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 <linux/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 /* We limit the number of times that we pass into dp_process_received_packet()
73 * to avoid blowing out the stack in the event that we have a loop. */
75 int count; /* Count. */
76 bool looping; /* Loop detected? */
79 #define DP_MAX_LOOPS 5
81 /* We use a separate counter for each CPU for both interrupt and non-interrupt
82 * context in order to keep the limit deterministic for a given packet. */
83 struct percpu_loop_counters {
84 struct loop_counter counters[2];
87 static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters);
89 static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
91 /* Must be called with rcu_read_lock or dp_mutex. */
92 struct datapath *get_dp(int dp_idx)
94 if (dp_idx < 0 || dp_idx >= ODP_MAX)
96 return rcu_dereference(dps[dp_idx]);
98 EXPORT_SYMBOL_GPL(get_dp);
100 static struct datapath *get_dp_locked(int dp_idx)
104 mutex_lock(&dp_mutex);
107 mutex_lock(&dp->mutex);
108 mutex_unlock(&dp_mutex);
112 /* Must be called with rcu_read_lock or RTNL lock. */
113 const char *dp_name(const struct datapath *dp)
115 return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
118 static inline size_t br_nlmsg_size(void)
120 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
121 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
122 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
123 + nla_total_size(4) /* IFLA_MASTER */
124 + nla_total_size(4) /* IFLA_MTU */
125 + nla_total_size(4) /* IFLA_LINK */
126 + nla_total_size(1); /* IFLA_OPERSTATE */
129 static int dp_fill_ifinfo(struct sk_buff *skb,
130 const struct dp_port *port,
131 int event, unsigned int flags)
133 const struct datapath *dp = port->dp;
134 int ifindex = vport_get_ifindex(port->vport);
135 int iflink = vport_get_iflink(port->vport);
136 struct ifinfomsg *hdr;
137 struct nlmsghdr *nlh;
145 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
149 hdr = nlmsg_data(nlh);
150 hdr->ifi_family = AF_BRIDGE;
152 hdr->ifi_type = ARPHRD_ETHER;
153 hdr->ifi_index = ifindex;
154 hdr->ifi_flags = vport_get_flags(port->vport);
157 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
158 NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
159 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
160 #ifdef IFLA_OPERSTATE
161 NLA_PUT_U8(skb, IFLA_OPERSTATE,
162 vport_is_running(port->vport)
163 ? vport_get_operstate(port->vport)
167 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
168 vport_get_addr(port->vport));
170 if (ifindex != iflink)
171 NLA_PUT_U32(skb, IFLA_LINK,iflink);
173 return nlmsg_end(skb, nlh);
176 nlmsg_cancel(skb, nlh);
180 static void dp_ifinfo_notify(int event, struct dp_port *port)
185 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
189 err = dp_fill_ifinfo(skb, port, event, 0);
191 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
192 WARN_ON(err == -EMSGSIZE);
196 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
200 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
203 static void release_dp(struct kobject *kobj)
205 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
209 static struct kobj_type dp_ktype = {
210 .release = release_dp
213 static int create_dp(int dp_idx, const char __user *devnamep)
215 struct odp_port internal_dev_port;
216 char devname[IFNAMSIZ];
222 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
226 } else if (retval >= IFNAMSIZ) {
231 snprintf(devname, sizeof devname, "of%d", dp_idx);
235 mutex_lock(&dp_mutex);
237 if (!try_module_get(THIS_MODULE))
240 /* Exit early if a datapath with that number already exists.
241 * (We don't use -EEXIST because that's ambiguous with 'devname'
242 * conflicting with an existing network device name.) */
248 dp = kzalloc(sizeof *dp, GFP_KERNEL);
251 INIT_LIST_HEAD(&dp->port_list);
252 mutex_init(&dp->mutex);
254 for (i = 0; i < DP_N_QUEUES; i++)
255 skb_queue_head_init(&dp->queues[i]);
256 init_waitqueue_head(&dp->waitqueue);
258 /* Initialize kobject for bridge. This will be added as
259 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
260 dp->ifobj.kset = NULL;
261 kobject_init(&dp->ifobj, &dp_ktype);
263 /* Allocate table. */
265 rcu_assign_pointer(dp->table, tbl_create(0));
269 /* Set up our datapath device. */
270 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
271 strcpy(internal_dev_port.devname, devname);
272 internal_dev_port.flags = ODP_PORT_INTERNAL;
273 err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
278 goto err_destroy_table;
282 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
283 if (!dp->stats_percpu)
284 goto err_destroy_local_port;
286 rcu_assign_pointer(dps[dp_idx], dp);
287 mutex_unlock(&dp_mutex);
294 err_destroy_local_port:
295 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
297 tbl_destroy(dp->table, NULL);
301 module_put(THIS_MODULE);
303 mutex_unlock(&dp_mutex);
309 static void do_destroy_dp(struct datapath *dp)
311 struct dp_port *p, *n;
314 list_for_each_entry_safe (p, n, &dp->port_list, node)
315 if (p->port_no != ODPP_LOCAL)
316 dp_detach_port(p, 1);
320 rcu_assign_pointer(dps[dp->dp_idx], NULL);
322 dp_detach_port(dp->ports[ODPP_LOCAL], 1);
324 tbl_destroy(dp->table, flow_free_tbl);
326 for (i = 0; i < DP_N_QUEUES; i++)
327 skb_queue_purge(&dp->queues[i]);
328 for (i = 0; i < DP_MAX_GROUPS; i++)
329 kfree(dp->groups[i]);
330 free_percpu(dp->stats_percpu);
331 kobject_put(&dp->ifobj);
332 module_put(THIS_MODULE);
335 static int destroy_dp(int dp_idx)
341 mutex_lock(&dp_mutex);
351 mutex_unlock(&dp_mutex);
