->length))
1794 execute.actions = (struct nlattr __force *)compat_ptr(actions);
1795 execute.data = (const void __force *)compat_ptr(data);
1797 return do_execute(dp, &execute);
1800 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1802 int dp_idx = iminor(f->f_dentry->d_inode);
1803 struct datapath *dp;
1807 case ODP_DP_DESTROY:
1808 case ODP_FLOW_FLUSH:
1809 /* Ioctls that don't need any translation at all. */
1810 return openvswitch_ioctl(f, cmd, argp);
1813 case ODP_VPORT_ATTACH:
1814 case ODP_VPORT_DETACH:
1816 case ODP_VPORT_MTU_SET:
1817 case ODP_VPORT_MTU_GET:
1818 case ODP_VPORT_ETHER_SET:
1819 case ODP_VPORT_ETHER_GET:
1820 case ODP_VPORT_STATS_SET:
1821 case ODP_VPORT_STATS_GET:
1823 case ODP_GET_DROP_FRAGS:
1824 case ODP_SET_DROP_FRAGS:
1825 case ODP_SET_LISTEN_MASK:
1826 case ODP_GET_LISTEN_MASK:
1827 case ODP_SET_SFLOW_PROBABILITY:
1828 case ODP_GET_SFLOW_PROBABILITY:
1829 case ODP_VPORT_QUERY:
1830 /* Ioctls that just need their pointer argument extended. */
1831 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1834 dp = get_dp_locked(dp_idx);
1840 case ODP_VPORT_DUMP32:
1841 err = compat_dump_port(dp, compat_ptr(argp));
1844 case ODP_FLOW_PUT32:
1845 err = compat_put_flow(dp, compat_ptr(argp));
1848 case ODP_FLOW_DEL32:
1849 err = compat_del_flow(dp, compat_ptr(argp));
1852 case ODP_FLOW_GET32:
1853 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1856 case ODP_FLOW_DUMP32:
1857 err = compat_dump_flow(dp, compat_ptr(argp));
1861 err = compat_execute(dp, compat_ptr(argp));
1868 mutex_unlock(&dp->mutex);
1874 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1875 size_t nbytes, loff_t *ppos)
1877 int listeners = get_listen_mask(f);
1878 int dp_idx = iminor(f->f_dentry->d_inode);
1879 struct datapath *dp = get_dp_locked(dp_idx);
1880 struct sk_buff *skb;
1887 if (nbytes == 0 || !listeners)
1893 for (i = 0; i < DP_N_QUEUES; i++) {
1894 if (listeners & (1 << i)) {
1895 skb = skb_dequeue(&dp->queues[i]);
1901 if (f->f_flags & O_NONBLOCK) {
1906 wait_event_interruptible(dp->waitqueue,
1907 dp_has_packet_of_interest(dp,
1910 if (signal_pending(current)) {
1911 retval = -ERESTARTSYS;
1916 mutex_unlock(&dp->mutex);
1919 iov.iov_len = min_t(size_t, skb->len, nbytes);
1920 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1928 mutex_unlock(&dp->mutex);
1932 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1934 int dp_idx = iminor(file->f_dentry->d_inode);
1935 struct datapath *dp = get_dp_locked(dp_idx);
1940 poll_wait(file, &dp->waitqueue, wait);
1941 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1942 mask |= POLLIN | POLLRDNORM;
1943 mutex_unlock(&dp->mutex);
1945 mask = POLLIN | POLLRDNORM | POLLHUP;
1950 static struct file_operations openvswitch_fops = {
1951 .owner = THIS_MODULE,
1952 .read = openvswitch_read,
1953 .poll = openvswitch_poll,
1954 .unlocked_ioctl = openvswitch_ioctl,
1955 #ifdef CONFIG_COMPAT
1956 .compat_ioctl = openvswitch_compat_ioctl,
1962 static int __init dp_init(void)
1964 struct sk_buff *dummy_skb;
1967 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1969 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1977 goto error_flow_exit;
1979 err = register_netdevice_notifier(&dp_device_notifier);
1981 goto error_vport_exit;
1983 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
1985 goto error_unreg_notifier;
1989 error_unreg_notifier:
1990 unregister_netdevice_notifier(&dp_device_notifier);
1999 static void dp_cleanup(void)
2002 unregister_chrdev(major, "openvswitch");
2003 unregister_netdevice_notifier(&dp_device_notifier);
2008 module_init(dp_init);
2009 module_exit(dp_cleanup);
2011 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2012 MODULE_LICENSE("GPL");