Loading...
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
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->dp = parms->dp;
139 vport->port_no = parms->port_no;
140 vport->ops = ops;
141 INIT_HLIST_NODE(&vport->dp_hash_node);
142
143 if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
144 kfree(vport);
145 return ERR_PTR(-EINVAL);
146 }
147
148 return vport;
149}
150EXPORT_SYMBOL_GPL(ovs_vport_alloc);
151
152/**
153 * ovs_vport_free - uninitialize and free vport
154 *
155 * @vport: vport to free
156 *
157 * Frees a vport allocated with vport_alloc() when it is no longer needed.
158 *
159 * The caller must ensure that an RCU grace period has passed since the last
160 * time @vport was in a datapath.
161 */
162void ovs_vport_free(struct vport *vport)
163{
164 /* vport is freed from RCU callback or error path, Therefore
165 * it is safe to use raw dereference.
166 */
167 kfree(rcu_dereference_raw(vport->upcall_portids));
168 kfree(vport);
169}
170EXPORT_SYMBOL_GPL(ovs_vport_free);
171
172static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
173{
174 struct vport_ops *ops;
175
176 list_for_each_entry(ops, &vport_ops_list, list)
177 if (ops->type == parms->type)
178 return ops;
179
180 return NULL;
181}
182
183/**
184 * ovs_vport_add - add vport device (for kernel callers)
185 *
186 * @parms: Information about new vport.
187 *
188 * Creates a new vport with the specified configuration (which is dependent on
189 * device type). ovs_mutex must be held.
190 */
191struct vport *ovs_vport_add(const struct vport_parms *parms)
192{
193 struct vport_ops *ops;
194 struct vport *vport;
195
196 ops = ovs_vport_lookup(parms);
197 if (ops) {
198 struct hlist_head *bucket;
199
200 if (!try_module_get(ops->owner))
201 return ERR_PTR(-EAFNOSUPPORT);
202
203 vport = ops->create(parms);
204 if (IS_ERR(vport)) {
205 module_put(ops->owner);
206 return vport;
207 }
208
209 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
210 ovs_vport_name(vport));
211 hlist_add_head_rcu(&vport->hash_node, bucket);
212 return vport;
213 }
214
215 /* Unlock to attempt module load and return -EAGAIN if load
216 * was successful as we need to restart the port addition
217 * workflow.
218 */
219 ovs_unlock();
220 request_module("vport-type-%d", parms->type);
221 ovs_lock();
222
223 if (!ovs_vport_lookup(parms))
224 return ERR_PTR(-EAFNOSUPPORT);
225 else
226 return ERR_PTR(-EAGAIN);
227}
228
229/**
230 * ovs_vport_set_options - modify existing vport device (for kernel callers)
231 *
232 * @vport: vport to modify.
233 * @options: New configuration.
234 *
235 * Modifies an existing device with the specified configuration (which is
236 * dependent on device type). ovs_mutex must be held.
237 */
238int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
239{
240 if (!vport->ops->set_options)
241 return -EOPNOTSUPP;
242 return vport->ops->set_options(vport, options);
243}
244
245/**
246 * ovs_vport_del - delete existing vport device
247 *
248 * @vport: vport to delete.
249 *
250 * Detaches @vport from its datapath and destroys it. ovs_mutex must
251 * be held.
252 */
253void ovs_vport_del(struct vport *vport)
254{
255 hlist_del_rcu(&vport->hash_node);
256 module_put(vport->ops->owner);
257 vport->ops->destroy(vport);
258}
259
260/**
261 * ovs_vport_get_stats - retrieve device stats
262 *
263 * @vport: vport from which to retrieve the stats
264 * @stats: location to store stats
265 *
266 * Retrieves transmit, receive, and error stats for the given device.
267 *
268 * Must be called with ovs_mutex or rcu_read_lock.
