vport.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright (c) 2007-2014 Nicira, Inc.
  4 */
  5
  6#include <linux/etherdevice.h>
  7#include <linux/if.h>
  8#include <linux/if_vlan.h>
  9#include <linux/jhash.h>
 10#include <linux/kernel.h>
 11#include <linux/list.h>
 12#include <linux/mutex.h>
 13#include <linux/percpu.h>
 14#include <linux/rcupdate.h>
 15#include <linux/rtnetlink.h>
 16#include <linux/compat.h>
 17#include <net/net_namespace.h>
 18#include <linux/module.h>
 19
 20#include "datapath.h"
 21#include "vport.h"
 22#include "vport-internal_dev.h"
 23
 24static LIST_HEAD(vport_ops_list);
 25
 26/* Protected by RCU read lock for reading, ovs_mutex for writing. */
 27static struct hlist_head *dev_table;
 28#define VPORT_HASH_BUCKETS 1024
 29
 30/**
 31 *	ovs_vport_init - initialize vport subsystem
 32 *
 33 * Called at module load time to initialize the vport subsystem.
 34 */
 35int ovs_vport_init(void)
 36{
 37	dev_table = kcalloc(VPORT_HASH_BUCKETS, sizeof(struct hlist_head),
 38			    GFP_KERNEL);
 39	if (!dev_table)
 40		return -ENOMEM;
 41
 42	return 0;
 43}
 44
 45/**
 46 *	ovs_vport_exit - shutdown vport subsystem
 47 *
 48 * Called at module exit time to shutdown the vport subsystem.
 49 */
 50void ovs_vport_exit(void)
 51{
 52	kfree(dev_table);
 53}
 54
 55static struct hlist_head *hash_bucket(const struct net *net, const char *name)
 56{
 57	unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
 58	return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
 59}
 60
 61int __ovs_vport_ops_register(struct vport_ops *ops)
 62{
 63	int err = -EEXIST;
 64	struct vport_ops *o;
 65
 66	ovs_lock();
 67	list_for_each_entry(o, &vport_ops_list, list)
 68		if (ops->type == o->type)
 69			goto errout;
 70
 71	list_add_tail(&ops->list, &vport_ops_list);
 72	err = 0;
 73errout:
 74	ovs_unlock();
 75	return err;
 76}
 77EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
 78
 79void ovs_vport_ops_unregister(struct vport_ops *ops)
 80{
 81	ovs_lock();
 82	list_del(&ops->list);
 83	ovs_unlock();
 84}
 85EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
 86
 87/**
 88 *	ovs_vport_locate - find a port that has already been created
 89 *
 90 * @net: network namespace
 91 * @name: name of port to find
 92 *
 93 * Must be called with ovs or RCU read lock.
 94 */
 95struct vport *ovs_vport_locate(const 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				 lockdep_ovsl_is_held())
102		if (!strcmp(name, ovs_vport_name(vport)) &&
103		    net_eq(ovs_dp_get_net(vport->dp), net))
104			return vport;
105
106	return NULL;
107}
108
109/**
110 *	ovs_vport_alloc - allocate and initialize new vport
111 *
112 * @priv_size: Size of private data area to allocate.
113 * @ops: vport device ops
114 * @parms: information about new vport.
115 *
116 * Allocate and initialize a new vport defined by @ops.  The vport will contain
117 * a private data area of size @priv_size that can be accessed using
118 * vport_priv().  Some parameters of the vport will be initialized from @parms.
119 * @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	int err;
128
129	alloc_size = sizeof(struct vport);
130	if (priv_size) {
131		alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
132		alloc_size += priv_size;
133	}
134
135	vport = kzalloc(alloc_size, GFP_KERNEL);
136	if (!vport)
137		return ERR_PTR(-ENOMEM);
138
139	vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
140	if (!vport->upcall_stats) {
141		err = -ENOMEM;
142		goto err_kfree_vport;
143	}
144
145	vport->dp = parms->dp;
146	vport->port_no = parms->port_no;
147	vport->ops = ops;
148	INIT_HLIST_NODE(&vport->dp_hash_node);
149
150	if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
151		err = -EINVAL;
152		goto err_free_percpu;
153	}
154
155	return vport;
156
157err_free_percpu:
158	free_percpu(vport->upcall_stats);
159err_kfree_vport:
160	kfree(vport);
161	return ERR_PTR(err);
162}
163EXPORT_SYMBOL_GPL(ovs_vport_alloc);
164
165/**
166 *	ovs_vport_free - uninitialize and free vport
167 *
168 * @vport: vport to free
169 *
170 * Frees a vport allocated with vport_alloc() when it is no longer needed.
