net/decnet/dn_table.c

   1 /*
   2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
   3  *              operating system.  DECnet is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              DECnet Routing Forwarding Information Base (Routing Tables)
   7  *
   8  * Author:      Steve Whitehouse <SteveW@ACM.org>
   9  *              Mostly copied from the IPv4 routing code
  10  *
  11  *
  12  * Changes:
  13  *
  14  */
  15 #include <linux/string.h>
  16 #include <linux/net.h>
  17 #include <linux/socket.h>
  18 #include <linux/slab.h>
  19 #include <linux/sockios.h>
  20 #include <linux/init.h>
  21 #include <linux/skbuff.h>
  22 #include <linux/netlink.h>
  23 #include <linux/rtnetlink.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/timer.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/atomic.h>
  29 #include <asm/uaccess.h>
  30 #include <linux/route.h> /* RTF_xxx */
  31 #include <net/neighbour.h>
  32 #include <net/netlink.h>
  33 #include <net/dst.h>
  34 #include <net/flow.h>
  35 #include <net/fib_rules.h>
  36 #include <net/dn.h>
  37 #include <net/dn_route.h>
  38 #include <net/dn_fib.h>
  39 #include <net/dn_neigh.h>
  40 #include <net/dn_dev.h>
  41
  42 struct dn_zone
  43 {
  44         struct dn_zone          *dz_next;
  45         struct dn_fib_node      **dz_hash;
  46         int                     dz_nent;
  47         int                     dz_divisor;
  48         u32                     dz_hashmask;
  49 #define DZ_HASHMASK(dz) ((dz)->dz_hashmask)
  50         int                     dz_order;
  51         __le16                  dz_mask;
  52 #define DZ_MASK(dz)     ((dz)->dz_mask)
  53 };
  54
  55 struct dn_hash
  56 {
  57         struct dn_zone  *dh_zones[17];
  58         struct dn_zone  *dh_zone_list;
  59 };
  60
  61 #define dz_key_0(key)           ((key).datum = 0)
  62
  63 #define for_nexthops(fi) { int nhsel; const struct dn_fib_nh *nh;\
  64         for(nhsel = 0, nh = (fi)->fib_nh; nhsel < (fi)->fib_nhs; nh++, nhsel++)
  65
  66 #define endfor_nexthops(fi) }
  67
  68 #define DN_MAX_DIVISOR 1024
  69 #define DN_S_ZOMBIE 1
  70 #define DN_S_ACCESSED 2
  71
  72 #define DN_FIB_SCAN(f, fp) \
  73 for( ; ((f) = *(fp)) != NULL; (fp) = &(f)->fn_next)
  74
  75 #define DN_FIB_SCAN_KEY(f, fp, key) \
  76 for( ; ((f) = *(fp)) != NULL && dn_key_eq((f)->fn_key, (key)); (fp) = &(f)->fn_next)
  77
  78 #define RT_TABLE_MIN 1
  79 #define DN_FIB_TABLE_HASHSZ 256
  80 static struct hlist_head dn_fib_table_hash[DN_FIB_TABLE_HASHSZ];
  81 static DEFINE_RWLOCK(dn_fib_tables_lock);
  82
  83 static struct kmem_cache *dn_hash_kmem __read_mostly;
  84 static int dn_fib_hash_zombies;
  85
  86 static inline dn_fib_idx_t dn_hash(dn_fib_key_t key, struct dn_zone *dz)
  87 {
  88         u16 h = le16_to_cpu(key.datum)>>(16 - dz->dz_order);
  89         h ^= (h >> 10);
  90         h ^= (h >> 6);
  91         h &= DZ_HASHMASK(dz);
  92         return *(dn_fib_idx_t *)&h;
  93 }
  94
  95 static inline dn_fib_key_t dz_key(__le16 dst, struct dn_zone *dz)
  96 {
  97         dn_fib_key_t k;
  98         k.datum = dst & DZ_MASK(dz);
  99         return k;
 100 }
 101
 102 static inline struct dn_fib_node **dn_chain_p(dn_fib_key_t key, struct dn_zone *dz)
 103 {
 104         return &dz->dz_hash[dn_hash(key, dz).datum];
 105 }
 106
 107 static inline struct dn_fib_node *dz_chain(dn_fib_key_t key, struct dn_zone *dz)
 108 {
 109         return dz->dz_hash[dn_hash(key, dz).datum];
 110 }
 111
 112 static inline int dn_key_eq(dn_fib_key_t a, dn_fib_key_t b)
 113 {
 114         return a.datum == b.datum;
 115 }
 116
 117 static inline int dn_key_leq(dn_fib_key_t a, dn_fib_key_t b)
 118 {
 119         return a.datum <= b.datum;
 120 }
 121
 122 static inline void dn_rebuild_zone(struct dn_zone *dz,
 123                                    struct dn_fib_node **old_ht,
 124                                    int old_divisor)
 125 {
 126         struct dn_fib_node *f, **fp, *next;
 127         int i;
 128
 129         for(i = 0; i < old_divisor; i++) {
 130                 for(f = old_ht[i]; f; f = next) {
