contrib/pf/pfctl/pfctl_optimize.c

   1 /*      $OpenBSD: pfctl_optimize.c,v 1.13 2006/10/31 14:17:45 mcbride Exp $ */
   2
   3 /*
   4  * Copyright (c) 2004 Mike Frantzen <frantzen@openbsd.org>
   5  *
   6  * Permission to use, copy, modify, and distribute this software for any
   7  * purpose with or without fee is hereby granted, provided that the above
   8  * copyright notice and this permission notice appear in all copies.
   9  *
  10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  17  */
  18
  19 #include <sys/cdefs.h>
  20 __FBSDID("$FreeBSD$");
  21
  22 #include <sys/types.h>
  23 #include <sys/ioctl.h>
  24 #include <sys/socket.h>
  25
  26 #include <net/if.h>
  27 #include <net/pfvar.h>
  28
  29 #include <netinet/in.h>
  30 #include <arpa/inet.h>
  31
  32 #include <assert.h>
  33 #include <ctype.h>
  34 #include <err.h>
  35 #include <errno.h>
  36 #include <stddef.h>
  37 #include <stdio.h>
  38 #include <stdlib.h>
  39 #include <string.h>
  40
  41 #include "pfctl_parser.h"
  42 #include "pfctl.h"
  43
  44 /* The size at which a table becomes faster than individual rules */
  45 #define TABLE_THRESHOLD         6
  46
  47
  48 /* #define OPT_DEBUG    1 */
  49 #ifdef OPT_DEBUG
  50 # define DEBUG(str, v...) \
  51         printf("%s: " str "\n", __FUNCTION__ , ## v)
  52 #else
  53 # define DEBUG(str, v...) ((void)0)
  54 #endif
  55
  56
  57 /*
  58  * A container that lets us sort a superblock to optimize the skip step jumps
  59  */
  60 struct pf_skip_step {
  61         int                             ps_count;       /* number of items */
  62         TAILQ_HEAD( , pf_opt_rule)      ps_rules;
  63         TAILQ_ENTRY(pf_skip_step)       ps_entry;
  64 };
  65
  66
  67 /*
  68  * A superblock is a block of adjacent rules of similar action.  If there
  69  * are five PASS rules in a row, they all become members of a superblock.
  70  * Once we have a superblock, we are free to re-order any rules within it
  71  * in order to improve performance; if a packet is passed, it doesn't matter
  72  * who passed it.
  73  */
  74 struct superblock {
  75         TAILQ_HEAD( , pf_opt_rule)               sb_rules;
  76         TAILQ_ENTRY(superblock)                  sb_entry;
  77         struct superblock                       *sb_profiled_block;
  78         TAILQ_HEAD(skiplist, pf_skip_step)       sb_skipsteps[PF_SKIP_COUNT];
  79 };
  80 TAILQ_HEAD(superblocks, superblock);
  81
  82
  83 /*
  84  * Description of the PF rule structure.
  85  */
  86 enum {
  87     BARRIER,    /* the presence of the field puts the rule in it's own block */
  88     BREAK,      /* the field may not differ between rules in a superblock */
  89     NOMERGE,    /* the field may not differ between rules when combined */
  90     COMBINED,   /* the field may itself be combined with other rules */
  91     DC,         /* we just don't care about the field */
  92     NEVER};     /* we should never see this field set?!? */
  93 struct pf_rule_field {
  94         const char      *prf_name;
  95         int              prf_type;
  96         size_t           prf_offset;
  97         size_t           prf_size;
  98 } pf_rule_desc[] = {
  99 #define PF_RULE_FIELD(field, ty)        \
 100     {#field,                            \
 101     ty,                                 \
 102     offsetof(struct pf_rule, field),    \
 103     sizeof(((struct pf_rule *)0)->field)}
 104
 105
 106     /*
 107      * The presence of these fields in a rule put the rule in it's own
 108      * superblock.  Thus it will not be optimized.  It also prevents the
 109      * rule from being re-ordered at all.
 110      */
 111     PF_RULE_FIELD(label,                BARRIER),
 112     PF_RULE_FIELD(prob,                 BARRIER),
 113     PF_RULE_FIELD(max_states,           BARRIER),
 114     PF_RULE_FIELD(max_src_nodes,        BARRIER),
 115     PF_RULE_FIELD(max_src_states,       BARRIER),
 116     PF_RULE_FIELD(max_src_conn,         BARRIER),
 117     PF_RULE_FIELD(max_src_conn_rate,    BARRIER),
 118     PF_RULE_FIELD(anchor,               BARRIER),       /* for now */
 119
 120     /*
 121      * These fields must be the same between all rules in the same superblock.
 122      * These rules are allowed to be re-ordered but only among like rules.
 123      * For instance we can re-order all 'tag "foo"' rules because they have the
 124      * same tag.  But we can not re-order between a 'tag "foo"' and a
 125      * 'tag "bar"' since that would change the meaning of the ruleset.
 126      */
 127     PF_RULE_FIELD(tagname,              BREAK),
 128     PF_RULE_FIELD(keep_state,           BREAK),
 129     PF_RULE_FIELD(qname,                BREAK),
 130     PF_RULE_FIELD(pqname,               BREAK),
 131     PF_RULE_FIELD(rt,                   BREAK),
 132     PF_RULE_FIELD(allow_opts,           BREAK),
 133     PF_RULE_FIELD(rule_flag,            BREAK),
 134     PF_RULE_FIELD(action,               BREAK),
 135     PF_RULE_FIELD(log,                  BREAK),
 136     PF_RULE_FIELD(quick,                BREAK),
 137     PF_RULE_FIELD(return_ttl,           BREAK),
 138     PF_RULE_FIELD(overload_tblname,     BREAK),
 139     PF_RULE_FIELD(flush,                BREAK),
 140     PF_RULE_FIELD(rpool,                BREAK),
 141     PF_RULE_FIELD(logif,                BREAK),
 142
 143     /*
 144      * Any fields not listed in this structure act as BREAK fields
 145      */
 146
 147
 148     /*
 149      * These fields must not differ when we merge two rules together but
 150      * their difference isn't enough to put the rules in different superblocks.
 151      * There are no problems re-ordering any rules with these fields.
 152      */
 153     PF_RULE_FIELD(af,                   NOMERGE),
 154     PF_RULE_FIELD(ifnot,                NOMERGE),
 155     PF_RULE_FIELD(ifname,               NOMERGE),       /* hack for IF groups */
 156     PF_RULE_FIELD(match_tag_not,        NOMERGE),
 157     PF_RULE_FIELD(match_tagname,        NOMERGE),
 158     PF_RULE_FIELD(os_fingerprint,       NOMERGE),
 159     PF_RULE_FIELD(timeout,              NOMERGE),
 160     PF_RULE_FIELD(return_icmp,          NOMERGE),
 161     PF_RULE_FIELD(return_icmp6,         NOMERGE),
 162     PF_RULE_FIELD(uid,                  NOMERGE),
 163     PF_RULE_FIELD(gid,                  NOMERGE),
 164     PF_RULE_FIELD(direction,            NOMERGE),
 165     PF_RULE_FIELD(proto,                NOMERGE),
 166     PF_RULE_FIELD(type,                 NOMERGE),
 167     PF_RULE_FIELD(code,                 NOMERGE),
 168     PF_RULE_FIELD(flags,                NOMERGE),
 169     PF_RULE_FIELD(flagset,              NOMERGE),
 170     PF_RULE_FIELD(tos,                  NOMERGE),
 171     PF_RULE_FIELD(src.port,             NOMERGE),
 172     PF_RULE_FIELD(dst.port,             NOMERGE),
 173     PF_RULE_FIELD(src.port_op,          NOMERGE),
 174     PF_RULE_FIELD(dst.port_op,          NOMERGE),
 175     PF_RULE_FIELD(src.neg,              NOMERGE),
 176     PF_RULE_FIELD(dst.neg,              NOMERGE),
 177
 178     /* These fields can be merged */
 179     PF_RULE_FIELD(src.addr,             COMBINED),
 180     PF_RULE_FIELD(dst.addr,             COMBINED),
 181
 182     /* We just don't care about these fields.  They're set by the kernel */
 183     PF_RULE_FIELD(skip,                 DC),
 184     PF_RULE_FIELD(evaluations,          DC),
 185     PF_RULE_FIELD(packets,              DC),
 186     PF_RULE_FIELD(bytes,                DC),
 187     PF_RULE_FIELD(kif,                  DC),
 188     PF_RULE_FIELD(states,               DC),
 189     PF_RULE_FIELD(src_nodes,            DC),
 190     PF_RULE_FIELD(nr,                   DC),
 191     PF_RULE_FIELD(entries,              DC),
 192     PF_RULE_FIELD(qid,                  DC),
 193     PF_RULE_FIELD(pqid,                 DC),
 194     PF_RULE_FIELD(anchor_relative,      DC),
 195     PF_RULE_FIELD(anchor_wildcard,      DC),
 196     PF_RULE_FIELD(tag,                  DC),
 197     PF_RULE_FIELD(match_tag,            DC),
 198     PF_RULE_FIELD(overload_tbl,         DC),
 199
 200     /* These fields should never be set in a PASS/BLOCK rule */
 201     PF_RULE_FIELD(natpass,              NEVER),
 202     PF_RULE_FIELD(max_mss,              NEVER),
 203     PF_RULE_FIELD(min_ttl,              NEVER),
 204 };
 205
 206
