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"
51 #include "loop_counter.h"
52 #include "odp-compat.h"
54 #include "vport-internal_dev.h"
58 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
59 EXPORT_SYMBOL(dp_ioctl_hook);
61 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
64 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
67 * It is safe to access the datapath and vport structures with just
70 static struct datapath __rcu *dps[ODP_MAX];
71 static DEFINE_MUTEX(dp_mutex);
73 static int new_vport(struct datapath *, struct odp_port *, int port_no);
75 /* Must be called with rcu_read_lock or dp_mutex. */
76 struct datapath *get_dp(int dp_idx)
78 if (dp_idx < 0 || dp_idx >= ODP_MAX)
80 return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
81 lockdep_is_held(&dp_mutex));
83 EXPORT_SYMBOL_GPL(get_dp);
85 static struct datapath *get_dp_locked(int dp_idx)
89 mutex_lock(&dp_mutex);
92 mutex_lock(&dp->mutex);
93 mutex_unlock(&dp_mutex);
97 static struct tbl *get_table_protected(struct datapath *dp)
99 return rcu_dereference_protected(dp->table, lockdep_is_held(&dp->mutex));
102 /* Must be called with rcu_read_lock or RTNL lock. */
103 const char *dp_name(const struct datapath *dp)
105 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
108 static inline size_t br_nlmsg_size(void)
110 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
111 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
112 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
113 + nla_total_size(4) /* IFLA_MASTER */
114 + nla_total_size(4) /* IFLA_MTU */
115 + nla_total_size(4) /* IFLA_LINK */
116 + nla_total_size(1); /* IFLA_OPERSTATE */
119 static int dp_fill_ifinfo(struct sk_buff *skb,
120 const struct vport *port,
121 int event, unsigned int flags)
123 const struct datapath *dp = port->dp;
124 int ifindex = vport_get_ifindex(port);
125 int iflink = vport_get_iflink(port);
126 struct ifinfomsg *hdr;
127 struct nlmsghdr *nlh;
135 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
139 hdr = nlmsg_data(nlh);
140 hdr->ifi_family = AF_BRIDGE;
142 hdr->ifi_type = ARPHRD_ETHER;
143 hdr->ifi_index = ifindex;
144 hdr->ifi_flags = vport_get_flags(port);
147 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
148 NLA_PUT_U32(skb, IFLA_MASTER,
149 vport_get_ifindex(rtnl_dereference(dp->ports[ODPP_LOCAL])));
150 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
151 #ifdef IFLA_OPERSTATE
152 NLA_PUT_U8(skb, IFLA_OPERSTATE,
153 vport_is_running(port)
154 ? vport_get_operstate(port)
158 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
160 if (ifindex != iflink)
161 NLA_PUT_U32(skb, IFLA_LINK,iflink);
163 return nlmsg_end(skb, nlh);
166 nlmsg_cancel(skb, nlh);
170 static void dp_ifinfo_notify(int event, struct vport *port)
175 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
179 err = dp_fill_ifinfo(skb, port, event, 0);
181 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
182 WARN_ON(err == -EMSGSIZE);
186 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
190 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
193 static void release_dp(struct kobject *kobj)
195 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
199 static struct kobj_type dp_ktype = {
200 .release = release_dp
203 static int create_dp(int dp_idx, const char __user *devnamep)
205 struct odp_port internal_dev_port;
206 char devname[IFNAMSIZ];
212 int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
216 } else if (retval >= IFNAMSIZ) {
221 snprintf(devname, sizeof devname, "of%d", dp_idx);
225 mutex_lock(&dp_mutex);
227 if (!try_module_get(THIS_MODULE))
230 /* Exit early if a datapath with that number already exists.
231 * (We don't use -EEXIST because that's ambiguous with 'devname'
232 * conflicting with an existing network device name.) */
238 dp = kzalloc(sizeof *dp, GFP_KERNEL);
241 INIT_LIST_HEAD(&dp->port_list);
242 mutex_init(&dp->mutex);
244 for (i = 0; i < DP_N_QUEUES; i++)
245 skb_queue_head_init(&dp->queues[i]);
246 init_waitqueue_head(&dp->waitqueue);
248 /* Initialize kobject for bridge. This will be added as
249 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
250 dp->ifobj.kset = NULL;
251 kobject_init(&dp->ifobj, &dp_ktype);
253 /* Allocate table. */
255 rcu_assign_pointer(dp->table, tbl_create(0));
259 /* Set up our datapath device. */
260 BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
261 strcpy(internal_dev_port.devname, devname);
262 strcpy(internal_dev_port.type, "internal");
263 err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
268 goto err_destroy_table;
272 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
273 if (!dp->stats_percpu) {
275 goto err_destroy_local_port;
278 rcu_assign_pointer(dps[dp_idx], dp);
281 mutex_unlock(&dp_mutex);
286 err_destroy_local_port:
287 dp_detach_port(dp->ports[ODPP_LOCAL]);
289 tbl_destroy(dp->table, NULL);
293 module_put(THIS_MODULE);
295 mutex_unlock(&dp_mutex);
301 static void do_destroy_dp(struct datapath *dp)
306 list_for_each_entry_safe (p, n, &dp->port_list, node)
307 if (p->port_no != ODPP_LOCAL)
312 rcu_assign_pointer(dps[dp->dp_idx], NULL);
314 dp_detach_port(dp->ports[ODPP_LOCAL]);
316 tbl_destroy(dp->table, flow_free_tbl);
318 for (i = 0; i < DP_N_QUEUES; i++)
319 skb_queue_purge(&dp->queues[i]);
320 free_percpu(dp->stats_percpu);
321 kobject_put(&dp->ifobj);
322 module_put(THIS_MODULE);
325 static int destroy_dp(int dp_idx)
331 mutex_lock(&dp_mutex);
341 mutex_unlock(&dp_mutex);
346 /* Called with RTNL lock and dp_mutex. */
347 static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
349 struct vport_parms parms;
352 parms.name = odp_port->devname;
353 parms.type = odp_port->type;
354 parms.config = odp_port->config;
356 parms.port_no = port_no;
359 vport = vport_add(&parms);
363 return PTR_ERR(vport);
365 rcu_assign_pointer(dp->ports[port_no], vport);
366 list_add_rcu(&vport->node, &dp->port_list);
369 dp_ifinfo_notify(RTM_NEWLINK, vport);
374 static int attach_port(int dp_idx, struct odp_port __user *portp)
377 struct odp_port port;
382 if (copy_from_user(&port, portp, sizeof port))
384 port.devname[IFNAMSIZ - 1] = '\0';
385 port.type[VPORT_TYPE_SIZE - 1] = '\0';
388 dp = get_dp_locked(dp_idx);
391 goto out_unlock_rtnl;
393 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
394 if (!dp->ports[port_no])
400 err = new_vport(dp, &port, port_no);
404 set_internal_devs_mtu(dp);
405 dp_sysfs_add_if(dp->ports[port_no]);
407 err = put_user(port_no, &portp->port);
410 mutex_unlock(&dp->mutex);
417 int dp_detach_port(struct vport *p)
423 if (p->port_no != ODPP_LOCAL)
425 dp_ifinfo_notify(RTM_DELLINK, p);
427 /* First drop references to device. */
429 list_del_rcu(&p->node);
430 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
432 /* Then destroy it. */
440 static int detach_port(int dp_idx, int port_no)
447 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
451 dp = get_dp_locked(dp_idx);
454 goto out_unlock_rtnl;
456 p = dp->ports[port_no];
461 err = dp_detach_port(p);
464 mutex_unlock(&dp->mutex);
471 /* Must be called with rcu_read_lock. */
472 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
474 struct datapath *dp = p->dp;
475 struct dp_stats_percpu *stats;
476 int stats_counter_off;
477 struct sw_flow_actions *acts;
478 struct loop_counter *loop;
481 OVS_CB(skb)->vport = p;
483 if (!OVS_CB(skb)->flow) {
484 struct odp_flow_key key;
485 struct tbl_node *flow_node;
488 /* Extract flow from 'skb' into 'key'. */
489 error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
490 if (unlikely(error)) {
495 if (is_frag && dp->drop_frags) {
497 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
502 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
503 flow_hash(&key), flow_cmp);
504 if (unlikely(!flow_node)) {
505 dp_output_control(dp, skb, _ODPL_MISS_NR,
506 (__force u64)OVS_CB(skb)->tun_id);
507 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
511 OVS_CB(skb)->flow = flow_cast(flow_node);
514 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
515 flow_used(OVS_CB(skb)->flow, skb);
517 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
519 /* Check whether we've looped too much. */
520 loop = loop_get_counter();
521 if (unlikely(++loop->count > MAX_LOOPS))
522 loop->looping = true;
523 if (unlikely(loop->looping)) {
524 loop_suppress(dp, acts);
529 /* Execute actions. */
