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 #include <linux/init.h>
12 #include <linux/module.h>
14 #include <linux/if_arp.h>
15 #include <linux/if_vlan.h>
18 #include <linux/delay.h>
19 #include <linux/time.h>
20 #include <linux/etherdevice.h>
21 #include <linux/kernel.h>
22 #include <linux/kthread.h>
23 #include <linux/mutex.h>
24 #include <linux/percpu.h>
25 #include <linux/rcupdate.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/version.h>
29 #include <linux/ethtool.h>
30 #include <linux/random.h>
31 #include <linux/wait.h>
32 #include <asm/system.h>
33 #include <asm/div64.h>
35 #include <asm/highmem.h>
36 #include <linux/netfilter_bridge.h>
37 #include <linux/netfilter_ipv4.h>
38 #include <linux/inetdevice.h>
39 #include <linux/list.h>
40 #include <linux/rculist.h>
41 #include <linux/workqueue.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 "odp-compat.h"
52 #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 /* Number of milliseconds between runs of the maintenance thread. */
73 #define MAINT_SLEEP_MSECS 1000
75 static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
77 /* Must be called with rcu_read_lock or dp_mutex. */
78 struct datapath *get_dp(int dp_idx)
80 if (dp_idx < 0 || dp_idx >= ODP_MAX)
82 return rcu_dereference(dps[dp_idx]);
84 EXPORT_SYMBOL_GPL(get_dp);
86 static struct datapath *get_dp_locked(int dp_idx)
90 mutex_lock(&dp_mutex);
93 mutex_lock(&dp->mutex);
94 mutex_unlock(&dp_mutex);
98 /* Must be called with rcu_read_lock or RTNL lock. */
99 const char *dp_name(const struct datapath *dp)
101 return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
104 static inline size_t br_nlmsg_size(void)
106 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
107 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
108 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
109 + nla_total_size(4) /* IFLA_MASTER */
110 + nla_total_size(4) /* IFLA_MTU */
111 + nla_total_size(4) /* IFLA_LINK */
112 + nla_total_size(1); /* IFLA_OPERSTATE */
115 static int dp_fill_ifinfo(struct sk_buff *skb,
116 const struct dp_port *port,
117 int event, unsigned int flags)
119 const struct datapath *dp = port->dp;
120 int ifindex = vport_get_ifindex(port->vport);
121 int iflink = vport_get_iflink(port->vport);
122 struct ifinfomsg *hdr;
123 struct nlmsghdr *nlh;
131 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
135 hdr = nlmsg_data(nlh);
136 hdr->ifi_family = AF_BRIDGE;
138 hdr->ifi_type = ARPHRD_ETHER;
139 hdr->ifi_index = ifindex;
140 hdr->ifi_flags = vport_get_flags(port->vport);
143 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
144 NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
145 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
146 #ifdef IFLA_OPERSTATE
147 NLA_PUT_U8(skb, IFLA_OPERSTATE,
148 vport_is_running(port->vport)
149 ? vport_get_operstate(port->vport)
153 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
154 vport_get_addr(port->vport));
156 if (ifindex != iflink)
157 NLA_PUT_U32(skb, IFLA_LINK,iflink);
159 return nlmsg_end(skb, nlh);
162 nlmsg_cancel(skb, nlh);
166 static void dp_ifinfo_notify(int event, struct dp_port *port)
171 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
175 err = dp_fill_ifinfo(skb, port, event, 0);
177 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
178 WARN_ON(err == -EMSGSIZE);
182 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
186 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
189 static void release_dp(struct kobject *kobj)
191 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
195 static struct kobj_type dp_ktype = {
196 .release = release_dp
199 static int create_dp(int dp_idx, const char __user *devnamep)
201 struct odp_port internal_dev_port;
202 char devname[IFNAMSIZ];
208 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
212 } else if (retval >= IFNAMSIZ) {
217 snprintf(devname, sizeof devname, "of%d", dp_idx);
221 mutex_lock(&dp_mutex);
223 if (!try_module_get(THIS_MODULE))
226 /* Exit early if a datapath with that number already exists.
