History

geneve: add initial netdev driver for GENEVE tunnels

This is an initial implementation of a netdev driver for GENEVE tunnels. This implementation uses a fixed UDP port, and only supports point-to-point links with specific partner endpoints. Only IPv4 links are supported at this time.

John W. Linville committed 11 years ago

Commit 2d07dc71 parent d37d29c4 files · +527 −0Browse files

Unified · Split

4 files
  1. drivers/net/Kconfig +14 −0
  2. drivers/net/Makefile +1 −0
  3. drivers/net/geneve.c +503 −0
  4. include/uapi/linux/if_link.h +9 −0
drivers/net/Kconfig+14−0
@@ -179,6 +179,20 @@ config VXLAN · 178 unmodified lines
179179 To compile this driver as a module, choose M here: the module
180180 will be called vxlan.
181181
182+config GENEVE
183+ tristate "Generic Network Virtualization Encapsulation netdev"
184+ depends on INET && GENEVE_CORE
185+ select NET_IP_TUNNEL
186+ ---help---
187+ This allows one to create geneve virtual interfaces that provide
188+ Layer 2 Networks over Layer 3 Networks. GENEVE is often used
189+ to tunnel virtual network infrastructure in virtualized environments.
190+ For more information see:
191+ http://tools.ietf.org/html/draft-gross-geneve-02
192+
193+ To compile this driver as a module, choose M here: the module
194+ will be called geneve.
195+
182196 config NETCONSOLE
183197 tristate "Network console logging support"
184198 ---help---
drivers/net/Makefile+1−0
@@ -23,6 +23,7 @@ obj-$(CONFIG_TUN) += tun.o · 22 unmodified lines
2323 obj-$(CONFIG_VETH) += veth.o
2424 obj-$(CONFIG_VIRTIO_NET) += virtio_net.o
2525 obj-$(CONFIG_VXLAN) += vxlan.o
26+obj-$(CONFIG_GENEVE) += geneve.o
2627 obj-$(CONFIG_NLMON) += nlmon.o
2728
2829 #
drivers/net/geneve.c+503−0
@@ -0,0 +1,503 @@
1+/*
2+ * GENEVE: Generic Network Virtualization Encapsulation
3+ *
4+ * Copyright (c) 2015 Red Hat, Inc.
5+ *
6+ * This program is free software; you can redistribute it and/or modify
7+ * it under the terms of the GNU General Public License version 2 as
8+ * published by the Free Software Foundation.
9+ */
10+
11+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12+
13+#include <linux/kernel.h>
14+#include <linux/module.h>
15+#include <linux/netdevice.h>
16+#include <linux/etherdevice.h>
17+#include <linux/hash.h>
18+#include <net/rtnetlink.h>
19+#include <net/geneve.h>
20+
21+#define GENEVE_NETDEV_VER "0.6"
22+
23+#define GENEVE_UDP_PORT 6081
24+
25+#define GENEVE_N_VID (1u << 24)
26+#define GENEVE_VID_MASK (GENEVE_N_VID - 1)
27+
28+#define VNI_HASH_BITS 10
29+#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
30+
31+static bool log_ecn_error = true;
32+module_param(log_ecn_error, bool, 0644);
33+MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34+
35+/* per-network namespace private data for this module */
36+struct geneve_net {
37+ struct list_head geneve_list;
38+ struct hlist_head vni_list[VNI_HASH_SIZE];
39+};
40+
41+/* Pseudo network device */
42+struct geneve_dev {
43+ struct hlist_node hlist; /* vni hash table */
44+ struct net *net; /* netns for packet i/o */
45+ struct net_device *dev; /* netdev for geneve tunnel */
46+ struct geneve_sock *sock; /* socket used for geneve tunnel */
47+ u8 vni[3]; /* virtual network ID for tunnel */
48+ struct sockaddr_in remote; /* IPv4 address for link partner */
