1 /* $Id: mandocdb.c,v 1.258 2018/02/23 18:25:57 schwarze Exp $ */
2 /*
3 * Copyright (c) 2011, 2012 Kristaps Dzonsons <kristaps@bsd.lv>
4 * Copyright (c) 2011-2017 Ingo Schwarze <schwarze@openbsd.org>
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 #include "config.h"
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 #if HAVE_ERR
28 #include <err.h>
29 #endif
30 #include <errno.h>
31 #include <fcntl.h>
32 #if HAVE_FTS
33 #include <fts.h>
34 #else
35 #include "compat_fts.h"
36 #endif
37 #include <limits.h>
38 #if HAVE_SANDBOX_INIT
39 #include <sandbox.h>
40 #endif
41 #include <stdarg.h>
42 #include <stddef.h>
43 #include <stdio.h>
44 #include <stdint.h>
45 #include <stdlib.h>
46 #include <string.h>
47 #include <unistd.h>
48
49 #include "mandoc_aux.h"
50 #include "mandoc_ohash.h"
51 #include "mandoc.h"
52 #include "roff.h"
53 #include "mdoc.h"
54 #include "man.h"
55 #include "manconf.h"
56 #include "mansearch.h"
57 #include "dba_array.h"
58 #include "dba.h"
59
60 extern const char *const mansearch_keynames[];
61
62 enum op {
63 OP_DEFAULT = 0, /* new dbs from dir list or default config */
64 OP_CONFFILE, /* new databases from custom config file */
65 OP_UPDATE, /* delete/add entries in existing database */
66 OP_DELETE, /* delete entries from existing database */
67 OP_TEST /* change no databases, report potential problems */
68 };
69
70 struct str {
71 const struct mpage *mpage; /* if set, the owning parse */
72 uint64_t mask; /* bitmask in sequence */
73 char key[]; /* rendered text */
74 };
75
76 struct inodev {
77 ino_t st_ino;
78 dev_t st_dev;
79 };
80
81 struct mpage {
82 struct inodev inodev; /* used for hashing routine */
83 struct dba_array *dba;
84 char *sec; /* section from file content */
85 char *arch; /* architecture from file content */
86 char *title; /* title from file content */
87 char *desc; /* description from file content */
88 struct mpage *next; /* singly linked list */
89 struct mlink *mlinks; /* singly linked list */
90 int name_head_done;
91 enum form form; /* format from file content */
92 };
93
94 struct mlink {
95 char file[PATH_MAX]; /* filename rel. to manpath */
96 char *dsec; /* section from directory */
97 char *arch; /* architecture from directory */
98 char *name; /* name from file name (not empty) */
99 char *fsec; /* section from file name suffix */
100 struct mlink *next; /* singly linked list */
101 struct mpage *mpage; /* parent */
102 int gzip; /* filename has a .gz suffix */
103 enum form dform; /* format from directory */
104 enum form fform; /* format from file name suffix */
105 };
106
107 typedef int (*mdoc_fp)(struct mpage *, const struct roff_meta *,
108 const struct roff_node *);
109
110 struct mdoc_handler {
111 mdoc_fp fp; /* optional handler */
112 uint64_t mask; /* set unless handler returns 0 */
113 int taboo; /* node flags that must not be set */
114 };
115
116
117 int mandocdb(int, char *[]);
118
119 static void dbadd(struct dba *, struct mpage *);
120 static void dbadd_mlink(const struct mlink *mlink);
121 static void dbprune(struct dba *);
122 static void dbwrite(struct dba *);
123 static void filescan(const char *);
124 #if HAVE_FTS_COMPARE_CONST
125 static int fts_compare(const FTSENT *const *, const FTSENT *const *);
126 #else
127 static int fts_compare(const FTSENT **, const FTSENT **);
128 #endif
129 static void mlink_add(struct mlink *, const struct stat *);
130 static void mlink_check(struct mpage *, struct mlink *);
131 static void mlink_free(struct mlink *);
132 static void mlinks_undupe(struct mpage *);
133 static void mpages_free(void);
134 static void mpages_merge(struct dba *, struct mparse *);
135 static void parse_cat(struct mpage *, int);
136 static void parse_man(struct mpage *, const struct roff_meta *,
137 const struct roff_node *);
138 static void parse_mdoc(struct mpage *, const struct roff_meta *,
139 const struct roff_node *);
140 static int parse_mdoc_head(struct mpage *, const struct roff_meta *,
141 const struct roff_node *);
142 static int parse_mdoc_Fa(struct mpage *, const struct roff_meta *,
143 const struct roff_node *);
144 static int parse_mdoc_Fd(struct mpage *, const struct roff_meta *,
145 const struct roff_node *);
146 static void parse_mdoc_fname(struct mpage *, const struct roff_node *);
147 static int parse_mdoc_Fn(struct mpage *, const struct roff_meta *,
148 const struct roff_node *);
149 static int parse_mdoc_Fo(struct mpage *, const struct roff_meta *,
150 const struct roff_node *);
151 static int parse_mdoc_Nd(struct mpage *, const struct roff_meta *,
152 const struct roff_node *);
153 static int parse_mdoc_Nm(struct mpage *, const struct roff_meta *,
154 const struct roff_node *);
155 static int parse_mdoc_Sh(struct mpage *, const struct roff_meta *,
156 const struct roff_node *);
157 static int parse_mdoc_Va(struct mpage *, const struct roff_meta *,
158 const struct roff_node *);
159 static int parse_mdoc_Xr(struct mpage *, const struct roff_meta *,
160 const struct roff_node *);
161 static void putkey(const struct mpage *, char *, uint64_t);
162 static void putkeys(const struct mpage *, char *, size_t, uint64_t);
163 static void putmdockey(const struct mpage *,
164 const struct roff_node *, uint64_t, int);
165 static int render_string(char **, size_t *);
166 static void say(const char *, const char *, ...)
167 __attribute__((__format__ (__printf__, 2, 3)));
168 static int set_basedir(const char *, int);
169 static int treescan(void);
170 static size_t utf8(unsigned int, char [7]);
171
172 static int nodb; /* no database changes */
173 static int mparse_options; /* abort the parse early */
174 static int use_all; /* use all found files */
175 static int debug; /* print what we're doing */
176 static int warnings; /* warn about crap */
177 static int write_utf8; /* write UTF-8 output; else ASCII */
178 static int exitcode; /* to be returned by main */
179 static enum op op; /* operational mode */
180 static char basedir[PATH_MAX]; /* current base directory */
181 static struct mpage *mpage_head; /* list of distinct manual pages */
182 static struct ohash mpages; /* table of distinct manual pages */
183 static struct ohash mlinks; /* table of directory entries */
184 static struct ohash names; /* table of all names */
185 static struct ohash strings; /* table of all strings */
186 static uint64_t name_mask;
187
188 static const struct mdoc_handler __mdocs[MDOC_MAX - MDOC_Dd] = {
189 { NULL, 0, NODE_NOPRT }, /* Dd */
190 { NULL, 0, NODE_NOPRT }, /* Dt */
191 { NULL, 0, NODE_NOPRT }, /* Os */
192 { parse_mdoc_Sh, TYPE_Sh, 0 }, /* Sh */
193 { parse_mdoc_head, TYPE_Ss, 0 }, /* Ss */
194 { NULL, 0, 0 }, /* Pp */
195 { NULL, 0, 0 }, /* D1 */
196 { NULL, 0, 0 }, /* Dl */
197 { NULL, 0, 0 }, /* Bd */
198 { NULL, 0, 0 }, /* Ed */
199 { NULL, 0, 0 }, /* Bl */
200 { NULL, 0, 0 }, /* El */
201 { NULL, 0, 0 }, /* It */
202 { NULL, 0, 0 }, /* Ad */
203 { NULL, TYPE_An, 0 }, /* An */
204 { NULL, 0, 0 }, /* Ap */
205 { NULL, TYPE_Ar, 0 }, /* Ar */
206 { NULL, TYPE_Cd, 0 }, /* Cd */
207 { NULL, TYPE_Cm, 0 }, /* Cm */
208 { NULL, TYPE_Dv, 0 }, /* Dv */
209 { NULL, TYPE_Er, 0 }, /* Er */
210 { NULL, TYPE_Ev, 0 }, /* Ev */
211 { NULL, 0, 0 }, /* Ex */
212 { parse_mdoc_Fa, 0, 0 }, /* Fa */
213 { parse_mdoc_Fd, 0, 0 }, /* Fd */
214 { NULL, TYPE_Fl, 0 }, /* Fl */
215 { parse_mdoc_Fn, 0, 0 }, /* Fn */
216 { NULL, TYPE_Ft | TYPE_Vt, 0 }, /* Ft */
217 { NULL, TYPE_Ic, 0 }, /* Ic */
218 { NULL, TYPE_In, 0 }, /* In */
219 { NULL, TYPE_Li, 0 }, /* Li */
220 { parse_mdoc_Nd, 0, 0 }, /* Nd */
221 { parse_mdoc_Nm, 0, 0 }, /* Nm */
222 { NULL, 0, 0 }, /* Op */
223 { NULL, 0, 0 }, /* Ot */
224 { NULL, TYPE_Pa, NODE_NOSRC }, /* Pa */
225 { NULL, 0, 0 }, /* Rv */
226 { NULL, TYPE_St, 0 }, /* St */
227 { parse_mdoc_Va, TYPE_Va, 0 }, /* Va */
228 { parse_mdoc_Va, TYPE_Vt, 0 }, /* Vt */
229 { parse_mdoc_Xr, 0, 0 }, /* Xr */
230 { NULL, 0, 0 }, /* %A */
231 { NULL, 0, 0 }, /* %B */
232 { NULL, 0, 0 }, /* %D */
233 { NULL, 0, 0 }, /* %I */
234 { NULL, 0, 0 }, /* %J */
235 { NULL, 0, 0 }, /* %N */
236 { NULL, 0, 0 }, /* %O */
237 { NULL, 0, 0 }, /* %P */
238 { NULL, 0, 0 }, /* %R */
239 { NULL, 0, 0 }, /* %T */
240 { NULL, 0, 0 }, /* %V */
241 { NULL, 0, 0 }, /* Ac */
242 { NULL, 0, 0 }, /* Ao */
243 { NULL, 0, 0 }, /* Aq */
244 { NULL, TYPE_At, 0 }, /* At */
245 { NULL, 0, 0 }, /* Bc */
246 { NULL, 0, 0 }, /* Bf */
247 { NULL, 0, 0 }, /* Bo */
248 { NULL, 0, 0 }, /* Bq */
249 { NULL, TYPE_Bsx, NODE_NOSRC }, /* Bsx */
250 { NULL, TYPE_Bx, NODE_NOSRC }, /* Bx */
251 { NULL, 0, 0 }, /* Db */
252 { NULL, 0, 0 }, /* Dc */
253 { NULL, 0, 0 }, /* Do */
254 { NULL, 0, 0 }, /* Dq */
255 { NULL, 0, 0 }, /* Ec */
256 { NULL, 0, 0 }, /* Ef */
257 { NULL, TYPE_Em, 0 }, /* Em */
258 { NULL, 0, 0 }, /* Eo */
259 { NULL, TYPE_Fx, NODE_NOSRC }, /* Fx */
260 { NULL, TYPE_Ms, 0 }, /* Ms */
261 { NULL, 0, 0 }, /* No */
262 { NULL, 0, 0 }, /* Ns */
263 { NULL, TYPE_Nx, NODE_NOSRC }, /* Nx */
264 { NULL, TYPE_Ox, NODE_NOSRC }, /* Ox */
265 { NULL, 0, 0 }, /* Pc */
266 { NULL, 0, 0 }, /* Pf */
267 { NULL, 0, 0 }, /* Po */
268 { NULL, 0, 0 }, /* Pq */
269 { NULL, 0, 0 }, /* Qc */
270 { NULL, 0, 0 }, /* Ql */
271 { NULL, 0, 0 }, /* Qo */
272 { NULL, 0, 0 }, /* Qq */
273 { NULL, 0, 0 }, /* Re */
274 { NULL, 0, 0 }, /* Rs */
275 { NULL, 0, 0 }, /* Sc */
276 { NULL, 0, 0 }, /* So */
277 { NULL, 0, 0 }, /* Sq */
278 { NULL, 0, 0 }, /* Sm */
279 { NULL, 0, 0 }, /* Sx */
280 { NULL, TYPE_Sy, 0 }, /* Sy */
281 { NULL, TYPE_Tn, 0 }, /* Tn */
282 { NULL, 0, NODE_NOSRC }, /* Ux */
283 { NULL, 0, 0 }, /* Xc */
284 { NULL, 0, 0 }, /* Xo */
285 { parse_mdoc_Fo, 0, 0 }, /* Fo */
286 { NULL, 0, 0 }, /* Fc */
287 { NULL, 0, 0 }, /* Oo */
288 { NULL, 0, 0 }, /* Oc */
289 { NULL, 0, 0 }, /* Bk */
290 { NULL, 0, 0 }, /* Ek */
291 { NULL, 0, 0 }, /* Bt */
292 { NULL, 0, 0 }, /* Hf */
293 { NULL, 0, 0 }, /* Fr */
294 { NULL, 0, 0 }, /* Ud */
295 { NULL, TYPE_Lb, NODE_NOSRC }, /* Lb */
296 { NULL, 0, 0 }, /* Lp */
297 { NULL, TYPE_Lk, 0 }, /* Lk */
298 { NULL, TYPE_Mt, NODE_NOSRC }, /* Mt */
299 { NULL, 0, 0 }, /* Brq */
300 { NULL, 0, 0 }, /* Bro */
301 { NULL, 0, 0 }, /* Brc */
302 { NULL, 0, 0 }, /* %C */
303 { NULL, 0, 0 }, /* Es */
304 { NULL, 0, 0 }, /* En */
305 { NULL, TYPE_Dx, NODE_NOSRC }, /* Dx */
306 { NULL, 0, 0 }, /* %Q */
307 { NULL, 0, 0 }, /* %U */
308 { NULL, 0, 0 }, /* Ta */
309 };
310 static const struct mdoc_handler *const mdocs = __mdocs - MDOC_Dd;
311
312
313 int
314 mandocdb(int argc, char *argv[])
315 {
316 struct manconf conf;
317 struct mparse *mp;
318 struct dba *dba;
319 const char *path_arg, *progname;
320 size_t j, sz;
321 int ch, i;
322
323 #if HAVE_PLEDGE
324 if (pledge("stdio rpath wpath cpath", NULL) == -1) {
325 warn("pledge");
326 return (int)MANDOCLEVEL_SYSERR;
327 }
328 #endif
329
330 #if HAVE_SANDBOX_INIT
331 if (sandbox_init(kSBXProfileNoInternet, SANDBOX_NAMED, NULL) == -1) {
332 warnx("sandbox_init");
333 return (int)MANDOCLEVEL_SYSERR;
334 }
335 #endif
336
337 memset(&conf, 0, sizeof(conf));
338
339 /*
340 * We accept a few different invocations.
341 * The CHECKOP macro makes sure that invocation styles don't
342 * clobber each other.
343 */
344 #define CHECKOP(_op, _ch) do \
345 if (OP_DEFAULT != (_op)) { \
346 warnx("-%c: Conflicting option", (_ch)); \
347 goto usage; \
348 } while (/*CONSTCOND*/0)
349
350 path_arg = NULL;
351 op = OP_DEFAULT;
352
353 while (-1 != (ch = getopt(argc, argv, "aC:Dd:npQT:tu:v")))
354 switch (ch) {
355 case 'a':
356 use_all = 1;
357 break;
358 case 'C':
359 CHECKOP(op, ch);
360 path_arg = optarg;
361 op = OP_CONFFILE;
362 break;
363 case 'D':
364 debug++;
365 break;
366 case 'd':
367 CHECKOP(op, ch);
368 path_arg = optarg;
369 op = OP_UPDATE;
370 break;
371 case 'n':
372 nodb = 1;
373 break;
374 case 'p':
375 warnings = 1;
376 break;
377 case 'Q':
378 mparse_options |= MPARSE_QUICK;
379 break;
380 case 'T':
381 if (strcmp(optarg, "utf8")) {
382 warnx("-T%s: Unsupported output format",
383 optarg);
384 goto usage;
385 }
386 write_utf8 = 1;
387 break;
388 case 't':
389 CHECKOP(op, ch);
390 dup2(STDOUT_FILENO, STDERR_FILENO);
391 op = OP_TEST;
392 nodb = warnings = 1;
393 break;
394 case 'u':
395 CHECKOP(op, ch);
396 path_arg = optarg;
397 op = OP_DELETE;
398 break;
399 case 'v':
400 /* Compatibility with espie@'s makewhatis. */
401 break;
402 default:
403 goto usage;
404 }
405
406 argc -= optind;
407 argv += optind;
408
409 #if HAVE_PLEDGE
410 if (nodb) {
411 if (pledge("stdio rpath", NULL) == -1) {
412 warn("pledge");
413 return (int)MANDOCLEVEL_SYSERR;
414 }
415 }
416 #endif
417
418 if (OP_CONFFILE == op && argc > 0) {
419 warnx("-C: Too many arguments");
420 goto usage;
421 }
422
423 exitcode = (int)MANDOCLEVEL_OK;
424 mchars_alloc();
425 mp = mparse_alloc(mparse_options, MANDOCERR_MAX, NULL,
426 MANDOC_OS_OTHER, NULL);
427 mandoc_ohash_init(&mpages, 6, offsetof(struct mpage, inodev));
428 mandoc_ohash_init(&mlinks, 6, offsetof(struct mlink, file));
429
430 if (OP_UPDATE == op || OP_DELETE == op || OP_TEST == op) {
431
432 /*
433 * Most of these deal with a specific directory.
434 * Jump into that directory first.
435 */
436 if (OP_TEST != op && 0 == set_basedir(path_arg, 1))
437 goto out;
438
439 dba = nodb ? dba_new(128) : dba_read(MANDOC_DB);
440 if (dba != NULL) {
441 /*
442 * The existing database is usable. Process
443 * all files specified on the command-line.
444 */
445 use_all = 1;
446 for (i = 0; i < argc; i++)
447 filescan(argv[i]);
448 if (nodb == 0)
449 dbprune(dba);
450 } else {
451 /* Database missing or corrupt. */
452 if (op != OP_UPDATE || errno != ENOENT)
453 say(MANDOC_DB, "%s: Automatically recreating"
454 " from scratch", strerror(errno));
455 exitcode = (int)MANDOCLEVEL_OK;
456 op = OP_DEFAULT;
457 if (0 == treescan())
458 goto out;
459 dba = dba_new(128);
460 }
461 if (OP_DELETE != op)
462 mpages_merge(dba, mp);
463 if (nodb == 0)
464 dbwrite(dba);
465 dba_free(dba);
466 } else {
467 /*
468 * If we have arguments, use them as our manpaths.
469 * If we don't, use man.conf(5).
470 */
471 if (argc > 0) {
472 conf.manpath.paths = mandoc_reallocarray(NULL,
473 argc, sizeof(char *));
474 conf.manpath.sz = (size_t)argc;
475 for (i = 0; i < argc; i++)
476 conf.manpath.paths[i] = mandoc_strdup(argv[i]);
477 } else
478 manconf_parse(&conf, path_arg, NULL, NULL);
479
480 if (conf.manpath.sz == 0) {
481 exitcode = (int)MANDOCLEVEL_BADARG;
482 say("", "Empty manpath");
483 }
484
485 /*
486 * First scan the tree rooted at a base directory, then
487 * build a new database and finally move it into place.
488 * Ignore zero-length directories and strip trailing
489 * slashes.
490 */
491 for (j = 0; j < conf.manpath.sz; j++) {
492 sz = strlen(conf.manpath.paths[j]);
493 if (sz && conf.manpath.paths[j][sz - 1] == '/')
494 conf.manpath.paths[j][--sz] = '\0';
495 if (0 == sz)
496 continue;
497
498 if (j) {
499 mandoc_ohash_init(&mpages, 6,
500 offsetof(struct mpage, inodev));
501 mandoc_ohash_init(&mlinks, 6,
502 offsetof(struct mlink, file));
503 }
504
505 if ( ! set_basedir(conf.manpath.paths[j], argc > 0))
506 continue;
507 if (0 == treescan())
508 continue;
509 dba = dba_new(128);
510 mpages_merge(dba, mp);
511 if (nodb == 0)
512 dbwrite(dba);
513 dba_free(dba);
514
515 if (j + 1 < conf.manpath.sz) {
516 mpages_free();
517 ohash_delete(&mpages);
518 ohash_delete(&mlinks);
519 }
520 }
521 }
522 out:
523 manconf_free(&conf);
524 mparse_free(mp);
525 mchars_free();
526 mpages_free();
527 ohash_delete(&mpages);
528 ohash_delete(&mlinks);
529 return exitcode;
530 usage:
531 progname = getprogname();
532 fprintf(stderr, "usage: %s [-aDnpQ] [-C file] [-Tutf8]\n"
533 " %s [-aDnpQ] [-Tutf8] dir ...\n"
534 " %s [-DnpQ] [-Tutf8] -d dir [file ...]\n"
535 " %s [-Dnp] -u dir [file ...]\n"
536 " %s [-Q] -t file ...\n",
537 progname, progname, progname, progname, progname);
538
539 return (int)MANDOCLEVEL_BADARG;
540 }
541
542 /*
543 * To get a singly linked list in alpha order while inserting entries
544 * at the beginning, process directory entries in reverse alpha order.
545 */
546 static int
547 #if HAVE_FTS_COMPARE_CONST
548 fts_compare(const FTSENT *const *a, const FTSENT *const *b)
549 #else
550 fts_compare(const FTSENT **a, const FTSENT **b)
551 #endif
552 {
553 return -strcmp((*a)->fts_name, (*b)->fts_name);
554 }
555
556 /*
557 * Scan a directory tree rooted at "basedir" for manpages.
558 * We use fts(), scanning directory parts along the way for clues to our
559 * section and architecture.
560 *
561 * If use_all has been specified, grok all files.
562 * If not, sanitise paths to the following:
563 *
564 * [./]man*[/<arch>]/<name>.<section>
565 * or
566 * [./]cat<section>[/<arch>]/<name>.0
567 *
568 * TODO: accommodate for multi-language directories.
569 */
570 static int
571 treescan(void)
572 {
573 char buf[PATH_MAX];
574 FTS *f;
575 FTSENT *ff;
576 struct mlink *mlink;
577 int gzip;
578 enum form dform;
579 char *dsec, *arch, *fsec, *cp;
580 const char *path;
581 const char *argv[2];
582
583 argv[0] = ".";
584 argv[1] = NULL;
585
586 f = fts_open((char * const *)argv, FTS_PHYSICAL | FTS_NOCHDIR,
587 fts_compare);
588 if (f == NULL) {
589 exitcode = (int)MANDOCLEVEL_SYSERR;
590 say("", "&fts_open");
591 return 0;
592 }
593
594 dsec = arch = NULL;
595 dform = FORM_NONE;
596
597 while ((ff = fts_read(f)) != NULL) {
598 path = ff->fts_path + 2;
599 switch (ff->fts_info) {
600
601 /*
602 * Symbolic links require various sanity checks,
603 * then get handled just like regular files.
604 */
605 case FTS_SL:
606 if (realpath(path, buf) == NULL) {
607 if (warnings)
608 say(path, "&realpath");
609 continue;
610 }
611 if (strstr(buf, basedir) != buf
612 #ifdef HOMEBREWDIR
613 && strstr(buf, HOMEBREWDIR) != buf
614 #endif
615 ) {
616 if (warnings) say("",
617 "%s: outside base directory", buf);
618 continue;
619 }
620 /* Use logical inode to avoid mpages dupe. */
621 if (stat(path, ff->fts_statp) == -1) {
622 if (warnings)
623 say(path, "&stat");
624 continue;
625 }
626 /* FALLTHROUGH */
627
628 /*
629 * If we're a regular file, add an mlink by using the
630 * stored directory data and handling the filename.
631 */
632 case FTS_F:
633 if ( ! strcmp(path, MANDOC_DB))
634 continue;
635 if ( ! use_all && ff->fts_level < 2) {
636 if (warnings)
637 say(path, "Extraneous file");
638 continue;
639 }
640 gzip = 0;
641 fsec = NULL;
642 while (fsec == NULL) {
643 fsec = strrchr(ff->fts_name, '.');
644 if (fsec == NULL || strcmp(fsec+1, "gz"))
645 break;
646 gzip = 1;
647 *fsec = '\0';
648 fsec = NULL;
649 }
650 if (fsec == NULL) {
651 if ( ! use_all) {
652 if (warnings)
653 say(path,
654 "No filename suffix");
655 continue;
656 }
657 } else if ( ! strcmp(++fsec, "html")) {
658 if (warnings)
659 say(path, "Skip html");
660 continue;
661 } else if ( ! strcmp(fsec, "ps")) {
662 if (warnings)
663 say(path, "Skip ps");
664 continue;
665 } else if ( ! strcmp(fsec, "pdf")) {
666 if (warnings)
667 say(path, "Skip pdf");
668 continue;
669 } else if ( ! use_all &&
670 ((dform == FORM_SRC &&
671 strncmp(fsec, dsec, strlen(dsec))) ||
672 (dform == FORM_CAT && strcmp(fsec, "0")))) {
673 if (warnings)
674 say(path, "Wrong filename suffix");
675 continue;
676 } else
677 fsec[-1] = '\0';
678
679 mlink = mandoc_calloc(1, sizeof(struct mlink));
680 if (strlcpy(mlink->file, path,
681 sizeof(mlink->file)) >=
682 sizeof(mlink->file)) {
683 say(path, "Filename too long");
684 free(mlink);
685 continue;
686 }
687 mlink->dform = dform;
688 mlink->dsec = dsec;
689 mlink->arch = arch;
690 mlink->name = ff->fts_name;
691 mlink->fsec = fsec;
692 mlink->gzip = gzip;
693 mlink_add(mlink, ff->fts_statp);
694 continue;
695
696 case FTS_D:
697 case FTS_DP:
698 break;
699
700 default:
701 if (warnings)
702 say(path, "Not a regular file");
703 continue;
704 }
705
706 switch (ff->fts_level) {
707 case 0:
708 /* Ignore the root directory. */
709 break;
710 case 1:
711 /*
712 * This might contain manX/ or catX/.
713 * Try to infer this from the name.
714 * If we're not in use_all, enforce it.
715 */
716 cp = ff->fts_name;
717 if (ff->fts_info == FTS_DP) {
718 dform = FORM_NONE;
719 dsec = NULL;
720 break;
721 }
722
723 if ( ! strncmp(cp, "man", 3)) {
724 dform = FORM_SRC;
725 dsec = cp + 3;
726 } else if ( ! strncmp(cp, "cat", 3)) {
727 dform = FORM_CAT;
728 dsec = cp + 3;
729 } else {
730 dform = FORM_NONE;
731 dsec = NULL;
732 }
733
734 if (dsec != NULL || use_all)
735 break;
736
737 if (warnings)
738 say(path, "Unknown directory part");
739 fts_set(f, ff, FTS_SKIP);
740 break;
741 case 2:
742 /*
743 * Possibly our architecture.
744 * If we're descending, keep tabs on it.
745 */
746 if (ff->fts_info != FTS_DP && dsec != NULL)
747 arch = ff->fts_name;
748 else
749 arch = NULL;
750 break;
751 default:
752 if (ff->fts_info == FTS_DP || use_all)
753 break;
754 if (warnings)
755 say(path, "Extraneous directory part");
756 fts_set(f, ff, FTS_SKIP);
757 break;
758 }
759 }
760
761 fts_close(f);
762 return 1;
763 }
764
765 /*
766 * Add a file to the mlinks table.
767 * Do not verify that it's a "valid" looking manpage (we'll do that
768 * later).
769 *
770 * Try to infer the manual section, architecture, and page name from the
771 * path, assuming it looks like
772 *
773 * [./]man*[/<arch>]/<name>.<section>
774 * or
775 * [./]cat<section>[/<arch>]/<name>.0
776 *
777 * See treescan() for the fts(3) version of this.
778 */
779 static void
780 filescan(const char *file)
781 {
782 char buf[PATH_MAX];
783 struct stat st;
784 struct mlink *mlink;
785 char *p, *start;
786
787 assert(use_all);
788
789 if (0 == strncmp(file, "./", 2))
790 file += 2;
791
792 /*
793 * We have to do lstat(2) before realpath(3) loses
794 * the information whether this is a symbolic link.
795 * We need to know that because for symbolic links,
796 * we want to use the orginal file name, while for
797 * regular files, we want to use the real path.
798 */
799 if (-1 == lstat(file, &st)) {
800 exitcode = (int)MANDOCLEVEL_BADARG;
801 say(file, "&lstat");
802 return;
803 } else if (0 == ((S_IFREG | S_IFLNK) & st.st_mode)) {
804 exitcode = (int)MANDOCLEVEL_BADARG;
805 say(file, "Not a regular file");
806 return;
807 }
808
809 /*
810 * We have to resolve the file name to the real path
811 * in any case for the base directory check.
812 */
813 if (NULL == realpath(file, buf)) {
814 exitcode = (int)MANDOCLEVEL_BADARG;
815 say(file, "&realpath");
816 return;
817 }
818
819 if (OP_TEST == op)
820 start = buf;
821 else if (strstr(buf, basedir) == buf)
822 start = buf + strlen(basedir);
823 #ifdef HOMEBREWDIR
824 else if (strstr(buf, HOMEBREWDIR) == buf)
825 start = buf;
826 #endif
827 else {
828 exitcode = (int)MANDOCLEVEL_BADARG;
829 say("", "%s: outside base directory", buf);
830 return;
831 }
832
833 /*
834 * Now we are sure the file is inside our tree.
835 * If it is a symbolic link, ignore the real path
836 * and use the original name.
837 * This implies passing stuff like "cat1/../man1/foo.1"
838 * on the command line won't work. So don't do that.
839 * Note the stat(2) can still fail if the link target
840 * doesn't exist.
841 */
842 if (S_IFLNK & st.st_mode) {
843 if (-1 == stat(buf, &st)) {
844 exitcode = (int)MANDOCLEVEL_BADARG;
845 say(file, "&stat");
846 return;
847 }
848 if (strlcpy(buf, file, sizeof(buf)) >= sizeof(buf)) {
849 say(file, "Filename too long");
850 return;
851 }
852 start = buf;
853 if (OP_TEST != op && strstr(buf, basedir) == buf)
854 start += strlen(basedir);
855 }
856
857 mlink = mandoc_calloc(1, sizeof(struct mlink));
858 mlink->dform = FORM_NONE;
859 if (strlcpy(mlink->file, start, sizeof(mlink->file)) >=
860 sizeof(mlink->file)) {
861 say(start, "Filename too long");
862 free(mlink);
863 return;
864 }
865
866 /*
867 * In test mode or when the original name is absolute
868 * but outside our tree, guess the base directory.
869 */
870
871 if (op == OP_TEST || (start == buf && *start == '/')) {
872 if (strncmp(buf, "man/", 4) == 0)
873 start = buf + 4;
874 else if ((start = strstr(buf, "/man/")) != NULL)
875 start += 5;
876 else
877 start = buf;
878 }
879
880 /*
881 * First try to guess our directory structure.
882 * If we find a separator, try to look for man* or cat*.
883 * If we find one of these and what's underneath is a directory,
884 * assume it's an architecture.
885 */
886 if (NULL != (p = strchr(start, '/'))) {
887 *p++ = '\0';
888 if (0 == strncmp(start, "man", 3)) {
889 mlink->dform = FORM_SRC;
890 mlink->dsec = start + 3;
891 } else if (0 == strncmp(start, "cat", 3)) {
892 mlink->dform = FORM_CAT;
893 mlink->dsec = start + 3;
894 }
895
896 start = p;
897 if (NULL != mlink->dsec && NULL != (p = strchr(start, '/'))) {
898 *p++ = '\0';
899 mlink->arch = start;
900 start = p;
901 }
902 }
903
904 /*
905 * Now check the file suffix.
906 * