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
32#include "datapath.h"
33#include "vport.h"
34#include "vport-internal_dev.h"
35
36static void ovs_vport_record_error(struct vport *,
37				   enum vport_err_type err_type);
38
39/* List of statically compiled vport implementations.  Don't forget to also
40 * add yours to the list at the bottom of vport.h. */
41static const struct vport_ops *vport_ops_list[] = {
42	&ovs_netdev_vport_ops,
43	&ovs_internal_vport_ops,
44
45#ifdef CONFIG_OPENVSWITCH_GRE
46	&ovs_gre_vport_ops,
47#endif
48#ifdef CONFIG_OPENVSWITCH_VXLAN
49	&ovs_vxlan_vport_ops,
50#endif
51#ifdef CONFIG_OPENVSWITCH_GENEVE
52	&ovs_geneve_vport_ops,
53#endif
54};
55
56/* Protected by RCU read lock for reading, ovs_mutex for writing. */
57static struct hlist_head *dev_table;
58#define VPORT_HASH_BUCKETS 1024
59
60/**
61 *	ovs_vport_init - initialize vport subsystem
62 *
63 * Called at module load time to initialize the vport subsystem.
64 */
65int ovs_vport_init(void)
66{
67	dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
68			    GFP_KERNEL);
69	if (!dev_table)
70		return -ENOMEM;
71
72	return 0;
73}
74
75/**
76 *	ovs_vport_exit - shutdown vport subsystem
77 *
78 * Called at module exit time to shutdown the vport subsystem.
79 */
80void ovs_vport_exit(void)
81{
82	kfree(dev_table);
83}
84
85static struct hlist_head *hash_bucket(struct net *net, const char *name)
86{
87	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
88	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
89}
90
91/**
92 *	ovs_vport_locate - find a port that has already been created
93 *
94 * @name: name of port to find
95 *
96 * Must be called with ovs or RCU read lock.
97 */
98struct vport *ovs_vport_locate(struct net *net, const char *name)
99{
100	struct hlist_head *bucket = hash_bucket(net, name);
101	struct vport *vport;
102
103	hlist_for_each_entry_rcu(vport, bucket, hash_node)
104		if (!strcmp(name, vport->ops->get_name(vport)) &&
105		    net_eq(ovs_dp_get_net(vport->dp), net))
106			return vport;
107
108	return NULL;
109}
110
111/**
112 *	ovs_vport_alloc - allocate and initialize new vport
113 *
114 * @priv_size: Size of private data area to allocate.
115 * @ops: vport device ops
116 *
117 * Allocate and initialize a new vport defined by @ops.  The vport will contain
118 * a private data area of size @priv_size that can be accessed using
119 * vport_priv().  vports that are no longer needed should be released with
120 * vport_free().
121 */
122struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
123			  const struct vport_parms *parms)
124{
125	struct vport *vport;
126	size_t alloc_size;
127
128	alloc_size = sizeof(struct vport);
129	if (priv_size) {
130		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
131		alloc_size += priv_size;
132	}
133
134	vport = kzalloc(alloc_size, GFP_KERNEL);
135	if (!vport)
136		return ERR_PTR(-ENOMEM);
137
138	vport->dp = parms->dp;
139	vport->port_no = parms->port_no;
140	vport->ops = ops;
141	INIT_HLIST_NODE(&vport->dp_hash_node);
142
143	if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
144		kfree(vport);
145		return ERR_PTR(-EINVAL);
146	}
147
148	vport->percpu_stats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
149	if (!vport->percpu_stats) {
150		kfree(vport);
151		return ERR_PTR(-ENOMEM);
152	}
153
154	return vport;
155}
156
157/**
158 *	ovs_vport_free - uninitialize and free vport
159 *
160 * @vport: vport to free
161 *
162 * Frees a vport allocated with vport_alloc() when it is no longer needed.
163 *
164 * The caller must ensure that an RCU grace period has passed since the last
165 * time @vport was in a datapath.
166 */
167void ovs_vport_free(struct vport *vport)
168{
169	/* vport is freed from RCU callback or error path, Therefore
170	 * it is safe to use raw dereference.
171	 */
172	kfree(rcu_dereference_raw(vport->upcall_portids));
173	free_percpu(vport->percpu_stats);
174	kfree(vport);
175}
176
177/**
178 *	ovs_vport_add - add vport device (for kernel callers)
179 *
180 * @parms: Information about new vport.
181 *
182 * Creates a new vport with the specified configuration (which is dependent on
183 * device type).  ovs_mutex must be held.