356 static void release_dp_port(struct kobject *kobj)
358 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
362 static struct kobj_type brport_ktype = {
364 .sysfs_ops = &brport_sysfs_ops,
366 .release = release_dp_port
369 /* Called with RTNL lock and dp_mutex. */
370 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
376 vport = vport_locate(odp_port->devname);
380 if (odp_port->flags & ODP_PORT_INTERNAL)
381 vport = vport_add(odp_port->devname, "internal", NULL);
383 vport = vport_add(odp_port->devname, "netdev", NULL);
388 return PTR_ERR(vport);
391 p = kzalloc(sizeof(*p), GFP_KERNEL);
395 p->port_no = port_no;
398 atomic_set(&p->sflow_pool, 0);
400 err = vport_attach(vport, p);
406 rcu_assign_pointer(dp->ports[port_no], p);
407 list_add_rcu(&p->node, &dp->port_list);
410 /* Initialize kobject for bridge. This will be added as
411 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
413 kobject_init(&p->kobj, &brport_ktype);
415 dp_ifinfo_notify(RTM_NEWLINK, p);
420 static int attach_port(int dp_idx, struct odp_port __user *portp)
423 struct odp_port port;
428 if (copy_from_user(&port, portp, sizeof port))
430 port.devname[IFNAMSIZ - 1] = '\0';
433 dp = get_dp_locked(dp_idx);
436 goto out_unlock_rtnl;
438 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
439 if (!dp->ports[port_no])
445 err = new_dp_port(dp, &port, port_no);
449 set_internal_devs_mtu(dp);
450 dp_sysfs_add_if(dp->ports[port_no]);
452 err = put_user(port_no, &portp->port);
455 mutex_unlock(&dp->mutex);
462 int dp_detach_port(struct dp_port *p, int may_delete)
464 struct vport *vport = p->vport;
469 if (p->port_no != ODPP_LOCAL)
471 dp_ifinfo_notify(RTM_DELLINK, p);
473 /* First drop references to device. */
475 list_del_rcu(&p->node);
476 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
478 err = vport_detach(vport);
482 /* Then wait until no one is still using it, and destroy it. */
486 const char *port_type = vport_get_type(vport);
488 if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
495 kobject_put(&p->kobj);
500 static int detach_port(int dp_idx, int port_no)
507 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
511 dp = get_dp_locked(dp_idx);
514 goto out_unlock_rtnl;
516 p = dp->ports[port_no];
521 err = dp_detach_port(p, 1);
524 mutex_unlock(&dp->mutex);
531 static void suppress_loop(struct datapath *dp, struct sw_flow_actions *actions)
534 printk(KERN_WARNING "%s: flow looped %d times, dropping\n",
535 dp_name(dp), DP_MAX_LOOPS);
536 actions->n_actions = 0;
539 /* Must be called with rcu_read_lock. */
540 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
542 struct datapath *dp = p->dp;
543 struct dp_stats_percpu *stats;
544 int stats_counter_off;
545 struct odp_flow_key key;
546 struct tbl_node *flow_node;
547 struct sw_flow *flow;
548 struct sw_flow_actions *acts;
549 struct loop_counter *loop;
551 OVS_CB(skb)->dp_port = p;
553 /* Extract flow from 'skb' into 'key'. */
554 if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
555 if (dp->drop_frags) {
557 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
563 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp);
564 if (unlikely(!flow_node)) {
565 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
566 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
570 flow = flow_cast(flow_node);
571 flow_used(flow, skb);
573 acts = rcu_dereference(flow->sf_acts);
575 /* Check whether we've looped too much. */
576 loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()];
577 if (unlikely(++loop->count > DP_MAX_LOOPS))
578 loop->looping = true;
579 if (unlikely(loop->looping)) {
580 suppress_loop(dp, acts);
584 /* Execute actions. */
585 execute_actions(dp, skb, &key, acts->actions, acts->n_actions, GFP_ATOMIC);
586 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
588 /* Check whether sub-actions looped too much. */
589 if (unlikely(loop->looping))
590 suppress_loop(dp, acts);
593 /* Decrement loop counter. */
595 loop->looping = false;
596 put_cpu_var(dp_loop_counters);
599 /* Update datapath statistics. */
601 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
602 (*(u64 *)((u8 *)stats + stats_counter_off))++;
606 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
607 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
608 * can't call this function directly because it isn't exported in all
610 int vswitch_skb_checksum_setup(struct sk_buff *skb)
615 __u16 csum_start, csum_offset;
617 if (!skb->proto_csum_blank)
620 if (skb->protocol != htons(ETH_P_IP))
623 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
627 th = skb_network_header(skb) + 4 * iph->ihl;
629 csum_start = th - skb->head;
630 switch (iph->protocol) {
632 csum_offset = offsetof(struct tcphdr, check);
635 csum_offset = offsetof(struct udphdr, check);
639 printk(KERN_ERR "Attempting to checksum a non-"
640 "TCP/UDP packet, dropping a protocol"
641 " %d packet", iph->protocol);
645 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
648 skb->ip_summed = CHECKSUM_PARTIAL;
649 skb->proto_csum_blank = 0;
651 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
652 skb->csum_start = csum_start;
653 skb->csum_offset = csum_offset;
655 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
656 skb->csum = csum_offset;
664 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
666 /* Types of checksums that we can receive (these all refer to L4 checksums):
667 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
668 * (though not verified) checksum in packet but not in skb->csum. Packets
669 * from the bridge local port will also have this type.