227 * (We don't use -EEXIST because that's ambiguous with 'devname'
228 * conflicting with an existing network device name.) */
234 dp = kzalloc(sizeof *dp, GFP_KERNEL);
237 INIT_LIST_HEAD(&dp->port_list);
238 mutex_init(&dp->mutex);
240 for (i = 0; i < DP_N_QUEUES; i++)
241 skb_queue_head_init(&dp->queues[i]);
242 init_waitqueue_head(&dp->waitqueue);
244 /* Initialize kobject for bridge. This will be added as
245 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
246 dp->ifobj.kset = NULL;
247 kobject_init(&dp->ifobj, &dp_ktype);
249 /* Allocate table. */
251 rcu_assign_pointer(dp->table, tbl_create(0));
255 /* Set up our datapath device. */
256 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
257 strcpy(internal_dev_port.devname, devname);
258 internal_dev_port.flags = ODP_PORT_INTERNAL;
259 err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
264 goto err_destroy_table;
268 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
269 if (!dp->stats_percpu)
270 goto err_destroy_local_port;
272 rcu_assign_pointer(dps[dp_idx], dp);
273 mutex_unlock(&dp_mutex);
280 err_destroy_local_port:
281 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
283 tbl_destroy(dp->table, NULL);
287 module_put(THIS_MODULE);
289 mutex_unlock(&dp_mutex);
295 static void do_destroy_dp(struct datapath *dp)
297 struct dp_port *p, *n;
300 list_for_each_entry_safe (p, n, &dp->port_list, node)
301 if (p->port_no != ODPP_LOCAL)
302 dp_detach_port(p, 1);
306 rcu_assign_pointer(dps[dp->dp_idx], NULL);
308 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
310 tbl_destroy(dp->table, flow_free_tbl);
312 for (i = 0; i < DP_N_QUEUES; i++)
313 skb_queue_purge(&dp->queues[i]);
314 for (i = 0; i < DP_MAX_GROUPS; i++)
315 kfree(dp->groups[i]);
316 free_percpu(dp->stats_percpu);
317 kobject_put(&dp->ifobj);
318 module_put(THIS_MODULE);
321 static int destroy_dp(int dp_idx)
327 mutex_lock(&dp_mutex);
337 mutex_unlock(&dp_mutex);
342 static void release_dp_port(struct kobject *kobj)
344 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
348 static struct kobj_type brport_ktype = {
350 .sysfs_ops = &brport_sysfs_ops,
352 .release = release_dp_port
355 /* Called with RTNL lock and dp_mutex. */
356 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
362 vport = vport_locate(odp_port->devname);
366 if (odp_port->flags & ODP_PORT_INTERNAL)
367 vport = vport_add(odp_port->devname, "internal", NULL);
369 vport = vport_add(odp_port->devname, "netdev", NULL);
374 return PTR_ERR(vport);
377 p = kzalloc(sizeof(*p), GFP_KERNEL);
381 p->port_no = port_no;
383 atomic_set(&p->sflow_pool, 0);
385 err = vport_attach(vport, p);
391 rcu_assign_pointer(dp->ports[port_no], p);
392 list_add_rcu(&p->node, &dp->port_list);
395 /* Initialize kobject for bridge. This will be added as
396 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
398 kobject_init(&p->kobj, &brport_ktype);
400 dp_ifinfo_notify(RTM_NEWLINK, p);
405 static int attach_port(int dp_idx, struct odp_port __user *portp)
408 struct odp_port port;
413 if (copy_from_user(&port, portp, sizeof port))
415 port.devname[IFNAMSIZ - 1] = '\0';
418 dp = get_dp_locked(dp_idx);
421 goto out_unlock_rtnl;
423 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
424 if (!dp->ports[port_no])
430 err = new_dp_port(dp, &port, port_no);
434 set_internal_devs_mtu(dp);
435 dp_sysfs_add_if(dp->ports[port_no]);
437 err = put_user(port_no, &portp->port);
440 mutex_unlock(&dp->mutex);
447 int dp_detach_port(struct dp_port *p, int may_delete)
449 struct vport *vport = p->vport;
454 if (p->port_no != ODPP_LOCAL)
456 dp_ifinfo_notify(RTM_DELLINK, p);
458 /* First drop references to device. */
460 list_del_rcu(&p->node);
461 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
463 err = vport_detach(vport);
467 /* Then wait until no one is still using it, and destroy it. */
471 const char *port_type = vport_get_type(vport);
473 if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
480 kobject_put(&p->kobj);
485 static int detach_port(int dp_idx, int port_no)
492 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
496 dp = get_dp_locked(dp_idx);
499 goto out_unlock_rtnl;
501 p = dp->ports[port_no];
506 err = dp_detach_port(p, 1);
509 mutex_unlock(&dp->mutex);
516 /* Must be called with rcu_read_lock. */
517 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
519 struct datapath *dp = p->dp;
520 struct dp_stats_percpu *stats;
521 int stats_counter_off;
522 struct odp_flow_key key;
523 struct tbl_node *flow_node;
525 WARN_ON_ONCE(skb_shared(skb));
526 skb_warn_if_lro(skb);
528 OVS_CB(skb)->dp_port = p;
530 if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
531 if (dp->drop_frags) {
533 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
538 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp);
540 struct sw_flow *flow = flow_cast(flow_node);
541 struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
542 flow_used(flow, skb);
543 execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
545 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
547 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
548 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
553 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
554 (*(u64 *)((u8 *)stats + stats_counter_off))++;
558 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
559 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
560 * can't call this function directly because it isn't exported in all
562 int vswitch_skb_checksum_setup(struct sk_buff *skb)
567 __u16 csum_start, csum_offset;
569 if (!skb->proto_csum_blank)
572 if (skb->protocol != htons(ETH_P_IP))
575 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
579 th = skb_network_header(skb) + 4 * iph->ihl;
581 csum_start = th - skb->head;
582 switch (iph->protocol) {
584 csum_offset = offsetof(struct tcphdr, check);
587 csum_offset = offsetof(struct udphdr, check);
591 printk(KERN_ERR "Attempting to checksum a non-"
592 "TCP/UDP packet, dropping a protocol"
593 " %d packet", iph->protocol);
597 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
600 skb->ip_summed = CHECKSUM_PARTIAL;
601 skb->proto_csum_blank = 0;
603 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
604 skb->csum_start = csum_start;
605 skb->csum_offset = csum_offset;
607 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
608 skb->csum = csum_offset;
616 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
618 /* Types of checksums that we can receive (these all refer to L4 checksums):
619 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
620 * (though not verified) checksum in packet but not in skb->csum. Packets
621 * from the bridge local port will also have this type.
622 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
623 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
624 * a valid skb->csum. Importantly, both contain a full checksum (not
625 * verified) in the packet itself. The only difference is that if the
626 * packet gets to L4 processing on this machine (not in DomU) we won't
627 * have to recompute the checksum to verify. Most hardware devices do not
628 * produce packets with this type, even if they support receive checksum
629 * offloading (they produce type #5).
630 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
631 * be computed if it is sent off box. Unfortunately on earlier kernels,
632 * this case is impossible to distinguish from #2, despite having opposite
633 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
634 * to distinguish the different states.
635 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
636 * generated locally by a Xen DomU and has a partial checksum. If it is
637 * handled on this machine (Dom0 or DomU), then the checksum will not be
638 * computed. If it goes off box, the checksum in the packet needs to be
639 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
640 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
641 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
642 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
643 * full checksum or using a protocol without a checksum. skb->csum is
644 * undefined. This is common from devices with receive checksum
645 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
646 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
648 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
649 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
650 * based on whether it is on the transmit or receive path. After the datapath
651 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
652 * checksum, we will panic. Since we can receive packets with checksums, we
653 * assume that all CHECKSUM_HW packets have checksums and map them to
654 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
655 * packet is processed by the local IP stack, in which case it will need to
656 * be reverified). If we receive a packet with CHECKSUM_HW that really means
657 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
658 * shouldn't be any devices that do this with bridging. */
660 compute_ip_summed(struct sk_buff *skb, bool xmit)
662 /* For our convenience these defines change repeatedly between kernel
663 * versions, so we can't just copy them over... */
664 switch (skb->ip_summed) {
666 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
668 case CHECKSUM_UNNECESSARY:
669 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
672 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
673 * However, on the receive side we should only get CHECKSUM_PARTIAL
674 * packets from Xen, which uses some special fields to represent this
675 * (see below). Since we can only make one type work, pick the one
676 * that actually happens in practice.
678 * On the transmit side (basically after skb_checksum_setup()
679 * has been run or on internal dev transmit), packets with
680 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
683 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
685 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
689 case CHECKSUM_COMPLETE:
690 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
692 case CHECKSUM_PARTIAL:
693 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
697 printk(KERN_ERR "openvswitch: unknown checksum type %d\n",
699 /* None seems the safest... */
700 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
703 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
704 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
705 * kernels. It should not be set on the transmit path though. */
706 if (skb->proto_csum_blank)
707 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
709 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
713 /* This function closely resembles skb_forward_csum() used by the bridge. It
714 * is slightly different because we are only concerned with bridging and not
715 * other types of forwarding and can get away with slightly more optimal
718 forward_ip_summed(struct sk_buff *skb)
721 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
722 skb->ip_summed = CHECKSUM_NONE;
726 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
727 * unless we broke up a GSO packet. */
729 queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
730 int queue_no, u32 arg)
732 struct sk_buff *nskb;
736 if (OVS_CB(skb)->dp_port)
737 port_no = OVS_CB(skb)->dp_port->port_no;
739 port_no = ODPP_LOCAL;
742 struct odp_msg *header;
747 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
748 /* Until 2.6.22, the start of the transport header was
749 * also the start of data to be checksummed. Linux
750 * 2.6.22 introduced the csum_start field for this
751 * purpose, but we should point the transport header to
752 * it anyway for backward compatibility, as
753 * dev_queue_xmit() does even in 2.6.28. */
754 skb_set_transport_header(skb, skb->csum_start - skb_headroom(skb));
757 err = skb_cow(skb, sizeof *header);
761 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
762 header->type = queue_no;
763 header->length = skb->len;
764 header->port = port_no;
765 header->reserved = 0;
767 skb_queue_tail(queue, skb);
775 while ((skb = nskb) != NULL) {
783 dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
786 struct dp_stats_percpu *stats;
787 struct sk_buff_head *queue;
790 WARN_ON_ONCE(skb_shared(skb));
791 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
792 queue = &dp->queues[queue_no];
794 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
797 forward_ip_summed(skb);
799 err = vswitch_skb_checksum_setup(skb);
803 /* Break apart GSO packets into their component pieces. Otherwise
804 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
805 if (skb_is_gso(skb)) {
806 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
810 if (unlikely(IS_ERR(skb))) {
