net/mctp/route.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Management Component Transport Protocol (MCTP) - routing
   4  * implementation.
   5  *
   6  * This is currently based on a simple routing table, with no dst cache. The
   7  * number of routes should stay fairly small, so the lookup cost is small.
   8  *
   9  * Copyright (c) 2021 Code Construct
  10  * Copyright (c) 2021 Google
  11  */
  12
  13 #include <linux/idr.h>
  14 #include <linux/kconfig.h>
  15 #include <linux/mctp.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/rtnetlink.h>
  18 #include <linux/skbuff.h>
  19
  20 #include <uapi/linux/if_arp.h>
  21
  22 #include <net/mctp.h>
  23 #include <net/mctpdevice.h>
  24 #include <net/netlink.h>
  25 #include <net/sock.h>
  26
  27 #include <trace/events/mctp.h>
  28
  29 static const unsigned int mctp_message_maxlen = 64 * 1024;
  30 static const unsigned long mctp_key_lifetime = 6 * CONFIG_HZ;
  31
  32 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev);
  33
  34 /* route output callbacks */
  35 static int mctp_route_discard(struct mctp_route *route, struct sk_buff *skb)
  36 {
  37         kfree_skb(skb);
  38         return 0;
  39 }
  40
  41 static struct mctp_sock *mctp_lookup_bind(struct net *net, struct sk_buff *skb)
  42 {
  43         struct mctp_skb_cb *cb = mctp_cb(skb);
  44         struct mctp_hdr *mh;
  45         struct sock *sk;
  46         u8 type;
  47
  48         WARN_ON(!rcu_read_lock_held());
  49
  50         /* TODO: look up in skb->cb? */
  51         mh = mctp_hdr(skb);
  52
  53         if (!skb_headlen(skb))
  54                 return NULL;
  55
  56         type = (*(u8 *)skb->data) & 0x7f;
  57
  58         sk_for_each_rcu(sk, &net->mctp.binds) {
  59                 struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
  60
  61                 if (msk->bind_net != MCTP_NET_ANY && msk->bind_net != cb->net)
  62                         continue;
  63
  64                 if (msk->bind_type != type)
  65                         continue;
  66
  67                 if (msk->bind_addr != MCTP_ADDR_ANY &&
  68                     msk->bind_addr != mh->dest)
  69                         continue;
  70
  71                 return msk;
  72         }
  73
  74         return NULL;
  75 }
  76
  77 static bool mctp_key_match(struct mctp_sk_key *key, mctp_eid_t local,
  78                            mctp_eid_t peer, u8 tag)
  79 {
  80         if (key->local_addr != local)
  81                 return false;
  82
  83         if (key->peer_addr != peer)
  84                 return false;
  85
  86         if (key->tag != tag)
  87                 return false;
  88
  89         return true;
  90 }
  91
  92 /* returns a key (with key->lock held, and refcounted), or NULL if no such
  93  * key exists.
  94  */
  95 static struct mctp_sk_key *mctp_lookup_key(struct net *net, struct sk_buff *skb,
  96                                            mctp_eid_t peer,
  97                                            unsigned long *irqflags)
  98         __acquires(&key->lock)
  99 {
 100         struct mctp_sk_key *key, *ret;
 101         unsigned long flags;
 102         struct mctp_hdr *mh;
 103         u8 tag;
 104
 105         mh = mctp_hdr(skb);
 106         tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 107
 108         ret = NULL;
 109         spin_lock_irqsave(&net->mctp.keys_lock, flags);
 110
 111         hlist_for_each_entry(key, &net->mctp.keys, hlist) {
 112                 if (!mctp_key_match(key, mh->dest, peer, tag))
 113                         continue;
 114
 115                 spin_lock(&key->lock);
 116                 if (key->valid) {
 117                         refcount_inc(&key->refs);
 118                         ret = key;
 119                         break;
 120                 }
 121                 spin_unlock(&key->lock);
 122         }
 123
 124         if (ret) {
 125                 spin_unlock(&net->mctp.keys_lock);
 126                 *irqflags = flags;
 127         } else {
 128                 spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 129         }
 130
 131         return ret;
 132 }
 133
 134 static struct mctp_sk_key *mctp_key_alloc(struct mctp_sock *msk,
 135                                           mctp_eid_t local, mctp_eid_t peer,
 136                                           u8 tag, gfp_t gfp)
 137 {
 138         struct mctp_sk_key *key;
 139
 140         key = kzalloc(sizeof(*key), gfp);
 141         if (!key)
 142                 return NULL;
 143
 144         key->peer_addr = peer;
 145         key->local_addr = local;
 146         key->tag = tag;
 147         key->sk = &msk->sk;
 148         key->valid = true;
 149         spin_lock_init(&key->lock);
 150         refcount_set(&key->refs, 1);
 151
 152         return key;
 153 }
 154
 155 void mctp_key_unref(struct mctp_sk_key *key)
 156 {
 157         unsigned long flags;
 158
 159         if (!refcount_dec_and_test(&key->refs))
 160                 return;
 161
 162         /* even though no refs exist here, the lock allows us to stay
 163          * consistent with the locking requirement of mctp_dev_release_key
 164          */
 165         spin_lock_irqsave(&key->lock, flags);
 166         mctp_dev_release_key(key->dev, key);
 167         spin_unlock_irqrestore(&key->lock, flags);
 168
 169         kfree(key);
 170 }
 171
 172 static int mctp_key_add(struct mctp_sk_key *key, struct mctp_sock *msk)
 173 {
 174         struct net *net = sock_net(&msk->sk);
 175         struct mctp_sk_key *tmp;
 176         unsigned long flags;
 177         int rc = 0;
 178
 179         spin_lock_irqsave(&net->mctp.keys_lock, flags);
 180
 181         hlist_for_each_entry(tmp, &net->mctp.keys, hlist) {
 182                 if (mctp_key_match(tmp, key->local_addr, key->peer_addr,
 183                                    key->tag)) {
 184                         spin_lock(&tmp->lock);
 185                         if (tmp->valid)
 186                                 rc = -EEXIST;
 187                         spin_unlock(&tmp->lock);
 188                         if (rc)
 189                                 break;
 190                 }
 191         }
 192
 193         if (!rc) {
 194                 refcount_inc(&key->refs);
 195                 key->expiry = jiffies + mctp_key_lifetime;
 196                 timer_reduce(&msk->key_expiry, key->expiry);
 197
 198                 hlist_add_head(&key->hlist, &net->mctp.keys);
 199                 hlist_add_head(&key->sklist, &msk->keys);
 200         }
 201
 202         spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 203
 204         return rc;
 205 }
 206
 207 /* We're done with the key; unset valid and remove from lists. There may still
 208  * be outstanding refs on the key though...
 209  */
 210 static void __mctp_key_unlock_drop(struct mctp_sk_key *key, struct net *net,
 211                                    unsigned long flags)
 212         __releases(&key->lock)
 213 {
 214         struct sk_buff *skb;
 215
 216         skb = key->reasm_head;
 217         key->reasm_head = NULL;
 218         key->reasm_dead = true;
 219         key->valid = false;
 220         mctp_dev_release_key(key->dev, key);
 221         spin_unlock_irqrestore(&key->lock, flags);
 222
 223         spin_lock_irqsave(&net->mctp.keys_lock, flags);
 224         hlist_del(&key->hlist);
 225         hlist_del(&key->sklist);
 226         spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 227
 228         /* one unref for the lists */
 229         mctp_key_unref(key);
 230
 231         /* and one for the local reference */
 232         mctp_key_unref(key);
 233
 234         kfree_skb(skb);
 235 }
 236
 237 #ifdef CONFIG_MCTP_FLOWS
 238 static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key)
 239 {
 240         struct mctp_flow *flow;
 241
 242         flow = skb_ext_add(skb, SKB_EXT_MCTP);
 243         if (!flow)
 244                 return;
 245
 246         refcount_inc(&key->refs);
 247         flow->key = key;
 248 }
 249
 250 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev)
 251 {
 252         struct mctp_sk_key *key;
 253         struct mctp_flow *flow;
 254
 255         flow = skb_ext_find(skb, SKB_EXT_MCTP);
 256         if (!flow)
 257                 return;
 258
 259         key = flow->key;
 260
 261         if (WARN_ON(key->dev && key->dev != dev))
 262                 return;
 263
 264         mctp_dev_set_key(dev, key);
 265 }
 266 #else
 267 static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key) {}
 268 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev) {}
 269 #endif
 270
 271 static int mctp_frag_queue(struct mctp_sk_key *key, struct sk_buff *skb)
 272 {
 273         struct mctp_hdr *hdr = mctp_hdr(skb);
 274         u8 exp_seq, this_seq;
 275
 276         this_seq = (hdr->flags_seq_tag >> MCTP_HDR_SEQ_SHIFT)
 277                 & MCTP_HDR_SEQ_MASK;
 278
 279         if (!key->reasm_head) {
 280                 key->reasm_head = skb;
 281                 key->reasm_tailp = &(skb_shinfo(skb)->frag_list);
 282                 key->last_seq = this_seq;
 283                 return 0;
 284         }
 285
 286         exp_seq = (key->last_seq + 1) & MCTP_HDR_SEQ_MASK;
 287
 288         if (this_seq != exp_seq)
 289                 return -EINVAL;
 290
 291         if (key->reasm_head->len + skb->len > mctp_message_maxlen)
 292                 return -EINVAL;
 293
 294         skb->next = NULL;
 295         skb->sk = NULL;
 296         *key->reasm_tailp = skb;
 297         key->reasm_tailp = &skb->next;
 298
 299         key->last_seq = this_seq;
 300
 301         key->reasm_head->data_len += skb->len;
 302         key->reasm_head->len += skb->len;
 303         key->reasm_head->truesize += skb->truesize;
 304
 305         return 0;
 306 }
 307
 308 static int mctp_route_input(struct mctp_route *route, struct sk_buff *skb)
 309 {
 310         struct net *net = dev_net(skb->dev);
 311         struct mctp_sk_key *key;
 312         struct mctp_sock *msk;
 313         struct mctp_hdr *mh;
 314         unsigned long f;
 315         u8 tag, flags;
 316         int rc;
 317
 318         msk = NULL;
 319         rc = -EINVAL;
 320
 321         /* we may be receiving a locally-routed packet; drop source sk
 322          * accounting
 323          */
 324         skb_orphan(skb);
 325
 326         /* ensure we have enough data for a header and a type */
 327         if (skb->len < sizeof(struct mctp_hdr) + 1)
 328                 goto out;
 329
 330         /* grab header, advance data ptr */
 331         mh = mctp_hdr(skb);
 332         skb_pull(skb, sizeof(struct mctp_hdr));
 333
 334         if (mh->ver != 1)
 335                 goto out;
 336
 337         flags = mh->flags_seq_tag & (MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM);
 338         tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 339
 340         rcu_read_lock();
 341
 342         /* lookup socket / reasm context, exactly matching (src,dest,tag).
 343          * we hold a ref on the key, and key->lock held.
 344          */
 345         key = mctp_lookup_key(net, skb, mh->src, &f);
 346
 347         if (flags & MCTP_HDR_FLAG_SOM) {
 348                 if (key) {
 349                         msk = container_of(key->sk, struct mctp_sock, sk);
 350                 } else {
 351                         /* first response to a broadcast? do a more general
 352                          * key lookup to find the socket, but don't use this
 353                          * key for reassembly - we'll create a more specific
 354                          * one for future packets if required (ie, !EOM).
 355                          */
 356                         key = mctp_lookup_key(net, skb, MCTP_ADDR_ANY, &f);
 357                         if (key) {
 358                                 msk = container_of(key->sk,
 359                                                    struct mctp_sock, sk);
 360                                 spin_unlock_irqrestore(&key->lock, f);
 361                                 mctp_key_unref(key);
 362                                 key = NULL;
 363                         }
 364                 }
 365
 366                 if (!key && !msk && (tag & MCTP_HDR_FLAG_TO))
 367                         msk = mctp_lookup_bind(net, skb);
 368
 369                 if (!msk) {
 370                         rc = -ENOENT;
 371                         goto out_unlock;
 372                 }
 373
 374                 /* single-packet message? deliver to socket, clean up any
 375                  * pending key.
 376                  */
 377                 if (flags & MCTP_HDR_FLAG_EOM) {
 378                         sock_queue_rcv_skb(&msk->sk, skb);
 379                         if (key) {
 380                                 /* we've hit a pending reassembly; not much we
 381                                  * can do but drop it
 382                                  */
 383                                 trace_mctp_key_release(key,
 384                                                        MCTP_TRACE_KEY_REPLIED);
 385                                 __mctp_key_unlock_drop(key, net, f);
 386                                 key = NULL;
 387                         }
 388                         rc = 0;
 389                         goto out_unlock;
 390                 }
 391
 392                 /* broadcast response or a bind() - create a key for further
 393                  * packets for this message
 394                  */
 395                 if (!key) {
 396                         key = mctp_key_alloc(msk, mh->dest, mh->src,
 397                                              tag, GFP_ATOMIC);
 398                         if (!key) {
 399                                 rc = -ENOMEM;
 400                                 goto out_unlock;
 401                         }
 402
 403                         /* we can queue without the key lock here, as the
 404                          * key isn't observable yet
 405                          */
 406                         mctp_frag_queue(key, skb);
 407
 408                         /* if the key_add fails, we've raced with another
 409                          * SOM packet with the same src, dest and tag. There's
 410                          * no way to distinguish future packets, so all we
 411                          * can do is drop; we'll free the skb on exit from
 412                          * this function.
 413                          */
 414                         rc = mctp_key_add(key, msk);
 415                         if (rc) {
 416                                 kfree(key);
 417                         } else {
 418                                 trace_mctp_key_acquire(key);
 419
 420                                 /* we don't need to release key->lock on exit */
 421                                 mctp_key_unref(key);
 422                         }
 423                         key = NULL;
 424
 425                 } else {
 426                         if (key->reasm_head || key->reasm_dead) {
 427                                 /* duplicate start? drop everything */
 428                                 trace_mctp_key_release(key,
 429                                                        MCTP_TRACE_KEY_INVALIDATED);
 430                                 __mctp_key_unlock_drop(key, net, f);
 431                                 rc = -EEXIST;
 432                                 key = NULL;
 433                         } else {
 434                                 rc = mctp_frag_queue(key, skb);
 435                         }
 436                 }
 437
 438         } else if (key) {
 439                 /* this packet continues a previous message; reassemble
 440                  * using the message-specific key
 441                  */
 442
 443                 /* we need to be continuing an existing reassembly... */
 444                 if (!key->reasm_head)
 445                         rc = -EINVAL;
 446                 else
 447                         rc = mctp_frag_queue(key, skb);
 448
 449                 /* end of message? deliver to socket, and we're done with
 450                  * the reassembly/response key
 451                  */
 452                 if (!rc && flags & MCTP_HDR_FLAG_EOM) {
 453                         sock_queue_rcv_skb(key->sk, key->reasm_head);
 454                         key->reasm_head = NULL;
 455                         trace_mctp_key_release(key, MCTP_TRACE_KEY_REPLIED);
 456                         __mctp_key_unlock_drop(key, net, f);
 457                         key = NULL;
 458                 }
 459
 460         } else {
 461                 /* not a start, no matching key */
 462                 rc = -ENOENT;
 463         }
 464
 465 out_unlock:
 466         rcu_read_unlock();
 467         if (key) {
 468                 spin_unlock_irqrestore(&key->lock, f);
 469                 mctp_key_unref(key);
 470         }
 471 out:
 472         if (rc)
 473                 kfree_skb(skb);
 474         return rc;
 475 }
 476
 477 static unsigned int mctp_route_mtu(struct mctp_route *rt)
 478 {
 479         return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
 480 }
 481
 482 static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
 483 {
 484         struct mctp_skb_cb *cb = mctp_cb(skb);
 485         struct mctp_hdr *hdr = mctp_hdr(skb);
 486         char daddr_buf[MAX_ADDR_LEN];
 487         char *daddr = NULL;
 488         unsigned int mtu;
 489         int rc;
 490
 491         skb->protocol = htons(ETH_P_MCTP);
 492
 493         mtu = READ_ONCE(skb->dev->mtu);
 494         if (skb->len > mtu) {
 495                 kfree_skb(skb);
 496                 return -EMSGSIZE;
 497         }
 498
 499         if (cb->ifindex) {
 500                 /* direct route; use the hwaddr we stashed in sendmsg */
 501                 daddr = cb->haddr;
 502         } else {
 503                 /* If lookup fails let the device handle daddr==NULL */
 504                 if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
 505                         daddr = daddr_buf;
 506         }
 507
 508         rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
 509                              daddr, skb->dev->dev_addr, skb->len);
 510         if (rc) {
 511                 kfree_skb(skb);
 512                 return -EHOSTUNREACH;
 513         }
 514
 515         mctp_flow_prepare_output(skb, route->dev);
 516
 517         rc = dev_queue_xmit(skb);
 518         if (rc)
 519                 rc = net_xmit_errno(rc);
 520
 521         return rc;
 522 }
 523
 524 /* route alloc/release */
 525 static void mctp_route_release(struct mctp_route *rt)
 526 {
 527         if (refcount_dec_and_test(&rt->refs)) {
 528                 mctp_dev_put(rt->dev);
 529                 kfree_rcu(rt, rcu);
 530         }
 531 }
 532
 533 /* returns a route with the refcount at 1 */
 534 static struct mctp_route *mctp_route_alloc(void)
 535 {
 536         struct mctp_route *rt;
 537
 538         rt = kzalloc(sizeof(*rt), GFP_KERNEL);
 539         if (!rt)
 540                 return NULL;
 541
 542         INIT_LIST_HEAD(&rt->list);
 543         refcount_set(&rt->refs, 1);
 544         rt->output = mctp_route_discard;
 545
 546         return rt;
 547 }
 548
 549 unsigned int mctp_default_net(struct net *net)
 550 {
 551         return READ_ONCE(net->mctp.default_net);
 552 }
 553
 554 int mctp_default_net_set(struct net *net, unsigned int index)
 555 {
 556         if (index == 0)
 557                 return -EINVAL;
 558         WRITE_ONCE(net->mctp.default_net, index);
 559         return 0;
 560 }
 561
 562 /* tag management */
 563 static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
 564                              struct mctp_sock *msk)
 565 {
 566         struct netns_mctp *mns = &net->mctp;
 567
 568         lockdep_assert_held(&mns->keys_lock);
 569
 570         key->expiry = jiffies + mctp_key_lifetime;
 571         timer_reduce(&msk->key_expiry, key->expiry);
 572
 573         /* we hold the net->key_lock here, allowing updates to both
 574          * then net and sk
 575          */
 576         hlist_add_head_rcu(&key->hlist, &mns->keys);
 577         hlist_add_head_rcu(&key->sklist, &msk->keys);
 578         refcount_inc(&key->refs);
 579 }
 580
 581 /* Allocate a locally-owned tag value for (saddr, daddr), and reserve
 582  * it for the socket msk
 583  */
 584 static struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
 585                                                 mctp_eid_t saddr,
 586                                                 mctp_eid_t daddr, u8 *tagp)
 587 {
 588         struct net *net = sock_net(&msk->sk);
 589         struct netns_mctp *mns = &net->mctp;
 590         struct mctp_sk_key *key, *tmp;
 591         unsigned long flags;
 592         u8 tagbits;
 593
 594         /* for NULL destination EIDs, we may get a response from any peer */
 595         if (daddr == MCTP_ADDR_NULL)
 596                 daddr = MCTP_ADDR_ANY;
 597
 598         /* be optimistic, alloc now */
 599         key = mctp_key_alloc(msk, saddr, daddr, 0, GFP_KERNEL);
 600         if (!key)
 601                 return ERR_PTR(-ENOMEM);
 602
 603         /* 8 possible tag values */
 604         tagbits = 0xff;
 605
 606         spin_lock_irqsave(&mns->keys_lock, flags);
 607
 608         /* Walk through the existing keys, looking for potential conflicting
 609          * tags. If we find a conflict, clear that bit from tagbits
 610          */
 611         hlist_for_each_entry(tmp, &mns->keys, hlist) {
 612                 /* We can check the lookup fields (*_addr, tag) without the
 613                  * lock held, they don't change over the lifetime of the key.
 614                  */
 615
 616                 /* if we don't own the tag, it can't conflict */
 617                 if (tmp->tag & MCTP_HDR_FLAG_TO)
 618                         continue;
 619
 620                 if (!((tmp->peer_addr == daddr ||
 621                        tmp->peer_addr == MCTP_ADDR_ANY) &&
 622                        tmp->local_addr == saddr))
 623                         continue;
 624
 625                 spin_lock(&tmp->lock);
 626                 /* key must still be valid. If we find a match, clear the
 627                  * potential tag value
 628                  */
 629                 if (tmp->valid)
 630                         tagbits &= ~(1 << tmp->tag);
 631                 spin_unlock(&tmp->lock);
 632
 633                 if (!tagbits)
 634                         break;
 635         }
 636
 637         if (tagbits) {
 638                 key->tag = __ffs(tagbits);
 639                 mctp_reserve_tag(net, key, msk);
 640                 trace_mctp_key_acquire(key);
 641
 642                 *tagp = key->tag;
 643         }
 644
 645         spin_unlock_irqrestore(&mns->keys_lock, flags);
 646
 647         if (!tagbits) {
 648                 kfree(key);
 649                 return ERR_PTR(-EBUSY);
 650         }
 651
 652         return key;
 653 }
 654
 655 /* routing lookups */
 656 static bool mctp_rt_match_eid(struct mctp_route *rt,
 657                               unsigned int net, mctp_eid_t eid)
 658 {
 659         return READ_ONCE(rt->dev->net) == net &&
 660                 rt->min <= eid && rt->max >= eid;
 661 }
 662
 663 /* compares match, used for duplicate prevention */
 664 static bool mctp_rt_compare_exact(struct mctp_route *rt1,
 665                                   struct mctp_route *rt2)
 666 {
 667         ASSERT_RTNL();
 668         return rt1->dev->net == rt2->dev->net &&
 669                 rt1->min == rt2->min &&
 670                 rt1->max == rt2->max;
 671 }
 672
 673 struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
 674                                      mctp_eid_t daddr)
 675 {
 676         struct mctp_route *tmp, *rt = NULL;
 677
 678         list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 679                 /* TODO: add metrics */
 680                 if (mctp_rt_match_eid(tmp, dnet, daddr)) {
 681                         if (refcount_inc_not_zero(&tmp->refs)) {
 682                                 rt = tmp;
 683                                 break;
 684                         }
 685                 }
 686         }
 687
 688         return rt;
 689 }
 690
 691 static struct mctp_route *mctp_route_lookup_null(struct net *net,
 692                                                  struct net_device *dev)
 693 {
 694         struct mctp_route *rt;
 695
 696         list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
 697                 if (rt->dev->dev == dev && rt->type == RTN_LOCAL &&
 698                     refcount_inc_not_zero(&rt->refs))
 699                         return rt;
 700         }
 701
 702         return NULL;
 703 }
 704
 705 static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
 706                                   unsigned int mtu, u8 tag)
 707 {
 708         const unsigned int hlen = sizeof(struct mctp_hdr);
 709         struct mctp_hdr *hdr, *hdr2;
 710         unsigned int pos, size;
 711         struct sk_buff *skb2;
 712         int rc;
 713         u8 seq;
 714
 715         hdr = mctp_hdr(skb);
 716         seq = 0;
 717         rc = 0;
 718
 719         if (mtu < hlen + 1) {
 720                 kfree_skb(skb);
 721                 return -EMSGSIZE;
 722         }
 723
 724         /* we've got the header */
 725         skb_pull(skb, hlen);
 726
 727         for (pos = 0; pos < skb->len;) {
 728                 /* size of message payload */
 729                 size = min(mtu - hlen, skb->len - pos);
 730
 731                 skb2 = alloc_skb(MCTP_HEADER_MAXLEN + hlen + size, GFP_KERNEL);
 732                 if (!skb2) {
 733                         rc = -ENOMEM;
 734                         break;
 735                 }
 736
 737                 /* generic skb copy */
 738                 skb2->protocol = skb->protocol;
 739                 skb2->priority = skb->priority;
 740                 skb2->dev = skb->dev;
 741                 memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
 742
 743                 if (skb->sk)
 744                         skb_set_owner_w(skb2, skb->sk);
 745
 746                 /* establish packet */
 747                 skb_reserve(skb2, MCTP_HEADER_MAXLEN);
 748                 skb_reset_network_header(skb2);
 749                 skb_put(skb2, hlen + size);
 750                 skb2->transport_header = skb2->network_header + hlen;
 751
 752                 /* copy header fields, calculate SOM/EOM flags & seq */
 753                 hdr2 = mctp_hdr(skb2);
 754                 hdr2->ver = hdr->ver;
 755                 hdr2->dest = hdr->dest;
 756                 hdr2->src = hdr->src;
 757                 hdr2->flags_seq_tag = tag &
 758                         (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 759
 760                 if (pos == 0)
 761                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
 762
 763                 if (pos + size == skb->len)
 764                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
 765
 766                 hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
 767
 768                 /* copy message payload */
 769                 skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
 770
 771                 /* do route */
 772                 rc = rt->output(rt, skb2);
 773                 if (rc)
 774                         break;
 775
 776                 seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
 777                 pos += size;
 778         }
 779
 780         consume_skb(skb);
 781         return rc;
 782 }
 783
 784 int mctp_local_output(struct sock *sk, struct mctp_route *rt,
 785                       struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
 786 {
 787         struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
