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 | |
24 | static LIST_HEAD(vport_ops_list); |
25 | |
26 | /* Protected by RCU read lock for reading, ovs_mutex for writing. */ |
27 | static 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 | */ |
35 | int ovs_vport_init(void) |
36 | { |
37 | dev_table = kcalloc(VPORT_HASH_BUCKETS, size: 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 | */ |
50 | void ovs_vport_exit(void) |
51 | { |
52 | kfree(objp: dev_table); |
53 | } |
54 | |
55 | static struct hlist_head *hash_bucket(const struct net *net, const char *name) |
56 | { |
57 | unsigned int hash = jhash(key: name, strlen(name), initval: (unsigned long) net); |
58 | return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)]; |
59 | } |
60 | |
61 | int __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(new: &ops->list, head: &vport_ops_list); |
72 | err = 0; |
73 | errout: |
74 | ovs_unlock(); |
75 | return err; |
76 | } |
77 | EXPORT_SYMBOL_GPL(__ovs_vport_ops_register); |
78 | |
79 | void ovs_vport_ops_unregister(struct vport_ops *ops) |
80 | { |
81 | ovs_lock(); |
82 | list_del(entry: &ops->list); |
83 | ovs_unlock(); |
84 | } |
85 | EXPORT_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 | */ |
95 | struct 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(net1: ovs_dp_get_net(dp: vport->dp), net2: 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 | */ |
122 | struct 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(size: alloc_size, GFP_KERNEL); |
136 | if (!vport) |
137 | return ERR_PTR(error: -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(h: &vport->dp_hash_node); |
149 | |
150 | if (ovs_vport_set_upcall_portids(vport, pids: parms->upcall_portids)) { |
151 | err = -EINVAL; |
152 | goto err_free_percpu; |
153 | } |
154 | |
155 | return vport; |
156 | |
157 | err_free_percpu: |
158 | free_percpu(pdata: vport->upcall_stats); |
159 | err_kfree_vport: |
160 | kfree(objp: vport); |
161 | return ERR_PTR(error: err); |
162 | } |
163 | EXPORT_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 | */ |
175 | void 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(pdata: vport->upcall_stats); |
182 | kfree(objp: vport); |
183 | } |
184 | EXPORT_SYMBOL_GPL(ovs_vport_free); |
185 | |
186 | static 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 | */ |
205 | struct 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(module: ops->owner)) |
215 | return ERR_PTR(error: -EAFNOSUPPORT); |
216 | |
217 | vport = ops->create(parms); |
218 | if (IS_ERR(ptr: vport)) { |
219 | module_put(module: ops->owner); |
220 | return vport; |
221 | } |
222 | |
223 | bucket = hash_bucket(net: ovs_dp_get_net(dp: vport->dp), |
224 | name: ovs_vport_name(vport)); |
225 | hlist_add_head_rcu(n: &vport->hash_node, h: 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(error: -EAFNOSUPPORT); |
239 | else |
240 | return ERR_PTR(error: -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 | */ |
252 | int 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 | */ |
267 | void ovs_vport_del(struct vport *vport) |
268 | { |
269 | hlist_del_rcu(n: &vport->hash_node); |
270 | module_put(module: 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 | */ |
284 | void 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(dev: vport->dev, storage: &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 | */ |
311 | int ovs_vport_get_upcall_stats(struct vport *vport, struct sk_buff *skb) |
312 | { |
313 | struct nlattr *nla; |
314 | int i; |
315 | |
316 | __u64 tx_success = 0; |
317 | __u64 tx_fail = 0; |
318 | |
