1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * net/psample/psample.c - Netlink channel for packet sampling
4 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
5 */
6
7 #include <linux/types.h>
8 #include <linux/kernel.h>
9 #include <linux/skbuff.h>
10 #include <linux/module.h>
11 #include <net/net_namespace.h>
12 #include <net/sock.h>
13 #include <net/netlink.h>
14 #include <net/genetlink.h>
15 #include <net/psample.h>
16 #include <linux/spinlock.h>
17 #include <net/ip_tunnels.h>
18 #include <net/dst_metadata.h>
19
20 #define PSAMPLE_MAX_PACKET_SIZE 0xffff
21
22 static LIST_HEAD(psample_groups_list);
23 static DEFINE_SPINLOCK(psample_groups_lock);
24
25 /* multicast groups */
26 enum psample_nl_multicast_groups {
27 PSAMPLE_NL_MCGRP_CONFIG,
28 PSAMPLE_NL_MCGRP_SAMPLE,
29 };
30
31 static const struct genl_multicast_group psample_nl_mcgrps[] = {
32 [PSAMPLE_NL_MCGRP_CONFIG] = { .name = PSAMPLE_NL_MCGRP_CONFIG_NAME },
33 [PSAMPLE_NL_MCGRP_SAMPLE] = { .name = PSAMPLE_NL_MCGRP_SAMPLE_NAME },
34 };
35
36 static struct genl_family psample_nl_family __ro_after_init;
37
psample_group_nl_fill(struct sk_buff * msg,struct psample_group * group,enum psample_command cmd,u32 portid,u32 seq,int flags)38 static int psample_group_nl_fill(struct sk_buff *msg,
39 struct psample_group *group,
40 enum psample_command cmd, u32 portid, u32 seq,
41 int flags)
42 {
43 void *hdr;
44 int ret;
45
46 hdr = genlmsg_put(msg, portid, seq, &psample_nl_family, flags, cmd);
47 if (!hdr)
48 return -EMSGSIZE;
49
50 ret = nla_put_u32(msg, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
51 if (ret < 0)
52 goto error;
53
54 ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_REFCOUNT, group->refcount);
55 if (ret < 0)
56 goto error;
57
58 ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_SEQ, group->seq);
59 if (ret < 0)
60 goto error;
61
62 genlmsg_end(msg, hdr);
63 return 0;
64
65 error:
66 genlmsg_cancel(msg, hdr);
67 return -EMSGSIZE;
68 }
69
psample_nl_cmd_get_group_dumpit(struct sk_buff * msg,struct netlink_callback * cb)70 static int psample_nl_cmd_get_group_dumpit(struct sk_buff *msg,
71 struct netlink_callback *cb)
72 {
73 struct psample_group *group;
74 int start = cb->args[0];
75 int idx = 0;
76 int err;
77
78 spin_lock_bh(&psample_groups_lock);
79 list_for_each_entry(group, &psample_groups_list, list) {
80 if (!net_eq(group->net, sock_net(msg->sk)))
81 continue;
82 if (idx < start) {
83 idx++;
84 continue;
85 }
86 err = psample_group_nl_fill(msg, group, PSAMPLE_CMD_NEW_GROUP,
87 NETLINK_CB(cb->skb).portid,
88 cb->nlh->nlmsg_seq, NLM_F_MULTI);
89 if (err)
90 break;
91 idx++;
92 }
93
94 spin_unlock_bh(&psample_groups_lock);
95 cb->args[0] = idx;
96 return msg->len;
97 }
98
99 static const struct genl_small_ops psample_nl_ops[] = {
100 {
101 .cmd = PSAMPLE_CMD_GET_GROUP,
102 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
103 .dumpit = psample_nl_cmd_get_group_dumpit,
104 /* can be retrieved by unprivileged users */
105 }
106 };
107
108 static struct genl_family psample_nl_family __ro_after_init = {
109 .name = PSAMPLE_GENL_NAME,
110 .version = PSAMPLE_GENL_VERSION,
111 .maxattr = PSAMPLE_ATTR_MAX,
112 .netnsok = true,
113 .module = THIS_MODULE,
114 .mcgrps = psample_nl_mcgrps,
115 .small_ops = psample_nl_ops,
116 .n_small_ops = ARRAY_SIZE(psample_nl_ops),
117 .n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps),
118 };
119
psample_group_notify(struct psample_group * group,enum psample_command cmd)120 static void psample_group_notify(struct psample_group *group,
121 enum psample_command cmd)
122 {
123 struct sk_buff *msg;
