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
 418                         trace_mctp_key_acquire(key);
 419
 420                         /* we don't need to release key->lock on exit */
 421                         mctp_key_unref(key);
 422                         key = NULL;
 423
 424                 } else {
 425                         if (key->reasm_head || key->reasm_dead) {
 426                                 /* duplicate start? drop everything */
 427                                 trace_mctp_key_release(key,
 428                                                        MCTP_TRACE_KEY_INVALIDATED);
 429                                 __mctp_key_unlock_drop(key, net, f);
 430                                 rc = -EEXIST;
 431                                 key = NULL;
 432                         } else {
 433                                 rc = mctp_frag_queue(key, skb);
 434                         }
 435                 }
 436
 437         } else if (key) {
 438                 /* this packet continues a previous message; reassemble
 439                  * using the message-specific key
 440                  */
 441
 442                 /* we need to be continuing an existing reassembly... */
 443                 if (!key->reasm_head)
 444                         rc = -EINVAL;
 445                 else
 446                         rc = mctp_frag_queue(key, skb);
 447
 448                 /* end of message? deliver to socket, and we're done with
 449                  * the reassembly/response key
 450                  */
 451                 if (!rc && flags & MCTP_HDR_FLAG_EOM) {
 452                         sock_queue_rcv_skb(key->sk, key->reasm_head);
 453                         key->reasm_head = NULL;
 454                         trace_mctp_key_release(key, MCTP_TRACE_KEY_REPLIED);
 455                         __mctp_key_unlock_drop(key, net, f);
 456                         key = NULL;
 457                 }
 458
 459         } else {
 460                 /* not a start, no matching key */
 461                 rc = -ENOENT;
 462         }
 463
 464 out_unlock:
 465         rcu_read_unlock();
 466         if (key) {
 467                 spin_unlock_irqrestore(&key->lock, f);
 468                 mctp_key_unref(key);
 469         }
 470 out:
 471         if (rc)
 472                 kfree_skb(skb);
 473         return rc;
 474 }
 475
 476 static unsigned int mctp_route_mtu(struct mctp_route *rt)
 477 {
 478         return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
 479 }
 480
 481 static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
 482 {
 483         struct mctp_skb_cb *cb = mctp_cb(skb);
 484         struct mctp_hdr *hdr = mctp_hdr(skb);
 485         char daddr_buf[MAX_ADDR_LEN];
 486         char *daddr = NULL;
 487         unsigned int mtu;
 488         int rc;
 489
 490         skb->protocol = htons(ETH_P_MCTP);
 491
 492         mtu = READ_ONCE(skb->dev->mtu);
 493         if (skb->len > mtu) {
 494                 kfree_skb(skb);
 495                 return -EMSGSIZE;
 496         }
 497
 498         if (cb->ifindex) {
 499                 /* direct route; use the hwaddr we stashed in sendmsg */
 500                 daddr = cb->haddr;
 501         } else {
 502                 /* If lookup fails let the device handle daddr==NULL */
 503                 if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
 504                         daddr = daddr_buf;
 505         }
 506
 507         rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
 508                              daddr, skb->dev->dev_addr, skb->len);
 509         if (rc) {
 510                 kfree_skb(skb);
 511                 return -EHOSTUNREACH;
 512         }
 513
 514         mctp_flow_prepare_output(skb, route->dev);
 515
 516         rc = dev_queue_xmit(skb);
 517         if (rc)
 518                 rc = net_xmit_errno(rc);
 519
 520         return rc;
 521 }
 522
 523 /* route alloc/release */
 524 static void mctp_route_release(struct mctp_route *rt)
 525 {
 526         if (refcount_dec_and_test(&rt->refs)) {
 527                 mctp_dev_put(rt->dev);
 528                 kfree_rcu(rt, rcu);
 529         }
 530 }
 531
 532 /* returns a route with the refcount at 1 */
 533 static struct mctp_route *mctp_route_alloc(void)
 534 {
 535         struct mctp_route *rt;
 536
 537         rt = kzalloc(sizeof(*rt), GFP_KERNEL);
 538         if (!rt)
 539                 return NULL;
 540
 541         INIT_LIST_HEAD(&rt->list);
 542         refcount_set(&rt->refs, 1);
 543         rt->output = mctp_route_discard;
 544
 545         return rt;
