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"
58 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
59 EXPORT_SYMBOL(dp_ioctl_hook);
61 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
64 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
67 * It is safe to access the datapath and dp_port structures with just
70 static struct datapath *dps[ODP_MAX];
71 static DEFINE_MUTEX(dp_mutex);
73 static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
75 /* Must be called with rcu_read_lock or dp_mutex. */
76 struct datapath *get_dp(int dp_idx)
78 if (dp_idx < 0 || dp_idx >= ODP_MAX)
80 return rcu_dereference(dps[dp_idx]);
82 EXPORT_SYMBOL_GPL(get_dp);
84 static struct datapath *get_dp_locked(int dp_idx)
88 mutex_lock(&dp_mutex);
91 mutex_lock(&dp->mutex);
92 mutex_unlock(&dp_mutex);
96 /* Must be called with rcu_read_lock or RTNL lock. */
97 const char *dp_name(const struct datapath *dp)
99 return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
102 static inline size_t br_nlmsg_size(void)
104 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
105 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
106 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
107 + nla_total_size(4) /* IFLA_MASTER */
108 + nla_total_size(4) /* IFLA_MTU */
109 + nla_total_size(4) /* IFLA_LINK */
110 + nla_total_size(1); /* IFLA_OPERSTATE */
113 static int dp_fill_ifinfo(struct sk_buff *skb,
114 const struct dp_port *port,
115 int event, unsigned int flags)
117 const struct datapath *dp = port->dp;
118 int ifindex = vport_get_ifindex(port->vport);
119 int iflink = vport_get_iflink(port->vport);
120 struct ifinfomsg *hdr;
121 struct nlmsghdr *nlh;
129 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
133 hdr = nlmsg_data(nlh);
134 hdr->ifi_family = AF_BRIDGE;
136 hdr->ifi_type = ARPHRD_ETHER;
137 hdr->ifi_index = ifindex;
138 hdr->ifi_flags = vport_get_flags(port->vport);
141 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
142 NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
143 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
144 #ifdef IFLA_OPERSTATE
145 NLA_PUT_U8(skb, IFLA_OPERSTATE,
146 vport_is_running(port->vport)
147 ? vport_get_operstate(port->vport)
151 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
152 vport_get_addr(port->vport));
154 if (ifindex != iflink)
155 NLA_PUT_U32(skb, IFLA_LINK,iflink);
157 return nlmsg_end(skb, nlh);
160 nlmsg_cancel(skb, nlh);
164 static void dp_ifinfo_notify(int event, struct dp_port *port)
169 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
173 err = dp_fill_ifinfo(skb, port, event, 0);
175 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
176 WARN_ON(err == -EMSGSIZE);
180 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
184 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
187 static void release_dp(struct kobject *kobj)
189 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
193 static struct kobj_type dp_ktype = {
194 .release = release_dp
197 static int create_dp(int dp_idx, const char __user *devnamep)
199 struct odp_port internal_dev_port;
200 char devname[IFNAMSIZ];
206 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
210 } else if (retval >= IFNAMSIZ) {
215 snprintf(devname, sizeof devname, "of%d", dp_idx);
219 mutex_lock(&dp_mutex);
221 if (!try_module_get(THIS_MODULE))
224 /* Exit early if a datapath with that number already exists.
225 * (We don't use -EEXIST because that's ambiguous with 'devname'
226 * conflicting with an existing network device name.) */
232 dp = kzalloc(sizeof *dp, GFP_KERNEL);
235 INIT_LIST_HEAD(&dp->port_list);
236 mutex_init(&dp->mutex);
238 for (i = 0; i < DP_N_QUEUES; i++)
239 skb_queue_head_init(&dp->queues[i]);
240 init_waitqueue_head(&dp->waitqueue);
242 /* Initialize kobject for bridge. This will be added as
243 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
244 dp->ifobj.kset = NULL;
245 kobject_init(&dp->ifobj, &dp_ktype);
247 /* Allocate table. */
249 rcu_assign_pointer(dp->table, tbl_create(0));
253 /* Set up our datapath device. */
254 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
255 strcpy(internal_dev_port.devname, devname);
256 internal_dev_port.flags = ODP_PORT_INTERNAL;
257 err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
262 goto err_destroy_table;
266 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
267 if (!dp->stats_percpu)
268 goto err_destroy_local_port;
270 rcu_assign_pointer(dps[dp_idx], dp);
271 mutex_unlock(&dp_mutex);
278 err_destroy_local_port:
279 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
281 tbl_destroy(dp->table, NULL);
285 module_put(THIS_MODULE);
287 mutex_unlock(&dp_mutex);
293 static void do_destroy_dp(struct datapath *dp)
295 struct dp_port *p, *n;
298 list_for_each_entry_safe (p, n, &dp->port_list, node)
299 if (p->port_no != ODPP_LOCAL)
300 dp_detach_port(p, 1);
304 rcu_assign_pointer(dps[dp->dp_idx], NULL);
306 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
308 tbl_destroy(dp->table, flow_free_tbl);
310 for (i = 0; i < DP_N_QUEUES; i++)
311 skb_queue_purge(&dp->queues[i]);
312 free_percpu(dp->stats_percpu);
313 kobject_put(&dp->ifobj);
314 module_put(THIS_MODULE);
317 static int destroy_dp(int dp_idx)
323 mutex_lock(&dp_mutex);
333 mutex_unlock(&dp_mutex);
338 static void release_dp_port(struct kobject *kobj)
340 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