 207
 208 int     add_opt_table(struct pfctl *, struct pf_opt_tbl **, sa_family_t,
 209             struct pf_rule_addr *);
 210 int     addrs_combineable(struct pf_rule_addr *, struct pf_rule_addr *);
 211 int     addrs_equal(struct pf_rule_addr *, struct pf_rule_addr *);
 212 int     block_feedback(struct pfctl *, struct superblock *);
 213 int     combine_rules(struct pfctl *, struct superblock *);
 214 void    comparable_rule(struct pf_rule *, const struct pf_rule *, int);
 215 int     construct_superblocks(struct pfctl *, struct pf_opt_queue *,
 216             struct superblocks *);
 217 void    exclude_supersets(struct pf_rule *, struct pf_rule *);
 218 int     interface_group(const char *);
 219 int     load_feedback_profile(struct pfctl *, struct superblocks *);
 220 int     optimize_superblock(struct pfctl *, struct superblock *);
 221 int     pf_opt_create_table(struct pfctl *, struct pf_opt_tbl *);
 222 void    remove_from_skipsteps(struct skiplist *, struct superblock *,
 223             struct pf_opt_rule *, struct pf_skip_step *);
 224 int     remove_identical_rules(struct pfctl *, struct superblock *);
 225 int     reorder_rules(struct pfctl *, struct superblock *, int);
 226 int     rules_combineable(struct pf_rule *, struct pf_rule *);
 227 void    skip_append(struct superblock *, int, struct pf_skip_step *,
 228             struct pf_opt_rule *);
 229 int     skip_compare(int, struct pf_skip_step *, struct pf_opt_rule *);
 230 void    skip_init(void);
 231 int     skip_cmp_af(struct pf_rule *, struct pf_rule *);
 232 int     skip_cmp_dir(struct pf_rule *, struct pf_rule *);
 233 int     skip_cmp_dst_addr(struct pf_rule *, struct pf_rule *);
 234 int     skip_cmp_dst_port(struct pf_rule *, struct pf_rule *);
 235 int     skip_cmp_ifp(struct pf_rule *, struct pf_rule *);
 236 int     skip_cmp_proto(struct pf_rule *, struct pf_rule *);
 237 int     skip_cmp_src_addr(struct pf_rule *, struct pf_rule *);
 238 int     skip_cmp_src_port(struct pf_rule *, struct pf_rule *);
 239 int     superblock_inclusive(struct superblock *, struct pf_opt_rule *);
 240 void    superblock_free(struct pfctl *, struct superblock *);
 241
 242
 243 int (*skip_comparitors[PF_SKIP_COUNT])(struct pf_rule *, struct pf_rule *);
 244 const char *skip_comparitors_names[PF_SKIP_COUNT];
 245 #define PF_SKIP_COMPARITORS {                           \
 246     { "ifp", PF_SKIP_IFP, skip_cmp_ifp },               \
 247     { "dir", PF_SKIP_DIR, skip_cmp_dir },               \
 248     { "af", PF_SKIP_AF, skip_cmp_af },                  \
 249     { "proto", PF_SKIP_PROTO, skip_cmp_proto },         \
 250     { "saddr", PF_SKIP_SRC_ADDR, skip_cmp_src_addr },   \
 251     { "sport", PF_SKIP_SRC_PORT, skip_cmp_src_port },   \
 252     { "daddr", PF_SKIP_DST_ADDR, skip_cmp_dst_addr },   \
 253     { "dport", PF_SKIP_DST_PORT, skip_cmp_dst_port }    \
 254 }
 255
 256 struct pfr_buffer table_buffer;
 257 int table_identifier;
 258
 259
 260 int
 261 pfctl_optimize_ruleset(struct pfctl *pf, struct pf_ruleset *rs)
 262 {
 263         struct superblocks superblocks;
 264         struct pf_opt_queue opt_queue;
 265         struct superblock *block;
 266         struct pf_opt_rule *por;
 267         struct pf_rule *r;
 268         struct pf_rulequeue *old_rules;
 269
 270         DEBUG("optimizing ruleset");
 271         memset(&table_buffer, 0, sizeof(table_buffer));
 272         skip_init();
 273         TAILQ_INIT(&opt_queue);
 274
 275         old_rules = rs->rules[PF_RULESET_FILTER].active.ptr;
 276         rs->rules[PF_RULESET_FILTER].active.ptr =
 277             rs->rules[PF_RULESET_FILTER].inactive.ptr;
 278         rs->rules[PF_RULESET_FILTER].inactive.ptr = old_rules;
 279
 280         /*
 281          * XXX expanding the pf_opt_rule format throughout pfctl might allow
 282          * us to avoid all this copying.
 283          */
 284         while ((r = TAILQ_FIRST(rs->rules[PF_RULESET_FILTER].inactive.ptr))
 285             != NULL) {
 286                 TAILQ_REMOVE(rs->rules[PF_RULESET_FILTER].inactive.ptr, r,
 287                     entries);
 288                 if ((por = calloc(1, sizeof(*por))) == NULL)
 289                         err(1, "calloc");
 290                 memcpy(&por->por_rule, r, sizeof(*r));
 291                 if (TAILQ_FIRST(&r->rpool.list) != NULL) {
 292                         TAILQ_INIT(&por->por_rule.rpool.list);
 293                         pfctl_move_pool(&r->rpool, &por->por_rule.rpool);
 294                 } else
 295                         bzero(&por->por_rule.rpool,
 296                             sizeof(por->por_rule.rpool));
 297
 298
 299                 TAILQ_INSERT_TAIL(&opt_queue, por, por_entry);
 300         }
 301
 302         TAILQ_INIT(&superblocks);
 303         if (construct_superblocks(pf, &opt_queue, &superblocks))
 304                 goto error;
 305
 306         if (pf->optimize & PF_OPTIMIZE_PROFILE) {
 307                 if (load_feedback_profile(pf, &superblocks))
 308                         goto error;
 309         }
 310
 311         TAILQ_FOREACH(block, &superblocks, sb_entry) {
 312                 if (optimize_superblock(pf, block))
 313                         goto error;
 314         }
 315
 316         rs->anchor->refcnt = 0;
 317         while ((block = TAILQ_FIRST(&superblocks))) {
 318                 TAILQ_REMOVE(&superblocks, block, sb_entry);
 319
 320                 while ((por = TAILQ_FIRST(&block->sb_rules))) {
 321                         TAILQ_REMOVE(&block->sb_rules, por, por_entry);
 322                         por->por_rule.nr = rs->anchor->refcnt++;
 323                         if ((r = calloc(1, sizeof(*r))) == NULL)
 324                                 err(1, "calloc");
 325                         memcpy(r, &por->por_rule, sizeof(*r));
 326                         TAILQ_INIT(&r->rpool.list);
 327                         pfctl_move_pool(&por->por_rule.rpool, &r->rpool);
 328                         TAILQ_INSERT_TAIL(
 329                             rs->rules[PF_RULESET_FILTER].active.ptr,
 330                             r, entries);
 331                         free(por);
 332                 }
 333                 free(block);
 334         }
 335
 336         return (0);
 337
 338 error:
 339         while ((por = TAILQ_FIRST(&opt_queue))) {
 340                 TAILQ_REMOVE(&opt_queue, por, por_entry);
 341                 if (por->por_src_tbl) {
 342                         pfr_buf_clear(por->por_src_tbl->pt_buf);
 343                         free(por->por_src_tbl->pt_buf);
 344                         free(por->por_src_tbl);
 345                 }
 346                 if (por->por_dst_tbl) {
 347                         pfr_buf_clear(por->por_dst_tbl->pt_buf);
 348                         free(por->por_dst_tbl->pt_buf);
 349                         free(por->por_dst_tbl);
 350                 }
 351                 free(por);
 352         }
 353         while ((block = TAILQ_FIRST(&superblocks))) {
 354                 TAILQ_REMOVE(&superblocks, block, sb_entry);
 355                 superblock_free(pf, block);
 356         }
 357         return (1);
 358 }
 359
 360
 361 /*
 362  * Go ahead and optimize a superblock
 363  */
 364 int
 365 optimize_superblock(struct pfctl *pf, struct superblock *block)
 366 {
 367 #ifdef OPT_DEBUG
 368         struct pf_opt_rule *por;
 369 #endif /* OPT_DEBUG */
 370
 371         /* We have a few optimization passes:
 372          *   1) remove duplicate rules or rules that are a subset of other
 373          *      rules
 374          *   2) combine otherwise identical rules with different IP addresses
 375          *      into a single rule and put the addresses in a table.
 376          *   3) re-order the rules to improve kernel skip steps
 377          *   4) re-order the 'quick' rules based on feedback from the
 378          *      active ruleset statistics
 379          *
 380          * XXX combine_rules() doesn't combine v4 and v6 rules.  would just
 381          *     have to keep af in the table container, make af 'COMBINE' and
 382          *     twiddle the af on the merged rule
 383          * XXX maybe add a weighting to the metric on skipsteps when doing
 384          *     reordering.  sometimes two sequential tables will be better
 385          *     that four consecutive interfaces.
