1/* $OpenBSD: mandocdb.c,v 1.221 2024/05/14 21:12:44 schwarze Exp $ */ 2/* 3 * Copyright (c) 2011-2021, 2024 Ingo Schwarze <schwarze@openbsd.org> 4 * Copyright (c) 2011, 2012 Kristaps Dzonsons <kristaps@bsd.lv> 5 * Copyright (c) 2016 Ed Maste <emaste@freebsd.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHORS DISCLAIM ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 * 19 * Implementation of the makewhatis(8) program. 20 */ 21#include <sys/types.h> 22#include <sys/mman.h> 23#include <sys/stat.h> 24 25#include <assert.h> 26#include <ctype.h> 27#include <err.h> 28#include <errno.h> 29#include <fcntl.h> 30#include <fts.h> 31#include <limits.h> 32#include <stdarg.h> 33#include <stddef.h> 34#include <stdio.h> 35#include <stdint.h> 36#include <stdlib.h> 37#include <string.h> 38#include <unistd.h> 39 40#include "mandoc_aux.h" 41#include "mandoc_ohash.h" 42#include "mandoc.h" 43#include "roff.h" 44#include "mdoc.h" 45#include "man.h" 46#include "mandoc_parse.h" 47#include "manconf.h" 48#include "mansearch.h" 49#include "dba_array.h" 50#include "dba.h" 51 52extern const char *const mansearch_keynames[]; 53 54enum op { 55 OP_DEFAULT = 0, /* new dbs from dir list or default config */ 56 OP_CONFFILE, /* new databases from custom config file */ 57 OP_UPDATE, /* delete/add entries in existing database */ 58 OP_DELETE, /* delete entries from existing database */ 59 OP_TEST /* change no databases, report potential problems */ 60}; 61 62struct str { 63 const struct mpage *mpage; /* if set, the owning parse */ 64 uint64_t mask; /* bitmask in sequence */ 65 char key[]; /* rendered text */ 66}; 67 68struct inodev { 69 ino_t st_ino; 70 dev_t st_dev; 71}; 72 73struct mpage { 74 struct inodev inodev; /* used for hashing routine */ 75 struct dba_array *dba; 76 char *sec; /* section from file content */ 77 char *arch; /* architecture from file content */ 78 char *title; /* title from file content */ 79 char *desc; /* description from file content */ 80 struct mpage *next; /* singly linked list */ 81 struct mlink *mlinks; /* singly linked list */ 82 int name_head_done; 83 enum form form; /* format from file content */ 84}; 85 86struct mlink { 87 char file[PATH_MAX]; /* filename rel. to manpath */ 88 char *dsec; /* section from directory */ 89 char *arch; /* architecture from directory */ 90 char *name; /* name from file name (not empty) */ 91 char *fsec; /* section from file name suffix */ 92 struct mlink *next; /* singly linked list */ 93 struct mpage *mpage; /* parent */ 94 int gzip; /* filename has a .gz suffix */ 95 enum form dform; /* format from directory */ 96 enum form fform; /* format from file name suffix */ 97}; 98 99typedef int (*mdoc_fp)(struct mpage *, const struct roff_meta *, 100 const struct roff_node *); 101 102struct mdoc_handler { 103 mdoc_fp fp; /* optional handler */ 104 uint64_t mask; /* set unless handler returns 0 */ 105 int taboo; /* node flags that must not be set */ 106}; 107 108 109int mandocdb(int, char *[]); 110 111static void dbadd(struct dba *, struct mpage *); 112static void dbadd_mlink(const struct mlink *); 113static void dbprune(struct dba *); 114static void dbwrite(struct dba *); 115static void filescan(const char *); 116static int fts_compare(const FTSENT **, const FTSENT **); 117static void mlink_add(struct mlink *, const struct stat *); 118static void mlink_check(struct mpage *, struct mlink *); 119static void mlink_free(struct mlink *); 120static void mlinks_undupe(struct mpage *); 121static void mpages_free(void); 122static void mpages_merge(struct dba *, struct mparse *); 123static void parse_cat(struct mpage *, int); 124static void parse_man(struct mpage *, const struct roff_meta *, 125 const struct roff_node *); 126static void parse_mdoc(struct mpage *, const struct roff_meta *, 127 const struct roff_node *); 128static int parse_mdoc_head(struct mpage *, const struct roff_meta *, 129 const struct roff_node *); 130static int parse_mdoc_Fa(struct mpage *, const struct roff_meta *, 131 const struct roff_node *); 132static int parse_mdoc_Fd(struct mpage *, const struct roff_meta *, 133 const struct roff_node *); 134static void parse_mdoc_fname(struct mpage *, const struct roff_node *); 135static int parse_mdoc_Fn(struct mpage *, const struct roff_meta *, 136 const struct roff_node *); 137static int parse_mdoc_Fo(struct mpage *, const struct roff_meta *, 138 const struct roff_node *); 139static int parse_mdoc_Nd(struct mpage *, const struct roff_meta *, 140 const struct roff_node *); 141static int parse_mdoc_Nm(struct mpage *, const struct roff_meta *, 142 const struct roff_node *); 143static int parse_mdoc_Sh(struct mpage *, const struct roff_meta *, 144 const struct roff_node *); 145static int parse_mdoc_Va(struct mpage *, const struct roff_meta *, 146 const struct roff_node *); 147static int parse_mdoc_Xr(struct mpage *, const struct roff_meta *, 148 const struct roff_node *); 149static void putkey(const struct mpage *, char *, uint64_t); 150static void putkeys(const struct mpage *, char *, size_t, uint64_t); 151static void putmdockey(const struct mpage *, 152 const struct roff_node *, uint64_t, int); 153static int render_string(char **, size_t *); 154static void say(const char *, const char *, ...) 155 __attribute__((__format__ (__printf__, 2, 3))); 156static int set_basedir(const char *, int); 157static int treescan(void); 158static size_t utf8(unsigned int, char[5]); 159 160static int nodb; /* no database changes */ 161static int mparse_options; /* abort the parse early */ 162static int use_all; /* use all found files */ 163static int debug; /* print what we're doing */ 164static int warnings; /* warn about crap */ 165static int write_utf8; /* write UTF-8 output; else ASCII */ 166static int exitcode; /* to be returned by main */ 167static enum op op; /* operational mode */ 168static char basedir[PATH_MAX]; /* current base directory */ 169static size_t basedir_len; /* strlen(basedir) */ 170static struct mpage *mpage_head; /* list of distinct manual pages */ 171static struct ohash mpages; /* table of distinct manual pages */ 172static struct ohash mlinks; /* table of directory entries */ 173static struct ohash names; /* table of all names */ 174static struct ohash strings; /* table of all strings */ 175static uint64_t name_mask; 176 177static const struct mdoc_handler mdoc_handlers[MDOC_MAX - MDOC_Dd] = { 178 { NULL, 0, NODE_NOPRT }, /* Dd */ 179 { NULL, 0, NODE_NOPRT }, /* Dt */ 180 { NULL, 0, NODE_NOPRT }, /* Os */ 181 { parse_mdoc_Sh, TYPE_Sh, 0 }, /* Sh */ 182 { parse_mdoc_head, TYPE_Ss, 0 }, /* Ss */ 183 { NULL, 0, 0 }, /* Pp */ 184 { NULL, 0, 0 }, /* D1 */ 185 { NULL, 0, 0 }, /* Dl */ 186 { NULL, 0, 0 }, /* Bd */ 187 { NULL, 0, 0 }, /* Ed */ 188 { NULL, 0, 0 }, /* Bl */ 189 { NULL, 0, 0 }, /* El */ 190 { NULL, 0, 0 }, /* It */ 191 { NULL, 0, 0 }, /* Ad */ 192 { NULL, TYPE_An, 0 }, /* An */ 193 { NULL, 0, 0 }, /* Ap */ 194 { NULL, TYPE_Ar, 0 }, /* Ar */ 195 { NULL, TYPE_Cd, 0 }, /* Cd */ 196 { NULL, TYPE_Cm, 0 }, /* Cm */ 197 { NULL, TYPE_Dv, 0 }, /* Dv */ 198 { NULL, TYPE_Er, 0 }, /* Er */ 199 { NULL, TYPE_Ev, 0 }, /* Ev */ 200 { NULL, 0, 0 }, /* Ex */ 201 { parse_mdoc_Fa, 0, 0 }, /* Fa */ 202 { parse_mdoc_Fd, 0, 0 }, /* Fd */ 203 { NULL, TYPE_Fl, 0 }, /* Fl */ 204 { parse_mdoc_Fn, 0, 0 }, /* Fn */ 205 { NULL, TYPE_Ft | TYPE_Vt, 0 }, /* Ft */ 206 { NULL, TYPE_Ic, 0 }, /* Ic */ 207 { NULL, TYPE_In, 0 }, /* In */ 208 { NULL, TYPE_Li, 0 }, /* Li */ 209 { parse_mdoc_Nd, 0, 0 }, /* Nd */ 210 { parse_mdoc_Nm, 0, 0 }, /* Nm */ 211 { NULL, 0, 0 }, /* Op */ 212 { NULL, 0, 0 }, /* Ot */ 213 { NULL, TYPE_Pa, NODE_NOSRC }, /* Pa */ 214 { NULL, 0, 0 }, /* Rv */ 215 { NULL, TYPE_St, 0 }, /* St */ 216 { parse_mdoc_Va, TYPE_Va, 0 }, /* Va */ 217 { parse_mdoc_Va, TYPE_Vt, 0 }, /* Vt */ 218 { parse_mdoc_Xr, 