530 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
533 /* Check whether sub-actions looped too much. */
534 if (unlikely(loop->looping))
535 loop_suppress(dp, acts);
538 /* Decrement loop counter. */
540 loop->looping = false;
544 /* Update datapath statistics. */
546 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
548 write_seqcount_begin(&stats->seqlock);
549 (*(u64 *)((u8 *)stats + stats_counter_off))++;
550 write_seqcount_end(&stats->seqlock);
555 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
556 * unless we broke up a GSO packet. */
557 static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
558 int queue_no, u64 arg)
560 struct sk_buff *nskb;
564 if (OVS_CB(skb)->vport)
565 port_no = OVS_CB(skb)->vport->port_no;
567 port_no = ODPP_LOCAL;
570 struct odp_msg *header;
575 err = skb_cow(skb, sizeof *header);
579 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
580 header->type = queue_no;
581 header->length = skb->len;
582 header->port = port_no;
584 skb_queue_tail(queue, skb);
592 while ((skb = nskb) != NULL) {
599 int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
602 struct dp_stats_percpu *stats;
603 struct sk_buff_head *queue;
606 WARN_ON_ONCE(skb_shared(skb));
607 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
608 queue = &dp->queues[queue_no];
610 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
613 forward_ip_summed(skb);
615 err = vswitch_skb_checksum_setup(skb);
619 /* Break apart GSO packets into their component pieces. Otherwise
620 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
621 if (skb_is_gso(skb)) {
622 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
632 err = queue_control_packets(skb, queue, queue_no, arg);
633 wake_up_interruptible(&dp->waitqueue);
640 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
642 write_seqcount_begin(&stats->seqlock);
644 write_seqcount_end(&stats->seqlock);
651 static int flush_flows(struct datapath *dp)
653 struct tbl *old_table = get_table_protected(dp);
654 struct tbl *new_table;
656 new_table = tbl_create(0);
660 rcu_assign_pointer(dp->table, new_table);
662 tbl_deferred_destroy(old_table, flow_free_tbl);
667 static int validate_actions(const struct nlattr *actions, u32 actions_len)
669 const struct nlattr *a;
672 nla_for_each_attr(a, actions, actions_len, rem) {
673 static const u32 action_lens[ODPAT_MAX + 1] = {
675 [ODPAT_CONTROLLER] = 8,
676 [ODPAT_SET_DL_TCI] = 2,
677 [ODPAT_STRIP_VLAN] = 0,
678 [ODPAT_SET_DL_SRC] = ETH_ALEN,
679 [ODPAT_SET_DL_DST] = ETH_ALEN,
680 [ODPAT_SET_NW_SRC] = 4,
681 [ODPAT_SET_NW_DST] = 4,
682 [ODPAT_SET_NW_TOS] = 1,
683 [ODPAT_SET_TP_SRC] = 2,
684 [ODPAT_SET_TP_DST] = 2,
685 [ODPAT_SET_TUNNEL] = 8,
686 [ODPAT_SET_PRIORITY] = 4,
687 [ODPAT_POP_PRIORITY] = 0,
688 [ODPAT_DROP_SPOOFED_ARP] = 0,
690 int type = nla_type(a);
692 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
699 case ODPAT_CONTROLLER:
700 case ODPAT_STRIP_VLAN:
701 case ODPAT_SET_DL_SRC:
702 case ODPAT_SET_DL_DST:
703 case ODPAT_SET_NW_SRC:
704 case ODPAT_SET_NW_DST:
705 case ODPAT_SET_TP_SRC:
706 case ODPAT_SET_TP_DST:
707 case ODPAT_SET_TUNNEL:
708 case ODPAT_SET_PRIORITY:
709 case ODPAT_POP_PRIORITY:
710 case ODPAT_DROP_SPOOFED_ARP:
711 /* No validation needed. */
715 if (nla_get_u32(a) >= DP_MAX_PORTS)
718 case ODPAT_SET_DL_TCI:
719 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
723 case ODPAT_SET_NW_TOS:
724 if (nla_get_u8(a) & INET_ECN_MASK)
739 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
741 struct sw_flow_actions *actions;
744 actions = flow_actions_alloc(flow->actions_len);
745 error = PTR_ERR(actions);
750 if (copy_from_user(actions->actions,
751 (struct nlattr __user *)flow->actions,
753 goto error_free_actions;
754 error = validate_actions(actions->actions, actions->actions_len);
756 goto error_free_actions;
763 return ERR_PTR(error);
766 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
769 struct timespec offset_ts, used, now_mono;
771 ktime_get_ts(&now_mono);
772 jiffies_to_timespec(jiffies - flow->used, &offset_ts);
773 set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
774 now_mono.tv_nsec - offset_ts.tv_nsec);
776 stats->used_sec = used.tv_sec;
777 stats->used_nsec = used.tv_nsec;
780 stats->used_nsec = 0;
783 stats->n_packets = flow->packet_count;
784 stats->n_bytes = flow->byte_count;
786 stats->tcp_flags = flow->tcp_flags;
790 static void clear_stats(struct sw_flow *flow)
794 flow->packet_count = 0;
795 flow->byte_count = 0;
798 static int expand_table(struct datapath *dp)
800 struct tbl *old_table = get_table_protected(dp);
801 struct tbl *new_table;
803 new_table = tbl_expand(old_table);