 386          * XXX need to adjust the skipstep count of everything after PROTO,
 387          *     since they aren't actually checked on a proto mismatch in
 388          *     pf_test_{tcp, udp, icmp}()
 389          * XXX should i treat proto=0, af=0 or dir=0 special in skepstep
 390          *     calculation since they are a DC?
 391          * XXX keep last skiplist of last superblock to influence this
 392          *     superblock.  '5 inet6 log' should make '3 inet6' come before '4
 393          *     inet' in the next superblock.
 394          * XXX would be useful to add tables for ports
 395          * XXX we can also re-order some mutually exclusive superblocks to
 396          *     try merging superblocks before any of these optimization passes.
 397          *     for instance a single 'log in' rule in the middle of non-logging
 398          *     out rules.
 399          */
 400
 401         /* shortcut.  there will be alot of 1-rule superblocks */
 402         if (!TAILQ_NEXT(TAILQ_FIRST(&block->sb_rules), por_entry))
 403                 return (0);
 404
 405 #ifdef OPT_DEBUG
 406         printf("--- Superblock ---\n");
 407         TAILQ_FOREACH(por, &block->sb_rules, por_entry) {
 408                 printf("  ");
 409                 print_rule(&por->por_rule, por->por_rule.anchor ?
 410                     por->por_rule.anchor->name : "", 1);
 411         }
 412 #endif /* OPT_DEBUG */
 413
 414
 415         if (remove_identical_rules(pf, block))
 416                 return (1);
 417         if (combine_rules(pf, block))
 418                 return (1);
 419         if ((pf->optimize & PF_OPTIMIZE_PROFILE) &&
 420             TAILQ_FIRST(&block->sb_rules)->por_rule.quick &&
 421             block->sb_profiled_block) {
 422                 if (block_feedback(pf, block))
 423                         return (1);
 424         } else if (reorder_rules(pf, block, 0)) {
 425                 return (1);
 426         }
 427
 428         /*
 429          * Don't add any optimization passes below reorder_rules().  It will
 430          * have divided superblocks into smaller blocks for further refinement
 431          * and doesn't put them back together again.  What once was a true
 432          * superblock might have been split into multiple superblocks.
 433          */
 434
 435 #ifdef OPT_DEBUG
 436         printf("--- END Superblock ---\n");
 437 #endif /* OPT_DEBUG */
 438         return (0);
 439 }
 440
 441
 442 /*
 443  * Optimization pass #1: remove identical rules
 444  */
 445 int
 446 remove_identical_rules(struct pfctl *pf, struct superblock *block)
 447 {
 448         struct pf_opt_rule *por1, *por2, *por_next, *por2_next;
 449         struct pf_rule a, a2, b, b2;
 450
 451         for (por1 = TAILQ_FIRST(&block->sb_rules); por1; por1 = por_next) {
 452                 por_next = TAILQ_NEXT(por1, por_entry);
 453                 for (por2 = por_next; por2; por2 = por2_next) {
 454                         por2_next = TAILQ_NEXT(por2, por_entry);
 455                         comparable_rule(&a, &por1->por_rule, DC);
 456                         comparable_rule(&b, &por2->por_rule, DC);
 457                         memcpy(&a2, &a, sizeof(a2));
 458                         memcpy(&b2, &b, sizeof(b2));
 459
 460                         exclude_supersets(&a, &b);
 461                         exclude_supersets(&b2, &a2);
 462                         if (memcmp(&a, &b, sizeof(a)) == 0) {
 463                                 DEBUG("removing identical rule  nr%d = *nr%d*",
 464                                     por1->por_rule.nr, por2->por_rule.nr);
 465                                 TAILQ_REMOVE(&block->sb_rules, por2, por_entry);
 466                                 if (por_next == por2)
 467                                         por_next = TAILQ_NEXT(por1, por_entry);
 468                                 free(por2);
 469                         } else if (memcmp(&a2, &b2, sizeof(a2)) == 0) {
 470                                 DEBUG("removing identical rule  *nr%d* = nr%d",
 471                                     por1->por_rule.nr, por2->por_rule.nr);
 472                                 TAILQ_REMOVE(&block->sb_rules, por1, por_entry);
 473                                 free(por1);
 474                                 break;
 475                         }
 476                 }
 477         }
 478
 479         return (0);
 480 }
 481
 482
 483 /*
 484  * Optimization pass #2: combine similar rules with different addresses
 485  * into a single rule and a table
 486  */
 487 int
 488 combine_rules(struct pfctl *pf, struct superblock *block)
 489 {
 490         struct pf_opt_rule *p1, *p2, *por_next;
 491         int src_eq, dst_eq;
 492
 493         if ((pf->loadopt & PFCTL_FLAG_TABLE) == 0) {
 494                 warnx("Must enable table loading for optimizations");
 495                 return (1);
 496         }
 497
 498         /* First we make a pass to combine the rules.  O(n log n) */
 499         TAILQ_FOREACH(p1, &block->sb_rules, por_entry) {
 500                 for (p2 = TAILQ_NEXT(p1, por_entry); p2; p2 = por_next) {
 501                         por_next = TAILQ_NEXT(p2, por_entry);
 502
 503                         src_eq = addrs_equal(&p1->por_rule.src,
 504                             &p2->por_rule.src);
 505                         dst_eq = addrs_equal(&p1->por_rule.dst,
 506                             &p2->por_rule.dst);
 507
 508                         if (src_eq && !dst_eq && p1->por_src_tbl == NULL &&
 509                             p2->por_dst_tbl == NULL &&
 510                             p2->por_src_tbl == NULL &&
 511                             rules_combineable(&p1->por_rule, &p2->por_rule) &&
 512                             addrs_combineable(&p1->por_rule.dst,
 513                             &p2->por_rule.dst)) {
 514                                 DEBUG("can combine rules  nr%d = nr%d",
 515                                     p1->por_rule.nr, p2->por_rule.nr);
 516                                 if (p1->por_dst_tbl == NULL &&
 517                                     add_opt_table(pf, &p1->por_dst_tbl,
 518                                     p1->por_rule.af, &p1->por_rule.dst))
 519                                         return (1);
 520                                 if (add_opt_table(pf, &p1->por_dst_tbl,
 521                                     p1->por_rule.af, &p2->por_rule.dst))
 522                                         return (1);
 523                                 p2->por_dst_tbl = p1->por_dst_tbl;
 524                                 if (p1->por_dst_tbl->pt_rulecount >=
 525                                     TABLE_THRESHOLD) {
 526                                         TAILQ_REMOVE(&block->sb_rules, p2,
 527                                             por_entry);
 528                                         free(p2);
 529                                 }
 530                         } else if (!src_eq && dst_eq && p1->por_dst_tbl == NULL
 531                             && p2->por_src_tbl == NULL &&
 532                             p2->por_dst_tbl == NULL &&
 533                             rules_combineable(&p1->por_rule, &p2->por_rule) &&
 534                             addrs_combineable(&p1->por_rule.src,
 535                             &p2->por_rule.src)) {
 536                                 DEBUG("can combine rules  nr%d = nr%d",
 537                                     p1->por_rule.nr, p2->por_rule.nr);
 538                                 if (p1->por_src_tbl == NULL &&
 539                                     add_opt_table(pf, &p1->por_src_tbl,
 540                                     p1->por_rule.af, &p1->por_rule.src))
 541                                         return (1);
 542                                 if (add_opt_table(pf, &p1->por_src_tbl,
 543                                     p1->por_rule.af, &p2->por_rule.src))
 544                                         return (1);
 545                                 p2->por_src_tbl = p1->por_src_tbl;
 546                                 if (p1->por_src_tbl->pt_rulecount >=
 547                                     TABLE_THRESHOLD) {
 548                                         TAILQ_REMOVE(&block->sb_rules, p2,
 549                                             por_entry);
 550                                         free(p2);
 551                                 }
 552                         }
 553                 }
 554         }
 555
 556
 557         /*
 558          * Then we make a final pass to create a valid table name and
 559          * insert the name into the rules.