344 static struct kobj_type brport_ktype = {
346 .sysfs_ops = &brport_sysfs_ops,
348 .release = release_dp_port
351 /* Called with RTNL lock and dp_mutex. */
352 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
358 vport = vport_locate(odp_port->devname);
362 if (odp_port->flags & ODP_PORT_INTERNAL)
363 vport = vport_add(odp_port->devname, "internal", NULL);
365 vport = vport_add(odp_port->devname, "netdev", NULL);
370 return PTR_ERR(vport);
373 p = kzalloc(sizeof(*p), GFP_KERNEL);
377 p->port_no = port_no;
380 atomic_set(&p->sflow_pool, 0);
382 err = vport_attach(vport, p);
388 rcu_assign_pointer(dp->ports[port_no], p);
389 list_add_rcu(&p->node, &dp->port_list);
392 /* Initialize kobject for bridge. This will be added as
393 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
395 kobject_init(&p->kobj, &brport_ktype);
397 dp_ifinfo_notify(RTM_NEWLINK, p);
402 static int attach_port(int dp_idx, struct odp_port __user *portp)
405 struct odp_port port;
410 if (copy_from_user(&port, portp, sizeof port))
412 port.devname[IFNAMSIZ - 1] = '\0';
415 dp = get_dp_locked(dp_idx);
418 goto out_unlock_rtnl;
420 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
421 if (!dp->ports[port_no])
427 err = new_dp_port(dp, &port, port_no);
431 set_internal_devs_mtu(dp);
432 dp_sysfs_add_if(dp->ports[port_no]);
434 err = put_user(port_no, &portp->port);
437 mutex_unlock(&dp->mutex);
444 int dp_detach_port(struct dp_port *p, int may_delete)
446 struct vport *vport = p->vport;
451 if (p->port_no != ODPP_LOCAL)
453 dp_ifinfo_notify(RTM_DELLINK, p);
455 /* First drop references to device. */
457 list_del_rcu(&p->node);
458 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
460 err = vport_detach(vport);
464 /* Then wait until no one is still using it, and destroy it. */
468 const char *port_type = vport_get_type(vport);
470 if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
477 kobject_put(&p->kobj);
482 static int detach_port(int dp_idx, int port_no)
489 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
493 dp = get_dp_locked(dp_idx);
496 goto out_unlock_rtnl;
498 p = dp->ports[port_no];
503 err = dp_detach_port(p, 1);
506 mutex_unlock(&dp->mutex);
513 /* Must be called with rcu_read_lock. */
514 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
516 struct datapath *dp = p->dp;
517 struct dp_stats_percpu *stats;
518 int stats_counter_off;
519 struct sw_flow_actions *acts;
520 struct loop_counter *loop;
523 OVS_CB(skb)->dp_port = p;
525 if (!OVS_CB(skb)->flow) {
526 struct odp_flow_key key;
527 struct tbl_node *flow_node;
530 /* Extract flow from 'skb' into 'key'. */
531 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
532 if (unlikely(error)) {
537 if (is_frag && dp->drop_frags) {
539 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
544 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
545 flow_hash(&key), flow_cmp);
546 if (unlikely(!flow_node)) {
547 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
548 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
552 OVS_CB(skb)->flow = flow_cast(flow_node);
555 flow_used(OVS_CB(skb)->flow, skb);
557 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
559 /* Check whether we've looped too much. */
560 loop = loop_get_counter();
561 if (unlikely(++loop->count > MAX_LOOPS))
562 loop->looping = true;
563 if (unlikely(loop->looping)) {
564 loop_suppress(dp, acts);
568 /* Execute actions. */
569 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
571 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
573 /* Check whether sub-actions looped too much. */
574 if (unlikely(loop->looping))
575 loop_suppress(dp, acts);
578 /* Decrement loop counter. */
580 loop->looping = false;
584 /* Update datapath statistics. */
586 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
588 write_seqcount_begin(&stats->seqlock);
589 (*(u64 *)((u8 *)stats + stats_counter_off))++;
590 write_seqcount_end(&stats->seqlock);
595 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
596 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
597 * can't call this function directly because it isn't exported in all
599 int vswitch_skb_checksum_setup(struct sk_buff *skb)
604 __u16 csum_start, csum_offset;
606 if (!skb->proto_csum_blank)
609 if (skb->protocol != htons(ETH_P_IP))
612 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
616 th = skb_network_header(skb) + 4 * iph->ihl;
618 csum_start = th - skb->head;
619 switch (iph->protocol) {
621 csum_offset = offsetof(struct tcphdr, check);
624 csum_offset = offsetof(struct udphdr, check);
628 pr_err("Attempting to checksum a non-TCP/UDP packet, "
629 "dropping a protocol %d packet",
634 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
637 skb->ip_summed = CHECKSUM_PARTIAL;
638 skb->proto_csum_blank = 0;
640 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
641 skb->csum_start = csum_start;
642 skb->csum_offset = csum_offset;
644 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
645 skb->csum = csum_offset;
653 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
655 /* Types of checksums that we can receive (these all refer to L4 checksums):
656 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
657 * (though not verified) checksum in packet but not in skb->csum. Packets
658 * from the bridge local port will also have this type.