171 *
172 * The caller must ensure that an RCU grace period has passed since the last
173 * time @vport was in a datapath.
174 */
175void ovs_vport_free(struct vport *vport)
176{
177	/* vport is freed from RCU callback or error path, Therefore
178	 * it is safe to use raw dereference.
179	 */
180	kfree(rcu_dereference_raw(vport->upcall_portids));
181	free_percpu(vport->upcall_stats);
182	kfree(vport);
183}
184EXPORT_SYMBOL_GPL(ovs_vport_free);
185
186static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
187{
188	struct vport_ops *ops;
189
190	list_for_each_entry(ops, &vport_ops_list, list)
191		if (ops->type == parms->type)
192			return ops;
193
194	return NULL;
195}
196
197/**
198 *	ovs_vport_add - add vport device (for kernel callers)
199 *
200 * @parms: Information about new vport.
201 *
202 * Creates a new vport with the specified configuration (which is dependent on
203 * device type).  ovs_mutex must be held.
204 */
205struct vport *ovs_vport_add(const struct vport_parms *parms)
206{
207	struct vport_ops *ops;
208	struct vport *vport;
209
210	ops = ovs_vport_lookup(parms);
211	if (ops) {
212		struct hlist_head *bucket;
213
214		if (!try_module_get(ops->owner))
215			return ERR_PTR(-EAFNOSUPPORT);
216
217		vport = ops->create(parms);
218		if (IS_ERR(vport)) {
219			module_put(ops->owner);
220			return vport;
221		}
222
223		bucket = hash_bucket(ovs_dp_get_net(vport->dp),
224				     ovs_vport_name(vport));
225		hlist_add_head_rcu(&vport->hash_node, bucket);
226		return vport;
227	}
228
229	/* Unlock to attempt module load and return -EAGAIN if load
230	 * was successful as we need to restart the port addition
231	 * workflow.
232	 */
233	ovs_unlock();
234	request_module("vport-type-%d", parms->type);
235	ovs_lock();
236
237	if (!ovs_vport_lookup(parms))
238		return ERR_PTR(-EAFNOSUPPORT);
239	else
240		return ERR_PTR(-EAGAIN);
241}
242
243/**
244 *	ovs_vport_set_options - modify existing vport device (for kernel callers)
245 *
246 * @vport: vport to modify.
247 * @options: New configuration.
248 *
249 * Modifies an existing device with the specified configuration (which is
250 * dependent on device type).  ovs_mutex must be held.
251 */
252int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
253{
254	if (!vport->ops->set_options)
255		return -EOPNOTSUPP;
256	return vport->ops->set_options(vport, options);
257}
258
259/**
260 *	ovs_vport_del - delete existing vport device
261 *
262 * @vport: vport to delete.
263 *
264 * Detaches @vport from its datapath and destroys it.  ovs_mutex must
265 * be held.
266 */
267void ovs_vport_del(struct vport *vport)
268{
269	hlist_del_rcu(&vport->hash_node);
270	module_put(vport->ops->owner);
271	vport->ops->destroy(vport);
272}
273
274/**
275 *	ovs_vport_get_stats - retrieve device stats
276 *
277 * @vport: vport from which to retrieve the stats
278 * @stats: location to store stats
279 *
280 * Retrieves transmit, receive, and error stats for the given device.
281 *
282 * Must be called with ovs_mutex or rcu_read_lock.
283 */
284void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
285{
286	const struct rtnl_link_stats64 *dev_stats;
287	struct rtnl_link_stats64 temp;
288
289	dev_stats = dev_get_stats(vport->dev, &temp);
290	stats->rx_errors  = dev_stats->rx_errors;
291	stats->tx_errors  = dev_stats->tx_errors;
292	stats->tx_dropped = dev_stats->tx_dropped;
293	stats->rx_dropped = dev_stats->rx_dropped;
294
295	stats->rx_bytes	  = dev_stats->rx_bytes;
296	stats->rx_packets = dev_stats->rx_packets;
297	stats->tx_bytes	  = dev_stats->tx_bytes;
298	stats->tx_packets = dev_stats->tx_packets;
299}
300
301/**
302 *	ovs_vport_get_upcall_stats - retrieve upcall stats
303 *
304 * @vport: vport from which to retrieve the stats.
