xref: /linux/net/openvswitch/vport.c (revision 90ab5ee94171b3e28de6bb42ee30b527014e0be7)
1 /*
2  * Copyright (c) 2007-2011 Nicira Networks.
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/dcache.h>
20 #include <linux/etherdevice.h>
21 #include <linux/if.h>
22 #include <linux/if_vlan.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 <linux/version.h>
31 
32 #include "vport.h"
33 #include "vport-internal_dev.h"
34 
35 /* List of statically compiled vport implementations.  Don't forget to also
36  * add yours to the list at the bottom of vport.h. */
37 static const struct vport_ops *vport_ops_list[] = {
38 	&ovs_netdev_vport_ops,
39 	&ovs_internal_vport_ops,
40 };
41 
42 /* Protected by RCU read lock for reading, RTNL lock for writing. */
43 static 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  */
51 int 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  */
66 void ovs_vport_exit(void)
67 {
68 	kfree(dev_table);
69 }
70 
71 static struct hlist_head *hash_bucket(const char *name)
72 {
73 	unsigned int hash = full_name_hash(name, strlen(name));
74 	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
75 }
76 
77 /**
78  *	ovs_vport_locate - find a port that has already been created
79  *
80  * @name: name of port to find
81  *
82  * Must be called with RTNL or RCU read lock.
83  */
84 struct vport *ovs_vport_locate(const char *name)
85 {
86 	struct hlist_head *bucket = hash_bucket(name);
87 	struct vport *vport;
88 	struct hlist_node *node;
89 
90 	hlist_for_each_entry_rcu(vport, node, bucket, hash_node)
91 		if (!strcmp(name, vport->ops->get_name(vport)))
92 			return vport;
93 
94 	return NULL;
95 }
96 
97 /**
98  *	ovs_vport_alloc - allocate and initialize new vport
99  *
100  * @priv_size: Size of private data area to allocate.
101  * @ops: vport device ops
102  *
103  * Allocate and initialize a new vport defined by @ops.  The vport will contain
104  * a private data area of size @priv_size that can be accessed using
105  * vport_priv().  vports that are no longer needed should be released with
106  * vport_free().
107  */
108 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
109 			  const struct vport_parms *parms)
110 {
111 	struct vport *vport;
112 	size_t alloc_size;
113 
114 	alloc_size = sizeof(struct vport);
115 	if (priv_size) {
116 		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
117 		alloc_size += priv_size;
118 	}
119 
120 	vport = kzalloc(alloc_size, GFP_KERNEL);
121 	if (!vport)
122 		return ERR_PTR(-ENOMEM);
123 
124 	vport->dp = parms->dp;
125 	vport->port_no = parms->port_no;
126 	vport->upcall_pid = parms->upcall_pid;
127 	vport->ops = ops;
128 
129 	vport->percpu_stats = alloc_percpu(struct vport_percpu_stats);
130 	if (!vport->percpu_stats) {
131 		kfree(vport);
132 		return ERR_PTR(-ENOMEM);
133 	}
134 
135 	spin_lock_init(&vport->stats_lock);
136 
137 	return vport;
138 }
139 
140 /**
141  *	ovs_vport_free - uninitialize and free vport
142  *
143  * @vport: vport to free
144  *
145  * Frees a vport allocated with vport_alloc() when it is no longer needed.
146  *
147  * The caller must ensure that an RCU grace period has passed since the last
148  * time @vport was in a datapath.
149  */
150 void ovs_vport_free(struct vport *vport)
151 {
152 	free_percpu(vport->percpu_stats);
153 	kfree(vport);
154 }
155 
156 /**
157  *	ovs_vport_add - add vport device (for kernel callers)
158  *
159  * @parms: Information about new vport.
160  *
161  * Creates a new vport with the specified configuration (which is dependent on
162  * device type).  RTNL lock must be held.
163  */
164 struct vport *ovs_vport_add(const struct vport_parms *parms)
165 {
166 	struct vport *vport;
167 	int err = 0;
168 	int i;
169 
170 	ASSERT_RTNL();
171 
172 	for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
173 		if (vport_ops_list[i]->type == parms->type) {
174 			vport = vport_ops_list[i]->create(parms);
175 			if (IS_ERR(vport)) {
176 				err = PTR_ERR(vport);
177 				goto out;
178 			}
179 
180 			hlist_add_head_rcu(&vport->hash_node,
181 					   hash_bucket(vport->ops->get_name(vport)));
182 			return vport;
183 		}
184 	}
185 
186 	err = -EAFNOSUPPORT;
187 
188 out:
189 	return ERR_PTR(err);
190 }
191 
192 /**
193  *	ovs_vport_set_options - modify existing vport device (for kernel callers)
194  *
195  * @vport: vport to modify.
196  * @port: New configuration.
197  *
198  * Modifies an existing device with the specified configuration (which is
199  * dependent on device type).  RTNL lock must be held.
200  */
201 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
202 {
203 	ASSERT_RTNL();
204 
205 	if (!vport->ops->set_options)
206 		return -EOPNOTSUPP;
207 	return vport->ops->set_options(vport, options);
208 }
209 
210 /**
211  *	ovs_vport_del - delete existing vport device
212  *
213  * @vport: vport to delete.
214  *
215  * Detaches @vport from its datapath and destroys it.  It is possible to fail
216  * for reasons such as lack of memory.  RTNL lock must be held.