 131                         next = f->fn_next;
 132                         for(fp = dn_chain_p(f->fn_key, dz);
 133                                 *fp && dn_key_leq((*fp)->fn_key, f->fn_key);
 134                                 fp = &(*fp)->fn_next)
 135                                 /* NOTHING */;
 136                         f->fn_next = *fp;
 137                         *fp = f;
 138                 }
 139         }
 140 }
 141
 142 static void dn_rehash_zone(struct dn_zone *dz)
 143 {
 144         struct dn_fib_node **ht, **old_ht;
 145         int old_divisor, new_divisor;
 146         u32 new_hashmask;
 147
 148         old_divisor = dz->dz_divisor;
 149
 150         switch (old_divisor) {
 151         case 16:
 152                 new_divisor = 256;
 153                 new_hashmask = 0xFF;
 154                 break;
 155         default:
 156                 printk(KERN_DEBUG "DECnet: dn_rehash_zone: BUG! %d\n",
 157                        old_divisor);
 158         case 256:
 159                 new_divisor = 1024;
 160                 new_hashmask = 0x3FF;
 161                 break;
 162         }
 163
 164         ht = kcalloc(new_divisor, sizeof(struct dn_fib_node*), GFP_KERNEL);
 165         if (ht == NULL)
 166                 return;
 167
 168         write_lock_bh(&dn_fib_tables_lock);
 169         old_ht = dz->dz_hash;
 170         dz->dz_hash = ht;
 171         dz->dz_hashmask = new_hashmask;
 172         dz->dz_divisor = new_divisor;
 173         dn_rebuild_zone(dz, old_ht, old_divisor);
 174         write_unlock_bh(&dn_fib_tables_lock);
 175         kfree(old_ht);
 176 }
 177
 178 static void dn_free_node(struct dn_fib_node *f)
 179 {
 180         dn_fib_release_info(DN_FIB_INFO(f));
 181         kmem_cache_free(dn_hash_kmem, f);
 182 }
 183
 184
 185 static struct dn_zone *dn_new_zone(struct dn_hash *table, int z)
 186 {
 187         int i;
 188         struct dn_zone *dz = kzalloc(sizeof(struct dn_zone), GFP_KERNEL);
 189         if (!dz)
 190                 return NULL;
 191
 192         if (z) {
 193                 dz->dz_divisor = 16;
 194                 dz->dz_hashmask = 0x0F;
 195         } else {
 196                 dz->dz_divisor = 1;
 197                 dz->dz_hashmask = 0;
 198         }
 199
 200         dz->dz_hash = kcalloc(dz->dz_divisor, sizeof(struct dn_fib_node *), GFP_KERNEL);
 201         if (!dz->dz_hash) {
 202                 kfree(dz);
 203                 return NULL;
 204         }
 205
 206         dz->dz_order = z;
 207         dz->dz_mask = dnet_make_mask(z);
 208
 209         for(i = z + 1; i <= 16; i++)
 210                 if (table->dh_zones[i])
 211                         break;
 212
 213         write_lock_bh(&dn_fib_tables_lock);
 214         if (i>16) {
 215                 dz->dz_next = table->dh_zone_list;
 216                 table->dh_zone_list = dz;
 217         } else {
 218                 dz->dz_next = table->dh_zones[i]->dz_next;
 219                 table->dh_zones[i]->dz_next = dz;
 220         }
 221         table->dh_zones[z] = dz;
 222         write_unlock_bh(&dn_fib_tables_lock);
 223         return dz;
 224 }
 225
 226
 227 static int dn_fib_nh_match(struct rtmsg *r, struct nlmsghdr *nlh, struct dn_kern_rta *rta, struct dn_fib_info *fi)
 228 {
 229         struct rtnexthop *nhp;
 230         int nhlen;
 231
 232         if (rta->rta_priority && *rta->rta_priority != fi->fib_priority)
 233                 return 1;
 234
 235         if (rta->rta_oif || rta->rta_gw) {
 236                 if ((!rta->rta_oif || *rta->rta_oif == fi->fib_nh->nh_oif) &&
 237                     (!rta->rta_gw  || memcmp(rta->rta_gw, &fi->fib_nh->nh_gw, 2) == 0))
 238                         return 0;
 239                 return 1;
 240         }
 241
 242         if (rta->rta_mp == NULL)
 243                 return 0;
 244
 245         nhp = RTA_DATA(rta->rta_mp);
 246         nhlen = RTA_PAYLOAD(rta->rta_mp);
 247
 248         for_nexthops(fi) {
 249                 int attrlen = nhlen - sizeof(struct rtnexthop);
 250                 __le16 gw;
 251