124 int err;
125
126 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
127 if (!msg)
128 return;
129
130 err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
131 if (!err)
132 genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
133 PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
134 else
135 nlmsg_free(msg);
136 }
137
psample_group_create(struct net * net,u32 group_num)138 static struct psample_group *psample_group_create(struct net *net,
139 u32 group_num)
140 {
141 struct psample_group *group;
142
143 group = kzalloc(sizeof(*group), GFP_ATOMIC);
144 if (!group)
145 return NULL;
146
147 group->net = net;
148 group->group_num = group_num;
149 list_add_tail(&group->list, &psample_groups_list);
150
151 psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
152 return group;
153 }
154
psample_group_destroy(struct psample_group * group)155 static void psample_group_destroy(struct psample_group *group)
156 {
157 psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
158 list_del(&group->list);
159 kfree_rcu(group, rcu);
160 }
161
162 static struct psample_group *
psample_group_lookup(struct net * net,u32 group_num)163 psample_group_lookup(struct net *net, u32 group_num)
164 {
165 struct psample_group *group;
166
167 list_for_each_entry(group, &psample_groups_list, list)
168 if ((group->group_num == group_num) && (group->net == net))
169 return group;
170 return NULL;
171 }
172
psample_group_get(struct net * net,u32 group_num)173 struct psample_group *psample_group_get(struct net *net, u32 group_num)
174 {
175 struct psample_group *group;
176
177 spin_lock_bh(&psample_groups_lock);
178
179 group = psample_group_lookup(net, group_num);
180 if (!group) {
181 group = psample_group_create(net, group_num);
182 if (!group)
183 goto out;
184 }
185 group->refcount++;
186
187 out:
188 spin_unlock_bh(&psample_groups_lock);
189 return group;
190 }
191 EXPORT_SYMBOL_GPL(psample_group_get);
192
psample_group_take(struct psample_group * group)193 void psample_group_take(struct psample_group *group)
194 {
195 spin_lock_bh(&psample_groups_lock);
196 group->refcount++;
197 spin_unlock_bh(&psample_groups_lock);
198 }
199 EXPORT_SYMBOL_GPL(psample_group_take);
200
psample_group_put(struct psample_group * group)201 void psample_group_put(struct psample_group *group)
202 {
203 spin_lock_bh(&psample_groups_lock);
204
205 if (--group->refcount == 0)
206 psample_group_destroy(group);
207
208 spin_unlock_bh(&psample_groups_lock);
209 }
210 EXPORT_SYMBOL_GPL(psample_group_put);
211
212 #ifdef CONFIG_INET
__psample_ip_tun_to_nlattr(struct sk_buff * skb,struct ip_tunnel_info * tun_info)213 static int __psample_ip_tun_to_nlattr(struct sk_buff *skb,
214 struct ip_tunnel_info *tun_info)
215 {
216 unsigned short tun_proto = ip_tunnel_info_af(tun_info);
217 const void *tun_opts = ip_tunnel_info_opts(tun_info);
218 const struct ip_tunnel_key *tun_key = &tun_info->key;
219 int tun_opts_len = tun_info->options_len;
220
221 if (tun_key->tun_flags & TUNNEL_KEY &&
222 nla_put_be64(skb, PSAMPLE_TUNNEL_KEY_ATTR_ID, tun_key->tun_id,
223 PSAMPLE_TUNNEL_KEY_ATTR_PAD))
224 return -EMSGSIZE;
225
226 if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE &&
227 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE))
228 return -EMSGSIZE;
229
230 switch (tun_proto) {
231 case AF_INET:
232 if (tun_key->u.ipv4.src &&
233 nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_SRC,
234 tun_key->u.ipv4.src))
235 return -EMSGSIZE;
236 if (tun_key->u.ipv4.dst &&
237 nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_DST,
238 tun_key->u.ipv4.dst))
239 return -EMSGSIZE;