184 */
185struct vport *ovs_vport_add(const struct vport_parms *parms)
186{
187	struct vport *vport;
188	int err = 0;
189	int i;
190
191	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
192		if (vport_ops_list[i]->type == parms->type) {
193			struct hlist_head *bucket;
194
195			vport = vport_ops_list[i]->create(parms);
196			if (IS_ERR(vport)) {
197				err = PTR_ERR(vport);
198				goto out;
199			}
200
201			bucket = hash_bucket(ovs_dp_get_net(vport->dp),
202					     vport->ops->get_name(vport));
203			hlist_add_head_rcu(&vport->hash_node, bucket);
204			return vport;
205		}
206	}
207
208	err = -EAFNOSUPPORT;
209
210out:
211	return ERR_PTR(err);
212}
213
214/**
215 *	ovs_vport_set_options - modify existing vport device (for kernel callers)
216 *
217 * @vport: vport to modify.
218 * @options: New configuration.
219 *
220 * Modifies an existing device with the specified configuration (which is
221 * dependent on device type).  ovs_mutex must be held.
222 */
223int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
224{
225	if (!vport->ops->set_options)
226		return -EOPNOTSUPP;
227	return vport->ops->set_options(vport, options);
228}
229
230/**
231 *	ovs_vport_del - delete existing vport device
232 *
233 * @vport: vport to delete.
234 *
235 * Detaches @vport from its datapath and destroys it.  It is possible to fail
236 * for reasons such as lack of memory.  ovs_mutex must be held.
237 */
238void ovs_vport_del(struct vport *vport)
239{
240	ASSERT_OVSL();
241
242	hlist_del_rcu(&vport->hash_node);
243
244	vport->ops->destroy(vport);
245}
246
247/**
248 *	ovs_vport_get_stats - retrieve device stats
249 *
250 * @vport: vport from which to retrieve the stats
251 * @stats: location to store stats
252 *
253 * Retrieves transmit, receive, and error stats for the given device.
254 *
255 * Must be called with ovs_mutex or rcu_read_lock.
256 */
257void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
258{
259	int i;
260
261	memset(stats, 0, sizeof(*stats));
262
263	/* We potentially have 2 sources of stats that need to be combined:
264	 * those we have collected (split into err_stats and percpu_stats) from
265	 * set_stats() and device error stats from netdev->get_stats() (for
266	 * errors that happen  downstream and therefore aren't reported through
267	 * our vport_record_error() function).
268	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
269	 * netdev-stats can be directly read over netlink-ioctl.
270	 */
271
272	stats->rx_errors  = atomic_long_read(&vport->err_stats.rx_errors);
273	stats->tx_errors  = atomic_long_read(&vport->err_stats.tx_errors);
274	stats->tx_dropped = atomic_long_read(&vport->err_stats.tx_dropped);
275	stats->rx_dropped = atomic_long_read(&vport->err_stats.rx_dropped);
276
277	for_each_possible_cpu(i) {
278		const struct pcpu_sw_netstats *percpu_stats;
279		struct pcpu_sw_netstats local_stats;
280		unsigned int start;
281
282		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
283
284		do {
285			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
286			local_stats = *percpu_stats;
287		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
288
289		stats->rx_bytes		+= local_stats.rx_bytes;
290		stats->rx_packets	+= local_stats.rx_packets;
291		stats->tx_bytes		+= local_stats.tx_bytes;
292		stats->tx_packets	+= local_stats.tx_packets;
293	}
294}
295
296/**
297 *	ovs_vport_get_options - retrieve device options
298 *
299 * @vport: vport from which to retrieve the options.
300 * @skb: sk_buff where options should be appended.
301 *
302 * Retrieves the configuration of the given device, appending an
303 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
304 * vport-specific attributes to @skb.
305 *
306 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
307 * negative error code if a real error occurred.  If an error occurs, @skb is
308 * left unmodified.
309 *
310 * Must be called with ovs_mutex or rcu_read_lock.
311 */
312int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
313{
314	struct nlattr *nla;
315	int err;
316
317	if (!vport->ops->get_options)
318		return 0;
319
320	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
321	if (!nla)
322		return -EMSGSIZE;
323
324	err = vport->ops->get_options(vport, skb);
325	if (err) {
326		nla_nest_cancel(skb, nla);
327		return err;
328	}
329
330	nla_nest_end(skb, nla);
331	return 0;
332}
333
334/**
335 *	ovs_vport_set_upcall_portids - set upcall portids of @vport.
336 *
337 * @vport: vport to modify.
338 * @ids: new configuration, an array of port ids.
339 *
340 * Sets the vport's upcall_portids to @ids.
341 *
342 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
343 * as an array of U32.
344 *
345 * Must be called with ovs_mutex.
346 */
347int ovs_vport_set_upcall_portids(struct vport *vport,  struct nlattr *ids)
348{
349	struct vport_portids *old, *vport_portids;
350
351	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
352		return -EINVAL;
353
354	old = ovsl_dereference(vport->upcall_portids);
355
356	vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
357				GFP_KERNEL);
358	if (!vport_portids)
359		return -ENOMEM;
360
361	vport_portids->n_ids = nla_len(ids) / sizeof(u32);
362	vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
363	nla_memcpy(vport_portids->ids, ids, nla_len(ids));
364
365	rcu_assign_pointer(vport->upcall_portids, vport_portids);
366
367	if (old)
368		kfree_rcu(old, rcu);
369	return 0;
370}
371
372/**
373 *	ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
374 *
375 * @vport: vport from which to retrieve the portids.
