The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/net/radix.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1988, 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 4. Neither the name of the University nor the names of its contributors
   14  *    may be used to endorse or promote products derived from this software
   15  *    without specific prior written permission.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  *
   29  *      @(#)radix.c     8.5 (Berkeley) 5/19/95
   30  * $FreeBSD$
   31  */
   32 
   33 /*
   34  * Routines to build and maintain radix trees for routing lookups.
   35  */
   36 #include <sys/param.h>
   37 #ifdef  _KERNEL
   38 #include <sys/lock.h>
   39 #include <sys/mutex.h>
   40 #include <sys/rwlock.h>
   41 #include <sys/systm.h>
   42 #include <sys/malloc.h>
   43 #include <sys/syslog.h>
   44 #include <net/radix.h>
   45 #include "opt_mpath.h"
   46 #ifdef RADIX_MPATH
   47 #include <net/radix_mpath.h>
   48 #endif
   49 #else /* !_KERNEL */
   50 #include <stdio.h>
   51 #include <strings.h>
   52 #include <stdlib.h>
   53 #define log(x, arg...)  fprintf(stderr, ## arg)
   54 #define panic(x)        fprintf(stderr, "PANIC: %s", x), exit(1)
   55 #define min(a, b) ((a) < (b) ? (a) : (b) )
   56 #include <net/radix.h>
   57 #endif /* !_KERNEL */
   58 
   59 static int      rn_walktree_from(struct radix_node_head *h, void *a, void *m,
   60                     walktree_f_t *f, void *w);
   61 static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *);
   62 static struct radix_node
   63          *rn_insert(void *, struct radix_node_head *, int *,
   64              struct radix_node [2]),
   65          *rn_newpair(void *, int, struct radix_node[2]),
   66          *rn_search(void *, struct radix_node *),
   67          *rn_search_m(void *, struct radix_node *, void *);
   68 
   69 static int      max_keylen;
   70 static struct radix_mask *rn_mkfreelist;
   71 static struct radix_node_head *mask_rnhead;
   72 /*
   73  * Work area -- the following point to 3 buffers of size max_keylen,
   74  * allocated in this order in a block of memory malloc'ed by rn_init.
   75  * rn_zeros, rn_ones are set in rn_init and used in readonly afterwards.
   76  * addmask_key is used in rn_addmask in rw mode and not thread-safe.
   77  */
   78 static char *rn_zeros, *rn_ones, *addmask_key;
   79 
   80 #define MKGet(m) {                                              \
   81         if (rn_mkfreelist) {                                    \
   82                 m = rn_mkfreelist;                              \
   83                 rn_mkfreelist = (m)->rm_mklist;                 \
   84         } else                                                  \
   85                 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask)); }
   86  
   87 #define MKFree(m) { (m)->rm_mklist = rn_mkfreelist; rn_mkfreelist = (m);}
   88 
   89 #define rn_masktop (mask_rnhead->rnh_treetop)
   90 
   91 static int      rn_lexobetter(void *m_arg, void *n_arg);
   92 static struct radix_mask *
   93                 rn_new_radix_mask(struct radix_node *tt,
   94                     struct radix_mask *next);
   95 static int      rn_satisfies_leaf(char *trial, struct radix_node *leaf,
   96                     int skip);
   97 
   98 /*
   99  * The data structure for the keys is a radix tree with one way
  100  * branching removed.  The index rn_bit at an internal node n represents a bit
  101  * position to be tested.  The tree is arranged so that all descendants
  102  * of a node n have keys whose bits all agree up to position rn_bit - 1.
  103  * (We say the index of n is rn_bit.)
  104  *
  105  * There is at least one descendant which has a one bit at position rn_bit,
  106  * and at least one with a zero there.
  107  *
  108  * A route is determined by a pair of key and mask.  We require that the
  109  * bit-wise logical and of the key and mask to be the key.
  110  * We define the index of a route to associated with the mask to be
  111  * the first bit number in the mask where 0 occurs (with bit number 0
  112  * representing the highest order bit).
  113  *
  114  * We say a mask is normal if every bit is 0, past the index of the mask.
  115  * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit,
  116  * and m is a normal mask, then the route applies to every descendant of n.
  117  * If the index(m) < rn_bit, this implies the trailing last few bits of k
  118  * before bit b are all 0, (and hence consequently true of every descendant
  119  * of n), so the route applies to all descendants of the node as well.
  120  *
  121  * Similar logic shows that a non-normal mask m such that
  122  * index(m) <= index(n) could potentially apply to many children of n.
  123  * Thus, for each non-host route, we attach its mask to a list at an internal
  124  * node as high in the tree as we can go.
  125  *
  126  * The present version of the code makes use of normal routes in short-
  127  * circuiting an explict mask and compare operation when testing whether
  128  * a key satisfies a normal route, and also in remembering the unique leaf
  129  * that governs a subtree.
  130  */
  131 
  132 /*
  133  * Most of the functions in this code assume that the key/mask arguments
  134  * are sockaddr-like structures, where the first byte is an u_char
  135  * indicating the size of the entire structure.
  136  *
  137  * To make the assumption more explicit, we use the LEN() macro to access
  138  * this field. It is safe to pass an expression with side effects
  139  * to LEN() as the argument is evaluated only once.
  140  * We cast the result to int as this is the dominant usage.