815 /* XXX This case might not be possible. It's hard to
816 * tell from the skb_gso_segment() code and comment. */
820 err = queue_control_packets(skb, queue, queue_no, arg);
821 wake_up_interruptible(&dp->waitqueue);
828 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
835 static int flush_flows(struct datapath *dp)
837 struct tbl *old_table = rcu_dereference(dp->table);
838 struct tbl *new_table;
840 new_table = tbl_create(0);
844 rcu_assign_pointer(dp->table, new_table);
846 tbl_deferred_destroy(old_table, flow_free_tbl);
851 static int validate_actions(const struct sw_flow_actions *actions)
855 for (i = 0; i < actions->n_actions; i++) {
856 const union odp_action *a = &actions->actions[i];
859 if (a->output.port >= DP_MAX_PORTS)
863 case ODPAT_OUTPUT_GROUP:
864 if (a->output_group.group >= DP_MAX_GROUPS)
868 case ODPAT_SET_VLAN_VID:
869 if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
873 case ODPAT_SET_VLAN_PCP:
874 if (a->vlan_pcp.vlan_pcp
875 & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
879 case ODPAT_SET_NW_TOS:
880 if (a->nw_tos.nw_tos & INET_ECN_MASK)
885 if (a->type >= ODPAT_N_ACTIONS)
894 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
896 struct sw_flow_actions *actions;
899 actions = flow_actions_alloc(flow->n_actions);
900 error = PTR_ERR(actions);
905 if (copy_from_user(actions->actions, flow->actions,
906 flow->n_actions * sizeof(union odp_action)))
907 goto error_free_actions;
908 error = validate_actions(actions);
910 goto error_free_actions;
917 return ERR_PTR(error);
920 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
922 if (flow->used.tv_sec) {
923 stats->used_sec = flow->used.tv_sec;
924 stats->used_nsec = flow->used.tv_nsec;
927 stats->used_nsec = 0;
929 stats->n_packets = flow->packet_count;
930 stats->n_bytes = flow->byte_count;
931 stats->ip_tos = flow->ip_tos;
932 stats->tcp_flags = flow->tcp_flags;
936 static void clear_stats(struct sw_flow *flow)
938 flow->used.tv_sec = flow->used.tv_nsec = 0;
941 flow->packet_count = 0;
942 flow->byte_count = 0;
945 static int expand_table(struct datapath *dp)
947 struct tbl *old_table = rcu_dereference(dp->table);
948 struct tbl *new_table;
950 new_table = tbl_expand(old_table);
951 if (IS_ERR(new_table))
952 return PTR_ERR(new_table);
954 rcu_assign_pointer(dp->table, new_table);
955 tbl_deferred_destroy(old_table, NULL);
960 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
961 struct odp_flow_stats *stats)
963 struct tbl_node *flow_node;
964 struct sw_flow *flow;
968 memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved);
970 table = rcu_dereference(dp->table);
971 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
974 struct sw_flow_actions *acts;
977 if (!(uf->flags & ODPPF_CREATE))
980 /* Expand table, if necessary, to make room. */
981 if (tbl_count(table) >= tbl_n_buckets(table)) {
982 error = expand_table(dp);
985 table = rcu_dereference(dp->table);
990 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
993 flow->key = uf->flow.key;
994 spin_lock_init(&flow->lock);
997 /* Obtain actions. */
998 acts = get_actions(&uf->flow);
999 error = PTR_ERR(acts);
1001 goto error_free_flow;
1002 rcu_assign_pointer(flow->sf_acts, acts);
1004 /* Put flow in bucket. */
1005 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1007 goto error_free_flow_acts;
1009 memset(stats, 0, sizeof(struct odp_flow_stats));
1011 /* We found a matching flow. */
1012 struct sw_flow_actions *old_acts, *new_acts;
1014 flow = flow_cast(flow_node);
1016 /* Bail out if we're not allowed to modify an existing flow. */
1018 if (!(uf->flags & ODPPF_MODIFY))
1022 new_acts = get_actions(&uf->flow);
1023 error = PTR_ERR(new_acts);
1024 if (IS_ERR(new_acts))
1026 old_acts = rcu_dereference(flow->sf_acts);
1027 if (old_acts->n_actions != new_acts->n_actions ||
1028 memcmp(old_acts->actions, new_acts->actions,
1029 sizeof(union odp_action) * old_acts->n_actions)) {
1030 rcu_assign_pointer(flow->sf_acts, new_acts);
1031 flow_deferred_free_acts(old_acts);
1036 /* Fetch stats, then clear them if necessary. */
1037 spin_lock_bh(&flow->lock);
1038 get_stats(flow, stats);
1039 if (uf->flags & ODPPF_ZERO_STATS)
1041 spin_unlock_bh(&flow->lock);
1046 error_free_flow_acts:
1047 kfree(flow->sf_acts);
1049 kmem_cache_free(flow_cache, flow);
1054 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1056 struct odp_flow_stats stats;
1057 struct odp_flow_put uf;
1060 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1063 error = do_put_flow(dp, &uf, &stats);
1067 if (copy_to_user(&ufp->flow.stats, &stats,
1068 sizeof(struct odp_flow_stats)))
1074 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1075 struct odp_flow_stats __user *ustats,
1076 union odp_action __user *actions,
1077 u32 __user *n_actionsp)
1079 struct sw_flow_actions *sf_acts;
1080 struct odp_flow_stats stats;
1083 spin_lock_bh(&flow->lock);
1084 get_stats(flow, &stats);
1085 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1086 flow->tcp_flags = 0;
1088 spin_unlock_bh(&flow->lock);
1090 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1091 get_user(n_actions, n_actionsp))
1097 sf_acts = rcu_dereference(flow->sf_acts);
1098 if (put_user(sf_acts->n_actions, n_actionsp) ||
1099 (actions && copy_to_user(actions, sf_acts->actions,
1100 sizeof(union odp_action) *
1101 min(sf_acts->n_actions, n_actions))))
1107 static int answer_query(struct sw_flow *flow, u32 query_flags,
1108 struct odp_flow __user *ufp)
1110 union odp_action *actions;
1112 if (get_user(actions, &ufp->actions))
1115 return do_answer_query(flow, query_flags,