269 */
270void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
271{
272 const struct rtnl_link_stats64 *dev_stats;
273 struct rtnl_link_stats64 temp;
274
275 dev_stats = dev_get_stats(vport->dev, &temp);
276 stats->rx_errors = dev_stats->rx_errors;
277 stats->tx_errors = dev_stats->tx_errors;
278 stats->tx_dropped = dev_stats->tx_dropped;
279 stats->rx_dropped = dev_stats->rx_dropped;
280
281 stats->rx_bytes = dev_stats->rx_bytes;
282 stats->rx_packets = dev_stats->rx_packets;
283 stats->tx_bytes = dev_stats->tx_bytes;
284 stats->tx_packets = dev_stats->tx_packets;
285}
286
287/**
288 * ovs_vport_get_options - retrieve device options
289 *
290 * @vport: vport from which to retrieve the options.
291 * @skb: sk_buff where options should be appended.
292 *
293 * Retrieves the configuration of the given device, appending an
294 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
295 * vport-specific attributes to @skb.
296 *
297 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
298 * negative error code if a real error occurred. If an error occurs, @skb is
299 * left unmodified.
300 *
301 * Must be called with ovs_mutex or rcu_read_lock.
302 */
303int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
304{
305 struct nlattr *nla;
306 int err;
307
308 if (!vport->ops->get_options)
309 return 0;
310
311 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
312 if (!nla)
313 return -EMSGSIZE;
314
315 err = vport->ops->get_options(vport, skb);
316 if (err) {
317 nla_nest_cancel(skb, nla);
318 return err;
319 }
320
321 nla_nest_end(skb, nla);
322 return 0;
323}
324
325/**
326 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
327 *
328 * @vport: vport to modify.
329 * @ids: new configuration, an array of port ids.
330 *
331 * Sets the vport's upcall_portids to @ids.
332 *
333 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
334 * as an array of U32.
335 *
336 * Must be called with ovs_mutex.
337 */
338int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
339{
340 struct vport_portids *old, *vport_portids;
341
342 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
343 return -EINVAL;
344
345 old = ovsl_dereference(vport->upcall_portids);
346
347 vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
348 GFP_KERNEL);
349 if (!vport_portids)
350 return -ENOMEM;
351
352 vport_portids->n_ids = nla_len(ids) / sizeof(u32);
353 vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
354 nla_memcpy(vport_portids->ids, ids, nla_len(ids));
355
356 rcu_assign_pointer(vport->upcall_portids, vport_portids);
357
358 if (old)
359 kfree_rcu(old, rcu);
360 return 0;
361}
362
363/**
364 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
365 *
366 * @vport: vport from which to retrieve the portids.
367 * @skb: sk_buff where portids should be appended.
368 *
369 * Retrieves the configuration of the given vport, appending the
370 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
371 * portids to @skb.
372 *
373 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
374 * If an error occurs, @skb is left unmodified. Must be called with
375 * ovs_mutex or rcu_read_lock.
376 */
377int ovs_vport_get_upcall_portids(const struct vport *vport,
378 struct sk_buff *skb)
379{
380 struct vport_portids *ids;
381
382 ids = rcu_dereference_ovsl(vport->upcall_portids);
383
384 if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
385 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
386 ids->n_ids * sizeof(u32), (void *)ids->ids);
387 else
388 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
389}
390
391/**
392 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
393 *
394 * @vport: vport from which the missed packet is received.
395 * @skb: skb that the missed packet was received.
396 *
397 * Uses the skb_get_hash() to select the upcall portid to send the
398 * upcall.
399 *
400 * Returns the portid of the target socket. Must be called with rcu_read_lock.
401 */
402u32 ovs_vport_find_upcall_portid(const struct vport *vport,
403 struct sk_buff *skb)
404{
405 struct vport_portids *ids;
406 u32 ids_index;
407 u32 hash;
408
409 ids = rcu_dereference(vport->upcall_portids);
410
411 /* If there is only one portid, select it in the fast-path. */
412 if (ids->n_ids == 1)
413 return ids->ids[0];
414
415 hash = skb_get_hash(skb);
416 ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
417 return ids->ids[ids_index];
418}
419
420/**
421 * ovs_vport_receive - pass up received packet to the datapath for processing
422 *
423 * @vport: vport that received the packet
424 * @skb: skb that was received
425 * @tun_info: tunnel (if any) that carried packet
426 *
427 * Must be called with rcu_read_lock. The packet cannot be shared and
428 * skb->data should point to the Ethernet header.