  141  */
  142 #define LEN(x) ( (int) (*(const u_char *)(x)) )
  143 
  144 /*
  145  * XXX THIS NEEDS TO BE FIXED
  146  * In the code, pointers to keys and masks are passed as either
  147  * 'void *' (because callers use to pass pointers of various kinds), or
  148  * 'caddr_t' (which is fine for pointer arithmetics, but not very
  149  * clean when you dereference it to access data). Furthermore, caddr_t
  150  * is really 'char *', while the natural type to operate on keys and
  151  * masks would be 'u_char'. This mismatch require a lot of casts and
  152  * intermediate variables to adapt types that clutter the code.
  153  */
  154 
  155 /*
  156  * Search a node in the tree matching the key.
  157  */
  158 static struct radix_node *
  159 rn_search(v_arg, head)
  160         void *v_arg;
  161         struct radix_node *head;
  162 {
  163         register struct radix_node *x;
  164         register caddr_t v;
  165 
  166         for (x = head, v = v_arg; x->rn_bit >= 0;) {
  167                 if (x->rn_bmask & v[x->rn_offset])
  168                         x = x->rn_right;
  169                 else
  170                         x = x->rn_left;
  171         }
  172         return (x);
  173 }
  174 
  175 /*
  176  * Same as above, but with an additional mask.
  177  * XXX note this function is used only once.
  178  */
  179 static struct radix_node *
  180 rn_search_m(v_arg, head, m_arg)
  181         struct radix_node *head;
  182         void *v_arg, *m_arg;
  183 {
  184         register struct radix_node *x;
  185         register caddr_t v = v_arg, m = m_arg;
  186 
  187         for (x = head; x->rn_bit >= 0;) {
  188                 if ((x->rn_bmask & m[x->rn_offset]) &&
  189                     (x->rn_bmask & v[x->rn_offset]))
  190                         x = x->rn_right;
  191                 else
  192                         x = x->rn_left;
  193         }
  194         return x;
  195 }
  196 
  197 int
  198 rn_refines(m_arg, n_arg)
  199         void *m_arg, *n_arg;
  200 {
  201         register caddr_t m = m_arg, n = n_arg;
  202         register caddr_t lim, lim2 = lim = n + LEN(n);
  203         int longer = LEN(n++) - LEN(m++);
  204         int masks_are_equal = 1;
  205 
  206         if (longer > 0)
  207                 lim -= longer;
  208         while (n < lim) {
  209                 if (*n & ~(*m))
  210                         return 0;
  211                 if (*n++ != *m++)
  212                         masks_are_equal = 0;
  213         }
  214         while (n < lim2)
  215                 if (*n++)
  216                         return 0;
  217         if (masks_are_equal && (longer < 0))
  218                 for (lim2 = m - longer; m < lim2; )
  219                         if (*m++)
  220                                 return 1;
  221         return (!masks_are_equal);
  222 }
  223 
  224 struct radix_node *
  225 rn_lookup(v_arg, m_arg, head)
  226         void *v_arg, *m_arg;
  227         struct radix_node_head *head;
  228 {
  229         register struct radix_node *x;
  230         caddr_t netmask = 0;
  231 
  232         if (m_arg) {
  233                 x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_offset);
  234                 if (x == 0)
  235                         return (0);
  236                 netmask = x->rn_key;
  237         }
  238         x = rn_match(v_arg, head);
  239         if (x && netmask) {
  240                 while (x && x->rn_mask != netmask)
  241                         x = x->rn_dupedkey;
  242         }
  243         return x;
  244 }
  245 
  246 static int
  247 rn_satisfies_leaf(trial, leaf, skip)
  248         char *trial;
  249         register struct radix_node *leaf;
  250         int skip;
  251 {
  252         register char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
  253         char *cplim;
  254         int length = min(LEN(cp), LEN(cp2));
  255 
  256         if (cp3 == NULL)
  257                 cp3 = rn_ones;
  258         else
  259                 length = min(length, LEN(cp3));
  260         cplim = cp + length; cp3 += skip; cp2 += skip;
  261         for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
  262                 if ((*cp ^ *cp2) & *cp3)
  263                         return 0;
  264         return 1;
  265 }
  266 
  267 struct radix_node *
  268 rn_match(v_arg, head)
  269         void *v_arg;
  270         struct radix_node_head *head;
  271 {
  272         caddr_t v = v_arg;
  273         register struct radix_node *t = head->rnh_treetop, *x;
  274         register caddr_t cp = v, cp2;
  275         caddr_t cplim;
  276         struct radix_node *saved_t, *top = t;
  277         int off = t->rn_offset, vlen = LEN(cp), matched_off;
  278         register int test, b, rn_bit;
  279 
  280         /*
  281          * Open code rn_search(v, top) to avoid overhead of extra
  282          * subroutine call.
  283          */
  284         for (; t->rn_bit >= 0; ) {
  285                 if (t->rn_bmask & cp[t->rn_offset])
  286                         t = t->rn_right;
  287                 else
  288                         t = t->rn_left;
  289         }
  290         /*
  291          * See if we match exactly as a host destination
  292          * or at least learn how many bits match, for normal mask finesse.
  293          *
  294          * It doesn't hurt us to limit how many bytes to check
  295          * to the length of the mask, since if it matches we had a genuine
  296          * match and the leaf we have is the most specific one anyway;
  297          * if it didn't match with a shorter length it would fail
  298          * with a long one.  This wins big for class B&C netmasks which
  299          * are probably the most common case...
  300          */
  301         if (t->rn_mask)
  302                 vlen = *(u_char *)t->rn_mask;
  303         cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
  304         for (; cp < cplim; cp++, cp2++)
  305                 if (*cp != *cp2)
  306                         goto on1;
  307         /*
  308          * This extra grot is in case we are explicitly asked
  309          * to look up the default.  Ugh!