659 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
660 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
661 * a valid skb->csum. Importantly, both contain a full checksum (not
662 * verified) in the packet itself. The only difference is that if the
663 * packet gets to L4 processing on this machine (not in DomU) we won't
664 * have to recompute the checksum to verify. Most hardware devices do not
665 * produce packets with this type, even if they support receive checksum
666 * offloading (they produce type #5).
667 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
668 * be computed if it is sent off box. Unfortunately on earlier kernels,
669 * this case is impossible to distinguish from #2, despite having opposite
670 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
671 * to distinguish the different states.
672 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
673 * generated locally by a Xen DomU and has a partial checksum. If it is
674 * handled on this machine (Dom0 or DomU), then the checksum will not be
675 * computed. If it goes off box, the checksum in the packet needs to be
676 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
677 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
678 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
679 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
680 * full checksum or using a protocol without a checksum. skb->csum is
681 * undefined. This is common from devices with receive checksum
682 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
683 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
685 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
686 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
687 * based on whether it is on the transmit or receive path. After the datapath
688 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
689 * checksum, we will panic. Since we can receive packets with checksums, we
690 * assume that all CHECKSUM_HW packets have checksums and map them to
691 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
692 * packet is processed by the local IP stack, in which case it will need to
693 * be reverified). If we receive a packet with CHECKSUM_HW that really means
694 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
695 * shouldn't be any devices that do this with bridging. */
696 void compute_ip_summed(struct sk_buff *skb, bool xmit)
698 /* For our convenience these defines change repeatedly between kernel
699 * versions, so we can't just copy them over... */
700 switch (skb->ip_summed) {
702 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
704 case CHECKSUM_UNNECESSARY:
705 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
708 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
709 * However, on the receive side we should only get CHECKSUM_PARTIAL
710 * packets from Xen, which uses some special fields to represent this
711 * (see below). Since we can only make one type work, pick the one
712 * that actually happens in practice.
714 * On the transmit side (basically after skb_checksum_setup()
715 * has been run or on internal dev transmit), packets with
716 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
719 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
721 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
725 case CHECKSUM_COMPLETE:
726 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
728 case CHECKSUM_PARTIAL:
729 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
733 pr_err("unknown checksum type %d\n", skb->ip_summed);
734 /* None seems the safest... */
735 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
738 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
739 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
740 * kernels. It should not be set on the transmit path though. */
741 if (skb->proto_csum_blank)
742 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
744 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
748 /* This function closely resembles skb_forward_csum() used by the bridge. It
749 * is slightly different because we are only concerned with bridging and not
750 * other types of forwarding and can get away with slightly more optimal
752 void forward_ip_summed(struct sk_buff *skb)
755 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
756 skb->ip_summed = CHECKSUM_NONE;
760 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
761 * unless we broke up a GSO packet. */
762 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
763 int queue_no, u32 arg)
765 struct sk_buff *nskb;
769 if (OVS_CB(skb)->dp_port)
770 port_no = OVS_CB(skb)->dp_port->port_no;
772 port_no = ODPP_LOCAL;
775 struct odp_msg *header;
780 err = skb_cow(skb, sizeof *header);
784 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
785 header->type = queue_no;
786 header->length = skb->len;
787 header->port = port_no;
788 header->reserved = 0;
790 skb_queue_tail(queue, skb);
798 while ((skb = nskb) != NULL) {
805 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
808 struct dp_stats_percpu *stats;
809 struct sk_buff_head *queue;
812 WARN_ON_ONCE(skb_shared(skb));
813 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
814 queue = &dp->queues[queue_no];
