xref: /linux/net/openvswitch/vport.c (revision d91517839e5d95adc0cf4b28caa7af62a71de526)
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 	int i;
125 
126 	alloc_size = sizeof(struct vport);
127 	if (priv_size) {
128 		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
129 		alloc_size += priv_size;
130 	}
131 
132 	vport = kzalloc(alloc_size, GFP_KERNEL);
133 	if (!vport)
134 		return ERR_PTR(-ENOMEM);
135 
136 	vport->dp = parms->dp;
137 	vport->port_no = parms->port_no;
138 	vport->upcall_portid = parms->upcall_portid;
139 	vport->ops = ops;
140 	INIT_HLIST_NODE(&vport->dp_hash_node);
141 
142 	vport->percpu_stats = alloc_percpu(struct pcpu_sw_netstats);
143 	if (!vport->percpu_stats) {
144 		kfree(vport);
145 		return ERR_PTR(-ENOMEM);
146 	}
147 
148 	for_each_possible_cpu(i) {
149 		struct pcpu_sw_netstats *vport_stats;
150 		vport_stats = per_cpu_ptr(vport->percpu_stats, i);
151 		u64_stats_init(&vport_stats->syncp);
152 	}
153 
154 
155 	spin_lock_init(&vport->stats_lock);
156 
157 	return vport;
158 }
159 
160 /**
161  *	ovs_vport_free - uninitialize and free vport
162  *
163  * @vport: vport to free
164  *
165  * Frees a vport allocated with vport_alloc() when it is no longer needed.
166  *
167  * The caller must ensure that an RCU grace period has passed since the last
168  * time @vport was in a datapath.
169  */
170 void ovs_vport_free(struct vport *vport)
171 {
172 	free_percpu(vport->percpu_stats);
173 	kfree(vport);
174 }
175 
176 /**
177  *	ovs_vport_add - add vport device (for kernel callers)
178  *
179  * @parms: Information about new vport.
180  *
181  * Creates a new vport with the specified configuration (which is dependent on
182  * device type).  ovs_mutex must be held.
183  */
184 struct vport *ovs_vport_add(const struct vport_parms *parms)
185 {
186 	struct vport *vport;
187 	int err = 0;
188 	int i;
189 
190 	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
191 		if (vport_ops_list[i]->type == parms->type) {
192 			struct hlist_head *bucket;
193 
194 			vport = vport_ops_list[i]->create(parms);
195 			if (IS_ERR(vport)) {
196 				err = PTR_ERR(vport);
197 				goto out;
198 			}
199 
200 			bucket = hash_bucket(ovs_dp_get_net(vport->dp),
201 					     vport->ops->get_name(vport));
202 			hlist_add_head_rcu(&vport->hash_node, bucket);
203 			return vport;
204 		}
205 	}
206 
207 	err = -EAFNOSUPPORT;
208 
209 out:
210 	return ERR_PTR(err);
211 }
212 
213 /**
214  *	ovs_vport_set_options - modify existing vport device (for kernel callers)
215  *
216  * @vport: vport to modify.
217  * @options: New configuration.
218  *
219  * Modifies an existing device with the specified configuration (which is
220  * dependent on device type).  ovs_mutex must be held.
221  */
222 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
223 {
224 	if (!vport->ops->set_options)
225 		return -EOPNOTSUPP;
226 	return vport->ops->set_options(vport, options);
227 }
228 
229 /**
230  *	ovs_vport_del - delete existing vport device
231  *
232  * @vport: vport to delete.
233  *
234  * Detaches @vport from its datapath and destroys it.  It is possible to fail
235  * for reasons such as lack of memory.  ovs_mutex must be held.
236  */
237 void ovs_vport_del(struct vport *vport)
238 {
239 	ASSERT_OVSL();
240 
241 	hlist_del_rcu(&vport->hash_node);
242 
243 	vport->ops->destroy(vport);
244 }
245 
246 /**
247  *	ovs_vport_get_stats - retrieve device stats
248  *
249  * @vport: vport from which to retrieve the stats
250  * @stats: location to store stats
251  *
252  * Retrieves transmit, receive, and error stats for the given device.
253  *
254  * Must be called with ovs_mutex or rcu_read_lock.
255  */
256 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
257 {
258 	int i;
259 
260 	memset(stats, 0, sizeof(*stats));
261 
262 	/* We potentially have 2 sources of stats that need to be combined:
263 	 * those we have collected (split into err_stats and percpu_stats) from
264 	 * set_stats() and device error stats from netdev->get_stats() (for
265 	 * errors that happen  downstream and therefore aren't reported through
266 	 * our vport_record_error() function).
267 	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
268 	 * netdev-stats can be directly read over netlink-ioctl.
269 	 */
270 
271 	spin_lock_bh(&vport->stats_lock);
272 
273 	stats->rx_errors	= vport->err_stats.rx_errors;
274 	stats->tx_errors	= vport->err_stats.tx_errors;
275 	stats->tx_dropped	= vport->err_stats.tx_dropped;
276 	stats->rx_dropped	= vport->err_stats.rx_dropped;
277 
278 	spin_unlock_bh(&vport->stats_lock);
279 
280 	for_each_possible_cpu(i) {
281 		const struct pcpu_sw_netstats *percpu_stats;
282 		struct pcpu_sw_netstats local_stats;
283 		unsigned int start;
284 
285 		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
286 
287 		do {
288 			start = u64_stats_fetch_begin_bh(&percpu_stats->syncp);
289 			local_stats = *percpu_stats;
290 		} while (u64_stats_fetch_retry_bh(&percpu_stats->syncp, start));
291 
292 		stats->rx_bytes		+= local_stats.rx_bytes;
293 		stats->rx_packets	+= local_stats.rx_packets;
294 		stats->tx_bytes		+= local_stats.tx_bytes;
295 		stats->tx_packets	+= local_stats.tx_packets;
296 	}
297 }
298 
299 /**
300  *	ovs_vport_get_options - retrieve device options
301  *
302  * @vport: vport from which to retrieve the options.
303  * @skb: sk_buff where options should be appended.
304  *
305  * Retrieves the configuration of the given device, appending an
306  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
307  * vport-specific attributes to @skb.
308  *
309  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
310  * negative error code if a real error occurred.  If an error occurs, @skb is
311  * left unmodified.
312  *
313  * Must be called with ovs_mutex or rcu_read_lock.
314  */
315 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
316 {
317 	struct nlattr *nla;
318 	int err;
319 
320 	if (!vport->ops->get_options)
321 		return 0;
322 
323 	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
324 	if (!nla)
325 		return -EMSGSIZE;
326 
327 	err = vport->ops->get_options(vport, skb);
328 	if (err) {
329 		nla_nest_cancel(skb, nla);
330 		return err;
331 	}
332 
333 	nla_nest_end(skb, nla);
334 	return 0;
335 }
336 
337 /**
338  *	ovs_vport_receive - pass up received packet to the datapath for processing
339  *
340  * @vport: vport that received the packet
341  * @skb: skb that was received
342  * @tun_key: tunnel (if any) that carried packet
343  *
344  * Must be called with rcu_read_lock.  The packet cannot be shared and
345  * skb->data should point to the Ethernet header.
346  */
347 void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
348 		       struct ovs_key_ipv4_tunnel *tun_key)
349 {
350 	struct pcpu_sw_netstats *stats;
351 
352 	stats = this_cpu_ptr(vport->percpu_stats);
353 	u64_stats_update_begin(&stats->syncp);
354 	stats->rx_packets++;
355 	stats->rx_bytes += skb->len;
356 	u64_stats_update_end(&stats->syncp);
357 
358 	OVS_CB(skb)->tun_key = tun_key;
359 	ovs_dp_process_received_packet(vport, skb);
360 }
361 
362 /**
363  *	ovs_vport_send - send a packet on a device
364  *
365  * @vport: vport on which to send the packet
366  * @skb: skb to send
367  *
368  * Sends the given packet and returns the length of data sent.  Either ovs
369  * lock or rcu_read_lock must be held.
370  */
371 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
372 {
373 	int sent = vport->ops->send(vport, skb);
374 
375 	if (likely(sent > 0)) {
376 		struct pcpu_sw_netstats *stats;
377 
378 		stats = this_cpu_ptr(vport->percpu_stats);
379 
380 		u64_stats_update_begin(&stats->syncp);
381 		stats->tx_packets++;
382 		stats->tx_bytes += sent;
383 		u64_stats_update_end(&stats->syncp);
384 	} else if (sent < 0) {
385 		ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
386 		kfree_skb(skb);
387 	} else
388 		ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
389 
390 	return sent;
391 }
392 
393 /**
394  *	ovs_vport_record_error - indicate device error to generic stats layer
395  *
396  * @vport: vport that encountered the error
397  * @err_type: one of enum vport_err_type types to indicate the error type
398  *
399  * If using the vport generic stats layer indicate that an error of the given
400  * type has occurred.
401  */
402 static void ovs_vport_record_error(struct vport *vport,
403 				   enum vport_err_type err_type)
404 {
405 	spin_lock(&vport->stats_lock);
406 
407 	switch (err_type) {
408 	case VPORT_E_RX_DROPPED:
409 		vport->err_stats.rx_dropped++;
410 		break;
411 
412 	case VPORT_E_RX_ERROR:
413 		vport->err_stats.rx_errors++;
414 		break;
415 
416 	case VPORT_E_TX_DROPPED:
417 		vport->err_stats.tx_dropped++;
418 		break;
419 
420 	case VPORT_E_TX_ERROR:
421 		vport->err_stats.tx_errors++;
422 		break;
423 	}
424 
425 	spin_unlock(&vport->stats_lock);
426 }
427 
428 static void free_vport_rcu(struct rcu_head *rcu)
429 {
430 	struct vport *vport = container_of(rcu, struct vport, rcu);
431 
432 	ovs_vport_free(vport);
433 }
434 
435 void ovs_vport_deferred_free(struct vport *vport)
436 {
437 	if (!vport)
438 		return;
439 
440 	call_rcu(&vport->rcu, free_vport_rcu);
441 }
442