49+ struct list_head next; /* geneve's per namespace list */
50+};
51+
52+static int geneve_net_id;
53+
54+static inline __u32 geneve_net_vni_hash(u8 vni[3])
55+{
56+ __u32 vnid;
57+
58+ vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
59+ return hash_32(vnid, VNI_HASH_BITS);
60+}
61+
62+/* geneve receive/decap routine */
63+static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
64+{
65+ struct genevehdr *gnvh = geneve_hdr(skb);
66+ struct geneve_dev *dummy, *geneve = NULL;
67+ struct geneve_net *gn;
68+ struct iphdr *iph = NULL;
69+ struct pcpu_sw_netstats *stats;
70+ struct hlist_head *vni_list_head;
71+ int err = 0;
72+ __u32 hash;
73+
74+ iph = ip_hdr(skb); /* Still outer IP header... */
75+
76+ gn = gs->rcv_data;
77+
78+ /* Find the device for this VNI */
79+ hash = geneve_net_vni_hash(gnvh->vni);
80+ vni_list_head = &gn->vni_list[hash];
81+ hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
82+ if (!memcmp(gnvh->vni, dummy->vni, sizeof(dummy->vni)) &&
83+ iph->saddr == dummy->remote.sin_addr.s_addr) {
84+ geneve = dummy;
85+ break;
86+ }
87+ }
88+ if (!geneve)
89+ goto drop;
90+
91+ /* Drop packets w/ critical options,
92+ * since we don't support any...
93+ */
94+ if (gnvh->critical)
95+ goto drop;
96+
97+ skb_reset_mac_header(skb);
98+ skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
99+ skb->protocol = eth_type_trans(skb, geneve->dev);
100+ skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
101+
102+ /* Ignore packet loops (and multicast echo) */
103+ if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
104+ goto drop;
105+
106+ skb_reset_network_header(skb);
107+
108+ iph = ip_hdr(skb); /* Now inner IP header... */
109+ err = IP_ECN_decapsulate(iph, skb);
110+
111+ if (unlikely(err)) {
112+ if (log_ecn_error)
113+ net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
114+ &iph->saddr, iph->tos);
115+ if (err > 1) {
116+ ++geneve->dev->stats.rx_frame_errors;
117+ ++geneve->dev->stats.rx_errors;
118+ goto drop;
119+ }
120+ }
121+
122+ stats = this_cpu_ptr(geneve->dev->tstats);
123+ u64_stats_update_begin(&stats->syncp);
124+ stats->rx_packets++;
125+ stats->rx_bytes += skb->len;
126+ u64_stats_update_end(&stats->syncp);
127+
128+ netif_rx(skb);
129+
130+ return;
131+drop:
132+ /* Consume bad packet */
133+ kfree_skb(skb);
134+}
135+
136+/* Setup stats when device is created */
137+static int geneve_init(struct net_device *dev)
138+{
139+ dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
140+ if (!dev->tstats)
141+ return -ENOMEM;
142+
143+ return 0;
144+}
145+
146+static void geneve_uninit(struct net_device *dev)
147+{
148+ free_percpu(dev->tstats);
149+}
150+
151+static int geneve_open(struct net_device *dev)
152+{
153+ struct geneve_dev *geneve = netdev_priv(dev);
154+ struct net *net = geneve->net;
155+ struct geneve_net *gn = net_generic(geneve->net, geneve_net_id);
156+ struct geneve_sock *gs;
157+
158+ gs = geneve_sock_add(net, htons(GENEVE_UDP_PORT), geneve_rx, gn,
159+ false, false);
160+ if (IS_ERR(gs))
161+ return PTR_ERR(gs);
162+
163+ geneve->sock = gs;
164+
165+ return 0;
166+}
167+
168+static int geneve_stop(struct net_device *dev)
169+{
170+ struct geneve_dev *geneve = netdev_priv(dev);
171+ struct geneve_sock *gs = geneve->sock;
172+
173+ geneve_sock_release(gs);
174+
175+ return 0;
176+}
177+