 252                 if (attrlen < 0 || (nhlen -= nhp->rtnh_len) < 0)
 253                         return -EINVAL;
 254                 if (nhp->rtnh_ifindex && nhp->rtnh_ifindex != nh->nh_oif)
 255                         return 1;
 256                 if (attrlen) {
 257                         gw = dn_fib_get_attr16(RTNH_DATA(nhp), attrlen, RTA_GATEWAY);
 258
 259                         if (gw && gw != nh->nh_gw)
 260                                 return 1;
 261                 }
 262                 nhp = RTNH_NEXT(nhp);
 263         } endfor_nexthops(fi);
 264
 265         return 0;
 266 }
 267
 268 static inline size_t dn_fib_nlmsg_size(struct dn_fib_info *fi)
 269 {
 270         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 271                          + nla_total_size(4) /* RTA_TABLE */
 272                          + nla_total_size(2) /* RTA_DST */
 273                          + nla_total_size(4); /* RTA_PRIORITY */
 274
 275         /* space for nested metrics */
 276         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 277
 278         if (fi->fib_nhs) {
 279                 /* Also handles the special case fib_nhs == 1 */
 280
 281                 /* each nexthop is packed in an attribute */
 282                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 283
 284                 /* may contain a gateway attribute */
 285                 nhsize += nla_total_size(4);
 286
 287                 /* all nexthops are packed in a nested attribute */
 288                 payload += nla_total_size(fi->fib_nhs * nhsize);
 289         }
 290
 291         return payload;
 292 }
 293
 294 static int dn_fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 295                         u32 tb_id, u8 type, u8 scope, void *dst, int dst_len,
 296                         struct dn_fib_info *fi, unsigned int flags)
 297 {
 298         struct rtmsg *rtm;
 299         struct nlmsghdr *nlh;
 300         unsigned char *b = skb_tail_pointer(skb);
 301
 302         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*rtm), flags);
 303         rtm = NLMSG_DATA(nlh);
 304         rtm->rtm_family = AF_DECnet;
 305         rtm->rtm_dst_len = dst_len;
 306         rtm->rtm_src_len = 0;
 307         rtm->rtm_tos = 0;
 308         rtm->rtm_table = tb_id;
 309         RTA_PUT_U32(skb, RTA_TABLE, tb_id);
 310         rtm->rtm_flags = fi->fib_flags;
 311         rtm->rtm_scope = scope;
 312         rtm->rtm_type  = type;
 313         if (rtm->rtm_dst_len)
 314                 RTA_PUT(skb, RTA_DST, 2, dst);
 315         rtm->rtm_protocol = fi->fib_protocol;
 316         if (fi->fib_priority)
 317                 RTA_PUT(skb, RTA_PRIORITY, 4, &fi->fib_priority);
 318         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
 319                 goto rtattr_failure;
 320         if (fi->fib_nhs == 1) {
 321                 if (fi->fib_nh->nh_gw)
 322                         RTA_PUT(skb, RTA_GATEWAY, 2, &fi->fib_nh->nh_gw);
 323                 if (fi->fib_nh->nh_oif)
 324                         RTA_PUT(skb, RTA_OIF, sizeof(int), &fi->fib_nh->nh_oif);
 325         }
 326         if (fi->fib_nhs > 1) {
 327                 struct rtnexthop *nhp;
 328                 struct rtattr *mp_head;
 329                 if (skb_tailroom(skb) <= RTA_SPACE(0))
 330                         goto rtattr_failure;
 331                 mp_head = (struct rtattr *)skb_put(skb, RTA_SPACE(0));
 332
 333                 for_nexthops(fi) {
 334                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
 335                                 goto rtattr_failure;
 336                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
 337                         nhp->rtnh_flags = nh->nh_flags & 0xFF;
 338                         nhp->rtnh_hops = nh->nh_weight - 1;
 339                         nhp->rtnh_ifindex = nh->nh_oif;
 340                         if (nh->nh_gw)
 341                                 RTA_PUT(skb, RTA_GATEWAY, 2, &nh->nh_gw);
 342                         nhp->rtnh_len = skb_tail_pointer(skb) - (unsigned char *)nhp;