  310          *
  311          * Never return the root node itself, it seems to cause a
  312          * lot of confusion.
  313          */
  314         if (t->rn_flags & RNF_ROOT)
  315                 t = t->rn_dupedkey;
  316         return t;
  317 on1:
  318         test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
  319         for (b = 7; (test >>= 1) > 0;)
  320                 b--;
  321         matched_off = cp - v;
  322         b += matched_off << 3;
  323         rn_bit = -1 - b;
  324         /*
  325          * If there is a host route in a duped-key chain, it will be first.
  326          */
  327         if ((saved_t = t)->rn_mask == 0)
  328                 t = t->rn_dupedkey;
  329         for (; t; t = t->rn_dupedkey)
  330                 /*
  331                  * Even if we don't match exactly as a host,
  332                  * we may match if the leaf we wound up at is
  333                  * a route to a net.
  334                  */
  335                 if (t->rn_flags & RNF_NORMAL) {
  336                         if (rn_bit <= t->rn_bit)
  337                                 return t;
  338                 } else if (rn_satisfies_leaf(v, t, matched_off))
  339                                 return t;
  340         t = saved_t;
  341         /* start searching up the tree */
  342         do {
  343                 register struct radix_mask *m;
  344                 t = t->rn_parent;
  345                 m = t->rn_mklist;
  346                 /*
  347                  * If non-contiguous masks ever become important
  348                  * we can restore the masking and open coding of
  349                  * the search and satisfaction test and put the
  350                  * calculation of "off" back before the "do".
  351                  */
  352                 while (m) {
  353                         if (m->rm_flags & RNF_NORMAL) {
  354                                 if (rn_bit <= m->rm_bit)
  355                                         return (m->rm_leaf);
  356                         } else {
  357                                 off = min(t->rn_offset, matched_off);
  358                                 x = rn_search_m(v, t, m->rm_mask);
  359                                 while (x && x->rn_mask != m->rm_mask)
  360                                         x = x->rn_dupedkey;
  361                                 if (x && rn_satisfies_leaf(v, x, off))
  362                                         return x;
  363                         }
  364                         m = m->rm_mklist;
  365                 }
  366         } while (t != top);
  367         return 0;
  368 }
  369 
  370 #ifdef RN_DEBUG
  371 int     rn_nodenum;
  372 struct  radix_node *rn_clist;
  373 int     rn_saveinfo;
  374 int     rn_debug =  1;
  375 #endif
  376 
  377 /*
  378  * Whenever we add a new leaf to the tree, we also add a parent node,
  379  * so we allocate them as an array of two elements: the first one must be
  380  * the leaf (see RNTORT() in route.c), the second one is the parent.
  381  * This routine initializes the relevant fields of the nodes, so that
  382  * the leaf is the left child of the parent node, and both nodes have
  383  * (almost) all all fields filled as appropriate.
  384  * (XXX some fields are left unset, see the '#if 0' section).
  385  * The function returns a pointer to the parent node.
  386  */
  387 
  388 static struct radix_node *
  389 rn_newpair(v, b, nodes)
  390         void *v;
  391         int b;
  392         struct radix_node nodes[2];
  393 {
  394         register struct radix_node *tt = nodes, *t = tt + 1;
  395         t->rn_bit = b;
  396         t->rn_bmask = 0x80 >> (b & 7);
  397         t->rn_left = tt;
  398         t->rn_offset = b >> 3;
  399 
  400 #if 0  /* XXX perhaps we should fill these fields as well. */
  401         t->rn_parent = t->rn_right = NULL;
  402 
  403         tt->rn_mask = NULL;
  404         tt->rn_dupedkey = NULL;
  405         tt->rn_bmask = 0;
  406 #endif
  407         tt->rn_bit = -1;
  408         tt->rn_key = (caddr_t)v;
  409         tt->rn_parent = t;
  410         tt->rn_flags = t->rn_flags = RNF_ACTIVE;
  411         tt->rn_mklist = t->rn_mklist = 0;
  412 #ifdef RN_DEBUG
  413         tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
  414         tt->rn_twin = t;
  415         tt->rn_ybro = rn_clist;
  416         rn_clist = tt;
  417 #endif
  418         return t;
  419 }
  420 
  421 static struct radix_node *
  422 rn_insert(v_arg, head, dupentry, nodes)
  423         void *v_arg;
  424         struct radix_node_head *head;
  425         int *dupentry;
  426         struct radix_node nodes[2];
  427 {
  428         caddr_t v = v_arg;
  429         struct radix_node *top = head->rnh_treetop;
  430         int head_off = top->rn_offset, vlen = LEN(v);
  431         register struct radix_node *t = rn_search(v_arg, top);
  432         register caddr_t cp = v + head_off;
  433         register int b;
  434         struct radix_node *tt;
  435         /*
  436          * Find first bit at which v and t->rn_key differ
  437          */
  438     {
  439         register caddr_t cp2 = t->rn_key + head_off;
  440         register int cmp_res;
  441         caddr_t cplim = v + vlen;
  442 
  443         while (cp < cplim)
  444                 if (*cp2++ != *cp++)
  445                         goto on1;
  446         *dupentry = 1;
  447         return t;
  448 on1:
  449         *dupentry = 0;
  450         cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
  451         for (b = (cp - v) << 3; cmp_res; b--)
  452                 cmp_res >>= 1;
  453     }
  454     {
  455         register struct radix_node *p, *x = top;
  456         cp = v;
  457         do {
  458                 p = x;
  459                 if (cp[x->rn_offset] & x->rn_bmask)
  460                         x = x->rn_right;
  461                 else
  462                         x = x->rn_left;
  463         } while (b > (unsigned) x->rn_bit);
  464                                 /* x->rn_bit < b && x->rn_bit >= 0 */
  465 #ifdef RN_DEBUG
  466         if (rn_debug)
