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 /* This skb was already validated on the upper/virtual dev */ 149 if ((x->xso.dev != dev) && (x->xso.real_dev == dev)) 150 return skb; 151 152 local_irq_save(flags); 153 sd = this_cpu_ptr(&softnet_data); 154 err = !skb_queue_empty(&sd->xfrm_backlog); 155 local_irq_restore(flags); 156 157 if (err) { 158 *again = true; 159 return skb; 160 } 161 162 if (skb_is_gso(skb) && (unlikely(x->xso.dev != dev) || 163 unlikely(xmit_xfrm_check_overflow(skb)))) { 164 struct sk_buff *segs; 165 166 /* Packet got rerouted, fixup features and segment it. */ 167 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP); 168 169 segs = skb_gso_segment(skb, esp_features); 170 if (IS_ERR(segs)) { 171 kfree_skb(skb); 172 dev_core_stats_tx_dropped_inc(dev); 173 return NULL; 174 } else { 175 consume_skb(skb); 176 skb = segs; 177 } 178 } 179 180 if (!skb->next) { 181 esp_features |= skb->dev->gso_partial_features; 182 xfrm_outer_mode_prep(x, skb); 183 184 xo->flags |= XFRM_DEV_RESUME; 185 186 err = x->type_offload->xmit(x, skb, esp_features); 187 if (err) { 188 if (err == -EINPROGRESS) 189 return NULL; 190 191 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 192 kfree_skb(skb); 193 return NULL; 194 } 195 196 skb_push(skb, skb->data - skb_mac_header(skb)); 197 198 return skb; 199 } 200 201 skb_list_walk_safe(skb, skb2, nskb) { 202 esp_features |= skb->dev->gso_partial_features; 203 skb_mark_not_on_list(skb2); 204 205 xo = xfrm_offload(skb2); 206 xo->flags |= XFRM_DEV_RESUME; 207 208 xfrm_outer_mode_prep(x, skb2); 209 210 err = x->type_offload->xmit(x, skb2, esp_features); 211 if (!err) { 212 skb2->next = nskb; 213 } else if (err != -EINPROGRESS) { 214 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 215 skb2->next = nskb; 216 kfree_skb_list(skb2); 217 return NULL; 218 } else { 219 if (skb == skb2) 220 skb = nskb; 221 else 222 pskb->next = nskb; 223 224 continue; 225 } 226 227 skb_push(skb2, skb2->data - skb_mac_header(skb2)); 228 pskb = skb2; 229 } 230 231 return skb; 232 } 233 EXPORT_SYMBOL_GPL(validate_xmit_xfrm); 234 235 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x, 236 struct xfrm_user_offload *xuo, 237 struct netlink_ext_ack *extack) 238 { 239 int err; 240 struct dst_entry *dst; 241 struct net_device *dev; 242 struct xfrm_dev_offload *xso = &x->xso; 243 xfrm_address_t *saddr; 244 xfrm_address_t *daddr; 245 bool is_packet_offload; 246 247 if (xuo->flags & 248 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) { 249 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 250 return -EINVAL; 251 } 252 253 if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) || 254 (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) { 255 NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction"); 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 xfrm_set_type_offload(x); 309 if (!x->type_offload) { 310 NL_SET_ERR_MSG(extack, "Type doesn't support offload"); 311 dev_put(dev); 312 return -EINVAL; 313 } 314 315 xso->dev = dev; 316 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC); 317 xso->real_dev = dev; 318 319 if (xuo->flags & XFRM_OFFLOAD_INBOUND) 320 xso->dir = XFRM_DEV_OFFLOAD_IN; 321 else 322 xso->dir = XFRM_DEV_OFFLOAD_OUT; 323 324 if (is_packet_offload) 325 xso->type = XFRM_DEV_OFFLOAD_PACKET; 326 else 327 xso->type = XFRM_DEV_OFFLOAD_CRYPTO; 328 329 err = dev->xfrmdev_ops->xdo_dev_state_add(x, extack); 330 if (err) { 331 xso->dev = NULL; 332 xso->dir = 0; 333 xso->real_dev = NULL; 334 netdev_put(dev, &xso->dev_tracker); 335 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 336 337 xfrm_unset_type_offload(x); 338 /* User explicitly requested packet offload mode and configured 339 * policy in addition to the XFRM state. So be civil to users, 340 * and return an error instead of taking fallback path. 341 */ 342 if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) { 343 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state"); 344 return err; 345 } 346 } 347 348 return 0; 349 } 350 EXPORT_SYMBOL_GPL(xfrm_dev_state_add); 351 352 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp, 353 struct xfrm_user_offload *xuo, u8 dir, 354 struct netlink_ext_ack *extack) 355 { 356 struct xfrm_dev_offload *xdo = &xp->xdo; 357 struct net_device *dev; 358 int err; 359 360 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) { 361 /* We support only packet offload mode and it means 362 * that user must set XFRM_OFFLOAD_PACKET bit. 363 */ 364 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 365 return -EINVAL; 366 } 367 368 dev = dev_get_by_index(net, xuo->ifindex); 369 if (!dev) 370 return -EINVAL; 371 372 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) { 373 xdo->dev = NULL; 374 dev_put(dev); 375 NL_SET_ERR_MSG(extack, "Policy offload is not supported"); 376 return -EINVAL; 377 } 378 379 xdo->dev = dev; 380 netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC); 381 xdo->real_dev = dev; 