 560          */
 561         for (p1 = TAILQ_FIRST(&block->sb_rules); p1; p1 = por_next) {
 562                 por_next = TAILQ_NEXT(p1, por_entry);
 563                 assert(p1->por_src_tbl == NULL || p1->por_dst_tbl == NULL);
 564
 565                 if (p1->por_src_tbl && p1->por_src_tbl->pt_rulecount >=
 566                     TABLE_THRESHOLD) {
 567                         if (p1->por_src_tbl->pt_generated) {
 568                                 /* This rule is included in a table */
 569                                 TAILQ_REMOVE(&block->sb_rules, p1, por_entry);
 570                                 free(p1);
 571                                 continue;
 572                         }
 573                         p1->por_src_tbl->pt_generated = 1;
 574
 575                         if ((pf->opts & PF_OPT_NOACTION) == 0 &&
 576                             pf_opt_create_table(pf, p1->por_src_tbl))
 577                                 return (1);
 578
 579                         pf->tdirty = 1;
 580
 581                         if (pf->opts & PF_OPT_VERBOSE)
 582                                 print_tabledef(p1->por_src_tbl->pt_name,
 583                                     PFR_TFLAG_CONST, 1,
 584                                     &p1->por_src_tbl->pt_nodes);
 585
 586                         memset(&p1->por_rule.src.addr, 0,
 587                             sizeof(p1->por_rule.src.addr));
 588                         p1->por_rule.src.addr.type = PF_ADDR_TABLE;
 589                         strlcpy(p1->por_rule.src.addr.v.tblname,
 590                             p1->por_src_tbl->pt_name,
 591                             sizeof(p1->por_rule.src.addr.v.tblname));
 592
 593                         pfr_buf_clear(p1->por_src_tbl->pt_buf);
 594                         free(p1->por_src_tbl->pt_buf);
 595                         p1->por_src_tbl->pt_buf = NULL;
 596                 }
 597                 if (p1->por_dst_tbl && p1->por_dst_tbl->pt_rulecount >=
 598                     TABLE_THRESHOLD) {
 599                         if (p1->por_dst_tbl->pt_generated) {
 600                                 /* This rule is included in a table */
 601                                 TAILQ_REMOVE(&block->sb_rules, p1, por_entry);
 602                                 free(p1);
 603                                 continue;
 604                         }
 605                         p1->por_dst_tbl->pt_generated = 1;
 606
 607                         if ((pf->opts & PF_OPT_NOACTION) == 0 &&
 608                             pf_opt_create_table(pf, p1->por_dst_tbl))
 609                                 return (1);
 610                         pf->tdirty = 1;
 611
 612                         if (pf->opts & PF_OPT_VERBOSE)
 613                                 print_tabledef(p1->por_dst_tbl->pt_name,
 614                                     PFR_TFLAG_CONST, 1,
 615                                     &p1->por_dst_tbl->pt_nodes);
 616
 617                         memset(&p1->por_rule.dst.addr, 0,
 618                             sizeof(p1->por_rule.dst.addr));
 619                         p1->por_rule.dst.addr.type = PF_ADDR_TABLE;
 620                         strlcpy(p1->por_rule.dst.addr.v.tblname,
 621                             p1->por_dst_tbl->pt_name,
 622                             sizeof(p1->por_rule.dst.addr.v.tblname));
 623
 624                         pfr_buf_clear(p1->por_dst_tbl->pt_buf);
 625                         free(p1->por_dst_tbl->pt_buf);
 626                         p1->por_dst_tbl->pt_buf = NULL;
 627                 }
 628         }
 629
 630         return (0);
 631 }
 632
 633
 634 /*
 635  * Optimization pass #3: re-order rules to improve skip steps
 636  */
 637 int
 638 reorder_rules(struct pfctl *pf, struct superblock *block, int depth)
 639 {
 640         struct superblock *newblock;
 641         struct pf_skip_step *skiplist;
 642         struct pf_opt_rule *por;
 643         int i, largest, largest_list, rule_count = 0;
 644         TAILQ_HEAD( , pf_opt_rule) head;
 645
 646         /*
 647          * Calculate the best-case skip steps.  We put each rule in a list
 648          * of other rules with common fields
 649          */
 650         for (i = 0; i < PF_SKIP_COUNT; i++) {
 651                 TAILQ_FOREACH(por, &block->sb_rules, por_entry) {
 652                         TAILQ_FOREACH(skiplist, &block->sb_skipsteps[i],
 653                             ps_entry) {
 654                                 if (skip_compare(i, skiplist, por) == 0)
 655                                         break;
 656                         }
 657                         if (skiplist == NULL) {
 658                                 if ((skiplist = calloc(1, sizeof(*skiplist))) ==
 659                                     NULL)
 660                                         err(1, "calloc");
 661                                 TAILQ_INIT(&skiplist->ps_rules);
 662                                 TAILQ_INSERT_TAIL(&block->sb_skipsteps[i],
 663                                     skiplist, ps_entry);
 664                         }
 665                         skip_append(block, i, skiplist, por);
 666                 }
 667         }
 668
 669         TAILQ_FOREACH(por, &block->sb_rules, por_entry)
 670                 rule_count++;
 671
 672         /*
 673          * Now we're going to ignore any fields that are identical between
 674          * all of the rules in the superblock and those fields which differ
 675          * between every rule in the superblock.
 676          */
 677         largest = 0;
 678         for (i = 0; i < PF_SKIP_COUNT; i++) {
 679                 skiplist = TAILQ_FIRST(&block->sb_skipsteps[i]);
 680                 if (skiplist->ps_count == rule_count) {
 681                         DEBUG("(%d) original skipstep '%s' is all rules",
 682                             depth, skip_comparitors_names[i]);
 683                         skiplist->ps_count = 0;
 684                 } else if (skiplist->ps_count == 1) {
 685                         skiplist->ps_count = 0;
 686                 } else {
 687                         DEBUG("(%d) original skipstep '%s' largest jump is %d",
 688                             depth, skip_comparitors_names[i],
 689                             skiplist->ps_count);
 690                         if (skiplist->ps_count > largest)
 691                                 largest = skiplist->ps_count;
 692                 }
 693         }
 694         if (largest == 0) {
 695                 /* Ugh.  There is NO commonality in the superblock on which
 696                  * optimize the skipsteps optimization.
 697                  */
 698                 goto done;
 699         }
 700
 701         /*
 702          * Now we're going to empty the superblock rule list and re-create
 703          * it based on a more optimal skipstep order.
 704          */
 705         TAILQ_INIT(&head);
 706         while ((por = TAILQ_FIRST(&block->sb_rules))) {
 707                 TAILQ_REMOVE(&block->sb_rules, por, por_entry);
 708                 TAILQ_INSERT_TAIL(&head, por, por_entry);
 709         }
 710
 711
 712         while (!TAILQ_EMPTY(&head)) {
 713                 largest = 1;
 714
 715                 /*
 716                  * Find the most useful skip steps remaining
 717                  */
 718                 for (i = 0; i < PF_SKIP_COUNT; i++) {
 719                         skiplist = TAILQ_FIRST(&block->sb_skipsteps[i]);
 720                         if (skiplist->ps_count > largest) {
 721                                 largest = skiplist->ps_count;
 722                                 largest_list = i;
 723                         }
 724                 }
 725
 726                 if (largest <= 1) {
 727                         /*
 728                          * Nothing useful left.  Leave remaining rules in order.
 729                          */
 730                         DEBUG("(%d) no more commonality for skip steps", depth);
 731                         while ((por = TAILQ_FIRST(&head))) {
 732                                 TAILQ_REMOVE(&head, por, por_entry);
 733                                 TAILQ_INSERT_TAIL(&block->sb_rules, por,
 734                                     por_entry);
 735                         }
 736                 } else {
 737                         /*
 738                          * There is commonality.  Extract those common rules
 739                          * and place them in the ruleset adjacent to each
 740                          * other.
 741                          */
 742                         skiplist = TAILQ_FIRST(&block->sb_skipsteps[
 743                             largest_list]);
 744                         DEBUG("(%d) skipstep '%s' largest jump is %d @ #%d",
 745                             depth, skip_comparitors_names[largest_list],
 746                             largest, TAILQ_FIRST(&TAILQ_FIRST(&block->
 747                             sb_skipsteps [largest_list])->ps_rules)->
 748                             por_rule.nr);
 749                         TAILQ_REMOVE(&block->sb_skipsteps[largest_list],
 750                             skiplist, ps_entry);
 751
 752
 753                         /*
 754                          * There may be further commonality inside these
 755                          * rules.  So we'll split them off into they're own
 756                          * superblock and pass it back into the optimizer.
 757                          */
 758                         if (skiplist->ps_count > 2) {
 759                                 if ((newblock = calloc(1, sizeof(*newblock)))
 760                                     == NULL) {
 761                                         warn("calloc");
 762                                         return (1);
 763                                 }
 764                                 TAILQ_INIT(&newblock->sb_rules);
 765                                 for (i = 0; i < PF_SKIP_COUNT; i++)
 766                                         TAILQ_INIT(&newblock->sb_skipsteps[i]);
 767                                 TAILQ_INSERT_BEFORE(block, newblock, sb_entry);
 768                                 DEBUG("(%d) splitting off %d rules from superblock @ #%d",
 769                                     depth, skiplist->ps_count,
 770                                     TAILQ_FIRST(&skiplist->ps_rules)->
 771                                     por_rule.nr);
 772                         } else {
 773                                 newblock = block;
 774                         }
 775
 776                         while ((por = TAILQ_FIRST(&skiplist->ps_rules))) {
 777                                 TAILQ_REMOVE(&head, por, por_entry);
 778                                 TAILQ_REMOVE(&skiplist->ps_rules, por,
 779                                     por_skip_entry[largest_list]);
 780                                 TAILQ_INSERT_TAIL(&newblock->sb_rules, por,
 781                                     por_entry);
 782
 783                                 /* Remove this rule from all other skiplists */
 784                                 remove_from_skipsteps(&block->sb_skipsteps[
 785                                     largest_list], block, por, skiplist);
 786                         }
 787                         free(skiplist);
 788                         if (newblock != block)
 789                                 if (reorder_rules(pf, newblock, depth + 1))
 790                                         return (1);
 791                 }
 792         }
 793
 794 done:
 795         for (i = 0; i < PF_SKIP_COUNT; i++) {
 796                 while ((skiplist = TAILQ_FIRST(&block->sb_skipsteps[i]))) {
 797                         TAILQ_REMOVE(&block->sb_skipsteps[i], skiplist,
 798                             ps_entry);
 799                         free(skiplist);
 800                 }
 801         }
 802
 803         return (0);
 804 }
 805
 806
 807 /*
 808  * Optimization pass #4: re-order 'quick' rules based on feedback from the
 809  * currently running ruleset
 810  */
 811 int
 812 block_feedback(struct pfctl *pf, struct superblock *block)
 813 {
 814         TAILQ_HEAD( , pf_opt_rule) queue;
 815         struct pf_opt_rule *por1, *por2;
 816         u_int64_t total_count = 0;
 817         struct pf_rule a, b;
 818
 819
 820         /*
 821          * Walk through all of the profiled superblock's rules and copy
 822          * the counters onto our rules.
