1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * xfrm_device.c - IPsec device offloading code. 4 * 5 * Copyright (c) 2015 secunet Security Networks AG 6 * 7 * Author: 8 * Steffen Klassert <steffen.klassert@secunet.com> 9 */ 10 11 #include <linux/errno.h> 12 #include <linux/module.h> 13 #include <linux/netdevice.h> 14 #include <linux/skbuff.h> 15 #include <linux/slab.h> 16 #include <linux/spinlock.h> 17 #include <net/dst.h> 18 #include <net/gso.h> 19 #include <net/xfrm.h> 20 #include <linux/notifier.h> 21 22 #ifdef CONFIG_XFRM_OFFLOAD 23 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb, 24 unsigned int hsize) 25 { 26 struct xfrm_offload *xo = xfrm_offload(skb); 27 28 skb_reset_mac_len(skb); 29 if (xo->flags & XFRM_GSO_SEGMENT) 30 skb->transport_header -= x->props.header_len; 31 32 pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len); 33 } 34 35 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb, 36 unsigned int hsize) 37 38 { 39 struct xfrm_offload *xo = xfrm_offload(skb); 40 41 if (xo->flags & XFRM_GSO_SEGMENT) 42 skb->transport_header = skb->network_header + hsize; 43 44 skb_reset_mac_len(skb); 45 pskb_pull(skb, 46 skb->mac_len + x->props.header_len - x->props.enc_hdr_len); 47 } 48 49 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb, 50 unsigned int hsize) 51 { 52 struct xfrm_offload *xo = xfrm_offload(skb); 53 int phlen = 0; 54 55 if (xo->flags & XFRM_GSO_SEGMENT) 56 skb->transport_header = skb->network_header + hsize; 57 58 skb_reset_mac_len(skb); 59 if (x->sel.family != AF_INET6) { 60 phlen = IPV4_BEET_PHMAXLEN; 61 if (x->outer_mode.family == AF_INET6) 62 phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr); 63 } 64 65 pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen)); 66 } 67 68 /* Adjust pointers into the packet when IPsec is done at layer2 */ 69 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb) 70 { 71 switch (x->outer_mode.encap) { 72 case XFRM_MODE_IPTFS: 73 case XFRM_MODE_TUNNEL: 74 if (x->outer_mode.family == AF_INET) 75 return __xfrm_mode_tunnel_prep(x, skb, 76 sizeof(struct iphdr)); 77 if (x->outer_mode.family == AF_INET6) 78 return __xfrm_mode_tunnel_prep(x, skb, 79 sizeof(struct ipv6hdr)); 80 break; 81 case XFRM_MODE_TRANSPORT: 82 if (x->outer_mode.family == AF_INET) 83 return __xfrm_transport_prep(x, skb, 84 sizeof(struct iphdr)); 85 if (x->outer_mode.family == AF_INET6) 86 return __xfrm_transport_prep(x, skb, 87 sizeof(struct ipv6hdr)); 88 break; 89 case XFRM_MODE_BEET: 90 if (x->outer_mode.family == AF_INET) 91 return __xfrm_mode_beet_prep(x, skb, 92 sizeof(struct iphdr)); 93 if (x->outer_mode.family == AF_INET6) 94 return __xfrm_mode_beet_prep(x, skb, 95 sizeof(struct ipv6hdr)); 96 break; 97 case XFRM_MODE_ROUTEOPTIMIZATION: 98 case XFRM_MODE_IN_TRIGGER: 99 break; 100 } 101 } 102 103 static inline bool xmit_xfrm_check_overflow(struct sk_buff *skb) 104 { 105 struct xfrm_offload *xo = xfrm_offload(skb); 106 __u32 seq = xo->seq.low; 107 108 seq += skb_shinfo(skb)->gso_segs; 109 if (unlikely(seq < xo->seq.low)) 110 return true; 111 112 return false; 113 } 114 115 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again) 116 { 117 int err; 118 unsigned long flags; 119 struct xfrm_state *x; 120 struct softnet_data *sd; 121 struct sk_buff *skb2, *nskb, *pskb = NULL; 122 netdev_features_t esp_features = features; 123 struct xfrm_offload *xo = xfrm_offload(skb); 124 struct net_device *dev = skb->dev; 125 struct sec_path *sp; 126 127 if (!xo || (xo->flags & XFRM_XMIT)) 128 return skb; 129 130 if (!(features & NETIF_F_HW_ESP)) 131 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK); 132 133 sp = skb_sec_path(skb); 134 x = sp->xvec[sp->len - 1]; 135 if (xo->flags & XFRM_GRO || x->xso.dir == XFRM_DEV_OFFLOAD_IN) 136 return skb; 137 138 /* The packet was sent to HW IPsec packet offload engine, 139 * but to wrong device. Drop the packet, so it won't skip 140 * XFRM stack. 