  467                 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
  468 #endif
  469         t = rn_newpair(v_arg, b, nodes); 
  470         tt = t->rn_left;
  471         if ((cp[p->rn_offset] & p->rn_bmask) == 0)
  472                 p->rn_left = t;
  473         else
  474                 p->rn_right = t;
  475         x->rn_parent = t;
  476         t->rn_parent = p; /* frees x, p as temp vars below */
  477         if ((cp[t->rn_offset] & t->rn_bmask) == 0) {
  478                 t->rn_right = x;
  479         } else {
  480                 t->rn_right = tt;
  481                 t->rn_left = x;
  482         }
  483 #ifdef RN_DEBUG
  484         if (rn_debug)
  485                 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
  486 #endif
  487     }
  488         return (tt);
  489 }
  490 
  491 struct radix_node *
  492 rn_addmask(n_arg, search, skip)
  493         int search, skip;
  494         void *n_arg;
  495 {
  496         caddr_t netmask = (caddr_t)n_arg;
  497         register struct radix_node *x;
  498         register caddr_t cp, cplim;
  499         register int b = 0, mlen, j;
  500         int maskduplicated, m0, isnormal;
  501         struct radix_node *saved_x;
  502         static int last_zeroed = 0;
  503 
  504         if ((mlen = LEN(netmask)) > max_keylen)
  505                 mlen = max_keylen;
  506         if (skip == 0)
  507                 skip = 1;
  508         if (mlen <= skip)
  509                 return (mask_rnhead->rnh_nodes);
  510         if (skip > 1)
  511                 bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
  512         if ((m0 = mlen) > skip)
  513                 bcopy(netmask + skip, addmask_key + skip, mlen - skip);
  514         /*
  515          * Trim trailing zeroes.
  516          */
  517         for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
  518                 cp--;
  519         mlen = cp - addmask_key;
  520         if (mlen <= skip) {
  521                 if (m0 >= last_zeroed)
  522                         last_zeroed = mlen;
  523                 return (mask_rnhead->rnh_nodes);
  524         }
  525         if (m0 < last_zeroed)
  526                 bzero(addmask_key + m0, last_zeroed - m0);
  527         *addmask_key = last_zeroed = mlen;
  528         x = rn_search(addmask_key, rn_masktop);
  529         if (bcmp(addmask_key, x->rn_key, mlen) != 0)
  530                 x = 0;
  531         if (x || search)
  532                 return (x);
  533         R_Zalloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
  534         if ((saved_x = x) == 0)
  535                 return (0);
  536         netmask = cp = (caddr_t)(x + 2);
  537         bcopy(addmask_key, cp, mlen);
  538         x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
  539         if (maskduplicated) {
  540                 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
  541                 Free(saved_x);
  542                 return (x);
  543         }
  544         /*
  545          * Calculate index of mask, and check for normalcy.
  546          * First find the first byte with a 0 bit, then if there are
  547          * more bits left (remember we already trimmed the trailing 0's),
  548          * the pattern must be one of those in normal_chars[], or we have
  549          * a non-contiguous mask.
  550          */
  551         cplim = netmask + mlen;
  552         isnormal = 1;
  553         for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
  554                 cp++;
  555         if (cp != cplim) {
  556                 static char normal_chars[] = {
  557                         0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff};
  558 
  559                 for (j = 0x80; (j & *cp) != 0; j >>= 1)
  560                         b++;
  561                 if (*cp != normal_chars[b] || cp != (cplim - 1))
  562                         isnormal = 0;
  563         }
  564         b += (cp - netmask) << 3;
  565         x->rn_bit = -1 - b;
  566         if (isnormal)
  567                 x->rn_flags |= RNF_NORMAL;
  568         return (x);
  569 }
  570 
  571 static int      /* XXX: arbitrary ordering for non-contiguous masks */
  572 rn_lexobetter(m_arg, n_arg)
  573         void *m_arg, *n_arg;
  574 {
  575         register u_char *mp = m_arg, *np = n_arg, *lim;
  576 
  577         if (LEN(mp) > LEN(np))
  578                 return 1;  /* not really, but need to check longer one first */
  579         if (LEN(mp) == LEN(np))
  580                 for (lim = mp + LEN(mp); mp < lim;)
  581                         if (*mp++ > *np++)
  582                                 return 1;
  583         return 0;
  584 }
  585 
  586 static struct radix_mask *
  587 rn_new_radix_mask(tt, next)
  588         register struct radix_node *tt;
  589         register struct radix_mask *next;
  590 {
  591         register struct radix_mask *m;
  592 
  593         MKGet(m);
  594         if (m == 0) {
  595                 log(LOG_ERR, "Mask for route not entered\n");
  596                 return (0);
  597         }
  598         bzero(m, sizeof *m);
  599         m->rm_bit = tt->rn_bit;
  600         m->rm_flags = tt->rn_flags;
  601         if (tt->rn_flags & RNF_NORMAL)
  602                 m->rm_leaf = tt;
  603         else
  604                 m->rm_mask = tt->rn_mask;
  605         m->rm_mklist = next;
  606         tt->rn_mklist = m;
  607         return m;
  608 }
  609 
  610 struct radix_node *
  611 rn_addroute(v_arg, n_arg, head, treenodes)
  612         void *v_arg, *n_arg;
  613         struct radix_node_head *head;
  614         struct radix_node treenodes[2];
  615 {
  616         caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
  617         register struct radix_node *t, *x = 0, *tt;
  618         struct radix_node *saved_tt, *top = head->rnh_treetop;
  619         short b = 0, b_leaf = 0;
  620         int keyduplicated;
  621         caddr_t mmask;
  622         struct radix_mask *m, **mp;
  623 
  624         /*
  625          * In dealing with non-contiguous masks, there may be
  626          * many different routes which have the same mask.