 823          */
 824         TAILQ_FOREACH(por1, &block->sb_profiled_block->sb_rules, por_entry) {
 825                 comparable_rule(&a, &por1->por_rule, DC);
 826                 total_count += por1->por_rule.packets[0] +
 827                     por1->por_rule.packets[1];
 828                 TAILQ_FOREACH(por2, &block->sb_rules, por_entry) {
 829                         if (por2->por_profile_count)
 830                                 continue;
 831                         comparable_rule(&b, &por2->por_rule, DC);
 832                         if (memcmp(&a, &b, sizeof(a)) == 0) {
 833                                 por2->por_profile_count =
 834                                     por1->por_rule.packets[0] +
 835                                     por1->por_rule.packets[1];
 836                                 break;
 837                         }
 838                 }
 839         }
 840         superblock_free(pf, block->sb_profiled_block);
 841         block->sb_profiled_block = NULL;
 842
 843         /*
 844          * Now we pull all of the rules off the superblock and re-insert them
 845          * in sorted order.
 846          */
 847
 848         TAILQ_INIT(&queue);
 849         while ((por1 = TAILQ_FIRST(&block->sb_rules)) != NULL) {
 850                 TAILQ_REMOVE(&block->sb_rules, por1, por_entry);
 851                 TAILQ_INSERT_TAIL(&queue, por1, por_entry);
 852         }
 853
 854         while ((por1 = TAILQ_FIRST(&queue)) != NULL) {
 855                 TAILQ_REMOVE(&queue, por1, por_entry);
 856 /* XXX I should sort all of the unused rules based on skip steps */
 857                 TAILQ_FOREACH(por2, &block->sb_rules, por_entry) {
 858                         if (por1->por_profile_count > por2->por_profile_count) {
 859                                 TAILQ_INSERT_BEFORE(por2, por1, por_entry);
 860                                 break;
 861                         }
 862                 }
 863 #ifdef __FreeBSD__
 864                 if (por2 == NULL)
 865 #else
 866                 if (por2 == TAILQ_END(&block->sb_rules))
 867 #endif
 868                         TAILQ_INSERT_TAIL(&block->sb_rules, por1, por_entry);
 869         }
 870
 871         return (0);
 872 }
 873
 874
 875 /*
 876  * Load the current ruleset from the kernel and try to associate them with
 877  * the ruleset we're optimizing.
 878  */
 879 int
 880 load_feedback_profile(struct pfctl *pf, struct superblocks *superblocks)
 881 {
 882         struct superblock *block, *blockcur;
 883         struct superblocks prof_superblocks;
 884         struct pf_opt_rule *por;
 885         struct pf_opt_queue queue;
 886         struct pfioc_rule pr;
 887         struct pf_rule a, b;
 888         int nr, mnr;
 889
 890         TAILQ_INIT(&queue);
 891         TAILQ_INIT(&prof_superblocks);
 892
 893         memset(&pr, 0, sizeof(pr));
 894         pr.rule.action = PF_PASS;
 895         if (ioctl(pf->dev, DIOCGETRULES, &pr)) {
 896                 warn("DIOCGETRULES");
 897                 return (1);
 898         }
 899         mnr = pr.nr;
 900
 901         DEBUG("Loading %d active rules for a feedback profile", mnr);
 902         for (nr = 0; nr < mnr; ++nr) {
 903                 struct pf_ruleset *rs;
 904                 if ((por = calloc(1, sizeof(*por))) == NULL) {
 905                         warn("calloc");
 906                         return (1);
 907                 }
 908                 pr.nr = nr;
 909                 if (ioctl(pf->dev, DIOCGETRULE, &pr)) {
 910                         warn("DIOCGETRULES");
 911                         return (1);
 912                 }
 913                 memcpy(&por->por_rule, &pr.rule, sizeof(por->por_rule));
 914                 rs = pf_find_or_create_ruleset(pr.anchor_call);
 915                 por->por_rule.anchor = rs->anchor;
 916                 if (TAILQ_EMPTY(&por->por_rule.rpool.list))
 917                         memset(&por->por_rule.rpool, 0,
 918                             sizeof(por->por_rule.rpool));
 919                 TAILQ_INSERT_TAIL(&queue, por, por_entry);
 920
 921                 /* XXX pfctl_get_pool(pf->dev, &pr.rule.rpool, nr, pr.ticket,
 922                  *         PF_PASS, pf->anchor) ???
 923                  * ... pfctl_clear_pool(&pr.rule.rpool)
 924                  */
 925         }
 926
 927         if (construct_superblocks(pf, &queue, &prof_superblocks))
 928                 return (1);
 929
 930
 931         /*
 932          * Now we try to associate the active ruleset's superblocks with
 933          * the superblocks we're compiling.
 934          */
 935         block = TAILQ_FIRST(superblocks);
 936         blockcur = TAILQ_FIRST(&prof_superblocks);
 937         while (block && blockcur) {
 938                 comparable_rule(&a, &TAILQ_FIRST(&block->sb_rules)->por_rule,
 939                     BREAK);
 940                 comparable_rule(&b, &TAILQ_FIRST(&blockcur->sb_rules)->por_rule,
 941                     BREAK);
 942                 if (memcmp(&a, &b, sizeof(a)) == 0) {
 943                         /* The two superblocks lined up */
 944                         block->sb_profiled_block = blockcur;
 945                 } else {
 946                         DEBUG("superblocks don't line up between #%d and #%d",
 947                             TAILQ_FIRST(&block->sb_rules)->por_rule.nr,
 948                             TAILQ_FIRST(&blockcur->sb_rules)->por_rule.nr);
 949                         break;
 950                 }
 951                 block = TAILQ_NEXT(block, sb_entry);
 952                 blockcur = TAILQ_NEXT(blockcur, sb_entry);
 953         }
 954
 955
 956
 957         /* Free any superblocks we couldn't link */
 958         while (blockcur) {
 959                 block = TAILQ_NEXT(blockcur, sb_entry);
 960                 superblock_free(pf, blockcur);
 961                 blockcur = block;
 962         }
 963         return (0);
 964 }
 965
 966
 967 /*
 968  * Compare a rule to a skiplist to see if the rule is a member
 969  */
 970 int
 971 skip_compare(int skipnum, struct pf_skip_step *skiplist,
 972     struct pf_opt_rule *por)
 973 {
 974         struct pf_rule *a, *b;
 975         if (skipnum >= PF_SKIP_COUNT || skipnum < 0)
 976                 errx(1, "skip_compare() out of bounds");
 977         a = &por->por_rule;
 978         b = &TAILQ_FIRST(&skiplist->ps_rules)->por_rule;
 979
 980         return ((skip_comparitors[skipnum])(a, b));
 981 }
 982
 983
 984 /*
 985  * Add a rule to a skiplist
 986  */
 987 void
 988 skip_append(struct superblock *superblock, int skipnum,
 989     struct pf_skip_step *skiplist, struct pf_opt_rule *por)
 990 {
 991         struct pf_skip_step *prev;
 992
 993         skiplist->ps_count++;
 994         TAILQ_INSERT_TAIL(&skiplist->ps_rules, por, por_skip_entry[skipnum]);
 995
 996         /* Keep the list of skiplists sorted by whichever is larger */
 997         while ((prev = TAILQ_PREV(skiplist, skiplist, ps_entry)) &&
 998             prev->ps_count < skiplist->ps_count) {
 999                 TAILQ_REMOVE(&superblock->sb_skipsteps[skipnum],
1000                     skiplist, ps_entry);
1001                 TAILQ_INSERT_BEFORE(prev, skiplist, ps_entry);
1002         }
1003 }
1004
1005
1006 /*
1007  * Remove a rule from the other skiplist calculations.