816 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
819 forward_ip_summed(skb);
821 err = vswitch_skb_checksum_setup(skb);
825 /* Break apart GSO packets into their component pieces. Otherwise
826 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
827 if (skb_is_gso(skb)) {
828 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
832 if (unlikely(IS_ERR(skb))) {
837 /* XXX This case might not be possible. It's hard to
838 * tell from the skb_gso_segment() code and comment. */
842 err = queue_control_packets(skb, queue, queue_no, arg);
843 wake_up_interruptible(&dp->waitqueue);
850 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
852 write_seqcount_begin(&stats->seqlock);
854 write_seqcount_end(&stats->seqlock);
861 static int flush_flows(struct datapath *dp)
863 struct tbl *old_table = rcu_dereference(dp->table);
864 struct tbl *new_table;
866 new_table = tbl_create(0);
870 rcu_assign_pointer(dp->table, new_table);
872 tbl_deferred_destroy(old_table, flow_free_tbl);
877 static int validate_actions(const struct sw_flow_actions *actions)
881 for (i = 0; i < actions->n_actions; i++) {
882 const union odp_action *a = &actions->actions[i];
885 case ODPAT_CONTROLLER:
886 case ODPAT_STRIP_VLAN:
887 case ODPAT_SET_DL_SRC:
888 case ODPAT_SET_DL_DST:
889 case ODPAT_SET_NW_SRC:
890 case ODPAT_SET_NW_DST:
891 case ODPAT_SET_TP_SRC:
892 case ODPAT_SET_TP_DST:
893 case ODPAT_SET_TUNNEL:
894 case ODPAT_SET_PRIORITY:
895 case ODPAT_POP_PRIORITY:
896 case ODPAT_DROP_SPOOFED_ARP:
897 /* No validation needed. */
901 if (a->output.port >= DP_MAX_PORTS)
905 case ODPAT_SET_DL_TCI:
906 if (a->dl_tci.tci & htons(VLAN_CFI_MASK))
910 case ODPAT_SET_NW_TOS:
911 if (a->nw_tos.nw_tos & INET_ECN_MASK)
923 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
925 struct sw_flow_actions *actions;
928 actions = flow_actions_alloc(flow->n_actions);
929 error = PTR_ERR(actions);
934 if (copy_from_user(actions->actions, flow->actions,
935 flow->n_actions * sizeof(union odp_action)))
936 goto error_free_actions;
937 error = validate_actions(actions);
939 goto error_free_actions;
946 return ERR_PTR(error);
949 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
952 struct timespec offset_ts, used, now_mono;
954 ktime_get_ts(&now_mono);
955 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
956 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
957 now_mono.tv_nsec - offset_ts.tv_nsec);
959 stats->used_sec = used.tv_sec;
960 stats->used_nsec = used.tv_nsec;
963 stats->used_nsec = 0;
966 stats->n_packets = flow->packet_count;
967 stats->n_bytes = flow->byte_count;
969 stats->tcp_flags = flow->tcp_flags;
973 static void clear_stats(struct sw_flow *flow)
977 flow->packet_count = 0;
978 flow->byte_count = 0;
981 static int expand_table(struct datapath *dp)
983 struct tbl *old_table = rcu_dereference(dp->table);
984 struct tbl *new_table;
986 new_table = tbl_expand(old_table);
987 if (IS_ERR(new_table))
988 return PTR_ERR(new_table);
990 rcu_assign_pointer(dp->table, new_table);
991 tbl_deferred_destroy(old_table, NULL);
996 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
997 struct odp_flow_stats *stats)
999 struct tbl_node *flow_node;
1000 struct sw_flow *flow;
1004 table = rcu_dereference(dp->table);
1005 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
1008 struct sw_flow_actions *acts;
1011 if (!(uf->flags & ODPPF_CREATE))
1014 /* Expand table, if necessary, to make room. */
1015 if (tbl_count(table) >= tbl_n_buckets(table)) {
1016 error = expand_table(dp);
1019 table = rcu_dereference(dp->table);
1022 /* Allocate flow. */
1023 flow = flow_alloc();
1025 error = PTR_ERR(flow);
1028 flow->key = uf->flow.key;
1031 /* Obtain actions. */
1032 acts = get_actions(&uf->flow);
1033 error = PTR_ERR(acts);
1035 goto error_free_flow;
1036 rcu_assign_pointer(flow->sf_acts, acts);
1038 /* Put flow in bucket. */
1039 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1041 goto error_free_flow_acts;
1043 memset(stats, 0, sizeof(struct odp_flow_stats));
1045 /* We found a matching flow. */
1046 struct sw_flow_actions *old_acts, *new_acts;
1048 flow = flow_cast(flow_node);
1050 /* Bail out if we're not allowed to modify an existing flow. */
1052 if (!(uf->flags & ODPPF_MODIFY))
1056 new_acts = get_actions(&uf->flow);
1057 error = PTR_ERR(new_acts);
1058 if (IS_ERR(new_acts))
1060 old_acts = rcu_dereference(flow->sf_acts);
1061 if (old_acts->n_actions != new_acts->n_actions ||
1062 memcmp(old_acts->actions, new_acts->actions,
1063 sizeof(union odp_action) * old_acts->n_actions)) {
1064 rcu_assign_pointer(flow->sf_acts, new_acts);
1065 flow_deferred_free_acts(old_acts);
1070 /* Fetch stats, then clear them if necessary. */
1071 spin_lock_bh(&flow->lock);
1072 get_stats(flow, stats);
1073 if (uf->flags & ODPPF_ZERO_STATS)
1075 spin_unlock_bh(&flow->lock);
1080 error_free_flow_acts:
1081 kfree(flow->sf_acts);
1083 flow->sf_acts = NULL;
1089 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1091 struct odp_flow_stats stats;
1092 struct odp_flow_put uf;
1095 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1098 error = do_put_flow(dp, &uf, &stats);
1102 if (copy_to_user(&ufp->flow.stats, &stats,
1103 sizeof(struct odp_flow_stats)))
