2 * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
16 #include <linux/if_arp.h>
17 #include <linux/if_vlan.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/kernel.h>
24 #include <linux/kthread.h>
25 #include <linux/mutex.h>
26 #include <linux/percpu.h>
27 #include <linux/rcupdate.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/version.h>
31 #include <linux/ethtool.h>
32 #include <linux/wait.h>
33 #include <asm/system.h>
34 #include <asm/div64.h>
36 #include <linux/highmem.h>
37 #include <linux/netfilter_bridge.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/inetdevice.h>
40 #include <linux/list.h>
41 #include <linux/rculist.h>
42 #include <linux/dmi.h>
43 #include <net/inet_ecn.h>
44 #include <linux/compat.h>
46 #include "openvswitch/datapath-protocol.h"
50 #include "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 free_percpu(dp->stats_percpu);
329 kobject_put(&dp->ifobj);
330 module_put(THIS_MODULE);
333 static int destroy_dp(int dp_idx)
339 mutex_lock(&dp_mutex);
349 mutex_unlock(&dp_mutex);
354 static void release_dp_port(struct kobject *kobj)
356 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
360 static struct kobj_type brport_ktype = {
362 .sysfs_ops = &brport_sysfs_ops,
364 .release = release_dp_port
367 /* Called with RTNL lock and dp_mutex. */
368 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
374 vport = vport_locate(odp_port->devname);
378 if (odp_port->flags & ODP_PORT_INTERNAL)
379 vport = vport_add(odp_port->devname, "internal", NULL);
381 vport = vport_add(odp_port->devname, "netdev", NULL);
386 return PTR_ERR(vport);
389 p = kzalloc(sizeof(*p), GFP_KERNEL);
393 p->port_no = port_no;
396 atomic_set(&p->sflow_pool, 0);
398 err = vport_attach(vport, p);
404 rcu_assign_pointer(dp->ports[port_no], p);
405 list_add_rcu(&p->node, &dp->port_list);
408 /* Initialize kobject for bridge. This will be added as
409 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
411 kobject_init(&p->kobj, &brport_ktype);
413 dp_ifinfo_notify(RTM_NEWLINK, p);
418 static int attach_port(int dp_idx, struct odp_port __user *portp)
421 struct odp_port port;
426 if (copy_from_user(&port, portp, sizeof port))
428 port.devname[IFNAMSIZ - 1] = '\0';
431 dp = get_dp_locked(dp_idx);
434 goto out_unlock_rtnl;
436 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
437 if (!dp->ports[port_no])
443 err = new_dp_port(dp, &port, port_no);
447 set_internal_devs_mtu(dp);
448 dp_sysfs_add_if(dp->ports[port_no]);
450 err = put_user(port_no, &portp->port);
453 mutex_unlock(&dp->mutex);
460 int dp_detach_port(struct dp_port *p, int may_delete)
462 struct vport *vport = p->vport;
467 if (p->port_no != ODPP_LOCAL)
469 dp_ifinfo_notify(RTM_DELLINK, p);
471 /* First drop references to device. */
473 list_del_rcu(&p->node);
474 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
476 err = vport_detach(vport);
480 /* Then wait until no one is still using it, and destroy it. */
484 const char *port_type = vport_get_type(vport);
486 if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
493 kobject_put(&p->kobj);
498 static int detach_port(int dp_idx, int port_no)
505 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
509 dp = get_dp_locked(dp_idx);
512 goto out_unlock_rtnl;
514 p = dp->ports[port_no];
519 err = dp_detach_port(p, 1);
522 mutex_unlock(&dp->mutex);
529 static void suppress_loop(struct datapath *dp, struct sw_flow_actions *actions)
532 pr_warn("%s: flow looped %d times, dropping\n",
533 dp_name(dp), DP_MAX_LOOPS);
534 actions->n_actions = 0;
537 /* Must be called with rcu_read_lock. */
538 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
540 struct datapath *dp = p->dp;
541 struct dp_stats_percpu *stats;
542 int stats_counter_off;
543 struct sw_flow_actions *acts;
544 struct loop_counter *loop;
547 OVS_CB(skb)->dp_port = p;
549 if (!OVS_CB(skb)->flow) {
550 struct odp_flow_key key;
551 struct tbl_node *flow_node;
554 /* Extract flow from 'skb' into 'key'. */
555 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
556 if (unlikely(error)) {
561 if (is_frag && dp->drop_frags) {
563 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
568 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
569 flow_hash(&key), flow_cmp);
570 if (unlikely(!flow_node)) {
571 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
572 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
576 OVS_CB(skb)->flow = flow_cast(flow_node);
579 flow_used(OVS_CB(skb)->flow, skb);
581 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
583 /* Check whether we've looped too much. */
584 loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()];
585 if (unlikely(++loop->count > DP_MAX_LOOPS))
586 loop->looping = true;
587 if (unlikely(loop->looping)) {
588 suppress_loop(dp, acts);
592 /* Execute actions. */
593 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
595 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