 343                 } endfor_nexthops(fi);
 344                 mp_head->rta_type = RTA_MULTIPATH;
 345                 mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
 346         }
 347
 348         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
 349         return skb->len;
 350
 351
 352 nlmsg_failure:
 353 rtattr_failure:
 354         nlmsg_trim(skb, b);
 355         return -EMSGSIZE;
 356 }
 357
 358
 359 static void dn_rtmsg_fib(int event, struct dn_fib_node *f, int z, u32 tb_id,
 360                         struct nlmsghdr *nlh, struct netlink_skb_parms *req)
 361 {
 362         struct sk_buff *skb;
 363         u32 pid = req ? req->pid : 0;
 364         int err = -ENOBUFS;
 365
 366         skb = nlmsg_new(dn_fib_nlmsg_size(DN_FIB_INFO(f)), GFP_KERNEL);
 367         if (skb == NULL)
 368                 goto errout;
 369
 370         err = dn_fib_dump_info(skb, pid, nlh->nlmsg_seq, event, tb_id,
 371                                f->fn_type, f->fn_scope, &f->fn_key, z,
 372                                DN_FIB_INFO(f), 0);
 373         if (err < 0) {
 374                 /* -EMSGSIZE implies BUG in dn_fib_nlmsg_size() */
 375                 WARN_ON(err == -EMSGSIZE);
 376                 kfree_skb(skb);
 377                 goto errout;
 378         }
 379         rtnl_notify(skb, &init_net, pid, RTNLGRP_DECnet_ROUTE, nlh, GFP_KERNEL);
 380         return;
 381 errout:
 382         if (err < 0)
 383                 rtnl_set_sk_err(&init_net, RTNLGRP_DECnet_ROUTE, err);
 384 }
 385
 386 static __inline__ int dn_hash_dump_bucket(struct sk_buff *skb,
 387                                 struct netlink_callback *cb,
 388                                 struct dn_fib_table *tb,
 389                                 struct dn_zone *dz,
 390                                 struct dn_fib_node *f)
 391 {
 392         int i, s_i;
 393
 394         s_i = cb->args[4];
 395         for(i = 0; f; i++, f = f->fn_next) {
 396                 if (i < s_i)
 397                         continue;
 398                 if (f->fn_state & DN_S_ZOMBIE)
 399                         continue;
 400                 if (dn_fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
 401                                 cb->nlh->nlmsg_seq,
 402                                 RTM_NEWROUTE,
 403                                 tb->n,
 404                                 (f->fn_state & DN_S_ZOMBIE) ? 0 : f->fn_type,
 405                                 f->fn_scope, &f->fn_key, dz->dz_order,
 406                                 f->fn_info, NLM_F_MULTI) < 0) {
 407                         cb->args[4] = i;
 408                         return -1;
 409                 }
 410         }
 411         cb->args[4] = i;
 412         return skb->len;
 413 }
 414
 415 static __inline__ int dn_hash_dump_zone(struct sk_buff *skb,
 416                                 struct netlink_callback *cb,
 417                                 struct dn_fib_table *tb,
 418                                 struct dn_zone *dz)
 419 {
 420         int h, s_h;
 421
 422         s_h = cb->args[3];
 423         for(h = 0; h < dz->dz_divisor; h++) {
 424                 if (h < s_h)
 425                         continue;
 426                 if (h > s_h)
 427                         memset(&cb->args[4], 0, sizeof(cb->args) - 4*sizeof(cb->args[0]));
 428                 if (dz->dz_hash == NULL || dz->dz_hash[h] == NULL)
 429                         continue;
 430                 if (dn_hash_dump_bucket(skb, cb, tb, dz, dz->dz_hash[h]) < 0) {
 431                         cb->args[3] = h;
 432                         return -1;
 433                 }
 434         }
 435         cb->args[3] = h;
 436         return skb->len;
 437 }
 438
 439 static int dn_fib_table_dump(struct dn_fib_table *tb, struct sk_buff *skb,
 440                                 struct netlink_callback *cb)
 441 {
 442         int m, s_m;
 443         struct dn_zone *dz;
 444         struct dn_hash *table = (struct dn_hash *)tb->data;
 445
 446         s_m = cb->args[2];
 447         read_lock(&dn_fib_tables_lock);