1109 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1110 struct odp_flow_stats __user *ustats,
1111 union odp_action __user *actions,
1112 u32 __user *n_actionsp)
1114 struct sw_flow_actions *sf_acts;
1115 struct odp_flow_stats stats;
1118 spin_lock_bh(&flow->lock);
1119 get_stats(flow, &stats);
1120 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1121 flow->tcp_flags = 0;
1123 spin_unlock_bh(&flow->lock);
1125 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1126 get_user(n_actions, n_actionsp))
1132 sf_acts = rcu_dereference(flow->sf_acts);
1133 if (put_user(sf_acts->n_actions, n_actionsp) ||
1134 (actions && copy_to_user(actions, sf_acts->actions,
1135 sizeof(union odp_action) *
1136 min(sf_acts->n_actions, n_actions))))
1142 static int answer_query(struct sw_flow *flow, u32 query_flags,
1143 struct odp_flow __user *ufp)
1145 union odp_action *actions;
1147 if (get_user(actions, &ufp->actions))
1150 return do_answer_query(flow, query_flags,
1151 &ufp->stats, actions, &ufp->n_actions);
1154 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1156 struct tbl *table = rcu_dereference(dp->table);
1157 struct tbl_node *flow_node;
1160 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1162 return ERR_PTR(-ENOENT);
1164 error = tbl_remove(table, flow_node);
1166 return ERR_PTR(error);
1168 /* XXX Returned flow_node's statistics might lose a few packets, since
1169 * other CPUs can be using this flow. We used to synchronize_rcu() to
1170 * make sure that we get completely accurate stats, but that blows our
1171 * performance, badly. */
1172 return flow_cast(flow_node);
1175 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1177 struct sw_flow *flow;
1181 if (copy_from_user(&uf, ufp, sizeof uf))
1184 flow = do_del_flow(dp, &uf.key);
1186 return PTR_ERR(flow);
1188 error = answer_query(flow, 0, ufp);
1189 flow_deferred_free(flow);
1193 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1195 struct tbl *table = rcu_dereference(dp->table);
1198 for (i = 0; i < flowvec->n_flows; i++) {
1199 struct odp_flow __user *ufp = &flowvec->flows[i];
1201 struct tbl_node *flow_node;
1204 if (copy_from_user(&uf, ufp, sizeof uf))
1207 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1209 error = put_user(ENOENT, &ufp->stats.error);
1211 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1215 return flowvec->n_flows;
1218 struct list_flows_cbdata {
1219 struct odp_flow __user *uflows;
1224 static int list_flow(struct tbl_node *node, void *cbdata_)
1226 struct sw_flow *flow = flow_cast(node);
1227 struct list_flows_cbdata *cbdata = cbdata_;
1228 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1231 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1233 error = answer_query(flow, 0, ufp);
1237 if (cbdata->listed_flows >= cbdata->n_flows)
1238 return cbdata->listed_flows;
1242 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1244 struct list_flows_cbdata cbdata;
1247 if (!flowvec->n_flows)
1250 cbdata.uflows = flowvec->flows;
1251 cbdata.n_flows = flowvec->n_flows;
1252 cbdata.listed_flows = 0;
1254 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1255 return error ? error : cbdata.listed_flows;
1258 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1259 int (*function)(struct datapath *,
1260 const struct odp_flowvec *))
1262 struct odp_flowvec __user *uflowvec;
1263 struct odp_flowvec flowvec;
1266 uflowvec = (struct odp_flowvec __user *)argp;
1267 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1270 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1273 retval = function(dp, &flowvec);
1274 return (retval < 0 ? retval
1275 : retval == flowvec.n_flows ? 0
1276 : put_user(retval, &uflowvec->n_flows));
1279 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1281 struct odp_flow_key key;
1282 struct sk_buff *skb;
1283 struct sw_flow_actions *actions;
1289 if (execute->length < ETH_HLEN || execute->length > 65535)
1292 actions = flow_actions_alloc(execute->n_actions);
1293 if (IS_ERR(actions)) {
1294 err = PTR_ERR(actions);
1299 if (copy_from_user(actions->actions, execute->actions,
1300 execute->n_actions * sizeof *execute->actions))
1301 goto error_free_actions;
1303 err = validate_actions(actions);
1305 goto error_free_actions;
1308 skb = alloc_skb(execute->length, GFP_KERNEL);
1310 goto error_free_actions;
1313 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1315 goto error_free_skb;
1317 skb_reset_mac_header(skb);
1320 /* Normally, setting the skb 'protocol' field would be handled by a
1321 * call to eth_type_trans(), but it assumes there's a sending
1322 * device, which we may not have. */
1323 if (ntohs(eth->h_proto) >= 1536)
1324 skb->protocol = eth->h_proto;
1326 skb->protocol = htons(ETH_P_802_2);
1328 err = flow_extract(skb, -1, &key, &is_frag);
1330 goto error_free_skb;
1333 err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions);
1347 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1349 struct odp_execute execute;
1351 if (copy_from_user(&execute, executep, sizeof execute))
1354 return do_execute(dp, &execute);
1357 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1359 struct tbl *table = rcu_dereference(dp->table);
1360 struct odp_stats stats;
1363 stats.n_flows = tbl_count(table);