319 | for_each_possible_cpu(i) { |
320 | const struct vport_upcall_stats_percpu *stats; |
321 | unsigned int start; |
322 | |
323 | stats = per_cpu_ptr(vport->upcall_stats, i); |
324 | do { |
325 | start = u64_stats_fetch_begin(syncp: &stats->syncp); |
326 | tx_success += u64_stats_read(p: &stats->n_success); |
327 | tx_fail += u64_stats_read(p: &stats->n_fail); |
328 | } while (u64_stats_fetch_retry(syncp: &stats->syncp, start)); |
329 | } |
330 | |
331 | nla = nla_nest_start_noflag(skb, attrtype: OVS_VPORT_ATTR_UPCALL_STATS); |
332 | if (!nla) |
333 | return -EMSGSIZE; |
334 | |
335 | if (nla_put_u64_64bit(skb, attrtype: OVS_VPORT_UPCALL_ATTR_SUCCESS, value: tx_success, |
336 | padattr: OVS_VPORT_ATTR_PAD)) { |
337 | nla_nest_cancel(skb, start: nla); |
338 | return -EMSGSIZE; |
339 | } |
340 | |
341 | if (nla_put_u64_64bit(skb, attrtype: OVS_VPORT_UPCALL_ATTR_FAIL, value: tx_fail, |
342 | padattr: OVS_VPORT_ATTR_PAD)) { |
343 | nla_nest_cancel(skb, start: nla); |
344 | return -EMSGSIZE; |
345 | } |
346 | nla_nest_end(skb, start: 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 | */ |
367 | int 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, attrtype: 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, start: nla); |
382 | return err; |
383 | } |
384 | |
385 | nla_nest_end(skb, start: 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 | */ |
402 | int 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(nla: ids) || nla_len(nla: ids) % sizeof(u32)) |
407 | return -EINVAL; |
408 | |
409 | old = ovsl_dereference(vport->upcall_portids); |
410 | |
411 | vport_portids = kmalloc(size: sizeof(*vport_portids) + nla_len(nla: ids), |
412 | GFP_KERNEL); |
413 | if (!vport_portids) |
414 | return -ENOMEM; |
415 | |
416 | vport_portids->n_ids = nla_len(nla: ids) / sizeof(u32); |
417 | vport_portids->rn_ids = reciprocal_value(d: vport_portids->n_ids); |
418 | nla_memcpy(dest: vport_portids->ids, src: ids, count: nla_len(nla: ids)); |
419 | |
420 | rcu_assign_pointer(vport->upcall_portids, vport_portids); |
421 | |
422 | if (old) |
423 | kfree_rcu(old, rcu); |
424 | return 0; |
425 | } |
426 | |
427 | /** |
428 | * ovs_vport_get_upcall_portids - get the upcall_portids of @vport. |
429 | * |
430 | * @vport: vport from which to retrieve the portids. |
431 | * @skb: sk_buff where portids should be appended. |
432 | * |
433 | * Retrieves the configuration of the given vport, appending the |
434 | * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall |
435 | * portids to @skb. |
436 | * |
437 | * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room. |
438 | * If an error occurs, @skb is left unmodified. Must be called with |
439 | * ovs_mutex or rcu_read_lock. |
440 | */ |
441 | int ovs_vport_get_upcall_portids(const struct vport *vport, |
442 | struct sk_buff *skb) |
443 | { |
444 | struct vport_portids *ids; |
445 | |
446 | ids = rcu_dereference_ovsl(vport->upcall_portids); |
447 | |
448 | if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS) |
449 | return nla_put(skb, attrtype: OVS_VPORT_ATTR_UPCALL_PID, |
450 | attrlen: ids->n_ids * sizeof(u32), data: (void *)ids->ids); |
451 | else |
452 | return nla_put_u32(skb, attrtype: OVS_VPORT_ATTR_UPCALL_PID, value: ids->ids[0]); |
453 | } |
454 | |
455 | /** |
456 | * ovs_vport_find_upcall_portid - find the upcall portid to send upcall. |
457 | * |
458 | * @vport: vport from which the missed packet is received. |
459 | * @skb: skb that the missed packet was received. |
460 | * |
461 | * Uses the skb_get_hash() to select the upcall portid to send the |
462 | * upcall. |