  627          * We will find it useful to have a unique pointer to
  628          * the mask to speed avoiding duplicate references at
  629          * nodes and possibly save time in calculating indices.
  630          */
  631         if (netmask)  {
  632                 if ((x = rn_addmask(netmask, 0, top->rn_offset)) == 0)
  633                         return (0);
  634                 b_leaf = x->rn_bit;
  635                 b = -1 - x->rn_bit;
  636                 netmask = x->rn_key;
  637         }
  638         /*
  639          * Deal with duplicated keys: attach node to previous instance
  640          */
  641         saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
  642         if (keyduplicated) {
  643                 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
  644 #ifdef RADIX_MPATH
  645                         /* permit multipath, if enabled for the family */
  646                         if (rn_mpath_capable(head) && netmask == tt->rn_mask) {
  647                                 /*
  648                                  * go down to the end of multipaths, so that
  649                                  * new entry goes into the end of rn_dupedkey
  650                                  * chain.
  651                                  */
  652                                 do {
  653                                         t = tt;
  654                                         tt = tt->rn_dupedkey;
  655                                 } while (tt && t->rn_mask == tt->rn_mask);
  656                                 break;
  657                         }
  658 #endif
  659                         if (tt->rn_mask == netmask)
  660                                 return (0);
  661                         if (netmask == 0 ||
  662                             (tt->rn_mask &&
  663                              ((b_leaf < tt->rn_bit) /* index(netmask) > node */
  664                               || rn_refines(netmask, tt->rn_mask)
  665                               || rn_lexobetter(netmask, tt->rn_mask))))
  666                                 break;
  667                 }
  668                 /*
  669                  * If the mask is not duplicated, we wouldn't
  670                  * find it among possible duplicate key entries
  671                  * anyway, so the above test doesn't hurt.
  672                  *
  673                  * We sort the masks for a duplicated key the same way as
  674                  * in a masklist -- most specific to least specific.
  675                  * This may require the unfortunate nuisance of relocating
  676                  * the head of the list.
  677                  *
  678                  * We also reverse, or doubly link the list through the
  679                  * parent pointer.
  680                  */
  681                 if (tt == saved_tt) {
  682                         struct  radix_node *xx = x;
  683                         /* link in at head of list */
  684                         (tt = treenodes)->rn_dupedkey = t;
  685                         tt->rn_flags = t->rn_flags;
  686                         tt->rn_parent = x = t->rn_parent;
  687                         t->rn_parent = tt;                      /* parent */
  688                         if (x->rn_left == t)
  689                                 x->rn_left = tt;
  690                         else
  691                                 x->rn_right = tt;
  692                         saved_tt = tt; x = xx;
  693                 } else {
  694                         (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
  695                         t->rn_dupedkey = tt;
  696                         tt->rn_parent = t;                      /* parent */
  697                         if (tt->rn_dupedkey)                    /* parent */
  698                                 tt->rn_dupedkey->rn_parent = tt; /* parent */
  699                 }
  700 #ifdef RN_DEBUG
  701                 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
  702                 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
  703 #endif
  704                 tt->rn_key = (caddr_t) v;
  705                 tt->rn_bit = -1;
  706                 tt->rn_flags = RNF_ACTIVE;
  707         }
  708         /*
  709          * Put mask in tree.
  710          */
  711         if (netmask) {
  712                 tt->rn_mask = netmask;
  713                 tt->rn_bit = x->rn_bit;
  714                 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
  715         }
  716         t = saved_tt->rn_parent;
  717         if (keyduplicated)
  718                 goto on2;
  719         b_leaf = -1 - t->rn_bit;
  720         if (t->rn_right == saved_tt)
  721                 x = t->rn_left;
  722         else
  723                 x = t->rn_right;
  724         /* Promote general routes from below */
  725         if (x->rn_bit < 0) {
  726             for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
  727                 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) {
  728                         *mp = m = rn_new_radix_mask(x, 0);
  729                         if (m)
  730                                 mp = &m->rm_mklist;
  731                 }
  732         } else if (x->rn_mklist) {
  733                 /*
  734                  * Skip over masks whose index is > that of new node
  735                  */
  736                 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
  737                         if (m->rm_bit >= b_leaf)
  738                                 break;
  739                 t->rn_mklist = m; *mp = 0;
  740         }
  741 on2:
  742         /* Add new route to highest possible ancestor's list */
  743         if ((netmask == 0) || (b > t->rn_bit ))
  744                 return tt; /* can't lift at all */
  745         b_leaf = tt->rn_bit;
  746         do {
  747                 x = t;
  748                 t = t->rn_parent;
  749         } while (b <= t->rn_bit && x != top);
  750         /*
  751          * Search through routes associated with node to
  752          * insert new route according to index.
  753          * Need same criteria as when sorting dupedkeys to avoid
  754          * double loop on deletion.