178+static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
179+{
180+ struct geneve_dev *geneve = netdev_priv(dev);
181+ struct geneve_sock *gs = geneve->sock;
182+ struct rtable *rt = NULL;
183+ const struct iphdr *iip; /* interior IP header */
184+ struct flowi4 fl4;
185+ int err;
186+ __be16 sport;
187+ __u8 tos, ttl = 0;
188+
189+ iip = ip_hdr(skb);
190+
191+ skb_reset_mac_header(skb);
192+
193+ /* TODO: port min/max limits should be configurable */
194+ sport = udp_flow_src_port(dev_net(dev), skb, 0, 0, true);
195+
196+ memset(&fl4, 0, sizeof(fl4));
197+ fl4.daddr = geneve->remote.sin_addr.s_addr;
198+ rt = ip_route_output_key(geneve->net, &fl4);
199+ if (IS_ERR(rt)) {
200+ netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
201+ dev->stats.tx_carrier_errors++;
202+ goto tx_error;
203+ }
204+ if (rt->dst.dev == dev) { /* is this necessary? */
205+ netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
206+ dev->stats.collisions++;
207+ goto rt_tx_error;
208+ }
209+
210+ /* TODO: tos and ttl should be configurable */
211+
212+ tos = ip_tunnel_ecn_encap(0, iip, skb);
213+
214+ if (IN_MULTICAST(ntohl(fl4.daddr)))
215+ ttl = 1;
216+
217+ ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
218+
219+ /* no need to handle local destination and encap bypass...yet... */
220+
221+ err = geneve_xmit_skb(gs, rt, skb, fl4.saddr, fl4.daddr,
222+ tos, ttl, 0, sport, htons(GENEVE_UDP_PORT), 0,
223+ geneve->vni, 0, NULL, false,
224+ !net_eq(geneve->net, dev_net(geneve->dev)));
225+ if (err < 0)
226+ ip_rt_put(rt);
227+
228+ iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
229+
230+ return NETDEV_TX_OK;
231+
232+rt_tx_error:
233+ ip_rt_put(rt);
234+tx_error:
235+ dev->stats.tx_errors++;
236+ dev_kfree_skb(skb);
237+ return NETDEV_TX_OK;
238+}
239+
240+static const struct net_device_ops geneve_netdev_ops = {
241+ .ndo_init = geneve_init,
242+ .ndo_uninit = geneve_uninit,
243+ .ndo_open = geneve_open,
244+ .ndo_stop = geneve_stop,
245+ .ndo_start_xmit = geneve_xmit,
246+ .ndo_get_stats64 = ip_tunnel_get_stats64,
247+ .ndo_change_mtu = eth_change_mtu,
248+ .ndo_validate_addr = eth_validate_addr,
249+ .ndo_set_mac_address = eth_mac_addr,
250+};
251+
252+static void geneve_get_drvinfo(struct net_device *dev,
253+ struct ethtool_drvinfo *drvinfo)
254+{
255+ strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
256+ strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
257+}
258+
259+static const struct ethtool_ops geneve_ethtool_ops = {
260+ .get_drvinfo = geneve_get_drvinfo,
261+ .get_link = ethtool_op_get_link,
262+};
263+
264+/* Info for udev, that this is a virtual tunnel endpoint */
265+static struct device_type geneve_type = {
266+ .name = "geneve",
267+};
268+
269+/* Initialize the device structure. */
270+static void geneve_setup(struct net_device *dev)
271+{
272+ ether_setup(dev);
273+
274+ dev->netdev_ops = &geneve_netdev_ops;
275+ dev->ethtool_ops = &geneve_ethtool_ops;
276+ dev->destructor = free_netdev;
277+
278+ SET_NETDEV_DEVTYPE(dev, &geneve_type);
279+
280+ dev->tx_queue_len = 0;
281+ dev->features |= NETIF_F_LLTX;
282+ dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
283+ dev->features |= NETIF_F_RXCSUM;
284+ dev->features |= NETIF_F_GSO_SOFTWARE;
285+
286+ dev->vlan_features = dev->features;