597 /* Check whether sub-actions looped too much. */
598 if (unlikely(loop->looping))
599 suppress_loop(dp, acts);
602 /* Decrement loop counter. */
604 loop->looping = false;
605 put_cpu_var(dp_loop_counters);
608 /* Update datapath statistics. */
610 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
612 write_seqcount_begin(&stats->seqlock);
613 (*(u64 *)((u8 *)stats + stats_counter_off))++;
614 write_seqcount_end(&stats->seqlock);
619 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
620 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
621 * can't call this function directly because it isn't exported in all
623 int vswitch_skb_checksum_setup(struct sk_buff *skb)
628 __u16 csum_start, csum_offset;
630 if (!skb->proto_csum_blank)
633 if (skb->protocol != htons(ETH_P_IP))
636 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
640 th = skb_network_header(skb) + 4 * iph->ihl;
642 csum_start = th - skb->head;
643 switch (iph->protocol) {
645 csum_offset = offsetof(struct tcphdr, check);
648 csum_offset = offsetof(struct udphdr, check);
652 pr_err("Attempting to checksum a non-TCP/UDP packet, "
653 "dropping a protocol %d packet",
658 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
661 skb->ip_summed = CHECKSUM_PARTIAL;
662 skb->proto_csum_blank = 0;
664 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
665 skb->csum_start = csum_start;
666 skb->csum_offset = csum_offset;
668 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
669 skb->csum = csum_offset;
677 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
679 /* Types of checksums that we can receive (these all refer to L4 checksums):
680 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
681 * (though not verified) checksum in packet but not in skb->csum. Packets
682 * from the bridge local port will also have this type.
683 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
684 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
685 * a valid skb->csum. Importantly, both contain a full checksum (not
686 * verified) in the packet itself. The only difference is that if the
687 * packet gets to L4 processing on this machine (not in DomU) we won't
688 * have to recompute the checksum to verify. Most hardware devices do not
689 * produce packets with this type, even if they support receive checksum
690 * offloading (they produce type #5).
691 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
692 * be computed if it is sent off box. Unfortunately on earlier kernels,
693 * this case is impossible to distinguish from #2, despite having opposite
694 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
695 * to distinguish the different states.
696 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
697 * generated locally by a Xen DomU and has a partial checksum. If it is
698 * handled on this machine (Dom0 or DomU), then the checksum will not be
699 * computed. If it goes off box, the checksum in the packet needs to be
700 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
701 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
702 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
703 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
704 * full checksum or using a protocol without a checksum. skb->csum is
705 * undefined. This is common from devices with receive checksum
706 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
707 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
709 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
710 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
711 * based on whether it is on the transmit or receive path. After the datapath
712 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
713 * checksum, we will panic. Since we can receive packets with checksums, we
714 * assume that all CHECKSUM_HW packets have checksums and map them to
715 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
716 * packet is processed by the local IP stack, in which case it will need to
717 * be reverified). If we receive a packet with CHECKSUM_HW that really means
718 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
719 * shouldn't be any devices that do this with bridging. */
720 void compute_ip_summed(struct sk_buff *skb, bool xmit)
722 /* For our convenience these defines change repeatedly between kernel
723 * versions, so we can't just copy them over... */
724 switch (skb->ip_summed) {
726 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
728 case CHECKSUM_UNNECESSARY:
729 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
732 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
733 * However, on the receive side we should only get CHECKSUM_PARTIAL
734 * packets from Xen, which uses some special fields to represent this
735 * (see below). Since we can only make one type work, pick the one
736 * that actually happens in practice.