305 * @skb: sk_buff where upcall stats should be appended.
306 *
307 * Retrieves upcall stats for the given device.
308 *
309 * Must be called with ovs_mutex or rcu_read_lock.
310 */
311int ovs_vport_get_upcall_stats(struct vport *vport, struct sk_buff *skb)
312{
313	u64 tx_success = 0, tx_fail = 0;
314	struct nlattr *nla;
315	int i;
316
317	for_each_possible_cpu(i) {
318		const struct vport_upcall_stats_percpu *stats;
319		u64 n_success, n_fail;
320		unsigned int start;
321
322		stats = per_cpu_ptr(vport->upcall_stats, i);
323		do {
324			start = u64_stats_fetch_begin(&stats->syncp);
325			n_success = u64_stats_read(&stats->n_success);
326			n_fail = u64_stats_read(&stats->n_fail);
327		} while (u64_stats_fetch_retry(&stats->syncp, start));
328		tx_success += n_success;
329		tx_fail += n_fail;
330	}
331
332	nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_UPCALL_STATS);
333	if (!nla)
334		return -EMSGSIZE;
335
336	if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_SUCCESS, tx_success,
337			      OVS_VPORT_ATTR_PAD)) {
338		nla_nest_cancel(skb, nla);
339		return -EMSGSIZE;
340	}
341
342	if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_FAIL, tx_fail,
343			      OVS_VPORT_ATTR_PAD)) {
344		nla_nest_cancel(skb, nla);
345		return -EMSGSIZE;
346	}
347	nla_nest_end(skb, nla);
348
349	return 0;
350}
351
352/**
353 *	ovs_vport_get_options - retrieve device options
354 *
355 * @vport: vport from which to retrieve the options.
356 * @skb: sk_buff where options should be appended.
357 *
358 * Retrieves the configuration of the given device, appending an
359 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
360 * vport-specific attributes to @skb.
361 *
362 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
363 * negative error code if a real error occurred.  If an error occurs, @skb is
364 * left unmodified.
365 *
366 * Must be called with ovs_mutex or rcu_read_lock.
367 */
368int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
369{
370	struct nlattr *nla;
371	int err;
372
373	if (!vport->ops->get_options)
374		return 0;
375
376	nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
377	if (!nla)
378		return -EMSGSIZE;
379
380	err = vport->ops->get_options(vport, skb);
381	if (err) {
382		nla_nest_cancel(skb, nla);
383		return err;
384	}
385
386	nla_nest_end(skb, nla);
387	return 0;
388}
389
390/**
391 *	ovs_vport_set_upcall_portids - set upcall portids of @vport.
392 *
393 * @vport: vport to modify.
394 * @ids: new configuration, an array of port ids.
395 *
396 * Sets the vport's upcall_portids to @ids.
397 *
398 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
399 * as an array of U32.
400 *
401 * Must be called with ovs_mutex.
402 */
403int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
404{
405	struct vport_portids *old, *vport_portids;
406
407	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
408		return -EINVAL;
409
410	old = ovsl_dereference(vport->upcall_portids);
411
412	vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
413				GFP_KERNEL);
414	if (!vport_portids)
415		return -ENOMEM;
416
417	vport_portids->n_ids = nla_len(ids) / sizeof(u32);
418	vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
419	nla_memcpy(vport_portids->ids, ids, nla_len(ids));
420
421	rcu_assign_pointer(vport->upcall_portids, vport_portids);
422
423	if (old)
424		kfree_rcu(old, rcu);
425	return 0;
426}
427
428/**
429 *	ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
430 *
431 * @vport: vport from which to retrieve the portids.
432 * @skb: sk_buff where portids should be appended.
433 *
434 * Retrieves the configuration of the given vport, appending the
435 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
436 * portids to @skb.
437 *
438 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
439 * If an error occurs, @skb is left unmodified.  Must be called with
440 * ovs_mutex or rcu_read_lock.