141 */ 142 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET && x->xso.dev != dev) { 143 kfree_skb(skb); 144 dev_core_stats_tx_dropped_inc(dev); 145 return NULL; 146 } 147 148 local_irq_save(flags); 149 sd = this_cpu_ptr(&softnet_data); 150 err = !skb_queue_empty(&sd->xfrm_backlog); 151 local_irq_restore(flags); 152 153 if (err) { 154 *again = true; 155 return skb; 156 } 157 158 if (skb_is_gso(skb) && unlikely(xmit_xfrm_check_overflow(skb))) { 159 struct sk_buff *segs; 160 161 /* Packet got rerouted, fixup features and segment it. */ 162 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP); 163 164 segs = skb_gso_segment(skb, esp_features); 165 if (IS_ERR(segs)) { 166 kfree_skb(skb); 167 dev_core_stats_tx_dropped_inc(dev); 168 return NULL; 169 } else { 170 consume_skb(skb); 171 skb = segs; 172 } 173 } 174 175 if (!skb->next) { 176 esp_features |= skb->dev->gso_partial_features; 177 xfrm_outer_mode_prep(x, skb); 178 179 xo->flags |= XFRM_DEV_RESUME; 180 181 err = x->type_offload->xmit(x, skb, esp_features); 182 if (err) { 183 if (err == -EINPROGRESS) 184 return NULL; 185 186 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 187 kfree_skb(skb); 188 return NULL; 189 } 190 191 skb_push(skb, skb->data - skb_mac_header(skb)); 192 193 return skb; 194 } 195 196 skb_list_walk_safe(skb, skb2, nskb) { 197 esp_features |= skb->dev->gso_partial_features; 198 skb_mark_not_on_list(skb2); 199 200 xo = xfrm_offload(skb2); 201 xo->flags |= XFRM_DEV_RESUME; 202 203 xfrm_outer_mode_prep(x, skb2); 204 205 err = x->type_offload->xmit(x, skb2, esp_features); 206 if (!err) { 207 skb2->next = nskb; 208 } else if (err != -EINPROGRESS) { 209 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 210 skb2->next = nskb; 211 kfree_skb_list(skb2); 212 return NULL; 213 } else { 214 if (skb == skb2) 215 skb = nskb; 216 else 217 pskb->next = nskb; 218 219 continue; 220 } 221 222 skb_push(skb2, skb2->data - skb_mac_header(skb2)); 223 pskb = skb2; 224 } 225 226 return skb; 227 } 228 EXPORT_SYMBOL_GPL(validate_xmit_xfrm); 229 230 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x, 231 struct xfrm_user_offload *xuo, 232 struct netlink_ext_ack *extack) 233 { 234 int err; 235 struct dst_entry *dst; 236 struct net_device *dev; 237 struct xfrm_dev_offload *xso = &x->xso; 238 xfrm_address_t *saddr; 239 xfrm_address_t *daddr; 240 bool is_packet_offload; 241 242 if (xuo->flags & 243 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) { 244 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 245 return -EINVAL; 246 } 247 248 if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) || 249 (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) { 250 NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction"); 251 return -EINVAL; 252 } 253 254 if (xuo->flags & XFRM_OFFLOAD_INBOUND && x->if_id) { 255 NL_SET_ERR_MSG(extack, "XFRM if_id is not supported in RX path"); 256 return -EINVAL; 257 } 258 259 is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET; 260 261 /* We don't yet support TFC padding. */ 262 if (x->tfcpad) { 263 NL_SET_ERR_MSG(extack, "TFC padding can't be offloaded"); 264 return -EINVAL; 265 } 266 267 dev = dev_get_by_index(net, xuo->ifindex); 268 if (!dev) { 269 struct xfrm_dst_lookup_params params; 270 271 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) { 272 saddr = &x->props.saddr; 273 daddr = &x->id.daddr; 274 } else { 275 saddr = &x->id.daddr; 276 daddr = &x->props.saddr; 277 } 278 279 memset(¶ms, 0, sizeof(params)); 280 params.net = net; 281 params.saddr = saddr; 282 params.daddr = daddr; 283 params.mark = xfrm_smark_get(0, x); 284 dst = __xfrm_dst_lookup(x->props.family, ¶ms); 285 if (IS_ERR(dst)) 286 return (is_packet_offload) ? -EINVAL : 0; 287 288 dev = dst->dev; 289 290 dev_hold(dev); 291 dst_release(dst); 292 } 293 294 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) { 295 xso->dev = NULL; 296 dev_put(dev); 297 return (is_packet_offload) ? -EINVAL : 0; 298 } 299 300 if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN && 301 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 302 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN"); 303 xso->dev = NULL; 304 dev_put(dev); 305 return -EINVAL; 306 } 307 308 if (!x->type_offload) { 309 NL_SET_ERR_MSG(extack, "Type doesn't support offload"); 310 dev_put(dev); 311 return -EINVAL; 312 } 313 314 xso->dev = dev; 315 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC); 316 317 if (xuo->flags & XFRM_OFFLOAD_INBOUND) 318 xso->dir = XFRM_DEV_OFFLOAD_IN; 319 else 320 xso->dir = XFRM_DEV_OFFLOAD_OUT; 321 322 if (is_packet_offload) 323 xso->type = XFRM_DEV_OFFLOAD_PACKET; 324 else 325 xso->type = XFRM_DEV_OFFLOAD_CRYPTO; 326 327 err = dev->xfrmdev_ops->xdo_dev_state_add(dev, x, extack); 328 if (err) { 329 xso->dev = NULL; 330 xso->dir = 0; 331 netdev_put(dev, &xso->dev_tracker); 332 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 333 334 xfrm_unset_type_offload(x); 335 /* User explicitly requested packet offload mode and configured 336 * policy in addition to the XFRM state. So be civil to users, 337 * and return an error instead of taking fallback path. 338 */ 339 if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) { 340 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state"); 341 return err; 342 } 343 } 344 345 return 0; 346 } 347 EXPORT_SYMBOL_GPL(xfrm_dev_state_add); 348 349 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp, 350 struct xfrm_user_offload *xuo, u8 dir, 351 struct netlink_ext_ack *extack) 352 { 353 struct xfrm_dev_offload *xdo = &xp->xdo; 354 struct net_device *dev; 355 int err; 356 357 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) { 358 /* We support only packet offload mode and it means 359 * that user must set XFRM_OFFLOAD_PACKET bit. 360 */ 361 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 362 return -EINVAL; 363 } 364 365 dev = dev_get_by_index(net, xuo->ifindex); 366 if (!dev) 367 return -EINVAL; 368 369 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) { 370 xdo->dev = NULL; 371 dev_put(dev); 372 NL_SET_ERR_MSG(extack, "Policy offload is not supported"); 373 return -EINVAL; 374 } 375 376 xdo->dev = dev; 377 netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC); 378 xdo->type = XFRM_DEV_OFFLOAD_PACKET; 379 switch (dir) { 380 case XFRM_POLICY_IN: 381 xdo->dir = XFRM_DEV_OFFLOAD_IN; 382 break; 383 case XFRM_POLICY_OUT: 384 xdo->dir = XFRM_DEV_OFFLOAD_OUT; 385 break; 386 case XFRM_POLICY_FWD: 387 xdo->dir = XFRM_DEV_OFFLOAD_FWD; 388 break; 389 default: 390 xdo->dev = NULL; 391 netdev_put(dev, &xdo->dev_tracker); 392 NL_SET_ERR_MSG(extack, "Unrecognized offload direction"); 393 return -EINVAL; 394 } 395 396 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack); 397 if (err) { 398 xdo->dev = NULL; 399 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 400 xdo->dir = 0; 401 netdev_put(dev, &xdo->dev_tracker); 402 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy"); 403 return err; 404 } 405 406 return 0; 407 } 408 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add); 409 410 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 411 { 412 int mtu; 413 struct dst_entry *dst = skb_dst(skb); 414 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 415 struct net_device *dev = x->xso.dev; 416 bool check_tunnel_size; 417 418 if (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED) 419 return