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/xfrm.h> 19 #include <linux/notifier.h> 20 21 #ifdef CONFIG_XFRM_OFFLOAD 22 static void __xfrm_transport_prep(struct xfrm_state *x, struct sk_buff *skb, 23 unsigned int hsize) 24 { 25 struct xfrm_offload *xo = xfrm_offload(skb); 26 27 skb_reset_mac_len(skb); 28 if (xo->flags & XFRM_GSO_SEGMENT) 29 skb->transport_header -= x->props.header_len; 30 31 pskb_pull(skb, skb_transport_offset(skb) + x->props.header_len); 32 } 33 34 static void __xfrm_mode_tunnel_prep(struct xfrm_state *x, struct sk_buff *skb, 35 unsigned int hsize) 36 37 { 38 struct xfrm_offload *xo = xfrm_offload(skb); 39 40 if (xo->flags & XFRM_GSO_SEGMENT) 41 skb->transport_header = skb->network_header + hsize; 42 43 skb_reset_mac_len(skb); 44 pskb_pull(skb, skb->mac_len + x->props.header_len); 45 } 46 47 static void __xfrm_mode_beet_prep(struct xfrm_state *x, struct sk_buff *skb, 48 unsigned int hsize) 49 { 50 struct xfrm_offload *xo = xfrm_offload(skb); 51 int phlen = 0; 52 53 if (xo->flags & XFRM_GSO_SEGMENT) 54 skb->transport_header = skb->network_header + hsize; 55 56 skb_reset_mac_len(skb); 57 if (x->sel.family != AF_INET6) { 58 phlen = IPV4_BEET_PHMAXLEN; 59 if (x->outer_mode.family == AF_INET6) 60 phlen += sizeof(struct ipv6hdr) - sizeof(struct iphdr); 61 } 62 63 pskb_pull(skb, skb->mac_len + hsize + (x->props.header_len - phlen)); 64 } 65 66 /* Adjust pointers into the packet when IPsec is done at layer2 */ 67 static void xfrm_outer_mode_prep(struct xfrm_state *x, struct sk_buff *skb) 68 { 69 switch (x->outer_mode.encap) { 70 case XFRM_MODE_TUNNEL: 71 if (x->outer_mode.family == AF_INET) 72 return __xfrm_mode_tunnel_prep(x, skb, 73 sizeof(struct iphdr)); 74 if (x->outer_mode.family == AF_INET6) 75 return __xfrm_mode_tunnel_prep(x, skb, 76 sizeof(struct ipv6hdr)); 77 break; 78 case XFRM_MODE_TRANSPORT: 79 if (x->outer_mode.family == AF_INET) 80 return __xfrm_transport_prep(x, skb, 81 sizeof(struct iphdr)); 82 if (x->outer_mode.family == AF_INET6) 83 return __xfrm_transport_prep(x, skb, 84 sizeof(struct ipv6hdr)); 85 break; 86 case XFRM_MODE_BEET: 87 if (x->outer_mode.family == AF_INET) 88 return __xfrm_mode_beet_prep(x, skb, 89 sizeof(struct iphdr)); 90 if (x->outer_mode.family == AF_INET6) 91 return __xfrm_mode_beet_prep(x, skb, 92 sizeof(struct ipv6hdr)); 93 break; 94 case XFRM_MODE_ROUTEOPTIMIZATION: 95 case XFRM_MODE_IN_TRIGGER: 96 break; 97 } 98 } 99 100 struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, bool *again) 101 { 102 int err; 103 unsigned long flags; 104 struct xfrm_state *x; 105 struct softnet_data *sd; 106 struct sk_buff *skb2, *nskb, *pskb = NULL; 107 netdev_features_t esp_features = features; 108 struct xfrm_offload *xo = xfrm_offload(skb); 109 struct net_device *dev = skb->dev; 110 struct sec_path *sp; 111 112 if (!xo || (xo->flags & XFRM_XMIT)) 113 return skb; 114 115 if (!(features & NETIF_F_HW_ESP)) 116 esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK); 117 118 sp = skb_sec_path(skb); 119 x = sp->xvec[sp->len - 1]; 120 if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND) 121 return skb; 122 123 /* This skb was already validated on the upper/virtual dev */ 124 if ((x->xso.dev != dev) && (x->xso.real_dev == dev)) 125 return skb; 126 127 local_irq_save(flags); 128 sd = this_cpu_ptr(&softnet_data); 129 err = !skb_queue_empty(&sd->xfrm_backlog); 130 local_irq_restore(flags); 131 132 if (err) { 133 *again = true; 134 return skb; 135 } 136 137 if (skb_is_gso(skb) && unlikely(x->xso.dev != dev)) { 138 struct sk_buff *segs; 139 140 /* Packet got rerouted, fixup features and segment it. */ 141 esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP); 142 143 segs = skb_gso_segment(skb, esp_features); 144 if (IS_ERR(segs)) { 145 kfree_skb(skb); 146 dev_core_stats_tx_dropped_inc(dev); 147 return NULL; 148 } else { 149 consume_skb(skb); 150 skb = segs; 151 } 152 } 153 154 if (!skb->next) { 155 esp_features |= skb->dev->gso_partial_features; 156 xfrm_outer_mode_prep(x, skb); 157 158 xo->flags |= XFRM_DEV_RESUME; 159 160 err = x->type_offload->xmit(x, skb, esp_features); 161 if (err) { 162 if (err == -EINPROGRESS) 163 return NULL; 164 165 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 166 kfree_skb(skb); 167 return NULL; 168 } 169 170 skb_push(skb, skb->data - skb_mac_header(skb)); 171 172 return skb; 173 } 174 175 skb_list_walk_safe(skb, skb2, nskb) { 176 esp_features |= skb->dev->gso_partial_features; 177 skb_mark_not_on_list(skb2); 178 179 xo = xfrm_offload(skb2); 180 xo->flags |= XFRM_DEV_RESUME; 181 182 xfrm_outer_mode_prep(x, skb2); 183 184 err = x->type_offload->xmit(x, skb2, esp_features); 185 if (!err) { 186 skb2->next = nskb; 187 } else if (err != -EINPROGRESS) { 188 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR); 189 skb2->next = nskb; 190 kfree_skb_list(skb2); 191 return NULL; 192 } else { 193 if (skb == skb2) 194 skb = nskb; 195 else 196 pskb->next = nskb; 197 198 continue; 199 } 200 201 skb_push(skb2, skb2->data - skb_mac_header(skb2)); 202 pskb = skb2; 203 } 204 205 return skb; 206 } 207 EXPORT_SYMBOL_GPL(validate_xmit_xfrm); 208 209 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x, 210 struct xfrm_user_offload *xuo) 211 { 212 int err; 213 struct dst_entry *dst; 214 struct net_device *dev; 215 struct xfrm_state_offload *xso = &x->xso; 216 xfrm_address_t *saddr; 217 xfrm_address_t *daddr; 218 219 if (!x->type_offload) 220 return -EINVAL; 221 222 /* We don't yet support UDP encapsulation and TFC padding. */ 223 if (x->encap || x->tfcpad) 224 return -EINVAL; 225 226 if (xuo->flags & ~(XFRM_OFFLOAD_IPV6 | XFRM_OFFLOAD_INBOUND)) 227 return -EINVAL; 228 229 dev = dev_get_by_index(net, xuo->ifindex); 230 if (!dev) { 231 if (!(xuo->flags & XFRM_OFFLOAD_INBOUND)) { 232 saddr = &x->props.saddr; 233 daddr = &x->id.daddr; 234 } else { 235 saddr = &x->id.daddr; 236 daddr = &x->props.saddr; 237 } 238 239 dst = __xfrm_dst_lookup(net, 0, 0, saddr, daddr, 240 x->props.family, 241 xfrm_smark_get(0, x)); 242 if (IS_ERR(dst)) 243 return 0; 244 245 dev = dst->dev; 246 247 dev_hold(dev); 248 dst_release(dst); 249 } 250 251 if (!dev->xfrmdev_ops || !dev->xfrmdev_ops->xdo_dev_state_add) { 252 xso->dev = NULL; 253 dev_put(dev); 254 return 0; 255 } 256 257 if (x->props.flags & XFRM_STATE_ESN && 258 !dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 259 xso->dev = NULL; 260 dev_put(dev); 261 return -EINVAL; 262 } 263 264 xso->dev = dev; 265 netdev_tracker_alloc(dev, &xso->dev_tracker, GFP_ATOMIC); 266 xso->real_dev = dev; 267 xso->num_exthdrs = 1; 268 /* Don't forward bit that is not implemented */ 269 xso->flags = xuo->flags & ~XFRM_OFFLOAD_IPV6; 270 271 err = dev->xfrmdev_ops->xdo_dev_state_add(x); 272 if (err) { 273 xso->num_exthdrs = 0; 274 xso->flags = 0; 275 xso->dev = NULL; 276 xso->real_dev = NULL; 277 dev_put_track(dev, &xso->dev_tracker); 278 279 if (err != -EOPNOTSUPP) 280 return err; 281 } 282 283 return 0; 284 } 285 