240 break;
241 case AF_INET6:
242 if (!ipv6_addr_any(&tun_key->u.ipv6.src) &&
243 nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_SRC,
244 &tun_key->u.ipv6.src))
245 return -EMSGSIZE;
246 if (!ipv6_addr_any(&tun_key->u.ipv6.dst) &&
247 nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_DST,
248 &tun_key->u.ipv6.dst))
249 return -EMSGSIZE;
250 break;
251 }
252 if (tun_key->tos &&
253 nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TOS, tun_key->tos))
254 return -EMSGSIZE;
255 if (nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TTL, tun_key->ttl))
256 return -EMSGSIZE;
257 if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
258 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
259 return -EMSGSIZE;
260 if ((tun_key->tun_flags & TUNNEL_CSUM) &&
261 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_CSUM))
262 return -EMSGSIZE;
263 if (tun_key->tp_src &&
264 nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src))
265 return -EMSGSIZE;
266 if (tun_key->tp_dst &&
267 nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst))
268 return -EMSGSIZE;
269 if ((tun_key->tun_flags & TUNNEL_OAM) &&
270 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_OAM))
271 return -EMSGSIZE;
272 if (tun_opts_len) {
273 if (tun_key->tun_flags & TUNNEL_GENEVE_OPT &&
274 nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_GENEVE_OPTS,
275 tun_opts_len, tun_opts))
276 return -EMSGSIZE;
277 else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT &&
278 nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
279 tun_opts_len, tun_opts))
280 return -EMSGSIZE;
281 }
282
283 return 0;
284 }
285
psample_ip_tun_to_nlattr(struct sk_buff * skb,struct ip_tunnel_info * tun_info)286 static int psample_ip_tun_to_nlattr(struct sk_buff *skb,
287 struct ip_tunnel_info *tun_info)
288 {
289 struct nlattr *nla;
290 int err;
291
292 nla = nla_nest_start_noflag(skb, PSAMPLE_ATTR_TUNNEL);
293 if (!nla)
294 return -EMSGSIZE;
295
296 err = __psample_ip_tun_to_nlattr(skb, tun_info);
297 if (err) {
298 nla_nest_cancel(skb, nla);
299 return err;
300 }
301
302 nla_nest_end(skb, nla);
303
304 return 0;
305 }
306
psample_tunnel_meta_len(struct ip_tunnel_info * tun_info)307 static int psample_tunnel_meta_len(struct ip_tunnel_info *tun_info)
308 {
309 unsigned short tun_proto = ip_tunnel_info_af(tun_info);
310 const struct ip_tunnel_key *tun_key = &tun_info->key;
311 int tun_opts_len = tun_info->options_len;
312 int sum = nla_total_size(0); /* PSAMPLE_ATTR_TUNNEL */
313
314 if (tun_key->tun_flags & TUNNEL_KEY)
315 sum += nla_total_size_64bit(sizeof(u64));
316
317 if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE)
318 sum += nla_total_size(0);
319
320 switch (tun_proto) {
321 case AF_INET:
322 if (tun_key->u.ipv4.src)
323 sum += nla_total_size(sizeof(u32));
324 if (tun_key->u.ipv4.dst)
325 sum += nla_total_size(sizeof(u32));
326 break;
327 case AF_INET6:
328 if (!ipv6_addr_any(&tun_key->u.ipv6.src))
329 sum += nla_total_size(sizeof(struct in6_addr));
330 if (!ipv6_addr_any(&tun_key->u.ipv6.dst))
331 sum += nla_total_size(sizeof(struct in6_addr));
332 break;
333 }
334 if (tun_key->tos)
335 sum += nla_total_size(sizeof(u8));
336 sum += nla_total_size(sizeof(u8)); /* TTL */
337 if (tun_key->tun_flags & TUNNEL_DONT_FRAGMENT)
338 sum += nla_total_size(0);
339 if (tun_key->tun_flags & TUNNEL_CSUM)
340 sum += nla_total_size(0);
341 if (tun_key->tp_src)
342 sum += nla_total_size(sizeof(u16));
343 if (tun_key->tp_dst)
344 sum += nla_total_size(sizeof(u16));
345 if (tun_key->tun_flags & TUNNEL_OAM)
346 sum += nla_total_size(0);
347 if (tun_opts_len) {