 448         for(dz = table->dh_zone_list, m = 0; dz; dz = dz->dz_next, m++) {
 449                 if (m < s_m)
 450                         continue;
 451                 if (m > s_m)
 452                         memset(&cb->args[3], 0, sizeof(cb->args) - 3*sizeof(cb->args[0]));
 453
 454                 if (dn_hash_dump_zone(skb, cb, tb, dz) < 0) {
 455                         cb->args[2] = m;
 456                         read_unlock(&dn_fib_tables_lock);
 457                         return -1;
 458                 }
 459         }
 460         read_unlock(&dn_fib_tables_lock);
 461         cb->args[2] = m;
 462
 463         return skb->len;
 464 }
 465
 466 int dn_fib_dump(struct sk_buff *skb, struct netlink_callback *cb)
 467 {
 468         struct net *net = sock_net(skb->sk);
 469         unsigned int h, s_h;
 470         unsigned int e = 0, s_e;
 471         struct dn_fib_table *tb;
 472         struct hlist_node *node;
 473         int dumped = 0;
 474
 475         if (!net_eq(net, &init_net))
 476                 return 0;
 477
 478         if (NLMSG_PAYLOAD(cb->nlh, 0) >= sizeof(struct rtmsg) &&
 479                 ((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED)
 480                         return dn_cache_dump(skb, cb);
 481
 482         s_h = cb->args[0];
 483         s_e = cb->args[1];
 484
 485         for (h = s_h; h < DN_FIB_TABLE_HASHSZ; h++, s_h = 0) {
 486                 e = 0;
 487                 hlist_for_each_entry(tb, node, &dn_fib_table_hash[h], hlist) {
 488                         if (e < s_e)
 489                                 goto next;
 490                         if (dumped)
 491                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 492                                                  2 * sizeof(cb->args[0]));
 493                         if (tb->dump(tb, skb, cb) < 0)
 494                                 goto out;
 495                         dumped = 1;
 496 next:
 497                         e++;
 498                 }
 499         }
 500 out:
 501         cb->args[1] = e;
 502         cb->args[0] = h;
 503
 504         return skb->len;
 505 }
 506
 507 static int dn_fib_table_insert(struct dn_fib_table *tb, struct rtmsg *r, struct dn_kern_rta *rta, struct nlmsghdr *n, struct netlink_skb_parms *req)
 508 {
 509         struct dn_hash *table = (struct dn_hash *)tb->data;
 510         struct dn_fib_node *new_f, *f, **fp, **del_fp;
 511         struct dn_zone *dz;
 512         struct dn_fib_info *fi;
 513         int z = r->rtm_dst_len;
 514         int type = r->rtm_type;
 515         dn_fib_key_t key;
 516         int err;
 517
 518         if (z > 16)
 519                 return -EINVAL;
 520
 521         dz = table->dh_zones[z];
 522         if (!dz && !(dz = dn_new_zone(table, z)))
 523                 return -ENOBUFS;
 524
 525         dz_key_0(key);
 526         if (rta->rta_dst) {
 527                 __le16 dst;
 528                 memcpy(&dst, rta->rta_dst, 2);
 529                 if (dst & ~DZ_MASK(dz))
 530                         return -EINVAL;
 531                 key = dz_key(dst, dz);
 532         }
 533
 534         if ((fi = dn_fib_create_info(r, rta, n, &err)) == NULL)
 535                 return err;
 536
 537         if (dz->dz_nent > (dz->dz_divisor << 2) &&
 538                         dz->dz_divisor > DN_MAX_DIVISOR &&
 539                         (z==16 || (1<<z) > dz->dz_divisor))
 540                 dn_rehash_zone(dz);
 541
 542         fp = dn_chain_p(key, dz);
 543
 544         DN_FIB_SCAN(f, fp) {
 545                 if (dn_key_leq(key, f->fn_key))
 546                         break;
 547         }
 548
 549         del_fp = NULL;
 550
 551         if (f && (f->fn_state & DN_S_ZOMBIE) &&
 552                         dn_key_eq(f->fn_key, key)) {
 553                 del_fp = fp;
 554                 fp = &f->fn_next;
 555                 f = *fp;
 556                 goto create;
 557         }
 558
 559         DN_FIB_SCAN_KEY(f, fp, key) {
 560                 if (fi->fib_priority <= DN_FIB_INFO(f)->fib_priority)