1364 stats.cur_capacity = tbl_n_buckets(table);
1365 stats.max_capacity = TBL_MAX_BUCKETS;
1366 stats.n_ports = dp->n_ports;
1367 stats.max_ports = DP_MAX_PORTS;
1368 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1369 for_each_possible_cpu(i) {
1370 const struct dp_stats_percpu *percpu_stats;
1371 struct dp_stats_percpu local_stats;
1374 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1377 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1378 local_stats = *percpu_stats;
1379 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1381 stats.n_frags += local_stats.n_frags;
1382 stats.n_hit += local_stats.n_hit;
1383 stats.n_missed += local_stats.n_missed;
1384 stats.n_lost += local_stats.n_lost;
1386 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1387 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1388 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1391 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1392 int dp_min_mtu(const struct datapath *dp)
1399 list_for_each_entry_rcu (p, &dp->port_list, node) {
1402 /* Skip any internal ports, since that's what we're trying to
1404 if (is_internal_vport(p->vport))
1407 dev_mtu = vport_get_mtu(p->vport);
1408 if (!mtu || dev_mtu < mtu)
1412 return mtu ? mtu : ETH_DATA_LEN;
1415 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1416 * be called with RTNL lock. */
1417 void set_internal_devs_mtu(const struct datapath *dp)
1424 mtu = dp_min_mtu(dp);
1426 list_for_each_entry_rcu (p, &dp->port_list, node) {
1427 if (is_internal_vport(p->vport))
1428 vport_set_mtu(p->vport, mtu);
1432 static int put_port(const struct dp_port *p, struct odp_port __user *uop)
1436 memset(&op, 0, sizeof op);
1439 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1442 op.port = p->port_no;
1443 op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
1445 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1448 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1450 struct odp_port port;
1452 if (copy_from_user(&port, uport, sizeof port))
1455 if (port.devname[0]) {
1456 struct vport *vport;
1457 struct dp_port *dp_port;
1460 port.devname[IFNAMSIZ - 1] = '\0';
1465 vport = vport_locate(port.devname);
1471 dp_port = vport_get_dp_port(vport);
1472 if (!dp_port || dp_port->dp != dp) {
1477 port.port = dp_port->port_no;
1486 if (port.port >= DP_MAX_PORTS)
1488 if (!dp->ports[port.port])
1492 return put_port(dp->ports[port.port], uport);
1495 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1502 list_for_each_entry_rcu (p, &dp->port_list, node) {
1503 if (put_port(p, &uports[idx]))
1505 if (idx++ >= n_ports)
1512 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1514 struct odp_portvec pv;
1517 if (copy_from_user(&pv, upv, sizeof pv))
1520 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1524 return put_user(retval, &upv->n_ports);
1527 static int get_listen_mask(const struct file *f)
1529 return (long)f->private_data;
1532 static void set_listen_mask(struct file *f, int listen_mask)
1534 f->private_data = (void*)(long)listen_mask;
1537 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1540 int dp_idx = iminor(f->f_dentry->d_inode);
1541 struct datapath *dp;
1542 int drop_frags, listeners, port_no;
1543 unsigned int sflow_probability;
1546 /* Handle commands with special locking requirements up front. */
1549 err = create_dp(dp_idx, (char __user *)argp);
1552 case ODP_DP_DESTROY:
1553 err = destroy_dp(dp_idx);
1556 case ODP_PORT_ATTACH:
1557 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1560 case ODP_PORT_DETACH:
1561 err = get_user(port_no, (int __user *)argp);
1563 err = detach_port(dp_idx, port_no);
1567 err = vport_user_add((struct odp_vport_add __user *)argp);
1571 err = vport_user_mod((struct odp_vport_mod __user *)argp);
1575 err = vport_user_del((char __user *)argp);
1578 case ODP_VPORT_STATS_GET:
1579 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1582 case ODP_VPORT_STATS_SET:
1583 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1586 case ODP_VPORT_ETHER_GET:
1587 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1590 case ODP_VPORT_ETHER_SET:
1591 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1594 case ODP_VPORT_MTU_GET:
1595 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1598 case ODP_VPORT_MTU_SET:
1599 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1603 dp = get_dp_locked(dp_idx);
1610 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1613 case ODP_GET_DROP_FRAGS:
1614 err = put_user(dp->drop_frags, (int __user *)argp);
1617 case ODP_SET_DROP_FRAGS:
1618 err = get_user(drop_frags, (int __user *)argp);
1622 if (drop_frags != 0 && drop_frags != 1)
1624 dp->drop_frags = drop_frags;
1628 case ODP_GET_LISTEN_MASK:
1629 err = put_user(get_listen_mask(f), (int __user *)argp);
1632 case ODP_SET_LISTEN_MASK:
1633 err = get_user(listeners, (int __user *)argp);
1637 if (listeners & ~ODPL_ALL)
1640 set_listen_mask(f, listeners);
1643 case ODP_GET_SFLOW_PROBABILITY:
1644 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1647 case ODP_SET_SFLOW_PROBABILITY:
1648 err = get_user(sflow_probability, (unsigned int __user *)argp);
1650 dp->sflow_probability = sflow_probability;
1653 case ODP_PORT_QUERY:
1654 err = query_port(dp, (struct odp_port __user *)argp);
1658 err = list_ports(dp, (struct odp_portvec __user *)argp);
1661 case ODP_FLOW_FLUSH:
1662 err = flush_flows(dp);
1666 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1670 err = del_flow(dp, (struct odp_flow __user *)argp);
1674 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1678 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1682 err = execute_packet(dp, (struct odp_execute __user *)argp);
1689 mutex_unlock(&dp->mutex);
1694 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1697 for (i = 0; i < DP_N_QUEUES; i++) {
1698 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1704 #ifdef CONFIG_COMPAT
1705 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1707 struct compat_odp_portvec pv;
1710 if (copy_from_user(&pv, upv, sizeof pv))
1713 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1717 return put_user(retval, &upv->n_ports);
1720 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1722 compat_uptr_t actions;
1724 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1725 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1726 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1727 __get_user(actions, &compat->actions) ||
1728 __get_user(flow->n_actions, &compat->n_actions) ||
1729 __get_user(flow->flags, &compat->flags))
1732 flow->actions = compat_ptr(actions);
1736 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1738 struct odp_flow_stats stats;
1739 struct odp_flow_put fp;
1742 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1743 get_user(fp.flags, &ufp->flags))
1746 error = do_put_flow(dp, &fp, &stats);
1750 if (copy_to_user(&ufp->flow.stats, &stats,
1751 sizeof(struct odp_flow_stats)))
1757 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1758 struct compat_odp_flow __user *ufp)
1760 compat_uptr_t actions;
1762 if (get_user(actions, &ufp->actions))
1765 return do_answer_query(flow, query_flags, &ufp->stats,
1766 compat_ptr(actions), &ufp->n_actions);
1769 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1771 struct sw_flow *flow;
1775 if (compat_get_flow(&uf, ufp))
1778 flow = do_del_flow(dp, &uf.key);
1780 return PTR_ERR(flow);
1782 error = compat_answer_query(flow, 0, ufp);
1783 flow_deferred_free(flow);
1787 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1789 struct tbl *table = rcu_dereference(dp->table);
1792 for (i = 0; i < n_flows; i++) {
1793 struct compat_odp_flow __user *ufp = &flows[i];
1795 struct tbl_node *flow_node;
1798 if (compat_get_flow(&uf, ufp))
1801 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1803 error = put_user(ENOENT, &ufp->stats.error);
1805 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1812 struct compat_list_flows_cbdata {
1813 struct compat_odp_flow __user *uflows;
1818 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1820 struct sw_flow *flow = flow_cast(node);
1821 struct compat_list_flows_cbdata *cbdata = cbdata_;
1822 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1825 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1827 error = compat_answer_query(flow, 0, ufp);
1831 if (cbdata->listed_flows >= cbdata->n_flows)
1832 return cbdata->listed_flows;
1836 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1838 struct compat_list_flows_cbdata cbdata;
1844 cbdata.uflows = flows;
1845 cbdata.n_flows = n_flows;
1846 cbdata.listed_flows = 0;
1848 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1849 return error ? error : cbdata.listed_flows;
1852 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1853 int (*function)(struct datapath *,
1854 struct compat_odp_flow *,
1857 struct compat_odp_flowvec __user *uflowvec;
1858 struct compat_odp_flow __user *flows;
1859 struct compat_odp_flowvec flowvec;
1862 uflowvec = compat_ptr(argp);
1863 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1864 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1867 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1870 flows = compat_ptr(flowvec.flows);
1871 if (!access_ok(VERIFY_WRITE, flows,
1872 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1875 retval = function(dp, flows, flowvec.n_flows);
1876 return (retval < 0 ? retval
1877 : retval == flowvec.n_flows ? 0
1878 : put_user(retval, &uflowvec->n_flows));
1881 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1883 struct odp_execute execute;
1884 compat_uptr_t actions;
1887 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1888 __get_user(actions, &uexecute->actions) ||
1889 __get_user(execute.n_actions, &uexecute->n_actions) ||
1890 __get_user(data, &uexecute->data) ||
1891 __get_user(execute.length, &uexecute->length))
1894 execute.actions = compat_ptr(actions);
1895 execute.data = compat_ptr(data);
1897 return do_execute(dp, &execute);
1900 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1902 int dp_idx = iminor(f->f_dentry->d_inode);
1903 struct datapath *dp;
1907 case ODP_DP_DESTROY:
1908 case ODP_FLOW_FLUSH:
1909 /* Ioctls that don't need any translation at all. */
1910 return openvswitch_ioctl(f, cmd, argp);
1913 case ODP_PORT_ATTACH:
1914 case ODP_PORT_DETACH:
1916 case ODP_VPORT_MTU_SET:
1917 case ODP_VPORT_MTU_GET:
1918 case ODP_VPORT_ETHER_SET:
1919 case ODP_VPORT_ETHER_GET:
1920 case ODP_VPORT_STATS_SET:
1921 case ODP_VPORT_STATS_GET:
1923 case ODP_GET_DROP_FRAGS:
1924 case ODP_SET_DROP_FRAGS:
1925 case ODP_SET_LISTEN_MASK:
1926 case ODP_GET_LISTEN_MASK:
1927 case ODP_SET_SFLOW_PROBABILITY:
1928 case ODP_GET_SFLOW_PROBABILITY:
1929 case ODP_PORT_QUERY:
1930 /* Ioctls that just need their pointer argument extended. */
1931 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1933 case ODP_VPORT_ADD32:
1934 return compat_vport_user_add(compat_ptr(argp));
1936 case ODP_VPORT_MOD32:
1937 return compat_vport_user_mod(compat_ptr(argp));
1940 dp = get_dp_locked(dp_idx);
1946 case ODP_PORT_LIST32:
1947 err = compat_list_ports(dp, compat_ptr(argp));
1950 case ODP_FLOW_PUT32:
1951 err = compat_put_flow(dp, compat_ptr(argp));
1954 case ODP_FLOW_DEL32:
1955 err = compat_del_flow(dp, compat_ptr(argp));
1958 case ODP_FLOW_GET32:
1959 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1962 case ODP_FLOW_LIST32:
1963 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1967 err = compat_execute(dp, compat_ptr(argp));
1974 mutex_unlock(&dp->mutex);
1980 /* Unfortunately this function is not exported so this is a verbatim copy
1981 * from net/core/datagram.c in 2.6.30. */
1982 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1983 u8 __user *to, int len,
1986 int start = skb_headlen(skb);
1988 int i, copy = start - offset;
1995 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1999 if ((len -= copy) == 0)
2006 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2009 WARN_ON(start > offset + len);
2011 end = start + skb_shinfo(skb)->frags[i].size;
2012 if ((copy = end - offset) > 0) {
2016 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2017 struct page *page = frag->page;
2022 csum2 = csum_and_copy_to_user(vaddr +
2029 *csump = csum_block_add(*csump, csum2, pos);
2039 if (skb_shinfo(skb)->frag_list) {
2040 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2042 for (; list; list=list->next) {
2045 WARN_ON(start > offset + len);
2047 end = start + list->len;
2048 if ((copy = end - offset) > 0) {
2052 if (skb_copy_and_csum_datagram(list,
2057 *csump = csum_block_add(*csump, csum2, pos);
2058 if ((len -= copy) == 0)
2074 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2077 /* XXX is there sufficient synchronization here? */
2078 int listeners = get_listen_mask(f);
2079 int dp_idx = iminor(f->f_dentry->d_inode);
2080 struct datapath *dp = get_dp(dp_idx);
2081 struct sk_buff *skb;
2082 size_t copy_bytes, tot_copy_bytes;
2088 if (nbytes == 0 || !listeners)
2094 for (i = 0; i < DP_N_QUEUES; i++) {
2095 if (listeners & (1 << i)) {
2096 skb = skb_dequeue(&dp->queues[i]);
2102 if (f->f_flags & O_NONBLOCK) {
2107 wait_event_interruptible(dp->waitqueue,
2108 dp_has_packet_of_interest(dp,
2111 if (signal_pending(current)) {
2112 retval = -ERESTARTSYS;
2117 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2120 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2121 if (copy_bytes == skb->len) {
2123 unsigned int csum_start, csum_offset;
2125 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2126 csum_start = skb->csum_start - skb_headroom(skb);
2127 csum_offset = skb->csum_offset;
2129 csum_start = skb_transport_header(skb) - skb->data;
2130 csum_offset = skb->csum;
2132 BUG_ON(csum_start >= skb_headlen(skb));
2133 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2134 copy_bytes - csum_start, &csum);
2136 __sum16 __user *csump;
2138 copy_bytes = csum_start;
2139 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2141 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
2142 put_user(csum_fold(csum), csump);
2145 retval = skb_checksum_help(skb);
2149 struct iovec __user iov;
2152 iov.iov_len = copy_bytes;
2153 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2157 retval = tot_copy_bytes;
2165 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2167 /* XXX is there sufficient synchronization here? */
2168 int dp_idx = iminor(file->f_dentry->d_inode);
2169 struct datapath *dp = get_dp(dp_idx);
2174 poll_wait(file, &dp->waitqueue, wait);
2175 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2176 mask |= POLLIN | POLLRDNORM;
2178 mask = POLLIN | POLLRDNORM | POLLHUP;
2183 struct file_operations openvswitch_fops = {
2184 /* XXX .aio_read = openvswitch_aio_read, */
2185 .read = openvswitch_read,
2186 .poll = openvswitch_poll,
2187 .unlocked_ioctl = openvswitch_ioctl,
2188 #ifdef CONFIG_COMPAT
2189 .compat_ioctl = openvswitch_compat_ioctl,
2191 /* XXX .fasync = openvswitch_fasync, */
2196 static int __init dp_init(void)
2198 struct sk_buff *dummy_skb;
2201 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2203 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2211 goto error_flow_exit;
2213 err = register_netdevice_notifier(&dp_device_notifier);
2215 goto error_vport_exit;
2217 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2219 goto error_unreg_notifier;
2223 error_unreg_notifier:
2224 unregister_netdevice_notifier(&dp_device_notifier);
2233 static void dp_cleanup(void)
2236 unregister_chrdev(major, "openvswitch");
2237 unregister_netdevice_notifier(&dp_device_notifier);
2242 module_init(dp_init);
2243 module_exit(dp_cleanup);
2245 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2246 MODULE_LICENSE("GPL");