Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 = kzalloc_objs(struct hlist_head, VPORT_HASH_BUCKETS);
38 if (!dev_table)
39 return -ENOMEM;
40
41 return 0;
42}
43
44/**
45 * ovs_vport_exit - shutdown vport subsystem
46 *
47 * Called at module exit time to shutdown the vport subsystem.
48 */
49void ovs_vport_exit(void)
50{
51 kfree(dev_table);
52}
53
54static struct hlist_head *hash_bucket(const struct net *net, const char *name)
55{
56 unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
57 return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
58}
59
60int __ovs_vport_ops_register(struct vport_ops *ops)
61{
62 int err = -EEXIST;
63 struct vport_ops *o;
64
65 ovs_lock();
66 list_for_each_entry(o, &vport_ops_list, list)
67 if (ops->type == o->type)
68 goto errout;
69
70 list_add_tail(&ops->list, &vport_ops_list);
71 err = 0;
72errout:
73 ovs_unlock();
74 return err;
75}
76EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
77
78void ovs_vport_ops_unregister(struct vport_ops *ops)
79{
80 ovs_lock();
81 list_del(&ops->list);
82 ovs_unlock();
83}
84EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
85
86/**
87 * ovs_vport_locate - find a port that has already been created
88 *
89 * @net: network namespace
90 * @name: name of port to find
91 *
92 * Must be called with ovs or RCU read lock.
93 */
94struct vport *ovs_vport_locate(const struct net *net, const char *name)
95{
96 struct hlist_head *bucket = hash_bucket(net, name);
97 struct vport *vport;
98
99 hlist_for_each_entry_rcu(vport, bucket, hash_node,
100 lockdep_ovsl_is_held())
101 if (!strcmp(name, ovs_vport_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 * @parms: information about new vport.
114 *
115 * Allocate and initialize a new vport defined by @ops. The vport will contain
116 * a private data area of size @priv_size that can be accessed using
117 * vport_priv(). Some parameters of the vport will be initialized from @parms.
118 * @vports that are no longer needed should be released with
119 * vport_free().
120 */
121struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
122 const struct vport_parms *parms)
123{
124 struct vport *vport;
125 size_t alloc_size;
126 int err;
127
128 alloc_size = sizeof(struct vport);
129 if (priv_size) {
130 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
131 alloc_size += priv_size;
132 }
133
134 vport = kzalloc(alloc_size, GFP_KERNEL);
135 if (!vport)
136 return ERR_PTR(-ENOMEM);
137
138 vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
139 if (!vport->upcall_stats) {
140 err = -ENOMEM;
141 goto err_kfree_vport;
142 }
143
144 vport->dp = parms->dp;
145 vport->port_no = parms->port_no;
146 vport->ops = ops;
147 INIT_HLIST_NODE(&vport->dp_hash_node);
148
149 if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
150 err = -EINVAL;
151 goto err_free_percpu;
152 }
153
154 return vport;
155
156err_free_percpu:
157 free_percpu(vport->upcall_stats);
158err_kfree_vport:
159 kfree(vport);
160 return ERR_PTR(err);
161}
162EXPORT_SYMBOL_GPL(ovs_vport_alloc);
163
164/**
165 * ovs_vport_free - uninitialize and free vport
166 *
167 * @vport: vport to free
168 *
169 * Frees a vport allocated with vport_alloc() when it is no longer needed.
170 *
171 * The caller must ensure that an RCU grace period has passed since the last
172 * time @vport was in a datapath.
173 */
174void ovs_vport_free(struct vport *vport)
175{
176 /* vport is freed from RCU callback or error path, Therefore
177 * it is safe to use raw dereference.