738 * On the transmit side (basically after skb_checksum_setup()
739 * has been run or on internal dev transmit), packets with
740 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
743 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
745 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
749 case CHECKSUM_COMPLETE:
750 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
752 case CHECKSUM_PARTIAL:
753 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
757 pr_err("unknown checksum type %d\n", skb->ip_summed);
758 /* None seems the safest... */
759 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
762 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
763 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
764 * kernels. It should not be set on the transmit path though. */
765 if (skb->proto_csum_blank)
766 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
768 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
772 /* This function closely resembles skb_forward_csum() used by the bridge. It
773 * is slightly different because we are only concerned with bridging and not
774 * other types of forwarding and can get away with slightly more optimal
776 void forward_ip_summed(struct sk_buff *skb)
779 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
780 skb->ip_summed = CHECKSUM_NONE;
784 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
785 * unless we broke up a GSO packet. */
786 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
787 int queue_no, u32 arg)
789 struct sk_buff *nskb;
793 if (OVS_CB(skb)->dp_port)
794 port_no = OVS_CB(skb)->dp_port->port_no;
796 port_no = ODPP_LOCAL;
799 struct odp_msg *header;
804 err = skb_cow(skb, sizeof *header);
808 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
809 header->type = queue_no;
810 header->length = skb->len;
811 header->port = port_no;
812 header->reserved = 0;
814 skb_queue_tail(queue, skb);
822 while ((skb = nskb) != NULL) {
829 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
832 struct dp_stats_percpu *stats;
833 struct sk_buff_head *queue;
836 WARN_ON_ONCE(skb_shared(skb));
837 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
838 queue = &dp->queues[queue_no];
840 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
843 forward_ip_summed(skb);
845 err = vswitch_skb_checksum_setup(skb);
849 /* Break apart GSO packets into their component pieces. Otherwise
850 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
851 if (skb_is_gso(skb)) {
852 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
856 if (unlikely(IS_ERR(skb))) {
861 /* XXX This case might not be possible. It's hard to
862 * tell from the skb_gso_segment() code and comment. */
866 err = queue_control_packets(skb, queue, queue_no, arg);
867 wake_up_interruptible(&dp->waitqueue);
874 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
876 write_seqcount_begin(&stats->seqlock);
878 write_seqcount_end(&stats->seqlock);
885 static int flush_flows(struct datapath *dp)
887 struct tbl *old_table = rcu_dereference(dp->table);
888 struct tbl *new_table;
890 new_table = tbl_create(0);
894 rcu_assign_pointer(dp->table, new_table);
896 tbl_deferred_destroy(old_table, flow_free_tbl);
901 static int validate_actions(const struct sw_flow_actions *actions)
905 for (i = 0; i < actions->n_actions; i++) {
906 const union odp_action *a = &actions->actions[i];
909 case ODPAT_CONTROLLER:
910 case ODPAT_STRIP_VLAN:
911 case ODPAT_SET_DL_SRC:
912 case ODPAT_SET_DL_DST:
913 case ODPAT_SET_NW_SRC:
914 case ODPAT_SET_NW_DST:
915 case ODPAT_SET_TP_SRC:
916 case ODPAT_SET_TP_DST:
917 case ODPAT_SET_TUNNEL:
918 case ODPAT_SET_PRIORITY:
919 case ODPAT_POP_PRIORITY:
920 case ODPAT_DROP_SPOOFED_ARP:
921 /* No validation needed. */
925 if (a->output.port >= DP_MAX_PORTS)
929 case ODPAT_SET_DL_TCI:
930 if (a->dl_tci.tci & htons(VLAN_CFI_MASK))
934 case ODPAT_SET_NW_TOS:
935 if (a->nw_tos.nw_tos & INET_ECN_MASK)
947 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
949 struct sw_flow_actions *actions;
952 actions = flow_actions_alloc(flow->n_actions);
953 error = PTR_ERR(actions);
958 if (copy_from_user(actions->actions, flow->actions,
959 flow->n_actions * sizeof(union odp_action)))
960 goto error_free_actions;
961 error = validate_actions(actions);
963 goto error_free_actions;
970 return ERR_PTR(error);
973 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
976 struct timespec offset_ts, used, now_mono;
978 ktime_get_ts(&now_mono);
979 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
980 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
981 now_mono.tv_nsec - offset_ts.tv_nsec);
983 stats->used_sec = used.tv_sec;
984 stats->used_nsec = used.tv_nsec;
987 stats->used_nsec = 0;
990 stats->n_packets = flow->packet_count;
991 stats->n_bytes = flow->byte_count;
993 stats->tcp_flags = flow->tcp_flags;
997 static void clear_stats(struct sw_flow *flow)
1000 flow->tcp_flags = 0;
1001 flow->packet_count = 0;
1002 flow->byte_count = 0;
1005 static int expand_table(struct datapath *dp)
1007 struct tbl *old_table = rcu_dereference(dp->table);
1008 struct tbl *new_table;
1010 new_table = tbl_expand(old_table);
1011 if (IS_ERR(new_table))
1012 return PTR_ERR(new_table);
1014 rcu_assign_pointer(dp->table, new_table);
1015 tbl_deferred_destroy(old_table, NULL);
1020 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
1021 struct odp_flow_stats *stats)
1023 struct tbl_node *flow_node;
1024 struct sw_flow *flow;
1028 table = rcu_dereference(dp->table);
1029 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
1032 struct sw_flow_actions *acts;
1035 if (!(uf->flags & ODPPF_CREATE))
1038 /* Expand table, if necessary, to make room. */
1039 if (tbl_count(table) >= tbl_n_buckets(table)) {
1040 error = expand_table(dp);
1043 table = rcu_dereference(dp->table);
1046 /* Allocate flow. */
1047 flow = flow_alloc();
1049 error = PTR_ERR(flow);
1052 flow->key = uf->flow.key;
1055 /* Obtain actions. */
1056 acts = get_actions(&uf->flow);
1057 error = PTR_ERR(acts);
1059 goto error_free_flow;
1060 rcu_assign_pointer(flow->sf_acts, acts);
1062 /* Put flow in bucket. */
1063 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1065 goto error_free_flow_acts;
1067 memset(stats, 0, sizeof(struct odp_flow_stats));
1069 /* We found a matching flow. */
1070 struct sw_flow_actions *old_acts, *new_acts;
1072 flow = flow_cast(flow_node);
1074 /* Bail out if we're not allowed to modify an existing flow. */
1076 if (!(uf->flags & ODPPF_MODIFY))
1080 new_acts = get_actions(&uf->flow);
1081 error = PTR_ERR(new_acts);
1082 if (IS_ERR(new_acts))
1084 old_acts = rcu_dereference(flow->sf_acts);
1085 if (old_acts->n_actions != new_acts->n_actions ||
1086 memcmp(old_acts->actions, new_acts->actions,
1087 sizeof(union odp_action) * old_acts->n_actions)) {
1088 rcu_assign_pointer(flow->sf_acts, new_acts);
1089 flow_deferred_free_acts(old_acts);
1094 /* Fetch stats, then clear them if necessary. */
1095 spin_lock_bh(&flow->lock);
1096 get_stats(flow, stats);
1097 if (uf->flags & ODPPF_ZERO_STATS)
1099 spin_unlock_bh(&flow->lock);
1104 error_free_flow_acts:
1105 kfree(flow->sf_acts);
1107 flow->sf_acts = NULL;
1113 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1115 struct odp_flow_stats stats;
1116 struct odp_flow_put uf;
1119 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1122 error = do_put_flow(dp, &uf, &stats);
1126 if (copy_to_user(&ufp->flow.stats, &stats,
1127 sizeof(struct odp_flow_stats)))
1133 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1134 struct odp_flow_stats __user *ustats,
1135 union odp_action __user *actions,
1136 u32 __user *n_actionsp)
1138 struct sw_flow_actions *sf_acts;
1139 struct odp_flow_stats stats;
1142 spin_lock_bh(&flow->lock);
1143 get_stats(flow, &stats);
1144 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1145 flow->tcp_flags = 0;
1147 spin_unlock_bh(&flow->lock);
1149 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1150 get_user(n_actions, n_actionsp))
1156 sf_acts = rcu_dereference(flow->sf_acts);
1157 if (put_user(sf_acts->n_actions, n_actionsp) ||
1158 (actions && copy_to_user(actions, sf_acts->actions,
1159 sizeof(union odp_action) *
1160 min(sf_acts->n_actions, n_actions))))
1166 static int answer_query(struct sw_flow *flow, u32 query_flags,
1167 struct odp_flow __user *ufp)
1169 union odp_action *actions;
1171 if (get_user(actions, &ufp->actions))
1174 return do_answer_query(flow, query_flags,
1175 &ufp->stats, actions, &ufp->n_actions);
1178 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1180 struct tbl *table = rcu_dereference(dp->table);
1181 struct tbl_node *flow_node;
1184 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1186 return ERR_PTR(-ENOENT);
1188 error = tbl_remove(table, flow_node);
1190 return ERR_PTR(error);
1192 /* XXX Returned flow_node's statistics might lose a few packets, since
1193 * other CPUs can be using this flow. We used to synchronize_rcu() to
1194 * make sure that we get completely accurate stats, but that blows our
1195 * performance, badly. */
1196 return flow_cast(flow_node);
1199 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1201 struct sw_flow *flow;
1205 if (copy_from_user(&uf, ufp, sizeof uf))
1208 flow = do_del_flow(dp, &uf.key);
1210 return PTR_ERR(flow);
1212 error = answer_query(flow, 0, ufp);
1213 flow_deferred_free(flow);
1217 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1219 struct tbl *table = rcu_dereference(dp->table);
1222 for (i = 0; i < flowvec->n_flows; i++) {
1223 struct odp_flow __user *ufp = &flowvec->flows[i];
1225 struct tbl_node *flow_node;
1228 if (copy_from_user(&uf, ufp, sizeof uf))
1231 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1233 error = put_user(ENOENT, &ufp->stats.error);
1235 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1239 return flowvec->n_flows;
1242 struct list_flows_cbdata {
1243 struct odp_flow __user *uflows;
1248 static int list_flow(struct tbl_node *node, void *cbdata_)
1250 struct sw_flow *flow = flow_cast(node);
1251 struct list_flows_cbdata *cbdata = cbdata_;
1252 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1255 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1257 error = answer_query(flow, 0, ufp);
1261 if (cbdata->listed_flows >= cbdata->n_flows)
1262 return cbdata->listed_flows;
1266 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1268 struct list_flows_cbdata cbdata;
1271 if (!flowvec->n_flows)
1274 cbdata.uflows = flowvec->flows;
1275 cbdata.n_flows = flowvec->n_flows;
1276 cbdata.listed_flows = 0;
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;
1313 if (execute->length < ETH_HLEN || execute->length > 65535)
1316 actions = flow_actions_alloc(execute->n_actions);
1317 if (IS_ERR(actions)) {
1318 err = PTR_ERR(actions);
1323 if (copy_from_user(actions->actions, execute->actions,
1324 execute->n_actions * sizeof *execute->actions))
1325 goto error_free_actions;
1327 err = validate_actions(actions);
1329 goto error_free_actions;
1332 skb = alloc_skb(execute->length, GFP_KERNEL);
1334 goto error_free_actions;
1337 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1339 goto error_free_skb;
1341 skb_reset_mac_header(skb);
1344 /* Normally, setting the skb 'protocol' field would be handled by a
1345 * call to eth_type_trans(), but it assumes there's a sending
1346 * device, which we may not have. */
1347 if (ntohs(eth->h_proto) >= 1536)
1348 skb->protocol = eth->h_proto;
1350 skb->protocol = htons(ETH_P_802_2);
1352 err = flow_extract(skb, -1, &key, &is_frag);
1354 goto error_free_skb;
1357 err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions);
1371 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1373 struct odp_execute execute;
1375 if (copy_from_user(&execute, executep, sizeof execute))
1378 return do_execute(dp, &execute);
1381 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1383 struct tbl *table = rcu_dereference(dp->table);
1384 struct odp_stats stats;
1387 stats.n_flows = tbl_count(table);
1388 stats.cur_capacity = tbl_n_buckets(table);
1389 stats.max_capacity = TBL_MAX_BUCKETS;
1390 stats.n_ports = dp->n_ports;
1391 stats.max_ports = DP_MAX_PORTS;
1392 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1393 for_each_possible_cpu(i) {
1394 const struct dp_stats_percpu *percpu_stats;
1395 struct dp_stats_percpu local_stats;
1398 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1401 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1402 local_stats = *percpu_stats;
1403 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1405 stats.n_frags += local_stats.n_frags;
1406 stats.n_hit += local_stats.n_hit;
1407 stats.n_missed += local_stats.n_missed;
1408 stats.n_lost += local_stats.n_lost;
1410 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1411 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1412 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1415 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1416 int dp_min_mtu(const struct datapath *dp)
1423 list_for_each_entry_rcu (p, &dp->port_list, node) {
1426 /* Skip any internal ports, since that's what we're trying to
1428 if (is_internal_vport(p->vport))
1431 dev_mtu = vport_get_mtu(p->vport);
1432 if (!mtu || dev_mtu < mtu)
1436 return mtu ? mtu : ETH_DATA_LEN;
1439 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1440 * be called with RTNL lock. */
1441 void set_internal_devs_mtu(const struct datapath *dp)
1448 mtu = dp_min_mtu(dp);
1450 list_for_each_entry_rcu (p, &dp->port_list, node) {
1451 if (is_internal_vport(p->vport))
1452 vport_set_mtu(p->vport, mtu);
1456 static int put_port(const struct dp_port *p, struct odp_port __user *uop)
1460 memset(&op, 0, sizeof op);
1463 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1466 op.port = p->port_no;
1467 op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
1469 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1472 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1474 struct odp_port port;
1476 if (copy_from_user(&port, uport, sizeof port))
1479 if (port.devname[0]) {
1480 struct vport *vport;
1481 struct dp_port *dp_port;
1484 port.devname[IFNAMSIZ - 1] = '\0';
1489 vport = vport_locate(port.devname);
1495 dp_port = vport_get_dp_port(vport);
1496 if (!dp_port || dp_port->dp != dp) {
1501 port.port = dp_port->port_no;
1510 if (port.port >= DP_MAX_PORTS)
1512 if (!dp->ports[port.port])
1516 return put_port(dp->ports[port.port], uport);
1519 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1526 list_for_each_entry_rcu (p, &dp->port_list, node) {
1527 if (put_port(p, &uports[idx]))
1529 if (idx++ >= n_ports)
1536 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1538 struct odp_portvec pv;
1541 if (copy_from_user(&pv, upv, sizeof pv))
1544 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1548 return put_user(retval, &upv->n_ports);
1551 static int get_listen_mask(const struct file *f)
1553 return (long)f->private_data;
1556 static void set_listen_mask(struct file *f, int listen_mask)
1558 f->private_data = (void*)(long)listen_mask;
1561 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1564 int dp_idx = iminor(f->f_dentry->d_inode);
1565 struct datapath *dp;
1566 int drop_frags, listeners, port_no;
1567 unsigned int sflow_probability;
1570 /* Handle commands with special locking requirements up front. */
1573 err = create_dp(dp_idx, (char __user *)argp);
1576 case ODP_DP_DESTROY:
1577 err = destroy_dp(dp_idx);
1580 case ODP_PORT_ATTACH:
1581 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1584 case ODP_PORT_DETACH:
1585 err = get_user(port_no, (int __user *)argp);
1587 err = detach_port(dp_idx, port_no);
1591 err = vport_user_add((struct odp_vport_add __user *)argp);
1595 err = vport_user_mod((struct odp_vport_mod __user *)argp);
1599 err = vport_user_del((char __user *)argp);
1602 case ODP_VPORT_STATS_GET:
1603 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1606 case ODP_VPORT_STATS_SET:
1607 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1610 case ODP_VPORT_ETHER_GET:
1611 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1614 case ODP_VPORT_ETHER_SET:
1615 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1618 case ODP_VPORT_MTU_GET:
1619 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1622 case ODP_VPORT_MTU_SET:
1623 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1627 dp = get_dp_locked(dp_idx);
1634 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1637 case ODP_GET_DROP_FRAGS:
1638 err = put_user(dp->drop_frags, (int __user *)argp);
1641 case ODP_SET_DROP_FRAGS:
1642 err = get_user(drop_frags, (int __user *)argp);
1646 if (drop_frags != 0 && drop_frags != 1)
1648 dp->drop_frags = drop_frags;
1652 case ODP_GET_LISTEN_MASK:
1653 err = put_user(get_listen_mask(f), (int __user *)argp);
1656 case ODP_SET_LISTEN_MASK:
1657 err = get_user(listeners, (int __user *)argp);
1661 if (listeners & ~ODPL_ALL)
1664 set_listen_mask(f, listeners);
1667 case ODP_GET_SFLOW_PROBABILITY:
1668 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1671 case ODP_SET_SFLOW_PROBABILITY:
1672 err = get_user(sflow_probability, (unsigned int __user *)argp);
1674 dp->sflow_probability = sflow_probability;
1677 case ODP_PORT_QUERY:
1678 err = query_port(dp, (struct odp_port __user *)argp);
1682 err = list_ports(dp, (struct odp_portvec __user *)argp);
1685 case ODP_FLOW_FLUSH:
1686 err = flush_flows(dp);
1690 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1694 err = del_flow(dp, (struct odp_flow __user *)argp);
1698 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1702 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1706 err = execute_packet(dp, (struct odp_execute __user *)argp);
1713 mutex_unlock(&dp->mutex);
1718 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1721 for (i = 0; i < DP_N_QUEUES; i++) {
1722 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1728 #ifdef CONFIG_COMPAT
1729 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1731 struct compat_odp_portvec pv;
1734 if (copy_from_user(&pv, upv, sizeof pv))
1737 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1741 return put_user(retval, &upv->n_ports);
1744 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1746 compat_uptr_t actions;
1748 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1749 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1750 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1751 __get_user(actions, &compat->actions) ||
1752 __get_user(flow->n_actions, &compat->n_actions) ||
1753 __get_user(flow->flags, &compat->flags))
1756 flow->actions = compat_ptr(actions);
1760 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1762 struct odp_flow_stats stats;
1763 struct odp_flow_put fp;
1766 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1767 get_user(fp.flags, &ufp->flags))
1770 error = do_put_flow(dp, &fp, &stats);
1774 if (copy_to_user(&ufp->flow.stats, &stats,
1775 sizeof(struct odp_flow_stats)))
1781 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1782 struct compat_odp_flow __user *ufp)
1784 compat_uptr_t actions;
1786 if (get_user(actions, &ufp->actions))
1789 return do_answer_query(flow, query_flags, &ufp->stats,
1790 compat_ptr(actions), &ufp->n_actions);
1793 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1795 struct sw_flow *flow;
1799 if (compat_get_flow(&uf, ufp))
1802 flow = do_del_flow(dp, &uf.key);
1804 return PTR_ERR(flow);
1806 error = compat_answer_query(flow, 0, ufp);
1807 flow_deferred_free(flow);
1811 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1813 struct tbl *table = rcu_dereference(dp->table);
1816 for (i = 0; i < n_flows; i++) {
1817 struct compat_odp_flow __user *ufp = &flows[i];
1819 struct tbl_node *flow_node;
1822 if (compat_get_flow(&uf, ufp))
1825 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1827 error = put_user(ENOENT, &ufp->stats.error);
1829 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1836 struct compat_list_flows_cbdata {
1837 struct compat_odp_flow __user *uflows;
1842 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1844 struct sw_flow *flow = flow_cast(node);
1845 struct compat_list_flows_cbdata *cbdata = cbdata_;
1846 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1849 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1851 error = compat_answer_query(flow, 0, ufp);
1855 if (cbdata->listed_flows >= cbdata->n_flows)
1856 return cbdata->listed_flows;
1860 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1862 struct compat_list_flows_cbdata cbdata;
1868 cbdata.uflows = flows;
1869 cbdata.n_flows = n_flows;
1870 cbdata.listed_flows = 0;
1872 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1873 return error ? error : cbdata.listed_flows;
1876 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1877 int (*function)(struct datapath *,
1878 struct compat_odp_flow *,
1881 struct compat_odp_flowvec __user *uflowvec;
1882 struct compat_odp_flow __user *flows;
1883 struct compat_odp_flowvec flowvec;
1886 uflowvec = compat_ptr(argp);
1887 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1888 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1891 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1894 flows = compat_ptr(flowvec.flows);
1895 if (!access_ok(VERIFY_WRITE, flows,
1896 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1899 retval = function(dp, flows, flowvec.n_flows);
1900 return (retval < 0 ? retval
1901 : retval == flowvec.n_flows ? 0
1902 : put_user(retval, &uflowvec->n_flows));
1905 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1907 struct odp_execute execute;
1908 compat_uptr_t actions;
1911 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1912 __get_user(actions, &uexecute->actions) ||
1913 __get_user(execute.n_actions, &uexecute->n_actions) ||
1914 __get_user(data, &uexecute->data) ||
1915 __get_user(execute.length, &uexecute->length))
1918 execute.actions = compat_ptr(actions);
1919 execute.data = compat_ptr(data);
1921 return do_execute(dp, &execute);
1924 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1926 int dp_idx = iminor(f->f_dentry->d_inode);
1927 struct datapath *dp;
1931 case ODP_DP_DESTROY:
1932 case ODP_FLOW_FLUSH:
1933 /* Ioctls that don't need any translation at all. */
1934 return openvswitch_ioctl(f, cmd, argp);
1937 case ODP_PORT_ATTACH:
1938 case ODP_PORT_DETACH:
1940 case ODP_VPORT_MTU_SET:
1941 case ODP_VPORT_MTU_GET:
1942 case ODP_VPORT_ETHER_SET:
1943 case ODP_VPORT_ETHER_GET:
1944 case ODP_VPORT_STATS_SET:
1945 case ODP_VPORT_STATS_GET:
1947 case ODP_GET_DROP_FRAGS:
1948 case ODP_SET_DROP_FRAGS:
1949 case ODP_SET_LISTEN_MASK:
1950 case ODP_GET_LISTEN_MASK:
1951 case ODP_SET_SFLOW_PROBABILITY:
1952 case ODP_GET_SFLOW_PROBABILITY:
1953 case ODP_PORT_QUERY:
1954 /* Ioctls that just need their pointer argument extended. */
1955 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1957 case ODP_VPORT_ADD32:
1958 return compat_vport_user_add(compat_ptr(argp));
1960 case ODP_VPORT_MOD32:
1961 return compat_vport_user_mod(compat_ptr(argp));
1964 dp = get_dp_locked(dp_idx);
1970 case ODP_PORT_LIST32:
1971 err = compat_list_ports(dp, compat_ptr(argp));
1974 case ODP_FLOW_PUT32:
1975 err = compat_put_flow(dp, compat_ptr(argp));
1978 case ODP_FLOW_DEL32:
1979 err = compat_del_flow(dp, compat_ptr(argp));
1982 case ODP_FLOW_GET32:
1983 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1986 case ODP_FLOW_LIST32:
1987 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1991 err = compat_execute(dp, compat_ptr(argp));
1998 mutex_unlock(&dp->mutex);
2004 /* Unfortunately this function is not exported so this is a verbatim copy
2005 * from net/core/datagram.c in 2.6.30. */
2006 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
2007 u8 __user *to, int len,
2010 int start = skb_headlen(skb);
2012 int i, copy = start - offset;
2019 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
2023 if ((len -= copy) == 0)
2030 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2033 WARN_ON(start > offset + len);
2035 end = start + skb_shinfo(skb)->frags[i].size;
2036 if ((copy = end - offset) > 0) {
2040 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2041 struct page *page = frag->page;
2046 csum2 = csum_and_copy_to_user(vaddr +
2053 *csump = csum_block_add(*csump, csum2, pos);
2063 if (skb_shinfo(skb)->frag_list) {
2064 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2066 for (; list; list=list->next) {
2069 WARN_ON(start > offset + len);
2071 end = start + list->len;
2072 if ((copy = end - offset) > 0) {
2076 if (skb_copy_and_csum_datagram(list,
2081 *csump = csum_block_add(*csump, csum2, pos);
2082 if ((len -= copy) == 0)
2098 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2101 /* XXX is there sufficient synchronization here? */
2102 int listeners = get_listen_mask(f);
2103 int dp_idx = iminor(f->f_dentry->d_inode);
2104 struct datapath *dp = get_dp(dp_idx);
2105 struct sk_buff *skb;
2106 size_t copy_bytes, tot_copy_bytes;
2112 if (nbytes == 0 || !listeners)
2118 for (i = 0; i < DP_N_QUEUES; i++) {
2119 if (listeners & (1 << i)) {
2120 skb = skb_dequeue(&dp->queues[i]);
2126 if (f->f_flags & O_NONBLOCK) {
2131 wait_event_interruptible(dp->waitqueue,
2132 dp_has_packet_of_interest(dp,
2135 if (signal_pending(current)) {
2136 retval = -ERESTARTSYS;
2141 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2144 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2145 if (copy_bytes == skb->len) {
2147 unsigned int csum_start, csum_offset;
2149 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2150 csum_start = skb->csum_start - skb_headroom(skb);
2151 csum_offset = skb->csum_offset;
2153 csum_start = skb_transport_header(skb) - skb->data;
2154 csum_offset = skb->csum;
2156 BUG_ON(csum_start >= skb_headlen(skb));
2157 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2158 copy_bytes - csum_start, &csum);
2160 __sum16 __user *csump;
2162 copy_bytes = csum_start;
2163 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2165 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
2166 put_user(csum_fold(csum), csump);
2169 retval = skb_checksum_help(skb);
2173 struct iovec __user iov;
2176 iov.iov_len = copy_bytes;
2177 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2181 retval = tot_copy_bytes;
2189 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2191 /* XXX is there sufficient synchronization here? */
2192 int dp_idx = iminor(file->f_dentry->d_inode);
2193 struct datapath *dp = get_dp(dp_idx);
2198 poll_wait(file, &dp->waitqueue, wait);
2199 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2200 mask |= POLLIN | POLLRDNORM;
2202 mask = POLLIN | POLLRDNORM | POLLHUP;
2207 struct file_operations openvswitch_fops = {
2208 /* XXX .aio_read = openvswitch_aio_read, */
2209 .read = openvswitch_read,
2210 .poll = openvswitch_poll,
2211 .unlocked_ioctl = openvswitch_ioctl,
2212 #ifdef CONFIG_COMPAT
2213 .compat_ioctl = openvswitch_compat_ioctl,
2215 /* XXX .fasync = openvswitch_fasync, */
2220 static int __init dp_init(void)
2222 struct sk_buff *dummy_skb;
2225 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2227 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2235 goto error_flow_exit;
2237 err = register_netdevice_notifier(&dp_device_notifier);
2239 goto error_vport_exit;
2241 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2243 goto error_unreg_notifier;
2247 error_unreg_notifier:
2248 unregister_netdevice_notifier(&dp_device_notifier);
2257 static void dp_cleanup(void)
2260 unregister_chrdev(major, "openvswitch");
2261 unregister_netdevice_notifier(&dp_device_notifier);
2266 module_init(dp_init);
2267 module_exit(dp_cleanup);
2269 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2270 MODULE_LICENSE("GPL");