1/*
2 * Copyright (c) 2007-2014 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#include <linux/etherdevice.h>
20#include <linux/if.h>
21#include <linux/if_vlan.h>
22#include <linux/jhash.h>
23#include <linux/kernel.h>
24#include <linux/list.h>
25#include <linux/mutex.h>
26#include <linux/percpu.h>
27#include <linux/rcupdate.h>
28#include <linux/rtnetlink.h>
29#include <linux/compat.h>
30#include <net/net_namespace.h>
31#include <linux/module.h>
32
33#include "datapath.h"
34#include "vport.h"
35#include "vport-internal_dev.h"
36
37static void ovs_vport_record_error(struct vport *,
38				   enum vport_err_type err_type);
39
40static LIST_HEAD(vport_ops_list);
41
42/* Protected by RCU read lock for reading, ovs_mutex for writing. */
43static struct hlist_head *dev_table;
44#define VPORT_HASH_BUCKETS 1024
45
46/**
47 *	ovs_vport_init - initialize vport subsystem
48 *
49 * Called at module load time to initialize the vport subsystem.
50 */
51int ovs_vport_init(void)
52{
53	dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
54			    GFP_KERNEL);
55	if (!dev_table)
56		return -ENOMEM;
57
58	return 0;
59}
60
61/**
62 *	ovs_vport_exit - shutdown vport subsystem
63 *
64 * Called at module exit time to shutdown the vport subsystem.
65 */
66void ovs_vport_exit(void)
67{
68	kfree(dev_table);
69}
70
71static struct hlist_head *hash_bucket(const struct net *net, const char *name)
72{
73	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
74	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
75}
76
77int ovs_vport_ops_register(struct vport_ops *ops)
78{
79	int err = -EEXIST;
80	struct vport_ops *o;
81
82	ovs_lock();
83	list_for_each_entry(o, &vport_ops_list, list)
84		if (ops->type == o->type)
85			goto errout;
86
87	list_add_tail(&ops->list, &vport_ops_list);
88	err = 0;
89errout:
90	ovs_unlock();
91	return err;
92}
93EXPORT_SYMBOL_GPL(ovs_vport_ops_register);
94
95void ovs_vport_ops_unregister(struct vport_ops *ops)
96{
97	ovs_lock();
98	list_del(&ops->list);
99	ovs_unlock();
100}
101EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
102
103/**
104 *	ovs_vport_locate - find a port that has already been created
105 *
106 * @name: name of port to find
107 *
108 * Must be called with ovs or RCU read lock.
109 */
110struct vport *ovs_vport_locate(const struct net *net, const char *name)
111{
112	struct hlist_head *bucket = hash_bucket(net, name);
113	struct vport *vport;
114
115	hlist_for_each_entry_rcu(vport, bucket, hash_node)
116		if (!strcmp(name, vport->ops->get_name(vport)) &&
117		    net_eq(ovs_dp_get_net(vport->dp), net))
118			return vport;
119
120	return NULL;
121}
122
123/**
124 *	ovs_vport_alloc - allocate and initialize new vport
125 *
126 * @priv_size: Size of private data area to allocate.
127 * @ops: vport device ops
128 *
129 * Allocate and initialize a new vport defined by @ops.  The vport will contain
130 * a private data area of size @priv_size that can be accessed using
131 * vport_priv().  vports that are no longer needed should be released with
132 * vport_free().
133 */
134struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
135			  const struct vport_parms *parms)
136{
137	struct vport *vport;
138	size_t alloc_size;
139
140	alloc_size = sizeof(struct vport);
141	if (priv_size) {
142		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
143		alloc_size += priv_size;
144	}
145
146	vport = kzalloc(alloc_size, GFP_KERNEL);
147	if (!vport)
148		return ERR_PTR(-ENOMEM);
149
150	vport->dp = parms->dp;
151	vport->port_no = parms->port_no;
152	vport->ops = ops;
153	INIT_HLIST_NODE(&vport->dp_hash_node);
154
155	if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
156		kfree(vport);
157		return ERR_PTR(-EINVAL);
158	}
159
160	vport->percpu_stats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
161	if (!vport->percpu_stats) {
162		kfree(vport);
163		return ERR_PTR(-ENOMEM);
164	}
165
166	return vport;
167}
168EXPORT_SYMBOL_GPL(ovs_vport_alloc);
169
170/**
171 *	ovs_vport_free - uninitialize and free vport
172 *
173 * @vport: vport to free
174 *
175 * Frees a vport allocated with vport_alloc() when it is no longer needed.