178 */
179 kfree(rcu_dereference_raw(vport->upcall_portids));
180 free_percpu(vport->upcall_stats);
181 kfree(vport);
182}
183EXPORT_SYMBOL_GPL(ovs_vport_free);
184
185static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
186{
187 struct vport_ops *ops;
188
189 list_for_each_entry(ops, &vport_ops_list, list)
190 if (ops->type == parms->type)
191 return ops;
192
193 return NULL;
194}
195
196/**
197 * ovs_vport_add - add vport device (for kernel callers)
198 *
199 * @parms: Information about new vport.
200 *
201 * Creates a new vport with the specified configuration (which is dependent on
202 * device type). ovs_mutex must be held.
203 */
204struct vport *ovs_vport_add(const struct vport_parms *parms)
205{
206 struct vport_ops *ops;
207 struct vport *vport;
208
209 ops = ovs_vport_lookup(parms);
210 if (ops) {
211 struct hlist_head *bucket;
212
213 if (!try_module_get(ops->owner))
214 return ERR_PTR(-EAFNOSUPPORT);
215
216 vport = ops->create(parms);
217 if (IS_ERR(vport)) {
218 module_put(ops->owner);
219 return vport;
220 }
221
222 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
223 ovs_vport_name(vport));
224 hlist_add_head_rcu(&vport->hash_node, bucket);
225 return vport;
226 }
227
228 /* Unlock to attempt module load and return -EAGAIN if load
229 * was successful as we need to restart the port addition
230 * workflow.
231 */
232 ovs_unlock();
233 request_module("vport-type-%d", parms->type);
234 ovs_lock();
235
236 if (!ovs_vport_lookup(parms))
237 return ERR_PTR(-EAFNOSUPPORT);
238 else
239 return ERR_PTR(-EAGAIN);
240}
241
242/**
243 * ovs_vport_set_options - modify existing vport device (for kernel callers)
244 *
245 * @vport: vport to modify.
246 * @options: New configuration.
247 *
248 * Modifies an existing device with the specified configuration (which is
249 * dependent on device type). ovs_mutex must be held.
250 */
251int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
252{
253 if (!vport->ops->set_options)
254 return -EOPNOTSUPP;
255 return vport->ops->set_options(vport, options);
256}
257
258/**
259 * ovs_vport_del - delete existing vport device
260 *
261 * @vport: vport to delete.
262 *
263 * Detaches @vport from its datapath and destroys it. ovs_mutex must
264 * be held.
265 */
266void ovs_vport_del(struct vport *vport)
267{
268 hlist_del_rcu(&vport->hash_node);
269 module_put(vport->ops->owner);
270 vport->ops->destroy(vport);
271}
272
273/**
274 * ovs_vport_get_stats - retrieve device stats
275 *
276 * @vport: vport from which to retrieve the stats
277 * @stats: location to store stats
278 *
279 * Retrieves transmit, receive, and error stats for the given device.
280 *
281 * Must be called with ovs_mutex or rcu_read_lock.
282 */
283void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
284{
285 const struct rtnl_link_stats64 *dev_stats;
286 struct rtnl_link_stats64 temp;
287
288 dev_stats = dev_get_stats(vport->dev, &temp);
289 stats->rx_errors = dev_stats->rx_errors;
290 stats->tx_errors = dev_stats->tx_errors;
291 stats->tx_dropped = dev_stats->tx_dropped;
292 stats->rx_dropped = dev_stats->rx_dropped;
293
294 stats->rx_bytes = dev_stats->rx_bytes;
295 stats->rx_packets = dev_stats->rx_packets;
296 stats->tx_bytes = dev_stats->tx_bytes;
297 stats->tx_packets = dev_stats->tx_packets;
298}
299
300/**
301 * ovs_vport_get_upcall_stats - retrieve upcall stats
302 *
303 * @vport: vport from which to retrieve the stats.
304 * @skb: sk_buff where upcall stats should be appended.
305 *
306 * Retrieves upcall stats for the given device.
307 *
308 * Must be called with ovs_mutex or rcu_read_lock.
