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