287+ dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
288+
289+ dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
290+ dev->hw_features |= NETIF_F_GSO_SOFTWARE;
291+ dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
292+
293+ netif_keep_dst(dev);
294+ dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
295+}
296+
297+static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
298+ [IFLA_GENEVE_ID] = { .type = NLA_U32 },
299+ [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
300+};
301+
302+static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
303+{
304+ if (tb[IFLA_ADDRESS]) {
305+ if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
306+ return -EINVAL;
307+
308+ if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
309+ return -EADDRNOTAVAIL;
310+ }
311+
312+ if (!data)
313+ return -EINVAL;
314+
315+ if (data[IFLA_GENEVE_ID]) {
316+ __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
317+
318+ if (vni >= GENEVE_VID_MASK)
319+ return -ERANGE;
320+ }
321+
322+ return 0;
323+}
324+
325+static int geneve_newlink(struct net *net, struct net_device *dev,
326+ struct nlattr *tb[], struct nlattr *data[])
327+{
328+ struct geneve_net *gn = net_generic(net, geneve_net_id);
329+ struct geneve_dev *dummy, *geneve = netdev_priv(dev);
330+ struct hlist_head *vni_list_head;
331+ struct sockaddr_in remote; /* IPv4 address for link partner */
332+ __u32 vni, hash;
333+ int err;
334+
335+ if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
336+ return -EINVAL;
337+
338+ geneve->net = net;
339+ geneve->dev = dev;
340+
341+ vni = nla_get_u32(data[IFLA_GENEVE_ID]);
342+ geneve->vni[0] = (vni & 0x00ff0000) >> 16;
343+ geneve->vni[1] = (vni & 0x0000ff00) >> 8;
344+ geneve->vni[2] = vni & 0x000000ff;
345+
346+ geneve->remote.sin_addr.s_addr =
347+ nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
348+ if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
349+ return -EINVAL;
350+
351+ remote = geneve->remote;
352+ hash = geneve_net_vni_hash(geneve->vni);
353+ vni_list_head = &gn->vni_list[hash];
354+ hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
355+ if (!memcmp(geneve->vni, dummy->vni, sizeof(dummy->vni)) &&
356+ !memcmp(&remote, &dummy->remote, sizeof(dummy->remote)))
357+ return -EBUSY;
358+ }
359+
360+ if (tb[IFLA_ADDRESS] == NULL)
361+ eth_hw_addr_random(dev);
362+
363+ err = register_netdevice(dev);
364+ if (err)
365+ return err;
366+
367+ list_add(&geneve->next, &gn->geneve_list);
368+
369+ hlist_add_head_rcu(&geneve->hlist, &gn->vni_list[hash]);
370+
371+ return 0;
372+}
373+
374+static void geneve_dellink(struct net_device *dev, struct list_head *head)
375+{
376+ struct geneve_dev *geneve = netdev_priv(dev);
377+
378+ if (!hlist_unhashed(&geneve->hlist))
379+ hlist_del_rcu(&geneve->hlist);
380+
381+ list_del(&geneve->next);
382+ unregister_netdevice_queue(dev, head);
383+}
384+
385+static size_t geneve_get_size(const struct net_device *dev)
386+{
387+ return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
388+ nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
389+ 0;
390+}
391+
392+static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
393+{
394+ struct geneve_dev *geneve = netdev_priv(dev);
395+ __u32 vni;
396+
397+ vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
398+ if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