309 */
310int ovs_vport_get_upcall_stats(struct vport *vport, struct sk_buff *skb)
311{
312 u64 tx_success = 0, tx_fail = 0;
313 struct nlattr *nla;
314 int i;
315
316 for_each_possible_cpu(i) {
317 const struct vport_upcall_stats_percpu *stats;
318 u64 n_success, n_fail;
319 unsigned int start;
320
321 stats = per_cpu_ptr(vport->upcall_stats, i);
322 do {
323 start = u64_stats_fetch_begin(&stats->syncp);
324 n_success = u64_stats_read(&stats->n_success);
325 n_fail = u64_stats_read(&stats->n_fail);
326 } while (u64_stats_fetch_retry(&stats->syncp, start));
327 tx_success += n_success;
328 tx_fail += n_fail;
329 }
330
331 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_UPCALL_STATS);
332 if (!nla)
333 return -EMSGSIZE;
334
335 if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_SUCCESS, tx_success,
336 OVS_VPORT_ATTR_PAD)) {
337 nla_nest_cancel(skb, nla);
338 return -EMSGSIZE;
339 }
340
341 if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_FAIL, tx_fail,
342 OVS_VPORT_ATTR_PAD)) {
343 nla_nest_cancel(skb, nla);
344 return -EMSGSIZE;
345 }
346 nla_nest_end(skb, nla);
347
348 return 0;
349}
350
351/**
352 * ovs_vport_get_options - retrieve device options
353 *
354 * @vport: vport from which to retrieve the options.
355 * @skb: sk_buff where options should be appended.
356 *
357 * Retrieves the configuration of the given device, appending an
358 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
359 * vport-specific attributes to @skb.
360 *
361 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
362 * negative error code if a real error occurred. If an error occurs, @skb is
363 * left unmodified.
364 *
365 * Must be called with ovs_mutex or rcu_read_lock.
366 */
367int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
368{
369 struct nlattr *nla;
370 int err;
371
372 if (!vport->ops->get_options)
373 return 0;
374
375 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
376 if (!nla)
377 return -EMSGSIZE;
378
379 err = vport->ops->get_options(vport, skb);
380 if (err) {
381 nla_nest_cancel(skb, nla);
382 return err;
383 }
384
385 nla_nest_end(skb, nla);
386 return 0;
387}
388
389/**
390 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
391 *
392 * @vport: vport to modify.
393 * @ids: new configuration, an array of port ids.
394 *
395 * Sets the vport's upcall_portids to @ids.
396 *
397 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
398 * as an array of U32.
399 *
400 * Must be called with ovs_mutex.
401 */
402int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
403{
404 struct vport_portids *old, *vport_portids;
405
406 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
407 return -EINVAL;
408
409 if (nla_len(ids) / sizeof(u32) > nr_cpu_ids)
410 return -EINVAL;
411
412 old = ovsl_dereference(vport->upcall_portids);
413
414 vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
415 GFP_KERNEL);
416 if (!vport_portids)
417 return -ENOMEM;
418
419 vport_portids->n_ids = nla_len(ids) / sizeof(u32);
420 vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
421 nla_memcpy(vport_portids->ids, ids, nla_len(ids));
422
423 rcu_assign_pointer(vport->upcall_portids, vport_portids);
424
425 if (old)
426 kfree_rcu(old, rcu);
427 return 0;
428}
429
430/**
431 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
432 *
433 * @vport: vport from which to retrieve the portids.
434 * @skb: sk_buff where portids should be appended.
435 *
436 * Retrieves the configuration of the given vport, appending the
437 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
438 * portids to @skb.
439 *
440 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
441 * If an error occurs, @skb is left unmodified. Must be called with
442 * ovs_mutex or rcu_read_lock.
443 */
444int ovs_vport_get_upcall_portids(const struct vport *vport,
445 struct sk_buff *skb)
446{
447 struct vport_portids *ids;
448
449 ids = rcu_dereference_ovsl(vport->upcall_portids);
450
451 if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
452 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
453 ids->n_ids * sizeof(u32), (void *)ids->ids);
454 else
455 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
456}
457
458/**
459 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
460 *
461 * @vport: vport from which the missed packet is received.