EXPORT_SYMBOL_GPL(xfrm_dev_state_add); 286 287 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x) 288 { 289 int mtu; 290 struct dst_entry *dst = skb_dst(skb); 291 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 292 struct net_device *dev = x->xso.dev; 293 294 if (!x->type_offload || x->encap) 295 return false; 296 297 if ((!dev || (dev == xfrm_dst_path(dst)->dev)) && 298 (!xdst->child->xfrm)) { 299 mtu = xfrm_state_mtu(x, xdst->child_mtu_cached); 300 if (skb->len <= mtu) 301 goto ok; 302 303 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 304 goto ok; 305 } 306 307 return false; 308 309 ok: 310 if (dev && dev->xfrmdev_ops && dev->xfrmdev_ops->xdo_dev_offload_ok) 311 return x->xso.dev->xfrmdev_ops->xdo_dev_offload_ok(skb, x); 312 313 return true; 314 } 315 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok); 316 317 void xfrm_dev_resume(struct sk_buff *skb) 318 { 319 struct net_device *dev = skb->dev; 320 int ret = NETDEV_TX_BUSY; 321 struct netdev_queue *txq; 322 struct softnet_data *sd; 323 unsigned long flags; 324 325 rcu_read_lock(); 326 txq = netdev_core_pick_tx(dev, skb, NULL); 327 328 HARD_TX_LOCK(dev, txq, smp_processor_id()); 329 if (!netif_xmit_frozen_or_stopped(txq)) 330 skb = dev_hard_start_xmit(skb, dev, txq, &ret); 331 HARD_TX_UNLOCK(dev, txq); 332 333 if (!dev_xmit_complete(ret)) { 334 local_irq_save(flags); 335 sd = this_cpu_ptr(&softnet_data); 336 skb_queue_tail(&sd->xfrm_backlog, skb); 337 raise_softirq_irqoff(NET_TX_SOFTIRQ); 338 local_irq_restore(flags); 339 } 340 rcu_read_unlock(); 341 } 342 EXPORT_SYMBOL_GPL(xfrm_dev_resume); 343 344 void xfrm_dev_backlog(struct softnet_data *sd) 345 { 346 struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog; 347 struct sk_buff_head list; 348 struct sk_buff *skb; 349 350 if (skb_queue_empty(xfrm_backlog)) 351 return; 352 353 __skb_queue_head_init(&list); 354 355 spin_lock(&xfrm_backlog->lock); 356 skb_queue_splice_init(xfrm_backlog, &list); 357 spin_unlock(&xfrm_backlog->lock); 358 359 while (!skb_queue_empty(&list)) { 360 skb = __skb_dequeue(&list); 361 xfrm_dev_resume(skb); 362 } 363 364 } 365 #endif 366 367 static int xfrm_api_check(struct net_device *dev) 368 { 369 #ifdef CONFIG_XFRM_OFFLOAD 370 if ((dev->features & NETIF_F_HW_ESP_TX_CSUM) && 371 !(dev->features & NETIF_F_HW_ESP)) 372 return NOTIFY_BAD; 373 374 if ((dev->features & NETIF_F_HW_ESP) && 375 (!(dev->xfrmdev_ops && 376 dev->xfrmdev_ops->xdo_dev_state_add && 377 dev->xfrmdev_ops->xdo_dev_state_delete))) 378 return NOTIFY_BAD; 379 #else 380 if (dev->features & (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM)) 381 return NOTIFY_BAD; 382 #endif 383 384 return NOTIFY_DONE; 385 } 386 387 static int xfrm_dev_down(struct net_device *dev) 388 { 389 if (dev->features & NETIF_F_HW_ESP) 390 xfrm_dev_state_flush(dev_net(dev), dev, true); 391 392 return NOTIFY_DONE; 393 } 394 395 static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr) 396 { 397 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 398 399 switch (event) { 400 case NETDEV_REGISTER: 401 return xfrm_api_check(dev); 402 403 case NETDEV_FEAT_CHANGE: 404 return xfrm_api_check(dev); 405 406 case NETDEV_DOWN: 407 case NETDEV_UNREGISTER: 408 return xfrm_dev_down(dev); 409 } 410 return NOTIFY_DONE; 411 } 412 413 static struct notifier_block xfrm_dev_notifier = { 414 .notifier_call = xfrm_dev_event, 415 }; 416 417 void __init xfrm_dev_init(void) 418 { 419 register_netdevice_notifier(&xfrm_dev_notifier); 420 } 421