 788         struct mctp_skb_cb *cb = mctp_cb(skb);
 789         struct mctp_route tmp_rt;
 790         struct mctp_sk_key *key;
 791         struct net_device *dev;
 792         struct mctp_hdr *hdr;
 793         unsigned long flags;
 794         unsigned int mtu;
 795         mctp_eid_t saddr;
 796         bool ext_rt;
 797         int rc;
 798         u8 tag;
 799
 800         rc = -ENODEV;
 801
 802         if (rt) {
 803                 ext_rt = false;
 804                 dev = NULL;
 805
 806                 if (WARN_ON(!rt->dev))
 807                         goto out_release;
 808
 809         } else if (cb->ifindex) {
 810                 ext_rt = true;
 811                 rt = &tmp_rt;
 812
 813                 rcu_read_lock();
 814                 dev = dev_get_by_index_rcu(sock_net(sk), cb->ifindex);
 815                 if (!dev) {
 816                         rcu_read_unlock();
 817                         return rc;
 818                 }
 819
 820                 rt->dev = __mctp_dev_get(dev);
 821                 rcu_read_unlock();
 822
 823                 if (!rt->dev)
 824                         goto out_release;
 825
 826                 /* establish temporary route - we set up enough to keep
 827                  * mctp_route_output happy
 828                  */
 829                 rt->output = mctp_route_output;
 830                 rt->mtu = 0;
 831
 832         } else {
 833                 return -EINVAL;
 834         }
 835
 836         spin_lock_irqsave(&rt->dev->addrs_lock, flags);
 837         if (rt->dev->num_addrs == 0) {
 838                 rc = -EHOSTUNREACH;
 839         } else {
 840                 /* use the outbound interface's first address as our source */
 841                 saddr = rt->dev->addrs[0];
 842                 rc = 0;
 843         }
 844         spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
 845
 846         if (rc)
 847                 goto out_release;
 848
 849         if (req_tag & MCTP_HDR_FLAG_TO) {
 850                 key = mctp_alloc_local_tag(msk, saddr, daddr, &tag);
 851                 if (IS_ERR(key)) {
 852                         rc = PTR_ERR(key);
 853                         goto out_release;
 854                 }
 855                 mctp_skb_set_flow(skb, key);
 856                 /* done with the key in this scope */
 857                 mctp_key_unref(key);
 858                 tag |= MCTP_HDR_FLAG_TO;
 859         } else {
 860                 key = NULL;
 861                 tag = req_tag;
 862         }
 863
 864         skb->protocol = htons(ETH_P_MCTP);
 865         skb->priority = 0;
 866         skb_reset_transport_header(skb);
 867         skb_push(skb, sizeof(struct mctp_hdr));
 868         skb_reset_network_header(skb);
 869         skb->dev = rt->dev->dev;
 870
 871         /* cb->net will have been set on initial ingress */
 872         cb->src = saddr;
 873
 874         /* set up common header fields */
 875         hdr = mctp_hdr(skb);
 876         hdr->ver = 1;
 877         hdr->dest = daddr;
 878         hdr->src = saddr;
 879
 880         mtu = mctp_route_mtu(rt);
 881
 882         if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
 883                 hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM |
 884                         MCTP_HDR_FLAG_EOM | tag;
 885                 rc = rt->output(rt, skb);
 886         } else {
 887                 rc = mctp_do_fragment_route(rt, skb, mtu, tag);
 888         }
 889
 890 out_release:
 891         if (!ext_rt)
 892                 mctp_route_release(rt);
 893
 894         dev_put(dev);
 895
 896         return rc;
 897
 898 }
 899
 900 /* route management */
 901 static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 902                           unsigned int daddr_extent, unsigned int mtu,
 903                           unsigned char type)
 904 {
 905         int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
 906         struct net *net = dev_net(mdev->dev);
 907         struct mctp_route *rt, *ert;
 908
 909         if (!mctp_address_ok(daddr_start))
 910                 return -EINVAL;
 911
 912         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 913                 return -EINVAL;
 914
 915         switch (type) {
 916         case RTN_LOCAL:
 917                 rtfn = mctp_route_input;
 918                 break;
 919         case RTN_UNICAST:
 920                 rtfn = mctp_route_output;
 921                 break;
 922         default:
 923                 return -EINVAL;
 924         }
 925
 926         rt = mctp_route_alloc();
 927         if (!rt)
 928                 return -ENOMEM;
 929
 930         rt->min = daddr_start;
 931         rt->max = daddr_start + daddr_extent;
 932         rt->mtu = mtu;
 933         rt->dev = mdev;
 934         mctp_dev_hold(rt->dev);
 935         rt->type = type;
 936         rt->output = rtfn;
 937
 938         ASSERT_RTNL();
 939         /* Prevent duplicate identical routes. */
 940         list_for_each_entry(ert, &net->mctp.routes, list) {
 941                 if (mctp_rt_compare_exact(rt, ert)) {
 942                         mctp_route_release(rt);
 943                         return -EEXIST;
 944                 }
 945         }
 946
 947         list_add_rcu(&rt->list, &net->mctp.routes);
 948
 949         return 0;
 950 }
 951
 952 static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 953                              unsigned int daddr_extent, unsigned char type)
 954 {
 955         struct net *net = dev_net(mdev->dev);
 956         struct mctp_route *rt, *tmp;
 957         mctp_eid_t daddr_end;
 958         bool dropped;
 959
 960         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 961                 return -EINVAL;
 962
 963         daddr_end = daddr_start + daddr_extent;
 964         dropped = false;
 965
 966         ASSERT_RTNL();
 967
 968         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
 969                 if (rt->dev == mdev &&
 970                     rt->min == daddr_start && rt->max == daddr_end &&
 971                     rt->type == type) {
 972                         list_del_rcu(&rt->list);
 973                         /* TODO: immediate RTM_DELROUTE */
