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"
50 #include "loop_counter.h"
51 #include "odp-compat.h"
53 #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 dp_port structures with just
69 static struct datapath *dps[ODP_MAX];
70 static DEFINE_MUTEX(dp_mutex);
72 static int new_dp_port(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(dps[dp_idx]);
81 EXPORT_SYMBOL_GPL(get_dp);
83 static struct datapath *get_dp_locked(int dp_idx)
87 mutex_lock(&dp_mutex);
90 mutex_lock(&dp->mutex);
91 mutex_unlock(&dp_mutex);
95 /* Must be called with rcu_read_lock or RTNL lock. */
96 const char *dp_name(const struct datapath *dp)
98 return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
101 static inline size_t br_nlmsg_size(void)
103 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
104 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
105 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
106 + nla_total_size(4) /* IFLA_MASTER */
107 + nla_total_size(4) /* IFLA_MTU */
108 + nla_total_size(4) /* IFLA_LINK */
109 + nla_total_size(1); /* IFLA_OPERSTATE */
112 static int dp_fill_ifinfo(struct sk_buff *skb,
113 const struct dp_port *port,
114 int event, unsigned int flags)
116 const struct datapath *dp = port->dp;
117 int ifindex = vport_get_ifindex(port->vport);
118 int iflink = vport_get_iflink(port->vport);
119 struct ifinfomsg *hdr;
120 struct nlmsghdr *nlh;
128 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
132 hdr = nlmsg_data(nlh);
133 hdr->ifi_family = AF_BRIDGE;
135 hdr->ifi_type = ARPHRD_ETHER;
136 hdr->ifi_index = ifindex;
137 hdr->ifi_flags = vport_get_flags(port->vport);
140 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
141 NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
142 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
143 #ifdef IFLA_OPERSTATE
144 NLA_PUT_U8(skb, IFLA_OPERSTATE,
145 vport_is_running(port->vport)
146 ? vport_get_operstate(port->vport)
150 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
151 vport_get_addr(port->vport));
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 dp_port *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 internal_dev_port.flags = ODP_PORT_INTERNAL;
256 err = new_dp_port(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);
270 mutex_unlock(&dp_mutex);
277 err_destroy_local_port:
278 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
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)
294 struct dp_port *p, *n;
297 list_for_each_entry_safe (p, n, &dp->port_list, node)
298 if (p->port_no != ODPP_LOCAL)
299 dp_detach_port(p, 1);
303 rcu_assign_pointer(dps[dp->dp_idx], NULL);
305 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
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 static void release_dp_port(struct kobject *kobj)
339 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
343 static struct kobj_type brport_ktype = {
345 .sysfs_ops = &brport_sysfs_ops,
347 .release = release_dp_port
350 /* Called with RTNL lock and dp_mutex. */
351 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
357 vport = vport_locate(odp_port->devname);
361 if (odp_port->flags & ODP_PORT_INTERNAL)
362 vport = vport_add(odp_port->devname, "internal", NULL);
364 vport = vport_add(odp_port->devname, "netdev", NULL);
369 return PTR_ERR(vport);
372 p = kzalloc(sizeof(*p), GFP_KERNEL);
376 p->port_no = port_no;
379 atomic_set(&p->sflow_pool, 0);
381 err = vport_attach(vport, p);
387 rcu_assign_pointer(dp->ports[port_no], p);
388 list_add_rcu(&p->node, &dp->port_list);
391 /* Initialize kobject for bridge. This will be added as
392 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
394 kobject_init(&p->kobj, &brport_ktype);
396 dp_ifinfo_notify(RTM_NEWLINK, p);
401 static int attach_port(int dp_idx, struct odp_port __user *portp)
404 struct odp_port port;
409 if (copy_from_user(&port, portp, sizeof port))
411 port.devname[IFNAMSIZ - 1] = '\0';
414 dp = get_dp_locked(dp_idx);
417 goto out_unlock_rtnl;
419 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
420 if (!dp->ports[port_no])
426 err = new_dp_port(dp, &port, port_no);
430 set_internal_devs_mtu(dp);
431 dp_sysfs_add_if(dp->ports[port_no]);
433 err = put_user(port_no, &portp->port);
436 mutex_unlock(&dp->mutex);
443 int dp_detach_port(struct dp_port *p, int may_delete)
445 struct vport *vport = p->vport;
450 if (p->port_no != ODPP_LOCAL)
452 dp_ifinfo_notify(RTM_DELLINK, p);
454 /* First drop references to device. */
456 list_del_rcu(&p->node);
457 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
459 err = vport_detach(vport);
463 /* Then wait until no one is still using it, and destroy it. */
467 const char *port_type = vport_get_type(vport);
469 if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
476 kobject_put(&p->kobj);
481 static int detach_port(int dp_idx, int port_no)
488 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
492 dp = get_dp_locked(dp_idx);
495 goto out_unlock_rtnl;
497 p = dp->ports[port_no];
502 err = dp_detach_port(p, 1);
505 mutex_unlock(&dp->mutex);
512 /* Must be called with rcu_read_lock. */
513 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
515 struct datapath *dp = p->dp;
516 struct dp_stats_percpu *stats;
517 int stats_counter_off;
518 struct sw_flow_actions *acts;
519 struct loop_counter *loop;
522 OVS_CB(skb)->dp_port = p;
524 if (!OVS_CB(skb)->flow) {
525 struct odp_flow_key key;
526 struct tbl_node *flow_node;
529 /* Extract flow from 'skb' into 'key'. */
530 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
531 if (unlikely(error)) {
536 if (is_frag && dp->drop_frags) {