0, 0 }, /* Xr */ 219 { NULL, 0, 0 }, /* %A */ 220 { NULL, 0, 0 }, /* %B */ 221 { NULL, 0, 0 }, /* %D */ 222 { NULL, 0, 0 }, /* %I */ 223 { NULL, 0, 0 }, /* %J */ 224 { NULL, 0, 0 }, /* %N */ 225 { NULL, 0, 0 }, /* %O */ 226 { NULL, 0, 0 }, /* %P */ 227 { NULL, 0, 0 }, /* %R */ 228 { NULL, 0, 0 }, /* %T */ 229 { NULL, 0, 0 }, /* %V */ 230 { NULL, 0, 0 }, /* Ac */ 231 { NULL, 0, 0 }, /* Ao */ 232 { NULL, 0, 0 }, /* Aq */ 233 { NULL, TYPE_At, 0 }, /* At */ 234 { NULL, 0, 0 }, /* Bc */ 235 { NULL, 0, 0 }, /* Bf */ 236 { NULL, 0, 0 }, /* Bo */ 237 { NULL, 0, 0 }, /* Bq */ 238 { NULL, TYPE_Bsx, NODE_NOSRC }, /* Bsx */ 239 { NULL, TYPE_Bx, NODE_NOSRC }, /* Bx */ 240 { NULL, 0, 0 }, /* Db */ 241 { NULL, 0, 0 }, /* Dc */ 242 { NULL, 0, 0 }, /* Do */ 243 { NULL, 0, 0 }, /* Dq */ 244 { NULL, 0, 0 }, /* Ec */ 245 { NULL, 0, 0 }, /* Ef */ 246 { NULL, TYPE_Em, 0 }, /* Em */ 247 { NULL, 0, 0 }, /* Eo */ 248 { NULL, TYPE_Fx, NODE_NOSRC }, /* Fx */ 249 { NULL, TYPE_Ms, 0 }, /* Ms */ 250 { NULL, 0, 0 }, /* No */ 251 { NULL, 0, 0 }, /* Ns */ 252 { NULL, TYPE_Nx, NODE_NOSRC }, /* Nx */ 253 { NULL, TYPE_Ox, NODE_NOSRC }, /* Ox */ 254 { NULL, 0, 0 }, /* Pc */ 255 { NULL, 0, 0 }, /* Pf */ 256 { NULL, 0, 0 }, /* Po */ 257 { NULL, 0, 0 }, /* Pq */ 258 { NULL, 0, 0 }, /* Qc */ 259 { NULL, 0, 0 }, /* Ql */ 260 { NULL, 0, 0 }, /* Qo */ 261 { NULL, 0, 0 }, /* Qq */ 262 { NULL, 0, 0 }, /* Re */ 263 { NULL, 0, 0 }, /* Rs */ 264 { NULL, 0, 0 }, /* Sc */ 265 { NULL, 0, 0 }, /* So */ 266 { NULL, 0, 0 }, /* Sq */ 267 { NULL, 0, 0 }, /* Sm */ 268 { NULL, 0, 0 }, /* Sx */ 269 { NULL, TYPE_Sy, 0 }, /* Sy */ 270 { NULL, TYPE_Tn, 0 }, /* Tn */ 271 { NULL, 0, NODE_NOSRC }, /* Ux */ 272 { NULL, 0, 0 }, /* Xc */ 273 { NULL, 0, 0 }, /* Xo */ 274 { parse_mdoc_Fo, 0, 0 }, /* Fo */ 275 { NULL, 0, 0 }, /* Fc */ 276 { NULL, 0, 0 }, /* Oo */ 277 { NULL, 0, 0 }, /* Oc */ 278 { NULL, 0, 0 }, /* Bk */ 279 { NULL, 0, 0 }, /* Ek */ 280 { NULL, 0, 0 }, /* Bt */ 281 { NULL, 0, 0 }, /* Hf */ 282 { NULL, 0, 0 }, /* Fr */ 283 { NULL, 0, 0 }, /* Ud */ 284 { NULL, TYPE_Lb, NODE_NOSRC }, /* Lb */ 285 { NULL, 0, 0 }, /* Lp */ 286 { NULL, TYPE_Lk, 0 }, /* Lk */ 287 { NULL, TYPE_Mt, NODE_NOSRC }, /* Mt */ 288 { NULL, 0, 0 }, /* Brq */ 289 { NULL, 0, 0 }, /* Bro */ 290 { NULL, 0, 0 }, /* Brc */ 291 { NULL, 0, 0 }, /* %C */ 292 { NULL, 0, 0 }, /* Es */ 293 { NULL, 0, 0 }, /* En */ 294 { NULL, TYPE_Dx, NODE_NOSRC }, /* Dx */ 295 { NULL, 0, 0 }, /* %Q */ 296 { NULL, 0, 0 }, /* %U */ 297 { NULL, 0, 0 }, /* Ta */ 298}; 299 300 301int 302mandocdb(int argc, char *argv[]) 303{ 304 struct manconf conf; 305 struct mparse *mp; 306 struct dba *dba; 307 const char *path_arg, *progname; 308 size_t j, sz; 309 int ch, i; 310 311 if (pledge("stdio rpath wpath cpath", NULL) == -1) { 312 warn("pledge"); 313 return (int)MANDOCLEVEL_SYSERR; 314 } 315 316 memset(&conf, 0, sizeof(conf)); 317 318 /* 319 * We accept a few different invocations. 320 * The CHECKOP macro makes sure that invocation styles don't 321 * clobber each other. 322 */ 323#define CHECKOP(_op, _ch) do \ 324 if ((_op) != OP_DEFAULT) { \ 325 warnx("-%c: Conflicting option", (_ch)); \ 326 goto usage; \ 327 } while (/*CONSTCOND*/0) 328 329 mparse_options = MPARSE_UTF8 | MPARSE_LATIN1 | MPARSE_VALIDATE; 330 path_arg = NULL; 331 op = OP_DEFAULT; 332 333 while ((ch = getopt(argc, argv, "aC:Dd:npQT:tu:v")) != -1) 334 switch (ch) { 335 case 'a': 336 use_all = 1; 337 break; 338 case 'C': 339 CHECKOP(op, ch); 340 path_arg = optarg; 341 op = OP_CONFFILE; 342 break; 343 case 'D': 344 debug++; 345 break; 346 case 'd': 347 CHECKOP(op, ch); 348 path_arg = optarg; 349 op = OP_UPDATE; 350 break; 351 case 'n': 352 nodb = 1; 353 break; 354 case 'p': 355 warnings = 1; 356 break; 357 case 'Q': 358 mparse_options |= MPARSE_QUICK; 359 break; 360 case 'T': 361 if (strcmp(optarg, "utf8") != 0) { 362 warnx("-T%s: Unsupported output format", 363 optarg); 364 goto usage; 365 } 366 write_utf8 = 1; 367 break; 368 case 't': 369 CHECKOP(op, ch); 370 dup2(STDOUT_FILENO, STDERR_FILENO); 371 op = OP_TEST; 372 nodb = warnings = 1; 373 break; 374 case 'u': 375 CHECKOP(op, ch); 376 path_arg = optarg; 377 op = OP_DELETE; 378 break; 379 case 'v': 380 /* Compatibility with espie@'s makewhatis. */ 381 break; 382 default: 383 goto usage; 384 } 385 386 argc -= optind; 387 argv += optind; 388 389 if (nodb) { 390 if (pledge("stdio rpath", NULL) == -1) { 391 warn("pledge"); 392 return (int)MANDOCLEVEL_SYSERR; 393 } 394 } 395 396 if (op == OP_CONFFILE && argc > 0) { 397 warnx("-C: Too many arguments"); 398 goto usage; 399 } 400 401 exitcode = (int)MANDOCLEVEL_OK; 402 mchars_alloc(); 403 mp = mparse_alloc(mparse_options, MANDOC_OS_OTHER, NULL); 404 mandoc_ohash_init(&mpages, 6, offsetof(struct mpage, inodev)); 405 mandoc_ohash_init(&mlinks, 6, offsetof(struct mlink, file)); 406 407 if (op == OP_UPDATE || op == OP_DELETE || op == OP_TEST) { 408 409 /* 410 * Most of these deal with a specific directory. 411 * Jump into that directory first. 412 */ 413 if (op != OP_TEST && set_basedir(path_arg, 1) == 0) 414 goto out; 415 416 dba = nodb ? dba_new(128) : dba_read(MANDOC_DB); 417 if (dba != NULL) { 418 /* 419 * The existing database is usable. Process 420 * all files specified on the command-line. 421 */ 422 use_all = 1; 423 for (i = 0; i < argc; i++) 424 filescan(argv[i]); 425 if (nodb == 0) 426 dbprune(dba); 427 } else { 428 /* Database missing or corrupt. */ 429 if (op != OP_UPDATE || errno != ENOENT) 430 say(MANDOC_DB, "%s: Automatically recreating" 431 " from scratch", strerror(errno)); 432 exitcode = (int)MANDOCLEVEL_OK; 433 op = OP_DEFAULT; 434 if (treescan() == 0) 435 goto out; 436 dba = dba_new(128); 437 } 438 if (op != OP_DELETE) 439 mpages_merge(dba, mp); 440 if (nodb == 0) 441 dbwrite(dba); 442 dba_free(dba); 443 } else { 444 /* 445 * If we have arguments, use them as our manpaths. 446 * If we don't, use man.conf(5). 447 */ 448 if (argc > 0) { 449 conf.manpath.paths = mandoc_reallocarray(NULL, 450 argc, sizeof(char *)); 451 conf.manpath.sz = (size_t)argc; 452 for (i = 0; i < argc; i++) 453 conf.manpath.paths[i] = mandoc_strdup(argv[i]); 454 } else 455 manconf_parse(&conf, path_arg, NULL, NULL); 456 457 if (conf.manpath.sz == 0) { 458 exitcode = (int)MANDOCLEVEL_BADARG; 459 say("", "Empty manpath"); 460 } 461 462 /* 463 * First scan the tree rooted at a base directory, then 464 * build a new database and finally move it into place. 465 * Ignore zero-length directories and strip trailing 466 * slashes. 467 */ 468 for (j = 0; j < conf.manpath.sz; j++) { 469 sz = strlen(conf.manpath.paths[j]); 470 if (sz && conf.manpath.paths[j][sz - 1] == '/') 471 conf.manpath.paths[j][--sz] = '\0'; 472 if (sz == 0) 473 continue; 474 475 if (j) { 476 mandoc_ohash_init(&mpages, 6, 477 offsetof(struct mpage, inodev)); 478 mandoc_ohash_init(&mlinks, 6, 479 offsetof(struct mlink, file)); 480 } 481 482 if (set_basedir(conf.manpath.paths[j], argc > 0) == 0) 483 continue; 484 if (treescan() == 0) 485 continue; 486 dba = dba_new(128); 487 mpages_merge(dba, mp); 488 if (nodb == 0) 489 dbwrite(dba); 490 dba_free(dba); 491 492 if (j + 1 < conf.manpath.sz) { 493 mpages_free(); 494 ohash_delete(&mpages); 495 ohash_delete(&mlinks); 496 } 497 } 498 } 499out: 500 manconf_free(&conf); 501 mparse_free(mp); 502 mchars_free(); 503 mpages_free(); 504 ohash_delete(&mpages); 505 ohash_delete(&mlinks); 506 return exitcode; 507usage: 508 progname = getprogname(); 509 fprintf(stderr, "usage: %s [-aDnpQ] [-C file] [-Tutf8]\n" 510 " %s [-aDnpQ] [-Tutf8] dir ...\n" 511 " %s [-DnpQ] [-Tutf8] -d dir [file ...]\n" 512 " %s [-Dnp] -u dir [file ...]\n" 513 " %s [-Q] -t file ...\n", 514 progname, progname, progname, progname, progname); 515 516 return (int)MANDOCLEVEL_BADARG; 517} 518 519/* 520 * To get a singly linked list in alpha order while inserting entries 521 * at the beginning, process directory entries in reverse alpha order. 522 */ 523static int 524fts_compare(const FTSENT **a, const FTSENT **b) 525{ 526 return -strcmp((*a)->fts_name, (*b)->fts_name); 527} 528 529/* 530 * Scan a directory tree rooted at "basedir" for manpages. 531 * We use fts(), scanning directory parts along the way for clues to our 532 * section and architecture. 533 * 534 * If use_all has been specified, grok all files. 535 * If not, sanitise paths to the following: 536 * 537 * [./]man*[/<arch>]/<name>.<section> 538 * or 539 * [./]cat<section>[/<arch>]/<name>.0 540 * 541 * TODO: accommodate for multi-language directories. 542 */ 543static int 544treescan(void) 545{ 546 char buf[PATH_MAX]; 547 FTS *f; 548 FTSENT *ff; 549 struct mlink *mlink; 550 int gzip; 551 enum form dform; 552 char *dsec, *arch, *fsec, *cp; 553 const char *path; 554 const char *argv[2]; 555 556 argv[0] = "."; 557 argv[1] = NULL; 558 559 f = fts_open((char * const *)argv, FTS_PHYSICAL | FTS_NOCHDIR, 560 fts_compare); 561 if (f == NULL) { 562 exitcode = (int)MANDOCLEVEL_SYSERR; 563 say("", "&fts_open"); 564 return 0; 565 } 566 567 dsec = arch = NULL; 568 dform = FORM_NONE; 569 570 while ((ff = fts_read(f)) != NULL) { 571 path = ff->fts_path + 2; 572 switch (ff->fts_info) { 573 574 /* 575 * Symbolic links require various sanity checks, 576 * then get handled just like regular files. 577 */ 578 case FTS_SL: 579 if (realpath(path, buf) == NULL) { 580 if (warnings) 581 say(path, "&realpath"); 582 continue; 583 } 584 if (strncmp(buf, basedir, basedir_len) != 0) { 585 if (warnings) say("", 586 "%s: outside base directory", buf); 587 continue; 588 } 589 /* Use logical inode to avoid mpages dupe. */ 590 if (stat(path, ff->fts_statp) == -1) { 591 if (warnings) 592 say(path, "&stat"); 593 continue; 594 } 595 if ((ff->fts_statp->st_mode & S_IFMT) != S_IFREG) 596 continue; 597 /* FALLTHROUGH */ 598 599 /* 600 * If we're a regular file, add an mlink by using the 601 * stored directory data and handling the filename. 602 */ 603 case FTS_F: 604 if ( ! strcmp(path, MANDOC_DB)) 605 continue; 606 if ( ! use_all && ff->fts_level < 2) { 607 if (warnings) 608 say(path, "Extraneous file"); 609 continue; 610 } 611 gzip = 0; 612 fsec = NULL; 613 while (fsec == NULL) { 614 fsec = strrchr(ff->fts_name, '.'); 615 if (fsec == NULL || strcmp(fsec+1, "gz")) 616 break; 617 gzip = 1; 618 *fsec = '\0'; 619 fsec = NULL; 620 } 621 if (fsec == NULL) { 622 if ( ! use_all) { 623 if (warnings) 624 say(path, 625 "No filename suffix"); 626 continue; 627 } 628 } else if ( ! strcmp(++fsec, "html")) { 629 if (warnings) 630 say(path, "Skip html"); 631 continue; 632 } else if ( ! strcmp(fsec, "ps")) { 633 if (warnings) 634 say(path, "Skip ps"); 635 continue; 636 } else if ( ! strcmp(fsec, "pdf")) { 637 if (warnings) 638 say(path, "Skip pdf"); 639 continue; 640 } else if ( ! use_all && 641 ((dform == FORM_SRC && 642 strncmp(fsec, dsec, strlen(dsec))) || 643 (dform == FORM_CAT && strcmp(fsec, "0")))) { 644 if (warnings) 645 say(path, "Wrong filename suffix"); 646 continue; 647 } else 648 fsec[-1] = '\0'; 649 650 mlink = mandoc_calloc(1, sizeof(struct mlink)); 651 if (strlcpy(mlink->file, path, 652 sizeof(mlink->file)) >= 653 sizeof(mlink->file)) { 654 say(path, "Filename too long"); 655 free(mlink); 656 continue; 657 } 658 mlink->dform = dform; 659 mlink->dsec = dsec; 660 mlink->arch = arch; 661 mlink->name = ff->fts_name; 662 mlink->fsec = fsec; 663 mlink->gzip = gzip; 664 mlink_add(mlink, ff->fts_statp); 665 continue; 666 667 case FTS_D: 668 case FTS_DP: 669 break; 670 671 default: 672 if (warnings) 673 say(path, "Not a regular file"); 674 continue; 675 } 676 677 switch (ff->fts_level) { 678 case 0: 679 /* Ignore the root directory. */ 680 break; 681 case 1: 682 /* 683 * This might contain manX/ or catX/. 684 * Try to infer this from the name. 685 * If we're not in use_all, enforce it. 686 */ 687 cp = ff->fts_name; 688 if (ff->fts_info == FTS_DP) { 689 dform = FORM_NONE; 690 dsec = NULL; 691 break; 692 } 693 694 if ( ! strncmp(cp, "man", 3)) { 695 dform = FORM_SRC; 696 dsec = cp + 3; 697 } else if ( ! strncmp(cp, "cat", 3)) { 698 dform = FORM_CAT; 699 dsec = cp + 3; 700 } else { 701 dform = FORM_NONE; 702 dsec = NULL; 703 } 704 705 if (dsec != NULL || use_all) 706 break; 707 708 if (warnings) 709 say(path, "Unknown directory part"); 710 fts_set(f, ff, FTS_SKIP); 711 break; 712 case 2: 713 /* 714 * Possibly our architecture. 715 * If we're descending, keep tabs on it. 716 */ 717 if (ff->fts_info != FTS_DP && dsec != NULL) 718 arch = ff->fts_name; 719 else 720 arch = NULL; 721 break; 722 default: 723 if (ff->fts_info == FTS_DP || use_all) 724 break; 725 if (warnings) 726 say(path, "Extraneous directory part"); 727 fts_set(f, ff, FTS_SKIP); 728 break; 729 } 730 } 731 732 fts_close(f); 733 return 1; 734} 735 736/* 737 * Add a file to the mlinks table. 738 * Do not verify that it's a "valid" looking manpage (we'll do that 739 * later). 740 * 741 * Try to infer the manual section, architecture, and page name from the 742 * path, assuming it looks like 743 * 744 * [./]man*[/<arch>]/<name>.<section> 745 * or 746 * [./]cat<section>[/<arch>]/<name>.0 747 * 748 * See treescan() for the fts(3) version of this. 749 */ 750static void 751filescan(const char *infile) 752{ 753 struct stat st; 754 struct mlink *mlink; 755 char *linkfile, *p, *realdir, *start, *usefile; 756 size_t realdir_len; 757 758 assert(use_all); 759 760 if (strncmp(infile, "./", 2) == 0) 761 infile += 2; 762 763 /* 764 * We have to do lstat(2) before realpath(3) loses 765 * the information whether this is a symbolic link. 766 * We need to know that because for symbolic links, 767 * we want to use the original file name, while for 768 * regular files, we want to use the real path. 769 */ 770 if (lstat(infile, &st) == -1) { 771 exitcode = (int)MANDOCLEVEL_BADARG; 772 say(infile, "&lstat"); 773 return; 774 } else if (S_ISREG(st.st_mode) == 0 && S_ISLNK(st.st_mode) == 0) { 775 exitcode = (int)MANDOCLEVEL_BADARG; 776 say(infile, "Not a regular file"); 777 return; 778 } 779 780 /* 781 * We have to resolve the file name to the real path 782 * in any case for the base directory check. 783 */ 784 if ((usefile = realpath(infile, NULL)) == NULL) { 785 exitcode = (int)MANDOCLEVEL_BADARG; 786 say(infile, "&realpath"); 787 return; 788 } 789 790 if (op == OP_TEST) 791 start = usefile; 792 else if (strncmp(usefile, basedir, basedir_len) == 0) 793 start = usefile + basedir_len; 794 else { 795 exitcode = (int)MANDOCLEVEL_BADARG; 796 say("", "%s: outside base directory", infile); 797 free(usefile); 798 return; 799 } 800 801 /* 802 * Now we are sure the file is inside our tree. 803 * If it is a symbolic link, ignore the real path 804 * and use the original name. 805 */ 806 do { 807 if (S_ISLNK(st.st_mode) == 0) 808 break; 809 810 /* 811 * Some implementations of realpath(3) may succeed 812 * even if the target of the link does not exist, 813 * so check again for extra safety. 814 */ 815 if (stat(usefile, &st) == -1) { 816 exitcode = (int)MANDOCLEVEL_BADARG; 817 say(infile, "&stat"); 818 free(usefile); 819 return; 820 } 821 linkfile = mandoc_strdup(infile); 822 if (op == OP_TEST) { 823 free(usefile); 824 start = usefile = linkfile; 825 break; 826 } 827 if (strncmp(infile, basedir, basedir_len) == 0) { 828 free(usefile); 829 usefile = linkfile; 830 start = usefile + basedir_len; 831 break; 832 } 833 834 /* 835 * This symbolic link points into the basedir 836 * from the outside. Let's see whether any of 837 * the parent directories resolve to the basedir. 838 */ 839 p = strchr(linkfile, '\0'); 840 do { 841 while (*--p != '/') 842 continue; 843 *p = '\0'; 844 if ((realdir = realpath(linkfile, NULL)) == NULL) { 845 exitcode = (int)MANDOCLEVEL_BADARG; 846 say(infile, "&realpath"); 847 free(linkfile); 848 free(usefile); 849 return; 850 } 851 realdir_len = strlen(realdir) + 1; 852 free(realdir); 853 *p = '/'; 854 } while (realdir_len > basedir_len); 855 856 /* 857 * If one of the directories resolves to the basedir, 858 * use the rest of the original name. 859 * Otherwise, the best we can do 860 * is to use the filename pointed to. 861 */ 862 if (realdir_len == basedir_len) { 863 free(usefile); 864 usefile = linkfile; 865 start = p + 1; 866 } else { 867 free(linkfile); 868 start = usefile + basedir_len; 869 } 870 } while (/* CONSTCOND */ 0); 871 872 mlink = mandoc_calloc(1, sizeof(struct mlink)); 873 mlink->dform = FORM_NONE; 874 if (strlcpy(mlink->file, start, sizeof(mlink->file)) >= 875 sizeof(mlink->file)) { 876 say(start, "Filename too long"); 877 free(mlink); 878 free(usefile); 879 return; 880 } 881 882 /* 883 * In test mode or when the original name is absolute 884 * but outside our tree, guess the base directory. 885 */ 886 887 if (op == OP_TEST || (start == usefile && *start == '/')) { 888 if (strncmp(usefile, "man/", 4) == 0) 889 start = usefile + 4; 890 else if ((start = strstr(usefile, "/man/")) != NULL) 891 start += 5; 892 else 893 start = usefile; 894 } 895 896 /* 897 * First try to guess our directory structure. 898 * If we find a separator, try to look for man* or cat*. 899 * If we find one of these and what's underneath is a directory, 900 * assume it's an architecture. 901 */ 902 if ((p = strchr(start, '/')) != NULL) { 903 *p++ = '\0'; 904 if (strncmp(start, "man", 3) == 0) { 905 mlink->dform = FORM_SRC; 906 mlink->dsec = start + 3; 907 } else if (strncmp(start, "cat", 3) == 0) { 908 mlink->dform = FORM_CAT; 909 mlink->dsec = start + 3; 910 } 911 912 start = p; 913 if (mlink->dsec != NULL && (p = strchr(start, '/')) != NULL) { 914 *p++ = '\0'; 915 mlink->arch = start; 916 start = p; 917 } 918 } 919 920 /* 921 * Now check the file suffix. 922 * Suffix of `.0' indicates a catpage, `.1-9' is a manpage. 923 */ 924 p = strrchr(start, '\0'); 925 while (p-- > start && *p != '/' && *p != '.') 926 continue; 927 928 if (*p == '.') { 929 *p++ = '\0'; 930 mlink->fsec = p; 931 } 932 933 /* 934 * Now try to parse the name. 935 * Use the filename portion of the path. 936 */ 937 mlink->name = start; 938 if ((p = strrchr(start, '/')) != NULL) { 939 mlink->name = p + 1; 940 *p = '\0'; 941 } 942 mlink_add(mlink, &st); 943 free(usefile); 944} 945 946static void 947mlink_add(struct mlink *mlink, const struct stat *st) 948{ 949 struct inodev inodev; 950 struct mpage *mpage; 951 unsigned int slot; 952 953 assert(NULL != mlink->file); 954 955 mlink->dsec = mandoc_strdup(mlink->dsec ? mlink->dsec : ""); 956 mlink->arch = mandoc_strdup(mlink->arch ? mlink->arch : ""); 957 mlink->name = mandoc_strdup(mlink->name ? mlink->name : ""); 958 mlink->fsec = mandoc_strdup(mlink->fsec ? mlink->fsec : ""); 959 960 if ('0' == *mlink->fsec) { 961 free(mlink->fsec); 962 mlink->fsec = mandoc_strdup(mlink->dsec); 963 mlink->fform = FORM_CAT; 964 } else if ('1' <= *mlink->fsec && '9' >= *mlink->fsec) 965 mlink->fform = FORM_SRC; 966 else 967 mlink->fform = FORM_NONE; 968 969 slot = ohash_qlookup(&mlinks, mlink->file); 970 assert(NULL == ohash_find(&mlinks, slot)); 971 ohash_insert(&mlinks, slot, mlink); 972 973 memset(&inodev, 0, sizeof(inodev)); /* Clear padding. */ 974 inodev.st_ino = st->st_ino; 975 inodev.st_dev = st->st_dev; 976 slot = ohash_lookup_memory(&mpages, (char *)&inodev, 977 sizeof(struct inodev), inodev.st_ino); 978 mpage = ohash_find(&mpages, slot); 979 if (NULL == mpage) { 980 mpage = mandoc_calloc(1, sizeof(struct mpage)); 981 mpage->inodev.st_ino = inodev.st_ino; 982 mpage->inodev.st_dev = inodev.st_dev; 983 mpage->form = FORM_NONE; 984 mpage->next = mpage_head; 985 mpage_head = mpage; 986 ohash_insert(&mpages, slot, mpage); 987 } else 988 mlink->next = mpage->mlinks; 989 mpage->mlinks = mlink; 990 mlink->mpage = mpage; 991} 992 993static void 994mlink_free(struct mlink *mlink) 995{ 996 997 free(mlink->dsec); 998 free(mlink->arch); 999 free(mlink->name); 1000 free(mlink->fsec); 1001 free(mlink); 1002} 1003 1004static void 1005mpages_free(void) 1006{ 1007 struct mpage *mpage; 1008 struct mlink *mlink; 1009 1010 while ((mpage = mpage_head) != NULL) { 1011 while ((mlink = mpage->mlinks) != NULL) { 1012 mpage->mlinks = mlink->next; 1013 mlink_free(mlink); 1014 } 1015 mpage_head = mpage->next; 1016 free(mpage->sec); 1017 free(mpage->arch); 1018 free(mpage->title); 1019 free(mpage->desc); 1020 free(mpage); 1021 } 1022} 1023 1024/* 1025 * For each mlink to the mpage, check whether the path looks like 1026 * it is formatted, and if it does, check whether a source manual 1027 * exists by the same name, ignoring the suffix. 1028 * If both conditions hold, drop the mlink. 1029 */ 1030static void 1031mlinks_undupe(struct mpage *mpage) 1032{ 1033 char buf[PATH_MAX]; 1034 struct mlink **prev; 1035 struct mlink *mlink; 1036 char *bufp; 1037 1038 mpage->form = FORM_CAT; 1039 prev = &mpage->mlinks; 1040 while (NULL != (mlink = *prev)) { 1041 if (FORM_CAT != mlink->dform) { 1042 mpage->form = FORM_NONE; 1043 goto nextlink; 1044 } 1045 (void)strlcpy(buf, mlink->file, sizeof(buf)); 1046 bufp = strstr(buf, "cat"); 1047 assert(NULL != bufp); 1048 memcpy(bufp, "man", 3); 1049 if (NULL != (bufp = strrchr(buf, '.'))) 1050 *++bufp = '\0'; 1051 (void)strlcat(buf, mlink->dsec, sizeof(buf)); 1052 if (NULL == ohash_find(&mlinks, 1053 ohash_qlookup(&mlinks, buf))) 1054 goto nextlink; 1055 if (warnings) 1056 say(mlink->file, "Man source exists: %s", buf); 1057 if (use_all) 1058 goto nextlink; 1059 *prev = mlink->next; 1060 mlink_free(mlink); 1061 continue; 1062nextlink: 1063 prev = &(*prev)->next; 1064 } 1065} 1066 1067static void 1068mlink_check(struct mpage *mpage, struct mlink *mlink) 1069{ 1070 struct str *str; 1071 unsigned int slot; 1072 1073 /* 1074 * Check whether the manual section given in a file 1075 * agrees with the directory where the file is located. 1076 * Some manuals have suffixes like (3p) on their 1077 * section number either inside the file or in the 1078 * directory name, some are linked into more than one 1079 * section, like encrypt(1) = makekey(8). 1080 */ 1081 1082 if (FORM_SRC == mpage->form && 1083 strcasecmp(mpage->sec, mlink->dsec)) 1084 say(mlink->file, "Section \"%s\" manual in %s directory", 1085 mpage->sec, mlink->dsec); 1086 1087 /* 1088 * Manual page directories exist for each kernel 1089 * architecture as returned by machine(1). 1090 * However, many manuals only depend on the 1091 * application architecture as returned by arch(1). 1092 * For example, some (2/ARM) manuals are shared 1093 * across the "armish" and "zaurus" kernel 1094 * architectures. 1095 * A few manuals are even shared across completely 1096 * different architectures, for example fdformat(1) 1097 * on amd64, i386, and sparc64. 1098 */ 1099 1100 if (strcasecmp(mpage->arch, mlink->arch)) 1101 say(mlink->file, "Architecture \"%s\" manual in " 1102 "\"%s\" directory", mpage->arch, mlink->arch); 1103 1104 /* 1105 * XXX 1106 * parse_cat() doesn't set NAME_TITLE yet. 1107 */ 1108 1109 if (FORM_CAT == mpage->form) 1110 return; 1111 1112 /* 1113 * Check whether this mlink 1114 * appears as a name in the NAME section. 1115 */ 1116 1117 slot = ohash_qlookup(&names, mlink->name); 1118 str = ohash_find(&names, slot); 1119 assert(NULL != str); 1120 if ( ! (NAME_TITLE & str->mask)) 1121 say(mlink->file, "Name missing in NAME section"); 1122} 1123 1124/* 1125 * Run through the files in the global vector "mpages" 1126 * and add them to the database specified in "basedir". 1127 * 1128 * This handles the parsing scheme itself, using the cues of directory 1129 * and filename to determine whether the file is parsable or not. 1130 */ 1131static void 1132mpages_merge(struct dba *dba, struct mparse *mp) 1133{ 1134 struct mpage *mpage, *mpage_dest; 1135 struct mlink *mlink, *mlink_dest; 1136 struct roff_meta *meta; 1137 char *cp; 1138 int fd; 1139 1140 for (mpage = mpage_head; mpage != NULL; mpage = mpage->next) { 1141 mlinks_undupe(mpage); 1142 if ((mlink = mpage->mlinks) == NULL) 1143 continue; 1144 1145 name_mask = NAME_MASK; 1146 mandoc_ohash_init(&names, 4, offsetof(struct str, key)); 1147 mandoc_ohash_init(&strings, 6, offsetof(struct str, key)); 1148 mparse_reset(mp); 1149 meta = NULL; 1150 1151 if ((fd = mparse_open(mp, mlink->file)) == -1) { 1152 say(mlink->file, "&open"); 1153 goto nextpage; 1154 } 1155 1156 /* 1157 * Interpret the file as mdoc(7) or man(7) source 1158 * code, unless it is known to be formatted. 1159 */ 1160 if (mlink->dform != FORM_CAT || mlink->fform != FORM_CAT) { 1161 mparse_readfd(mp, fd, mlink->file); 1162 close(fd); 1163 fd = -1; 1164 meta = mparse_result(mp); 1165 } 1166 1167 if (meta != NULL && meta->sodest != NULL) { 1168 mlink_dest = ohash_find(&mlinks, 1169 ohash_qlookup(&mlinks, meta->sodest)); 1170 if (mlink_dest == NULL) { 1171 mandoc_asprintf(&cp, "%s.gz", meta->sodest); 1172 mlink_dest = ohash_find(&mlinks, 1173 ohash_qlookup(&mlinks, cp)); 1174 free(cp); 1175 } 1176 if (mlink_dest != NULL) { 1177 1178 /* The .so target exists. */ 1179 1180 mpage_dest = mlink_dest->mpage; 1181 while (1) { 1182 mlink->mpage = mpage_dest; 1183 1184 /* 1185 * If the target was already 1186 * processed, add the links 1187 * to the database now. 1188 * Otherwise, this will 1189 * happen when we come 1190 * to the target. 1191 */ 1192 1193 if (mpage_dest->dba != NULL) 1194 dbadd_mlink(mlink); 1195 1196 if (mlink->next == NULL) 1197 break; 1198 mlink = mlink->next; 1199 } 1200 1201 /* Move all links to the target. */ 1202 1203 mlink->next = mlink_dest->next; 1204 mlink_dest->next = mpage->mlinks; 1205 mpage->mlinks = NULL; 1206 goto nextpage; 1207 } 1208 meta->macroset = MACROSET_NONE; 1209 } 1210 if (meta != NULL && meta->macroset == MACROSET_MDOC) { 1211 mpage->form = FORM_SRC; 1212 mpage->sec = meta->msec; 1213 mpage->sec = mandoc_strdup( 1214 mpage->sec == NULL ? "" : mpage->sec); 1215 mpage->arch = meta->arch; 1216 mpage->arch = mandoc_strdup( 1217 mpage->arch == NULL ? "" : mpage->arch); 1218 mpage->title = mandoc_strdup(meta->title); 1219 } else if (meta != NULL && meta->macroset == MACROSET_MAN) { 1220 if (*meta->msec != '\0' || *meta->title != '\0') { 1221 mpage->form = FORM_SRC; 1222 mpage->sec = mandoc_strdup(meta->msec); 1223 mpage->arch = mandoc_strdup(mlink->arch); 1224 mpage->title = mandoc_strdup(meta->title); 1225 } else 1226 meta = NULL; 1227 } 1228 1229 assert(mpage->desc == NULL); 1230 if (meta == NULL || meta->sodest != NULL) { 1231 mpage->sec = mandoc_strdup(mlink->dsec); 1232 mpage->arch = mandoc_strdup(mlink->arch); 1233 mpage->title = mandoc_strdup(mlink->name); 1234 if (meta == NULL) { 1235 mpage->form = FORM_CAT; 1236 parse_cat(mpage, fd); 1237 } else 1238 mpage->form = FORM_SRC; 1239 } else if (meta->macroset == MACROSET_MDOC) 1240 parse_mdoc(mpage, meta, meta->first); 1241 else 1242 parse_man(mpage, meta, meta->first); 1243 if (mpage->desc == NULL) { 1244 mpage->desc = mandoc_strdup(mlink->name); 1245 if (warnings) 1246 say(mlink->file, "No one-line description, " 1247 "using filename \"%s\"", mlink->name); 1248 } 1249 1250 for (mlink = mpage->mlinks; 1251 mlink != NULL; 1252 mlink = mlink->next) { 1253 putkey(mpage, mlink->name, NAME_FILE); 1254 if (warnings && !use_all) 1255 mlink_check(mpage, mlink); 1256 } 1257 1258 dbadd(dba, mpage); 1259 1260nextpage: 1261 ohash_delete(&strings); 1262 ohash_delete(&names); 1263 } 1264} 1265 1266static void 1267parse_cat(struct mpage *mpage, int fd) 1268{ 1269 FILE *stream; 1270 struct mlink *mlink; 1271 char *line, *p, *title, *sec; 1272 size_t linesz, plen, titlesz; 1273 ssize_t len; 1274 int offs; 1275 1276 mlink = mpage->mlinks; 1277 stream = fd == -1 ? fopen(mlink->file, "r") : fdopen(fd, "r"); 1278 if (stream == NULL) { 1279 if (fd != -1) 1280 close(fd); 1281 if (warnings) 1282 say(mlink->file, "&fopen"); 1283 return; 1284 } 1285 1286 line = NULL; 1287 linesz = 0; 1288 1289 /* Parse the section number from the header line. */ 1290 1291 while (getline(&line, &linesz, stream) != -1) { 1292 if (*line == '\n') 1293 continue; 1294 if ((sec = strchr(line, '(')) == NULL) 1295 break; 1296 if ((p = strchr(++sec, ')')) == NULL) 1297 break; 1298 free(mpage->sec); 1299 mpage->sec = mandoc_strndup(sec, p - sec); 1300 if (warnings && *mlink->dsec != '\0' && 1301 strcasecmp(mpage->sec, mlink->dsec)) 1302 say(mlink->file, 1303 "Section \"%s\" manual in %s directory", 1304 mpage->sec, mlink->dsec); 1305 break; 1306 } 1307 1308 /* Skip to first blank line. */ 1309 1310 while (line == NULL || *line != '\n') 1311 if (getline(&line, &linesz, stream) == -1) 1312 break; 1313 1314 /* 1315 * Assume the first line that is not indented 1316 * is the first section header. Skip to it. 1317 */ 1318 1319 while (getline(&line, &linesz, stream) != -1) 1320 if (*line != '\n' && *line != ' ') 1321 break; 1322 1323 /* 1324 * Read up until the next section into a buffer. 1325 * Strip the leading and trailing newline from each read line, 1326 * appending a trailing space. 1327 * Ignore empty (whitespace-only) lines. 1328 */ 1329 1330 titlesz = 0; 1331 title = NULL; 1332 1333 while ((len = getline(&line, &linesz, stream)) != -1) { 1334 if (*line != ' ') 1335 break; 1336 offs = 0; 1337 while (isspace((unsigned char)line[offs])) 1338 offs++; 1339 if (line[offs] == '\0') 1340 continue; 1341 title = mandoc_realloc(title, titlesz + len - offs); 1342 memcpy(title + titlesz, line + offs, len - offs); 1343 titlesz += len - offs; 1344 title[titlesz - 1] = ' '; 1345 } 1346 free(line); 1347 1348 /* 1349 * If no page content can be found, or the input line 1350 * is already the next section header, or there is no 1351 * trailing newline, reuse the page title as the page 1352 * description. 1353 */ 1354 1355 if (NULL == title || '\0' == *title) { 1356 if (warnings) 1357 say(mlink->file, "Cannot find NAME section"); 1358 fclose(stream); 1359 free(title); 1360 return; 1361 } 1362 1363 title[titlesz - 1] = '\0'; 1364 1365 /* 1366 * Skip to the first dash. 1367 * Use the remaining line as the description (no more than 70 1368 * bytes). 1369 */ 1370 1371 if (NULL != (p = strstr(title, "- "))) { 1372 for (p += 2; ' ' == *p || '\b' == *p; p++) 1373 /* Skip to next word. */ ; 1374 } else { 1375 if (warnings) 1376 say(mlink->file, "No dash in title line, " 1377 "reusing \"%s\" as one-line description", title); 1378 p = title; 1379 } 1380 1381 plen = strlen(p); 1382 1383 /* Strip backspace-encoding from line. */ 1384 1385 while (NULL != (line = memchr(p, '\b', plen))) { 1386 len = line - p; 1387 if (0 == len) { 1388 memmove(line, line + 1, plen--); 1389 continue; 1390 } 1391 memmove(line - 1, line + 1, plen - len); 1392 plen -= 2; 1393 } 1394 1395 /* 1396 * Cut off excessive one-line descriptions. 1397 * Bad pages are not worth better heuristics. 1398 */ 1399 1400 mpage->desc = mandoc_strndup(p, 150); 1401 fclose(stream); 1402 free(title); 1403} 1404 1405/* 1406 * Put a type/word pair into the word database for this particular file. 1407 */ 1408static void 1409putkey(const struct mpage *mpage, char *value, uint64_t type) 1410{ 1411 putkeys(mpage, value, strlen(value), type); 1412} 1413 1414/* 1415 * Grok all nodes at or below a certain mdoc node into putkey(). 