xref: /openbmc/linux/net/openvswitch/vport.c (revision 15e47304)
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 /* List of statically compiled vport implementations.  Don't forget to also
37  * add yours to the list at the bottom of vport.h. */
38 static const struct vport_ops *vport_ops_list[] = {
39 	&ovs_netdev_vport_ops,
40 	&ovs_internal_vport_ops,
41 };
42 
43 /* Protected by RCU read lock for reading, RTNL lock for writing. */
44 static struct hlist_head *dev_table;
45 #define VPORT_HASH_BUCKETS 1024
46 
47 /**
48  *	ovs_vport_init - initialize vport subsystem
49  *
50  * Called at module load time to initialize the vport subsystem.
51  */
52 int ovs_vport_init(void)
53 {
54 	dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
55 			    GFP_KERNEL);
56 	if (!dev_table)
57 		return -ENOMEM;
58 
59 	return 0;
60 }
61 
62 /**
63  *	ovs_vport_exit - shutdown vport subsystem
64  *
65  * Called at module exit time to shutdown the vport subsystem.
66  */
67 void ovs_vport_exit(void)
68 {
69 	kfree(dev_table);
70 }
71 
72 static struct hlist_head *hash_bucket(struct net *net, const char *name)
73 {
74 	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
75 	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
76 }
77 
78 /**
79  *	ovs_vport_locate - find a port that has already been created
80  *
81  * @name: name of port to find
82  *
83  * Must be called with RTNL or RCU read lock.
84  */
85 struct vport *ovs_vport_locate(struct net *net, const char *name)
86 {
87 	struct hlist_head *bucket = hash_bucket(net, name);
88 	struct vport *vport;
89 	struct hlist_node *node;
90 
91 	hlist_for_each_entry_rcu(vport, node, bucket, hash_node)
92 		if (!strcmp(name, vport->ops->get_name(vport)) &&
93 		    net_eq(ovs_dp_get_net(vport->dp), net))
94 			return vport;
95 
96 	return NULL;
97 }
98 
99 /**
100  *	ovs_vport_alloc - allocate and initialize new vport
101  *
102  * @priv_size: Size of private data area to allocate.
103  * @ops: vport device ops
104  *
105  * Allocate and initialize a new vport defined by @ops.  The vport will contain
106  * a private data area of size @priv_size that can be accessed using
107  * vport_priv().  vports that are no longer needed should be released with
108  * vport_free().
109  */
110 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
111 			  const struct vport_parms *parms)
112 {
113 	struct vport *vport;
114 	size_t alloc_size;
115 
116 	alloc_size = sizeof(struct vport);
117 	if (priv_size) {
118 		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
119 		alloc_size += priv_size;
120 	}
121 
122 	vport = kzalloc(alloc_size, GFP_KERNEL);
123 	if (!vport)
124 		return ERR_PTR(-ENOMEM);
125 
126 	vport->dp = parms->dp;
127 	vport->port_no = parms->port_no;
128 	vport->upcall_portid = parms->upcall_portid;
129 	vport->ops = ops;
130 	INIT_HLIST_NODE(&vport->dp_hash_node);
131 
132 	vport->percpu_stats = alloc_percpu(struct vport_percpu_stats);
133 	if (!vport->percpu_stats) {
134 		kfree(vport);
135 		return ERR_PTR(-ENOMEM);
136 	}
137 
138 	spin_lock_init(&vport->stats_lock);
139 
140 	return vport;
141 }
142 
143 /**
144  *	ovs_vport_free - uninitialize and free vport
145  *
146  * @vport: vport to free
147  *
148  * Frees a vport allocated with vport_alloc() when it is no longer needed.
149  *
150  * The caller must ensure that an RCU grace period has passed since the last
151  * time @vport was in a datapath.
152  */
153 void ovs_vport_free(struct vport *vport)
154 {
155 	free_percpu(vport->percpu_stats);
156 	kfree(vport);
157 }
158 
159 /**
160  *	ovs_vport_add - add vport device (for kernel callers)
161  *
162  * @parms: Information about new vport.
163  *
164  * Creates a new vport with the specified configuration (which is dependent on
165  * device type).  RTNL lock must be held.
166  */
167 struct vport *ovs_vport_add(const struct vport_parms *parms)
168 {
169 	struct vport *vport;
170 	int err = 0;
171 	int i;
172 
173 	ASSERT_RTNL();
174 
175 	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
176 		if (vport_ops_list[i]->type == parms->type) {
177 			struct hlist_head *bucket;
178 
179 			vport = vport_ops_list[i]->create(parms);
180 			if (IS_ERR(vport)) {
181 				err = PTR_ERR(vport);
182 				goto out;
183 			}
184 
185 			bucket = hash_bucket(ovs_dp_get_net(vport->dp),
186 					     vport->ops->get_name(vport));
187 			hlist_add_head_rcu(&vport->hash_node, bucket);
188 			return vport;
189 		}
190 	}
191 
192 	err = -EAFNOSUPPORT;
193 
194 out:
195 	return ERR_PTR(err);
196 }
197 
198 /**
199  *	ovs_vport_set_options - modify existing vport device (for kernel callers)
200  *
201  * @vport: vport to modify.
202  * @port: New configuration.
203  *
204  * Modifies an existing device with the specified configuration (which is
205  * dependent on device type).  RTNL lock must be held.
206  */
207 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
208 {
209 	ASSERT_RTNL();
210 
211 	if (!vport->ops->set_options)
212 		return -EOPNOTSUPP;
213 	return vport->ops->set_options(vport, options);
214 }
215 
216 /**
217  *	ovs_vport_del - delete existing vport device
218  *
219  * @vport: vport to delete.
220  *
221  * Detaches @vport from its datapath and destroys it.  It is possible to fail
222  * for reasons such as lack of memory.  RTNL lock must be held.
223  */
224 void ovs_vport_del(struct vport *vport)
225 {
226 	ASSERT_RTNL();
227 
228 	hlist_del_rcu(&vport->hash_node);
229 
230 	vport->ops->destroy(vport);
231 }
232 
233 /**
234  *	ovs_vport_get_stats - retrieve device stats
235  *
236  * @vport: vport from which to retrieve the stats
237  * @stats: location to store stats
238  *
239  * Retrieves transmit, receive, and error stats for the given device.
240  *
241  * Must be called with RTNL lock or rcu_read_lock.
242  */
243 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
244 {
245 	int i;
246 
247 	memset(stats, 0, sizeof(*stats));
248 
249 	/* We potentially have 2 sources of stats that need to be combined:
250 	 * those we have collected (split into err_stats and percpu_stats) from
251 	 * set_stats() and device error stats from netdev->get_stats() (for
252 	 * errors that happen  downstream and therefore aren't reported through
253 	 * our vport_record_error() function).
254 	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
255 	 * netdev-stats can be directly read over netlink-ioctl.
256 	 */
257 
258 	spin_lock_bh(&vport->stats_lock);
259 
260 	stats->rx_errors	= vport->err_stats.rx_errors;
261 	stats->tx_errors	= vport->err_stats.tx_errors;
262 	stats->tx_dropped	= vport->err_stats.tx_dropped;
263 	stats->rx_dropped	= vport->err_stats.rx_dropped;
264 
265 	spin_unlock_bh(&vport->stats_lock);
266 
267 	for_each_possible_cpu(i) {
268 		const struct vport_percpu_stats *percpu_stats;
269 		struct vport_percpu_stats local_stats;
270 		unsigned int start;
271 
272 		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
273 
274 		do {
275 			start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
276 			local_stats = *percpu_stats;
277 		} while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
278 
279 		stats->rx_bytes		+= local_stats.rx_bytes;
280 		stats->rx_packets	+= local_stats.rx_packets;
281 		stats->tx_bytes		+= local_stats.tx_bytes;
282 		stats->tx_packets	+= local_stats.tx_packets;
283 	}
284 }
285 
286 /**
287  *	ovs_vport_get_options - retrieve device options
288  *
289  * @vport: vport from which to retrieve the options.
290  * @skb: sk_buff where options should be appended.
291  *
292  * Retrieves the configuration of the given device, appending an
293  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
294  * vport-specific attributes to @skb.
295  *
296  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
297  * negative error code if a real error occurred.  If an error occurs, @skb is
298  * left unmodified.
299  *
300  * Must be called with RTNL lock or rcu_read_lock.
301  */
302 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
303 {
304 	struct nlattr *nla;
305 
306 	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
307 	if (!nla)
308 		return -EMSGSIZE;
309 
310 	if (vport->ops->get_options) {
311 		int err = vport->ops->get_options(vport, skb);
312 		if (err) {
313 			nla_nest_cancel(skb, nla);
314 			return err;
315 		}
316 	}
317 
318 	nla_nest_end(skb, nla);
319 	return 0;
320 }
321 
322 /**
323  *	ovs_vport_receive - pass up received packet to the datapath for processing
324  *
325  * @vport: vport that received the packet
326  * @skb: skb that was received
327  *
328  * Must be called with rcu_read_lock.  The packet cannot be shared and
329  * skb->data should point to the Ethernet header.  The caller must have already
330  * called compute_ip_summed() to initialize the checksumming fields.
331  */
332 void ovs_vport_receive(struct vport *vport, struct sk_buff *skb)
333 {
334 	struct vport_percpu_stats *stats;
335 
336 	stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
337 
338 	u64_stats_update_begin(&stats->sync);
339 	stats->rx_packets++;
340 	stats->rx_bytes += skb->len;
341 	u64_stats_update_end(&stats->sync);
342 
343 	ovs_dp_process_received_packet(vport, skb);
344 }
345 
346 /**
347  *	ovs_vport_send - send a packet on a device
348  *
349  * @vport: vport on which to send the packet
350  * @skb: skb to send
351  *
352  * Sends the given packet and returns the length of data sent.  Either RTNL
353  * lock or rcu_read_lock must be held.
354  */
355 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
356 {
357 	int sent = vport->ops->send(vport, skb);
358 
359 	if (likely(sent)) {
360 		struct vport_percpu_stats *stats;
361 
362 		stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
363 
364 		u64_stats_update_begin(&stats->sync);
365 		stats->tx_packets++;
366 		stats->tx_bytes += sent;
367 		u64_stats_update_end(&stats->sync);
368 	}
369 	return sent;
370 }
371 
372 /**
373  *	ovs_vport_record_error - indicate device error to generic stats layer
374  *
375  * @vport: vport that encountered the error
376  * @err_type: one of enum vport_err_type types to indicate the error type
377  *
378  * If using the vport generic stats layer indicate that an error of the given
379  * type has occured.
380  */
381 void ovs_vport_record_error(struct vport *vport, enum vport_err_type err_type)
382 {
383 	spin_lock(&vport->stats_lock);
384 
385 	switch (err_type) {
386 	case VPORT_E_RX_DROPPED:
387 		vport->err_stats.rx_dropped++;
388 		break;
389 
390 	case VPORT_E_RX_ERROR:
391 		vport->err_stats.rx_errors++;
392 		break;
393 
394 	case VPORT_E_TX_DROPPED:
395 		vport->err_stats.tx_dropped++;
396 		break;
397 
398 	case VPORT_E_TX_ERROR:
399 		vport->err_stats.tx_errors++;
400 		break;
401 	};
402 
403 	spin_unlock(&vport->stats_lock);
404 }
405