804 if (IS_ERR(new_table))
805 return PTR_ERR(new_table);
807 rcu_assign_pointer(dp->table, new_table);
808 tbl_deferred_destroy(old_table, NULL);
813 static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
814 struct odp_flow_stats *stats)
816 struct tbl_node *flow_node;
817 struct sw_flow *flow;
821 table = get_table_protected(dp);
822 flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
825 struct sw_flow_actions *acts;
828 if (!(uf->flags & ODPPF_CREATE))
831 /* Expand table, if necessary, to make room. */
832 if (tbl_count(table) >= tbl_n_buckets(table)) {
833 error = expand_table(dp);
836 table = get_table_protected(dp);
842 error = PTR_ERR(flow);
845 flow->key = uf->flow.key;
848 /* Obtain actions. */
849 acts = get_actions(&uf->flow);
850 error = PTR_ERR(acts);
852 goto error_free_flow;
853 rcu_assign_pointer(flow->sf_acts, acts);
855 /* Put flow in bucket. */
856 error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
858 goto error_free_flow_acts;
860 memset(stats, 0, sizeof(struct odp_flow_stats));
862 /* We found a matching flow. */
863 struct sw_flow_actions *old_acts, *new_acts;
865 flow = flow_cast(flow_node);
867 /* Bail out if we're not allowed to modify an existing flow. */
869 if (!(uf->flags & ODPPF_MODIFY))
873 new_acts = get_actions(&uf->flow);
874 error = PTR_ERR(new_acts);
875 if (IS_ERR(new_acts))
878 old_acts = rcu_dereference_protected(flow->sf_acts,
879 lockdep_is_held(&dp->mutex));
880 if (old_acts->actions_len != new_acts->actions_len ||
881 memcmp(old_acts->actions, new_acts->actions,
882 old_acts->actions_len)) {
883 rcu_assign_pointer(flow->sf_acts, new_acts);
884 flow_deferred_free_acts(old_acts);
889 /* Fetch stats, then clear them if necessary. */
890 spin_lock_bh(&flow->lock);
891 get_stats(flow, stats);
892 if (uf->flags & ODPPF_ZERO_STATS)
894 spin_unlock_bh(&flow->lock);
899 error_free_flow_acts:
900 kfree(flow->sf_acts);
902 flow->sf_acts = NULL;
908 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
910 struct odp_flow_stats stats;
911 struct odp_flow_put uf;
914 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
917 error = do_put_flow(dp, &uf, &stats);
921 if (copy_to_user(&ufp->flow.stats, &stats,
922 sizeof(struct odp_flow_stats)))
928 static int do_answer_query(struct datapath *dp, struct sw_flow *flow,
930 struct odp_flow_stats __user *ustats,
931 struct nlattr __user *actions,
932 u32 __user *actions_lenp)
934 struct sw_flow_actions *sf_acts;
935 struct odp_flow_stats stats;
938 spin_lock_bh(&flow->lock);
939 get_stats(flow, &stats);
940 if (query_flags & ODPFF_ZERO_TCP_FLAGS)
943 spin_unlock_bh(&flow->lock);
945 if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
946 get_user(actions_len, actions_lenp))
952 sf_acts = rcu_dereference_protected(flow->sf_acts,
953 lockdep_is_held(&dp->mutex));
954 if (put_user(sf_acts->actions_len, actions_lenp) ||
955 (actions && copy_to_user(actions, sf_acts->actions,
956 min(sf_acts->actions_len, actions_len))))
962 static int answer_query(struct datapath *dp, struct sw_flow *flow,
963 u32 query_flags, struct odp_flow __user *ufp)
965 struct nlattr __user *actions;
967 if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
970 return do_answer_query(dp, flow, query_flags,
971 &ufp->stats, actions, &ufp->actions_len);
974 static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
976 struct tbl *table = get_table_protected(dp);
977 struct tbl_node *flow_node;
980 flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
982 return ERR_PTR(-ENOENT);
984 error = tbl_remove(table, flow_node);
986 return ERR_PTR(error);
988 /* XXX Returned flow_node's statistics might lose a few packets, since
989 * other CPUs can be using this flow. We used to synchronize_rcu() to
990 * make sure that we get completely accurate stats, but that blows our
991 * performance, badly. */
992 return flow_cast(flow_node);
995 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
997 struct sw_flow *flow;
1001 if (copy_from_user(&uf, ufp, sizeof uf))
1004 flow = do_del_flow(dp, &uf.key);
1006 return PTR_ERR(flow);
1008 error = answer_query(dp, flow, 0, ufp);
1009 flow_deferred_free(flow);
1013 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1015 struct tbl *table = get_table_protected(dp);
1018 for (i = 0; i < flowvec->n_flows; i++) {
1019 struct odp_flow __user *ufp = (struct odp_flow __user *)&flowvec->flows[i];
1021 struct tbl_node *flow_node;
1024 if (copy_from_user(&uf, ufp, sizeof uf))
1027 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1029 error = put_user(ENOENT, &ufp->stats.error);
1031 error = answer_query(dp, flow_cast(flow_node), uf.flags, ufp);
1035 return flowvec->n_flows;