429 */
430int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
431 const struct ip_tunnel_info *tun_info)
432{
433 struct sw_flow_key key;
434 int error;
435
436 OVS_CB(skb)->input_vport = vport;
437 OVS_CB(skb)->mru = 0;
438 OVS_CB(skb)->cutlen = 0;
439 if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
440 u32 mark;
441
442 mark = skb->mark;
443 skb_scrub_packet(skb, true);
444 skb->mark = mark;
445 tun_info = NULL;
446 }
447
448 /* Extract flow from 'skb' into 'key'. */
449 error = ovs_flow_key_extract(tun_info, skb, &key);
450 if (unlikely(error)) {
451 kfree_skb(skb);
452 return error;
453 }
454 ovs_dp_process_packet(skb, &key);
455 return 0;
456}
457
458static int packet_length(const struct sk_buff *skb,
459 struct net_device *dev)
460{
461 int length = skb->len - dev->hard_header_len;
462
463 if (!skb_vlan_tag_present(skb) &&
464 eth_type_vlan(skb->protocol))
465 length -= VLAN_HLEN;
466
467 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
468 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
469 * account for 802.1ad. e.g. is_skb_forwardable().
470 */
471
472 return length > 0 ? length : 0;
473}
474
475void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
476{
477 int mtu = vport->dev->mtu;
478
479 switch (vport->dev->type) {
480 case ARPHRD_NONE:
481 if (mac_proto == MAC_PROTO_ETHERNET) {
482 skb_reset_network_header(skb);
483 skb_reset_mac_len(skb);
484 skb->protocol = htons(ETH_P_TEB);
485 } else if (mac_proto != MAC_PROTO_NONE) {
486 WARN_ON_ONCE(1);
487 goto drop;
488 }
489 break;
490 case ARPHRD_ETHER:
491 if (mac_proto != MAC_PROTO_ETHERNET)
492 goto drop;
493 break;
494 default:
495 goto drop;
496 }
497
498 if (unlikely(packet_length(skb, vport->dev) > mtu &&
499 !skb_is_gso(skb))) {
500 vport->dev->stats.tx_errors++;
501 if (vport->dev->flags & IFF_UP)
502 net_warn_ratelimited("%s: dropped over-mtu packet: "
503 "%d > %d\n", vport->dev->name,
504 packet_length(skb, vport->dev),
505 mtu);
506 goto drop;
507 }
508
509 skb->dev = vport->dev;
510 skb->tstamp = 0;
511 vport->ops->send(skb);
512 return;
513
514drop:
515 kfree_skb(skb);
516}
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
36static 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. */
41static 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. */
54static 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 */
62int 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 */
77void ovs_vport_exit(void)
78{
79 kfree(dev_table);
80}
81
82static 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 */
95struct 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 */
119struct 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
125 alloc_size = sizeof(struct vport);
126 if (priv_size) {
127 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
128 alloc_size += priv_size;
129 }
130
131 vport = kzalloc(alloc_size, GFP_KERNEL);
132 if (!vport)
133 return ERR_PTR(-ENOMEM);
134
135 vport->dp = parms->dp;
136 vport->port_no = parms->port_no;
137 vport->upcall_portid = parms->upcall_portid;
138 vport->ops = ops;
139 INIT_HLIST_NODE(&vport->dp_hash_node);
140
141 vport->percpu_stats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
142 if (!vport->percpu_stats) {
143 kfree(vport);
144 return ERR_PTR(-ENOMEM);
145 }
146
147 spin_lock_init(&vport->stats_lock);
148
149 return vport;
150}
151
152/**
153 * ovs_vport_free - uninitialize and free vport
154 *
155 * @vport: vport to free
156 *
157 * Frees a vport allocated with vport_alloc() when it is no longer needed.