382 xdo->type = XFRM_DEV_OFFLOAD_PACKET; 383 switch (dir) { 384 case XFRM_POLICY_IN: 385 xdo->dir = XFRM_DEV_OFFLOAD_IN; 386 break; 387 case XFRM_POLICY_OUT: 388 xdo->dir = XFRM_DEV_OFFLOAD_OUT; 389 break; 390 case XFRM_POLICY_FWD: 391 xdo->dir = XFRM_DEV_OFFLOAD_FWD; 392 break; 393 default: 394 xdo->dev = NULL; 395 netdev_put(dev, &xdo->dev_tracker); 396 NL_SET_ERR_MSG(extack, "Unrecognized offload direction"); 397 return -EINVAL; 398 } 399 400 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack); 401 if (err) { 402 xdo->dev = NULL; 403 xdo->real_dev = NULL; 404 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 405 xdo->dir = 0; 406 netdev_put(dev, &xdo->dev_tracker); 407 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy"); 408 return err; 409 } 410 411 return 0; 412 } 413 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add); 414 415 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 416 { 417 int mtu; 418 struct dst_entry *dst = skb_dst(skb); 419 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 420 struct net_device *dev = x->xso.dev; 421 bool check_tunnel_size; 422 423 if (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED) 424 return false; 425 426 if ((dev == xfrm_dst_path(dst)->dev) && !xdst->child->xfrm) { 427 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 428 if (skb->len <= mtu) 429 goto ok; 430 431 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 432 goto ok; 433 } 434 435 return false; 436 437 ok: 438 check_tunnel_size = x->xso.type == XFRM_DEV_OFFLOAD_PACKET && 439 x->props.mode == XFRM_MODE_TUNNEL; 440 switch (x->props.family) { 441 case AF_INET: 442 /* Check for IPv4 options */ 443 if (ip_hdr(skb)->ihl != 5) 444 return false; 445 if (check_tunnel_size && xfrm4_tunnel_check_size(skb)) 446 return false; 447 break; 448 case AF_INET6: 449 /* Check for IPv6 extensions */ 450 if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr)) 451 return false; 452 if (check_tunnel_size && xfrm6_tunnel_check_size(skb)) 453 return false; 454 break; 455 default: 456 break; 457 } 458 459 if (dev->xfrmdev_ops->xdo_dev_offload_ok) 460 return dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 461 462 return true; 463 } 464 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 465 466 void xfrm_dev_resume(struct sk_buff *skb) 467 { 468 struct net_device *dev = skb->dev; 469 int ret = NETDEV_TX_BUSY; 470 struct netdev_queue *txq; 471 struct softnet_data *sd; 472 unsigned long flags; 473 474 rcu_read_lock(); 475 txq = netdev_core_pick_tx(dev, skb, NULL); 476 477 HARD_TX_LOCK(dev, txq, smp_processor_id()); 478 if (!netif_xmit_frozen_or_stopped(txq)) 479 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 480 HARD_TX_UNLOCK(dev, txq); 481 482 if (!dev_xmit_complete(ret)) { 483 local_irq_save(flags); 484 sd = this_cpu_ptr(&softnet_data); 485 skb_queue_tail(&sd->xfrm_backlog, skb); 486 raise_softirq_irqoff(NET_TX_SOFTIRQ); 487 local_irq_restore(flags); 488 } 489 rcu_read_unlock(); 490 } 491 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 492 493 void xfrm_dev_backlog(struct softnet_data *sd) 494 { 495 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 496 struct sk_buff_head list; 497 struct sk_buff *skb; 498 499 if (skb_queue_empty(xfrm_backlog)) 500 return; 501 502 __skb_queue_head_init(&list); 503 504 spin_lock(&xfrm_backlog->lock); 505 skb_queue_splice_init(xfrm_backlog, &list); 506 spin_unlock(&xfrm_backlog->lock); 507 508 while (!skb_queue_empty(&list)) { 509 skb = __skb_dequeue(&list); 510 xfrm_dev_resume(skb); 511 } 512 513 } 514 #endif 515 516 static int xfrm_api_check(struct net_device *dev) 517 { 518 #ifdef CONFIG_XFRM_OFFLOAD 519 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 520 !(dev->features & NETIF_F_HW_ESP)) 521 return NOTIFY_BAD; 522 523 if ((dev->features & NETIF_F_HW_ESP) && 524 (!(dev->xfrmdev_ops && 525 dev->xfrmdev_ops->xdo_dev_state_add && 526 dev->xfrmdev_ops->xdo_dev_state_delete))) 527 return NOTIFY_BAD; 528 #else 529 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 530 return NOTIFY_BAD; 531 #endif 532 533 return NOTIFY_DONE; 534 } 535 536 static int xfrm_dev_down(struct net_device *dev) 537 { 538 if (dev->features & NETIF_F_HW_ESP) { 539 xfrm_dev_state_flush(dev_net(dev), dev, true); 540 xfrm_dev_policy_flush(dev_net(dev), dev, true); 541 } 542 543 return NOTIFY_DONE; 544 } 545 546 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 547 { 548 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 549 550 switch (event) { 551 case NETDEV_REGISTER: 552 return xfrm_api_check(dev); 553 554 case NETDEV_FEAT_CHANGE: 555 return xfrm_api_check(dev); 556 557 case NETDEV_DOWN: 558 case NETDEV_UNREGISTER: 559 return xfrm_dev_down(dev); 560 } 561 return NOTIFY_DONE; 562 } 563 564 static struct notifier_block xfrm_dev_notifier = { 565 .notifier_call = xfrm_dev_event, 566 }; 567 568 void __init xfrm_dev_init(void) 569 { 570 register_netdevice_notifier(&xfrm_dev_notifier); 571 } 572