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 (!x->type_offload) { 248 NL_SET_ERR_MSG(extack, "Type doesn't support offload"); 249 return -EINVAL; 250 } 251 252 if (xuo->flags & 253 ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND | XFRM_OFFLOAD_PACKET)) { 254 NL_SET_ERR_MSG(extack, "Unrecognized flags in offload request"); 255 return -EINVAL; 256 } 257 258 if ((xuo->flags & XFRM_OFFLOAD_INBOUND && x->dir == XFRM_SA_DIR_OUT) || 259 (!(xuo->flags & XFRM_OFFLOAD_INBOUND) && x->dir == XFRM_SA_DIR_IN)) { 260 NL_SET_ERR_MSG(extack, "Mismatched SA and offload direction"); 261 return -EINVAL; 262 } 263 264 is_packet_offload = xuo->flags & XFRM_OFFLOAD_PACKET; 265 266 /* We don't yet support TFC padding. */ 267 if (x->tfcpad) { 268 NL_SET_ERR_MSG(extack, "TFC padding can't be offloaded"); 269 return -EINVAL; 270 } 271 272 dev = dev_get_by_index(net, xuo->ifindex); 273 if (!dev) { 274 struct xfrm_dst_lookup_params params; 275 276 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) { 277 saddr = &x->props.saddr; 278 daddr = &x->id.daddr; 279 } else { 280 saddr = &x->id.daddr; 281 daddr = &x->props.saddr; 282 } 283 284 memset(¶ms, 0, sizeof(params)); 285 params.net = net; 286 params.saddr = saddr; 287 params.daddr = daddr; 288 params.mark = xfrm_smark_get(0, x); 289 dst = __xfrm_dst_lookup(x->props.family, ¶ms); 290 if (IS_ERR(dst)) 291 return (is_packet_offload) ? -EINVAL : 0; 292 293 dev = dst->dev; 294 295 dev_hold(dev); 296 dst_release(dst); 297 } 298 299 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) { 300 xso->dev = NULL; 301 dev_put(dev); 302 return (is_packet_offload) ? -EINVAL : 0; 303 } 304 305 if (!is_packet_offload && x->props.flags & XFRM_STATE_ESN && 306 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 307 NL_SET_ERR_MSG(extack, "Device doesn't support offload with ESN"); 308 xso->dev = NULL; 309 dev_put(dev); 310 return -EINVAL; 311 } 312 313 xso->dev = dev; 314 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC); 315 xso->real_dev = dev; 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(x, extack); 328 if (err) { 329 xso->dev = NULL; 330 xso->dir = 0; 331 xso->real_dev = NULL; 332 netdev_put(dev, &xso->dev_tracker); 333 xso->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 334 335 /* 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->real_dev = dev; 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->real_dev = NULL; 401 xdo->type = XFRM_DEV_OFFLOAD_UNSPECIFIED; 402 xdo->dir = 0; 403 netdev_put(dev, &xdo->dev_tracker); 404 NL_SET_ERR_MSG_WEAK(extack, "Device failed to offload this policy"); 405 return err; 406 } 407 408 return 0; 409 } 410 EXPORT_SYMBOL_GPL(xfrm_dev_policy_add); 411 412 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 413 { 414 int mtu; 415 struct dst_entry *dst = skb_dst(skb); 416 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 417 struct net_device *dev = x->xso.dev; 418 419 if (!x->type_offload || 420 (x->xso.type == XFRM_DEV_OFFLOAD_UNSPECIFIED && x->encap)) 421 return false; 422 423 if (x->xso.type == XFRM_DEV_OFFLOAD_PACKET || 424 ((!dev || (dev == xfrm_dst_path(dst)->dev)) && 425 !xdst->child->xfrm)) { 426 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 427 if (skb->len <= mtu) 428 goto ok; 429 430 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 431 goto ok; 432 } 433 434 return false; 435 436 ok: 437 if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok) 438 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 439 440 return true; 441 } 442 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 443 444 void xfrm_dev_resume(struct sk_buff *skb) 445 { 446 struct net_device *dev = skb->dev; 447 int ret = NETDEV_TX_BUSY; 448 struct netdev_queue *txq; 449 struct softnet_data *sd; 450 unsigned long flags; 451 452 rcu_read_lock(); 453 txq = netdev_core_pick_tx(dev, skb, NULL); 454 455 HARD_TX_LOCK(dev, txq, smp_processor_id()); 456 if (!netif_xmit_frozen_or_stopped(txq)) 457 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 458 HARD_TX_UNLOCK(dev, txq); 459 460 if (!dev_xmit_complete(ret)) { 461 local_irq_save(flags); 462 sd = this_cpu_ptr(&softnet_data); 463 skb_queue_tail(&sd->xfrm_backlog, skb); 464 raise_softirq_irqoff(NET_TX_SOFTIRQ); 465 local_irq_restore(flags); 466 } 467 rcu_read_unlock(); 468 } 469 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 470 471 void xfrm_dev_backlog(struct softnet_data *sd) 472 { 473 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 474 struct sk_buff_head list; 475 struct sk_buff *skb; 476 477 if (skb_queue_empty(xfrm_backlog)) 478 return; 479 480 __skb_queue_head_init(&list); 481 482 spin_lock(&xfrm_backlog->lock); 483 skb_queue_splice_init(xfrm_backlog, &list); 484 spin_unlock(&xfrm_backlog->lock); 485 486 while (!skb_queue_empty(&list)) { 487 skb = __skb_dequeue(&list); 488 xfrm_dev_resume(skb); 489 } 490 491 } 492 #endif 493 494 static int xfrm_api_check(struct net_device *dev) 495 { 496 #ifdef CONFIG_XFRM_OFFLOAD 497 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 498 !(dev->features & NETIF_F_HW_ESP)) 499 return NOTIFY_BAD; 500 501 if ((dev->features & NETIF_F_HW_ESP) && 502 (!(dev->xfrmdev_ops && 503 dev->xfrmdev_ops->xdo_dev_state_add && 504 dev->xfrmdev_ops->xdo_dev_state_delete))) 505 return NOTIFY_BAD; 506 #else 507 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 508 return NOTIFY_BAD; 509 #endif 510 511 return NOTIFY_DONE; 512 } 513 514 static int xfrm_dev_down(struct net_device *dev) 515 { 516 if (dev->features & NETIF_F_HW_ESP) { 517 xfrm_dev_state_flush(dev_net(dev), dev, true); 518 xfrm_dev_policy_flush(dev_net(dev), dev, true); 519 } 520 521 return NOTIFY_DONE; 522 } 523 524 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 525 { 526 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 527 528 switch (event) { 529 case NETDEV_REGISTER: 530 return xfrm_api_check(dev); 531 532 case NETDEV_FEAT_CHANGE: 533 return xfrm_api_check(dev); 534 535 case NETDEV_DOWN: 536 case NETDEV_UNREGISTER: 537 return xfrm_dev_down(dev); 538 } 539 return NOTIFY_DONE; 540 } 541 542 static struct notifier_block xfrm_dev_notifier = { 543 .notifier_call = xfrm_dev_event, 544 }; 545 546 void __init xfrm_dev_init(void) 547 { 548 register_netdevice_notifier(&xfrm_dev_notifier); 549 } 550