348 if (tun_key->tun_flags & TUNNEL_GENEVE_OPT)
349 sum += nla_total_size(tun_opts_len);
350 else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT)
351 sum += nla_total_size(tun_opts_len);
352 }
353
354 return sum;
355 }
356 #endif
357
psample_sample_packet(struct psample_group * group,struct sk_buff * skb,u32 trunc_size,int in_ifindex,int out_ifindex,u32 sample_rate)358 void psample_sample_packet(struct psample_group *group, struct sk_buff *skb,
359 u32 trunc_size, int in_ifindex, int out_ifindex,
360 u32 sample_rate)
361 {
362 #ifdef CONFIG_INET
363 struct ip_tunnel_info *tun_info;
364 #endif
365 struct sk_buff *nl_skb;
366 int data_len;
367 int meta_len;
368 void *data;
369 int ret;
370
371 meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
372 (out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
373 nla_total_size(sizeof(u32)) + /* sample_rate */
374 nla_total_size(sizeof(u32)) + /* orig_size */
375 nla_total_size(sizeof(u32)) + /* group_num */
376 nla_total_size(sizeof(u32)); /* seq */
377
378 #ifdef CONFIG_INET
379 tun_info = skb_tunnel_info(skb);
380 if (tun_info)
381 meta_len += psample_tunnel_meta_len(tun_info);
382 #endif
383
384 data_len = min(skb->len, trunc_size);
385 if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
386 data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
387 - NLA_ALIGNTO;
388
389 nl_skb = genlmsg_new(meta_len + nla_total_size(data_len), GFP_ATOMIC);
390 if (unlikely(!nl_skb))
391 return;
392
393 data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
394 PSAMPLE_CMD_SAMPLE);
395 if (unlikely(!data))
396 goto error;
397
398 if (in_ifindex) {
399 ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
400 if (unlikely(ret < 0))
401 goto error;
402 }
403
404 if (out_ifindex) {
405 ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
406 if (unlikely(ret < 0))
407 goto error;
408 }
409
410 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
411 if (unlikely(ret < 0))
412 goto error;
413
414 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
415 if (unlikely(ret < 0))
416 goto error;
417
418 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
419 if (unlikely(ret < 0))
420 goto error;
421
422 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
423 if (unlikely(ret < 0))
424 goto error;
425
426 if (data_len) {
427 int nla_len = nla_total_size(data_len);
428 struct nlattr *nla;
429
430 nla = skb_put(nl_skb, nla_len);
431 nla->nla_type = PSAMPLE_ATTR_DATA;
432 nla->nla_len = nla_attr_size(data_len);
433
434 if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
435 goto error;
436 }
437
438 #ifdef CONFIG_INET
439 if (tun_info) {
440 ret = psample_ip_tun_to_nlattr(nl_skb, tun_info);
441 if (unlikely(ret < 0))
442 goto error;
443 }
444 #endif
445
446 genlmsg_end(nl_skb, data);
447 genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
448 PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);
449
450 return;
451 error:
452 pr_err_ratelimited("Could not create psample log message\n");
453 nlmsg_free(nl_skb);
454 }
455 EXPORT_SYMBOL_GPL(psample_sample_packet);
456
psample_module_init(void)457 static int __init psample_module_init(void)
458 {
459 return genl_register_family(&psample_nl_family);
460 }
461
psample_module_exit(void)462 static void __exit psample_module_exit(void)
463 {
464 genl_unregister_family(&psample_nl_family);
465 }
466
467 module_init(psample_module_init);
468 module_exit(psample_module_exit);
469
470 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
471 MODULE_DESCRIPTION("netlink channel for packet sampling");
472 MODULE_LICENSE("GPL v2");
473