399+ goto nla_put_failure;
400+
401+ if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
402+ geneve->remote.sin_addr.s_addr))
403+ goto nla_put_failure;
404+
405+ return 0;
406+
407+nla_put_failure:
408+ return -EMSGSIZE;
409+}
410+
411+static struct rtnl_link_ops geneve_link_ops __read_mostly = {
412+ .kind = "geneve",
413+ .maxtype = IFLA_GENEVE_MAX,
414+ .policy = geneve_policy,
415+ .priv_size = sizeof(struct geneve_dev),
416+ .setup = geneve_setup,
417+ .validate = geneve_validate,
418+ .newlink = geneve_newlink,
419+ .dellink = geneve_dellink,
420+ .get_size = geneve_get_size,
421+ .fill_info = geneve_fill_info,
422+};
423+
424+static __net_init int geneve_init_net(struct net *net)
425+{
426+ struct geneve_net *gn = net_generic(net, geneve_net_id);
427+ unsigned int h;
428+
429+ INIT_LIST_HEAD(&gn->geneve_list);
430+
431+ for (h = 0; h < VNI_HASH_SIZE; ++h)
432+ INIT_HLIST_HEAD(&gn->vni_list[h]);
433+
434+ return 0;
435+}
436+
437+static void __net_exit geneve_exit_net(struct net *net)
438+{
439+ struct geneve_net *gn = net_generic(net, geneve_net_id);
440+ struct geneve_dev *geneve, *next;
441+ struct net_device *dev, *aux;
442+ LIST_HEAD(list);
443+
444+ rtnl_lock();
445+
446+ /* gather any geneve devices that were moved into this ns */
447+ for_each_netdev_safe(net, dev, aux)
448+ if (dev->rtnl_link_ops == &geneve_link_ops)
449+ unregister_netdevice_queue(dev, &list);
450+
451+ /* now gather any other geneve devices that were created in this ns */
452+ list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
453+ /* If geneve->dev is in the same netns, it was already added
454+ * to the list by the previous loop.
455+ */
456+ if (!net_eq(dev_net(geneve->dev), net))
457+ unregister_netdevice_queue(geneve->dev, &list);
458+ }
459+
460+ /* unregister the devices gathered above */
461+ unregister_netdevice_many(&list);
462+ rtnl_unlock();
463+}
464+
465+static struct pernet_operations geneve_net_ops = {
466+ .init = geneve_init_net,
467+ .exit = geneve_exit_net,
468+ .id = &geneve_net_id,
469+ .size = sizeof(struct geneve_net),
470+};
471+
472+static int __init geneve_init_module(void)
473+{
474+ int rc;
475+
476+ rc = register_pernet_subsys(&geneve_net_ops);
477+ if (rc)
478+ goto out1;
479+
480+ rc = rtnl_link_register(&geneve_link_ops);
481+ if (rc)
482+ goto out2;
483+
484+ return 0;
485+out2:
486+ unregister_pernet_subsys(&geneve_net_ops);
487+out1:
488+ return rc;
489+}
490+late_initcall(geneve_init_module);
491+
492+static void __exit geneve_cleanup_module(void)
493+{
494+ rtnl_link_unregister(&geneve_link_ops);
495+ unregister_pernet_subsys(&geneve_net_ops);
496+}
497+module_exit(geneve_cleanup_module);
498+
499+MODULE_LICENSE("GPL");
500+MODULE_VERSION(GENEVE_NETDEV_VER);
501+MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
502+MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
503+MODULE_ALIAS_RTNL_LINK("geneve");
include/uapi/linux/if_link.h+9−0
@@ -390,6 +390,15 @@ struct ifla_vxlan_port_range { · 389 unmodified lines
390390 __be16 high;
391391 };
392392
393+/* GENEVE section */
394+enum {
395+ IFLA_GENEVE_UNSPEC,
396+ IFLA_GENEVE_ID,
397+ IFLA_GENEVE_REMOTE,
398+ __IFLA_GENEVE_MAX
399+};
400+#define IFLA_GENEVE_MAX (__IFLA_GENEVE_MAX - 1)
401+
393402 /* Bonding section */
394403
395404 enum {