1008  */
1009 void
1010 remove_from_skipsteps(struct skiplist *head, struct superblock *block,
1011     struct pf_opt_rule *por, struct pf_skip_step *active_list)
1012 {
1013         struct pf_skip_step *sk, *next;
1014         struct pf_opt_rule *p2;
1015         int i, found;
1016
1017         for (i = 0; i < PF_SKIP_COUNT; i++) {
1018                 sk = TAILQ_FIRST(&block->sb_skipsteps[i]);
1019                 if (sk == NULL || sk == active_list || sk->ps_count <= 1)
1020                         continue;
1021                 found = 0;
1022                 do {
1023                         TAILQ_FOREACH(p2, &sk->ps_rules, por_skip_entry[i])
1024                                 if (p2 == por) {
1025                                         TAILQ_REMOVE(&sk->ps_rules, p2,
1026                                             por_skip_entry[i]);
1027                                         found = 1;
1028                                         sk->ps_count--;
1029                                         break;
1030                                 }
1031                 } while (!found && (sk = TAILQ_NEXT(sk, ps_entry)));
1032                 if (found && sk) {
1033                         /* Does this change the sorting order? */
1034                         while ((next = TAILQ_NEXT(sk, ps_entry)) &&
1035                             next->ps_count > sk->ps_count) {
1036                                 TAILQ_REMOVE(head, sk, ps_entry);
1037                                 TAILQ_INSERT_AFTER(head, next, sk, ps_entry);
1038                         }
1039 #ifdef OPT_DEBUG
1040                         next = TAILQ_NEXT(sk, ps_entry);
1041                         assert(next == NULL || next->ps_count <= sk->ps_count);
1042 #endif /* OPT_DEBUG */
1043                 }
1044         }
1045 }
1046
1047
1048 /* Compare two rules AF field for skiplist construction */
1049 int
1050 skip_cmp_af(struct pf_rule *a, struct pf_rule *b)
1051 {
1052         if (a->af != b->af || a->af == 0)
1053                 return (1);
1054         return (0);
1055 }
1056
1057 /* Compare two rules DIRECTION field for skiplist construction */
1058 int
1059 skip_cmp_dir(struct pf_rule *a, struct pf_rule *b)
1060 {
1061         if (a->direction == 0 || a->direction != b->direction)
1062                 return (1);
1063         return (0);
1064 }
1065
1066 /* Compare two rules DST Address field for skiplist construction */
1067 int
1068 skip_cmp_dst_addr(struct pf_rule *a, struct pf_rule *b)
1069 {
1070         if (a->dst.neg != b->dst.neg ||
1071             a->dst.addr.type != b->dst.addr.type)
1072                 return (1);
1073         /* XXX if (a->proto != b->proto && a->proto != 0 && b->proto != 0
1074          *    && (a->proto == IPPROTO_TCP || a->proto == IPPROTO_UDP ||
1075          *    a->proto == IPPROTO_ICMP
1076          *      return (1);
1077          */
1078         switch (a->dst.addr.type) {
1079         case PF_ADDR_ADDRMASK:
1080                 if (memcmp(&a->dst.addr.v.a.addr, &b->dst.addr.v.a.addr,
1081                     sizeof(a->dst.addr.v.a.addr)) ||
1082                     memcmp(&a->dst.addr.v.a.mask, &b->dst.addr.v.a.mask,
1083                     sizeof(a->dst.addr.v.a.mask)) ||
1084                     (a->dst.addr.v.a.addr.addr32[0] == 0 &&
1085                     a->dst.addr.v.a.addr.addr32[1] == 0 &&
1086                     a->dst.addr.v.a.addr.addr32[2] == 0 &&
1087                     a->dst.addr.v.a.addr.addr32[3] == 0))
1088                         return (1);
1089                 return (0);
1090         case PF_ADDR_DYNIFTL:
1091                 if (strcmp(a->dst.addr.v.ifname, b->dst.addr.v.ifname) != 0 ||
1092                     a->dst.addr.iflags != a->dst.addr.iflags ||
1093                     memcmp(&a->dst.addr.v.a.mask, &b->dst.addr.v.a.mask,
1094                     sizeof(a->dst.addr.v.a.mask)))
1095                         return (1);
1096                 return (0);
1097         case PF_ADDR_NOROUTE:
1098         case PF_ADDR_URPFFAILED:
1099                 return (0);
1100         case PF_ADDR_TABLE:
1101                 return (strcmp(a->dst.addr.v.tblname, b->dst.addr.v.tblname));
1102         }
1103         return (1);
1104 }
1105
1106 /* Compare two rules DST port field for skiplist construction */
1107 int
1108 skip_cmp_dst_port(struct pf_rule *a, struct pf_rule *b)
1109 {
1110         /* XXX if (a->proto != b->proto && a->proto != 0 && b->proto != 0
1111          *    && (a->proto == IPPROTO_TCP || a->proto == IPPROTO_UDP ||
1112          *    a->proto == IPPROTO_ICMP
1113          *      return (1);
1114          */
1115         if (a->dst.port_op == PF_OP_NONE || a->dst.port_op != b->dst.port_op ||
1116             a->dst.port[0] != b->dst.port[0] ||
1117             a->dst.port[1] != b->dst.port[1])
1118                 return (1);
1119         return (0);
1120 }
1121
1122 /* Compare two rules IFP field for skiplist construction */
1123 int
1124 skip_cmp_ifp(struct pf_rule *a, struct pf_rule *b)
1125 {
1126         if (strcmp(a->ifname, b->ifname) || a->ifname[0] == '\0')
1127                 return (1);
1128         return (a->ifnot != b->ifnot);
1129 }
1130
1131 /* Compare two rules PROTO field for skiplist construction */
1132 int
1133 skip_cmp_proto(struct pf_rule *a, struct pf_rule *b)
1134 {
1135         return (a->proto != b->proto || a->proto == 0);
1136 }
1137
1138 /* Compare two rules SRC addr field for skiplist construction */
1139 int
1140 skip_cmp_src_addr(struct pf_rule *a, struct pf_rule *b)
1141 {
1142         if (a->src.neg != b->src.neg ||
1143             a->src.addr.type != b->src.addr.type)
1144                 return (1);
1145         /* XXX if (a->proto != b->proto && a->proto != 0 && b->proto != 0
1146          *    && (a->proto == IPPROTO_TCP || a->proto == IPPROTO_UDP ||
1147          *    a->proto == IPPROTO_ICMP
1148          *      return (1);
1149          */
1150         switch (a->src.addr.type) {
1151         case PF_ADDR_ADDRMASK:
1152                 if (memcmp(&a->src.addr.v.a.addr, &b->src.addr.v.a.addr,
1153                     sizeof(a->src.addr.v.a.addr)) ||
1154                     memcmp(&a->src.addr.v.a.mask, &b->src.addr.v.a.mask,
1155                     sizeof(a->src.addr.v.a.mask)) ||
1156                     (a->src.addr.v.a.addr.addr32[0] == 0 &&
1157                     a->src.addr.v.a.addr.addr32[1] == 0 &&
1158                     a->src.addr.v.a.addr.addr32[2] == 0 &&
1159                     a->src.addr.v.a.addr.addr32[3] == 0))
1160                         return (1);
1161                 return (0);
1162         case PF_ADDR_DYNIFTL:
1163                 if (strcmp(a->src.addr.v.ifname, b->src.addr.v.ifname) != 0 ||
1164                     a->src.addr.iflags != a->src.addr.iflags ||
1165                     memcmp(&a->src.addr.v.a.mask, &b->src.addr.v.a.mask,
1166                     sizeof(a->src.addr.v.a.mask)))
1167                         return (1);
1168                 return (0);
1169         case PF_ADDR_NOROUTE:
1170         case PF_ADDR_URPFFAILED:
1171                 return (0);
1172         case PF_ADDR_TABLE:
1173                 return (strcmp(a->src.addr.v.tblname, b->src.addr.v.tblname));
1174         }
1175         return (1);
1176 }
1177
1178 /* Compare two rules SRC port field for skiplist construction */
1179 int
1180 skip_cmp_src_port(struct pf_rule *a, struct pf_rule *b)
1181 {
1182         if (a->src.port_op == PF_OP_NONE || a->src.port_op != b->src.port_op ||
1183             a->src.port[0] != b->src.port[0] ||
1184             a->src.port[1] != b->src.port[1])
1185                 return (1);
1186         /* XXX if (a->proto != b->proto && a->proto != 0 && b->proto != 0
1187          *    && (a->proto == IPPROTO_TCP || a->proto == IPPROTO_UDP ||
1188          *    a->proto == IPPROTO_ICMP
1189          *      return (1);
1190          */
1191         return (0);
1192 }
1193
1194
1195 void
1196 skip_init(void)
1197 {
1198         struct {
1199                 char *name;
1200                 int skipnum;
1201                 int (*func)(struct pf_rule *, struct pf_rule *);
1202         } comps[] = PF_SKIP_COMPARITORS;
1203         int skipnum, i;
1204
1205         for (skipnum = 0; skipnum < PF_SKIP_COUNT; skipnum++) {
1206                 for (i = 0; i < sizeof(comps)/sizeof(*comps); i++)
1207                         if (comps[i].skipnum == skipnum) {
1208                                 skip_comparitors[skipnum] = comps[i].func;
1209                                 skip_comparitors_names[skipnum] = comps[i].name;
1210                         }
1211         }
1212         for (skipnum = 0; skipnum < PF_SKIP_COUNT; skipnum++)
1213                 if (skip_comparitors[skipnum] == NULL)
1214                         errx(1, "Need to add skip step comparitor to pfctl?!");
1215 }
1216
1217 /*
1218  * Add a host/netmask to a table
1219  */
1220 int
1221 add_opt_table(struct pfctl *pf, struct pf_opt_tbl **tbl, sa_family_t af,
1222     struct pf_rule_addr *addr)
1223 {
1224 #ifdef OPT_DEBUG
1225         char buf[128];
1226 #endif /* OPT_DEBUG */
1227         static int tablenum = 0;
1228         struct node_host node_host;
1229
1230         if (*tbl == NULL) {
1231                 if ((*tbl = calloc(1, sizeof(**tbl))) == NULL ||
1232                     ((*tbl)->pt_buf = calloc(1, sizeof(*(*tbl)->pt_buf))) ==
1233                     NULL)
1234                         err(1, "calloc");
1235                 (*tbl)->pt_buf->pfrb_type = PFRB_ADDRS;
1236                 SIMPLEQ_INIT(&(*tbl)->pt_nodes);
1237
1238                 /* This is just a temporary table name */
1239                 snprintf((*tbl)->pt_name, sizeof((*tbl)->pt_name), "%s%d",
1240                     PF_OPT_TABLE_PREFIX, tablenum++);
1241                 DEBUG("creating table <%s>", (*tbl)->pt_name);
1242         }
1243
1244         memset(&node_host, 0, sizeof(node_host));
1245         node_host.af = af;
1246         node_host.addr = addr->addr;
1247
1248 #ifdef OPT_DEBUG
1249         DEBUG("<%s> adding %s/%d", (*tbl)->pt_name, inet_ntop(af,
1250             &node_host.addr.v.a.addr, buf, sizeof(buf)),
1251             unmask(&node_host.addr.v.a.mask, af));
1252 #endif /* OPT_DEBUG */
1253
1254         if (append_addr_host((*tbl)->pt_buf, &node_host, 0, 0)) {
1255                 warn("failed to add host");
1256                 return (1);
1257         }
1258         if (pf->opts & PF_OPT_VERBOSE) {
1259                 struct node_tinit *ti;
1260
1261                 if ((ti = calloc(1, sizeof(*ti))) == NULL)
1262                         err(1, "malloc");
1263                 if ((ti->host = malloc(sizeof(*ti->host))) == NULL)
1264                         err(1, "malloc");
1265                 memcpy(ti->host, &node_host, sizeof(*ti->host));
1266                 SIMPLEQ_INSERT_TAIL(&(*tbl)->pt_nodes, ti, entries);
1267         }
1268
1269         (*tbl)->pt_rulecount++;
1270         if ((*tbl)->pt_rulecount == TABLE_THRESHOLD)
1271                 DEBUG("table <%s> now faster than skip steps", (*tbl)->pt_name);
1272
1273         return (0);
1274 }
1275
1276
1277 /*
1278  * Do the dirty work of choosing an unused table name and creating it.