217  */
218 void ovs_vport_del(struct vport *vport)
219 {
220 	ASSERT_RTNL();
221 
222 	hlist_del_rcu(&vport->hash_node);
223 
224 	vport->ops->destroy(vport);
225 }
226 
227 /**
228  *	ovs_vport_get_stats - retrieve device stats
229  *
230  * @vport: vport from which to retrieve the stats
231  * @stats: location to store stats
232  *
233  * Retrieves transmit, receive, and error stats for the given device.
234  *
235  * Must be called with RTNL lock or rcu_read_lock.
236  */
237 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
238 {
239 	int i;
240 
241 	memset(stats, 0, sizeof(*stats));
242 
243 	/* We potentially have 2 sources of stats that need to be combined:
244 	 * those we have collected (split into err_stats and percpu_stats) from
245 	 * set_stats() and device error stats from netdev->get_stats() (for
246 	 * errors that happen  downstream and therefore aren't reported through
247 	 * our vport_record_error() function).
248 	 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
249 	 * netdev-stats can be directly read over netlink-ioctl.
250 	 */
251 
252 	spin_lock_bh(&vport->stats_lock);
253 
254 	stats->rx_errors	= vport->err_stats.rx_errors;
255 	stats->tx_errors	= vport->err_stats.tx_errors;
256 	stats->tx_dropped	= vport->err_stats.tx_dropped;
257 	stats->rx_dropped	= vport->err_stats.rx_dropped;
258 
259 	spin_unlock_bh(&vport->stats_lock);
260 
261 	for_each_possible_cpu(i) {
262 		const struct vport_percpu_stats *percpu_stats;
263 		struct vport_percpu_stats local_stats;
264 		unsigned int start;
265 
266 		percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
267 
268 		do {
269 			start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
270 			local_stats = *percpu_stats;
271 		} while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
272 
273 		stats->rx_bytes		+= local_stats.rx_bytes;
274 		stats->rx_packets	+= local_stats.rx_packets;
275 		stats->tx_bytes		+= local_stats.tx_bytes;
276 		stats->tx_packets	+= local_stats.tx_packets;
277 	}
278 }
279 
280 /**
281  *	ovs_vport_get_options - retrieve device options
282  *
283  * @vport: vport from which to retrieve the options.
284  * @skb: sk_buff where options should be appended.
285  *
286  * Retrieves the configuration of the given device, appending an
287  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
288  * vport-specific attributes to @skb.
289  *
290  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
291  * negative error code if a real error occurred.  If an error occurs, @skb is
292  * left unmodified.
293  *
294  * Must be called with RTNL lock or rcu_read_lock.
295  */
296 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
297 {
298 	struct nlattr *nla;
299 
300 	nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
301 	if (!nla)
302 		return -EMSGSIZE;
303 
304 	if (vport->ops->get_options) {
305 		int err = vport->ops->get_options(vport, skb);
306 		if (err) {
307 			nla_nest_cancel(skb, nla);
308 			return err;
309 		}
310 	}
311 
312 	nla_nest_end(skb, nla);
313 	return 0;
314 }
315 
316 /**
317  *	ovs_vport_receive - pass up received packet to the datapath for processing
318  *
319  * @vport: vport that received the packet
320  * @skb: skb that was received
321  *
322  * Must be called with rcu_read_lock.  The packet cannot be shared and
323  * skb->data should point to the Ethernet header.  The caller must have already
324  * called compute_ip_summed() to initialize the checksumming fields.
325  */
326 void ovs_vport_receive(struct vport *vport, struct sk_buff *skb)
327 {
328 	struct vport_percpu_stats *stats;
329 
330 	stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
331 
332 	u64_stats_update_begin(&stats->sync);
333 	stats->rx_packets++;
334 	stats->rx_bytes += skb->len;
335 	u64_stats_update_end(&stats->sync);
336 
337 	ovs_dp_process_received_packet(vport, skb);
338 }
339 
340 /**
341  *	ovs_vport_send - send a packet on a device
342  *
343  * @vport: vport on which to send the packet
344  * @skb: skb to send
345  *
346  * Sends the given packet and returns the length of data sent.  Either RTNL
347  * lock or rcu_read_lock must be held.
348  */
349 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
350 {
351 	int sent = vport->ops->send(vport, skb);
352 
353 	if (likely(sent)) {
354 		struct vport_percpu_stats *stats;
355 
356 		stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
357 
358 		u64_stats_update_begin(&stats->sync);
359 		stats->tx_packets++;
360 		stats->tx_bytes += sent;
361 		u64_stats_update_end(&stats->sync);
362 	}
363 	return sent;
364 }
365 
366 /**
367  *	ovs_vport_record_error - indicate device error to generic stats layer
368  *
369  * @vport: vport that encountered the error
370  * @err_type: one of enum vport_err_type types to indicate the error type
371  *
372  * If using the vport generic stats layer indicate that an error of the given
373  * type has occured.
374  */
375 void ovs_vport_record_error(struct vport *vport, enum vport_err_type err_type)
376 {
377 	spin_lock(&vport->stats_lock);
378 
379 	switch (err_type) {
380 	case VPORT_E_RX_DROPPED:
381 		vport->err_stats.rx_dropped++;
382 		break;
383 
384 	case VPORT_E_RX_ERROR:
385 		vport->err_stats.rx_errors++;
386 		break;
387 
388 	case VPORT_E_TX_DROPPED:
389 		vport->err_stats.tx_dropped++;
390 		break;
391 
392 	case VPORT_E_TX_ERROR:
393 		vport->err_stats.tx_errors++;
394 		break;
395 	};
396 
397 	spin_unlock(&vport->stats_lock);
398 }
399