 546 }
 547
 548 unsigned int mctp_default_net(struct net *net)
 549 {
 550         return READ_ONCE(net->mctp.default_net);
 551 }
 552
 553 int mctp_default_net_set(struct net *net, unsigned int index)
 554 {
 555         if (index == 0)
 556                 return -EINVAL;
 557         WRITE_ONCE(net->mctp.default_net, index);
 558         return 0;
 559 }
 560
 561 /* tag management */
 562 static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
 563                              struct mctp_sock *msk)
 564 {
 565         struct netns_mctp *mns = &net->mctp;
 566
 567         lockdep_assert_held(&mns->keys_lock);
 568
 569         key->expiry = jiffies + mctp_key_lifetime;
 570         timer_reduce(&msk->key_expiry, key->expiry);
 571
 572         /* we hold the net->key_lock here, allowing updates to both
 573          * then net and sk
 574          */
 575         hlist_add_head_rcu(&key->hlist, &mns->keys);
 576         hlist_add_head_rcu(&key->sklist, &msk->keys);
 577         refcount_inc(&key->refs);
 578 }
 579
 580 /* Allocate a locally-owned tag value for (saddr, daddr), and reserve
 581  * it for the socket msk
 582  */
 583 static struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
 584                                                 mctp_eid_t saddr,
 585                                                 mctp_eid_t daddr, u8 *tagp)
 586 {
 587         struct net *net = sock_net(&msk->sk);
 588         struct netns_mctp *mns = &net->mctp;
 589         struct mctp_sk_key *key, *tmp;
 590         unsigned long flags;
 591         u8 tagbits;
 592
 593         /* for NULL destination EIDs, we may get a response from any peer */
 594         if (daddr == MCTP_ADDR_NULL)
 595                 daddr = MCTP_ADDR_ANY;
 596
 597         /* be optimistic, alloc now */
 598         key = mctp_key_alloc(msk, saddr, daddr, 0, GFP_KERNEL);
 599         if (!key)
 600                 return ERR_PTR(-ENOMEM);
 601
 602         /* 8 possible tag values */
 603         tagbits = 0xff;
 604
 605         spin_lock_irqsave(&mns->keys_lock, flags);
 606
 607         /* Walk through the existing keys, looking for potential conflicting
 608          * tags. If we find a conflict, clear that bit from tagbits
 609          */
 610         hlist_for_each_entry(tmp, &mns->keys, hlist) {
 611                 /* We can check the lookup fields (*_addr, tag) without the
 612                  * lock held, they don't change over the lifetime of the key.
 613                  */
 614
 615                 /* if we don't own the tag, it can't conflict */
 616                 if (tmp->tag & MCTP_HDR_FLAG_TO)
 617                         continue;
 618
 619                 if (!((tmp->peer_addr == daddr ||
 620                        tmp->peer_addr == MCTP_ADDR_ANY) &&
 621                        tmp->local_addr == saddr))
 622                         continue;
 623
 624                 spin_lock(&tmp->lock);
 625                 /* key must still be valid. If we find a match, clear the
 626                  * potential tag value
 627                  */
 628                 if (tmp->valid)
 629                         tagbits &= ~(1 << tmp->tag);
 630                 spin_unlock(&tmp->lock);
 631
 632                 if (!tagbits)
 633                         break;
 634         }
 635
 636         if (tagbits) {
 637                 key->tag = __ffs(tagbits);
 638                 mctp_reserve_tag(net, key, msk);
 639                 trace_mctp_key_acquire(key);
 640
 641                 *tagp = key->tag;
 642         }
 643
 644         spin_unlock_irqrestore(&mns->keys_lock, flags);
 645
 646         if (!tagbits) {
 647                 kfree(key);
 648                 return ERR_PTR(-EBUSY);
 649         }
 650
 651         return key;
 652 }
 653
 654 /* routing lookups */
 655 static bool mctp_rt_match_eid(struct mctp_route *rt,
 656                               unsigned int net, mctp_eid_t eid)
 657 {
 658         return READ_ONCE(rt->dev->net) == net &&
 659                 rt->min <= eid && rt->max >= eid;
 660 }
 661
 662 /* compares match, used for duplicate prevention */
 663 static bool mctp_rt_compare_exact(struct mctp_route *rt1,
 664                                   struct mctp_route *rt2)
 665 {
 666         ASSERT_RTNL();
 667         return rt1->dev->net == rt2->dev->net &&
 668                 rt1->min == rt2->min &&
 669                 rt1->max == rt2->max;
 670 }
 671
 672 struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
 673                                      mctp_eid_t daddr)
 674 {
 675         struct mctp_route *tmp, *rt = NULL;
 676