1279  * (be careful with the table name, it might already be used in another anchor)
1280  */
1281 int
1282 pf_opt_create_table(struct pfctl *pf, struct pf_opt_tbl *tbl)
1283 {
1284         static int tablenum;
1285         struct pfr_table *t;
1286
1287         if (table_buffer.pfrb_type == 0) {
1288                 /* Initialize the list of tables */
1289                 table_buffer.pfrb_type = PFRB_TABLES;
1290                 for (;;) {
1291                         pfr_buf_grow(&table_buffer, table_buffer.pfrb_size);
1292                         table_buffer.pfrb_size = table_buffer.pfrb_msize;
1293                         if (pfr_get_tables(NULL, table_buffer.pfrb_caddr,
1294                             &table_buffer.pfrb_size, PFR_FLAG_ALLRSETS))
1295                                 err(1, "pfr_get_tables");
1296                         if (table_buffer.pfrb_size <= table_buffer.pfrb_msize)
1297                                 break;
1298                 }
1299                 table_identifier = arc4random();
1300         }
1301
1302         /* XXX would be *really* nice to avoid duplicating identical tables */
1303
1304         /* Now we have to pick a table name that isn't used */
1305 again:
1306         DEBUG("translating temporary table <%s> to <%s%x_%d>", tbl->pt_name,
1307             PF_OPT_TABLE_PREFIX, table_identifier, tablenum);
1308         snprintf(tbl->pt_name, sizeof(tbl->pt_name), "%s%x_%d",
1309             PF_OPT_TABLE_PREFIX, table_identifier, tablenum);
1310         PFRB_FOREACH(t, &table_buffer) {
1311                 if (strcasecmp(t->pfrt_name, tbl->pt_name) == 0) {
1312                         /* Collision.  Try again */
1313                         DEBUG("wow, table <%s> in use.  trying again",
1314                             tbl->pt_name);
1315                         table_identifier = arc4random();
1316                         goto again;
1317                 }
1318         }
1319         tablenum++;
1320
1321
1322         if (pfctl_define_table(tbl->pt_name, PFR_TFLAG_CONST, 1,
1323             pf->anchor->name, tbl->pt_buf, pf->anchor->ruleset.tticket)) {
1324                 warn("failed to create table %s", tbl->pt_name);
1325                 return (1);
1326         }
1327         return (0);
1328 }
1329
1330 /*
1331  * Partition the flat ruleset into a list of distinct superblocks
1332  */
1333 int
1334 construct_superblocks(struct pfctl *pf, struct pf_opt_queue *opt_queue,
1335     struct superblocks *superblocks)
1336 {
1337         struct superblock *block = NULL;
1338         struct pf_opt_rule *por;
1339         int i;
1340
1341         while (!TAILQ_EMPTY(opt_queue)) {
1342                 por = TAILQ_FIRST(opt_queue);
1343                 TAILQ_REMOVE(opt_queue, por, por_entry);
1344                 if (block == NULL || !superblock_inclusive(block, por)) {
1345                         if ((block = calloc(1, sizeof(*block))) == NULL) {
1346                                 warn("calloc");
1347                                 return (1);
1348                         }
1349                         TAILQ_INIT(&block->sb_rules);
1350                         for (i = 0; i < PF_SKIP_COUNT; i++)
1351                                 TAILQ_INIT(&block->sb_skipsteps[i]);
1352                         TAILQ_INSERT_TAIL(superblocks, block, sb_entry);
1353                 }
1354                 TAILQ_INSERT_TAIL(&block->sb_rules, por, por_entry);
1355         }
1356
1357         return (0);
1358 }
1359
1360
1361 /*
1362  * Compare two rule addresses
1363  */
1364 int
1365 addrs_equal(struct pf_rule_addr *a, struct pf_rule_addr *b)
1366 {
1367         if (a->neg != b->neg)
1368                 return (0);
1369         return (memcmp(&a->addr, &b->addr, sizeof(a->addr)) == 0);
1370 }
1371
1372
1373 /*
1374  * The addresses are not equal, but can we combine them into one table?
1375  */
1376 int
1377 addrs_combineable(struct pf_rule_addr *a, struct pf_rule_addr *b)
1378 {
1379         if (a->addr.type != PF_ADDR_ADDRMASK ||
1380             b->addr.type != PF_ADDR_ADDRMASK)
1381                 return (0);
1382         if (a->neg != b->neg || a->port_op != b->port_op ||
1383             a->port[0] != b->port[0] || a->port[1] != b->port[1])
1384                 return (0);
1385         return (1);
1386 }
1387
1388
1389 /*
1390  * Are we allowed to combine these two rules
1391  */
1392 int
1393 rules_combineable(struct pf_rule *p1, struct pf_rule *p2)
1394 {
1395         struct pf_rule a, b;
1396
1397         comparable_rule(&a, p1, COMBINED);
1398         comparable_rule(&b, p2, COMBINED);
1399         return (memcmp(&a, &b, sizeof(a)) == 0);
1400 }
1401
1402
1403 /*
1404  * Can a rule be included inside a superblock
1405  */
1406 int
1407 superblock_inclusive(struct superblock *block, struct pf_opt_rule *por)
1408 {
1409         struct pf_rule a, b;
1410         int i, j;
1411
1412         /* First check for hard breaks */
1413         for (i = 0; i < sizeof(pf_rule_desc)/sizeof(*pf_rule_desc); i++) {
1414                 if (pf_rule_desc[i].prf_type == BARRIER) {
1415                         for (j = 0; j < pf_rule_desc[i].prf_size; j++)
1416                                 if (((char *)&por->por_rule)[j +
1417                                     pf_rule_desc[i].prf_offset] != 0)
1418                                         return (0);
1419                 }
1420         }
1421
1422         /* per-rule src-track is also a hard break */
1423         if (por->por_rule.rule_flag & PFRULE_RULESRCTRACK)
1424                 return (0);
1425
1426         /*
1427          * Have to handle interface groups seperately.  Consider the following
1428          * rules:
1429          *      block on EXTIFS to any port 22
1430          *      pass  on em0 to any port 22
1431          * (where EXTIFS is an arbitrary interface group)
1432          * The optimizer may decide to re-order the pass rule in front of the
1433          * block rule.  But what if EXTIFS includes em0???  Such a reordering
1434          * would change the meaning of the ruleset.
1435          * We can't just lookup the EXTIFS group and check if em0 is a member
1436          * because the user is allowed to add interfaces to a group during
1437          * runtime.