463 | * |
464 | * Returns the portid of the target socket. Must be called with rcu_read_lock. |
465 | */ |
466 | u32 ovs_vport_find_upcall_portid(const struct vport *vport, |
467 | struct sk_buff *skb) |
468 | { |
469 | struct vport_portids *ids; |
470 | u32 ids_index; |
471 | u32 hash; |
472 | |
473 | ids = rcu_dereference(vport->upcall_portids); |
474 | |
475 | /* If there is only one portid, select it in the fast-path. */ |
476 | if (ids->n_ids == 1) |
477 | return ids->ids[0]; |
478 | |
479 | hash = skb_get_hash(skb); |
480 | ids_index = hash - ids->n_ids * reciprocal_divide(a: hash, R: ids->rn_ids); |
481 | return ids->ids[ids_index]; |
482 | } |
483 | |
484 | /** |
485 | * ovs_vport_receive - pass up received packet to the datapath for processing |
486 | * |
487 | * @vport: vport that received the packet |
488 | * @skb: skb that was received |
489 | * @tun_info: tunnel (if any) that carried packet |
490 | * |
491 | * Must be called with rcu_read_lock. The packet cannot be shared and |
492 | * skb->data should point to the Ethernet header. |
493 | */ |
494 | int ovs_vport_receive(struct vport *vport, struct sk_buff *skb, |
495 | const struct ip_tunnel_info *tun_info) |
496 | { |
497 | struct sw_flow_key key; |
498 | int error; |
499 | |
500 | OVS_CB(skb)->input_vport = vport; |
501 | OVS_CB(skb)->mru = 0; |
502 | OVS_CB(skb)->cutlen = 0; |
503 | if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) { |
504 | u32 mark; |
505 | |
506 | mark = skb->mark; |
507 | skb_scrub_packet(skb, xnet: true); |
508 | skb->mark = mark; |
509 | tun_info = NULL; |
510 | } |
511 | |
512 | /* Extract flow from 'skb' into 'key'. */ |
513 | error = ovs_flow_key_extract(tun_info, skb, key: &key); |
514 | if (unlikely(error)) { |
515 | kfree_skb(skb); |
516 | return error; |
517 | } |
518 | ovs_dp_process_packet(skb, key: &key); |
519 | return 0; |
520 | } |
521 | |
522 | static int packet_length(const struct sk_buff *skb, |
523 | struct net_device *dev) |
524 | { |
525 | int length = skb->len - dev->hard_header_len; |
526 | |
527 | if (!skb_vlan_tag_present(skb) && |
528 | eth_type_vlan(ethertype: skb->protocol)) |
529 | length -= VLAN_HLEN; |
530 | |
531 | /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow |
532 | * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none |
533 | * account for 802.1ad. e.g. is_skb_forwardable(). |
534 | */ |
535 | |
536 | return length > 0 ? length : 0; |
537 | } |
538 | |
539 | void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto) |
540 | { |
541 | int mtu = vport->dev->mtu; |
542 | |
543 | switch (vport->dev->type) { |
544 | case ARPHRD_NONE: |
545 | if (mac_proto == MAC_PROTO_ETHERNET) { |
546 | skb_reset_network_header(skb); |
547 | skb_reset_mac_len(skb); |
548 | skb->protocol = htons(ETH_P_TEB); |
549 | } else if (mac_proto != MAC_PROTO_NONE) { |
550 | WARN_ON_ONCE(1); |
551 | goto drop; |
552 | } |
553 | break; |
554 | case ARPHRD_ETHER: |
555 | if (mac_proto != MAC_PROTO_ETHERNET) |
556 | goto drop; |
557 | break; |
558 | default: |
559 | goto drop; |
560 | } |
561 | |
562 | if (unlikely(packet_length(skb, vport->dev) > mtu && |
563 | !skb_is_gso(skb))) { |
564 | vport->dev->stats.tx_errors++; |
565 | if (vport->dev->flags & IFF_UP) |
566 | net_warn_ratelimited("%s: dropped over-mtu packet: " |
567 | "%d > %d\n" , vport->dev->name, |
568 | packet_length(skb, vport->dev), |
569 | mtu); |
570 | goto drop; |
571 | } |
572 | |
573 | skb->dev = vport->dev; |
574 | skb_clear_tstamp(skb); |
575 | vport->ops->send(skb); |
576 | return; |
577 | |
578 | drop: |
579 | kfree_skb(skb); |
580 | } |
581 | |