 677         list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 678                 /* TODO: add metrics */
 679                 if (mctp_rt_match_eid(tmp, dnet, daddr)) {
 680                         if (refcount_inc_not_zero(&tmp->refs)) {
 681                                 rt = tmp;
 682                                 break;
 683                         }
 684                 }
 685         }
 686
 687         return rt;
 688 }
 689
 690 static struct mctp_route *mctp_route_lookup_null(struct net *net,
 691                                                  struct net_device *dev)
 692 {
 693         struct mctp_route *rt;
 694
 695         list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
 696                 if (rt->dev->dev == dev && rt->type == RTN_LOCAL &&
 697                     refcount_inc_not_zero(&rt->refs))
 698                         return rt;
 699         }
 700
 701         return NULL;
 702 }
 703
 704 static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
 705                                   unsigned int mtu, u8 tag)
 706 {
 707         const unsigned int hlen = sizeof(struct mctp_hdr);
 708         struct mctp_hdr *hdr, *hdr2;
 709         unsigned int pos, size;
 710         struct sk_buff *skb2;
 711         int rc;
 712         u8 seq;
 713
 714         hdr = mctp_hdr(skb);
 715         seq = 0;
 716         rc = 0;
 717
 718         if (mtu < hlen + 1) {
 719                 kfree_skb(skb);
 720                 return -EMSGSIZE;
 721         }
 722
 723         /* we've got the header */
 724         skb_pull(skb, hlen);
 725
 726         for (pos = 0; pos < skb->len;) {
 727                 /* size of message payload */
 728                 size = min(mtu - hlen, skb->len - pos);
 729
 730                 skb2 = alloc_skb(MCTP_HEADER_MAXLEN + hlen + size, GFP_KERNEL);
 731                 if (!skb2) {
 732                         rc = -ENOMEM;
 733                         break;
 734                 }
 735
 736                 /* generic skb copy */
 737                 skb2->protocol = skb->protocol;
 738                 skb2->priority = skb->priority;
 739                 skb2->dev = skb->dev;
 740                 memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
 741
 742                 if (skb->sk)
 743                         skb_set_owner_w(skb2, skb->sk);
 744
 745                 /* establish packet */
 746                 skb_reserve(skb2, MCTP_HEADER_MAXLEN);
 747                 skb_reset_network_header(skb2);
 748                 skb_put(skb2, hlen + size);
 749                 skb2->transport_header = skb2->network_header + hlen;
 750
 751                 /* copy header fields, calculate SOM/EOM flags & seq */
 752                 hdr2 = mctp_hdr(skb2);
 753                 hdr2->ver = hdr->ver;
 754                 hdr2->dest = hdr->dest;
 755                 hdr2->src = hdr->src;
 756                 hdr2->flags_seq_tag = tag &
 757                         (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 758
 759                 if (pos == 0)
 760                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
 761
 762                 if (pos + size == skb->len)
 763                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
 764
 765                 hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
 766
 767                 /* copy message payload */
 768                 skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
 769
 770                 /* do route */
 771                 rc = rt->output(rt, skb2);
 772                 if (rc)
 773                         break;
 774
 775                 seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
 776                 pos += size;
 777         }
 778
 779         consume_skb(skb);
 780         return rc;
 781 }
 782
 783 int mctp_local_output(struct sock *sk, struct mctp_route *rt,
 784                       struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
 785 {
 786         struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
 787         struct mctp_skb_cb *cb = mctp_cb(skb);
 788         struct mctp_route tmp_rt;
 789         struct mctp_sk_key *key;
 790         struct net_device *dev;
 791         struct mctp_hdr *hdr;
 792         unsigned long flags;
 793         unsigned int mtu;
 794         mctp_eid_t saddr;
 795         bool ext_rt;
 796         int rc;
 797         u8 tag;
 798
 799         rc = -ENODEV;
 800
 801         if (rt) {
 802                 ext_rt = false;
 803                 dev = NULL;
 804
 805                 if (WARN_ON(!rt->dev))
 806                         goto out_release;
 807
 808         } else if (cb->ifindex) {
 809                 ext_rt = true;
 810                 rt = &tmp_rt;
 811
 812                 rcu_read_lock();
 813                 dev = dev_get_by_index_rcu(sock_net(sk), cb->ifindex);
 814                 if (!dev) {
 815                         rcu_read_unlock();
 816                         return rc;
 817                 }
 818
 819                 rt->dev = __mctp_dev_get(dev);
 820                 rcu_read_unlock();
 821
 822                 if (!rt->dev)
 823                         goto out_release;
 824
 825                 /* establish temporary route - we set up enough to keep
 826                  * mctp_route_output happy
 827                  */
 828                 rt->output = mctp_route_output;
 829                 rt->mtu = 0;
 830
 831         } else {
 832                 return -EINVAL;
 833         }
 834
 835         spin_lock_irqsave(&rt->dev->addrs_lock, flags);
 836         if (rt->dev->num_addrs == 0) {
 837                 rc = -EHOSTUNREACH;
 838         } else {
 839                 /* use the outbound interface's first address as our source */
 840                 saddr = rt->dev->addrs[0];
 841                 rc = 0;
 842         }
 843         spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
 844
 845         if (rc)
 846                 goto out_release;
 847
 848         if (req_tag & MCTP_HDR_FLAG_TO) {
 849                 key = mctp_alloc_local_tag(msk, saddr, daddr, &tag);
 850                 if (IS_ERR(key)) {
 851                         rc = PTR_ERR(key);
 852                         goto out_release;
 853                 }
 854                 mctp_skb_set_flow(skb, key);
 855                 /* done with the key in this scope */
 856                 mctp_key_unref(key);
 857                 tag |= MCTP_HDR_FLAG_TO;
 858         } else {
 859                 key = NULL;
 860                 tag = req_tag;
 861         }
 862
 863         skb->protocol = htons(ETH_P_MCTP);
 864         skb->priority = 0;
 865         skb_reset_transport_header(skb);
 866         skb_push(skb, sizeof(struct mctp_hdr));
 867         skb_reset_network_header(skb);
 868         skb->dev = rt->dev->dev;
 869
 870         /* cb->net will have been set on initial ingress */
 871         cb->src = saddr;
 872
 873         /* set up common header fields */
 874         hdr = mctp_hdr(skb);
 875         hdr->ver = 1;
 876         hdr->dest = daddr;
 877         hdr->src = saddr;
 878
 879         mtu = mctp_route_mtu(rt);
 880
 881         if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
 882                 hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM |
 883                         MCTP_HDR_FLAG_EOM | tag;
 884                 rc = rt->output(rt, skb);
 885         } else {
 886                 rc = mctp_do_fragment_route(rt, skb, mtu, tag);
 887         }
 888
 889 out_release:
 890         if (!ext_rt)
 891                 mctp_route_release(rt);
 892
 893         dev_put(dev);
 894
 895         return rc;
 896
 897 }
 898
 899 /* route management */
 900 static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 901                           unsigned int daddr_extent, unsigned int mtu,
 902                           unsigned char type)
 903 {
 904         int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
 905         struct net *net = dev_net(mdev->dev);
 906         struct mctp_route *rt, *ert;
 907
 908         if (!mctp_address_ok(daddr_start))
 909                 return -EINVAL;
 910
 911         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 912                 return -EINVAL;
 913
 914         switch (type) {
 915         case RTN_LOCAL:
 916                 rtfn = mctp_route_input;
 917                 break;
 918         case RTN_UNICAST:
 919                 rtfn = mctp_route_output;
 920                 break;
 921         default:
 922                 return -EINVAL;
 923         }
 924
 925         rt = mctp_route_alloc();
 926         if (!rt)
 927                 return -ENOMEM;
 928
 929         rt->min = daddr_start;
 930         rt->max = daddr_start + daddr_extent;
 931         rt->mtu = mtu;
 932         rt->dev = mdev;
 933         mctp_dev_hold(rt->dev);
 934         rt->type = type;
 935         rt->output = rtfn;
 936
 937         ASSERT_RTNL();
 938         /* Prevent duplicate identical routes. */
 939         list_for_each_entry(ert, &net->mctp.routes, list) {
 940                 if (mctp_rt_compare_exact(rt, ert)) {
 941                         mctp_route_release(rt);
 942                         return -EEXIST;