1038 struct list_flows_cbdata {
1039 struct datapath *dp;
1040 struct odp_flow __user *uflows;
1045 static int list_flow(struct tbl_node *node, void *cbdata_)
1047 struct sw_flow *flow = flow_cast(node);
1048 struct list_flows_cbdata *cbdata = cbdata_;
1049 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1052 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1054 error = answer_query(cbdata->dp, flow, 0, ufp);
1058 if (cbdata->listed_flows >= cbdata->n_flows)
1059 return cbdata->listed_flows;
1063 static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1065 struct list_flows_cbdata cbdata;
1068 if (!flowvec->n_flows)
1072 cbdata.uflows = (struct odp_flow __user *)flowvec->flows;
1073 cbdata.n_flows = flowvec->n_flows;
1074 cbdata.listed_flows = 0;
1076 error = tbl_foreach(get_table_protected(dp), list_flow, &cbdata);
1077 return error ? error : cbdata.listed_flows;
1080 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1081 int (*function)(struct datapath *,
1082 const struct odp_flowvec *))
1084 struct odp_flowvec __user *uflowvec;
1085 struct odp_flowvec flowvec;
1088 uflowvec = (struct odp_flowvec __user *)argp;
1089 if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1092 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1095 retval = function(dp, &flowvec);
1096 return (retval < 0 ? retval
1097 : retval == flowvec.n_flows ? 0
1098 : put_user(retval, &uflowvec->n_flows));
1101 static int do_execute(struct datapath *dp, const struct odp_execute *execute)
1103 struct odp_flow_key key;
1104 struct sk_buff *skb;
1105 struct sw_flow_actions *actions;
1111 if (execute->length < ETH_HLEN || execute->length > 65535)
1114 actions = flow_actions_alloc(execute->actions_len);
1115 if (IS_ERR(actions)) {
1116 err = PTR_ERR(actions);
1121 if (copy_from_user(actions->actions,
1122 (struct nlattr __user *)execute->actions, execute->actions_len))
1123 goto error_free_actions;
1125 err = validate_actions(actions->actions, execute->actions_len);
1127 goto error_free_actions;
1130 skb = alloc_skb(execute->length, GFP_KERNEL);
1132 goto error_free_actions;
1135 if (copy_from_user(skb_put(skb, execute->length),
1136 (const void __user *)execute->data,
1138 goto error_free_skb;
1140 skb_reset_mac_header(skb);
1143 /* Normally, setting the skb 'protocol' field would be handled by a
1144 * call to eth_type_trans(), but it assumes there's a sending
1145 * device, which we may not have. */
1146 if (ntohs(eth->h_proto) >= 1536)
1147 skb->protocol = eth->h_proto;
1149 skb->protocol = htons(ETH_P_802_2);
1151 err = flow_extract(skb, -1, &key, &is_frag);
1153 goto error_free_skb;
1156 err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
1170 static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
1172 struct odp_execute execute;
1174 if (copy_from_user(&execute, executep, sizeof execute))
1177 return do_execute(dp, &execute);
1180 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1182 struct tbl *table = get_table_protected(dp);
1183 struct odp_stats stats;
1186 stats.n_flows = tbl_count(table);
1187 stats.cur_capacity = tbl_n_buckets(table);
1188 stats.max_capacity = TBL_MAX_BUCKETS;
1189 stats.n_ports = dp->n_ports;
1190 stats.max_ports = DP_MAX_PORTS;
1191 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1192 for_each_possible_cpu(i) {
1193 const struct dp_stats_percpu *percpu_stats;
1194 struct dp_stats_percpu local_stats;
1197 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1200 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
1201 local_stats = *percpu_stats;
1202 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
1204 stats.n_frags += local_stats.n_frags;
1205 stats.n_hit += local_stats.n_hit;
1206 stats.n_missed += local_stats.n_missed;
1207 stats.n_lost += local_stats.n_lost;
1209 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1210 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1211 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1214 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1215 int dp_min_mtu(const struct datapath *dp)
1222 list_for_each_entry_rcu (p, &dp->port_list, node) {
1225 /* Skip any internal ports, since that's what we're trying to
1227 if (is_internal_vport(p))
1230 dev_mtu = vport_get_mtu(p);
1231 if (!mtu || dev_mtu < mtu)
1235 return mtu ? mtu : ETH_DATA_LEN;
1238 /* Sets the MTU of all datapath devices to the minimum of the ports. Must
1239 * be called with RTNL lock. */
1240 void set_internal_devs_mtu(const struct datapath *dp)
1247 mtu = dp_min_mtu(dp);
1249 list_for_each_entry_rcu (p, &dp->port_list, node) {
1250 if (is_internal_vport(p))
1251 vport_set_mtu(p, mtu);