376 * @skb: sk_buff where portids should be appended.
377 *
378 * Retrieves the configuration of the given vport, appending the
379 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
380 * portids to @skb.
381 *
382 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
383 * If an error occurs, @skb is left unmodified.  Must be called with
384 * ovs_mutex or rcu_read_lock.
385 */
386int ovs_vport_get_upcall_portids(const struct vport *vport,
387				 struct sk_buff *skb)
388{
389	struct vport_portids *ids;
390
391	ids = rcu_dereference_ovsl(vport->upcall_portids);
392
393	if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
394		return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
395			       ids->n_ids * sizeof(u32), (void *)ids->ids);
396	else
397		return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
398}
399
400/**
401 *	ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
402 *
403 * @vport: vport from which the missed packet is received.
404 * @skb: skb that the missed packet was received.
405 *
406 * Uses the skb_get_hash() to select the upcall portid to send the
407 * upcall.
408 *
409 * Returns the portid of the target socket.  Must be called with rcu_read_lock.
410 */
411u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
412{
413	struct vport_portids *ids;
414	u32 ids_index;
415	u32 hash;
416
417	ids = rcu_dereference(vport->upcall_portids);
418
419	if (ids->n_ids == 1 && ids->ids[0] == 0)
420		return 0;
421
422	hash = skb_get_hash(skb);
423	ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
424	return ids->ids[ids_index];
425}
426
427/**
428 *	ovs_vport_receive - pass up received packet to the datapath for processing
429 *
430 * @vport: vport that received the packet
431 * @skb: skb that was received
432 * @tun_key: tunnel (if any) that carried packet
433 *
434 * Must be called with rcu_read_lock.  The packet cannot be shared and
435 * skb->data should point to the Ethernet header.
436 */
437void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
438		       struct ovs_tunnel_info *tun_info)
439{
440	struct pcpu_sw_netstats *stats;
441	struct sw_flow_key key;
442	int error;
443
444	stats = this_cpu_ptr(vport->percpu_stats);
445	u64_stats_update_begin(&stats->syncp);
446	stats->rx_packets++;
447	stats->rx_bytes += skb->len;
448	u64_stats_update_end(&stats->syncp);
449
450	OVS_CB(skb)->input_vport = vport;
451	OVS_CB(skb)->egress_tun_info = NULL;
452	/* Extract flow from 'skb' into 'key'. */
453	error = ovs_flow_key_extract(tun_info, skb, &key);
454	if (unlikely(error)) {
455		kfree_skb(skb);
456		return;
457	}
458	ovs_dp_process_packet(skb, &key);
459}
460
461/**
462 *	ovs_vport_send - send a packet on a device
463 *
464 * @vport: vport on which to send the packet
465 * @skb: skb to send
466 *
467 * Sends the given packet and returns the length of data sent.  Either ovs
468 * lock or rcu_read_lock must be held.
469 */
470int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
471{
472	int sent = vport->ops->send(vport, skb);
473
474	if (likely(sent > 0)) {
475		struct pcpu_sw_netstats *stats;
476
477		stats = this_cpu_ptr(vport->percpu_stats);
478
479		u64_stats_update_begin(&stats->syncp);
480		stats->tx_packets++;
481		stats->tx_bytes += sent;
482		u64_stats_update_end(&stats->syncp);
483	} else if (sent < 0) {
484		ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
485		kfree_skb(skb);
486	} else
487		ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
488
489	return sent;
490}
491
492/**
493 *	ovs_vport_record_error - indicate device error to generic stats layer
494 *
495 * @vport: vport that encountered the error
496 * @err_type: one of enum vport_err_type types to indicate the error type
497 *
498 * If using the vport generic stats layer indicate that an error of the given
499 * type has occurred.
500 */
501static void ovs_vport_record_error(struct vport *vport,
502				   enum vport_err_type err_type)
503{
504	switch (err_type) {
505	case VPORT_E_RX_DROPPED:
506		atomic_long_inc(&vport->err_stats.rx_dropped);
507		break;
508
509	case VPORT_E_RX_ERROR:
510		atomic_long_inc(&vport->err_stats.rx_errors);
511		break;
512
513	case VPORT_E_TX_DROPPED:
514		atomic_long_inc(&vport->err_stats.tx_dropped);
515		break;
516
517	case VPORT_E_TX_ERROR:
518		atomic_long_inc(&vport->err_stats.tx_errors);
519		break;
520	}
521
522}
523
524static void free_vport_rcu(struct rcu_head *rcu)
525{
526	struct vport *vport = container_of(rcu, struct vport, rcu);
527
528	ovs_vport_free(vport);
529}
530
531void ovs_vport_deferred_free(struct vport *vport)
532{
533	if (!vport)
534		return;
535
536	call_rcu(&vport->rcu, free_vport_rcu);
537}
538