1438          * Ergo interface groups become a defacto superblock break :-(
1439          */
1440         if (interface_group(por->por_rule.ifname) ||
1441             interface_group(TAILQ_FIRST(&block->sb_rules)->por_rule.ifname)) {
1442                 if (strcasecmp(por->por_rule.ifname,
1443                     TAILQ_FIRST(&block->sb_rules)->por_rule.ifname) != 0)
1444                         return (0);
1445         }
1446
1447         comparable_rule(&a, &TAILQ_FIRST(&block->sb_rules)->por_rule, NOMERGE);
1448         comparable_rule(&b, &por->por_rule, NOMERGE);
1449         if (memcmp(&a, &b, sizeof(a)) == 0)
1450                 return (1);
1451
1452 #ifdef OPT_DEBUG
1453         for (i = 0; i < sizeof(por->por_rule); i++) {
1454                 int closest = -1;
1455                 if (((u_int8_t *)&a)[i] != ((u_int8_t *)&b)[i]) {
1456                         for (j = 0; j < sizeof(pf_rule_desc) /
1457                             sizeof(*pf_rule_desc); j++) {
1458                                 if (i >= pf_rule_desc[j].prf_offset &&
1459                                     i < pf_rule_desc[j].prf_offset +
1460                                     pf_rule_desc[j].prf_size) {
1461                                         DEBUG("superblock break @ %d due to %s",
1462                                             por->por_rule.nr,
1463                                             pf_rule_desc[j].prf_name);
1464                                         return (0);
1465                                 }
1466                                 if (i > pf_rule_desc[j].prf_offset) {
1467                                         if (closest == -1 ||
1468                                             i-pf_rule_desc[j].prf_offset <
1469                                             i-pf_rule_desc[closest].prf_offset)
1470                                                 closest = j;
1471                                 }
1472                         }
1473
1474                         if (closest >= 0)
1475                                 DEBUG("superblock break @ %d on %s+%xh",
1476                                     por->por_rule.nr,
1477                                     pf_rule_desc[closest].prf_name,
1478                                     i - pf_rule_desc[closest].prf_offset -
1479                                     pf_rule_desc[closest].prf_size);
1480                         else
1481                                 DEBUG("superblock break @ %d on field @ %d",
1482                                     por->por_rule.nr, i);
1483                         return (0);
1484                 }
1485         }
1486 #endif /* OPT_DEBUG */
1487
1488         return (0);
1489 }
1490
1491
1492 /*
1493  * Figure out if an interface name is an actual interface or actually a
1494  * group of interfaces.
1495  */
1496 int
1497 interface_group(const char *ifname)
1498 {
1499         if (ifname == NULL || !ifname[0])
1500                 return (0);
1501
1502         /* Real interfaces must end in a number, interface groups do not */
1503         if (isdigit(ifname[strlen(ifname) - 1]))
1504                 return (0);
1505         else
1506                 return (1);
1507 }
1508
1509
1510 /*
1511  * Make a rule that can directly compared by memcmp()
1512  */
1513 void
1514 comparable_rule(struct pf_rule *dst, const struct pf_rule *src, int type)
1515 {
1516         int i;
1517         /*
1518          * To simplify the comparison, we just zero out the fields that are
1519          * allowed to be different and then do a simple memcmp()
1520          */
1521         memcpy(dst, src, sizeof(*dst));
1522         for (i = 0; i < sizeof(pf_rule_desc)/sizeof(*pf_rule_desc); i++)
1523                 if (pf_rule_desc[i].prf_type >= type) {
1524 #ifdef OPT_DEBUG
1525                         assert(pf_rule_desc[i].prf_type != NEVER ||
1526                             *(((char *)dst) + pf_rule_desc[i].prf_offset) == 0);
1527 #endif /* OPT_DEBUG */
1528                         memset(((char *)dst) + pf_rule_desc[i].prf_offset, 0,
1529                             pf_rule_desc[i].prf_size);
1530                 }
1531 }
1532
1533
1534 /*
1535  * Remove superset information from two rules so we can directly compare them
1536  * with memcmp()
1537  */
1538 void
1539 exclude_supersets(struct pf_rule *super, struct pf_rule *sub)
1540 {
1541         if (super->ifname[0] == '\0')
1542                 memset(sub->ifname, 0, sizeof(sub->ifname));
1543         if (super->direction == PF_INOUT)
1544                 sub->direction = PF_INOUT;
1545         if ((super->proto == 0 || super->proto == sub->proto) &&
1546             super->flags == 0 && super->flagset == 0 && (sub->flags ||
1547             sub->flagset)) {
1548                 sub->flags = super->flags;
1549                 sub->flagset = super->flagset;
1550         }
1551         if (super->proto == 0)
1552                 sub->proto = 0;
1553
1554         if (super->src.port_op == 0) {
1555                 sub->src.port_op = 0;
1556                 sub->src.port[0] = 0;
1557                 sub->src.port[1] = 0;
1558         }
1559         if (super->dst.port_op == 0) {
1560                 sub->dst.port_op = 0;
1561                 sub->dst.port[0] = 0;
1562                 sub->dst.port[1] = 0;
1563         }
1564
1565         if (super->src.addr.type == PF_ADDR_ADDRMASK && !super->src.neg &&
1566             !sub->src.neg && super->src.addr.v.a.mask.addr32[0] == 0 &&
1567             super->src.addr.v.a.mask.addr32[1] == 0 &&
1568             super->src.addr.v.a.mask.addr32[2] == 0 &&
1569             super->src.addr.v.a.mask.addr32[3] == 0)
1570                 memset(&sub->src.addr, 0, sizeof(sub->src.addr));
1571         else if (super->src.addr.type == PF_ADDR_ADDRMASK &&
1572             sub->src.addr.type == PF_ADDR_ADDRMASK &&
1573             super->src.neg == sub->src.neg &&
1574             super->af == sub->af &&
1575             unmask(&super->src.addr.v.a.mask, super->af) <
1576             unmask(&sub->src.addr.v.a.mask, sub->af) &&
1577             super->src.addr.v.a.addr.addr32[0] ==
1578             (sub->src.addr.v.a.addr.addr32[0] &
1579             super->src.addr.v.a.mask.addr32[0]) &&
1580             super->src.addr.v.a.addr.addr32[1] ==
1581             (sub->src.addr.v.a.addr.addr32[1] &
1582             super->src.addr.v.a.mask.addr32[1]) &&
1583             super->src.addr.v.a.addr.addr32[2] ==
1584             (sub->src.addr.v.a.addr.addr32[2] &
1585             super->src.addr.v.a.mask.addr32[2]) &&
1586             super->src.addr.v.a.addr.addr32[3] ==
1587             (sub->src.addr.v.a.addr.addr32[3] &
1588             super->src.addr.v.a.mask.addr32[3])) {
1589                 /* sub->src.addr is a subset of super->src.addr/mask */
1590                 memcpy(&sub->src.addr, &super->src.addr, sizeof(sub->src.addr));
1591         }
1592
1593         if (super->dst.addr.type == PF_ADDR_ADDRMASK && !super->dst.neg &&
1594             !sub->dst.neg && super->dst.addr.v.a.mask.addr32[0] == 0 &&
1595             super->dst.addr.v.a.mask.addr32[1] == 0 &&
1596             super->dst.addr.v.a.mask.addr32[2] == 0 &&
1597             super->dst.addr.v.a.mask.addr32[3] == 0)
1598                 memset(&sub->dst.addr, 0, sizeof(sub->dst.addr));
1599         else if (super->dst.addr.type == PF_ADDR_ADDRMASK &&
1600             sub->dst.addr.type == PF_ADDR_ADDRMASK &&
1601             super->dst.neg == sub->dst.neg &&
1602             super->af == sub->af &&
1603             unmask(&super->dst.addr.v.a.mask, super->af) <
1604             unmask(&sub->dst.addr.v.a.mask, sub->af) &&
1605             super->dst.addr.v.a.addr.addr32[0] ==
1606             (sub->dst.addr.v.a.addr.addr32[0] &
1607             super->dst.addr.v.a.mask.addr32[0]) &&
1608             super->dst.addr.v.a.addr.addr32[1] ==
1609             (sub->dst.addr.v.a.addr.addr32[1] &
1610             super->dst.addr.v.a.mask.addr32[1]) &&
1611             super->dst.addr.v.a.addr.addr32[2] ==
1612             (sub->dst.addr.v.a.addr.addr32[2] &
1613             super->dst.addr.v.a.mask.addr32[2]) &&
1614             super->dst.addr.v.a.addr.addr32[3] ==
1615             (sub->dst.addr.v.a.addr.addr32[3] &
1616             super->dst.addr.v.a.mask.addr32[3])) {
1617                 /* sub->dst.addr is a subset of super->dst.addr/mask */
1618                 memcpy(&sub->dst.addr, &super->dst.addr, sizeof(sub->dst.addr));
1619         }
1620
1621         if (super->af == 0)
1622                 sub->af = 0;
1623 }
1624
1625
1626 void
1627 superblock_free(struct pfctl *pf, struct superblock *block)
1628 {
1629         struct pf_opt_rule *por;
1630         while ((por = TAILQ_FIRST(&block->sb_rules))) {
1631                 TAILQ_REMOVE(&block->sb_rules, por, por_entry);
1632                 if (por->por_src_tbl) {
1633                         if (por->por_src_tbl->pt_buf) {
1634                                 pfr_buf_clear(por->por_src_tbl->pt_buf);
1635                                 free(por->por_src_tbl->pt_buf);
1636                         }
1637                         free(por->por_src_tbl);
1638                 }
1639                 if (por->por_dst_tbl) {
1640                         if (por->por_dst_tbl->pt_buf) {
1641                                 pfr_buf_clear(por->por_dst_tbl->pt_buf);
1642                                 free(por->por_dst_tbl->pt_buf);
1643                         }
1644                         free(por->por_dst_tbl);
1645                 }
1646                 free(por);
1647         }
1648         if (block->sb_profiled_block)
1649                 superblock_free(pf, block->sb_profiled_block);
1650         free(block);
1651 }
1652