xref: /openbmc/linux/net/openvswitch/vport.c (revision afb46f79)
1 /*
2  * Copyright (c) 2007-2012 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 
36 static 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. */
41 static 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 };
52 
53 /* Protected by RCU read lock for reading, ovs_mutex for writing. */
54 static struct hlist_head *dev_table;
55 #define VPORT_HASH_BUCKETS 1024
56 
57 /**
58  *	ovs_vport_init - initialize vport subsystem
59  *
60  * Called at module load time to initialize the vport subsystem.
61  */
62 int ovs_vport_init(void)
63 {
64 	dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
65 			    GFP_KERNEL);
66 	if (!dev_table)
67 		return -ENOMEM;
68 
69 	return 0;
70 }
71 
72 /**
73  *	ovs_vport_exit - shutdown vport subsystem
74  *
75  * Called at module exit time to shutdown the vport subsystem.
76  */
77 void ovs_vport_exit(void)
78 {
79 	kfree(dev_table);
80 }
81 
82 static struct hlist_head *hash_bucket(struct net *net, const char *name)
83 {
84 	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
85 	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
86 }
87 
88 /**
89  *	ovs_vport_locate - find a port that has already been created
90  *
91  * @name: name of port to find
92  *
93  * Must be called with ovs or RCU read lock.
94  */
95 struct vport *ovs_vport_locate(struct net *net, const char *name)
96 {
97 	struct hlist_head *bucket = hash_bucket(net, name);
98 	struct vport *vport;
99 
100 	hlist_for_each_entry_rcu(vport, bucket, hash_node)
101 		if (!strcmp(name, vport->ops->get_name(vport)) &&
102 		    net_eq(ovs_dp_get_net(vport->dp), net))
103 			return vport;
104 
105 	return NULL;
106 }
107 
108 /**
109  *	ovs_vport_alloc - allocate and initialize new vport
110  *
111  * @priv_size: Size of private data area to allocate.
112  * @ops: vport device ops
113  *
114  * Allocate and initialize a new vport defined by @ops.  The vport will contain
115  * a private data area of size @priv_size that can be accessed using
116  * vport_priv().  vports that are no longer needed should be released with
117  * vport_free().
118  */
119 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
120 			  const struct vport_parms *parms)
121 {
122 	struct vport *vport;
123 	size_t alloc_size;
124 
125 	alloc_size = sizeof(struct vport);
126 	if (priv_size) {
127 		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
128 		alloc_size += priv_size;
129 	}
130 
131 	vport = kzalloc(alloc_size, GFP_KERNEL);
132 	if (!vport)
133 		return ERR_PTR(-ENOMEM);
134 
135 	vport->dp = parms->dp;
136 	vport->port_no = parms->port_no;
137 	vport->upcall_portid = parms->upcall_portid;
138 	vport->ops = ops;
139 	INIT_HLIST_NODE(&vport->dp_hash_node);
140 
141 	vport->percpu_stats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
142 	if (!vport->percpu_stats) {
143 		kfree(vport);
144 		return ERR_PTR(-ENOMEM);
145 	}
146 
147 	spin_lock_init(&vport->stats_lock);
148 
149 	return vport;
150 }
151 
152 /**
153  *	ovs_vport_free - uninitialize and free vport
154  *
155  * @vport: vport to free
156  *
157  * Frees a vport allocated with vport_alloc() when it is no longer needed.
158  *
159  * The caller must ensure that an RCU grace period has passed since the last
160  * time @vport was in a datapath.
161  */
162 void ovs_vport_free(struct vport *vport)
163 {
164 	free_percpu(vport->percpu_stats);
165 	kfree(vport);
166 }
167 
168 /**
169  *	ovs_vport_add - add vport device (for kernel callers)
170  *
171  * @parms: Information about new vport.
172  *
173  * Creates a new vport with the specified configuration (which is dependent on
174  * device type).  ovs_mutex must be held.