1116 &ufp->stats, actions, &ufp->n_actions);
1119 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1121 struct tbl *table = rcu_dereference(dp->table);
1122 struct tbl_node *flow_node;
1125 memset(key->reserved, 0, sizeof key->reserved);
1126 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1128 return ERR_PTR(-ENOENT);
1130 error = tbl_remove(table, flow_node);
1132 return ERR_PTR(error);
1134 /* XXX Returned flow_node's statistics might lose a few packets, since
1135 * other CPUs can be using this flow. We used to synchronize_rcu() to
1136 * make sure that we get completely accurate stats, but that blows our
1137 * performance, badly. */
1138 return flow_cast(flow_node);
1141 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1143 struct sw_flow *flow;
1147 if (copy_from_user(&uf, ufp, sizeof uf))
1150 flow = do_del_flow(dp, &uf.key);
1152 return PTR_ERR(flow);
1154 error = answer_query(flow, 0, ufp);
1155 flow_deferred_free(flow);
1159 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1161 struct tbl *table = rcu_dereference(dp->table);
1164 for (i = 0; i < flowvec->n_flows; i++) {
1165 struct odp_flow __user *ufp = &flowvec->flows[i];
1167 struct tbl_node *flow_node;
1170 if (copy_from_user(&uf, ufp, sizeof uf))
1172 memset(uf.key.reserved, 0, sizeof uf.key.reserved);
1174 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1176 error = put_user(ENOENT, &ufp->stats.error);
1178 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1182 return flowvec->n_flows;
1185 struct list_flows_cbdata {
1186 struct odp_flow __user *uflows;
1191 static int list_flow(struct tbl_node *node, void *cbdata_)
1193 struct sw_flow *flow = flow_cast(node);
1194 struct list_flows_cbdata *cbdata = cbdata_;
1195 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1198 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1200 error = answer_query(flow, 0, ufp);
1204 if (cbdata->listed_flows >= cbdata->n_flows)
1205 return cbdata->listed_flows;
1209 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1211 struct list_flows_cbdata cbdata;
1214 if (!flowvec->n_flows)
1217 cbdata.uflows = flowvec->flows;
1218 cbdata.n_flows = flowvec->n_flows;
1219 cbdata.listed_flows = 0;
1220 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1221 return error ? error : cbdata.listed_flows;
1224 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1225 int (*function)(struct datapath *,
1226 const struct odp_flowvec *))
1228 struct odp_flowvec __user *uflowvec;
1229 struct odp_flowvec flowvec;
1232 uflowvec = (struct odp_flowvec __user *)argp;
1233 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1236 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1239 retval = function(dp, &flowvec);
1240 return (retval < 0 ? retval
1241 : retval == flowvec.n_flows ? 0
1242 : put_user(retval, &uflowvec->n_flows));
1245 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1247 struct odp_flow_key key;
1248 struct sk_buff *skb;
1249 struct sw_flow_actions *actions;
1254 if (execute->length < ETH_HLEN || execute->length > 65535)
1258 actions = flow_actions_alloc(execute->n_actions);
1263 if (copy_from_user(actions->actions, execute->actions,
1264 execute->n_actions * sizeof *execute->actions))
1265 goto error_free_actions;
1267 err = validate_actions(actions);
1269 goto error_free_actions;
1272 skb = alloc_skb(execute->length, GFP_KERNEL);
1274 goto error_free_actions;
1276 if (execute->in_port < DP_MAX_PORTS)
1277 OVS_CB(skb)->dp_port = dp->ports[execute->in_port];
1279 OVS_CB(skb)->dp_port = NULL;
1282 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1284 goto error_free_skb;
1286 skb_reset_mac_header(skb);
1289 /* Normally, setting the skb 'protocol' field would be handled by a
1290 * call to eth_type_trans(), but it assumes there's a sending
1291 * device, which we may not have. */
1292 if (ntohs(eth->h_proto) >= 1536)
1293 skb->protocol = eth->h_proto;
1295 skb->protocol = htons(ETH_P_802_2);
1297 flow_extract(skb, execute->in_port, &key);
1300 err = execute_actions(dp, skb, &key, actions->actions,
1301 actions->n_actions, GFP_KERNEL);
1315 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1317 struct odp_execute execute;
1319 if (copy_from_user(&execute, executep, sizeof execute))
1322 return do_execute(dp, &execute);
1325 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1327 struct tbl *table = rcu_dereference(dp->table);
1328 struct odp_stats stats;
1331 stats.n_flows = tbl_count(table);
1332 stats.cur_capacity = tbl_n_buckets(table);
1333 stats.max_capacity = TBL_MAX_BUCKETS;
1334 stats.n_ports = dp->n_ports;
1335 stats.max_ports = DP_MAX_PORTS;
1336 stats.max_groups = DP_MAX_GROUPS;
1337 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1338 for_each_possible_cpu(i) {
1339 const struct dp_stats_percpu *s;
1340 s = per_cpu_ptr(dp->stats_percpu, i);
1341 stats.n_frags += s->n_frags;
1342 stats.n_hit += s->n_hit;
1343 stats.n_missed += s->n_missed;
1344 stats.n_lost += s->n_lost;
1346 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1347 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1348 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1351 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1352 int dp_min_mtu(const struct datapath *dp)
1359 list_for_each_entry_rcu (p, &dp->port_list, node) {
1362 /* Skip any internal ports, since that's what we're trying to
1364 if (is_internal_vport(p->vport))
1367 dev_mtu = vport_get_mtu(p->vport);
1368 if (!mtu || dev_mtu < mtu)
1372 return mtu ? mtu : ETH_DATA_LEN;
1375 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1376 * be called with RTNL lock. */