 974                         mctp_route_release(rt);
 975                         dropped = true;
 976                 }
 977         }
 978
 979         return dropped ? 0 : -ENOENT;
 980 }
 981
 982 int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
 983 {
 984         return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
 985 }
 986
 987 int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
 988 {
 989         return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
 990 }
 991
 992 /* removes all entries for a given device */
 993 void mctp_route_remove_dev(struct mctp_dev *mdev)
 994 {
 995         struct net *net = dev_net(mdev->dev);
 996         struct mctp_route *rt, *tmp;
 997
 998         ASSERT_RTNL();
 999         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1000                 if (rt->dev == mdev) {
1001                         list_del_rcu(&rt->list);
1002                         /* TODO: immediate RTM_DELROUTE */
1003                         mctp_route_release(rt);
1004                 }
1005         }
1006 }
1007
1008 /* Incoming packet-handling */
1009
1010 static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
1011                                 struct packet_type *pt,
1012                                 struct net_device *orig_dev)
1013 {
1014         struct net *net = dev_net(dev);
1015         struct mctp_dev *mdev;
1016         struct mctp_skb_cb *cb;
1017         struct mctp_route *rt;
1018         struct mctp_hdr *mh;
1019
1020         rcu_read_lock();
1021         mdev = __mctp_dev_get(dev);
1022         rcu_read_unlock();
1023         if (!mdev) {
1024                 /* basic non-data sanity checks */
1025                 goto err_drop;
1026         }
1027
1028         if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
1029                 goto err_drop;
1030
1031         skb_reset_transport_header(skb);
1032         skb_reset_network_header(skb);
1033
1034         /* We have enough for a header; decode and route */
1035         mh = mctp_hdr(skb);
1036         if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
1037                 goto err_drop;
1038
1039         /* MCTP drivers must populate halen/haddr */
1040         if (dev->type == ARPHRD_MCTP) {
1041                 cb = mctp_cb(skb);
1042         } else {
1043                 cb = __mctp_cb(skb);
1044                 cb->halen = 0;
1045         }
1046         cb->net = READ_ONCE(mdev->net);
1047         cb->ifindex = dev->ifindex;
1048
1049         rt = mctp_route_lookup(net, cb->net, mh->dest);
1050
1051         /* NULL EID, but addressed to our physical address */
1052         if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
1053                 rt = mctp_route_lookup_null(net, dev);
1054
1055         if (!rt)
1056                 goto err_drop;
1057
1058         rt->output(rt, skb);
1059         mctp_route_release(rt);
1060
1061         return NET_RX_SUCCESS;
1062
1063 err_drop:
1064         kfree_skb(skb);
1065         return NET_RX_DROP;
1066 }
1067
1068 static struct packet_type mctp_packet_type = {
1069         .type = cpu_to_be16(ETH_P_MCTP),
1070         .func = mctp_pkttype_receive,
1071 };
1072
1073 /* netlink interface */
1074
1075 static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
1076         [RTA_DST]               = { .type = NLA_U8 },
1077         [RTA_METRICS]           = { .type = NLA_NESTED },
1078         [RTA_OIF]               = { .type = NLA_U32 },
1079 };
1080
1081 /* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
1082  * tb must hold RTA_MAX+1 elements.
1083  */
1084 static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
1085                               struct netlink_ext_ack *extack,
1086                               struct nlattr **tb, struct rtmsg **rtm,
1087                               struct mctp_dev **mdev, mctp_eid_t *daddr_start)
1088 {
1089         struct net *net = sock_net(skb->sk);
1090         struct net_device *dev;
1091         unsigned int ifindex;
1092         int rc;
1093
1094         rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
1095                          rta_mctp_policy, extack);
1096         if (rc < 0) {
1097                 NL_SET_ERR_MSG(extack, "incorrect format");
1098                 return rc;
1099         }
1100
1101         if (!tb[RTA_DST]) {
1102                 NL_SET_ERR_MSG(extack, "dst EID missing");
1103                 return -EINVAL;
1104         }
1105         *daddr_start = nla_get_u8(tb[RTA_DST]);
1106
1107         if (!tb[RTA_OIF]) {
1108                 NL_SET_ERR_MSG(extack, "ifindex missing");
1109                 return -EINVAL;
1110         }
1111         ifindex = nla_get_u32(tb[RTA_OIF]);
1112
1113         *rtm = nlmsg_data(nlh);
1114         if ((*rtm)->rtm_family != AF_MCTP) {
1115                 NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1116                 return -EINVAL;
1117         }
1118
1119         dev = __dev_get_by_index(net, ifindex);
1120         if (!dev) {
1121                 NL_SET_ERR_MSG(extack, "bad ifindex");
1122                 return -ENODEV;
1123         }
1124         *mdev = mctp_dev_get_rtnl(dev);
1125         if (!*mdev)
1126                 return -ENODEV;
1127
1128         if (dev->flags & IFF_LOOPBACK) {
1129                 NL_SET_ERR_MSG(extack, "no routes to loopback");
1130                 return -EINVAL;
1131         }
1132
1133         return 0;
1134 }
1135
1136 static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1137         [RTAX_MTU]              = { .type = NLA_U32 },
1138 };
1139
1140 static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1141                          struct netlink_ext_ack *extack)
1142 {
1143         struct nlattr *tb[RTA_MAX + 1];
1144         struct nlattr *tbx[RTAX_MAX + 1];
1145         mctp_eid_t daddr_start;
1146         struct mctp_dev *mdev;
1147         struct rtmsg *rtm;
1148         unsigned int mtu;
1149         int rc;
1150
1151         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1152                                 &rtm, &mdev, &daddr_start);
1153         if (rc < 0)
1154                 return rc;
1155
1156         if (rtm->rtm_type != RTN_UNICAST) {