 561                         break;
 562         }
 563
 564         if (f && dn_key_eq(f->fn_key, key) &&
 565                         fi->fib_priority == DN_FIB_INFO(f)->fib_priority) {
 566                 struct dn_fib_node **ins_fp;
 567
 568                 err = -EEXIST;
 569                 if (n->nlmsg_flags & NLM_F_EXCL)
 570                         goto out;
 571
 572                 if (n->nlmsg_flags & NLM_F_REPLACE) {
 573                         del_fp = fp;
 574                         fp = &f->fn_next;
 575                         f = *fp;
 576                         goto replace;
 577                 }
 578
 579                 ins_fp = fp;
 580                 err = -EEXIST;
 581
 582                 DN_FIB_SCAN_KEY(f, fp, key) {
 583                         if (fi->fib_priority != DN_FIB_INFO(f)->fib_priority)
 584                                 break;
 585                         if (f->fn_type == type &&
 586                             f->fn_scope == r->rtm_scope &&
 587                             DN_FIB_INFO(f) == fi)
 588                                 goto out;
 589                 }
 590
 591                 if (!(n->nlmsg_flags & NLM_F_APPEND)) {
 592                         fp = ins_fp;
 593                         f = *fp;
 594                 }
 595         }
 596
 597 create:
 598         err = -ENOENT;
 599         if (!(n->nlmsg_flags & NLM_F_CREATE))
 600                 goto out;
 601
 602 replace:
 603         err = -ENOBUFS;
 604         new_f = kmem_cache_zalloc(dn_hash_kmem, GFP_KERNEL);
 605         if (new_f == NULL)
 606                 goto out;
 607
 608         new_f->fn_key = key;
 609         new_f->fn_type = type;
 610         new_f->fn_scope = r->rtm_scope;
 611         DN_FIB_INFO(new_f) = fi;
 612
 613         new_f->fn_next = f;
 614         write_lock_bh(&dn_fib_tables_lock);
 615         *fp = new_f;
 616         write_unlock_bh(&dn_fib_tables_lock);
 617         dz->dz_nent++;
 618
 619         if (del_fp) {
 620                 f = *del_fp;
 621                 write_lock_bh(&dn_fib_tables_lock);
 622                 *del_fp = f->fn_next;
 623                 write_unlock_bh(&dn_fib_tables_lock);
 624
 625                 if (!(f->fn_state & DN_S_ZOMBIE))
 626                         dn_rtmsg_fib(RTM_DELROUTE, f, z, tb->n, n, req);
 627                 if (f->fn_state & DN_S_ACCESSED)
 628                         dn_rt_cache_flush(-1);
 629                 dn_free_node(f);
 630                 dz->dz_nent--;
 631         } else {
 632                 dn_rt_cache_flush(-1);
 633         }
 634
 635         dn_rtmsg_fib(RTM_NEWROUTE, new_f, z, tb->n, n, req);
 636
 637         return 0;
 638 out:
 639         dn_fib_release_info(fi);
 640         return err;
 641 }
 642
 643
 644 static int dn_fib_table_delete(struct dn_fib_table *tb, struct rtmsg *r, struct dn_kern_rta *rta, struct nlmsghdr *n, struct netlink_skb_parms *req)
 645 {
 646         struct dn_hash *table = (struct dn_hash*)tb->data;
 647         struct dn_fib_node **fp, **del_fp, *f;
 648         int z = r->rtm_dst_len;
 649         struct dn_zone *dz;
 650         dn_fib_key_t key;
 651         int matched;
 652
 653
 654         if (z > 16)
 655                 return -EINVAL;
 656
 657         if ((dz = table->dh_zones[z]) == NULL)
 658                 return -ESRCH;
 659
 660         dz_key_0(key);
 661         if (rta->rta_dst) {
 662                 __le16 dst;
 663                 memcpy(&dst, rta->rta_dst, 2);
 664                 if (dst & ~DZ_MASK(dz))
 665                         return -EINVAL;
 666                 key = dz_key(dst, dz);
 667         }
 668
 669         fp = dn_chain_p(key, dz);
 670
 671         DN_FIB_SCAN(f, fp) {
 672                 if (dn_key_eq(f->fn_key, key))
 673                         break;
 674                 if (dn_key_leq(key, f->fn_key))
 675                         return -ESRCH;
 676         }
 677
 678         matched = 0;
 679         del_fp = NULL;
 680         DN_FIB_SCAN_KEY(f, fp, key) {
 681                 struct dn_fib_info *fi = DN_FIB_INFO(f);
 682