false; 420 421 if ((dev == xfrm_dst_path(dst)->dev) && !xdst->child->xfrm) { 422 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 423 if (skb->len <= mtu) 424 goto ok; 425 426 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 427 goto ok; 428 } 429 430 return false; 431 432 ok: 433 check_tunnel_size = x->xso.type == XFRM_DEV_OFFLOAD_PACKET && 434 x->props.mode == XFRM_MODE_TUNNEL; 435 switch (x->props.family) { 436 case AF_INET: 437 /* Check for IPv4 options */ 438 if (ip_hdr(skb)->ihl != 5) 439 return false; 440 if (check_tunnel_size && xfrm4_tunnel_check_size(skb)) 441 return false; 442 break; 443 case AF_INET6: 444 /* Check for IPv6 extensions */ 445 if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr)) 446 return false; 447 if (check_tunnel_size && xfrm6_tunnel_check_size(skb)) 448 return false; 449 break; 450 default: 451 break; 452 } 453 454 if (dev->xfrmdev_ops->xdo_dev_offload_ok) 455 return dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 456 457 return true; 458 } 459 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 460 461 void xfrm_dev_resume(struct sk_buff *skb) 462 { 463 struct net_device *dev = skb->dev; 464 int ret = NETDEV_TX_BUSY; 465 struct netdev_queue *txq; 466 struct softnet_data *sd; 467 unsigned long flags; 468 469 rcu_read_lock(); 470 txq = netdev_core_pick_tx(dev, skb, NULL); 471 472 HARD_TX_LOCK(dev, txq, smp_processor_id()); 473 if (!netif_xmit_frozen_or_stopped(txq)) 474 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 475 HARD_TX_UNLOCK(dev, txq); 476 477 if (!dev_xmit_complete(ret)) { 478 local_irq_save(flags); 479 sd = this_cpu_ptr(&softnet_data); 480 skb_queue_tail(&sd->xfrm_backlog, skb); 481 raise_softirq_irqoff(NET_TX_SOFTIRQ); 482 local_irq_restore(flags); 483 } 484 rcu_read_unlock(); 485 } 486 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 487 488 void xfrm_dev_backlog(struct softnet_data *sd) 489 { 490 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 491 struct sk_buff_head list; 492 struct sk_buff *skb; 493 494 if (skb_queue_empty(xfrm_backlog)) 495 return; 496 497 __skb_queue_head_init(&list); 498 499 spin_lock(&xfrm_backlog->lock); 500 skb_queue_splice_init(xfrm_backlog, &list); 501 spin_unlock(&xfrm_backlog->lock); 502 503 while (!skb_queue_empty(&list)) { 504 skb = __skb_dequeue(&list); 505 xfrm_dev_resume(skb); 506 } 507 508 } 509 #endif 510 511 static int xfrm_api_check(struct net_device *dev) 512 { 513 #ifdef CONFIG_XFRM_OFFLOAD 514 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 515 !(dev->features & NETIF_F_HW_ESP)) 516 return NOTIFY_BAD; 517 518 if ((dev->features & NETIF_F_HW_ESP) && 519 (!(dev->xfrmdev_ops && 520 dev->xfrmdev_ops->xdo_dev_state_add && 521 dev->xfrmdev_ops->xdo_dev_state_delete))) 522 return NOTIFY_BAD; 523 #else 524 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 525 return NOTIFY_BAD; 526 #endif 527 528 return NOTIFY_DONE; 529 } 530 531 static int xfrm_dev_down(struct net_device *dev) 532 { 533 if (dev->features & NETIF_F_HW_ESP) { 534 xfrm_dev_state_flush(dev_net(dev), dev, true); 535 xfrm_dev_policy_flush(dev_net(dev), dev, true); 536 } 537 538 return NOTIFY_DONE; 539 } 540 541 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 542 { 543 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 544 545 switch (event) { 546 case NETDEV_REGISTER: 547 return xfrm_api_check(dev); 548 549 case NETDEV_FEAT_CHANGE: 550 return xfrm_api_check(dev); 551 552 case NETDEV_DOWN: 553 case NETDEV_UNREGISTER: 554 return xfrm_dev_down(dev); 555 } 556 return NOTIFY_DONE; 557 } 558 559 static struct notifier_block xfrm_dev_notifier = { 560 .notifier_call = xfrm_dev_event, 561 }; 562 563 void __init xfrm_dev_init(void) 564 { 565 register_netdevice_notifier(&xfrm_dev_notifier); 566 } 567