670 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
671 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
672 * a valid skb->csum. Importantly, both contain a full checksum (not
673 * verified) in the packet itself. The only difference is that if the
674 * packet gets to L4 processing on this machine (not in DomU) we won't
675 * have to recompute the checksum to verify. Most hardware devices do not
676 * produce packets with this type, even if they support receive checksum
677 * offloading (they produce type #5).
678 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
679 * be computed if it is sent off box. Unfortunately on earlier kernels,
680 * this case is impossible to distinguish from #2, despite having opposite
681 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
682 * to distinguish the different states.
683 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
684 * generated locally by a Xen DomU and has a partial checksum. If it is
685 * handled on this machine (Dom0 or DomU), then the checksum will not be
686 * computed. If it goes off box, the checksum in the packet needs to be
687 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
688 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
689 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
690 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
691 * full checksum or using a protocol without a checksum. skb->csum is
692 * undefined. This is common from devices with receive checksum
693 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
694 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
696 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
697 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
698 * based on whether it is on the transmit or receive path. After the datapath
699 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
700 * checksum, we will panic. Since we can receive packets with checksums, we
701 * assume that all CHECKSUM_HW packets have checksums and map them to
702 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
703 * packet is processed by the local IP stack, in which case it will need to
704 * be reverified). If we receive a packet with CHECKSUM_HW that really means
705 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
706 * shouldn't be any devices that do this with bridging. */
707 void compute_ip_summed(struct sk_buff *skb, bool xmit)
709 /* For our convenience these defines change repeatedly between kernel
710 * versions, so we can't just copy them over... */
711 switch (skb->ip_summed) {
713 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
715 case CHECKSUM_UNNECESSARY:
716 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
719 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
720 * However, on the receive side we should only get CHECKSUM_PARTIAL
721 * packets from Xen, which uses some special fields to represent this
722 * (see below). Since we can only make one type work, pick the one
723 * that actually happens in practice.
725 * On the transmit side (basically after skb_checksum_setup()
726 * has been run or on internal dev transmit), packets with
727 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
730 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
732 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
736 case CHECKSUM_COMPLETE:
737 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
739 case CHECKSUM_PARTIAL:
740 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
744 printk(KERN_ERR "openvswitch: unknown checksum type %d\n",
746 /* None seems the safest... */
747 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
750 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
751 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
752 * kernels. It should not be set on the transmit path though. */
753 if (skb->proto_csum_blank)
754 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
756 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
760 /* This function closely resembles skb_forward_csum() used by the bridge. It
761 * is slightly different because we are only concerned with bridging and not
762 * other types of forwarding and can get away with slightly more optimal
764 void forward_ip_summed(struct sk_buff *skb)
767 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
768 skb->ip_summed = CHECKSUM_NONE;
772 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
773 * unless we broke up a GSO packet. */
774 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
775 int queue_no, u32 arg)
777 struct sk_buff *nskb;
781 if (OVS_CB(skb)->dp_port)
782 port_no = OVS_CB(skb)->dp_port->port_no;
784 port_no = ODPP_LOCAL;
787 struct odp_msg *header;
792 err = skb_cow(skb, sizeof *header);
796 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
797 header->type = queue_no;
798 header->length = skb->len;
799 header->port = port_no;
800 header->reserved = 0;
802 skb_queue_tail(queue, skb);
810 while ((skb = nskb) != NULL) {
817 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
820 struct dp_stats_percpu *stats;
821 struct sk_buff_head *queue;
824 WARN_ON_ONCE(skb_shared(skb));
825 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
826 queue = &dp->queues[queue_no];
828 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
831 forward_ip_summed(skb);
833 err = vswitch_skb_checksum_setup(skb);
837 /* Break apart GSO packets into their component pieces. Otherwise
838 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
839 if (skb_is_gso(skb)) {
840 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
844 if (unlikely(IS_ERR(skb))) {
849 /* XXX This case might not be possible. It's hard to
850 * tell from the skb_gso_segment() code and comment. */
854 err = queue_control_packets(skb, queue, queue_no, arg);