158 *
159 * The caller must ensure that an RCU grace period has passed since the last
160 * time @vport was in a datapath.
161 */
162void ovs_vport_free(struct vport *vport)
163{
164 free_percpu(vport->percpu_stats);
165 kfree(vport);
166}
167
168/**
169 * ovs_vport_add - add vport device (for kernel callers)
170 *
171 * @parms: Information about new vport.
172 *
173 * Creates a new vport with the specified configuration (which is dependent on
174 * device type). ovs_mutex must be held.
175 */
176struct vport *ovs_vport_add(const struct vport_parms *parms)
177{
178 struct vport *vport;
179 int err = 0;
180 int i;
181
182 for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
183 if (vport_ops_list[i]->type == parms->type) {
184 struct hlist_head *bucket;
185
186 vport = vport_ops_list[i]->create(parms);
187 if (IS_ERR(vport)) {
188 err = PTR_ERR(vport);
189 goto out;
190 }
191
192 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
193 vport->ops->get_name(vport));
194 hlist_add_head_rcu(&vport->hash_node, bucket);
195 return vport;
196 }
197 }
198
199 err = -EAFNOSUPPORT;
200
201out:
202 return ERR_PTR(err);
203}
204
205/**
206 * ovs_vport_set_options - modify existing vport device (for kernel callers)
207 *
208 * @vport: vport to modify.
209 * @options: New configuration.
210 *
211 * Modifies an existing device with the specified configuration (which is
212 * dependent on device type). ovs_mutex must be held.
213 */
214int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
215{
216 if (!vport->ops->set_options)
217 return -EOPNOTSUPP;
218 return vport->ops->set_options(vport, options);
219}
220
221/**
222 * ovs_vport_del - delete existing vport device
223 *
224 * @vport: vport to delete.
225 *
226 * Detaches @vport from its datapath and destroys it. It is possible to fail
227 * for reasons such as lack of memory. ovs_mutex must be held.
228 */
229void ovs_vport_del(struct vport *vport)
230{
231 ASSERT_OVSL();
232
233 hlist_del_rcu(&vport->hash_node);
234
235 vport->ops->destroy(vport);
236}
237
238/**
239 * ovs_vport_get_stats - retrieve device stats
240 *
241 * @vport: vport from which to retrieve the stats
242 * @stats: location to store stats
243 *
244 * Retrieves transmit, receive, and error stats for the given device.
245 *
246 * Must be called with ovs_mutex or rcu_read_lock.
247 */
248void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
249{
250 int i;
251
252 memset(stats, 0, sizeof(*stats));
253
254 /* We potentially have 2 sources of stats that need to be combined:
255 * those we have collected (split into err_stats and percpu_stats) from
256 * set_stats() and device error stats from netdev->get_stats() (for
257 * errors that happen downstream and therefore aren't reported through
258 * our vport_record_error() function).
259 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
260 * netdev-stats can be directly read over netlink-ioctl.
261 */
262
263 spin_lock_bh(&vport->stats_lock);
264
265 stats->rx_errors = vport->err_stats.rx_errors;
266 stats->tx_errors = vport->err_stats.tx_errors;
267 stats->tx_dropped = vport->err_stats.tx_dropped;
268 stats->rx_dropped = vport->err_stats.rx_dropped;
269
270 spin_unlock_bh(&vport->stats_lock);
271
272 for_each_possible_cpu(i) {
273 const struct pcpu_sw_netstats *percpu_stats;
274 struct pcpu_sw_netstats local_stats;
275 unsigned int start;
276
277 percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
278
279 do {
280 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
281 local_stats = *percpu_stats;
282 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
283
284 stats->rx_bytes += local_stats.rx_bytes;
285 stats->rx_packets += local_stats.rx_packets;
286 stats->tx_bytes += local_stats.tx_bytes;
287 stats->tx_packets += local_stats.tx_packets;
288 }
289}
290
291/**
292 * ovs_vport_get_options - retrieve device options
293 *
294 * @vport: vport from which to retrieve the options.