 943                 }
 944         }
 945
 946         list_add_rcu(&rt->list, &net->mctp.routes);
 947
 948         return 0;
 949 }
 950
 951 static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 952                              unsigned int daddr_extent, unsigned char type)
 953 {
 954         struct net *net = dev_net(mdev->dev);
 955         struct mctp_route *rt, *tmp;
 956         mctp_eid_t daddr_end;
 957         bool dropped;
 958
 959         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 960                 return -EINVAL;
 961
 962         daddr_end = daddr_start + daddr_extent;
 963         dropped = false;
 964
 965         ASSERT_RTNL();
 966
 967         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
 968                 if (rt->dev == mdev &&
 969                     rt->min == daddr_start && rt->max == daddr_end &&
 970                     rt->type == type) {
 971                         list_del_rcu(&rt->list);
 972                         /* TODO: immediate RTM_DELROUTE */
 973                         mctp_route_release(rt);
 974                         dropped = true;
 975                 }
 976         }
 977
 978         return dropped ? 0 : -ENOENT;
 979 }
 980
 981 int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
 982 {
 983         return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
 984 }
 985
 986 int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
 987 {
 988         return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
 989 }
 990
 991 /* removes all entries for a given device */
 992 void mctp_route_remove_dev(struct mctp_dev *mdev)
 993 {
 994         struct net *net = dev_net(mdev->dev);
 995         struct mctp_route *rt, *tmp;
 996
 997         ASSERT_RTNL();
 998         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
 999                 if (rt->dev == mdev) {
1000                         list_del_rcu(&rt->list);
1001                         /* TODO: immediate RTM_DELROUTE */
1002                         mctp_route_release(rt);
1003                 }
1004         }
1005 }
1006
1007 /* Incoming packet-handling */
1008
1009 static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
1010                                 struct packet_type *pt,
1011                                 struct net_device *orig_dev)
1012 {
1013         struct net *net = dev_net(dev);
1014         struct mctp_dev *mdev;
1015         struct mctp_skb_cb *cb;
1016         struct mctp_route *rt;
1017         struct mctp_hdr *mh;
1018
1019         rcu_read_lock();
1020         mdev = __mctp_dev_get(dev);
1021         rcu_read_unlock();
1022         if (!mdev) {
1023                 /* basic non-data sanity checks */
1024                 goto err_drop;
1025         }
1026
1027         if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
1028                 goto err_drop;
1029
1030         skb_reset_transport_header(skb);
1031         skb_reset_network_header(skb);
1032
1033         /* We have enough for a header; decode and route */
1034         mh = mctp_hdr(skb);
1035         if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
1036                 goto err_drop;
1037
1038         /* MCTP drivers must populate halen/haddr */
1039         if (dev->type == ARPHRD_MCTP) {
1040                 cb = mctp_cb(skb);
1041         } else {
1042                 cb = __mctp_cb(skb);
1043                 cb->halen = 0;
1044         }
1045         cb->net = READ_ONCE(mdev->net);
1046         cb->ifindex = dev->ifindex;
1047
1048         rt = mctp_route_lookup(net, cb->net, mh->dest);
1049
1050         /* NULL EID, but addressed to our physical address */
1051         if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
1052                 rt = mctp_route_lookup_null(net, dev);
1053
1054         if (!rt)
1055                 goto err_drop;
1056
1057         rt->output(rt, skb);
1058         mctp_route_release(rt);
1059
1060         return NET_RX_SUCCESS;
1061
1062 err_drop:
1063         kfree_skb(skb);
1064         return NET_RX_DROP;
1065 }
1066
1067 static struct packet_type mctp_packet_type = {
1068         .type = cpu_to_be16(ETH_P_MCTP),
1069         .func = mctp_pkttype_receive,
1070 };
1071
1072 /* netlink interface */
1073
1074 static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
1075         [RTA_DST]               = { .type = NLA_U8 },
1076         [RTA_METRICS]           = { .type = NLA_NESTED },
1077         [RTA_OIF]               = { .type = NLA_U32 },
1078 };
1079
1080 /* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
1081  * tb must hold RTA_MAX+1 elements.
