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 <[email protected]> 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 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 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_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 .ops = psample_nl_ops, 116 .n_ops = ARRAY_SIZE(psample_nl_ops), 117 .n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps), 118 }; 119 120 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 } 137 138 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 } 154 155 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 * 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 } 172 173 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); 192 193 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); 200 201 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 static int __psample_ip_tun_to_nlattr(struct sk_buff *skb, 213 struct ip_tunnel_info *tun_info) 214 { 215 unsigned short tun_proto = ip_tunnel_info_af(tun_info); 216 const void *tun_opts = ip_tunnel_info_opts(tun_info); 217 const struct ip_tunnel_key *tun_key = &tun_info->key; 218 int tun_opts_len = tun_info->options_len; 219 220 if (tun_key->tun_flags & TUNNEL_KEY && 221 nla_put_be64(skb, PSAMPLE_TUNNEL_KEY_ATTR_ID, tun_key->tun_id, 222 PSAMPLE_TUNNEL_KEY_ATTR_PAD)) 223 return -EMSGSIZE; 224 225 if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE && 226 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE)) 227 return -EMSGSIZE; 228 229 switch (tun_proto) { 230 case AF_INET: 231 if (tun_key->u.ipv4.src && 232 nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_SRC, 233 tun_key->u.ipv4.src)) 234 return -EMSGSIZE; 235 if (tun_key->u.ipv4.dst && 236 nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_DST, 237 tun_key->u.ipv4.dst)) 238 return -EMSGSIZE; 239 break; 240 case AF_INET6: 241 if (!ipv6_addr_any(&tun_key->u.ipv6.src) && 242 nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_SRC, 243 &tun_key->u.ipv6.src)) 244 return -EMSGSIZE; 245 if (!ipv6_addr_any(&tun_key->u.ipv6.dst) && 246 nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_DST, 247 &tun_key->u.ipv6.dst)) 248 return -EMSGSIZE; 249 break; 250 } 251 if (tun_key->tos && 252 nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TOS, tun_key->tos)) 253 return -EMSGSIZE; 254 if (nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TTL, tun_key->ttl)) 255 return -EMSGSIZE; 256 if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) && 257 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_DONT_FRAGMENT)) 258 return -EMSGSIZE; 259 if ((tun_key->tun_flags & TUNNEL_CSUM) && 260 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_CSUM)) 261 return -EMSGSIZE; 262 if (tun_key->tp_src && 263 nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src)) 264 return -EMSGSIZE; 265 if (tun_key->tp_dst && 266 nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst)) 267 return -EMSGSIZE; 268 if ((tun_key->tun_flags & TUNNEL_OAM) && 269 nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_OAM)) 270 return -EMSGSIZE; 271 if (tun_opts_len) { 272 if (tun_key->tun_flags & TUNNEL_GENEVE_OPT && 273 nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_GENEVE_OPTS, 274 tun_opts_len, tun_opts)) 275 return -EMSGSIZE; 276 else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT && 277 nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_ERSPAN_OPTS, 278 tun_opts_len, tun_opts)) 279 return -EMSGSIZE; 280 } 281 282 return 0; 283 } 284 285 static int psample_ip_tun_to_nlattr(struct sk_buff *skb, 286 struct ip_tunnel_info *tun_info) 287 { 288 struct nlattr *nla; 289 int err; 290 291 nla = nla_nest_start_noflag(skb, PSAMPLE_ATTR_TUNNEL); 292 if (!nla) 293 return -EMSGSIZE; 294 295 err = __psample_ip_tun_to_nlattr(skb, tun_info); 296 if (err) { 297 nla_nest_cancel(skb, nla); 298 return err; 299 } 300 301 nla_nest_end(skb, nla); 302 303 return 0; 304 } 305 306 static int psample_tunnel_meta_len(struct ip_tunnel_info *tun_info) 307 { 308 unsigned short tun_proto = ip_tunnel_info_af(tun_info); 309 const struct ip_tunnel_key *tun_key = &tun_info->key; 310 int tun_opts_len = tun_info->options_len; 311 int sum = 0; 312 313 if (tun_key->tun_flags & TUNNEL_KEY) 314 sum += nla_total_size(sizeof(u64)); 315 316 