1416 */ 1417static void 1418putmdockey(const struct mpage *mpage, 1419 const struct roff_node *n, uint64_t m, int taboo) 1420{ 1421 1422 for ( ; NULL != n; n = n->next) { 1423 if (n->flags & taboo) 1424 continue; 1425 if (NULL != n->child) 1426 putmdockey(mpage, n->child, m, taboo); 1427 if (n->type == ROFFT_TEXT) 1428 putkey(mpage, n->string, m); 1429 } 1430} 1431 1432static void 1433parse_man(struct mpage *mpage, const struct roff_meta *meta, 1434 const struct roff_node *n) 1435{ 1436 const struct roff_node *head, *body; 1437 char *start, *title; 1438 char byte; 1439 size_t sz; 1440 1441 if (n == NULL) 1442 return; 1443 1444 /* 1445 * We're only searching for one thing: the first text child in 1446 * the BODY of a NAME section. Since we don't keep track of 1447 * sections in -man, run some hoops to find out whether we're in 1448 * the correct section or not. 1449 */ 1450 1451 if (n->type == ROFFT_BODY && n->tok == MAN_SH) { 1452 body = n; 1453 if ((head = body->parent->head) != NULL && 1454 (head = head->child) != NULL && 1455 head->next == NULL && 1456 head->type == ROFFT_TEXT && 1457 strcmp(head->string, "NAME") == 0 && 1458 body->child != NULL) { 1459 1460 /* 1461 * Suck the entire NAME section into memory. 1462 * Yes, we might run away. 1463 * But too many manuals have big, spread-out 1464 * NAME sections over many lines. 1465 */ 1466 1467 title = NULL; 1468 deroff(&title, body); 1469 if (NULL == title) 1470 return; 1471 1472 /* 1473 * Go through a special heuristic dance here. 1474 * Conventionally, one or more manual names are 1475 * comma-specified prior to a whitespace, then a 1476 * dash, then a description. Try to puzzle out 1477 * the name parts here. 1478 */ 1479 1480 start = title; 1481 for ( ;; ) { 1482 sz = strcspn(start, " ,"); 1483 if ('\0' == start[sz]) 1484 break; 1485 1486 byte = start[sz]; 1487 start[sz] = '\0'; 1488 1489 /* 1490 * Assume a stray trailing comma in the 1491 * name list if a name begins with a dash. 1492 */ 1493 1494 if ('-' == start[0] || 1495 ('\\' == start[0] && '-' == start[1])) 1496 break; 1497 1498 putkey(mpage, start, NAME_TITLE); 1499 if ( ! (mpage->name_head_done || 1500 strcasecmp(start, meta->title))) { 1501 putkey(mpage, start, NAME_HEAD); 1502 mpage->name_head_done = 1; 1503 } 1504 1505 if (' ' == byte) { 1506 start += sz + 1; 1507 break; 1508 } 1509 1510 assert(',' == byte); 1511 start += sz + 1; 1512 while (' ' == *start) 1513 start++; 1514 } 1515 1516 if (start == title) { 1517 putkey(mpage, start, NAME_TITLE); 1518 if ( ! (mpage->name_head_done || 1519 strcasecmp(start, meta->title))) { 1520 putkey(mpage, start, NAME_HEAD); 1521 mpage->name_head_done = 1; 1522 } 1523 free(title); 1524 return; 1525 } 1526 1527 while (isspace((unsigned char)*start)) 1528 start++; 1529 1530 if (0 == strncmp(start, "-", 1)) 1531 start += 1; 1532 else if (0 == strncmp(start, "\\-\\-", 4)) 1533 start += 4; 1534 else if (0 == strncmp(start, "\\-", 2)) 1535 start += 2; 1536 else if (0 == strncmp(start, "\\(en", 4)) 1537 start += 4; 1538 else if (0 == strncmp(start, "\\(em", 4)) 1539 start += 4; 1540 1541 while (' ' == *start) 1542 start++; 1543 1544 /* 1545 * Cut off excessive one-line descriptions. 1546 * Bad pages are not worth better heuristics. 1547 */ 1548 1549 mpage->desc = mandoc_strndup(start, 150); 1550 free(title); 1551 return; 1552 } 1553 } 1554 1555 for (n = n->child; n; n = n->next) { 1556 if (NULL != mpage->desc) 1557 break; 1558 parse_man(mpage, meta, n); 1559 } 1560} 1561 1562static void 1563parse_mdoc(struct mpage *mpage, const struct roff_meta *meta, 1564 const struct roff_node *n) 1565{ 1566 const struct mdoc_handler *handler; 1567 1568 for (n = n->child; n != NULL; n = n->next) { 1569 if (n->tok == TOKEN_NONE || n->tok < ROFF_MAX) 1570 continue; 1571 assert(n->tok >= MDOC_Dd && n->tok < MDOC_MAX); 1572 handler = mdoc_handlers + (n->tok - MDOC_Dd); 1573 if (n->flags & handler->taboo) 1574 continue; 1575 1576 switch (n->type) { 1577 case ROFFT_ELEM: 1578 case ROFFT_BLOCK: 1579 case ROFFT_HEAD: 1580 case ROFFT_BODY: 1581 case ROFFT_TAIL: 1582 if (handler->fp != NULL && 1583 (*handler->fp)(mpage, meta, n) == 0) 1584 break; 1585 if (handler->mask) 1586 putmdockey(mpage, n->child, 1587 handler->mask, handler->taboo); 1588 break; 1589 default: 1590 continue; 1591 } 1592 if (NULL != n->child) 1593 parse_mdoc(mpage, meta, n); 1594 } 1595} 1596 1597static int 1598parse_mdoc_Fa(struct mpage *mpage, const struct roff_meta *meta, 1599 const struct roff_node *n) 1600{ 1601 uint64_t mask; 1602 1603 mask = TYPE_Fa; 1604 if (n->sec == SEC_SYNOPSIS) 1605 mask |= TYPE_Vt; 1606 1607 putmdockey(mpage, n->child, mask, 0); 1608 return 0; 1609} 1610 1611static int 1612parse_mdoc_Fd(struct mpage *mpage, const struct roff_meta *meta, 1613 const struct roff_node *n) 1614{ 1615 char *start, *end; 1616 size_t sz; 1617 1618 if (SEC_SYNOPSIS != n->sec || 1619 NULL == (n = n->child) || 1620 n->type != ROFFT_TEXT) 1621 return 0; 1622 1623 /* 1624 * Only consider those `Fd' macro fields that begin with an 1625 * "inclusion" token (versus, e.g., #define). 1626 */ 1627 1628 if (strcmp("#include", n->string)) 1629 return 0; 1630 1631 if ((n = n->next) == NULL || n->type != ROFFT_TEXT) 1632 return 0; 1633 1634 /* 1635 * Strip away the enclosing angle brackets and make sure we're 1636 * not zero-length. 1637 */ 1638 1639 start = n->string; 1640 if ('<' == *start || '"' == *start) 1641 start++; 1642 1643 if (0 == (sz = strlen(start))) 1644 return 0; 1645 1646 end = &start[(int)sz - 1]; 1647 if ('>' == *end || '"' == *end) 1648 end--; 1649 1650 if (end > start) 1651 putkeys(mpage, start, end - start + 1, TYPE_In); 1652 return 0; 1653} 1654 1655static void 1656parse_mdoc_fname(struct mpage *mpage, const struct roff_node *n) 1657{ 1658 char *cp; 1659 size_t sz; 1660 1661 if (n->type != ROFFT_TEXT) 1662 return; 1663 1664 /* Skip function pointer punctuation. */ 1665 1666 cp = n->string; 1667 while (*cp == '(' || *cp == '*') 1668 cp++; 1669 sz = strcspn(cp, "()"); 1670 1671 putkeys(mpage, cp, sz, TYPE_Fn); 1672 if (n->sec == SEC_SYNOPSIS) 1673 putkeys(mpage, cp, sz, NAME_SYN); 1674} 1675 1676static int 1677parse_mdoc_Fn(struct mpage *mpage, const struct roff_meta *meta, 1678 const struct roff_node *n) 1679{ 1680 uint64_t mask; 1681 1682 if (n->child == NULL) 1683 return 0; 1684 1685 parse_mdoc_fname(mpage, n->child); 1686 1687 n = n->child->next; 1688 if (n != NULL && n->type == ROFFT_TEXT) { 1689 mask = TYPE_Fa; 1690 if (n->sec == SEC_SYNOPSIS) 1691 mask |= TYPE_Vt; 1692 putmdockey(mpage, n, mask, 0); 1693 } 1694 1695 return 0; 1696} 1697 1698static int 1699parse_mdoc_Fo(struct mpage *mpage, const struct roff_meta *meta, 1700 const struct roff_node *n) 1701{ 1702 1703 if (n->type != ROFFT_HEAD) 1704 return 1; 1705 1706 if (n->child != NULL) 1707 parse_mdoc_fname(mpage, n->child); 1708 1709 return 0; 1710} 1711 1712static int 1713parse_mdoc_Va(struct mpage *mpage, const struct roff_meta *meta, 1714 const struct roff_node *n) 1715{ 1716 char *cp; 1717 1718 if (n->type != ROFFT_ELEM && n->type != ROFFT_BODY) 1719 return 0; 1720 1721 if (n->child != NULL && 1722 n->child->next == NULL && 1723 n->child->type == ROFFT_TEXT) 1724 return 1; 1725 1726 cp = NULL; 1727 deroff(&cp, n); 1728 if (cp != NULL) { 1729 putkey(mpage, cp, TYPE_Vt | (n->tok == MDOC_Va || 1730 n->type == ROFFT_BODY ? TYPE_Va : 0)); 1731 free(cp); 1732 } 1733 1734 return 0; 1735} 1736 1737static int 1738parse_mdoc_Xr(struct mpage *mpage, const struct roff_meta *meta, 1739 const struct roff_node *n) 1740{ 1741 char *cp; 1742 1743 if (NULL == (n = n->child)) 1744 return 0; 1745 1746 if (NULL == n->next) { 1747 putkey(mpage, n->string, TYPE_Xr); 1748 return 0; 1749 } 1750 1751 mandoc_asprintf(&cp, "%s(%s)", n->string, n->next->string); 1752 putkey(mpage, cp, TYPE_Xr); 1753 free(cp); 1754 return 0; 1755} 1756 1757static int 1758parse_mdoc_Nd(struct mpage *mpage, const struct roff_meta *meta, 1759 const struct roff_node *n) 1760{ 1761 1762 if (n->type == ROFFT_BODY) 1763 deroff(&mpage->desc, n); 1764 return 0; 1765} 1766 1767static int 1768parse_mdoc_Nm(struct mpage *mpage, const struct roff_meta *meta, 1769 const struct roff_node *n) 1770{ 1771 1772 if (SEC_NAME == n->sec) 1773 putmdockey(mpage, n->child, NAME_TITLE, 0); 1774 else if (n->sec == SEC_SYNOPSIS && n->type == ROFFT_HEAD) { 1775 if (n->child == NULL) 1776 putkey(mpage, meta->name, NAME_SYN); 1777 else 1778 putmdockey(mpage, n->child, NAME_SYN, 0); 1779 } 1780 if ( ! (mpage->name_head_done || 1781 n->child == NULL || n->child->string == NULL || 1782 strcasecmp(n->child->string, meta->title))) { 1783 putkey(mpage, n->child->string, NAME_HEAD); 1784 mpage->name_head_done = 1; 1785 } 1786 return 0; 1787} 1788 1789static int 1790parse_mdoc_Sh(struct mpage *mpage, const struct roff_meta *meta, 1791 const struct roff_node *n) 1792{ 1793 1794 return n->sec == SEC_CUSTOM && n->type == ROFFT_HEAD; 1795} 1796 1797static int 1798parse_mdoc_head(struct mpage *mpage, const struct roff_meta *meta, 1799 const struct roff_node *n) 1800{ 1801 1802 return n->type == ROFFT_HEAD; 1803} 1804 1805/* 1806 * Add a string to the hash table for the current manual. 1807 * Each string has a bitmask telling which macros it belongs to. 1808 * When we finish the manual, we'll dump the table. 1809 */ 1810static void 1811putkeys(const struct mpage *mpage, char *cp, size_t sz, uint64_t v) 1812{ 1813 struct ohash *htab; 1814 struct str *s; 1815 const char *end; 1816 unsigned int slot; 1817 int i, mustfree; 1818 1819 if (0 == sz) 1820 return; 1821 1822 mustfree = render_string(&cp, &sz); 1823 1824 if (TYPE_Nm & v) { 1825 htab = &names; 1826 v &= name_mask; 1827 if (v & NAME_FIRST) 1828 name_mask &= ~NAME_FIRST; 1829 if (debug > 1) 1830 say(mpage->mlinks->file, 1831 "Adding name %*s, bits=0x%llx", (int)sz, cp, 1832 (unsigned long long)v); 1833 } else { 1834 htab = &strings; 1835 if (debug > 1) 1836 for (i = 0; i < KEY_MAX; i++) 1837 if ((uint64_t)1 << i & v) 1838 say(mpage->mlinks->file, 1839 "Adding key %s=%*s", 1840 mansearch_keynames[i], (int)sz, cp); 1841 } 1842 1843 end = cp + sz; 1844 slot = ohash_qlookupi(htab, cp, &end); 1845 s = ohash_find(htab, slot); 1846 1847 if (NULL != s && mpage == s->mpage) { 1848 s->mask |= v; 1849 return; 1850 } else if (NULL == s) { 1851 s = mandoc_calloc(1, sizeof(struct str) + sz + 1); 1852 memcpy(s->key, cp, sz); 1853 ohash_insert(htab, slot, s); 1854 } 1855 s->mpage = mpage; 1856 s->mask = v; 1857 1858 if (mustfree) 1859 free(cp); 1860} 1861 1862/* 1863 * Take a Unicode codepoint and produce its UTF-8 encoding. 1864 * This isn't the best way to do this, but it works. 1865 * The magic numbers are from the UTF-8 packaging. 1866 * Read the UTF-8 spec or the utf8(7) manual page for details. 1867 */ 1868static size_t 1869utf8(unsigned int cp, char out[5]) 1870{ 1871 size_t rc; 1872 1873 if (cp <= 0x7f) { 1874 rc = 1; 1875 out[0] = (char)cp; 1876 } else if (cp <= 0x7ff) { 1877 rc = 2; 1878 out[0] = (cp >> 6 & 31) | 192; 1879 out[1] = (cp & 63) | 128; 1880 } else if (cp >= 0xd800 && cp <= 0xdfff) { 1881 rc = 0; /* reject UTF-16 surrogate */ 1882 } else if (cp <= 0xffff) { 1883 rc = 3; 1884 out[0] = (cp >> 12 & 15) | 224; 1885 out[1] = (cp >> 6 & 63) | 128; 1886 out[2] = (cp & 63) | 128; 1887 } else if (cp <= 0x10ffff) { 1888 rc = 4; 1889 out[0] = (cp >> 18 & 7) | 240; 1890 out[1] = (cp >> 12 & 63) | 128; 1891 out[2] = (cp >> 6 & 63) | 128; 1892 out[3] = (cp & 63) | 128; 1893 } else 1894 rc = 0; 1895 1896 out[rc] = '\0'; 1897 return rc; 1898} 1899 1900/* 1901 * If the string contains escape sequences, 1902 * replace it with an allocated rendering and return 1, 1903 * such that the caller can free it after use. 1904 * Otherwise, do nothing and return 0. 1905 */ 1906static int 1907render_string(char **public, size_t *psz) 1908{ 1909 const char *src, *scp, *addcp, *seq; 1910 char *dst; 1911 size_t ssz, dsz, addsz; 1912 char utfbuf[7], res[6]; 1913 int seqlen, unicode; 1914 1915 res[0] = '\\'; 1916 res[1] = '\t'; 1917 res[2] = ASCII_NBRSP; 1918 res[3] = ASCII_HYPH; 1919 res[4] = ASCII_BREAK; 1920 res[5] = '\0'; 1921 1922 src = scp = *public; 1923 ssz = *psz; 1924 dst = NULL; 1925 dsz = 0; 1926 1927 while (scp < src + *psz) { 1928 1929 /* Leave normal characters unchanged. */ 1930 1931 if (strchr(res, *scp) == NULL) { 1932 if (dst != NULL) 1933 dst[dsz++] = *scp; 1934 scp++; 1935 continue; 1936 } 1937 1938 /* 1939 * Found something that requires replacing, 1940 * make sure we have a destination buffer. 1941 */ 1942 1943 if (dst == NULL) { 1944 dst = mandoc_malloc(ssz + 1); 1945 dsz = scp - src; 1946 memcpy(dst, src, dsz); 1947 } 1948 1949 /* Handle single-char special characters. */ 1950 1951 switch (*scp) { 1952 case '\\': 1953 break; 1954 case '\t': 1955 case ASCII_NBRSP: 1956 dst[dsz++] = ' '; 1957 scp++; 1958 continue; 1959 case ASCII_HYPH: 1960 dst[dsz++] = '-'; 1961 /* FALLTHROUGH */ 1962 case ASCII_BREAK: 1963 scp++; 1964 continue; 1965 default: 1966 abort(); 1967 } 1968 1969 /* 1970 * Found an escape sequence. 1971 * Read past the slash, then parse it. 1972 * Ignore everything except characters. 1973 */ 1974 1975 scp++; 1976 switch (mandoc_escape(&scp, &seq, &seqlen)) { 1977 case ESCAPE_UNICODE: 1978 unicode = mchars_num2uc(seq + 1, seqlen - 1); 1979 break; 1980 case ESCAPE_NUMBERED: 1981 unicode = mchars_num2char(seq, seqlen); 1982 break; 1983 case ESCAPE_SPECIAL: 1984 unicode = mchars_spec2cp(seq, seqlen); 1985 break; 1986 default: 1987 unicode = -1; 1988 break; 1989 } 1990 if (unicode <= 0) 1991 continue; 1992 1993 /* 1994 * Render the special character 1995 * as either UTF-8 or ASCII. 1996 */ 1997 1998 if (write_utf8) { 1999 addsz = utf8(unicode, utfbuf); 2000 if (addsz == 0) 2001 continue; 2002 addcp = utfbuf; 2003 } else { 2004 addcp = mchars_uc2str(unicode); 2005 if (addcp == NULL) 2006 continue; 2007 if (*addcp == ASCII_NBRSP) 2008 addcp = " "; 2009 addsz = strlen(addcp); 2010 } 2011 2012 /* Copy the rendered glyph into the stream. */ 2013 2014 ssz += addsz; 2015 dst = mandoc_realloc(dst, ssz + 1); 2016 memcpy(dst + dsz, addcp, addsz); 2017 dsz += addsz; 2018 } 2019 if (dst != NULL) { 2020 *public = dst; 2021 *psz = dsz; 2022 } 2023 2024 /* Trim trailing whitespace and NUL-terminate. */ 2025 2026 while (*psz > 0 && (*public)[*psz - 1] == ' ') 2027 --*psz; 2028 if (dst != NULL) { 2029 (*public)[*psz] = '\0'; 2030 return 1; 2031 } else 2032 return 0; 2033} 2034 2035static void 2036dbadd_mlink(const struct mlink *mlink) 2037{ 2038 dba_page_alias(mlink->mpage->dba, mlink->name, NAME_FILE); 2039 dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->dsec); 2040 dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->fsec); 2041 dba_page_add(mlink->mpage->dba, DBP_ARCH, mlink->arch); 2042 dba_page_add(mlink->mpage->dba, DBP_FILE, mlink->file); 2043} 2044 2045/* 2046 * Flush the current page's terms (and their bits) into the database. 2047 * Also, handle escape sequences at the last possible moment. 