  755          */
  756         for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
  757                 if (m->rm_bit < b_leaf)
  758                         continue;
  759                 if (m->rm_bit > b_leaf)
  760                         break;
  761                 if (m->rm_flags & RNF_NORMAL) {
  762                         mmask = m->rm_leaf->rn_mask;
  763                         if (tt->rn_flags & RNF_NORMAL) {
  764 #if !defined(RADIX_MPATH)
  765                             log(LOG_ERR,
  766                                 "Non-unique normal route, mask not entered\n");
  767 #endif
  768                                 return tt;
  769                         }
  770                 } else
  771                         mmask = m->rm_mask;
  772                 if (mmask == netmask) {
  773                         m->rm_refs++;
  774                         tt->rn_mklist = m;
  775                         return tt;
  776                 }
  777                 if (rn_refines(netmask, mmask)
  778                     || rn_lexobetter(netmask, mmask))
  779                         break;
  780         }
  781         *mp = rn_new_radix_mask(tt, *mp);
  782         return tt;
  783 }
  784 
  785 struct radix_node *
  786 rn_delete(v_arg, netmask_arg, head)
  787         void *v_arg, *netmask_arg;
  788         struct radix_node_head *head;
  789 {
  790         register struct radix_node *t, *p, *x, *tt;
  791         struct radix_mask *m, *saved_m, **mp;
  792         struct radix_node *dupedkey, *saved_tt, *top;
  793         caddr_t v, netmask;
  794         int b, head_off, vlen;
  795 
  796         v = v_arg;
  797         netmask = netmask_arg;
  798         x = head->rnh_treetop;
  799         tt = rn_search(v, x);
  800         head_off = x->rn_offset;
  801         vlen =  LEN(v);
  802         saved_tt = tt;
  803         top = x;
  804         if (tt == 0 ||
  805             bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
  806                 return (0);
  807         /*
  808          * Delete our route from mask lists.
  809          */
  810         if (netmask) {
  811                 if ((x = rn_addmask(netmask, 1, head_off)) == 0)
  812                         return (0);
  813                 netmask = x->rn_key;
  814                 while (tt->rn_mask != netmask)
  815                         if ((tt = tt->rn_dupedkey) == 0)
  816                                 return (0);
  817         }
  818         if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
  819                 goto on1;
  820         if (tt->rn_flags & RNF_NORMAL) {
  821                 if (m->rm_leaf != tt || m->rm_refs > 0) {
  822                         log(LOG_ERR, "rn_delete: inconsistent annotation\n");
  823                         return 0;  /* dangling ref could cause disaster */
  824                 }
  825         } else {
  826                 if (m->rm_mask != tt->rn_mask) {
  827                         log(LOG_ERR, "rn_delete: inconsistent annotation\n");
  828                         goto on1;
  829                 }
  830                 if (--m->rm_refs >= 0)
  831                         goto on1;
  832         }
  833         b = -1 - tt->rn_bit;
  834         t = saved_tt->rn_parent;
  835         if (b > t->rn_bit)
  836                 goto on1; /* Wasn't lifted at all */
  837         do {
  838                 x = t;
  839                 t = t->rn_parent;
  840         } while (b <= t->rn_bit && x != top);
  841         for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
  842                 if (m == saved_m) {
  843                         *mp = m->rm_mklist;
  844                         MKFree(m);
  845                         break;
  846                 }
  847         if (m == 0) {
  848                 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
  849                 if (tt->rn_flags & RNF_NORMAL)
  850                         return (0); /* Dangling ref to us */
  851         }
  852 on1:
  853         /*
  854          * Eliminate us from tree
  855          */
  856         if (tt->rn_flags & RNF_ROOT)
  857                 return (0);
  858 #ifdef RN_DEBUG
  859         /* Get us out of the creation list */
  860         for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
  861         if (t) t->rn_ybro = tt->rn_ybro;
  862 #endif
  863         t = tt->rn_parent;
  864         dupedkey = saved_tt->rn_dupedkey;
  865         if (dupedkey) {
  866                 /*
  867                  * Here, tt is the deletion target and
  868                  * saved_tt is the head of the dupekey chain.
  869                  */
  870                 if (tt == saved_tt) {
  871                         /* remove from head of chain */
  872                         x = dupedkey; x->rn_parent = t;
  873                         if (t->rn_left == tt)
  874                                 t->rn_left = x;
  875                         else
  876                                 t->rn_right = x;
  877                 } else {
  878                         /* find node in front of tt on the chain */
  879                         for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
  880                                 p = p->rn_dupedkey;
  881                         if (p) {
  882                                 p->rn_dupedkey = tt->rn_dupedkey;
  883                                 if (tt->rn_dupedkey)            /* parent */
  884                                         tt->rn_dupedkey->rn_parent = p;
  885                                                                 /* parent */
  886                         } else log(LOG_ERR, "rn_delete: couldn't find us\n");
  887                 }
  888                 t = tt + 1;
  889                 if  (t->rn_flags & RNF_ACTIVE) {
  890 #ifndef RN_DEBUG
  891                         *++x = *t;
  892                         p = t->rn_parent;
  893 #else
  894                         b = t->rn_info;
  895                         *++x = *t;
  896                         t->rn_info = b;
  897                         p = t->rn_parent;
  898 #endif
  899                         if (p->rn_left == t)
  900                                 p->rn_left = x;
  901                         else
  902                                 p->rn_right = x;
  903                         x->rn_left->rn_parent = x;
  904                         x->rn_right->rn_parent = x;
  905                 }
  906                 goto out;
  907         }
  908         if (t->rn_left == tt)
  909                 x = t->rn_right;
  910         else
  911                 x = t->rn_left;
  912         p = t->rn_parent;
  913         if (p->rn_right == t)
  914                 p->rn_right = x;
  915         else
  916                 p->rn_left = x;
  917         x->rn_parent = p;
  918         /*
  919          * Demote routes attached to us.