1082  */
1083 static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
1084                               struct netlink_ext_ack *extack,
1085                               struct nlattr **tb, struct rtmsg **rtm,
1086                               struct mctp_dev **mdev, mctp_eid_t *daddr_start)
1087 {
1088         struct net *net = sock_net(skb->sk);
1089         struct net_device *dev;
1090         unsigned int ifindex;
1091         int rc;
1092
1093         rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
1094                          rta_mctp_policy, extack);
1095         if (rc < 0) {
1096                 NL_SET_ERR_MSG(extack, "incorrect format");
1097                 return rc;
1098         }
1099
1100         if (!tb[RTA_DST]) {
1101                 NL_SET_ERR_MSG(extack, "dst EID missing");
1102                 return -EINVAL;
1103         }
1104         *daddr_start = nla_get_u8(tb[RTA_DST]);
1105
1106         if (!tb[RTA_OIF]) {
1107                 NL_SET_ERR_MSG(extack, "ifindex missing");
1108                 return -EINVAL;
1109         }
1110         ifindex = nla_get_u32(tb[RTA_OIF]);
1111
1112         *rtm = nlmsg_data(nlh);
1113         if ((*rtm)->rtm_family != AF_MCTP) {
1114                 NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1115                 return -EINVAL;
1116         }
1117
1118         dev = __dev_get_by_index(net, ifindex);
1119         if (!dev) {
1120                 NL_SET_ERR_MSG(extack, "bad ifindex");
1121                 return -ENODEV;
1122         }
1123         *mdev = mctp_dev_get_rtnl(dev);
1124         if (!*mdev)
1125                 return -ENODEV;
1126
1127         if (dev->flags & IFF_LOOPBACK) {
1128                 NL_SET_ERR_MSG(extack, "no routes to loopback");
1129                 return -EINVAL;
1130         }
1131
1132         return 0;
1133 }
1134
1135 static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1136         [RTAX_MTU]              = { .type = NLA_U32 },
1137 };
1138
1139 static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1140                          struct netlink_ext_ack *extack)
1141 {
1142         struct nlattr *tb[RTA_MAX + 1];
1143         struct nlattr *tbx[RTAX_MAX + 1];
1144         mctp_eid_t daddr_start;
1145         struct mctp_dev *mdev;
1146         struct rtmsg *rtm;
1147         unsigned int mtu;
1148         int rc;
1149
1150         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1151                                 &rtm, &mdev, &daddr_start);
1152         if (rc < 0)
1153                 return rc;
1154
1155         if (rtm->rtm_type != RTN_UNICAST) {
1156                 NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1157                 return -EINVAL;
1158         }
1159
1160         mtu = 0;
1161         if (tb[RTA_METRICS]) {
1162                 rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1163                                       rta_metrics_policy, NULL);
1164                 if (rc < 0)
1165                         return rc;
1166                 if (tbx[RTAX_MTU])
1167                         mtu = nla_get_u32(tbx[RTAX_MTU]);
1168         }
1169
1170         if (rtm->rtm_type != RTN_UNICAST)
1171                 return -EINVAL;
1172
1173         rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1174                             rtm->rtm_type);
1175         return rc;
1176 }
1177
1178 static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1179                          struct netlink_ext_ack *extack)
1180 {
1181         struct nlattr *tb[RTA_MAX + 1];
1182         mctp_eid_t daddr_start;
1183         struct mctp_dev *mdev;
1184         struct rtmsg *rtm;
1185         int rc;
1186
1187         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1188                                 &rtm, &mdev, &daddr_start);
1189         if (rc < 0)
1190                 return rc;
1191
1192         /* we only have unicast routes */
1193         if (rtm->rtm_type != RTN_UNICAST)
1194                 return -EINVAL;
1195
1196         rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1197         return rc;
1198 }
1199
1200 static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1201                             u32 portid, u32 seq, int event, unsigned int flags)
1202 {
1203         struct nlmsghdr *nlh;
1204         struct rtmsg *hdr;
1205         void *metrics;
1206