1255 static int put_port(const struct vport *p, struct odp_port __user *uop)
1259 memset(&op, 0, sizeof op);
1262 strncpy(op.devname, vport_get_name(p), sizeof op.devname);
1263 strncpy(op.type, vport_get_type(p), sizeof op.type);
1266 op.port = p->port_no;
1268 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1271 static int query_port(struct datapath *dp, struct odp_port __user *uport)
1273 struct odp_port port;
1275 if (copy_from_user(&port, uport, sizeof port))
1278 if (port.devname[0]) {
1279 struct vport *vport;
1282 port.devname[IFNAMSIZ - 1] = '\0';
1287 vport = vport_locate(port.devname);
1292 if (vport->dp != dp) {
1297 port.port = vport->port_no;
1306 if (port.port >= DP_MAX_PORTS)
1308 if (!dp->ports[port.port])
1312 return put_port(dp->ports[port.port], uport);
1315 static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
1322 list_for_each_entry_rcu (p, &dp->port_list, node) {
1323 if (put_port(p, &uports[idx]))
1325 if (idx++ >= n_ports)
1332 static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
1334 struct odp_portvec pv;
1337 if (copy_from_user(&pv, upv, sizeof pv))
1340 retval = do_list_ports(dp, (struct odp_port __user *)pv.ports,
1345 return put_user(retval, &upv->n_ports);
1348 static int get_listen_mask(const struct file *f)
1350 return (long)f->private_data;
1353 static void set_listen_mask(struct file *f, int listen_mask)
1355 f->private_data = (void*)(long)listen_mask;
1358 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1361 int dp_idx = iminor(f->f_dentry->d_inode);
1362 struct datapath *dp;
1363 int drop_frags, listeners, port_no;
1364 unsigned int sflow_probability;
1367 /* Handle commands with special locking requirements up front. */
1370 err = create_dp(dp_idx, (char __user *)argp);
1373 case ODP_DP_DESTROY:
1374 err = destroy_dp(dp_idx);
1377 case ODP_VPORT_ATTACH:
1378 err = attach_port(dp_idx, (struct odp_port __user *)argp);
1381 case ODP_VPORT_DETACH:
1382 err = get_user(port_no, (int __user *)argp);
1384 err = detach_port(dp_idx, port_no);
1388 err = vport_user_mod((struct odp_port __user *)argp);
1391 case ODP_VPORT_STATS_GET:
1392 err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
1395 case ODP_VPORT_STATS_SET:
1396 err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
1399 case ODP_VPORT_ETHER_GET:
1400 err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
1403 case ODP_VPORT_ETHER_SET:
1404 err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
1407 case ODP_VPORT_MTU_GET:
1408 err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
1411 case ODP_VPORT_MTU_SET:
1412 err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
1416 dp = get_dp_locked(dp_idx);
1423 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1426 case ODP_GET_DROP_FRAGS:
1427 err = put_user(dp->drop_frags, (int __user *)argp);
1430 case ODP_SET_DROP_FRAGS:
1431 err = get_user(drop_frags, (int __user *)argp);
1435 if (drop_frags != 0 && drop_frags != 1)
1437 dp->drop_frags = drop_frags;
1441 case ODP_GET_LISTEN_MASK:
1442 err = put_user(get_listen_mask(f), (int __user *)argp);
1445 case ODP_SET_LISTEN_MASK:
1446 err = get_user(listeners, (int __user *)argp);
1450 if (listeners & ~ODPL_ALL)
1453 set_listen_mask(f, listeners);
1456 case ODP_GET_SFLOW_PROBABILITY:
1457 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1460 case ODP_SET_SFLOW_PROBABILITY:
1461 err = get_user(sflow_probability, (unsigned int __user *)argp);
1463 dp->sflow_probability = sflow_probability;
1466 case ODP_VPORT_QUERY:
1467 err = query_port(dp, (struct odp_port __user *)argp);
1470 case ODP_VPORT_LIST:
1471 err = list_ports(dp, (struct odp_portvec __user *)argp);
1474 case ODP_FLOW_FLUSH:
1475 err = flush_flows(dp);
1479 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1483 err = del_flow(dp, (struct odp_flow __user *)argp);
1487 err = do_flowvec_ioctl(dp, argp, do_query_flows);
1491 err = do_flowvec_ioctl(dp, argp, do_list_flows);
1495 err = execute_packet(dp, (struct odp_execute __user *)argp);
1502 mutex_unlock(&dp->mutex);
1507 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1510 for (i = 0; i < DP_N_QUEUES; i++) {
1511 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1517 #ifdef CONFIG_COMPAT
1518 static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
1520 struct compat_odp_portvec pv;
1523 if (copy_from_user(&pv, upv, sizeof pv))
1526 retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
1530 return put_user(retval, &upv->n_ports);
1533 static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
1535 compat_uptr_t actions;
1537 if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
1538 __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