  920          */
  921         if (t->rn_mklist) {
  922                 if (x->rn_bit >= 0) {
  923                         for (mp = &x->rn_mklist; (m = *mp);)
  924                                 mp = &m->rm_mklist;
  925                         *mp = t->rn_mklist;
  926                 } else {
  927                         /* If there are any key,mask pairs in a sibling
  928                            duped-key chain, some subset will appear sorted
  929                            in the same order attached to our mklist */
  930                         for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
  931                                 if (m == x->rn_mklist) {
  932                                         struct radix_mask *mm = m->rm_mklist;
  933                                         x->rn_mklist = 0;
  934                                         if (--(m->rm_refs) < 0)
  935                                                 MKFree(m);
  936                                         m = mm;
  937                                 }
  938                         if (m)
  939                                 log(LOG_ERR,
  940                                     "rn_delete: Orphaned Mask %p at %p\n",
  941                                     m, x);
  942                 }
  943         }
  944         /*
  945          * We may be holding an active internal node in the tree.
  946          */
  947         x = tt + 1;
  948         if (t != x) {
  949 #ifndef RN_DEBUG
  950                 *t = *x;
  951 #else
  952                 b = t->rn_info;
  953                 *t = *x;
  954                 t->rn_info = b;
  955 #endif
  956                 t->rn_left->rn_parent = t;
  957                 t->rn_right->rn_parent = t;
  958                 p = x->rn_parent;
  959                 if (p->rn_left == x)
  960                         p->rn_left = t;
  961                 else
  962                         p->rn_right = t;
  963         }
  964 out:
  965         tt->rn_flags &= ~RNF_ACTIVE;
  966         tt[1].rn_flags &= ~RNF_ACTIVE;
  967         return (tt);
  968 }
  969 
  970 /*
  971  * This is the same as rn_walktree() except for the parameters and the
  972  * exit.
  973  */
  974 static int
  975 rn_walktree_from(h, a, m, f, w)
  976         struct radix_node_head *h;
  977         void *a, *m;
  978         walktree_f_t *f;
  979         void *w;
  980 {
  981         int error;
  982         struct radix_node *base, *next;
  983         u_char *xa = (u_char *)a;
  984         u_char *xm = (u_char *)m;
  985         register struct radix_node *rn, *last = 0 /* shut up gcc */;
  986         int stopping = 0;
  987         int lastb;
  988 
  989         /*
  990          * rn_search_m is sort-of-open-coded here. We cannot use the
  991          * function because we need to keep track of the last node seen.
  992          */
  993         /* printf("about to search\n"); */
  994         for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) {
  995                 last = rn;
  996                 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n",
  997                        rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */
  998                 if (!(rn->rn_bmask & xm[rn->rn_offset])) {
  999                         break;
 1000                 }
 1001                 if (rn->rn_bmask & xa[rn->rn_offset]) {
 1002                         rn = rn->rn_right;
 1003                 } else {
 1004                         rn = rn->rn_left;
 1005                 }
 1006         }
 1007         /* printf("done searching\n"); */
 1008 
 1009         /*
 1010          * Two cases: either we stepped off the end of our mask,
 1011          * in which case last == rn, or we reached a leaf, in which
 1012          * case we want to start from the last node we looked at.
 1013          * Either way, last is the node we want to start from.
 1014          */
 1015         rn = last;
 1016         lastb = rn->rn_bit;
 1017 
 1018         /* printf("rn %p, lastb %d\n", rn, lastb);*/
 1019 
 1020         /*
 1021          * This gets complicated because we may delete the node
 1022          * while applying the function f to it, so we need to calculate
 1023          * the successor node in advance.
 1024          */
 1025         while (rn->rn_bit >= 0)
 1026                 rn = rn->rn_left;
 1027 
 1028         while (!stopping) {
 1029                 /* printf("node %p (%d)\n", rn, rn->rn_bit); */
 1030                 base = rn;
 1031                 /* If at right child go back up, otherwise, go right */
 1032                 while (rn->rn_parent->rn_right == rn
 1033                        && !(rn->rn_flags & RNF_ROOT)) {
 1034                         rn = rn->rn_parent;
 1035 
 1036                         /* if went up beyond last, stop */
 1037                         if (rn->rn_bit <= lastb) {
 1038                                 stopping = 1;
 1039                                 /* printf("up too far\n"); */
 1040                                 /*
 1041                                  * XXX we should jump to the 'Process leaves'
 1042                                  * part, because the values of 'rn' and 'next'
 1043                                  * we compute will not be used. Not a big deal
 1044                                  * because this loop will terminate, but it is
 1045                                  * inefficient and hard to understand!
 1046                                  */
 1047                         }
 1048                 }
 1049                 
 1050                 /* 
 1051                  * At the top of the tree, no need to traverse the right
 1052                  * half, prevent the traversal of the entire tree in the
 1053                  * case of default route.