538 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
543 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
544 flow_hash(&key), flow_cmp);
545 if (unlikely(!flow_node)) {
546 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
547 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
551 OVS_CB(skb)->flow = flow_cast(flow_node);
554 flow_used(OVS_CB(skb)->flow, skb);
556 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
558 /* Check whether we've looped too much. */
559 loop = loop_get_counter();
560 if (unlikely(++loop->count > MAX_LOOPS))
561 loop->looping = true;
562 if (unlikely(loop->looping)) {
563 loop_suppress(dp, acts);
567 /* Execute actions. */
568 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
570 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
572 /* Check whether sub-actions looped too much. */
573 if (unlikely(loop->looping))
574 loop_suppress(dp, acts);
577 /* Decrement loop counter. */
579 loop->looping = false;
583 /* Update datapath statistics. */
585 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
587 write_seqcount_begin(&stats->seqlock);
588 (*(u64 *)((u8 *)stats + stats_counter_off))++;
589 write_seqcount_end(&stats->seqlock);
594 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
595 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
596 * can't call this function directly because it isn't exported in all
598 int vswitch_skb_checksum_setup(struct sk_buff *skb)
603 __u16 csum_start, csum_offset;
605 if (!skb->proto_csum_blank)
608 if (skb->protocol != htons(ETH_P_IP))
611 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
615 th = skb_network_header(skb) + 4 * iph->ihl;
617 csum_start = th - skb->head;
618 switch (iph->protocol) {
620 csum_offset = offsetof(struct tcphdr, check);
623 csum_offset = offsetof(struct udphdr, check);
627 pr_err("Attempting to checksum a non-TCP/UDP packet, "
628 "dropping a protocol %d packet",
633 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
636 skb->ip_summed = CHECKSUM_PARTIAL;
637 skb->proto_csum_blank = 0;
639 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
640 skb->csum_start = csum_start;
641 skb->csum_offset = csum_offset;
643 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
644 skb->csum = csum_offset;
652 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
654 /* Types of checksums that we can receive (these all refer to L4 checksums):
655 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
656 * (though not verified) checksum in packet but not in skb->csum. Packets
657 * from the bridge local port will also have this type.
658 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
659 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
660 * a valid skb->csum. Importantly, both contain a full checksum (not
661 * verified) in the packet itself. The only difference is that if the
662 * packet gets to L4 processing on this machine (not in DomU) we won't
663 * have to recompute the checksum to verify. Most hardware devices do not
664 * produce packets with this type, even if they support receive checksum
665 * offloading (they produce type #5).
666 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
667 * be computed if it is sent off box. Unfortunately on earlier kernels,
668 * this case is impossible to distinguish from #2, despite having opposite
669 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
670 * to distinguish the different states.
671 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
672 * generated locally by a Xen DomU and has a partial checksum. If it is
673 * handled on this machine (Dom0 or DomU), then the checksum will not be
674 * computed. If it goes off box, the checksum in the packet needs to be
675 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
676 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
677 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
678 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
679 * full checksum or using a protocol without a checksum. skb->csum is
680 * undefined. This is common from devices with receive checksum
681 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
682 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
684 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
685 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
686 * based on whether it is on the transmit or receive path. After the datapath
687 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
688 * checksum, we will panic. Since we can receive packets with checksums, we
689 * assume that all CHECKSUM_HW packets have checksums and map them to
690 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
691 * packet is processed by the local IP stack, in which case it will need to
692 * be reverified). If we receive a packet with CHECKSUM_HW that really means
693 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
694 * shouldn't be any devices that do this with bridging. */
695 void compute_ip_summed(struct sk_buff *skb, bool xmit)
697 /* For our convenience these defines change repeatedly between kernel
698 * versions, so we can't just copy them over... */
699 switch (skb->ip_summed) {
701 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
703 case CHECKSUM_UNNECESSARY:
704 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
707 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
708 * However, on the receive side we should only get CHECKSUM_PARTIAL
709 * packets from Xen, which uses some special fields to represent this
710 * (see below). Since we can only make one type work, pick the one
711 * that actually happens in practice.