176 *
177 * The caller must ensure that an RCU grace period has passed since the last
178 * time @vport was in a datapath.
179 */
180void ovs_vport_free(struct vport *vport)
181{
182	/* vport is freed from RCU callback or error path, Therefore
183	 * it is safe to use raw dereference.
184	 */
185	kfree(rcu_dereference_raw(vport->upcall_portids));
186	free_percpu(vport->percpu_stats);
187	kfree(vport);
188}
189EXPORT_SYMBOL_GPL(ovs_vport_free);
190
191static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
192{
193	struct vport_ops *ops;
194
195	list_for_each_entry(ops, &vport_ops_list, list)
196		if (ops->type == parms->type)
197			return ops;
198
199	return NULL;
200}
201
202/**
203 *	ovs_vport_add - add vport device (for kernel callers)
204 *
205 * @parms: Information about new vport.
206 *
207 * Creates a new vport with the specified configuration (which is dependent on
208 * device type).  ovs_mutex must be held.
209 */
210struct vport *ovs_vport_add(const struct vport_parms *parms)
211{
212	struct vport_ops *ops;
213	struct vport *vport;
214
215	ops = ovs_vport_lookup(parms);
216	if (ops) {
217		struct hlist_head *bucket;
218
219		if (!try_module_get(ops->owner))
220			return ERR_PTR(-EAFNOSUPPORT);
221
222		vport = ops->create(parms);
223		if (IS_ERR(vport)) {
224			module_put(ops->owner);
225			return vport;
226		}
227
228		bucket = hash_bucket(ovs_dp_get_net(vport->dp),
229				     vport->ops->get_name(vport));
230		hlist_add_head_rcu(&vport->hash_node, bucket);
231		return vport;
232	}
233
234	/* Unlock to attempt module load and return -EAGAIN if load
235	 * was successful as we need to restart the port addition
236	 * workflow.
237	 */
238	ovs_unlock();
239	request_module("vport-type-%d", parms->type);
240	ovs_lock();
241
242	if (!ovs_vport_lookup(parms))
243		return ERR_PTR(-EAFNOSUPPORT);
244	else
245		return ERR_PTR(-EAGAIN);
246}
247
248/**
249 *	ovs_vport_set_options - modify existing vport device (for kernel callers)
250 *
251 * @vport: vport to modify.
252 * @options: New configuration.
253 *
254 * Modifies an existing device with the specified configuration (which is
255 * dependent on device type).  ovs_mutex must be held.
256 */
257int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
258{
259	if (!vport->ops->set_options)
260		return -EOPNOTSUPP;
261	return vport->ops->set_options(vport, options);
262}
263
264/**
265 *	ovs_vport_del - delete existing vport device
266 *
267 * @vport: vport to delete.
268 *
269 * Detaches @vport from its datapath and destroys it.  It is possible to fail
270 * for reasons such as lack of memory.  ovs_mutex must be held.
271 */
272void ovs_vport_del(struct vport *vport)
273{
274	ASSERT_OVSL();
275
276	hlist_del_rcu(&vport->hash_node);
277	module_put(vport->ops->owner);
278	vport->ops->destroy(vport);
279}
280
281/**
282 *	ovs_vport_get_stats - retrieve device stats
283 *
284 * @vport: vport from which to retrieve the stats
285 * @stats: location to store stats
286 *
287 * Retrieves transmit, receive, and error stats for the given device.
288 *
289 * Must be called with ovs_mutex or rcu_read_lock.
290 */
291void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
292{
293	int i;
294
295	memset(stats, 0, sizeof(*stats));
296
297	/* We potentially have 2 sources of stats that need to be combined:
298	 * those we have collected (split into err_stats and percpu_stats) from
299	 * set_stats() and device error stats from netdev->get_stats() (for
300	 * errors that happen  downstream and therefore aren't reported through
301	 * our vport_record_error() function).
302	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
303	 * netdev-stats can be directly read over netlink-ioctl.
304	 */
305
306	stats->rx_errors  = atomic_long_read(&vport->err_stats.rx_errors);
307	stats->tx_errors  = atomic_long_read(&vport->err_stats.tx_errors);
308	stats->tx_dropped = atomic_long_read(&vport->err_stats.tx_dropped);
309	stats->rx_dropped = atomic_long_read(&vport->err_stats.rx_dropped);
310
311	for_each_possible_cpu(i) {
312		const struct pcpu_sw_netstats *percpu_stats;
313		struct pcpu_sw_netstats local_stats;
314		unsigned int start;
315
316		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
317
318		do {
319			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
320			local_stats = *percpu_stats;
321		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
322
323		stats->rx_bytes		+= local_stats.rx_bytes;
324		stats->rx_packets	+= local_stats.rx_packets;
325		stats->tx_bytes		+= local_stats.tx_bytes;
326		stats->tx_packets	+= local_stats.tx_packets;
327	}
328}
329
330/**
331 *	ovs_vport_get_options - retrieve device options
332 *
333 * @vport: vport from which to retrieve the options.
334 * @skb: sk_buff where options should be appended.
335 *
336 * Retrieves the configuration of the given device, appending an
337 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
338 * vport-specific attributes to @skb.
339 *
340 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
341 * negative error code if a real error occurred.  If an error occurs, @skb is
342 * left unmodified.
343 *
344 * Must be called with ovs_mutex or rcu_read_lock.
345 */
346int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
347{
348	struct nlattr *nla;
349	int err;
350
351	if (!vport->ops->get_options)
352		return 0;
353
354	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
355	if (!nla)
356		return -EMSGSIZE;
357
358	err = vport->ops->get_options(vport, skb);
359	if (err) {
360		nla_nest_cancel(skb, nla);
361		return err;
362	}
363
364	nla_nest_end(skb, nla);
365	return 0;
366}
367
368/**
369 *	ovs_vport_set_upcall_portids - set upcall portids of @vport.
370 *
371 * @vport: vport to modify.
372 * @ids: new configuration, an array of port ids.
373 *
374 * Sets the vport's upcall_portids to @ids.
375 *
376 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
377 * as an array of U32.
378 *
379 * Must be called with ovs_mutex.
380 */
381int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
382{
383	struct vport_portids *old, *vport_portids;
384
385	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
386		return -EINVAL;
387
388	old = ovsl_dereference(vport->upcall_portids);
389
390	vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
391				GFP_KERNEL);
392	if (!vport_portids)
393		return -ENOMEM;
394
395	vport_portids->n_ids = nla_len(ids) / sizeof(u32);
396	vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
397	nla_memcpy(vport_portids->ids, ids, nla_len(ids));
398
399	rcu_assign_pointer(vport->upcall_portids, vport_portids);
400
401	if (old)
402		kfree_rcu(old, rcu);
403	return 0;
404}
405
406/**
407 *	ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
408 *
409 * @vport: vport from which to retrieve the portids.
410 * @skb: sk_buff where portids should be appended.
411 *
412 * Retrieves the configuration of the given vport, appending the
413 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
414 * portids to @skb.
415 *
416 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
417 * If an error occurs, @skb is left unmodified.  Must be called with
418 * ovs_mutex or rcu_read_lock.
419 */
420int ovs_vport_get_upcall_portids(const struct vport *vport,
421				 struct sk_buff *skb)
422{
423	struct vport_portids *ids;
424
425	ids = rcu_dereference_ovsl(vport->upcall_portids);
426
427	if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
428		return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
429			       ids->n_ids * sizeof(u32), (void *)ids->ids);
430	else
431		return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
432}
433
434/**
435 *	ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
436 *
437 * @vport: vport from which the missed packet is received.
438 * @skb: skb that the missed packet was received.
439 *
440 * Uses the skb_get_hash() to select the upcall portid to send the
441 * upcall.
442 *
443 * Returns the portid of the target socket.  Must be called with rcu_read_lock.
444 */
445u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
446{
447	struct vport_portids *ids;
448	u32 ids_index;
449	u32 hash;
450
451	ids = rcu_dereference(vport->upcall_portids);
452
453	if (ids->n_ids == 1 && ids->ids[0] == 0)
454		return 0;
455
456	hash = skb_get_hash(skb);
457	ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
458	return ids->ids[ids_index];
459}
460
461/**
462 *	ovs_vport_receive - pass up received packet to the datapath for processing
463 *
464 * @vport: vport that received the packet
465 * @skb: skb that was received
466 * @tun_key: tunnel (if any) that carried packet
467 *
468 * Must be called with rcu_read_lock.  The packet cannot be shared and
469 * skb->data should point to the Ethernet header.