2048 */ 2049static void 2050dbadd(struct dba *dba, struct mpage *mpage) 2051{ 2052 struct mlink *mlink; 2053 struct str *key; 2054 char *cp; 2055 uint64_t mask; 2056 size_t i; 2057 unsigned int slot; 2058 int mustfree; 2059 2060 mlink = mpage->mlinks; 2061 2062 if (nodb) { 2063 for (key = ohash_first(&names, &slot); NULL != key; 2064 key = ohash_next(&names, &slot)) 2065 free(key); 2066 for (key = ohash_first(&strings, &slot); NULL != key; 2067 key = ohash_next(&strings, &slot)) 2068 free(key); 2069 if (0 == debug) 2070 return; 2071 while (NULL != mlink) { 2072 fputs(mlink->name, stdout); 2073 if (NULL == mlink->next || 2074 strcmp(mlink->dsec, mlink->next->dsec) || 2075 strcmp(mlink->fsec, mlink->next->fsec) || 2076 strcmp(mlink->arch, mlink->next->arch)) { 2077 putchar('('); 2078 if ('\0' == *mlink->dsec) 2079 fputs(mlink->fsec, stdout); 2080 else 2081 fputs(mlink->dsec, stdout); 2082 if ('\0' != *mlink->arch) 2083 printf("/%s", mlink->arch); 2084 putchar(')'); 2085 } 2086 mlink = mlink->next; 2087 if (NULL != mlink) 2088 fputs(", ", stdout); 2089 } 2090 printf(" - %s\n", mpage->desc); 2091 return; 2092 } 2093 2094 if (debug) 2095 say(mlink->file, "Adding to database"); 2096 2097 cp = mpage->desc; 2098 i = strlen(cp); 2099 mustfree = render_string(&cp, &i); 2100 mpage->dba = dba_page_new(dba->pages, 2101 *mpage->arch == '\0' ? mlink->arch : mpage->arch, 2102 cp, mlink->file, mpage->form); 2103 if (mustfree) 2104 free(cp); 2105 dba_page_add(mpage->dba, DBP_SECT, mpage->sec); 2106 2107 while (mlink != NULL) { 2108 dbadd_mlink(mlink); 2109 mlink = mlink->next; 2110 } 2111 2112 for (key = ohash_first(&names, &slot); NULL != key; 2113 key = ohash_next(&names, &slot)) { 2114 assert(key->mpage == mpage); 2115 dba_page_alias(mpage->dba, key->key, key->mask); 2116 free(key); 2117 } 2118 for (key = ohash_first(&strings, &slot); NULL != key; 2119 key = ohash_next(&strings, &slot)) { 2120 assert(key->mpage == mpage); 2121 i = 0; 2122 for (mask = TYPE_Xr; mask <= TYPE_Lb; mask *= 2) { 2123 if (key->mask & mask) 2124 dba_macro_add(dba->macros, i, 2125 key->key, mpage->dba); 2126 i++; 2127 } 2128 free(key); 2129 } 2130} 2131 2132static void 2133dbprune(struct dba *dba) 2134{ 2135 struct dba_array *page, *files; 2136 char *file; 2137 2138 dba_array_FOREACH(dba->pages, page) { 2139 files = dba_array_get(page, DBP_FILE); 2140 dba_array_FOREACH(files, file) { 2141 if (*file < ' ') 2142 file++; 2143 if (ohash_find(&mlinks, ohash_qlookup(&mlinks, 2144 file)) != NULL) { 2145 if (debug) 2146 say(file, "Deleting from database"); 2147 dba_array_del(dba->pages); 2148 break; 2149 } 2150 } 2151 } 2152} 2153 2154/* 2155 * Write the database from memory to disk. 2156 */ 2157static void 2158dbwrite(struct dba *dba) 2159{ 2160 struct stat sb1, sb2; 2161 char tfn[33], *cp1, *cp2; 2162 off_t i; 2163 int fd1, fd2; 2164 2165 /* 2166 * Do not write empty databases, and delete existing ones 2167 * when makewhatis -u causes them to become empty. 2168 */ 2169 2170 dba_array_start(dba->pages); 2171 if (dba_array_next(dba->pages) == NULL) { 2172 if (unlink(MANDOC_DB) == -1 && errno != ENOENT) 2173 say(MANDOC_DB, "&unlink"); 2174 return; 2175 } 2176 2177 /* 2178 * Build the database in a temporary file, 2179 * then atomically move it into place. 2180 */ 2181 2182 if (dba_write(MANDOC_DB "~", dba) != -1) { 2183 if (rename(MANDOC_DB "~", MANDOC_DB) == -1) { 2184 exitcode = (int)MANDOCLEVEL_SYSERR; 2185 say(MANDOC_DB, "&rename"); 2186 unlink(MANDOC_DB "~"); 2187 } 2188 return; 2189 } 2190 2191 /* 2192 * We lack write permission and cannot replace the database 2193 * file, but let's at least check whether the data changed. 2194 */ 2195 2196 (void)strlcpy(tfn, "/tmp/mandocdb.XXXXXXXX", sizeof(tfn)); 2197 if (mkdtemp(tfn) == NULL) { 2198 exitcode = (int)MANDOCLEVEL_SYSERR; 2199 say("", "&%s", tfn); 2200 return; 2201 } 2202 cp1 = cp2 = MAP_FAILED; 2203 fd1 = fd2 = -1; 2204 (void)strlcat(tfn, "/" MANDOC_DB, sizeof(tfn)); 2205 if (dba_write(tfn, dba) == -1) { 2206 say(tfn, "&dba_write"); 2207 goto err; 2208 } 2209 if ((fd1 = open(MANDOC_DB, O_RDONLY)) == -1) { 2210 say(MANDOC_DB, "&open"); 2211 goto err; 2212 } 2213 if ((fd2 = open(tfn, O_RDONLY)) == -1) { 2214 say(tfn, "&open"); 2215 goto err; 2216 } 2217 if (fstat(fd1, &sb1) == -1) { 2218 say(MANDOC_DB, "&fstat"); 2219 goto err; 2220 } 2221 if (fstat(fd2, &sb2) == -1) { 2222 say(tfn, "&fstat"); 2223 goto err; 2224 } 2225 if (sb1.st_size != sb2.st_size) 2226 goto err; 2227 if ((cp1 = mmap(NULL, sb1.st_size, PROT_READ, MAP_PRIVATE, 2228 fd1, 0)) == MAP_FAILED) { 2229 say(MANDOC_DB, "&mmap"); 2230 goto err; 2231 } 2232 if ((cp2 = mmap(NULL, sb2.st_size, PROT_READ, MAP_PRIVATE, 2233 fd2, 0)) == MAP_FAILED) { 2234 say(tfn, "&mmap"); 2235 goto err; 2236 } 2237 for (i = 0; i < sb1.st_size; i++) 2238 if (cp1[i] != cp2[i]) 2239 goto err; 2240 goto out; 2241 2242err: 2243 exitcode = (int)MANDOCLEVEL_SYSERR; 2244 say(MANDOC_DB, "Data changed, but cannot replace database"); 2245 2246out: 2247 if (cp1 != MAP_FAILED) 2248 munmap(cp1, sb1.st_size); 2249 if (cp2 != MAP_FAILED) 2250 munmap(cp2, sb2.st_size); 2251 if (fd1 != -1) 2252 close(fd1); 2253 if (fd2 != -1) 2254 close(fd2); 2255 unlink(tfn); 2256 *strrchr(tfn, '/') = '\0'; 2257 rmdir(tfn); 2258} 2259 2260static int 2261set_basedir(const char *targetdir, int report_baddir) 2262{ 2263 static char startdir[PATH_MAX]; 2264 static int getcwd_status; /* 1 = ok, 2 = failure */ 2265 static int chdir_status; /* 1 = changed directory */ 2266 2267 /* 2268 * Remember the original working directory, if possible. 2269 * This will be needed if the second or a later directory 2270 * on the command line is given as a relative path. 2271 * Do not error out if the current directory is not 2272 * searchable: Maybe it won't be needed after all. 2273 */ 2274 if (getcwd_status == 0) { 2275 if (getcwd(startdir, sizeof(startdir)) == NULL) { 2276 getcwd_status = 2; 2277 (void)strlcpy(startdir, strerror(errno), 2278 sizeof(startdir)); 2279 } else 2280 getcwd_status = 1; 2281 } 2282 2283 /* 2284 * We are leaving the old base directory. 2285 * Do not use it any longer, not even for messages. 2286 */ 2287 *basedir = '\0'; 2288 basedir_len = 0; 2289 2290 /* 2291 * If and only if the directory was changed earlier and 2292 * the next directory to process is given as a relative path, 2293 * first go back, or bail out if that is impossible. 2294 */ 2295 if (chdir_status && *targetdir != '/') { 2296 if (getcwd_status == 2) { 2297 exitcode = (int)MANDOCLEVEL_SYSERR; 2298 say("", "getcwd: %s", startdir); 2299 return 0; 2300 } 2301 if (chdir(startdir) == -1) { 2302 exitcode = (int)MANDOCLEVEL_SYSERR; 2303 say("", "&chdir %s", startdir); 2304 return 0; 2305 } 2306 } 2307 2308 /* 2309 * Always resolve basedir to the canonicalized absolute 2310 * pathname and append a trailing slash, such that 2311 * we can reliably check whether files are inside. 2312 */ 2313 if (realpath(targetdir, basedir) == NULL) { 2314 if (report_baddir || errno != ENOENT) { 2315 exitcode = (int)MANDOCLEVEL_BADARG; 2316 say("", "&%s: realpath", targetdir); 2317 } 2318 *basedir = '\0'; 2319 return 0; 2320 } else if (chdir(basedir) == -1) { 2321 if (report_baddir || errno != ENOENT) { 2322 exitcode = (int)MANDOCLEVEL_BADARG; 2323 say("", "&chdir"); 2324 } 2325 *basedir = '\0'; 2326 return 0; 2327 } 2328 chdir_status = 1; 2329 basedir_len = strlen(basedir); 2330 if (basedir[basedir_len - 1] != '/') { 2331 if (basedir_len >= PATH_MAX - 1) { 2332 exitcode = (int)MANDOCLEVEL_SYSERR; 2333 say("", "Filename too long"); 2334 *basedir = '\0'; 2335 basedir_len = 0; 2336 return 0; 2337 } 2338 basedir[basedir_len++] = '/'; 2339 basedir[basedir_len] = '\0'; 2340 } 2341 return 1; 2342} 2343 2344static void 2345say(const char *file, const char *format, ...) 2346{ 2347 va_list ap; 2348 int use_errno; 2349 2350 if (*basedir != '\0') 2351 fprintf(stderr, "%s", basedir); 2352 if (*basedir != '\0' && *file != '\0') 2353 fputc('/', stderr); 2354 if (*file != '\0') 2355 fprintf(stderr, "%s", file); 2356 2357 use_errno = 1; 2358 if (format != NULL) { 2359 switch (*format) { 2360 case '&': 2361 format++; 2362 break; 2363 case '\0': 2364 format = NULL; 2365 break; 2366 default: 2367 use_errno = 0; 2368 break; 2369 } 2370 } 2371 if (format != NULL) { 2372 if (*basedir != '\0' || *file != '\0') 2373 fputs(": ", stderr); 2374 va_start(ap, format); 2375 vfprintf(stderr, format, ap); 2376 va_end(ap); 2377 } 2378 if (use_errno) { 2379 if (*basedir != '\0' || *file != '\0' || format != NULL) 2380 fputs(": ", stderr); 2381 perror(NULL); 2382 } else 2383 fputc('\n', stderr); 2384} 2385