713 * On the transmit side (basically after skb_checksum_setup()
714 * has been run or on internal dev transmit), packets with
715 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
718 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
720 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
724 case CHECKSUM_COMPLETE:
725 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
727 case CHECKSUM_PARTIAL:
728 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
732 pr_err("unknown checksum type %d\n", skb->ip_summed);
733 /* None seems the safest... */
734 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
737 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
738 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
739 * kernels. It should not be set on the transmit path though. */
740 if (skb->proto_csum_blank)
741 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
743 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
747 /* This function closely resembles skb_forward_csum() used by the bridge. It
748 * is slightly different because we are only concerned with bridging and not
749 * other types of forwarding and can get away with slightly more optimal
751 void forward_ip_summed(struct sk_buff *skb)
754 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
755 skb->ip_summed = CHECKSUM_NONE;
759 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
760 * unless we broke up a GSO packet. */
761 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
762 int queue_no, u32 arg)
764 struct sk_buff *nskb;
768 if (OVS_CB(skb)->dp_port)
769 port_no = OVS_CB(skb)->dp_port->port_no;
771 port_no = ODPP_LOCAL;
774 struct odp_msg *header;
779 err = skb_cow(skb, sizeof *header);
783 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
784 header->type = queue_no;
785 header->length = skb->len;
786 header->port = port_no;
787 header->reserved = 0;
789 skb_queue_tail(queue, skb);
797 while ((skb = nskb) != NULL) {
804 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
807 struct dp_stats_percpu *stats;
808 struct sk_buff_head *queue;
811 WARN_ON_ONCE(skb_shared(skb));
812 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
813 queue = &dp->queues[queue_no];
815 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
818 forward_ip_summed(skb);
820 err = vswitch_skb_checksum_setup(skb);
824 /* Break apart GSO packets into their component pieces. Otherwise
825 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
826 if (skb_is_gso(skb)) {
827 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
831 if (unlikely(IS_ERR(skb))) {
836 /* XXX This case might not be possible. It's hard to
837 * tell from the skb_gso_segment() code and comment. */
841 err = queue_control_packets(skb, queue, queue_no, arg);
842 wake_up_interruptible(&dp->waitqueue);
849 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
851 write_seqcount_begin(&stats->seqlock);
853 write_seqcount_end(&stats->seqlock);
860 static int flush_flows(struct datapath *dp)
862 struct tbl *old_table = rcu_dereference(dp->table);
863 struct tbl *new_table;
865 new_table = tbl_create(0);
869 rcu_assign_pointer(dp->table, new_table);
871 tbl_deferred_destroy(old_table, flow_free_tbl);
876 static int validate_actions(const struct sw_flow_actions *actions)
880 for (i = 0; i < actions->n_actions; i++) {
881 const union odp_action *a = &actions->actions[i];
884 case ODPAT_CONTROLLER:
885 case ODPAT_STRIP_VLAN:
886 case ODPAT_SET_DL_SRC:
887 case ODPAT_SET_DL_DST:
888 case ODPAT_SET_NW_SRC:
889 case ODPAT_SET_NW_DST:
890 case ODPAT_SET_TP_SRC:
891 case ODPAT_SET_TP_DST:
892 case ODPAT_SET_TUNNEL:
893 case ODPAT_SET_PRIORITY:
894 case ODPAT_POP_PRIORITY:
895 case ODPAT_DROP_SPOOFED_ARP:
896 /* No validation needed. */
900 if (a->output.port >= DP_MAX_PORTS)
904 case ODPAT_SET_DL_TCI:
905 if (a->dl_tci.tci & htons(VLAN_CFI_MASK))
909 case ODPAT_SET_NW_TOS:
910 if (a->nw_tos.nw_tos & INET_ECN_MASK)
922 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
924 struct sw_flow_actions *actions;
927 actions = flow_actions_alloc(flow->n_actions);
928 error = PTR_ERR(actions);
933 if (copy_from_user(actions->actions, flow->actions,
934 flow->n_actions * sizeof(union odp_action)))
935 goto error_free_actions;
936 error = validate_actions(actions);
938 goto error_free_actions;
945 return ERR_PTR(error);
948 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
951 struct timespec offset_ts, used, now_mono;
953 ktime_get_ts(&now_mono);
954 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
955 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
956 now_mono.tv_nsec - offset_ts.tv_nsec);
958 stats->used_sec = used.tv_sec;
959 stats->used_nsec = used.tv_nsec;
962 stats->used_nsec = 0;