855 wake_up_interruptible(&dp->waitqueue);
862 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
869 static int flush_flows(struct datapath *dp)
871 struct tbl *old_table = rcu_dereference(dp->table);
872 struct tbl *new_table;
874 new_table = tbl_create(0);
878 rcu_assign_pointer(dp->table, new_table);
880 tbl_deferred_destroy(old_table, flow_free_tbl);
885 static int validate_actions(const struct sw_flow_actions *actions)
889 for (i = 0; i < actions->n_actions; i++) {
890 const union odp_action *a = &actions->actions[i];
893 if (a->output.port >= DP_MAX_PORTS)
897 case ODPAT_OUTPUT_GROUP:
898 if (a->output_group.group >= DP_MAX_GROUPS)
902 case ODPAT_SET_VLAN_VID:
903 if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
907 case ODPAT_SET_VLAN_PCP:
908 if (a->vlan_pcp.vlan_pcp
909 & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
913 case ODPAT_SET_NW_TOS:
914 if (a->nw_tos.nw_tos & INET_ECN_MASK)
919 if (a->type >= ODPAT_N_ACTIONS)
928 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
930 struct sw_flow_actions *actions;
933 actions = flow_actions_alloc(flow->n_actions);
934 error = PTR_ERR(actions);
939 if (copy_from_user(actions->actions, flow->actions,
940 flow->n_actions * sizeof(union odp_action)))
941 goto error_free_actions;
942 error = validate_actions(actions);
944 goto error_free_actions;
951 return ERR_PTR(error);
954 static struct timespec get_time_offset(void)
956 struct timespec now_mono, now_jiffies;
958 ktime_get_ts(&now_mono);
959 jiffies_to_timespec(jiffies, &now_jiffies);
960 return timespec_sub(now_mono, now_jiffies);
963 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats,
964 struct timespec time_offset)
967 struct timespec flow_ts, used;
969 jiffies_to_timespec(flow->used, &flow_ts);
970 set_normalized_timespec(&used, flow_ts.tv_sec + time_offset.tv_sec,
971 flow_ts.tv_nsec + time_offset.tv_nsec);
973 stats->used_sec = used.tv_sec;
974 stats->used_nsec = used.tv_nsec;
977 stats->used_nsec = 0;
980 stats->n_packets = flow->packet_count;
981 stats->n_bytes = flow->byte_count;
983 stats->tcp_flags = flow->tcp_flags;
987 static void clear_stats(struct sw_flow *flow)
991 flow->packet_count = 0;
992 flow->byte_count = 0;
995 static int expand_table(struct datapath *dp)
997 struct tbl *old_table = rcu_dereference(dp->table);
998 struct tbl *new_table;
1000 new_table = tbl_expand(old_table);
1001 if (IS_ERR(new_table))
1002 return PTR_ERR(new_table);
1004 rcu_assign_pointer(dp->table, new_table);
1005 tbl_deferred_destroy(old_table, NULL);
1010 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
1011 struct odp_flow_stats *stats)
1013 struct tbl_node *flow_node;
1014 struct sw_flow *flow;
1018 memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved);
1020 table = rcu_dereference(dp->table);
1021 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
1024 struct sw_flow_actions *acts;
1027 if (!(uf->flags & ODPPF_CREATE))
1030 /* Expand table, if necessary, to make room. */
1031 if (tbl_count(table) >= tbl_n_buckets(table)) {
1032 error = expand_table(dp);
1035 table = rcu_dereference(dp->table);
1038 /* Allocate flow. */
1040 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
1043 flow->key = uf->flow.key;
1044 spin_lock_init(&flow->lock);
1047 /* Obtain actions. */
1048 acts = get_actions(&uf->flow);
1049 error = PTR_ERR(acts);
1051 goto error_free_flow;
1052 rcu_assign_pointer(flow->sf_acts, acts);
1054 /* Put flow in bucket. */
1055 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1057 goto error_free_flow_acts;
1059 memset(stats, 0, sizeof(struct odp_flow_stats));
1061 /* We found a matching flow. */
1062 struct sw_flow_actions *old_acts, *new_acts;
1064 flow = flow_cast(flow_node);
1066 /* Bail out if we're not allowed to modify an existing flow. */
1068 if (!(uf->flags & ODPPF_MODIFY))
1072 new_acts = get_actions(&uf->flow);
1073 error = PTR_ERR(new_acts);
1074 if (IS_ERR(new_acts))
1076 old_acts = rcu_dereference(flow->sf_acts);
1077 if (old_acts->n_actions != new_acts->n_actions ||
1078 memcmp(old_acts->actions, new_acts->actions,
1079 sizeof(union odp_action) * old_acts->n_actions)) {
1080 rcu_assign_pointer(flow->sf_acts, new_acts);
1081 flow_deferred_free_acts(old_acts);
1086 /* Fetch stats, then clear them if necessary. */
1087 spin_lock_bh(&flow->lock);
1088 get_stats(flow, stats, get_time_offset());
1089 if (uf->flags & ODPPF_ZERO_STATS)
1091 spin_unlock_bh(&flow->lock);
1096 error_free_flow_acts:
1097 kfree(flow->sf_acts);
1099 kmem_cache_free(flow_cache, flow);
1104 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1106 struct odp_flow_stats stats;
1107 struct odp_flow_put uf;
1110 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1113 error = do_put_flow(dp, &uf, &stats);
1117 if (copy_to_user(&ufp->flow.stats, &stats,
1118 sizeof(struct odp_flow_stats)))
1124 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1125 struct timespec time_offset,
1126 struct odp_flow_stats __user *ustats,
1127 union odp_action __user *actions,
1128 u32 __user *n_actionsp)
1130 struct sw_flow_actions *sf_acts;
1131 struct odp_flow_stats stats;
1134 spin_lock_bh(&flow->lock);
1135 get_stats(flow, &stats, time_offset);
1136 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1137 flow->tcp_flags = 0;
1139 spin_unlock_bh(&flow->lock);
1141 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1142 get_user(n_actions, n_actionsp))
1148 sf_acts = rcu_dereference(flow->sf_acts);
1149 if (put_user(sf_acts->n_actions, n_actionsp) ||
1150 (actions && copy_to_user(actions, sf_acts->actions,
1151 sizeof(union odp_action) *
1152 min(sf_acts->n_actions, n_actions))))
1158 static int answer_query(struct sw_flow *flow, u32 query_flags,
1159 struct timespec time_offset,
1160 struct odp_flow __user *ufp)