 1054                  */
 1055                 if (rn->rn_parent->rn_flags & RNF_ROOT)
 1056                         stopping = 1;
 1057 
 1058                 /* Find the next *leaf* since next node might vanish, too */
 1059                 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
 1060                         rn = rn->rn_left;
 1061                 next = rn;
 1062                 /* Process leaves */
 1063                 while ((rn = base) != 0) {
 1064                         base = rn->rn_dupedkey;
 1065                         /* printf("leaf %p\n", rn); */
 1066                         if (!(rn->rn_flags & RNF_ROOT)
 1067                             && (error = (*f)(rn, w)))
 1068                                 return (error);
 1069                 }
 1070                 rn = next;
 1071 
 1072                 if (rn->rn_flags & RNF_ROOT) {
 1073                         /* printf("root, stopping"); */
 1074                         stopping = 1;
 1075                 }
 1076 
 1077         }
 1078         return 0;
 1079 }
 1080 
 1081 static int
 1082 rn_walktree(h, f, w)
 1083         struct radix_node_head *h;
 1084         walktree_f_t *f;
 1085         void *w;
 1086 {
 1087         int error;
 1088         struct radix_node *base, *next;
 1089         register struct radix_node *rn = h->rnh_treetop;
 1090         /*
 1091          * This gets complicated because we may delete the node
 1092          * while applying the function f to it, so we need to calculate
 1093          * the successor node in advance.
 1094          */
 1095 
 1096         /* First time through node, go left */
 1097         while (rn->rn_bit >= 0)
 1098                 rn = rn->rn_left;
 1099         for (;;) {
 1100                 base = rn;
 1101                 /* If at right child go back up, otherwise, go right */
 1102                 while (rn->rn_parent->rn_right == rn
 1103                        && (rn->rn_flags & RNF_ROOT) == 0)
 1104                         rn = rn->rn_parent;
 1105                 /* Find the next *leaf* since next node might vanish, too */
 1106                 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
 1107                         rn = rn->rn_left;
 1108                 next = rn;
 1109                 /* Process leaves */
 1110                 while ((rn = base)) {
 1111                         base = rn->rn_dupedkey;
 1112                         if (!(rn->rn_flags & RNF_ROOT)
 1113                             && (error = (*f)(rn, w)))
 1114                                 return (error);
 1115                 }
 1116                 rn = next;
 1117                 if (rn->rn_flags & RNF_ROOT)
 1118                         return (0);
 1119         }
 1120         /* NOTREACHED */
 1121 }
 1122 
 1123 /*
 1124  * Allocate and initialize an empty tree. This has 3 nodes, which are
 1125  * part of the radix_node_head (in the order <left,root,right>) and are
 1126  * marked RNF_ROOT so they cannot be freed.
 1127  * The leaves have all-zero and all-one keys, with significant
 1128  * bits starting at 'off'.
 1129  * Return 1 on success, 0 on error.
 1130  */
 1131 int
 1132 rn_inithead(head, off)
 1133         void **head;
 1134         int off;
 1135 {
 1136         register struct radix_node_head *rnh;
 1137         register struct radix_node *t, *tt, *ttt;
 1138         if (*head)
 1139                 return (1);
 1140         R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh));
 1141         if (rnh == 0)
 1142                 return (0);
 1143 #ifdef _KERNEL
 1144         RADIX_NODE_HEAD_LOCK_INIT(rnh);
 1145 #endif
 1146         *head = rnh;
 1147         t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
 1148         ttt = rnh->rnh_nodes + 2;
 1149         t->rn_right = ttt;
 1150         t->rn_parent = t;
 1151         tt = t->rn_left;        /* ... which in turn is rnh->rnh_nodes */
 1152         tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
 1153         tt->rn_bit = -1 - off;
 1154         *ttt = *tt;
 1155         ttt->rn_key = rn_ones;
 1156         rnh->rnh_addaddr = rn_addroute;
 1157         rnh->rnh_deladdr = rn_delete;
 1158         rnh->rnh_matchaddr = rn_match;
 1159         rnh->rnh_lookup = rn_lookup;
 1160         rnh->rnh_walktree = rn_walktree;
 1161         rnh->rnh_walktree_from = rn_walktree_from;
 1162         rnh->rnh_treetop = t;
 1163         return (1);
 1164 }
 1165 
 1166 int
 1167 rn_detachhead(void **head)
 1168 {
 1169         struct radix_node_head *rnh;
 1170 
 1171         KASSERT((head != NULL && *head != NULL),
 1172             ("%s: head already freed", __func__));
 1173         rnh = *head;
 1174         
 1175         /* Free <left,root,right> nodes. */
 1176         Free(rnh);
 1177 
 1178         *head = NULL;
 1179         return (1);
 1180 }
 1181 
 1182 void
 1183 rn_init(int maxk)
 1184 {
 1185         char *cp, *cplim;
 1186 
 1187         max_keylen = maxk;
 1188         if (max_keylen == 0) {
 1189                 log(LOG_ERR,
 1190                     "rn_init: radix functions require max_keylen be set\n");
 1191                 return;
 1192         }
 1193         R_Malloc(rn_zeros, char *, 3 * max_keylen);
 1194         if (rn_zeros == NULL)
 1195                 panic("rn_init");
 1196         bzero(rn_zeros, 3 * max_keylen);
 1197         rn_ones = cp = rn_zeros + max_keylen;
 1198         addmask_key = cplim = rn_ones + max_keylen;
 1199         while (cp < cplim)
 1200                 *cp++ = -1;
 1201         if (rn_inithead((void **)(void *)&mask_rnhead, 0) == 0)
 1202                 panic("rn_init 2");
 1203 }

Cache object: b6fc8e9a24c971ed6a20127b2bbd7e68


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.