965 stats->n_packets = flow->packet_count;
966 stats->n_bytes = flow->byte_count;
968 stats->tcp_flags = flow->tcp_flags;
972 static void clear_stats(struct sw_flow *flow)
976 flow->packet_count = 0;
977 flow->byte_count = 0;
980 static int expand_table(struct datapath *dp)
982 struct tbl *old_table = rcu_dereference(dp->table);
983 struct tbl *new_table;
985 new_table = tbl_expand(old_table);
986 if (IS_ERR(new_table))
987 return PTR_ERR(new_table);
989 rcu_assign_pointer(dp->table, new_table);
990 tbl_deferred_destroy(old_table, NULL);
995 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
996 struct odp_flow_stats *stats)
998 struct tbl_node *flow_node;
999 struct sw_flow *flow;
1003 table = rcu_dereference(dp->table);
1004 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
1007 struct sw_flow_actions *acts;
1010 if (!(uf->flags & ODPPF_CREATE))
1013 /* Expand table, if necessary, to make room. */
1014 if (tbl_count(table) >= tbl_n_buckets(table)) {
1015 error = expand_table(dp);
1018 table = rcu_dereference(dp->table);
1021 /* Allocate flow. */
1022 flow = flow_alloc();
1024 error = PTR_ERR(flow);
1027 flow->key = uf->flow.key;
1030 /* Obtain actions. */
1031 acts = get_actions(&uf->flow);
1032 error = PTR_ERR(acts);
1034 goto error_free_flow;
1035 rcu_assign_pointer(flow->sf_acts, acts);
1037 /* Put flow in bucket. */
1038 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1040 goto error_free_flow_acts;
1042 memset(stats, 0, sizeof(struct odp_flow_stats));
1044 /* We found a matching flow. */
1045 struct sw_flow_actions *old_acts, *new_acts;
1047 flow = flow_cast(flow_node);
1049 /* Bail out if we're not allowed to modify an existing flow. */
1051 if (!(uf->flags & ODPPF_MODIFY))
1055 new_acts = get_actions(&uf->flow);
1056 error = PTR_ERR(new_acts);
1057 if (IS_ERR(new_acts))
1059 old_acts = rcu_dereference(flow->sf_acts);
1060 if (old_acts->n_actions != new_acts->n_actions ||
1061 memcmp(old_acts->actions, new_acts->actions,
1062 sizeof(union odp_action) * old_acts->n_actions)) {
1063 rcu_assign_pointer(flow->sf_acts, new_acts);
1064 flow_deferred_free_acts(old_acts);
1069 /* Fetch stats, then clear them if necessary. */
1070 spin_lock_bh(&flow->lock);
1071 get_stats(flow, stats);
1072 if (uf->flags & ODPPF_ZERO_STATS)
1074 spin_unlock_bh(&flow->lock);
1079 error_free_flow_acts:
1080 kfree(flow->sf_acts);
1082 flow->sf_acts = NULL;
1088 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1090 struct odp_flow_stats stats;
1091 struct odp_flow_put uf;
1094 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1097 error = do_put_flow(dp, &uf, &stats);
1101 if (copy_to_user(&ufp->flow.stats, &stats,
1102 sizeof(struct odp_flow_stats)))
1108 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1109 struct odp_flow_stats __user *ustats,
1110 union odp_action __user *actions,
1111 u32 __user *n_actionsp)
1113 struct sw_flow_actions *sf_acts;
1114 struct odp_flow_stats stats;
1117 spin_lock_bh(&flow->lock);
1118 get_stats(flow, &stats);
1119 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1120 flow->tcp_flags = 0;
1122 spin_unlock_bh(&flow->lock);
1124 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1125 get_user(n_actions, n_actionsp))
1131 sf_acts = rcu_dereference(flow->sf_acts);
1132 if (put_user(sf_acts->n_actions, n_actionsp) ||
1133 (actions && copy_to_user(actions, sf_acts->actions,
1134 sizeof(union odp_action) *
1135 min(sf_acts->n_actions, n_actions))))
1141 static int answer_query(struct sw_flow *flow, u32 query_flags,
1142 struct odp_flow __user *ufp)
1144 union odp_action *actions;
1146 if (get_user(actions, &ufp->actions))
1149 return do_answer_query(flow, query_flags,
1150 &ufp->stats, actions, &ufp->n_actions);
1153 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1155 struct tbl *table = rcu_dereference(dp->table);
1156 struct tbl_node *flow_node;
1159 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1161 return ERR_PTR(-ENOENT);
1163 error = tbl_remove(table, flow_node);
1165 return ERR_PTR(error);
1167 /* XXX Returned flow_node's statistics might lose a few packets, since
1168 * other CPUs can be using this flow. We used to synchronize_rcu() to
1169 * make sure that we get completely accurate stats, but that blows our
1170 * performance, badly. */
1171 return flow_cast(flow_node);
1174 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1176 struct sw_flow *flow;
1180 if (copy_from_user(&uf, ufp, sizeof uf))
1183 flow = do_del_flow(dp, &uf.key);
1185 return PTR_ERR(flow);
1187 error = answer_query(flow, 0, ufp);
1188 flow_deferred_free(flow);
1192 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1194 struct tbl *table = rcu_dereference(dp->table);
1197 for (i = 0; i < flowvec->n_flows; i++) {