470 */
471void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
472		       const struct ovs_tunnel_info *tun_info)
473{
474	struct pcpu_sw_netstats *stats;
475	struct sw_flow_key key;
476	int error;
477
478	stats = this_cpu_ptr(vport->percpu_stats);
479	u64_stats_update_begin(&stats->syncp);
480	stats->rx_packets++;
481	stats->rx_bytes += skb->len +
482			   (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
483	u64_stats_update_end(&stats->syncp);
484
485	OVS_CB(skb)->input_vport = vport;
486	OVS_CB(skb)->egress_tun_info = NULL;
487	/* Extract flow from 'skb' into 'key'. */
488	error = ovs_flow_key_extract(tun_info, skb, &key);
489	if (unlikely(error)) {
490		kfree_skb(skb);
491		return;
492	}
493	ovs_dp_process_packet(skb, &key);
494}
495EXPORT_SYMBOL_GPL(ovs_vport_receive);
496
497/**
498 *	ovs_vport_send - send a packet on a device
499 *
500 * @vport: vport on which to send the packet
501 * @skb: skb to send
502 *
503 * Sends the given packet and returns the length of data sent.  Either ovs
504 * lock or rcu_read_lock must be held.
505 */
506int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
507{
508	int sent = vport->ops->send(vport, skb);
509
510	if (likely(sent > 0)) {
511		struct pcpu_sw_netstats *stats;
512
513		stats = this_cpu_ptr(vport->percpu_stats);
514
515		u64_stats_update_begin(&stats->syncp);
516		stats->tx_packets++;
517		stats->tx_bytes += sent;
518		u64_stats_update_end(&stats->syncp);
519	} else if (sent < 0) {
520		ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
521	} else {
522		ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
523	}
524	return sent;
525}
526
527/**
528 *	ovs_vport_record_error - indicate device error to generic stats layer
529 *
530 * @vport: vport that encountered the error
531 * @err_type: one of enum vport_err_type types to indicate the error type
532 *
533 * If using the vport generic stats layer indicate that an error of the given
534 * type has occurred.
535 */
536static void ovs_vport_record_error(struct vport *vport,
537				   enum vport_err_type err_type)
538{
539	switch (err_type) {
540	case VPORT_E_RX_DROPPED:
541		atomic_long_inc(&vport->err_stats.rx_dropped);
542		break;
543
544	case VPORT_E_RX_ERROR:
545		atomic_long_inc(&vport->err_stats.rx_errors);
546		break;
547
548	case VPORT_E_TX_DROPPED:
549		atomic_long_inc(&vport->err_stats.tx_dropped);
550		break;
551
552	case VPORT_E_TX_ERROR:
553		atomic_long_inc(&vport->err_stats.tx_errors);
554		break;
555	}
556
557}
558
559static void free_vport_rcu(struct rcu_head *rcu)
560{
561	struct vport *vport = container_of(rcu, struct vport, rcu);
562
563	ovs_vport_free(vport);
564}
565
566void ovs_vport_deferred_free(struct vport *vport)
567{
568	if (!vport)
569		return;
570
571	call_rcu(&vport->rcu, free_vport_rcu);
572}
573EXPORT_SYMBOL_GPL(ovs_vport_deferred_free);
574
575int ovs_tunnel_get_egress_info(struct ovs_tunnel_info *egress_tun_info,
576			       struct net *net,
577			       const struct ovs_tunnel_info *tun_info,
578			       u8 ipproto,
579			       u32 skb_mark,
580			       __be16 tp_src,
581			       __be16 tp_dst)
582{
583	const struct ovs_key_ipv4_tunnel *tun_key;
584	struct rtable *rt;
585	struct flowi4 fl;
586
587	if (unlikely(!tun_info))
588		return -EINVAL;
589
590	tun_key = &tun_info->tunnel;
591
592	/* Route lookup to get srouce IP address.
593	 * The process may need to be changed if the corresponding process
594	 * in vports ops changed.
595	 */
596	rt = ovs_tunnel_route_lookup(net, tun_key, skb_mark, &fl, ipproto);
597	if (IS_ERR(rt))
598		return PTR_ERR(rt);
599
600	ip_rt_put(rt);
601
602	/* Generate egress_tun_info based on tun_info,
603	 * saddr, tp_src and tp_dst
604	 */
605	__ovs_flow_tun_info_init(egress_tun_info,
606				 fl.saddr, tun_key->ipv4_dst,
607				 tun_key->ipv4_tos,
608				 tun_key->ipv4_ttl,
609				 tp_src, tp_dst,
610				 tun_key->tun_id,
611				 tun_key->tun_flags,
612				 tun_info->options,
613				 tun_info->options_len);
614
615	return 0;
616}
617EXPORT_SYMBOL_GPL(ovs_tunnel_get_egress_info);
618
619int ovs_vport_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
620				  struct ovs_tunnel_info *info)
621{
622	/* get_egress_tun_info() is only implemented on tunnel ports. */
623	if (unlikely(!vport->ops->get_egress_tun_info))
624		return -EINVAL;
625
626	return vport->ops->get_egress_tun_info(vport, skb, info);
627}
628