 683                 if (f->fn_state & DN_S_ZOMBIE)
 684                         return -ESRCH;
 685
 686                 matched++;
 687
 688                 if (del_fp == NULL &&
 689                                 (!r->rtm_type || f->fn_type == r->rtm_type) &&
 690                                 (r->rtm_scope == RT_SCOPE_NOWHERE || f->fn_scope == r->rtm_scope) &&
 691                                 (!r->rtm_protocol ||
 692                                         fi->fib_protocol == r->rtm_protocol) &&
 693                                 dn_fib_nh_match(r, n, rta, fi) == 0)
 694                         del_fp = fp;
 695         }
 696
 697         if (del_fp) {
 698                 f = *del_fp;
 699                 dn_rtmsg_fib(RTM_DELROUTE, f, z, tb->n, n, req);
 700
 701                 if (matched != 1) {
 702                         write_lock_bh(&dn_fib_tables_lock);
 703                         *del_fp = f->fn_next;
 704                         write_unlock_bh(&dn_fib_tables_lock);
 705
 706                         if (f->fn_state & DN_S_ACCESSED)
 707                                 dn_rt_cache_flush(-1);
 708                         dn_free_node(f);
 709                         dz->dz_nent--;
 710                 } else {
 711                         f->fn_state |= DN_S_ZOMBIE;
 712                         if (f->fn_state & DN_S_ACCESSED) {
 713                                 f->fn_state &= ~DN_S_ACCESSED;
 714                                 dn_rt_cache_flush(-1);
 715                         }
 716                         if (++dn_fib_hash_zombies > 128)
 717                                 dn_fib_flush();
 718                 }
 719
 720                 return 0;
 721         }
 722
 723         return -ESRCH;
 724 }
 725
 726 static inline int dn_flush_list(struct dn_fib_node **fp, int z, struct dn_hash *table)
 727 {
 728         int found = 0;
 729         struct dn_fib_node *f;
 730
 731         while((f = *fp) != NULL) {
 732                 struct dn_fib_info *fi = DN_FIB_INFO(f);
 733
 734                 if (fi && ((f->fn_state & DN_S_ZOMBIE) || (fi->fib_flags & RTNH_F_DEAD))) {
 735                         write_lock_bh(&dn_fib_tables_lock);
 736                         *fp = f->fn_next;
 737                         write_unlock_bh(&dn_fib_tables_lock);
 738
 739                         dn_free_node(f);
 740                         found++;
 741                         continue;
 742                 }
 743                 fp = &f->fn_next;
 744         }
 745
 746         return found;
 747 }
 748
 749 static int dn_fib_table_flush(struct dn_fib_table *tb)
 750 {
 751         struct dn_hash *table = (struct dn_hash *)tb->data;
 752         struct dn_zone *dz;
 753         int found = 0;
 754
 755         dn_fib_hash_zombies = 0;
 756         for(dz = table->dh_zone_list; dz; dz = dz->dz_next) {
 757                 int i;
 758                 int tmp = 0;
 759                 for(i = dz->dz_divisor-1; i >= 0; i--)
 760                         tmp += dn_flush_list(&dz->dz_hash[i], dz->dz_order, table);
 761                 dz->dz_nent -= tmp;
 762                 found += tmp;
 763         }
 764
 765         return found;
 766 }
 767
 768 static int dn_fib_table_lookup(struct dn_fib_table *tb, const struct flowidn *flp, struct dn_fib_res *res)
 769 {
 770         int err;
 771         struct dn_zone *dz;
 772         struct dn_hash *t = (struct dn_hash *)tb->data;
 773
 774         read_lock(&dn_fib_tables_lock);
 775         for(dz = t->dh_zone_list; dz; dz = dz->dz_next) {
 776                 struct dn_fib_node *f;
 777                 dn_fib_key_t k = dz_key(flp->daddr, dz);
 778
 779                 for(f = dz_chain(k, dz); f; f = f->fn_next) {
 780                         if (!dn_key_eq(k, f->fn_key)) {
 781                                 if (dn_key_leq(k, f->fn_key))
 782                                         break;
 783                                 else
 784                                         continue;
 785                         }
 786
 787                         f->fn_state |= DN_S_ACCESSED;