1198 struct odp_flow __user *ufp = &flowvec->flows[i];
1200 struct tbl_node *flow_node;
1203 if (copy_from_user(&uf, ufp, sizeof uf))
1206 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1208 error = put_user(ENOENT, &ufp->stats.error);
1210 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1214 return flowvec->n_flows;
1217 struct list_flows_cbdata {
1218 struct odp_flow __user *uflows;
1223 static int list_flow(struct tbl_node *node, void *cbdata_)
1225 struct sw_flow *flow = flow_cast(node);
1226 struct list_flows_cbdata *cbdata = cbdata_;
1227 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1230 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1232 error = answer_query(flow, 0, ufp);
1236 if (cbdata->listed_flows >= cbdata->n_flows)
1237 return cbdata->listed_flows;
1241 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1243 struct list_flows_cbdata cbdata;
1246 if (!flowvec->n_flows)
1249 cbdata.uflows = flowvec->flows;
1250 cbdata.n_flows = flowvec->n_flows;
1251 cbdata.listed_flows = 0;
1253 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1254 return error ? error : cbdata.listed_flows;
1257 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1258 int (*function)(struct datapath *,
1259 const struct odp_flowvec *))
1261 struct odp_flowvec __user *uflowvec;
1262 struct odp_flowvec flowvec;
1265 uflowvec = (struct odp_flowvec __user *)argp;
1266 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1269 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1272 retval = function(dp, &flowvec);
1273 return (retval < 0 ? retval
1274 : retval == flowvec.n_flows ? 0
1275 : put_user(retval, &uflowvec->n_flows));
1278 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1280 struct odp_flow_key key;
1281 struct sk_buff *skb;
1282 struct sw_flow_actions *actions;
1288 if (execute->length < ETH_HLEN || execute->length > 65535)
1291 actions = flow_actions_alloc(execute->n_actions);
1292 if (IS_ERR(actions)) {
1293 err = PTR_ERR(actions);
1298 if (copy_from_user(actions->actions, execute->actions,
1299 execute->n_actions * sizeof *execute->actions))
1300 goto error_free_actions;
1302 err = validate_actions(actions);
1304 goto error_free_actions;
1307 skb = alloc_skb(execute->length, GFP_KERNEL);
1309 goto error_free_actions;
1312 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1314 goto error_free_skb;
1316 skb_reset_mac_header(skb);
1319 /* Normally, setting the skb 'protocol' field would be handled by a
1320 * call to eth_type_trans(), but it assumes there's a sending
1321 * device, which we may not have. */
1322 if (ntohs(eth->h_proto) >= 1536)
1323 skb->protocol = eth->h_proto;
1325 skb->protocol = htons(ETH_P_802_2);
1327 err = flow_extract(skb, -1, &key, &is_frag);
1329 goto error_free_skb;
1332 err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions);
1346 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1348 struct odp_execute execute;
1350 if (copy_from_user(&execute, executep, sizeof execute))
1353 return do_execute(dp, &execute);
1356 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1358 struct tbl *table = rcu_dereference(dp->table);
1359 struct odp_stats stats;
1362 stats.n_flows = tbl_count(table);
1363 stats.cur_capacity = tbl_n_buckets(table);
1364 stats.max_capacity = TBL_MAX_BUCKETS;
1365 stats.n_ports = dp->n_ports;
1366 stats.max_ports = DP_MAX_PORTS;
1367 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1368 for_each_possible_cpu(i) {
1369 const struct dp_stats_percpu *percpu_stats;
1370 struct dp_stats_percpu local_stats;
1373 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1376 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1377 local_stats = *percpu_stats;
1378 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1380 stats.n_frags += local_stats.n_frags;
1381 stats.n_hit += local_stats.n_hit;
1382 stats.n_missed += local_stats.n_missed;
1383 stats.n_lost += local_stats.n_lost;
1385 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1386 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1387 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1390 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1391 int dp_min_mtu(const struct datapath *dp)
1398 list_for_each_entry_rcu (p, &dp->port_list, node) {
1401 /* Skip any internal ports, since that's what we're trying to
1403 if (is_internal_vport(p->vport))
1406 dev_mtu = vport_get_mtu(p->vport);
1407 if (!mtu || dev_mtu < mtu)
1411 return mtu ? mtu : ETH_DATA_LEN;
1414 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1415 * be called with RTNL lock. */
1416 void set_internal_devs_mtu(const struct datapath *dp)
1423 mtu = dp_min_mtu(dp);
1425 list_for_each_entry_rcu (p, &dp->port_list, node) {