175  */
176 struct vport *ovs_vport_add(const struct vport_parms *parms)
177 {
178 	struct vport *vport;
179 	int err = 0;
180 	int i;
181 
182 	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
183 		if (vport_ops_list[i]->type == parms->type) {
184 			struct hlist_head *bucket;
185 
186 			vport = vport_ops_list[i]->create(parms);
187 			if (IS_ERR(vport)) {
188 				err = PTR_ERR(vport);
189 				goto out;
190 			}
191 
192 			bucket = hash_bucket(ovs_dp_get_net(vport->dp),
193 					     vport->ops->get_name(vport));
194 			hlist_add_head_rcu(&vport->hash_node, bucket);
195 			return vport;
196 		}
197 	}
198 
199 	err = -EAFNOSUPPORT;
200 
201 out:
202 	return ERR_PTR(err);
203 }
204 
205 /**
206  *	ovs_vport_set_options - modify existing vport device (for kernel callers)
207  *
208  * @vport: vport to modify.
209  * @options: New configuration.
210  *
211  * Modifies an existing device with the specified configuration (which is
212  * dependent on device type).  ovs_mutex must be held.
213  */
214 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
215 {
216 	if (!vport->ops->set_options)
217 		return -EOPNOTSUPP;
218 	return vport->ops->set_options(vport, options);
219 }
220 
221 /**
222  *	ovs_vport_del - delete existing vport device
223  *
224  * @vport: vport to delete.
225  *
226  * Detaches @vport from its datapath and destroys it.  It is possible to fail
227  * for reasons such as lack of memory.  ovs_mutex must be held.
228  */
229 void ovs_vport_del(struct vport *vport)
230 {
231 	ASSERT_OVSL();
232 
233 	hlist_del_rcu(&vport->hash_node);
234 
235 	vport->ops->destroy(vport);
236 }
237 
238 /**
239  *	ovs_vport_get_stats - retrieve device stats
240  *
241  * @vport: vport from which to retrieve the stats
242  * @stats: location to store stats
243  *
244  * Retrieves transmit, receive, and error stats for the given device.
245  *
246  * Must be called with ovs_mutex or rcu_read_lock.
247  */
248 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
249 {
250 	int i;
251 
252 	memset(stats, 0, sizeof(*stats));
253 
254 	/* We potentially have 2 sources of stats that need to be combined:
255 	 * those we have collected (split into err_stats and percpu_stats) from
256 	 * set_stats() and device error stats from netdev->get_stats() (for
257 	 * errors that happen  downstream and therefore aren't reported through
258 	 * our vport_record_error() function).
259 	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
260 	 * netdev-stats can be directly read over netlink-ioctl.
261 	 */
262 
263 	spin_lock_bh(&vport->stats_lock);
264 
265 	stats->rx_errors	= vport->err_stats.rx_errors;
266 	stats->tx_errors	= vport->err_stats.tx_errors;
267 	stats->tx_dropped	= vport->err_stats.tx_dropped;
268 	stats->rx_dropped	= vport->err_stats.rx_dropped;
269 
270 	spin_unlock_bh(&vport->stats_lock);
271 
272 	for_each_possible_cpu(i) {
273 		const struct pcpu_sw_netstats *percpu_stats;
274 		struct pcpu_sw_netstats local_stats;
275 		unsigned int start;
276 
277 		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
278 
279 		do {
280 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
281 			local_stats = *percpu_stats;
282 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
283 
284 		stats->rx_bytes		+= local_stats.rx_bytes;
285 		stats->rx_packets	+= local_stats.rx_packets;
286 		stats->tx_bytes		+= local_stats.tx_bytes;
287 		stats->tx_packets	+= local_stats.tx_packets;
288 	}
289 }
290 
291 /**
292  *	ovs_vport_get_options - retrieve device options
293  *
294  * @vport: vport from which to retrieve the options.
295  * @skb: sk_buff where options should be appended.
296  *
297  * Retrieves the configuration of the given device, appending an
298  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
299  * vport-specific attributes to @skb.
300  *
301  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
302  * negative error code if a real error occurred.  If an error occurs, @skb is
303  * left unmodified.
304  *
305  * Must be called with ovs_mutex or rcu_read_lock.
306  */
307 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
308 {
309 	struct nlattr *nla;
310 	int err;
311 
312 	if (!vport->ops->get_options)
313 		return 0;
314 
315 	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
316 	if (!nla)
317 		return -EMSGSIZE;
318 
319 	err = vport->ops->get_options(vport, skb);
320 	if (err) {
321 		nla_nest_cancel(skb, nla);
322 		return err;
323 	}
324 
325 	nla_nest_end(skb, nla);
326 	return 0;
327 }
328 
329 /**
330  *	ovs_vport_receive - pass up received packet to the datapath for processing
331  *
332  * @vport: vport that received the packet
333  * @skb: skb that was received
334  * @tun_key: tunnel (if any) that carried packet
335  *
336  * Must be called with rcu_read_lock.  The packet cannot be shared and
337  * skb->data should point to the Ethernet header.
338  */
339 void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
340 		       struct ovs_key_ipv4_tunnel *tun_key)
341 {
342 	struct pcpu_sw_netstats *stats;
343 
344 	stats = this_cpu_ptr(vport->percpu_stats);
345 	u64_stats_update_begin(&stats->syncp);
346 	stats->rx_packets++;
347 	stats->rx_bytes += skb->len;
348 	u64_stats_update_end(&stats->syncp);
349 
350 	OVS_CB(skb)->tun_key = tun_key;
351 	ovs_dp_process_received_packet(vport, skb);
352 }
353 
354 /**
355  *	ovs_vport_send - send a packet on a device
356  *
357  * @vport: vport on which to send the packet
358  * @skb: skb to send
359  *
360  * Sends the given packet and returns the length of data sent.  Either ovs
361  * lock or rcu_read_lock must be held.
362  */
363 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
364 {
365 	int sent = vport->ops->send(vport, skb);
366 
367 	if (likely(sent > 0)) {
368 		struct pcpu_sw_netstats *stats;
369 
370 		stats = this_cpu_ptr(vport->percpu_stats);
371 
372 		u64_stats_update_begin(&stats->syncp);
373 		stats->tx_packets++;
374 		stats->tx_bytes += sent;
375 		u64_stats_update_end(&stats->syncp);
376 	} else if (sent < 0) {
377 		ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
378 		kfree_skb(skb);
379 	} else
380 		ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
381 
382 	return sent;
383 }
384 
385 /**
386  *	ovs_vport_record_error - indicate device error to generic stats layer
387  *
388  * @vport: vport that encountered the error
389  * @err_type: one of enum vport_err_type types to indicate the error type
390  *
391  * If using the vport generic stats layer indicate that an error of the given
392  * type has occurred.
393  */
394 static void ovs_vport_record_error(struct vport *vport,
395 				   enum vport_err_type err_type)
396 {
397 	spin_lock(&vport->stats_lock);
398 
399 	switch (err_type) {
400 	case VPORT_E_RX_DROPPED:
401 		vport->err_stats.rx_dropped++;
402 		break;
403 
404 	case VPORT_E_RX_ERROR:
405 		vport->err_stats.rx_errors++;
406 		break;
407 
408 	case VPORT_E_TX_DROPPED:
409 		vport->err_stats.tx_dropped++;
410 		break;
411 
412 	case VPORT_E_TX_ERROR:
413 		vport->err_stats.tx_errors++;
414 		break;
415 	}
416 
417 	spin_unlock(&vport->stats_lock);
418 }
419 
420 static void free_vport_rcu(struct rcu_head *rcu)
421 {
422 	struct vport *vport = container_of(rcu, struct vport, rcu);
423 
424 	ovs_vport_free(vport);
425 }
426 
427 void ovs_vport_deferred_free(struct vport *vport)
428 {
429 	if (!vport)
430 		return;
431 
432 	call_rcu(&vport->rcu, free_vport_rcu);
433 }
434