2 * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
16 #include <linux/if_arp.h>
17 #include <linux/if_vlan.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/kernel.h>
24 #include <linux/kthread.h>
25 #include <linux/mutex.h>
26 #include <linux/percpu.h>
27 #include <linux/rcupdate.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/version.h>
31 #include <linux/ethtool.h>
32 #include <linux/wait.h>
33 #include <asm/system.h>
34 #include <asm/div64.h>
36 #include <linux/highmem.h>
37 #include <linux/netfilter_bridge.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/inetdevice.h>
40 #include <linux/list.h>
41 #include <linux/rculist.h>
42 #include <linux/dmi.h>
43 #include <net/inet_ecn.h>
44 #include <linux/compat.h>
46 #include "openvswitch/datapath-protocol.h"
50 #include "loop_counter.h"
51 #include "odp-compat.h"
53 #include "vport-internal_dev.h"
57 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
58 EXPORT_SYMBOL(dp_ioctl_hook);
60 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
63 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
66 * It is safe to access the datapath and dp_port structures with just
69 static struct datapath *dps[ODP_MAX];
70 static DEFINE_MUTEX(dp_mutex);
72 static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
74 /* Must be called with rcu_read_lock or dp_mutex. */
75 struct datapath *get_dp(int dp_idx)
77 if (dp_idx < 0 || dp_idx >= ODP_MAX)
79 return rcu_dereference(dps[dp_idx]);
81 EXPORT_SYMBOL_GPL(get_dp);
83 static struct datapath *get_dp_locked(int dp_idx)
87 mutex_lock(&dp_mutex);
90 mutex_lock(&dp->mutex);
91 mutex_unlock(&dp_mutex);
95 /* Must be called with rcu_read_lock or RTNL lock. */
96 const char *dp_name(const struct datapath *dp)
98 return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
101 static inline size_t br_nlmsg_size(void)
103 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
104 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
105 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
106 + nla_total_size(4) /* IFLA_MASTER */
107 + nla_total_size(4) /* IFLA_MTU */
108 + nla_total_size(4) /* IFLA_LINK */
109 + nla_total_size(1); /* IFLA_OPERSTATE */
112 static int dp_fill_ifinfo(struct sk_buff *skb,
113 const struct dp_port *port,
114 int event, unsigned int flags)
116 const struct datapath *dp = port->dp;
117 int ifindex = vport_get_ifindex(port->vport);
118 int iflink = vport_get_iflink(port->vport);
119 struct ifinfomsg *hdr;
120 struct nlmsghdr *nlh;
128 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
132 hdr = nlmsg_data(nlh);
133 hdr->ifi_family = AF_BRIDGE;
135 hdr->ifi_type = ARPHRD_ETHER;
136 hdr->ifi_index = ifindex;
137 hdr->ifi_flags = vport_get_flags(port->vport);
140 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
141 NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
142 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
143 #ifdef IFLA_OPERSTATE
144 NLA_PUT_U8(skb, IFLA_OPERSTATE,
145 vport_is_running(port->vport)
146 ? vport_get_operstate(port->vport)
150 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
151 vport_get_addr(port->vport));
153 if (ifindex != iflink)
154 NLA_PUT_U32(skb, IFLA_LINK,iflink);
156 return nlmsg_end(skb, nlh);
159 nlmsg_cancel(skb, nlh);
163 static void dp_ifinfo_notify(int event, struct dp_port *port)
168 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
172 err = dp_fill_ifinfo(skb, port, event, 0);
174 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
175 WARN_ON(err == -EMSGSIZE);
179 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
183 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
186 static void release_dp(struct kobject *kobj)
188 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
192 static struct kobj_type dp_ktype = {
193 .release = release_dp
196 static int create_dp(int dp_idx, const char __user *devnamep)
198 struct odp_port internal_dev_port;
199 char devname[IFNAMSIZ];
205 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
209 } else if (retval >= IFNAMSIZ) {
214 snprintf(devname, sizeof devname, "of%d", dp_idx);
218 mutex_lock(&dp_mutex);
220 if (!try_module_get(THIS_MODULE))
223 /* Exit early if a datapath with that number already exists.
224 * (We don't use -EEXIST because that's ambiguous with 'devname'
225 * conflicting with an existing network device name.) */
231 dp = kzalloc(sizeof *dp, GFP_KERNEL);
234 INIT_LIST_HEAD(&dp->port_list);
235 mutex_init(&dp->mutex);
237 for (i = 0; i < DP_N_QUEUES; i++)
238 skb_queue_head_init(&dp->queues[i]);
239 init_waitqueue_head(&dp->waitqueue);
241 /* Initialize kobject for bridge. This will be added as
242 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
243 dp->ifobj.kset = NULL;
244 kobject_init(&dp->ifobj, &dp_ktype);
246 /* Allocate table. */
248 rcu_assign_pointer(dp->table, tbl_create(0));
252 /* Set up our datapath device. */
253 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
254 strcpy(internal_dev_port.devname, devname);
255 strcpy(internal_dev_port.type, "internal");
256 err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
261 goto err_destroy_table;
265 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
266 if (!dp->stats_percpu)
267 goto err_destroy_local_port;
269 rcu_assign_pointer(dps[dp_idx], dp);
270 mutex_unlock(&dp_mutex);
277 err_destroy_local_port:
278 dp_detach_port(dp->ports[ODPP_LOCAL]);
280 tbl_destroy(dp->table, NULL);
284 module_put(THIS_MODULE);
286 mutex_unlock(&dp_mutex);
292 static void do_destroy_dp(struct datapath *dp)
294 struct dp_port *p, *n;
297 list_for_each_entry_safe (p, n, &dp->port_list, node)
298 if (p->port_no != ODPP_LOCAL)
303 rcu_assign_pointer(dps[dp->dp_idx], NULL);
305 dp_detach_port(dp->ports[ODPP_LOCAL]);
307 tbl_destroy(dp->table, flow_free_tbl);
309 for (i = 0; i < DP_N_QUEUES; i++)
310 skb_queue_purge(&dp->queues[i]);
311 free_percpu(dp->stats_percpu);
312 kobject_put(&dp->ifobj);
313 module_put(THIS_MODULE);
316 static int destroy_dp(int dp_idx)
322 mutex_lock(&dp_mutex);
332 mutex_unlock(&dp_mutex);
337 static void release_dp_port(struct kobject *kobj)
339 struct dp_port *p = container_of(kobj, struct dp_port, kobj);
343 static struct kobj_type brport_ktype = {
345 .sysfs_ops = &brport_sysfs_ops,
347 .release = release_dp_port
350 /* Called with RTNL lock and dp_mutex. */
351 static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
353 struct vport_parms parms;
358 parms.name = odp_port->devname;
359 parms.type = odp_port->type;
360 parms.config = odp_port->config;
363 vport = vport_add(&parms);
367 return PTR_ERR(vport);
369 p = kzalloc(sizeof(*p), GFP_KERNEL);
373 p->port_no = port_no;
376 atomic_set(&p->sflow_pool, 0);
378 err = vport_attach(vport, p);
384 rcu_assign_pointer(dp->ports[port_no], p);
385 list_add_rcu(&p->node, &dp->port_list);
388 /* Initialize kobject for bridge. This will be added as
389 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
391 kobject_init(&p->kobj, &brport_ktype);
393 dp_ifinfo_notify(RTM_NEWLINK, p);
398 static int attach_port(int dp_idx, struct odp_port __user *portp)
401 struct odp_port port;
406 if (copy_from_user(&port, portp, sizeof port))
408 port.devname[IFNAMSIZ - 1] = '\0';
409 port.type[VPORT_TYPE_SIZE - 1] = '\0';
412 dp = get_dp_locked(dp_idx);
415 goto out_unlock_rtnl;
417 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
418 if (!dp->ports[port_no])
424 err = new_dp_port(dp, &port, port_no);
428 set_internal_devs_mtu(dp);
429 dp_sysfs_add_if(dp->ports[port_no]);
431 err = put_user(port_no, &portp->port);
434 mutex_unlock(&dp->mutex);
441 int dp_detach_port(struct dp_port *p)
443 struct vport *vport = p->vport;
448 if (p->port_no != ODPP_LOCAL)
450 dp_ifinfo_notify(RTM_DELLINK, p);
452 /* First drop references to device. */
454 list_del_rcu(&p->node);
455 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
457 err = vport_detach(vport);
461 /* Then wait until no one is still using it, and destroy it. */
468 kobject_put(&p->kobj);
473 static int detach_port(int dp_idx, int port_no)
480 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
484 dp = get_dp_locked(dp_idx);
487 goto out_unlock_rtnl;
489 p = dp->ports[port_no];
494 err = dp_detach_port(p);
497 mutex_unlock(&dp->mutex);
504 /* Must be called with rcu_read_lock. */
505 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
507 struct datapath *dp = p->dp;
508 struct dp_stats_percpu *stats;
509 int stats_counter_off;
510 struct sw_flow_actions *acts;
511 struct loop_counter *loop;
514 OVS_CB(skb)->dp_port = p;
516 if (!OVS_CB(skb)->flow) {
517 struct odp_flow_key key;
518 struct tbl_node *flow_node;
521 /* Extract flow from 'skb' into 'key'. */
522 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
523 if (unlikely(error)) {
528 if (is_frag && dp->drop_frags) {
530 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
535 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
536 flow_hash(&key), flow_cmp);
537 if (unlikely(!flow_node)) {
538 dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
539 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
543 OVS_CB(skb)->flow = flow_cast(flow_node);
546 flow_used(OVS_CB(skb)->flow, skb);
548 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
550 /* Check whether we've looped too much. */
551 loop = loop_get_counter();
552 if (unlikely(++loop->count > MAX_LOOPS))
553 loop->looping = true;
554 if (unlikely(loop->looping)) {
555 loop_suppress(dp, acts);
559 /* Execute actions. */
560 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
562 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
564 /* Check whether sub-actions looped too much. */
565 if (unlikely(loop->looping))
566 loop_suppress(dp, acts);
569 /* Decrement loop counter. */
571 loop->looping = false;
575 /* Update datapath statistics. */
577 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
579 write_seqcount_begin(&stats->seqlock);
580 (*(u64 *)((u8 *)stats + stats_counter_off))++;
581 write_seqcount_end(&stats->seqlock);
586 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
587 /* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
588 * can't call this function directly because it isn't exported in all
590 int vswitch_skb_checksum_setup(struct sk_buff *skb)
595 __u16 csum_start, csum_offset;
597 if (!skb->proto_csum_blank)
600 if (skb->protocol != htons(ETH_P_IP))
603 if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
607 th = skb_network_header(skb) + 4 * iph->ihl;
609 csum_start = th - skb->head;
610 switch (iph->protocol) {
612 csum_offset = offsetof(struct tcphdr, check);
615 csum_offset = offsetof(struct udphdr, check);
619 pr_err("Attempting to checksum a non-TCP/UDP packet, "
620 "dropping a protocol %d packet",
625 if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
628 skb->ip_summed = CHECKSUM_PARTIAL;
629 skb->proto_csum_blank = 0;
631 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
632 skb->csum_start = csum_start;
633 skb->csum_offset = csum_offset;
635 skb_set_transport_header(skb, csum_start - skb_headroom(skb));
636 skb->csum = csum_offset;
644 #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
646 /* Types of checksums that we can receive (these all refer to L4 checksums):
647 * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
648 * (though not verified) checksum in packet but not in skb->csum. Packets
649 * from the bridge local port will also have this type.
650 * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
651 * also the GRE module. This is the same as CHECKSUM_NONE, except it has
652 * a valid skb->csum. Importantly, both contain a full checksum (not
653 * verified) in the packet itself. The only difference is that if the
654 * packet gets to L4 processing on this machine (not in DomU) we won't
655 * have to recompute the checksum to verify. Most hardware devices do not
656 * produce packets with this type, even if they support receive checksum
657 * offloading (they produce type #5).
658 * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
659 * be computed if it is sent off box. Unfortunately on earlier kernels,
660 * this case is impossible to distinguish from #2, despite having opposite
661 * meanings. Xen adds an extra field on earlier kernels (see #4) in order
662 * to distinguish the different states.
663 * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
664 * generated locally by a Xen DomU and has a partial checksum. If it is
665 * handled on this machine (Dom0 or DomU), then the checksum will not be
666 * computed. If it goes off box, the checksum in the packet needs to be
667 * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
668 * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
669 * kernels, this combination is replaced with CHECKSUM_PARTIAL.
670 * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
671 * full checksum or using a protocol without a checksum. skb->csum is
672 * undefined. This is common from devices with receive checksum
673 * offloading. This is somewhat similar to CHECKSUM_NONE, except that
674 * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
676 * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
677 * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
678 * based on whether it is on the transmit or receive path. After the datapath
679 * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
680 * checksum, we will panic. Since we can receive packets with checksums, we
681 * assume that all CHECKSUM_HW packets have checksums and map them to
682 * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
683 * packet is processed by the local IP stack, in which case it will need to
684 * be reverified). If we receive a packet with CHECKSUM_HW that really means
685 * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
686 * shouldn't be any devices that do this with bridging. */
687 void compute_ip_summed(struct sk_buff *skb, bool xmit)
689 /* For our convenience these defines change repeatedly between kernel
690 * versions, so we can't just copy them over... */
691 switch (skb->ip_summed) {
693 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
695 case CHECKSUM_UNNECESSARY:
696 OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
699 /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
700 * However, on the receive side we should only get CHECKSUM_PARTIAL
701 * packets from Xen, which uses some special fields to represent this
702 * (see below). Since we can only make one type work, pick the one
703 * that actually happens in practice.
705 * On the transmit side (basically after skb_checksum_setup()
706 * has been run or on internal dev transmit), packets with
707 * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
710 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
712 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
716 case CHECKSUM_COMPLETE:
717 OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
719 case CHECKSUM_PARTIAL:
720 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
724 pr_err("unknown checksum type %d\n", skb->ip_summed);
725 /* None seems the safest... */
726 OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
729 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
730 /* Xen has a special way of representing CHECKSUM_PARTIAL on older
731 * kernels. It should not be set on the transmit path though. */
732 if (skb->proto_csum_blank)
733 OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
735 WARN_ON_ONCE(skb->proto_csum_blank && xmit);
739 /* This function closely resembles skb_forward_csum() used by the bridge. It
740 * is slightly different because we are only concerned with bridging and not
741 * other types of forwarding and can get away with slightly more optimal
743 void forward_ip_summed(struct sk_buff *skb)
746 if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
747 skb->ip_summed = CHECKSUM_NONE;
751 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
752 * unless we broke up a GSO packet. */
753 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
754 int queue_no, u32 arg)
756 struct sk_buff *nskb;
760 if (OVS_CB(skb)->dp_port)
761 port_no = OVS_CB(skb)->dp_port->port_no;
763 port_no = ODPP_LOCAL;
766 struct odp_msg *header;
771 err = skb_cow(skb, sizeof *header);
775 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
776 header->type = queue_no;
777 header->length = skb->len;
778 header->port = port_no;
779 header->reserved = 0;
781 skb_queue_tail(queue, skb);
789 while ((skb = nskb) != NULL) {
796 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
799 struct dp_stats_percpu *stats;
800 struct sk_buff_head *queue;
803 WARN_ON_ONCE(skb_shared(skb));
804 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
805 queue = &dp->queues[queue_no];
807 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
810 forward_ip_summed(skb);
812 err = vswitch_skb_checksum_setup(skb);
816 /* Break apart GSO packets into their component pieces. Otherwise
817 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
818 if (skb_is_gso(skb)) {
819 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
823 if (unlikely(IS_ERR(skb))) {
828 /* XXX This case might not be possible. It's hard to
829 * tell from the skb_gso_segment() code and comment. */
833 err = queue_control_packets(skb, queue, queue_no, arg);
834 wake_up_interruptible(&dp->waitqueue);
841 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
843 write_seqcount_begin(&stats->seqlock);
845 write_seqcount_end(&stats->seqlock);
852 static int flush_flows(struct datapath *dp)
854 struct tbl *old_table = rcu_dereference(dp->table);
855 struct tbl *new_table;
857 new_table = tbl_create(0);
861 rcu_assign_pointer(dp->table, new_table);
863 tbl_deferred_destroy(old_table, flow_free_tbl);
868 static int validate_actions(const struct sw_flow_actions *actions)
872 for (i = 0; i < actions->n_actions; i++) {
873 const union odp_action *a = &actions->actions[i];
876 case ODPAT_CONTROLLER:
877 case ODPAT_STRIP_VLAN:
878 case ODPAT_SET_DL_SRC:
879 case ODPAT_SET_DL_DST:
880 case ODPAT_SET_NW_SRC:
881 case ODPAT_SET_NW_DST:
882 case ODPAT_SET_TP_SRC:
883 case ODPAT_SET_TP_DST:
884 case ODPAT_SET_TUNNEL:
885 case ODPAT_SET_PRIORITY:
886 case ODPAT_POP_PRIORITY:
887 case ODPAT_DROP_SPOOFED_ARP:
888 /* No validation needed. */
892 if (a->output.port >= DP_MAX_PORTS)
896 case ODPAT_SET_DL_TCI:
897 if (a->dl_tci.tci & htons(VLAN_CFI_MASK))
901 case ODPAT_SET_NW_TOS:
902 if (a->nw_tos.nw_tos & INET_ECN_MASK)
914 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
916 struct sw_flow_actions *actions;
919 actions = flow_actions_alloc(flow->n_actions);
920 error = PTR_ERR(actions);
925 if (copy_from_user(actions->actions, flow->actions,
926 flow->n_actions * sizeof(union odp_action)))
927 goto error_free_actions;
928 error = validate_actions(actions);
930 goto error_free_actions;
937 return ERR_PTR(error);
940 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
943 struct timespec offset_ts, used, now_mono;
945 ktime_get_ts(&now_mono);
946 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
947 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
948 now_mono.tv_nsec - offset_ts.tv_nsec);
950 stats->used_sec = used.tv_sec;
951 stats->used_nsec = used.tv_nsec;
954 stats->used_nsec = 0;
957 stats->n_packets = flow->packet_count;
958 stats->n_bytes = flow->byte_count;
960 stats->tcp_flags = flow->tcp_flags;
964 static void clear_stats(struct sw_flow *flow)
968 flow->packet_count = 0;
969 flow->byte_count = 0;
972 static int expand_table(struct datapath *dp)
974 struct tbl *old_table = rcu_dereference(dp->table);
975 struct tbl *new_table;
977 new_table = tbl_expand(old_table);
978 if (IS_ERR(new_table))
979 return PTR_ERR(new_table);
981 rcu_assign_pointer(dp->table, new_table);
982 tbl_deferred_destroy(old_table, NULL);
987 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
988 struct odp_flow_stats *stats)
990 struct tbl_node *flow_node;
991 struct sw_flow *flow;
995 table = rcu_dereference(dp->table);
996 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
999 struct sw_flow_actions *acts;
1002 if (!(uf->flags & ODPPF_CREATE))
1005 /* Expand table, if necessary, to make room. */
1006 if (tbl_count(table) >= tbl_n_buckets(table)) {
1007 error = expand_table(dp);
1010 table = rcu_dereference(dp->table);
1013 /* Allocate flow. */
1014 flow = flow_alloc();
1016 error = PTR_ERR(flow);
1019 flow->key = uf->flow.key;
1022 /* Obtain actions. */
1023 acts = get_actions(&uf->flow);
1024 error = PTR_ERR(acts);
1026 goto error_free_flow;
1027 rcu_assign_pointer(flow->sf_acts, acts);
1029 /* Put flow in bucket. */
1030 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
1032 goto error_free_flow_acts;
1034 memset(stats, 0, sizeof(struct odp_flow_stats));
1036 /* We found a matching flow. */
1037 struct sw_flow_actions *old_acts, *new_acts;
1039 flow = flow_cast(flow_node);
1041 /* Bail out if we're not allowed to modify an existing flow. */
1043 if (!(uf->flags & ODPPF_MODIFY))
1047 new_acts = get_actions(&uf->flow);
1048 error = PTR_ERR(new_acts);
1049 if (IS_ERR(new_acts))
1051 old_acts = rcu_dereference(flow->sf_acts);
1052 if (old_acts->n_actions != new_acts->n_actions ||
1053 memcmp(old_acts->actions, new_acts->actions,
1054 sizeof(union odp_action) * old_acts->n_actions)) {
1055 rcu_assign_pointer(flow->sf_acts, new_acts);
1056 flow_deferred_free_acts(old_acts);
1061 /* Fetch stats, then clear them if necessary. */
1062 spin_lock_bh(&flow->lock);
1063 get_stats(flow, stats);
1064 if (uf->flags & ODPPF_ZERO_STATS)
1066 spin_unlock_bh(&flow->lock);
1071 error_free_flow_acts:
1072 kfree(flow->sf_acts);
1074 flow->sf_acts = NULL;
1080 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
1082 struct odp_flow_stats stats;
1083 struct odp_flow_put uf;
1086 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
1089 error = do_put_flow(dp, &uf, &stats);
1093 if (copy_to_user(&ufp->flow.stats, &stats,
1094 sizeof(struct odp_flow_stats)))
1100 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
1101 struct odp_flow_stats __user *ustats,
1102 union odp_action __user *actions,
1103 u32 __user *n_actionsp)
1105 struct sw_flow_actions *sf_acts;
1106 struct odp_flow_stats stats;
1109 spin_lock_bh(&flow->lock);
1110 get_stats(flow, &stats);
1111 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
1112 flow->tcp_flags = 0;
1114 spin_unlock_bh(&flow->lock);
1116 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
1117 get_user(n_actions, n_actionsp))
1123 sf_acts = rcu_dereference(flow->sf_acts);
1124 if (put_user(sf_acts->n_actions, n_actionsp) ||
1125 (actions && copy_to_user(actions, sf_acts->actions,
1126 sizeof(union odp_action) *
1127 min(sf_acts->n_actions, n_actions))))
1133 static int answer_query(struct sw_flow *flow, u32 query_flags,
1134 struct odp_flow __user *ufp)
1136 union odp_action *actions;
1138 if (get_user(actions, &ufp->actions))
1141 return do_answer_query(flow, query_flags,
1142 &ufp->stats, actions, &ufp->n_actions);
1145 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
1147 struct tbl *table = rcu_dereference(dp->table);
1148 struct tbl_node *flow_node;
1151 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
1153 return ERR_PTR(-ENOENT);
1155 error = tbl_remove(table, flow_node);
1157 return ERR_PTR(error);
1159 /* XXX Returned flow_node's statistics might lose a few packets, since
1160 * other CPUs can be using this flow. We used to synchronize_rcu() to
1161 * make sure that we get completely accurate stats, but that blows our
1162 * performance, badly. */
1163 return flow_cast(flow_node);
1166 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
1168 struct sw_flow *flow;
1172 if (copy_from_user(&uf, ufp, sizeof uf))
1175 flow = do_del_flow(dp, &uf.key);
1177 return PTR_ERR(flow);
1179 error = answer_query(flow, 0, ufp);
1180 flow_deferred_free(flow);
1184 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1186 struct tbl *table = rcu_dereference(dp->table);
1189 for (i = 0; i < flowvec->n_flows; i++) {
1190 struct odp_flow __user *ufp = &flowvec->flows[i];
1192 struct tbl_node *flow_node;
1195 if (copy_from_user(&uf, ufp, sizeof uf))
1198 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1200 error = put_user(ENOENT, &ufp->stats.error);
1202 error = answer_query(flow_cast(flow_node), uf.flags, ufp);
1206 return flowvec->n_flows;
1209 struct list_flows_cbdata {
1210 struct odp_flow __user *uflows;
1215 static int list_flow(struct tbl_node *node, void *cbdata_)
1217 struct sw_flow *flow = flow_cast(node);
1218 struct list_flows_cbdata *cbdata = cbdata_;
1219 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1222 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1224 error = answer_query(flow, 0, ufp);
1228 if (cbdata->listed_flows >= cbdata->n_flows)
1229 return cbdata->listed_flows;
1233 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1235 struct list_flows_cbdata cbdata;
1238 if (!flowvec->n_flows)
1241 cbdata.uflows = flowvec->flows;
1242 cbdata.n_flows = flowvec->n_flows;
1243 cbdata.listed_flows = 0;
1245 error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
1246 return error ? error : cbdata.listed_flows;
1249 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1250 int (*function)(struct datapath *,
1251 const struct odp_flowvec *))
1253 struct odp_flowvec __user *uflowvec;
1254 struct odp_flowvec flowvec;
1257 uflowvec = (struct odp_flowvec __user *)argp;
1258 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1261 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1264 retval = function(dp, &flowvec);
1265 return (retval < 0 ? retval
1266 : retval == flowvec.n_flows ? 0
1267 : put_user(retval, &uflowvec->n_flows));
1270 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1272 struct odp_flow_key key;
1273 struct sk_buff *skb;
1274 struct sw_flow_actions *actions;
1280 if (execute->length < ETH_HLEN || execute->length > 65535)
1283 actions = flow_actions_alloc(execute->n_actions);
1284 if (IS_ERR(actions)) {
1285 err = PTR_ERR(actions);
1290 if (copy_from_user(actions->actions, execute->actions,
1291 execute->n_actions * sizeof *execute->actions))
1292 goto error_free_actions;
1294 err = validate_actions(actions);
1296 goto error_free_actions;
1299 skb = alloc_skb(execute->length, GFP_KERNEL);
1301 goto error_free_actions;
1304 if (copy_from_user(skb_put(skb, execute->length), execute->data,
1306 goto error_free_skb;
1308 skb_reset_mac_header(skb);
1311 /* Normally, setting the skb 'protocol' field would be handled by a
1312 * call to eth_type_trans(), but it assumes there's a sending
1313 * device, which we may not have. */
1314 if (ntohs(eth->h_proto) >= 1536)
1315 skb->protocol = eth->h_proto;
1317 skb->protocol = htons(ETH_P_802_2);
1319 err = flow_extract(skb, -1, &key, &is_frag);
1321 goto error_free_skb;
1324 err = execute_actions(dp, skb, &key, actions->actions, actions->n_actions);
1338 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1340 struct odp_execute execute;
1342 if (copy_from_user(&execute, executep, sizeof execute))
1345 return do_execute(dp, &execute);
1348 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1350 struct tbl *table = rcu_dereference(dp->table);
1351 struct odp_stats stats;
1354 stats.n_flows = tbl_count(table);
1355 stats.cur_capacity = tbl_n_buckets(table);
1356 stats.max_capacity = TBL_MAX_BUCKETS;
1357 stats.n_ports = dp->n_ports;
1358 stats.max_ports = DP_MAX_PORTS;
1359 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1360 for_each_possible_cpu(i) {
1361 const struct dp_stats_percpu *percpu_stats;
1362 struct dp_stats_percpu local_stats;
1365 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1368 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1369 local_stats = *percpu_stats;
1370 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1372 stats.n_frags += local_stats.n_frags;
1373 stats.n_hit += local_stats.n_hit;
1374 stats.n_missed += local_stats.n_missed;
1375 stats.n_lost += local_stats.n_lost;
1377 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1378 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1379 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1382 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1383 int dp_min_mtu(const struct datapath *dp)
1390 list_for_each_entry_rcu (p, &dp->port_list, node) {
1393 /* Skip any internal ports, since that's what we're trying to
1395 if (is_internal_vport(p->vport))
1398 dev_mtu = vport_get_mtu(p->vport);
1399 if (!mtu || dev_mtu < mtu)
1403 return mtu ? mtu : ETH_DATA_LEN;
1406 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1407 * be called with RTNL lock. */
1408 void set_internal_devs_mtu(const struct datapath *dp)
1415 mtu = dp_min_mtu(dp);
1417 list_for_each_entry_rcu (p, &dp->port_list, node) {
1418 if (is_internal_vport(p->vport))
1419 vport_set_mtu(p->vport, mtu);
1423 static int put_port(const struct dp_port *p, struct odp_port __user *uop)
1427 memset(&op, 0, sizeof op);
1430 strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
1431 strncpy(op.type, vport_get_type(p->vport), sizeof op.type);
1434 op.port = p->port_no;
1436 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1439 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1441 struct odp_port port;
1443 if (copy_from_user(&port, uport, sizeof port))
1446 if (port.devname[0]) {
1447 struct vport *vport;
1448 struct dp_port *dp_port;
1451 port.devname[IFNAMSIZ - 1] = '\0';
1456 vport = vport_locate(port.devname);
1462 dp_port = vport_get_dp_port(vport);
1463 if (!dp_port || dp_port->dp != dp) {
1468 port.port = dp_port->port_no;
1477 if (port.port >= DP_MAX_PORTS)
1479 if (!dp->ports[port.port])
1483 return put_port(dp->ports[port.port], uport);
1486 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1493 list_for_each_entry_rcu (p, &dp->port_list, node) {
1494 if (put_port(p, &uports[idx]))
1496 if (idx++ >= n_ports)
1503 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1505 struct odp_portvec pv;
1508 if (copy_from_user(&pv, upv, sizeof pv))
1511 retval = do_list_ports(dp, pv.ports, pv.n_ports);
1515 return put_user(retval, &upv->n_ports);
1518 static int get_listen_mask(const struct file *f)
1520 return (long)f->private_data;
1523 static void set_listen_mask(struct file *f, int listen_mask)
1525 f->private_data = (void*)(long)listen_mask;
1528 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1531 int dp_idx = iminor(f->f_dentry->d_inode);
1532 struct datapath *dp;
1533 int drop_frags, listeners, port_no;
1534 unsigned int sflow_probability;
1537 /* Handle commands with special locking requirements up front. */
1540 err = create_dp(dp_idx, (char __user *)argp);
1543 case ODP_DP_DESTROY:
1544 err = destroy_dp(dp_idx);
1547 case ODP_VPORT_ATTACH:
1548 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1551 case ODP_VPORT_DETACH:
1552 err = get_user(port_no, (int __user *)argp);
1554 err = detach_port(dp_idx, port_no);
1558 err = vport_user_mod((struct odp_port __user *)argp);
1561 case ODP_VPORT_STATS_GET:
1562 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1565 case ODP_VPORT_STATS_SET:
1566 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1569 case ODP_VPORT_ETHER_GET:
1570 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1573 case ODP_VPORT_ETHER_SET:
1574 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1577 case ODP_VPORT_MTU_GET:
1578 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1581 case ODP_VPORT_MTU_SET:
1582 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1586 dp = get_dp_locked(dp_idx);
1593 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1596 case ODP_GET_DROP_FRAGS:
1597 err = put_user(dp->drop_frags, (int __user *)argp);
1600 case ODP_SET_DROP_FRAGS:
1601 err = get_user(drop_frags, (int __user *)argp);
1605 if (drop_frags != 0 && drop_frags != 1)
1607 dp->drop_frags = drop_frags;
1611 case ODP_GET_LISTEN_MASK:
1612 err = put_user(get_listen_mask(f), (int __user *)argp);
1615 case ODP_SET_LISTEN_MASK:
1616 err = get_user(listeners, (int __user *)argp);
1620 if (listeners & ~ODPL_ALL)
1623 set_listen_mask(f, listeners);
1626 case ODP_GET_SFLOW_PROBABILITY:
1627 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1630 case ODP_SET_SFLOW_PROBABILITY:
1631 err = get_user(sflow_probability, (unsigned int __user *)argp);
1633 dp->sflow_probability = sflow_probability;
1636 case ODP_VPORT_QUERY:
1637 err = query_port(dp, (struct odp_port __user *)argp);
1640 case ODP_VPORT_LIST:
1641 err = list_ports(dp, (struct odp_portvec __user *)argp);
1644 case ODP_FLOW_FLUSH:
1645 err = flush_flows(dp);
1649 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1653 err = del_flow(dp, (struct odp_flow __user *)argp);
1657 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1661 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1665 err = execute_packet(dp, (struct odp_execute __user *)argp);
1672 mutex_unlock(&dp->mutex);
1677 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1680 for (i = 0; i < DP_N_QUEUES; i++) {
1681 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1687 #ifdef CONFIG_COMPAT
1688 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1690 struct compat_odp_portvec pv;
1693 if (copy_from_user(&pv, upv, sizeof pv))
1696 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1700 return put_user(retval, &upv->n_ports);
1703 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1705 compat_uptr_t actions;
1707 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1708 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1709 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1710 __get_user(actions, &compat->actions) ||
1711 __get_user(flow->n_actions, &compat->n_actions) ||
1712 __get_user(flow->flags, &compat->flags))
1715 flow->actions = compat_ptr(actions);
1719 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1721 struct odp_flow_stats stats;
1722 struct odp_flow_put fp;
1725 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1726 get_user(fp.flags, &ufp->flags))
1729 error = do_put_flow(dp, &fp, &stats);
1733 if (copy_to_user(&ufp->flow.stats, &stats,
1734 sizeof(struct odp_flow_stats)))
1740 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
1741 struct compat_odp_flow __user *ufp)
1743 compat_uptr_t actions;
1745 if (get_user(actions, &ufp->actions))
1748 return do_answer_query(flow, query_flags, &ufp->stats,
1749 compat_ptr(actions), &ufp->n_actions);
1752 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1754 struct sw_flow *flow;
1758 if (compat_get_flow(&uf, ufp))
1761 flow = do_del_flow(dp, &uf.key);
1763 return PTR_ERR(flow);
1765 error = compat_answer_query(flow, 0, ufp);
1766 flow_deferred_free(flow);
1770 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1772 struct tbl *table = rcu_dereference(dp->table);
1775 for (i = 0; i < n_flows; i++) {
1776 struct compat_odp_flow __user *ufp = &flows[i];
1778 struct tbl_node *flow_node;
1781 if (compat_get_flow(&uf, ufp))
1784 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1786 error = put_user(ENOENT, &ufp->stats.error);
1788 error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
1795 struct compat_list_flows_cbdata {
1796 struct compat_odp_flow __user *uflows;
1801 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1803 struct sw_flow *flow = flow_cast(node);
1804 struct compat_list_flows_cbdata *cbdata = cbdata_;
1805 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1808 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1810 error = compat_answer_query(flow, 0, ufp);
1814 if (cbdata->listed_flows >= cbdata->n_flows)
1815 return cbdata->listed_flows;
1819 static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
1821 struct compat_list_flows_cbdata cbdata;
1827 cbdata.uflows = flows;
1828 cbdata.n_flows = n_flows;
1829 cbdata.listed_flows = 0;
1831 error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
1832 return error ? error : cbdata.listed_flows;
1835 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1836 int (*function)(struct datapath *,
1837 struct compat_odp_flow *,
1840 struct compat_odp_flowvec __user *uflowvec;
1841 struct compat_odp_flow __user *flows;
1842 struct compat_odp_flowvec flowvec;
1845 uflowvec = compat_ptr(argp);
1846 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1847 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1850 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1853 flows = compat_ptr(flowvec.flows);
1854 if (!access_ok(VERIFY_WRITE, flows,
1855 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1858 retval = function(dp, flows, flowvec.n_flows);
1859 return (retval < 0 ? retval
1860 : retval == flowvec.n_flows ? 0
1861 : put_user(retval, &uflowvec->n_flows));
1864 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1866 struct odp_execute execute;
1867 compat_uptr_t actions;
1870 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1871 __get_user(actions, &uexecute->actions) ||
1872 __get_user(execute.n_actions, &uexecute->n_actions) ||
1873 __get_user(data, &uexecute->data) ||
1874 __get_user(execute.length, &uexecute->length))
1877 execute.actions = compat_ptr(actions);
1878 execute.data = compat_ptr(data);
1880 return do_execute(dp, &execute);
1883 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1885 int dp_idx = iminor(f->f_dentry->d_inode);
1886 struct datapath *dp;
1890 case ODP_DP_DESTROY:
1891 case ODP_FLOW_FLUSH:
1892 /* Ioctls that don't need any translation at all. */
1893 return openvswitch_ioctl(f, cmd, argp);
1896 case ODP_VPORT_ATTACH:
1897 case ODP_VPORT_DETACH:
1899 case ODP_VPORT_MTU_SET:
1900 case ODP_VPORT_MTU_GET:
1901 case ODP_VPORT_ETHER_SET:
1902 case ODP_VPORT_ETHER_GET:
1903 case ODP_VPORT_STATS_SET:
1904 case ODP_VPORT_STATS_GET:
1906 case ODP_GET_DROP_FRAGS:
1907 case ODP_SET_DROP_FRAGS:
1908 case ODP_SET_LISTEN_MASK:
1909 case ODP_GET_LISTEN_MASK:
1910 case ODP_SET_SFLOW_PROBABILITY:
1911 case ODP_GET_SFLOW_PROBABILITY:
1912 case ODP_VPORT_QUERY:
1913 /* Ioctls that just need their pointer argument extended. */
1914 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1917 dp = get_dp_locked(dp_idx);
1923 case ODP_VPORT_LIST32:
1924 err = compat_list_ports(dp, compat_ptr(argp));
1927 case ODP_FLOW_PUT32:
1928 err = compat_put_flow(dp, compat_ptr(argp));
1931 case ODP_FLOW_DEL32:
1932 err = compat_del_flow(dp, compat_ptr(argp));
1935 case ODP_FLOW_GET32:
1936 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1939 case ODP_FLOW_LIST32:
1940 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1944 err = compat_execute(dp, compat_ptr(argp));
1951 mutex_unlock(&dp->mutex);
1957 /* Unfortunately this function is not exported so this is a verbatim copy
1958 * from net/core/datagram.c in 2.6.30. */
1959 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1960 u8 __user *to, int len,
1963 int start = skb_headlen(skb);
1965 int i, copy = start - offset;
1972 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1976 if ((len -= copy) == 0)
1983 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1986 WARN_ON(start > offset + len);
1988 end = start + skb_shinfo(skb)->frags[i].size;
1989 if ((copy = end - offset) > 0) {
1993 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1994 struct page *page = frag->page;
1999 csum2 = csum_and_copy_to_user(vaddr +
2006 *csump = csum_block_add(*csump, csum2, pos);
2016 if (skb_shinfo(skb)->frag_list) {
2017 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2019 for (; list; list=list->next) {
2022 WARN_ON(start > offset + len);
2024 end = start + list->len;
2025 if ((copy = end - offset) > 0) {
2029 if (skb_copy_and_csum_datagram(list,
2034 *csump = csum_block_add(*csump, csum2, pos);
2035 if ((len -= copy) == 0)
2051 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
2054 /* XXX is there sufficient synchronization here? */
2055 int listeners = get_listen_mask(f);
2056 int dp_idx = iminor(f->f_dentry->d_inode);
2057 struct datapath *dp = get_dp(dp_idx);
2058 struct sk_buff *skb;
2059 size_t copy_bytes, tot_copy_bytes;
2065 if (nbytes == 0 || !listeners)
2071 for (i = 0; i < DP_N_QUEUES; i++) {
2072 if (listeners & (1 << i)) {
2073 skb = skb_dequeue(&dp->queues[i]);
2079 if (f->f_flags & O_NONBLOCK) {
2084 wait_event_interruptible(dp->waitqueue,
2085 dp_has_packet_of_interest(dp,
2088 if (signal_pending(current)) {
2089 retval = -ERESTARTSYS;
2094 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
2097 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2098 if (copy_bytes == skb->len) {
2100 unsigned int csum_start, csum_offset;
2102 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
2103 csum_start = skb->csum_start - skb_headroom(skb);
2104 csum_offset = skb->csum_offset;
2106 csum_start = skb_transport_header(skb) - skb->data;
2107 csum_offset = skb->csum;
2109 BUG_ON(csum_start >= skb_headlen(skb));
2110 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
2111 copy_bytes - csum_start, &csum);
2113 __sum16 __user *csump;
2115 copy_bytes = csum_start;
2116 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
2118 BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
2119 put_user(csum_fold(csum), csump);
2122 retval = skb_checksum_help(skb);
2126 struct iovec __user iov;
2129 iov.iov_len = copy_bytes;
2130 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
2134 retval = tot_copy_bytes;
2142 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
2144 /* XXX is there sufficient synchronization here? */
2145 int dp_idx = iminor(file->f_dentry->d_inode);
2146 struct datapath *dp = get_dp(dp_idx);
2151 poll_wait(file, &dp->waitqueue, wait);
2152 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
2153 mask |= POLLIN | POLLRDNORM;
2155 mask = POLLIN | POLLRDNORM | POLLHUP;
2160 struct file_operations openvswitch_fops = {
2161 /* XXX .aio_read = openvswitch_aio_read, */
2162 .read = openvswitch_read,
2163 .poll = openvswitch_poll,
2164 .unlocked_ioctl = openvswitch_ioctl,
2165 #ifdef CONFIG_COMPAT
2166 .compat_ioctl = openvswitch_compat_ioctl,
2168 /* XXX .fasync = openvswitch_fasync, */
2173 static int __init dp_init(void)
2175 struct sk_buff *dummy_skb;
2178 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2180 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2188 goto error_flow_exit;
2190 err = register_netdevice_notifier(&dp_device_notifier);
2192 goto error_vport_exit;
2194 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2196 goto error_unreg_notifier;
2200 error_unreg_notifier:
2201 unregister_netdevice_notifier(&dp_device_notifier);
2210 static void dp_cleanup(void)
2213 unregister_chrdev(major, "openvswitch");
2214 unregister_netdevice_notifier(&dp_device_notifier);
2219 module_init(dp_init);
2220 module_exit(dp_cleanup);
2222 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2223 MODULE_LICENSE("GPL");