1162 union odp_action *actions;
1164 if (get_user(actions, &ufp->actions))
1167 return do_answer_query(flow, query_flags, time_offset,
1168 &ufp->stats, actions, &ufp->n_actions);
1171 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1173 struct tbl *table = rcu_dereference(dp->table);
1174 struct tbl_node *flow_node;
1177 memset(key->reserved, 0, sizeof key->reserved);
1178 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1180 return ERR_PTR(-ENOENT);
1182 error = tbl_remove(table, flow_node);
1184 return ERR_PTR(error);
1186 /* XXX Returned flow_node's statistics might lose a few packets, since
1187 * other CPUs can be using this flow. We used to synchronize_rcu() to
1188 * make sure that we get completely accurate stats, but that blows our
1189 * performance, badly. */
1190 return flow_cast(flow_node);
1193 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1195 struct sw_flow *flow;
1199 if (copy_from_user(&uf, ufp, sizeof uf))
1202 flow = do_del_flow(dp, &uf.key);
1204 return PTR_ERR(flow);
1206 error = answer_query(flow, 0, get_time_offset(), ufp);
1207 flow_deferred_free(flow);
1211 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1213 struct tbl *table = rcu_dereference(dp->table);
1214 struct timespec time_offset;
1217 time_offset = get_time_offset();
1219 for (i = 0; i < flowvec->n_flows; i++) {
1220 struct odp_flow __user *ufp = &flowvec->flows[i];
1222 struct tbl_node *flow_node;
1225 if (copy_from_user(&uf, ufp, sizeof uf))
1227 memset(uf.key.reserved, 0, sizeof uf.key.reserved);
1229 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1231 error = put_user(ENOENT, &ufp->stats.error);
1233 error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
1237 return flowvec->n_flows;
1240 struct list_flows_cbdata {
1241 struct odp_flow __user *uflows;
1244 struct timespec time_offset;
1247 static int list_flow(struct tbl_node *node, void *cbdata_)
1249 struct sw_flow *flow = flow_cast(node);
1250 struct list_flows_cbdata *cbdata = cbdata_;
1251 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1254 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1256 error = answer_query(flow, 0, cbdata->time_offset, ufp);
1260 if (cbdata->listed_flows >= cbdata->n_flows)
1261 return cbdata->listed_flows;
1265 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1267 struct list_flows_cbdata cbdata;
1270 if (!flowvec->n_flows)
1273 cbdata.uflows = flowvec->flows;
1274 cbdata.n_flows = flowvec->n_flows;
1275 cbdata.listed_flows = 0;
1276 cbdata.time_offset = get_time_offset();
1278 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1279 return error ? error : cbdata.listed_flows;
1282 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1283 int (*function)(struct datapath *,
1284 const struct odp_flowvec *))
1286 struct odp_flowvec __user *uflowvec;
1287 struct odp_flowvec flowvec;
1290 uflowvec = (struct odp_flowvec __user *)argp;
1291 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1294 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1297 retval = function(dp, &flowvec);
1298 return (retval < 0 ? retval
1299 : retval == flowvec.n_flows ? 0
1300 : put_user(retval, &uflowvec->n_flows));
1303 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1305 struct odp_flow_key key;
1306 struct sk_buff *skb;
1307 struct sw_flow_actions *actions;
1312 if (execute->length < ETH_HLEN || execute->length > 65535)
1316 actions = flow_actions_alloc(execute->n_actions);
1321 if (copy_from_user(actions->actions, execute->actions,
1322 execute->n_actions * sizeof *execute->actions))
1323 goto error_free_actions;
1325 err = validate_actions(actions);
1327 goto error_free_actions;
1330 skb = alloc_skb(execute->length, GFP_KERNEL);
1332 goto error_free_actions;
1334 if (execute->in_port < DP_MAX_PORTS)
1335 OVS_CB(skb)->dp_port = dp->ports[execute->in_port];
1337 OVS_CB(skb)->dp_port = NULL;
1340 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1342 goto error_free_skb;
1344 skb_reset_mac_header(skb);
1347 /* Normally, setting the skb 'protocol' field would be handled by a
1348 * call to eth_type_trans(), but it assumes there's a sending
1349 * device, which we may not have. */
1350 if (ntohs(eth->h_proto) >= 1536)
1351 skb->protocol = eth->h_proto;
1353 skb->protocol = htons(ETH_P_802_2);
1355 flow_extract(skb, execute->in_port, &key);
1358 err = execute_actions(dp, skb, &key, actions->actions,
1359 actions->n_actions, GFP_KERNEL);
1373 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1375 struct odp_execute execute;
1377 if (copy_from_user(&execute, executep, sizeof execute))
1380 return do_execute(dp, &execute);
1383 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1385 struct tbl *table = rcu_dereference(dp->table);
1386 struct odp_stats stats;
1389 stats.n_flows = tbl_count(table);
1390 stats.cur_capacity = tbl_n_buckets(table);
1391 stats.max_capacity = TBL_MAX_BUCKETS;
1392 stats.n_ports = dp->n_ports;
1393 stats.max_ports = DP_MAX_PORTS;
1394 stats.max_groups = DP_MAX_GROUPS;
1395 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1396 for_each_possible_cpu(i) {
1397 const struct dp_stats_percpu *s;
1398 s = per_cpu_ptr(dp->stats_percpu, i);
1399 stats.n_frags += s->n_frags;
1400 stats.n_hit += s->n_hit;
1401 stats.n_missed += s->n_missed;
1402 stats.n_lost += s->n_lost;
1404 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1405 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1406 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1409 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1410 int dp_min_mtu(const struct datapath *dp)
1417 list_for_each_entry_rcu (p, &dp->port_list, node) {
1420 /* Skip any internal ports, since that's what we're trying to
1422 if (is_internal_vport(p->vport))
1425 dev_mtu = vport_get_mtu(p->vport);