1426 if (is_internal_vport(p->vport))
1427 vport_set_mtu(p->vport, mtu);
1431 static int put_port(const struct dp_port *p, struct odp_port __user *uop)
1435 memset(&op, 0, sizeof op);
1438 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1441 op.port = p->port_no;
1442 op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
1444 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1447 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1449 struct odp_port port;
1451 if (copy_from_user(&port, uport, sizeof port))
1454 if (port.devname[0]) {
1455 struct vport *vport;
1456 struct dp_port *dp_port;
1459 port.devname[IFNAMSIZ - 1] = '\0';
1464 vport = vport_locate(port.devname);
1470 dp_port = vport_get_dp_port(vport);
1471 if (!dp_port || dp_port->dp != dp) {
1476 port.port = dp_port->port_no;
1485 if (port.port >= DP_MAX_PORTS)
1487 if (!dp->ports[port.port])
1491 return put_port(dp->ports[port.port], uport);
1494 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1501 list_for_each_entry_rcu (p, &dp->port_list, node) {
1502 if (put_port(p, &uports[idx]))
1504 if (idx++ >= n_ports)
1511 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1513 struct odp_portvec pv;
1516 if (copy_from_user(&pv, upv, sizeof pv))
1519 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1523 return put_user(retval, &upv->n_ports);
1526 static int get_listen_mask(const struct file *f)
1528 return (long)f->private_data;
1531 static void set_listen_mask(struct file *f, int listen_mask)
1533 f->private_data = (void*)(long)listen_mask;
1536 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1539 int dp_idx = iminor(f->f_dentry->d_inode);
1540 struct datapath *dp;
1541 int drop_frags, listeners, port_no;
1542 unsigned int sflow_probability;
1545 /* Handle commands with special locking requirements up front. */
1548 err = create_dp(dp_idx, (char __user *)argp);
1551 case ODP_DP_DESTROY:
1552 err = destroy_dp(dp_idx);
1555 case ODP_PORT_ATTACH:
1556 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1559 case ODP_PORT_DETACH:
1560 err = get_user(port_no, (int __user *)argp);
1562 err = detach_port(dp_idx, port_no);
1566 err = vport_user_add((struct odp_vport_add __user *)argp);
1570 err = vport_user_mod((struct odp_vport_mod __user *)argp);
1574 err = vport_user_del((char __user *)argp);
1577 case ODP_VPORT_STATS_GET:
1578 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1581 case ODP_VPORT_STATS_SET:
1582 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1585 case ODP_VPORT_ETHER_GET:
1586 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1589 case ODP_VPORT_ETHER_SET:
1590 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1593 case ODP_VPORT_MTU_GET:
1594 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1597 case ODP_VPORT_MTU_SET:
1598 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1602 dp = get_dp_locked(dp_idx);
1609 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1612 case ODP_GET_DROP_FRAGS:
1613 err = put_user(dp->drop_frags, (int __user *)argp);
1616 case ODP_SET_DROP_FRAGS:
1617 err = get_user(drop_frags, (int __user *)argp);
1621 if (drop_frags != 0 && drop_frags != 1)
1623 dp->drop_frags = drop_frags;
1627 case ODP_GET_LISTEN_MASK:
1628 err = put_user(get_listen_mask(f), (int __user *)argp);
1631 case ODP_SET_LISTEN_MASK:
1632 err = get_user(listeners, (int __user *)argp);
1636 if (listeners & ~ODPL_ALL)
1639 set_listen_mask(f, listeners);
1642 case ODP_GET_SFLOW_PROBABILITY:
1643 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1646 case ODP_SET_SFLOW_PROBABILITY:
1647 err = get_user(sflow_probability, (unsigned int __user *)argp);
1649 dp->sflow_probability = sflow_probability;
1652 case ODP_PORT_QUERY:
1653 err = query_port(dp, (struct odp_port __user *)argp);
1657 err = list_ports(dp, (struct odp_portvec __user *)argp);
1660 case ODP_FLOW_FLUSH:
1661 err = flush_flows(dp);
1665 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1669 err = del_flow(dp, (struct odp_flow __user *)argp);
1673 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1677 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1681 err = execute_packet(dp, (struct odp_execute __user *)argp);
1688 mutex_unlock(&dp->mutex);
1693 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1696 for (i = 0; i < DP_N_QUEUES; i++) {
1697 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1703 #ifdef CONFIG_COMPAT
1704 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1706 struct compat_odp_portvec pv;
1709 if (copy_from_user(&pv, upv, sizeof pv))
1712 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1716 return put_user(retval, &upv->n_ports);
1719 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1721 compat_uptr_t actions;
