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(x->xso.dev != dev) || 159 unlikely(xmit_xfrm_check_overflow(skb)))) { 160 struct sk_buff *segs; 161 162 /* Packet got rerouted, fixup features and segment it. */ 163 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP); 164 165 segs = skb_gso_segment(skb, esp_features); 166 if (IS_ERR(segs)) { 167 kfree_skb(skb); 168 dev_core_stats_tx_dropped_inc(dev); 169 return NULL; 170 } else { 171 consume_skb(skb); 172 skb = segs; 173 } 174 } 175 176 if (!skb->next) { 177 esp_features |= skb->dev->gso_partial_features; 178 xfrm_outer_mode_prep(x, skb); 179 180 xo->flags |= XFRM_DEV_RESUME; 181 182 err = x->type_offload->xmit(x, skb, esp_features); 183 if (err) { 184 if (err == -EINPROGRESS) 185 return ERR_PTR(-EINPROGRESS); 186 187 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 188 kfree_skb(skb); 189 return NULL; 190 } 191 192 skb_push(skb, skb->data - skb_mac_header(skb)); 193 194 return skb; 195 } 196 197 skb_list_walk_safe(skb, skb2, nskb) { 198 esp_features |= skb->dev->gso_partial_features; 199 skb_mark_not_on_list(skb2); 200 201 xo = xfrm_offload(skb2); 202 xo->flags |= XFRM_DEV_RESUME; 203 204 xfrm_outer_mode_prep(x, skb2); 205 206 err = x->type_offload->xmit(x, skb2, esp_features); 207 if (!err) { 208 skb2->next = nskb; 209 } else if (err != -EINPROGRESS) { 210 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 211 skb2->next = nskb; 212 kfree_skb_list(skb2); 213 return NULL; 214 } else { 215 if (skb == skb2) 216 skb = nskb; 217 else 218 pskb->next = nskb; 219 220 continue; 221 } 222 223 skb_push(skb2, skb2->data - skb_mac_header(skb2)); 224 pskb = skb2; 225 } 226 227 /* skb_gso_segment() set skb->prev to the last segment, but async 228 * crypto may have stolen it above without updating ->prev. Repoint 229 * it at the last retained segment so validate_xmit_skb_list() does 230 * not chain onto a segment now owned by the crypto engine. 231 */ 232 if (skb) 233 skb->prev = pskb; 234 235 return skb ? skb : ERR_PTR(-EINPROGRESS); 236 } 237 EXPORT_SYMBOL_GPL(validate_xmit_xfrm); 238 239 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x, 240 const struct xfrm_user_offload *xuo, 241 struct netlink_ext_ack *extack) 242 { 243 int err; 244 struct dst_entry *dst; 245 struct net_device *dev; 246 struct xfrm_dev_offload *xso = &x->xso; 247 xfrm_address_t *saddr; 248 xfrm_address_t *daddr; 249 bool is_packet_offload; 250 251 if (xuo->flags & 252 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) { 253 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 254 return -EINVAL; 255 } 256 257 if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) || 258 (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) { 259 NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction"); 260 return -EINVAL; 261 } 262 263 if (xuo->flags & XFRM_OFFLOAD_INBOUND && x->if_id) { 264 NL_SET_ERR_MSG(extack, "XFRM if_id is not supported in RX path"); 265 return -EINVAL; 266 } 267 268 is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET; 269 270 /* We don't yet support TFC padding. */ 271 if (x->tfcpad) { 272 NL_SET_ERR_MSG(extack, "TFC padding can't be offloaded"); 273 return -EINVAL; 274 } 275 276 dev = dev_get_by_index(net, xuo->ifindex); 277 if (!dev) { 278 struct xfrm_dst_lookup_params params; 279 280 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) { 281 saddr = &x->props.saddr; 282 daddr = &x->id.daddr; 283 } else { 284 saddr = &x->id.daddr; 285 daddr = &x->props.saddr; 286 } 287 288 memset(¶ms, 0, sizeof(params)); 289 params.net = net; 290 params.saddr = saddr; 291 params.daddr = daddr; 292 params.mark = xfrm_smark_get(0, x); 293 dst = __xfrm_dst_lookup(x->props.family, ¶ms); 294 if (IS_ERR(dst)) 295 return (is_packet_offload) ? -EINVAL : 0; 296 297 dev = dst->dev; 298 299 dev_hold(dev); 300 dst_release(dst); 301 } 302 303 