441 */
442int ovs_vport_get_upcall_portids(const struct vport *vport,
443				 struct sk_buff *skb)
444{
445	struct vport_portids *ids;
446
447	ids = rcu_dereference_ovsl(vport->upcall_portids);
448
449	if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
450		return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
451			       ids->n_ids * sizeof(u32), (void *)ids->ids);
452	else
453		return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
454}
455
456/**
457 *	ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
458 *
459 * @vport: vport from which the missed packet is received.
460 * @skb: skb that the missed packet was received.
461 *
462 * Uses the skb_get_hash() to select the upcall portid to send the
463 * upcall.
464 *
465 * Returns the portid of the target socket.  Must be called with rcu_read_lock.
466 */
467u32 ovs_vport_find_upcall_portid(const struct vport *vport,
468				 struct sk_buff *skb)
469{
470	struct vport_portids *ids;
471	u32 ids_index;
472	u32 hash;
473
474	ids = rcu_dereference(vport->upcall_portids);
475
476	/* If there is only one portid, select it in the fast-path. */
477	if (ids->n_ids == 1)
478		return ids->ids[0];
479
480	hash = skb_get_hash(skb);
481	ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
482	return ids->ids[ids_index];
483}
484
485/**
486 *	ovs_vport_receive - pass up received packet to the datapath for processing
487 *
488 * @vport: vport that received the packet
489 * @skb: skb that was received
490 * @tun_info: tunnel (if any) that carried packet
491 *
492 * Must be called with rcu_read_lock.  The packet cannot be shared and
493 * skb->data should point to the Ethernet header.
494 */
495int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
496		      const struct ip_tunnel_info *tun_info)
497{
498	struct sw_flow_key key;
499	int error;
500
501	OVS_CB(skb)->input_vport = vport;
502	OVS_CB(skb)->mru = 0;
503	OVS_CB(skb)->cutlen = 0;
504	OVS_CB(skb)->probability = 0;
505	OVS_CB(skb)->upcall_pid = 0;
506	if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
507		u32 mark;
508
509		mark = skb->mark;
510		skb_scrub_packet(skb, true);
511		skb->mark = mark;
512		tun_info = NULL;
513	}
514
515	/* Extract flow from 'skb' into 'key'. */
516	error = ovs_flow_key_extract(tun_info, skb, &key);
517	if (unlikely(error)) {
518		kfree_skb(skb);
519		return error;
520	}
521	ovs_dp_process_packet(skb, &key);
522	return 0;
523}
524
525static int packet_length(const struct sk_buff *skb,
526			 struct net_device *dev)
527{
528	int length = skb->len - dev->hard_header_len;
529
530	if (!skb_vlan_tag_present(skb) &&
531	    eth_type_vlan(skb->protocol))
532		length -= VLAN_HLEN;
533
534	/* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
535	 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
536	 * account for 802.1ad. e.g. is_skb_forwardable().
537	 */
538
539	return length > 0 ? length : 0;
540}
541
542void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
543{
544	int mtu = vport->dev->mtu;
545
546	switch (vport->dev->type) {
547	case ARPHRD_NONE:
548		if (mac_proto == MAC_PROTO_ETHERNET) {
549			skb_reset_network_header(skb);
550			skb_reset_mac_len(skb);
551			skb->protocol = htons(ETH_P_TEB);
552		} else if (mac_proto != MAC_PROTO_NONE) {
553			WARN_ON_ONCE(1);
554			goto drop;
555		}
556		break;
557	case ARPHRD_ETHER:
558		if (mac_proto != MAC_PROTO_ETHERNET)
559			goto drop;
560		break;
561	default:
562		goto drop;
563	}
564
565	if (unlikely(packet_length(skb, vport->dev) > mtu &&
566		     !skb_is_gso(skb))) {
567		vport->dev->stats.tx_errors++;
568		if (vport->dev->flags & IFF_UP)
569			net_warn_ratelimited("%s: dropped over-mtu packet: "
570					     "%d > %d\n", vport->dev->name,
571					     packet_length(skb, vport->dev),
572					     mtu);
573		goto drop;
574	}
575
576	skb->dev = vport->dev;
577	skb_clear_tstamp(skb);
578	vport->ops->send(skb);
579	return;
580
581drop:
582	kfree_skb(skb);
583}