295 * @skb: sk_buff where options should be appended.
296 *
297 * Retrieves the configuration of the given device, appending an
298 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
299 * vport-specific attributes to @skb.
300 *
301 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
302 * negative error code if a real error occurred. If an error occurs, @skb is
303 * left unmodified.
304 *
305 * Must be called with ovs_mutex or rcu_read_lock.
306 */
307int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
308{
309 struct nlattr *nla;
310 int err;
311
312 if (!vport->ops->get_options)
313 return 0;
314
315 nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
316 if (!nla)
317 return -EMSGSIZE;
318
319 err = vport->ops->get_options(vport, skb);
320 if (err) {
321 nla_nest_cancel(skb, nla);
322 return err;
323 }
324
325 nla_nest_end(skb, nla);
326 return 0;
327}
328
329/**
330 * ovs_vport_receive - pass up received packet to the datapath for processing
331 *
332 * @vport: vport that received the packet
333 * @skb: skb that was received
334 * @tun_key: tunnel (if any) that carried packet
335 *
336 * Must be called with rcu_read_lock. The packet cannot be shared and
337 * skb->data should point to the Ethernet header.
338 */
339void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
340 struct ovs_key_ipv4_tunnel *tun_key)
341{
342 struct pcpu_sw_netstats *stats;
343
344 stats = this_cpu_ptr(vport->percpu_stats);
345 u64_stats_update_begin(&stats->syncp);
346 stats->rx_packets++;
347 stats->rx_bytes += skb->len;
348 u64_stats_update_end(&stats->syncp);
349
350 OVS_CB(skb)->tun_key = tun_key;
351 ovs_dp_process_received_packet(vport, skb);
352}
353
354/**
355 * ovs_vport_send - send a packet on a device
356 *
357 * @vport: vport on which to send the packet
358 * @skb: skb to send
359 *
360 * Sends the given packet and returns the length of data sent. Either ovs
361 * lock or rcu_read_lock must be held.
362 */
363int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
364{
365 int sent = vport->ops->send(vport, skb);
366
367 if (likely(sent > 0)) {
368 struct pcpu_sw_netstats *stats;
369
370 stats = this_cpu_ptr(vport->percpu_stats);
371
372 u64_stats_update_begin(&stats->syncp);
373 stats->tx_packets++;
374 stats->tx_bytes += sent;
375 u64_stats_update_end(&stats->syncp);
376 } else if (sent < 0) {
377 ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
378 kfree_skb(skb);
379 } else
380 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
381
382 return sent;
383}
384
385/**
386 * ovs_vport_record_error - indicate device error to generic stats layer
387 *
388 * @vport: vport that encountered the error
389 * @err_type: one of enum vport_err_type types to indicate the error type
390 *
391 * If using the vport generic stats layer indicate that an error of the given
392 * type has occurred.
393 */
394static void ovs_vport_record_error(struct vport *vport,
395 enum vport_err_type err_type)
396{
397 spin_lock(&vport->stats_lock);
398
399 switch (err_type) {
400 case VPORT_E_RX_DROPPED:
401 vport->err_stats.rx_dropped++;
402 break;
403
404 case VPORT_E_RX_ERROR:
405 vport->err_stats.rx_errors++;
406 break;
407
408 case VPORT_E_TX_DROPPED:
409 vport->err_stats.tx_dropped++;
410 break;
411
412 case VPORT_E_TX_ERROR:
413 vport->err_stats.tx_errors++;
414 break;
415 }
416
417 spin_unlock(&vport->stats_lock);
418}
419
420static void free_vport_rcu(struct rcu_head *rcu)
421{
422 struct vport *vport = container_of(rcu, struct vport, rcu);
423
424 ovs_vport_free(vport);
425}
426
427void ovs_vport_deferred_free(struct vport *vport)
428{
429 if (!vport)
430 return;
431
432 call_rcu(&vport->rcu, free_vport_rcu);
433}