462 * @skb: skb that the missed packet was received.
463 *
464 * Uses the skb_get_hash() to select the upcall portid to send the
465 * upcall.
466 *
467 * Returns the portid of the target socket. Must be called with rcu_read_lock.
468 */
469u32 ovs_vport_find_upcall_portid(const struct vport *vport,
470 struct sk_buff *skb)
471{
472 struct vport_portids *ids;
473 u32 ids_index;
474 u32 hash;
475
476 ids = rcu_dereference(vport->upcall_portids);
477
478 /* If there is only one portid, select it in the fast-path. */
479 if (ids->n_ids == 1)
480 return ids->ids[0];
481
482 hash = skb_get_hash(skb);
483 ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
484 return ids->ids[ids_index];
485}
486
487/**
488 * ovs_vport_receive - pass up received packet to the datapath for processing
489 *
490 * @vport: vport that received the packet
491 * @skb: skb that was received
492 * @tun_info: tunnel (if any) that carried packet
493 *
494 * Must be called with rcu_read_lock. The packet cannot be shared and
495 * skb->data should point to the Ethernet header.
496 */
497int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
498 const struct ip_tunnel_info *tun_info)
499{
500 struct sw_flow_key key;
501 int error;
502
503 OVS_CB(skb)->input_vport = vport;
504 OVS_CB(skb)->mru = 0;
505 OVS_CB(skb)->cutlen = 0;
506 OVS_CB(skb)->probability = 0;
507 OVS_CB(skb)->upcall_pid = 0;
508 if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
509 u32 mark;
510
511 mark = skb->mark;
512 skb_scrub_packet(skb, true);
513 skb->mark = mark;
514 tun_info = NULL;
515 }
516
517 /* Extract flow from 'skb' into 'key'. */
518 error = ovs_flow_key_extract(tun_info, skb, &key);
519 if (unlikely(error)) {
520 kfree_skb(skb);
521 return error;
522 }
523 ovs_dp_process_packet(skb, &key);
524 return 0;
525}
526
527static int packet_length(const struct sk_buff *skb,
528 struct net_device *dev)
529{
530 int length = skb->len - dev->hard_header_len;
531
532 if (!skb_vlan_tag_present(skb) &&
533 eth_type_vlan(skb->protocol))
534 length -= VLAN_HLEN;
535
536 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
537 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
538 * account for 802.1ad. e.g. is_skb_forwardable().
539 */
540
541 return length > 0 ? length : 0;
542}
543
544void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
545{
546 int mtu = vport->dev->mtu;
547
548 switch (vport->dev->type) {
549 case ARPHRD_NONE:
550 if (mac_proto == MAC_PROTO_ETHERNET) {
551 skb_reset_network_header(skb);
552 skb_reset_mac_len(skb);
553 skb->protocol = htons(ETH_P_TEB);
554 } else if (mac_proto != MAC_PROTO_NONE) {
555 WARN_ON_ONCE(1);
556 goto drop;
557 }
558 break;
559 case ARPHRD_ETHER:
560 if (mac_proto != MAC_PROTO_ETHERNET)
561 goto drop;
562 break;
563 default:
564 goto drop;
565 }
566
567 if (unlikely(packet_length(skb, vport->dev) > mtu &&
568 !skb_is_gso(skb))) {
569 vport->dev->stats.tx_errors++;
570 if (vport->dev->flags & IFF_UP)
571 net_warn_ratelimited("%s: dropped over-mtu packet: "
572 "%d > %d\n", vport->dev->name,
573 packet_length(skb, vport->dev),
574 mtu);
575 goto drop;
576 }
577
578 skb->dev = vport->dev;
579 skb_clear_tstamp(skb);
580 vport->ops->send(skb);
581 return;
582
583drop:
584 kfree_skb(skb);
585}