1157                 NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1158                 return -EINVAL;
1159         }
1160
1161         mtu = 0;
1162         if (tb[RTA_METRICS]) {
1163                 rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1164                                       rta_metrics_policy, NULL);
1165                 if (rc < 0)
1166                         return rc;
1167                 if (tbx[RTAX_MTU])
1168                         mtu = nla_get_u32(tbx[RTAX_MTU]);
1169         }
1170
1171         if (rtm->rtm_type != RTN_UNICAST)
1172                 return -EINVAL;
1173
1174         rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1175                             rtm->rtm_type);
1176         return rc;
1177 }
1178
1179 static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1180                          struct netlink_ext_ack *extack)
1181 {
1182         struct nlattr *tb[RTA_MAX + 1];
1183         mctp_eid_t daddr_start;
1184         struct mctp_dev *mdev;
1185         struct rtmsg *rtm;
1186         int rc;
1187
1188         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1189                                 &rtm, &mdev, &daddr_start);
1190         if (rc < 0)
1191                 return rc;
1192
1193         /* we only have unicast routes */
1194         if (rtm->rtm_type != RTN_UNICAST)
1195                 return -EINVAL;
1196
1197         rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1198         return rc;
1199 }
1200
1201 static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1202                             u32 portid, u32 seq, int event, unsigned int flags)
1203 {
1204         struct nlmsghdr *nlh;
1205         struct rtmsg *hdr;
1206         void *metrics;
1207
1208         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1209         if (!nlh)
1210                 return -EMSGSIZE;
1211
1212         hdr = nlmsg_data(nlh);
1213         hdr->rtm_family = AF_MCTP;
1214
1215         /* we use the _len fields as a number of EIDs, rather than
1216          * a number of bits in the address
1217          */
1218         hdr->rtm_dst_len = rt->max - rt->min;
1219         hdr->rtm_src_len = 0;
1220         hdr->rtm_tos = 0;
1221         hdr->rtm_table = RT_TABLE_DEFAULT;
1222         hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1223         hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1224         hdr->rtm_type = rt->type;
1225
1226         if (nla_put_u8(skb, RTA_DST, rt->min))
1227                 goto cancel;
1228
1229         metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1230         if (!metrics)
1231                 goto cancel;
1232
1233         if (rt->mtu) {
1234                 if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1235                         goto cancel;
1236         }
1237
1238         nla_nest_end(skb, metrics);
1239
1240         if (rt->dev) {
1241                 if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1242                         goto cancel;
1243         }
1244
1245         /* TODO: conditional neighbour physaddr? */
1246
1247         nlmsg_end(skb, nlh);
1248
1249         return 0;
1250
1251 cancel:
1252         nlmsg_cancel(skb, nlh);
1253         return -EMSGSIZE;
1254 }
1255
1256 static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1257 {
1258         struct net *net = sock_net(skb->sk);
1259         struct mctp_route *rt;
1260         int s_idx, idx;
1261
1262         /* TODO: allow filtering on route data, possibly under
1263          * cb->strict_check
1264          */
1265
1266         /* TODO: change to struct overlay */
1267         s_idx = cb->args[0];
1268         idx = 0;
1269
1270         rcu_read_lock();
1271         list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1272                 if (idx++ < s_idx)
1273                         continue;
1274                 if (mctp_fill_rtinfo(skb, rt,
1275                                      NETLINK_CB(cb->skb).portid,
1276                                      cb->nlh->nlmsg_seq,
1277                                      RTM_NEWROUTE, NLM_F_MULTI) < 0)
1278                         break;
1279         }
1280
1281         rcu_read_unlock();
1282         cb->args[0] = idx;
1283
1284         return skb->len;
1285 }
1286
1287 /* net namespace implementation */
1288 static int __net_init mctp_routes_net_init(struct net *net)
1289 {
1290         struct netns_mctp *ns = &net->mctp;
1291
1292         INIT_LIST_HEAD(&ns->routes);
1293         INIT_HLIST_HEAD(&ns->binds);
1294         mutex_init(&ns->bind_lock);
1295         INIT_HLIST_HEAD(&ns->keys);
1296         spin_lock_init(&ns->keys_lock);
1297         WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1298         return 0;
1299 }
1300
1301 static void __net_exit mctp_routes_net_exit(struct net *net)
1302 {
1303         struct mctp_route *rt;
1304
1305         rcu_read_lock();
1306         list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1307                 mctp_route_release(rt);
1308         rcu_read_unlock();
1309 }
1310
1311 static struct pernet_operations mctp_net_ops = {
1312         .init = mctp_routes_net_init,
1313         .exit = mctp_routes_net_exit,
1314 };
1315
1316 int __init mctp_routes_init(void)
1317 {
1318         dev_add_pack(&mctp_packet_type);
1319
1320         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1321                              NULL, mctp_dump_rtinfo, 0);
1322         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1323                              mctp_newroute, NULL, 0);
1324         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1325                              mctp_delroute, NULL, 0);
1326
1327         return register_pernet_subsys(&mctp_net_ops);
1328 }
1329
1330 void __exit mctp_routes_exit(void)
1331 {
1332         unregister_pernet_subsys(&mctp_net_ops);
1333         rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1334         rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1335         rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1336         dev_remove_pack(&mctp_packet_type);
1337 }
1338
1339 #if IS_ENABLED(CONFIG_MCTP_TEST)
1340 #include "test/route-test.c"
1341 #endif