1207         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1208         if (!nlh)
1209                 return -EMSGSIZE;
1210
1211         hdr = nlmsg_data(nlh);
1212         hdr->rtm_family = AF_MCTP;
1213
1214         /* we use the _len fields as a number of EIDs, rather than
1215          * a number of bits in the address
1216          */
1217         hdr->rtm_dst_len = rt->max - rt->min;
1218         hdr->rtm_src_len = 0;
1219         hdr->rtm_tos = 0;
1220         hdr->rtm_table = RT_TABLE_DEFAULT;
1221         hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1222         hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1223         hdr->rtm_type = rt->type;
1224
1225         if (nla_put_u8(skb, RTA_DST, rt->min))
1226                 goto cancel;
1227
1228         metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1229         if (!metrics)
1230                 goto cancel;
1231
1232         if (rt->mtu) {
1233                 if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1234                         goto cancel;
1235         }
1236
1237         nla_nest_end(skb, metrics);
1238
1239         if (rt->dev) {
1240                 if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1241                         goto cancel;
1242         }
1243
1244         /* TODO: conditional neighbour physaddr? */
1245
1246         nlmsg_end(skb, nlh);
1247
1248         return 0;
1249
1250 cancel:
1251         nlmsg_cancel(skb, nlh);
1252         return -EMSGSIZE;
1253 }
1254
1255 static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1256 {
1257         struct net *net = sock_net(skb->sk);
1258         struct mctp_route *rt;
1259         int s_idx, idx;
1260
1261         /* TODO: allow filtering on route data, possibly under
1262          * cb->strict_check
1263          */
1264
1265         /* TODO: change to struct overlay */
1266         s_idx = cb->args[0];
1267         idx = 0;
1268
1269         rcu_read_lock();
1270         list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1271                 if (idx++ < s_idx)
1272                         continue;
1273                 if (mctp_fill_rtinfo(skb, rt,
1274                                      NETLINK_CB(cb->skb).portid,
1275                                      cb->nlh->nlmsg_seq,
1276                                      RTM_NEWROUTE, NLM_F_MULTI) < 0)
1277                         break;
1278         }
1279
1280         rcu_read_unlock();
1281         cb->args[0] = idx;
1282
1283         return skb->len;
1284 }
1285
1286 /* net namespace implementation */
1287 static int __net_init mctp_routes_net_init(struct net *net)
1288 {
1289         struct netns_mctp *ns = &net->mctp;
1290
1291         INIT_LIST_HEAD(&ns->routes);
1292         INIT_HLIST_HEAD(&ns->binds);
1293         mutex_init(&ns->bind_lock);
1294         INIT_HLIST_HEAD(&ns->keys);
1295         spin_lock_init(&ns->keys_lock);
1296         WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1297         return 0;
1298 }
1299
1300 static void __net_exit mctp_routes_net_exit(struct net *net)
1301 {
1302         struct mctp_route *rt;
1303
1304         rcu_read_lock();
1305         list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1306                 mctp_route_release(rt);
1307         rcu_read_unlock();
1308 }
1309
1310 static struct pernet_operations mctp_net_ops = {
1311         .init = mctp_routes_net_init,
1312         .exit = mctp_routes_net_exit,
1313 };
1314
1315 int __init mctp_routes_init(void)
1316 {
1317         dev_add_pack(&mctp_packet_type);
1318
1319         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1320                              NULL, mctp_dump_rtinfo, 0);
1321         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1322                              mctp_newroute, NULL, 0);
1323         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1324                              mctp_delroute, NULL, 0);
1325
1326         return register_pernet_subsys(&mctp_net_ops);
1327 }
1328
1329 void __exit mctp_routes_exit(void)
1330 {
1331         unregister_pernet_subsys(&mctp_net_ops);
1332         rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1333         rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1334         rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1335         dev_remove_pack(&mctp_packet_type);
1336 }
1337
1338 #if IS_ENABLED(CONFIG_MCTP_TEST)
1339 #include "test/route-test.c"
1340 #endif