1539 __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
1540 __get_user(actions, &compat->actions) ||
1541 __get_user(flow->actions_len, &compat->actions_len) ||
1542 __get_user(flow->flags, &compat->flags))
1545 flow->actions = (struct nlattr __force *)compat_ptr(actions);
1549 static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
1551 struct odp_flow_stats stats;
1552 struct odp_flow_put fp;
1555 if (compat_get_flow(&fp.flow, &ufp->flow) ||
1556 get_user(fp.flags, &ufp->flags))
1559 error = do_put_flow(dp, &fp, &stats);
1563 if (copy_to_user(&ufp->flow.stats, &stats,
1564 sizeof(struct odp_flow_stats)))
1570 static int compat_answer_query(struct datapath *dp, struct sw_flow *flow,
1572 struct compat_odp_flow __user *ufp)
1574 compat_uptr_t actions;
1576 if (get_user(actions, &ufp->actions))
1579 return do_answer_query(dp, flow, query_flags, &ufp->stats,
1580 compat_ptr(actions), &ufp->actions_len);
1583 static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
1585 struct sw_flow *flow;
1589 if (compat_get_flow(&uf, ufp))
1592 flow = do_del_flow(dp, &uf.key);
1594 return PTR_ERR(flow);
1596 error = compat_answer_query(dp, flow, 0, ufp);
1597 flow_deferred_free(flow);
1601 static int compat_query_flows(struct datapath *dp,
1602 struct compat_odp_flow __user *flows,
1605 struct tbl *table = get_table_protected(dp);
1608 for (i = 0; i < n_flows; i++) {
1609 struct compat_odp_flow __user *ufp = &flows[i];
1611 struct tbl_node *flow_node;
1614 if (compat_get_flow(&uf, ufp))
1617 flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
1619 error = put_user(ENOENT, &ufp->stats.error);
1621 error = compat_answer_query(dp, flow_cast(flow_node),
1629 struct compat_list_flows_cbdata {
1630 struct datapath *dp;
1631 struct compat_odp_flow __user *uflows;
1636 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
1638 struct sw_flow *flow = flow_cast(node);
1639 struct compat_list_flows_cbdata *cbdata = cbdata_;
1640 struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1643 if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1645 error = compat_answer_query(cbdata->dp, flow, 0, ufp);
1649 if (cbdata->listed_flows >= cbdata->n_flows)
1650 return cbdata->listed_flows;
1654 static int compat_list_flows(struct datapath *dp,
1655 struct compat_odp_flow __user *flows, u32 n_flows)
1657 struct compat_list_flows_cbdata cbdata;
1664 cbdata.uflows = flows;
1665 cbdata.n_flows = n_flows;
1666 cbdata.listed_flows = 0;
1668 error = tbl_foreach(get_table_protected(dp), compat_list_flow, &cbdata);
1669 return error ? error : cbdata.listed_flows;
1672 static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1673 int (*function)(struct datapath *,
1674 struct compat_odp_flow __user *,
1677 struct compat_odp_flowvec __user *uflowvec;
1678 struct compat_odp_flow __user *flows;
1679 struct compat_odp_flowvec flowvec;
1682 uflowvec = compat_ptr(argp);
1683 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1684 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1687 if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
1690 flows = compat_ptr(flowvec.flows);
1691 if (!access_ok(VERIFY_WRITE, flows,
1692 flowvec.n_flows * sizeof(struct compat_odp_flow)))
1695 retval = function(dp, flows, flowvec.n_flows);
1696 return (retval < 0 ? retval
1697 : retval == flowvec.n_flows ? 0
1698 : put_user(retval, &uflowvec->n_flows));
1701 static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
1703 struct odp_execute execute;
1704 compat_uptr_t actions;
1707 if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
1708 __get_user(actions, &uexecute->actions) ||
1709 __get_user(execute.actions_len, &uexecute->actions_len) ||
1710 __get_user(data, &uexecute->data) ||
1711 __get_user(execute.length, &uexecute->length))
1714 execute.actions = (struct nlattr __force *)compat_ptr(actions);
1715 execute.data = (const void __force *)compat_ptr(data);
1717 return do_execute(dp, &execute);
1720 static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
1722 int dp_idx = iminor(f->f_dentry->d_inode);
1723 struct datapath *dp;
1727 case ODP_DP_DESTROY:
1728 case ODP_FLOW_FLUSH:
1729 /* Ioctls that don't need any translation at all. */
1730 return openvswitch_ioctl(f, cmd, argp);
1733 case ODP_VPORT_ATTACH:
1734 case ODP_VPORT_DETACH:
1736 case ODP_VPORT_MTU_SET:
1737 case ODP_VPORT_MTU_GET:
1738 case ODP_VPORT_ETHER_SET:
1739 case ODP_VPORT_ETHER_GET:
1740 case ODP_VPORT_STATS_SET:
1741 case ODP_VPORT_STATS_GET:
1743 case ODP_GET_DROP_FRAGS:
1744 case ODP_SET_DROP_FRAGS:
1745 case ODP_SET_LISTEN_MASK:
1746 case ODP_GET_LISTEN_MASK:
1747 case ODP_SET_SFLOW_PROBABILITY:
1748 case ODP_GET_SFLOW_PROBABILITY:
1749 case ODP_VPORT_QUERY:
1750 /* Ioctls that just need their pointer argument extended. */
1751 return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
1754 dp = get_dp_locked(dp_idx);
1760 case ODP_VPORT_LIST32:
1761 err = compat_list_ports(dp, compat_ptr(argp));
1764 case ODP_FLOW_PUT32:
1765 err = compat_put_flow(dp, compat_ptr(argp));
1768 case ODP_FLOW_DEL32:
1769 err = compat_del_flow(dp, compat_ptr(argp));
1772 case ODP_FLOW_GET32:
1773 err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
1776 case ODP_FLOW_LIST32:
1777 err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
1781 err = compat_execute(dp, compat_ptr(argp));
1788 mutex_unlock(&dp->mutex);
1794 /* Unfortunately this function is not exported so this is a verbatim copy
1795 * from net/core/datagram.c in 2.6.30. */
1796 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
1797 u8 __user *to, int len,
1800 int start = skb_headlen(skb);
1802 int i, copy = start - offset;
1809 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
1813 if ((len -= copy) == 0)
1820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1823 WARN_ON(start > offset + len);
1825 end = start + skb_shinfo(skb)->frags[i].size;
1826 if ((copy = end - offset) > 0) {
1830 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1831 struct page *page = frag->page;
1836 csum2 = csum_and_copy_to_user(vaddr +
1843 *csump = csum_block_add(*csump, csum2, pos);
1853 if (skb_shinfo(skb)->frag_list) {
1854 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1856 for (; list; list=list->next) {
1859 WARN_ON(start > offset + len);
1861 end = start + list->len;
1862 if ((copy = end - offset) > 0) {
1866 if (skb_copy_and_csum_datagram(list,
1871 *csump = csum_block_add(*csump, csum2, pos);
1872 if ((len -= copy) == 0)
1888 static ssize_t openvswitch_read(struct file *f, char __user *buf,
1889 size_t nbytes, loff_t *ppos)
1891 int listeners = get_listen_mask(f);
1892 int dp_idx = iminor(f->f_dentry->d_inode);
1893 struct datapath *dp = get_dp_locked(dp_idx);
1894 struct sk_buff *skb;
1895 size_t copy_bytes, tot_copy_bytes;
1901 if (nbytes == 0 || !listeners)
1907 for (i = 0; i < DP_N_QUEUES; i++) {
1908 if (listeners & (1 << i)) {
1909 skb = skb_dequeue(&dp->queues[i]);
1915 if (f->f_flags & O_NONBLOCK) {
1920 wait_event_interruptible(dp->waitqueue,
1921 dp_has_packet_of_interest(dp,
1924 if (signal_pending(current)) {
1925 retval = -ERESTARTSYS;
1930 mutex_unlock(&dp->mutex);
1932 copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
1935 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1936 if (copy_bytes == skb->len) {
1938 u16 csum_start, csum_offset;
1940 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
1941 csum_start -= skb_headroom(skb);
1943 BUG_ON(csum_start >= skb_headlen(skb));
1944 retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
1945 copy_bytes - csum_start, &csum);
1947 __sum16 __user *csump;
1949 copy_bytes = csum_start;
1950 csump = (__sum16 __user *)(buf + csum_start + csum_offset);
1952 BUG_ON((char __user *)csump + sizeof(__sum16) >
1954 put_user(csum_fold(csum), csump);
1957 retval = skb_checksum_help(skb);
1964 iov.iov_len = copy_bytes;
1965 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1969 retval = tot_copy_bytes;
1975 mutex_unlock(&dp->mutex);
1979 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1981 int dp_idx = iminor(file->f_dentry->d_inode);
1982 struct datapath *dp = get_dp_locked(dp_idx);
1987 poll_wait(file, &dp->waitqueue, wait);
1988 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1989 mask |= POLLIN | POLLRDNORM;
1990 mutex_unlock(&dp->mutex);
1992 mask = POLLIN | POLLRDNORM | POLLHUP;
1997 static struct file_operations openvswitch_fops = {
1998 .read = openvswitch_read,
1999 .poll = openvswitch_poll,
2000 .unlocked_ioctl = openvswitch_ioctl,
2001 #ifdef CONFIG_COMPAT
2002 .compat_ioctl = openvswitch_compat_ioctl,
2008 static int __init dp_init(void)
2010 struct sk_buff *dummy_skb;
2013 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2015 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2023 goto error_flow_exit;
2025 err = register_netdevice_notifier(&dp_device_notifier);
2027 goto error_vport_exit;
2029 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
2031 goto error_unreg_notifier;
2035 error_unreg_notifier:
2036 unregister_netdevice_notifier(&dp_device_notifier);
2045 static void dp_cleanup(void)
2048 unregister_chrdev(major, "openvswitch");
2049 unregister_netdevice_notifier(&dp_device_notifier);
2054 module_init(dp_init);
2055 module_exit(dp_cleanup);
2057 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2058 MODULE_LICENSE("GPL");