1723 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1724 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1725 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1726 __get_user(actions, &compat->actions) ||
1727 __get_user(flow->n_actions, &compat->n_actions) ||
1728 __get_user(flow->flags, &compat->flags))
1731 flow->actions = compat_ptr(actions);
1735 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1737 struct odp_flow_stats stats;
1738 struct odp_flow_put fp;
1741 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1742 get_user(fp.flags, &ufp->flags))
1745 error = do_put_flow(dp, &fp, &stats);
1749 if (copy_to_user(&ufp->flow.stats, &stats,
1750 sizeof(struct odp_flow_stats)))
1756 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1757 struct compat_odp_flow __user *ufp)
1759 compat_uptr_t actions;
1761 if (get_user(actions, &ufp->actions))
1764 return do_answer_query(flow, query_flags, &ufp->stats,
1765 compat_ptr(actions), &ufp->n_actions);
1768 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1770 struct sw_flow *flow;
1774 if (compat_get_flow(&uf, ufp))
1777 flow = do_del_flow(dp, &uf.key);
1779 return PTR_ERR(flow);
1781 error = compat_answer_query(flow, 0, ufp);
1782 flow_deferred_free(flow);
1786 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1788 struct tbl *table = rcu_dereference(dp->table);
1791 for (i = 0; i < n_flows; i++) {
1792 struct compat_odp_flow __user *ufp = &flows[i];
1794 struct tbl_node *flow_node;
1797 if (compat_get_flow(&uf, ufp))
1800 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1802 error = put_user(ENOENT, &ufp->stats.error);
1804 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1811 struct compat_list_flows_cbdata {
1812 struct compat_odp_flow __user *uflows;
1817 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1819 struct sw_flow *flow = flow_cast(node);
1820 struct compat_list_flows_cbdata *cbdata = cbdata_;
1821 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1824 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1826 error = compat_answer_query(flow, 0, ufp);
1830 if (cbdata->listed_flows >= cbdata->n_flows)
1831 return cbdata->listed_flows;
1835 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1837 struct compat_list_flows_cbdata cbdata;
1843 cbdata.uflows = flows;
1844 cbdata.n_flows = n_flows;
1845 cbdata.listed_flows = 0;
1847 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1848 return error ? error : cbdata.listed_flows;
1851 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1852 int (*function)(struct datapath *,
1853 struct compat_odp_flow *,
1856 struct compat_odp_flowvec __user *uflowvec;
1857 struct compat_odp_flow __user *flows;
1858 struct compat_odp_flowvec flowvec;
1861 uflowvec = compat_ptr(argp);
1862 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1863 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1866 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1869 flows = compat_ptr(flowvec.flows);
1870 if (!access_ok(VERIFY_WRITE, flows,
1871 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1874 retval = function(dp, flows, flowvec.n_flows);
1875 return (retval < 0 ? retval
1876 : retval == flowvec.n_flows ? 0
1877 : put_user(retval, &uflowvec->n_flows));
1880 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1882 struct odp_execute execute;
1883 compat_uptr_t actions;
1886 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1887 __get_user(actions, &uexecute->actions) ||
1888 __get_user(execute.n_actions, &uexecute->n_actions) ||
1889 __get_user(data, &uexecute->data) ||
1890 __get_user(execute.length, &uexecute->length))
1893 execute.actions = compat_ptr(actions);
1894 execute.data = compat_ptr(data);
1896 return do_execute(dp, &execute);
1899 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1901 int dp_idx = iminor(f->f_dentry->d_inode);
1902 struct datapath *dp;
1906 case ODP_DP_DESTROY:
1907 case ODP_FLOW_FLUSH:
1908 /* Ioctls that don't need any translation at all. */
1909 return openvswitch_ioctl(f, cmd, argp);
1912 case ODP_PORT_ATTACH:
1913 case ODP_PORT_DETACH:
1915 case ODP_VPORT_MTU_SET:
1916 case ODP_VPORT_MTU_GET:
1917 case ODP_VPORT_ETHER_SET:
1918 case ODP_VPORT_ETHER_GET:
1919 case ODP_VPORT_STATS_SET:
1920 case ODP_VPORT_STATS_GET:
1922 case ODP_GET_DROP_FRAGS:
1923 case ODP_SET_DROP_FRAGS:
1924 case ODP_SET_LISTEN_MASK:
1925 case ODP_GET_LISTEN_MASK:
1926 case ODP_SET_SFLOW_PROBABILITY:
1927 case ODP_GET_SFLOW_PROBABILITY:
1928 case ODP_PORT_QUERY:
1929 /* Ioctls that just need their pointer argument extended. */
1930 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1932 case ODP_VPORT_ADD32:
1933 return compat_vport_user_add(compat_ptr(argp));
1935 case ODP_VPORT_MOD32:
1936 return compat_vport_user_mod(compat_ptr(argp));
1939 dp = get_dp_locked(dp_idx);
1945 case ODP_PORT_LIST32:
1946 err = compat_list_ports(dp, compat_ptr(argp));
1949 case ODP_FLOW_PUT32:
1950 err = compat_put_flow(dp, compat_ptr(argp));
1953 case ODP_FLOW_DEL32:
1954 err = compat_del_flow(dp, compat_ptr(argp));
1957 case ODP_FLOW_GET32:
1958 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1961 case ODP_FLOW_LIST32:
1962 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1966 err = compat_execute(dp, compat_ptr(argp));
1973 mutex_unlock(&dp->mutex);
1979 /* Unfortunately this function is not exported so this is a verbatim copy
1980 * from net/core/datagram.c in 2.6.30. */
1981 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1982 u8 __user *to, int len,
1985 int start = skb_headlen(skb);
1987 int i, copy = start - offset;
1994 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1998 if ((len -= copy) == 0)
2005 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2008 WARN_ON(start > offset + len);
2010 end = start + skb_shinfo(skb)->frags[i].size;
2011 if ((copy = end - offset) > 0) {
2015 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2016 struct page *page = frag->page;
2021 csum2 = csum_and_copy_to_user(vaddr +
2028 *csump = csum_block_add(*csump, csum2, pos);
2038 if (skb_shinfo(skb)->frag_list) {
2039 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2041 for (; list; list=list->next) {
2044 WARN_ON(start > offset + len);
2046 end = start + list->len;
2047 if ((copy = end - offset) > 0) {
2051 if (skb_copy_and_csum_datagram(list,
2056 *csump = csum_block_add(*csump, csum2, pos);
2057 if ((len -= copy) == 0)
2073 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2076 /* XXX is there sufficient synchronization here? */
2077 int listeners = get_listen_mask(f);
2078 int dp_idx = iminor(f->f_dentry->d_inode);
2079 struct datapath *dp = get_dp(dp_idx);
2080 struct sk_buff *skb;
2081 size_t copy_bytes, tot_copy_bytes;
2087 if (nbytes == 0 || !listeners)
2093 for (i = 0; i < DP_N_QUEUES; i++) {
2094 if (listeners & (1 << i)) {
2095 skb = skb_dequeue(&dp->queues[i]);
2101 if (f->f_flags & O_NONBLOCK) {
2106 wait_event_interruptible(dp->waitqueue,
2107 dp_has_packet_of_interest(dp,
2110 if (signal_pending(current)) {
2111 retval = -ERESTARTSYS;
2116 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2119 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2120 if (copy_bytes == skb->len) {
2122 unsigned int csum_start, csum_offset;
2124 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2125 csum_start = skb->csum_start - skb_headroom(skb);
2126 csum_offset = skb->csum_offset;
2128 csum_start = skb_transport_header(skb) - skb->data;
2129 csum_offset = skb->csum;
2131 BUG_ON(csum_start >= skb_headlen(skb));
2132 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2133 copy_bytes - csum_start, &csum);
2135 __sum16 __user *csump;
2137 copy_bytes = csum_start;
2138 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2140 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
2141 put_user(csum_fold(csum), csump);
2144 retval = skb_checksum_help(skb);
2148 struct iovec __user iov;
2151 iov.iov_len = copy_bytes;
2152 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2156 retval = tot_copy_bytes;
2164 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2166 /* XXX is there sufficient synchronization here? */
2167 int dp_idx = iminor(file->f_dentry->d_inode);
2168 struct datapath *dp = get_dp(dp_idx);
2173 poll_wait(file, &dp->waitqueue, wait);
2174 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2175 mask |= POLLIN | POLLRDNORM;
2177 mask = POLLIN | POLLRDNORM | POLLHUP;
2182 struct file_operations openvswitch_fops = {
2183 /* XXX .aio_read = openvswitch_aio_read, */
2184 .read = openvswitch_read,
2185 .poll = openvswitch_poll,
2186 .unlocked_ioctl = openvswitch_ioctl,
2187 #ifdef CONFIG_COMPAT
2188 .compat_ioctl = openvswitch_compat_ioctl,
2190 /* XXX .fasync = openvswitch_fasync, */
2195 static int __init dp_init(void)
2197 struct sk_buff *dummy_skb;
2200 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2202 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2210 goto error_flow_exit;
2212 err = register_netdevice_notifier(&dp_device_notifier);
2214 goto error_vport_exit;
2216 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2218 goto error_unreg_notifier;
2222 error_unreg_notifier:
2223 unregister_netdevice_notifier(&dp_device_notifier);
2232 static void dp_cleanup(void)
2235 unregister_chrdev(major, "openvswitch");
2236 unregister_netdevice_notifier(&dp_device_notifier);
2241 module_init(dp_init);
2242 module_exit(dp_cleanup);
2244 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2245 MODULE_LICENSE("GPL");