1426 if (!mtu || dev_mtu < mtu)
1430 return mtu ? mtu : ETH_DATA_LEN;
1433 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1434 * be called with RTNL lock. */
1435 void set_internal_devs_mtu(const struct datapath *dp)
1442 mtu = dp_min_mtu(dp);
1444 list_for_each_entry_rcu (p, &dp->port_list, node) {
1445 if (is_internal_vport(p->vport))
1446 vport_set_mtu(p->vport, mtu);
1450 static int put_port(const struct dp_port *p, struct odp_port __user *uop)
1454 memset(&op, 0, sizeof op);
1457 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1460 op.port = p->port_no;
1461 op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
1463 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1466 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1468 struct odp_port port;
1470 if (copy_from_user(&port, uport, sizeof port))
1473 if (port.devname[0]) {
1474 struct vport *vport;
1475 struct dp_port *dp_port;
1478 port.devname[IFNAMSIZ - 1] = '\0';
1483 vport = vport_locate(port.devname);
1489 dp_port = vport_get_dp_port(vport);
1490 if (!dp_port || dp_port->dp != dp) {
1495 port.port = dp_port->port_no;
1504 if (port.port >= DP_MAX_PORTS)
1506 if (!dp->ports[port.port])
1510 return put_port(dp->ports[port.port], uport);
1513 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1520 list_for_each_entry_rcu (p, &dp->port_list, node) {
1521 if (put_port(p, &uports[idx]))
1523 if (idx++ >= n_ports)
1530 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1532 struct odp_portvec pv;
1535 if (copy_from_user(&pv, upv, sizeof pv))
1538 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1542 return put_user(retval, &upv->n_ports);
1545 /* RCU callback for freeing a dp_port_group */
1546 static void free_port_group(struct rcu_head *rcu)
1548 struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
1552 static int do_set_port_group(struct datapath *dp, u16 __user *ports,
1553 int n_ports, int group)
1555 struct dp_port_group *new_group, *old_group;
1559 if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS)
1563 new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL);
1567 new_group->n_ports = n_ports;
1569 if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports))
1572 old_group = rcu_dereference(dp->groups[group]);
1573 rcu_assign_pointer(dp->groups[group], new_group);
1575 call_rcu(&old_group->rcu, free_port_group);
1584 static int set_port_group(struct datapath *dp,
1585 const struct odp_port_group __user *upg)
1587 struct odp_port_group pg;
1589 if (copy_from_user(&pg, upg, sizeof pg))
1592 return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group);
1595 static int do_get_port_group(struct datapath *dp,
1596 u16 __user *ports, int n_ports, int group,
1597 u16 __user *n_portsp)
1599 struct dp_port_group *g;
1602 if (group >= DP_MAX_GROUPS)
1605 g = dp->groups[group];
1606 n_copy = g ? min_t(int, g->n_ports, n_ports) : 0;
1607 if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16)))
1610 if (put_user(g ? g->n_ports : 0, n_portsp))
1616 static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg)
1618 struct odp_port_group pg;
1620 if (copy_from_user(&pg, upg, sizeof pg))
1623 return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &upg->n_ports);
1626 static int get_listen_mask(const struct file *f)
1628 return (long)f->private_data;
1631 static void set_listen_mask(struct file *f, int listen_mask)
1633 f->private_data = (void*)(long)listen_mask;
1636 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1639 int dp_idx = iminor(f->f_dentry->d_inode);
1640 struct datapath *dp;
1641 int drop_frags, listeners, port_no;
1642 unsigned int sflow_probability;
1645 /* Handle commands with special locking requirements up front. */
1648 err = create_dp(dp_idx, (char __user *)argp);
1651 case ODP_DP_DESTROY:
1652 err = destroy_dp(dp_idx);
1655 case ODP_PORT_ATTACH:
1656 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1659 case ODP_PORT_DETACH:
1660 err = get_user(port_no, (int __user *)argp);
1662 err = detach_port(dp_idx, port_no);
1666 err = vport_user_add((struct odp_vport_add __user *)argp);
1670 err = vport_user_mod((struct odp_vport_mod __user *)argp);
1674 err = vport_user_del((char __user *)argp);
1677 case ODP_VPORT_STATS_GET:
1678 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1681 case ODP_VPORT_STATS_SET:
1682 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1685 case ODP_VPORT_ETHER_GET:
1686 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1689 case ODP_VPORT_ETHER_SET:
1690 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1693 case ODP_VPORT_MTU_GET:
1694 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1697 case ODP_VPORT_MTU_SET:
1698 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1702 dp = get_dp_locked(dp_idx);
1709 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1712 case ODP_GET_DROP_FRAGS:
1713 err = put_user(dp->drop_frags, (int __user *)argp);
1716 case ODP_SET_DROP_FRAGS:
1717 err = get_user(drop_frags, (int __user *)argp);
1721 if (drop_frags != 0 && drop_frags != 1)
1723 dp->drop_frags = drop_frags;
1727 case ODP_GET_LISTEN_MASK:
1728 err = put_user(get_listen_mask(f), (int __user *)argp);
1731 case ODP_SET_LISTEN_MASK:
1732 err = get_user(listeners, (int __user *)argp);
1736 if (listeners & ~ODPL_ALL)
1739 set_listen_mask(f, listeners);
1742 case ODP_GET_SFLOW_PROBABILITY:
1743 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1746 case ODP_SET_SFLOW_PROBABILITY:
1747 err = get_user(sflow_probability, (unsigned int __user *)argp);
1749 dp->sflow_probability = sflow_probability;
1752 case ODP_PORT_QUERY:
1753 err = query_port(dp, (struct odp_port __user *)argp);
1757 err = list_ports(dp, (struct odp_portvec __user *)argp);
1760 case ODP_PORT_GROUP_SET:
1761 err = set_port_group(dp, (struct odp_port_group __user *)argp);
1764 case ODP_PORT_GROUP_GET:
1765 err = get_port_group(dp, (struct odp_port_group __user *)argp);
1768 case ODP_FLOW_FLUSH:
1769 err = flush_flows(dp);
1773 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1777 err = del_flow(dp, (struct odp_flow __user *)argp);
1781 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1785 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1789 err = execute_packet(dp, (struct odp_execute __user *)argp);
1796 mutex_unlock(&dp->mutex);
1801 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1804 for (i = 0; i < DP_N_QUEUES; i++) {
1805 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1811 #ifdef CONFIG_COMPAT
1812 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1814 struct compat_odp_portvec pv;
1817 if (copy_from_user(&pv, upv, sizeof pv))
1820 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1824 return put_user(retval, &upv->n_ports);
1827 static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg)
1829 struct compat_odp_port_group pg;
1831 if (copy_from_user(&pg, upg, sizeof pg))
1834 return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group);
1837 static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg)
1839 struct compat_odp_port_group pg;
1841 if (copy_from_user(&pg, upg, sizeof pg))
1844 return do_get_port_group(dp, compat_ptr(pg.ports), pg.n_ports,
1845 pg.group, &upg->n_ports);
1848 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1850 compat_uptr_t actions;
1852 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1853 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1854 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1855 __get_user(actions, &compat->actions) ||
1856 __get_user(flow->n_actions, &compat->n_actions) ||
1857 __get_user(flow->flags, &compat->flags))
1860 flow->actions = compat_ptr(actions);
1864 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1866 struct odp_flow_stats stats;
1867 struct odp_flow_put fp;
1870 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1871 get_user(fp.flags, &ufp->flags))
1874 error = do_put_flow(dp, &fp, &stats);
1878 if (copy_to_user(&ufp->flow.stats, &stats,
1879 sizeof(struct odp_flow_stats)))
1885 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1886 struct timespec time_offset,
1887 struct compat_odp_flow __user *ufp)
1889 compat_uptr_t actions;
1891 if (get_user(actions, &ufp->actions))
1894 return do_answer_query(flow, query_flags, time_offset, &ufp->stats,
1895 compat_ptr(actions), &ufp->n_actions);
1898 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1900 struct sw_flow *flow;
1904 if (compat_get_flow(&uf, ufp))
1907 flow = do_del_flow(dp, &uf.key);
1909 return PTR_ERR(flow);
1911 error = compat_answer_query(flow, 0, get_time_offset(), ufp);
1912 flow_deferred_free(flow);
1916 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1918 struct tbl *table = rcu_dereference(dp->table);
1919 struct timespec time_offset;
1922 time_offset = get_time_offset();
1924 for (i = 0; i < n_flows; i++) {
1925 struct compat_odp_flow __user *ufp = &flows[i];
1927 struct tbl_node *flow_node;
1930 if (compat_get_flow(&uf, ufp))
1932 memset(uf.key.reserved, 0, sizeof uf.key.reserved);
1934 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1936 error = put_user(ENOENT, &ufp->stats.error);
1938 error = compat_answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
1945 struct compat_list_flows_cbdata {
1946 struct compat_odp_flow __user *uflows;
1949 struct timespec time_offset;
1952 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1954 struct sw_flow *flow = flow_cast(node);
1955 struct compat_list_flows_cbdata *cbdata = cbdata_;
1956 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1959 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1961 error = compat_answer_query(flow, 0, cbdata->time_offset, ufp);
1965 if (cbdata->listed_flows >= cbdata->n_flows)
1966 return cbdata->listed_flows;
1970 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1972 struct compat_list_flows_cbdata cbdata;
1978 cbdata.uflows = flows;
1979 cbdata.n_flows = n_flows;
1980 cbdata.listed_flows = 0;
1981 cbdata.time_offset = get_time_offset();
1983 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1984 return error ? error : cbdata.listed_flows;
1987 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1988 int (*function)(struct datapath *,
1989 struct compat_odp_flow *,
1992 struct compat_odp_flowvec __user *uflowvec;
1993 struct compat_odp_flow __user *flows;
1994 struct compat_odp_flowvec flowvec;
1997 uflowvec = compat_ptr(argp);
1998 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1999 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
2002 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
2005 flows = compat_ptr(flowvec.flows);
2006 if (!access_ok(VERIFY_WRITE, flows,
2007 flowvec.n_flows * sizeof(struct compat_odp_flow)))
2010 retval = function(dp, flows, flowvec.n_flows);
2011 return (retval < 0 ? retval
2012 : retval == flowvec.n_flows ? 0
2013 : put_user(retval, &uflowvec->n_flows));
2016 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
2018 struct odp_execute execute;
2019 compat_uptr_t actions;
2022 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
2023 __get_user(execute.in_port, &uexecute->in_port) ||
2024 __get_user(actions, &uexecute->actions) ||
2025 __get_user(execute.n_actions, &uexecute->n_actions) ||
2026 __get_user(data, &uexecute->data) ||
2027 __get_user(execute.length, &uexecute->length))
2030 execute.actions = compat_ptr(actions);
2031 execute.data = compat_ptr(data);
2033 return do_execute(dp, &execute);
2036 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
2038 int dp_idx = iminor(f->f_dentry->d_inode);
2039 struct datapath *dp;
