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 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); 317 318 if (xuo->flags & XFRM_OFFLOAD_INBOUND) 319 xso->dir = XFRM_DEV_OFFLOAD_IN; 320 else 321 xso->dir = XFRM_DEV_OFFLOAD_OUT; 322 323 if (is_packet_offload) 324 xso->type = XFRM_DEV_OFFLOAD_PACKET; 325 else 326 xso->type = XFRM_DEV_OFFLOAD_CRYPTO; 327 328 err = dev->xfrmdev_ops->xdo_dev_state_add(dev, x, extack); 329 if (err) { 330 xso->dev = NULL; 331 xso->dir = 0; 332 netdev_put(dev, &xso->dev_tracker); 333 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 334 335 xfrm_unset_type_offload(x); 336 /* User explicitly requested packet offload mode and configured 337 * policy in addition to the XFRM state. So be civil to users, 338 * and return an error instead of taking fallback path. 339 */ 340 if ((err != -EOPNOTSUPP && !is_packet_offload) || is_packet_offload) { 341 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this state"); 342 return err; 343 } 344 } 345 346 return 0; 347 } 348 EXPORT_SYMBOL_GPL(xfrm_dev_state_add); 349 350 int xfrm_dev_policy_add(struct net *net, struct xfrm_policy *xp, 351 struct xfrm_user_offload *xuo, u8 dir, 352 struct netlink_ext_ack *extack) 353 { 354 struct xfrm_dev_offload *xdo = &xp->xdo; 355 struct net_device *dev; 356 int err; 357 358 if (!xuo->flags || xuo->flags & ~XFRM_OFFLOAD_PACKET) { 359 /* We support only packet offload mode and it means 360 * that user must set XFRM_OFFLOAD_PACKET bit. 361 */ 362 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 363 return -EINVAL; 364 } 365 366 dev = dev_get_by_index(net, xuo->ifindex); 367 if (!dev) 368 return -EINVAL; 369 370 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_policy_add) { 371 xdo->dev = NULL; 372 dev_put(dev); 373 NL_SET_ERR_MSG(extack, "Policy offload is not supported"); 374 return -EINVAL; 375 } 376 377 xdo->dev = dev; 378 netdev_tracker_alloc(dev, &xdo->dev_tracker, GFP_ATOMIC); 379 xdo->type = XFRM_DEV_OFFLOAD_PACKET; 380 switch (dir) { 381 case XFRM_POLICY_IN: 382 xdo->dir = XFRM_DEV_OFFLOAD_IN; 383 break; 384 case XFRM_POLICY_OUT: 385 xdo->dir = XFRM_DEV_OFFLOAD_OUT; 386 break; 387 case XFRM_POLICY_FWD: 388 xdo->dir = XFRM_DEV_OFFLOAD_FWD; 389 break; 390 default: 391 xdo->dev = NULL; 392 netdev_put(dev, &xdo->dev_tracker); 393 NL_SET_ERR_MSG(extack, "Unrecognized offload direction"); 394 return -EINVAL; 395 } 396 397 err = dev->xfrmdev_ops->xdo_dev_policy_add(xp, extack); 398 if (err) { 399 xdo->dev = NULL; 400 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 401 xdo->dir = 0; 402 netdev_put(dev, &xdo->dev_tracker); 403 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy"); 404 return err; 405 } 406 407 return 0; 408 } 409 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add); 410 411 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 412 { 413 int mtu; 414 struct dst_entry *dst = skb_dst(skb); 415 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 416 struct net_device *dev = x->xso.dev; 417 bool check_tunnel_size; 418 419 if (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED) 420 return false; 421 422 if ((dev == xfrm_dst_path(dst)->dev) && !xdst->child->xfrm) { 423 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 424 if (skb->len <= mtu) 425 goto ok; 426 427 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 428 goto ok; 429 } 430 431 return false; 432 433 ok: 434 check_tunnel_size = x->xso.type == XFRM_DEV_OFFLOAD_PACKET && 435 x->props.mode == XFRM_MODE_TUNNEL; 436 switch (x->props.family) { 437 case AF_INET: 438 /* Check for IPv4 options */ 439 if (ip_hdr(skb)->ihl != 5) 440 return false; 441 if (check_tunnel_size && xfrm4_tunnel_check_size(skb)) 442 return false; 443 break; 444 case AF_INET6: 445 /* Check for IPv6 extensions */ 446 if (ipv6_ext_hdr(ipv6_hdr(skb)->nexthdr)) 447 return false; 448 if (check_tunnel_size && xfrm6_tunnel_check_size(skb)) 449 return false; 450 break; 451 default: 452 break; 453 } 454 455 if (dev->xfrmdev_ops->xdo_dev_offload_ok) 456 return dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 457 458 return true; 459 } 460 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 461 462 void xfrm_dev_resume(struct sk_buff *skb) 463 { 464 struct net_device *dev = skb->dev; 465 int ret = NETDEV_TX_BUSY; 466 struct netdev_queue *txq; 467 struct softnet_data *sd; 468 unsigned long flags; 469 470 rcu_read_lock(); 471 txq = netdev_core_pick_tx(dev, skb, NULL); 472 473 HARD_TX_LOCK(dev, txq, smp_processor_id()); 474 if (!netif_xmit_frozen_or_stopped(txq)) 475 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 476 HARD_TX_UNLOCK(dev, txq); 477 478 if (!dev_xmit_complete(ret)) { 479 local_irq_save(flags); 480 sd = this_cpu_ptr(&softnet_data); 481 skb_queue_tail(&sd->xfrm_backlog, skb); 482 raise_softirq_irqoff(NET_TX_SOFTIRQ); 483 local_irq_restore(flags); 484 } 485 rcu_read_unlock(); 486 } 487 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 488 489 void xfrm_dev_backlog(struct softnet_data *sd) 490 { 491 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 492 struct sk_buff_head list; 493 struct sk_buff *skb; 494 495 if (skb_queue_empty(xfrm_backlog)) 496 return; 497 498 __skb_queue_head_init(&list); 499 500 spin_lock(&xfrm_backlog->lock); 501 skb_queue_splice_init(xfrm_backlog, &list); 502 spin_unlock(&xfrm_backlog->lock); 503 504 while (!skb_queue_empty(&list)) { 505 skb = __skb_dequeue(&list); 506 xfrm_dev_resume(skb); 507 } 508 509 } 510 #endif 511 512 static int xfrm_api_check(struct net_device *dev) 513 { 514 #ifdef CONFIG_XFRM_OFFLOAD 515 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 516 !(dev->features & NETIF_F_HW_ESP)) 517 return NOTIFY_BAD; 518 519 if ((dev->features & NETIF_F_HW_ESP) && 520 (!(dev->xfrmdev_ops && 521 dev->xfrmdev_ops->xdo_dev_state_add && 522 dev->xfrmdev_ops->xdo_dev_state_delete))) 523 return NOTIFY_BAD; 524 #else 525 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 526 return NOTIFY_BAD; 527 #endif 528 529 return NOTIFY_DONE; 530 } 531 532 static int xfrm_dev_down(struct net_device *dev) 533 { 534 if (dev->features & NETIF_F_HW_ESP) { 535 xfrm_dev_state_flush(dev_net(dev), dev, true); 536 xfrm_dev_policy_flush(dev_net(dev), dev, true); 537 } 538 539 return NOTIFY_DONE; 540 } 541 542 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 543 { 544 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 545 546 switch (event) { 547 case NETDEV_REGISTER: 548 return xfrm_api_check(dev); 549 550 case NETDEV_FEAT_CHANGE: 551 return xfrm_api_check(dev); 552 553 case NETDEV_DOWN: 554 case NETDEV_UNREGISTER: 555 return xfrm_dev_down(dev); 556 } 557 return NOTIFY_DONE; 558 } 559 560 static struct notifier_block xfrm_dev_notifier = { 561 .notifier_call = xfrm_dev_event, 562 }; 563 564 void __init xfrm_dev_init(void) 565 { 566 register_netdevice_notifier(&xfrm_dev_notifier); 567 } 568