if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE) 317 sum += nla_total_size(0); 318 319 switch (tun_proto) { 320 case AF_INET: 321 if (tun_key->u.ipv4.src) 322 sum += nla_total_size(sizeof(u32)); 323 if (tun_key->u.ipv4.dst) 324 sum += nla_total_size(sizeof(u32)); 325 break; 326 case AF_INET6: 327 if (!ipv6_addr_any(&tun_key->u.ipv6.src)) 328 sum += nla_total_size(sizeof(struct in6_addr)); 329 if (!ipv6_addr_any(&tun_key->u.ipv6.dst)) 330 sum += nla_total_size(sizeof(struct in6_addr)); 331 break; 332 } 333 if (tun_key->tos) 334 sum += nla_total_size(sizeof(u8)); 335 sum += nla_total_size(sizeof(u8)); /* TTL */ 336 if (tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) 337 sum += nla_total_size(0); 338 if (tun_key->tun_flags & TUNNEL_CSUM) 339 sum += nla_total_size(0); 340 if (tun_key->tp_src) 341 sum += nla_total_size(sizeof(u16)); 342 if (tun_key->tp_dst) 343 sum += nla_total_size(sizeof(u16)); 344 if (tun_key->tun_flags & TUNNEL_OAM) 345 sum += nla_total_size(0); 346 if (tun_opts_len) { 347 if (tun_key->tun_flags & TUNNEL_GENEVE_OPT) 348 sum += nla_total_size(tun_opts_len); 349 else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT) 350 sum += nla_total_size(tun_opts_len); 351 } 352 353 return sum; 354 } 355 356 void psample_sample_packet(struct psample_group *group, struct sk_buff *skb, 357 u32 trunc_size, int in_ifindex, int out_ifindex, 358 u32 sample_rate) 359 { 360 struct ip_tunnel_info *tun_info; 361 struct sk_buff *nl_skb; 362 int data_len; 363 int meta_len; 364 void *data; 365 int ret; 366 367 meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) + 368 (out_ifindex ? nla_total_size(sizeof(u16)) : 0) + 369 nla_total_size(sizeof(u32)) + /* sample_rate */ 370 nla_total_size(sizeof(u32)) + /* orig_size */ 371 nla_total_size(sizeof(u32)) + /* group_num */ 372 nla_total_size(sizeof(u32)); /* seq */ 373 374 tun_info = skb_tunnel_info(skb); 375 if (tun_info) 376 meta_len += psample_tunnel_meta_len(tun_info); 377 378 data_len = min(skb->len, trunc_size); 379 if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE) 380 data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN 381 - NLA_ALIGNTO; 382 383 nl_skb = genlmsg_new(meta_len + nla_total_size(data_len), GFP_ATOMIC); 384 if (unlikely(!nl_skb)) 385 return; 386 387 data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0, 388 PSAMPLE_CMD_SAMPLE); 389 if (unlikely(!data)) 390 goto error; 391 392 if (in_ifindex) { 393 ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex); 394 if (unlikely(ret < 0)) 395 goto error; 396 } 397 398 if (out_ifindex) { 399 ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex); 400 if (unlikely(ret < 0)) 401 goto error; 402 } 403 404 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate); 405 if (unlikely(ret < 0)) 406 goto error; 407 408 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len); 409 if (unlikely(ret < 0)) 410 goto error; 411 412 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num); 413 if (unlikely(ret < 0)) 414 goto error; 415 416 ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++); 417 if (unlikely(ret < 0)) 418 goto error; 419 420 if (data_len) { 421 int nla_len = nla_total_size(data_len); 422 struct nlattr *nla; 423 424 nla = skb_put(nl_skb, nla_len); 425 nla->nla_type = PSAMPLE_ATTR_DATA; 426 nla->nla_len = nla_attr_size(data_len); 427 428 if (skb_copy_bits(skb, 0, nla_data(nla), data_len)) 429 goto error; 430 } 431 432 if (tun_info) { 433 ret = psample_ip_tun_to_nlattr(nl_skb, tun_info); 434 if (unlikely(ret < 0)) 435 goto error; 436 } 437 438 genlmsg_end(nl_skb, data); 439 genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0, 440 PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC); 441 442 return; 443 error: 444 pr_err_ratelimited("Could not create psample log message\n"); 445 nlmsg_free(nl_skb); 446 } 447 EXPORT_SYMBOL_GPL(psample_sample_packet); 448 449 static int __init psample_module_init(void) 450 { 451 return genl_register_family(&psample_nl_family); 452 } 453 454 static void __exit psample_module_exit(void) 455 { 456 genl_unregister_family(&psample_nl_family); 457 } 458 459 module_init(psample_module_init); 460 module_exit(psample_module_exit); 461 462 MODULE_AUTHOR("Yotam Gigi <[email protected]>"); 463 MODULE_DESCRIPTION("netlink channel for packet sampling"); 464 MODULE_LICENSE("GPL v2"); 465