 788
 789                         if (f->fn_state&DN_S_ZOMBIE)
 790                                 continue;
 791
 792                         if (f->fn_scope < flp->flowidn_scope)
 793                                 continue;
 794
 795                         err = dn_fib_semantic_match(f->fn_type, DN_FIB_INFO(f), flp, res);
 796
 797                         if (err == 0) {
 798                                 res->type = f->fn_type;
 799                                 res->scope = f->fn_scope;
 800                                 res->prefixlen = dz->dz_order;
 801                                 goto out;
 802                         }
 803                         if (err < 0)
 804                                 goto out;
 805                 }
 806         }
 807         err = 1;
 808 out:
 809         read_unlock(&dn_fib_tables_lock);
 810         return err;
 811 }
 812
 813
 814 struct dn_fib_table *dn_fib_get_table(u32 n, int create)
 815 {
 816         struct dn_fib_table *t;
 817         struct hlist_node *node;
 818         unsigned int h;
 819
 820         if (n < RT_TABLE_MIN)
 821                 return NULL;
 822
 823         if (n > RT_TABLE_MAX)
 824                 return NULL;
 825
 826         h = n & (DN_FIB_TABLE_HASHSZ - 1);
 827         rcu_read_lock();
 828         hlist_for_each_entry_rcu(t, node, &dn_fib_table_hash[h], hlist) {
 829                 if (t->n == n) {
 830                         rcu_read_unlock();
 831                         return t;
 832                 }
 833         }
 834         rcu_read_unlock();
 835
 836         if (!create)
 837                 return NULL;
 838
 839         if (in_interrupt() && net_ratelimit()) {
 840                 printk(KERN_DEBUG "DECnet: BUG! Attempt to create routing table from interrupt\n");
 841                 return NULL;
 842         }
 843
 844         t = kzalloc(sizeof(struct dn_fib_table) + sizeof(struct dn_hash),
 845                     GFP_KERNEL);
 846         if (t == NULL)
 847                 return NULL;
 848
 849         t->n = n;
 850         t->insert = dn_fib_table_insert;
 851         t->delete = dn_fib_table_delete;
 852         t->lookup = dn_fib_table_lookup;
 853         t->flush  = dn_fib_table_flush;
 854         t->dump = dn_fib_table_dump;
 855         hlist_add_head_rcu(&t->hlist, &dn_fib_table_hash[h]);
 856
 857         return t;
 858 }
 859
 860 struct dn_fib_table *dn_fib_empty_table(void)
 861 {
 862         u32 id;
 863
 864         for(id = RT_TABLE_MIN; id <= RT_TABLE_MAX; id++)
 865                 if (dn_fib_get_table(id, 0) == NULL)
 866                         return dn_fib_get_table(id, 1);
 867         return NULL;
 868 }
 869
 870 void dn_fib_flush(void)
 871 {
 872         int flushed = 0;
 873         struct dn_fib_table *tb;
 874         struct hlist_node *node;
 875         unsigned int h;
 876
 877         for (h = 0; h < DN_FIB_TABLE_HASHSZ; h++) {
 878                 hlist_for_each_entry(tb, node, &dn_fib_table_hash[h], hlist)
 879                         flushed += tb->flush(tb);
 880         }
 881
 882         if (flushed)
 883                 dn_rt_cache_flush(-1);
 884 }
 885
 886 void __init dn_fib_table_init(void)
 887 {
 888         dn_hash_kmem = kmem_cache_create("dn_fib_info_cache",
 889                                         sizeof(struct dn_fib_info),
 890                                         0, SLAB_HWCACHE_ALIGN,
 891                                         NULL);
 892 }
 893
 894 void __exit dn_fib_table_cleanup(void)
 895 {
 896         struct dn_fib_table *t;
 897         struct hlist_node *node, *next;
 898         unsigned int h;
 899
 900         write_lock(&dn_fib_tables_lock);
 901         for (h = 0; h < DN_FIB_TABLE_HASHSZ; h++) {
 902                 hlist_for_each_entry_safe(t, node, next, &dn_fib_table_hash[h],
 903                                           hlist) {
 904                         hlist_del(&t->hlist);
 905                         kfree(t);
 906                 }
 907         }
 908         write_unlock(&dn_fib_tables_lock);
 909 }