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) { 304 xso->dev = NULL; 305 dev_put(dev); 306 return (is_packet_offload) ? -EINVAL : 0; 307 } 308 309 if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN && 310 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 311 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN"); 312 xso->dev = NULL; 313 dev_put(dev); 314 return -EINVAL; 315 } 316 317 if (!x->type_offload) { 318 NL_SET_ERR_MSG(extack, "Type doesn't support offload"); 319 dev_put(dev); 320 return -EINVAL; 321 } 322 323 xso->dev = dev; 324 xso->ifindex = dev->ifindex; 325 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC); 326 327 if (xuo->flags & XFRM_OFFLOAD_INBOUND) 328 xso->dir = XFRM_DEV_OFFLOAD_IN; 329 else 330 xso->dir = XFRM_DEV_OFFLOAD_OUT; 331 332 if (is_packet_offload) 333 xso->type = XFRM_DEV_OFFLOAD_PACKET; 334 else 335 xso->type = XFRM_DEV_OFFLOAD_CRYPTO; 336 337 err = dev->xfrmdev_ops->xdo_dev_state_add(dev, x, extack); 338 if (err) { 339 xso->dev = NULL; 340 xso->dir = 0; 341 netdev_put(dev, &xso->dev_tracker); 342 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 343 344 xfrm_unset_type_offload(x); 345 /* User explicitly requested packet offload mode and configured 346 * policy in addition to the XFRM state. So be civil to users, 347 * and return an error instead of taking fallback path. 348 */ 349 if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) { 350 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state"); 351 return err; 352 } 353 } 354 355 return 0; 356 } 357 EXPORT_SYMBOL_GPL(xfrm_dev_state_add); 358 359 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp, 360 struct xfrm_user_offload *xuo, u8 dir, 361 struct netlink_ext_ack *extack) 362 { 363 struct xfrm_dev_offload *xdo = &xp->xdo; 364 struct net_device *dev; 365 int err; 366 367 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) { 368 /* We support only packet offload mode and it means 369 * that user must set XFRM_OFFLOAD_PACKET bit. 370 */ 371 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 372 return -EINVAL; 373 } 374 375 dev = dev_get_by_index(net, xuo->ifindex); 376 if (!dev) 377 return -EINVAL; 378 379 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) { 380 xdo->dev = NULL; 381 dev_put(dev); 382 NL_SET_ERR_MSG(extack, "Policy offload is not supported"); 383 return -EINVAL; 384 } 385 386 xdo->dev = dev; 387 netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC); 388 xdo->type = XFRM_DEV_OFFLOAD_PACKET; 389 switch (dir) { 390 case XFRM_POLICY_IN: 391 xdo->dir = XFRM_DEV_OFFLOAD_IN; 392 break; 393 case XFRM_POLICY_OUT: 394 xdo->dir = XFRM_DEV_OFFLOAD_OUT; 395 break; 396 case XFRM_POLICY_FWD: 397 xdo->dir = XFRM_DEV_OFFLOAD_FWD; 398 break; 399 default: 400 xdo->dev = NULL; 401 netdev_put(dev, &xdo->dev_tracker); 402 NL_SET_ERR_MSG(extack, "Unrecognized offload direction"); 403 return -EINVAL; 404 } 405 406 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack); 407 if (err) { 408 xdo->dev = NULL; 409 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 410 xdo->dir = 0; 411 netdev_put(dev, &xdo->dev_tracker); 412 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy"); 413 return err; 414 } 415 416 return 0; 417 } 418 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add); 419 420 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 421 { 422 int mtu; 423 struct dst_entry *dst = skb_dst(skb); 424 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 425 struct net_device *dev = x->xso.dev; 426 bool check_tunnel_size; 427 428 if (!x->type_offload || 429 (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED && x->encap)) 430 return false; 431 432 if ((!dev || dev == xfrm_dst_path(dst)->dev) && 433 !xdst->child->xfrm) { 434 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 435 if (skb->len <= mtu) 436 goto ok; 437 438 