1377 void set_internal_devs_mtu(const struct datapath *dp)
1384 mtu = dp_min_mtu(dp);
1386 list_for_each_entry_rcu (p, &dp->port_list, node) {
1387 if (is_internal_vport(p->vport))
1388 vport_set_mtu(p->vport, mtu);
1393 put_port(const struct dp_port *p, struct odp_port __user *uop)
1397 memset(&op, 0, sizeof op);
1400 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1403 op.port = p->port_no;
1404 op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
1406 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1410 query_port(struct datapath *dp, struct odp_port __user *uport)
1412 struct odp_port port;
1414 if (copy_from_user(&port, uport, sizeof port))
1417 if (port.devname[0]) {
1418 struct vport *vport;
1419 struct dp_port *dp_port;
1422 port.devname[IFNAMSIZ - 1] = '\0';
1427 vport = vport_locate(port.devname);
1433 dp_port = vport_get_dp_port(vport);
1434 if (!dp_port || dp_port->dp != dp) {
1439 port.port = dp_port->port_no;
1448 if (port.port >= DP_MAX_PORTS)
1450 if (!dp->ports[port.port])
1454 return put_port(dp->ports[port.port], uport);
1458 do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports)
1464 list_for_each_entry_rcu (p, &dp->port_list, node) {
1465 if (put_port(p, &uports[idx]))
1467 if (idx++ >= n_ports)
1475 list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1477 struct odp_portvec pv;
1480 if (copy_from_user(&pv, upv, sizeof pv))
1483 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1487 return put_user(retval, &upv->n_ports);
1490 /* RCU callback for freeing a dp_port_group */
1491 static void free_port_group(struct rcu_head *rcu)
1493 struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
1498 do_set_port_group(struct datapath *dp, u16 __user *ports, int n_ports, int group)
1500 struct dp_port_group *new_group, *old_group;
1504 if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS)
1508 new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL);
1512 new_group->n_ports = n_ports;
1514 if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports))
1517 old_group = rcu_dereference(dp->groups[group]);
1518 rcu_assign_pointer(dp->groups[group], new_group);
1520 call_rcu(&old_group->rcu, free_port_group);
1530 set_port_group(struct datapath *dp, const struct odp_port_group __user *upg)
1532 struct odp_port_group pg;
1534 if (copy_from_user(&pg, upg, sizeof pg))
1537 return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group);
1541 do_get_port_group(struct datapath *dp,
1542 u16 __user *ports, int n_ports, int group,
1543 u16 __user *n_portsp)
1545 struct dp_port_group *g;
1548 if (group >= DP_MAX_GROUPS)
1551 g = dp->groups[group];
1552 n_copy = g ? min_t(int, g->n_ports, n_ports) : 0;
1553 if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16)))
1556 if (put_user(g ? g->n_ports : 0, n_portsp))
1562 static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg)
1564 struct odp_port_group pg;
1566 if (copy_from_user(&pg, upg, sizeof pg))
1569 return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &upg->n_ports);
1572 static int get_listen_mask(const struct file *f)
1574 return (long)f->private_data;
1577 static void set_listen_mask(struct file *f, int listen_mask)
1579 f->private_data = (void*)(long)listen_mask;
1582 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1585 int dp_idx = iminor(f->f_dentry->d_inode);
1586 struct datapath *dp;
1587 int drop_frags, listeners, port_no;
1588 unsigned int sflow_probability;
1591 /* Handle commands with special locking requirements up front. */
1594 err = create_dp(dp_idx, (char __user *)argp);
1597 case ODP_DP_DESTROY:
1598 err = destroy_dp(dp_idx);
1601 case ODP_PORT_ATTACH:
1602 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1605 case ODP_PORT_DETACH:
1606 err = get_user(port_no, (int __user *)argp);
1608 err = detach_port(dp_idx, port_no);
1612 err = vport_user_add((struct odp_vport_add __user *)argp);
1616 err = vport_user_mod((struct odp_vport_mod __user *)argp);
1620 err = vport_user_del((char __user *)argp);
1623 case ODP_VPORT_STATS_GET:
1624 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1627 case ODP_VPORT_STATS_SET:
1628 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1631 case ODP_VPORT_ETHER_GET:
1632 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1635 case ODP_VPORT_ETHER_SET:
1636 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1639 case ODP_VPORT_MTU_GET:
1640 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1643 case ODP_VPORT_MTU_SET:
1644 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1648 dp = get_dp_locked(dp_idx);
1655 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1658 case ODP_GET_DROP_FRAGS:
1659 err = put_user(dp->drop_frags, (int __user *)argp);
1662 case ODP_SET_DROP_FRAGS:
1663 err = get_user(drop_frags, (int __user *)argp);
1667 if (drop_frags != 0 && drop_frags != 1)
1669 dp->drop_frags = drop_frags;
1673 case ODP_GET_LISTEN_MASK:
1674 err = put_user(get_listen_mask(f), (int __user *)argp);
1677 case ODP_SET_LISTEN_MASK:
1678 err = get_user(listeners, (int __user *)argp);
1682 if (listeners & ~ODPL_ALL)
1685 set_listen_mask(f, listeners);
1688 case ODP_GET_SFLOW_PROBABILITY:
1689 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1692 case ODP_SET_SFLOW_PROBABILITY:
1693 err = get_user(sflow_probability, (unsigned int __user *)argp);
1695 dp->sflow_probability = sflow_probability;
1698 case ODP_PORT_QUERY:
1699 err = query_port(dp, (struct odp_port __user *)argp);
1703 err = list_ports(dp, (struct odp_portvec __user *)argp);
1706 case ODP_PORT_GROUP_SET:
1707 err = set_port_group(dp, (struct odp_port_group __user *)argp);
1710 case ODP_PORT_GROUP_GET:
1711 err = get_port_group(dp, (struct odp_port_group __user *)argp);
1714 case ODP_FLOW_FLUSH:
1715 err = flush_flows(dp);
1719 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1723 err = del_flow(dp, (struct odp_flow __user *)argp);
1727 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1731 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1735 err = execute_packet(dp, (struct odp_execute __user *)argp);
1742 mutex_unlock(&dp->mutex);
1747 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1750 for (i = 0; i < DP_N_QUEUES; i++) {
1751 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1757 #ifdef CONFIG_COMPAT
1758 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1760 struct compat_odp_portvec pv;
1763 if (copy_from_user(&pv, upv, sizeof pv))
1766 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1770 return put_user(retval, &upv->n_ports);
1773 static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg)
1775 struct compat_odp_port_group pg;
1777 if (copy_from_user(&pg, upg, sizeof pg))
1780 return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group);
1783 static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg)
1785 struct compat_odp_port_group pg;
1787 if (copy_from_user(&pg, upg, sizeof pg))
1790 return do_get_port_group(dp, compat_ptr(pg.ports), pg.n_ports,
1791 pg.group, &upg->n_ports);
1794 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1796 compat_uptr_t actions;
1798 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1799 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1800 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1801 __get_user(actions, &compat->actions) ||
1802 __get_user(flow->n_actions, &compat->n_actions) ||
1803 __get_user(flow->flags, &compat->flags))
1806 flow->actions = compat_ptr(actions);
1810 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1812 struct odp_flow_stats stats;
1813 struct odp_flow_put fp;
1816 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1817 get_user(fp.flags, &ufp->flags))
1820 error = do_put_flow(dp, &fp, &stats);
1824 if (copy_to_user(&ufp->flow.stats, &stats,
1825 sizeof(struct odp_flow_stats)))
1831 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1832 struct compat_odp_flow __user *ufp)
1834 compat_uptr_t actions;
1836 if (get_user(actions, &ufp->actions))
1839 return do_answer_query(flow, query_flags, &ufp->stats,
1840 compat_ptr(actions), &ufp->n_actions);
1843 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1845 struct sw_flow *flow;
1849 if (compat_get_flow(&uf, ufp))
1852 flow = do_del_flow(dp, &uf.key);
1854 return PTR_ERR(flow);
1856 error = compat_answer_query(flow, 0, ufp);
1857 flow_deferred_free(flow);
1861 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1863 struct tbl *table = rcu_dereference(dp->table);
1866 for (i = 0; i < n_flows; i++) {
1867 struct compat_odp_flow __user *ufp = &flows[i];
1869 struct tbl_node *flow_node;
1872 if (compat_get_flow(&uf, ufp))
1874 memset(uf.key.reserved, 0, sizeof uf.key.reserved);
1876 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1878 error = put_user(ENOENT, &ufp->stats.error);
1880 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1887 struct compat_list_flows_cbdata {
1888 struct compat_odp_flow __user *uflows;
1893 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1895 struct sw_flow *flow = flow_cast(node);
1896 struct compat_list_flows_cbdata *cbdata = cbdata_;
1897 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1900 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1902 error = compat_answer_query(flow, 0, ufp);
1906 if (cbdata->listed_flows >= cbdata->n_flows)
1907 return cbdata->listed_flows;
1911 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1913 struct compat_list_flows_cbdata cbdata;
1919 cbdata.uflows = flows;
1920 cbdata.n_flows = n_flows;
1921 cbdata.listed_flows = 0;
1922 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1923 return error ? error : cbdata.listed_flows;
1926 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1927 int (*function)(struct datapath *,
1928 struct compat_odp_flow *,
1931 struct compat_odp_flowvec __user *uflowvec;
1932 struct compat_odp_flow __user *flows;
1933 struct compat_odp_flowvec flowvec;
1936 uflowvec = compat_ptr(argp);
1937 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1938 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1941 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1944 flows = compat_ptr(flowvec.flows);
1945 if (!access_ok(VERIFY_WRITE, flows,
1946 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1949 retval = function(dp, flows, flowvec.n_flows);
1950 return (retval < 0 ? retval
1951 : retval == flowvec.n_flows ? 0
1952 : put_user(retval, &uflowvec->n_flows));
1955 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1957 struct odp_execute execute;
1958 compat_uptr_t actions;
1961 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1962 __get_user(execute.in_port, &uexecute->in_port) ||
1963 __get_user(actions, &uexecute->actions) ||
1964 __get_user(execute.n_actions, &uexecute->n_actions) ||
1965 __get_user(data, &uexecute->data) ||
1966 __get_user(execute.length, &uexecute->length))
1969 execute.actions = compat_ptr(actions);
1970 execute.data = compat_ptr(data);
1972 return do_execute(dp, &execute);
1975 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1977 int dp_idx = iminor(f->f_dentry->d_inode);
1978 struct datapath *dp;
1982 case ODP_DP_DESTROY:
1983 case ODP_FLOW_FLUSH:
1984 /* Ioctls that don't need any translation at all. */
1985 return openvswitch_ioctl(f, cmd, argp);
1988 case ODP_PORT_ATTACH:
1989 case ODP_PORT_DETACH:
1991 case ODP_VPORT_MTU_SET:
1992 case ODP_VPORT_MTU_GET:
1993 case ODP_VPORT_ETHER_SET:
1994 case ODP_VPORT_ETHER_GET:
1995 case ODP_VPORT_STATS_SET:
1996 case ODP_VPORT_STATS_GET:
1998 case ODP_GET_DROP_FRAGS:
1999 case ODP_SET_DROP_FRAGS:
2000 case ODP_SET_LISTEN_MASK:
2001 case ODP_GET_LISTEN_MASK:
2002 case ODP_SET_SFLOW_PROBABILITY:
2003 case ODP_GET_SFLOW_PROBABILITY:
2004 case ODP_PORT_QUERY:
2005 /* Ioctls that just need their pointer argument extended. */
2006 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
2008 case ODP_VPORT_ADD32:
2009 return compat_vport_user_add(compat_ptr(argp));
2011 case ODP_VPORT_MOD32:
2012 return compat_vport_user_mod(compat_ptr(argp));
2015 dp = get_dp_locked(dp_idx);
2021 case ODP_PORT_LIST32:
2022 err = compat_list_ports(dp, compat_ptr(argp));
2025 case ODP_PORT_GROUP_SET32:
2026 err = compat_set_port_group(dp, compat_ptr(argp));
2029 case ODP_PORT_GROUP_GET32:
2030 err = compat_get_port_group(dp, compat_ptr(argp));
2033 case ODP_FLOW_PUT32:
2034 err = compat_put_flow(dp, compat_ptr(argp));
2037 case ODP_FLOW_DEL32:
2038 err = compat_del_flow(dp, compat_ptr(argp));
2041 case ODP_FLOW_GET32:
2042 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
2045 case ODP_FLOW_LIST32:
2046 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
2050 err = compat_execute(dp, compat_ptr(argp));
2057 mutex_unlock(&dp->mutex);
2063 /* Unfortunately this function is not exported so this is a verbatim copy
2064 * from net/core/datagram.c in 2.6.30. */
2065 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
2066 u8 __user *to, int len,
2069 int start = skb_headlen(skb);
2071 int i, copy = start - offset;
2078 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
2082 if ((len -= copy) == 0)
2089 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2092 WARN_ON(start > offset + len);
2094 end = start + skb_shinfo(skb)->frags[i].size;
2095 if ((copy = end - offset) > 0) {
2099 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2100 struct page *page = frag->page;
2105 csum2 = csum_and_copy_to_user(vaddr +
2112 *csump = csum_block_add(*csump, csum2, pos);
2122 if (skb_shinfo(skb)->frag_list) {
2123 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2125 for (; list; list=list->next) {
2128 WARN_ON(start > offset + len);
2130 end = start + list->len;
2131 if ((copy = end - offset) > 0) {
2135 if (skb_copy_and_csum_datagram(list,
2140 *csump = csum_block_add(*csump, csum2, pos);
2141 if ((len -= copy) == 0)
2157 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2160 /* XXX is there sufficient synchronization here? */
2161 int listeners = get_listen_mask(f);
2162 int dp_idx = iminor(f->f_dentry->d_inode);
2163 struct datapath *dp = get_dp(dp_idx);
2164 struct sk_buff *skb;
2165 size_t copy_bytes, tot_copy_bytes;
2171 if (nbytes == 0 || !listeners)
2177 for (i = 0; i < DP_N_QUEUES; i++) {
2178 if (listeners & (1 << i)) {
2179 skb = skb_dequeue(&dp->queues[i]);
2185 if (f->f_flags & O_NONBLOCK) {
2190 wait_event_interruptible(dp->waitqueue,
2191 dp_has_packet_of_interest(dp,
2194 if (signal_pending(current)) {
2195 retval = -ERESTARTSYS;
2200 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2203 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2205 int csum_start, csum_offset;
2207 csum_start = skb_transport_header(skb) - skb->data;
2208 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2209 csum_offset = skb->csum_offset;
2211 csum_offset = skb->csum;
2213 if (csum_start + csum_offset + sizeof(__sum16) <= copy_bytes) {
2214 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2215 copy_bytes - csum_start, &csum);
2218 __sum16 __user *csump;
2220 copy_bytes = csum_start;
2221 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2222 put_user(csum_fold(csum), csump);
2228 struct iovec __user iov;
2231 iov.iov_len = copy_bytes;
2232 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2236 retval = tot_copy_bytes;
2244 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2246 /* XXX is there sufficient synchronization here? */
2247 int dp_idx = iminor(file->f_dentry->d_inode);
2248 struct datapath *dp = get_dp(dp_idx);
2253 poll_wait(file, &dp->waitqueue, wait);
2254 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2255 mask |= POLLIN | POLLRDNORM;
2257 mask = POLLIN | POLLRDNORM | POLLHUP;
2262 struct file_operations openvswitch_fops = {
2263 /* XXX .aio_read = openvswitch_aio_read, */
2264 .read = openvswitch_read,
2265 .poll = openvswitch_poll,
2266 .unlocked_ioctl = openvswitch_ioctl,
2267 #ifdef CONFIG_COMPAT
2268 .compat_ioctl = openvswitch_compat_ioctl,
2270 /* XXX .fasync = openvswitch_fasync, */
2275 static int __init dp_init(void)
2277 struct sk_buff *dummy_skb;
2280 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2282 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2290 goto error_flow_exit;
2292 err = register_netdevice_notifier(&dp_device_notifier);
2294 goto error_vport_exit;
2296 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2298 goto error_unreg_notifier;
2302 error_unreg_notifier:
2303 unregister_netdevice_notifier(&dp_device_notifier);
2312 static void dp_cleanup(void)
2315 unregister_chrdev(major, "openvswitch");
2316 unregister_netdevice_notifier(&dp_device_notifier);
2321 module_init(dp_init);
2322 module_exit(dp_cleanup);
2324 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2325 MODULE_LICENSE("GPL");