2043 case ODP_DP_DESTROY:
2044 case ODP_FLOW_FLUSH:
2045 /* Ioctls that don't need any translation at all. */
2046 return openvswitch_ioctl(f, cmd, argp);
2049 case ODP_PORT_ATTACH:
2050 case ODP_PORT_DETACH:
2052 case ODP_VPORT_MTU_SET:
2053 case ODP_VPORT_MTU_GET:
2054 case ODP_VPORT_ETHER_SET:
2055 case ODP_VPORT_ETHER_GET:
2056 case ODP_VPORT_STATS_SET:
2057 case ODP_VPORT_STATS_GET:
2059 case ODP_GET_DROP_FRAGS:
2060 case ODP_SET_DROP_FRAGS:
2061 case ODP_SET_LISTEN_MASK:
2062 case ODP_GET_LISTEN_MASK:
2063 case ODP_SET_SFLOW_PROBABILITY:
2064 case ODP_GET_SFLOW_PROBABILITY:
2065 case ODP_PORT_QUERY:
2066 /* Ioctls that just need their pointer argument extended. */
2067 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
2069 case ODP_VPORT_ADD32:
2070 return compat_vport_user_add(compat_ptr(argp));
2072 case ODP_VPORT_MOD32:
2073 return compat_vport_user_mod(compat_ptr(argp));
2076 dp = get_dp_locked(dp_idx);
2082 case ODP_PORT_LIST32:
2083 err = compat_list_ports(dp, compat_ptr(argp));
2086 case ODP_PORT_GROUP_SET32:
2087 err = compat_set_port_group(dp, compat_ptr(argp));
2090 case ODP_PORT_GROUP_GET32:
2091 err = compat_get_port_group(dp, compat_ptr(argp));
2094 case ODP_FLOW_PUT32:
2095 err = compat_put_flow(dp, compat_ptr(argp));
2098 case ODP_FLOW_DEL32:
2099 err = compat_del_flow(dp, compat_ptr(argp));
2102 case ODP_FLOW_GET32:
2103 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
2106 case ODP_FLOW_LIST32:
2107 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
2111 err = compat_execute(dp, compat_ptr(argp));
2118 mutex_unlock(&dp->mutex);
2124 /* Unfortunately this function is not exported so this is a verbatim copy
2125 * from net/core/datagram.c in 2.6.30. */
2126 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
2127 u8 __user *to, int len,
2130 int start = skb_headlen(skb);
2132 int i, copy = start - offset;
2139 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
2143 if ((len -= copy) == 0)
2150 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2153 WARN_ON(start > offset + len);
2155 end = start + skb_shinfo(skb)->frags[i].size;
2156 if ((copy = end - offset) > 0) {
2160 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2161 struct page *page = frag->page;
2166 csum2 = csum_and_copy_to_user(vaddr +
2173 *csump = csum_block_add(*csump, csum2, pos);
2183 if (skb_shinfo(skb)->frag_list) {
2184 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2186 for (; list; list=list->next) {
2189 WARN_ON(start > offset + len);
2191 end = start + list->len;
2192 if ((copy = end - offset) > 0) {
2196 if (skb_copy_and_csum_datagram(list,
2201 *csump = csum_block_add(*csump, csum2, pos);
2202 if ((len -= copy) == 0)
2218 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2221 /* XXX is there sufficient synchronization here? */
2222 int listeners = get_listen_mask(f);
2223 int dp_idx = iminor(f->f_dentry->d_inode);
2224 struct datapath *dp = get_dp(dp_idx);
2225 struct sk_buff *skb;
2226 size_t copy_bytes, tot_copy_bytes;
2232 if (nbytes == 0 || !listeners)
2238 for (i = 0; i < DP_N_QUEUES; i++) {
2239 if (listeners & (1 << i)) {
2240 skb = skb_dequeue(&dp->queues[i]);
2246 if (f->f_flags & O_NONBLOCK) {
2251 wait_event_interruptible(dp->waitqueue,
2252 dp_has_packet_of_interest(dp,
2255 if (signal_pending(current)) {
2256 retval = -ERESTARTSYS;
2261 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2264 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2265 if (copy_bytes == skb->len) {
2267 unsigned int csum_start, csum_offset;
2269 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2270 csum_start = skb->csum_start - skb_headroom(skb);
2271 csum_offset = skb->csum_offset;
2273 csum_start = skb_transport_header(skb) - skb->data;
2274 csum_offset = skb->csum;
2276 BUG_ON(csum_start >= skb_headlen(skb));
2277 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2278 copy_bytes - csum_start, &csum);
2280 __sum16 __user *csump;
2282 copy_bytes = csum_start;
2283 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2285 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
2286 put_user(csum_fold(csum), csump);
2289 retval = skb_checksum_help(skb);
2293 struct iovec __user iov;
2296 iov.iov_len = copy_bytes;
2297 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2301 retval = tot_copy_bytes;
2309 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2311 /* XXX is there sufficient synchronization here? */
2312 int dp_idx = iminor(file->f_dentry->d_inode);
2313 struct datapath *dp = get_dp(dp_idx);
2318 poll_wait(file, &dp->waitqueue, wait);
2319 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2320 mask |= POLLIN | POLLRDNORM;
2322 mask = POLLIN | POLLRDNORM | POLLHUP;
2327 struct file_operations openvswitch_fops = {
2328 /* XXX .aio_read = openvswitch_aio_read, */
2329 .read = openvswitch_read,
2330 .poll = openvswitch_poll,
2331 .unlocked_ioctl = openvswitch_ioctl,
2332 #ifdef CONFIG_COMPAT
2333 .compat_ioctl = openvswitch_compat_ioctl,
2335 /* XXX .fasync = openvswitch_fasync, */
2340 static int __init dp_init(void)
2342 struct sk_buff *dummy_skb;
2345 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2347 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2355 goto error_flow_exit;
2357 err = register_netdevice_notifier(&dp_device_notifier);
2359 goto error_vport_exit;
2361 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2363 goto error_unreg_notifier;
2367 error_unreg_notifier:
2368 unregister_netdevice_notifier(&dp_device_notifier);
2377 static void dp_cleanup(void)
2380 unregister_chrdev(major, "openvswitch");
2381 unregister_netdevice_notifier(&dp_device_notifier);
2386 module_init(dp_init);
2387 module_exit(dp_cleanup);
2389 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2390 MODULE_LICENSE("GPL");