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 439 goto ok; 440 } 441 442 return false; 443 444 ok: 445 if (!dev) 446 return true; 447 448 check_tunnel_size = x->xso.type == XFRM_DEV_OFFLOAD_PACKET && 449 x->props.mode == XFRM_MODE_TUNNEL; 450 switch (skb_dst(skb)->ops->family) { 451 case AF_INET: 452 /* Check for IPv4 options */ 453 if (ip_hdr(skb)->ihl != 5) 454 return false; 455 if (check_tunnel_size && xfrm4_tunnel_check_size(skb)) 456 return false; 457 break; 458 case AF_INET6: 459 /* Check for IPv6 extensions */ 460 if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr)) 461 return false; 462 if (check_tunnel_size && xfrm6_tunnel_check_size(skb)) 463 return false; 464 break; 465 default: 466 break; 467 } 468 469 if (dev->xfrmdev_ops->xdo_dev_offload_ok) 470 return dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 471 472 return true; 473 } 474 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 475 476 void xfrm_dev_resume(struct sk_buff *skb) 477 { 478 struct net_device *dev = skb->dev; 479 int ret = NETDEV_TX_BUSY; 480 struct netdev_queue *txq; 481 struct softnet_data *sd; 482 unsigned long flags; 483 484 rcu_read_lock(); 485 txq = netdev_core_pick_tx(dev, skb, NULL); 486 487 HARD_TX_LOCK(dev, txq, smp_processor_id()); 488 if (!netif_xmit_frozen_or_stopped(txq)) 489 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 490 HARD_TX_UNLOCK(dev, txq); 491 492 if (!dev_xmit_complete(ret)) { 493 local_irq_save(flags); 494 sd = this_cpu_ptr(&softnet_data); 495 skb_queue_tail(&sd->xfrm_backlog, skb); 496 raise_softirq_irqoff(NET_TX_SOFTIRQ); 497 local_irq_restore(flags); 498 } 499 rcu_read_unlock(); 500 } 501 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 502 503 void xfrm_dev_backlog(struct softnet_data *sd) 504 { 505 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 506 struct sk_buff_head list; 507 struct sk_buff *skb; 508 509 if (skb_queue_empty(xfrm_backlog)) 510 return; 511 512 __skb_queue_head_init(&list); 513 514 spin_lock(&xfrm_backlog->lock); 515 skb_queue_splice_init(xfrm_backlog, &list); 516 spin_unlock(&xfrm_backlog->lock); 517 518 while (!skb_queue_empty(&list)) { 519 skb = __skb_dequeue(&list); 520 xfrm_dev_resume(skb); 521 } 522 523 } 524 #endif 525 526 static int xfrm_api_check(struct net_device *dev) 527 { 528 #ifdef CONFIG_XFRM_OFFLOAD 529 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 530 !(dev->features & NETIF_F_HW_ESP)) 531 return NOTIFY_BAD; 532 533 if ((dev->features & NETIF_F_HW_ESP) && 534 (!(dev->xfrmdev_ops && 535 dev->xfrmdev_ops->xdo_dev_state_add && 536 dev->xfrmdev_ops->xdo_dev_state_delete))) 537 return NOTIFY_BAD; 538 #else 539 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 540 return NOTIFY_BAD; 541 #endif 542 543 return NOTIFY_DONE; 544 } 545 546 static int xfrm_dev_down(struct net_device *dev) 547 { 548 if (dev->features & NETIF_F_HW_ESP) { 549 xfrm_dev_state_flush(dev_net(dev), dev, true); 550 xfrm_dev_policy_flush(dev_net(dev), dev, true); 551 } 552 553 return NOTIFY_DONE; 554 } 555 556 static int xfrm_dev_unregister(struct net_device *dev) 557 { 558 xfrm_dev_state_flush(dev_net(dev), dev, true); 559 xfrm_dev_policy_flush(dev_net(dev), dev, true); 560 561 return NOTIFY_DONE; 562 } 563 564 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 565 { 566 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 567 568 switch (event) { 569 case NETDEV_REGISTER: 570 return xfrm_api_check(dev); 571 572 case NETDEV_FEAT_CHANGE: 573 return xfrm_api_check(dev); 574 575 case NETDEV_DOWN: 576 return xfrm_dev_down(dev); 577 578 case NETDEV_UNREGISTER: 579 return xfrm_dev_unregister(dev); 580 } 581 return NOTIFY_DONE; 582 } 583 584 static struct notifier_block xfrm_dev_notifier = { 585 .notifier_call = xfrm_dev_event, 586 }; 587 588 void __init xfrm_dev_init(void) 589 { 590 register_netdevice_notifier(&xfrm_dev_notifier); 591 } 592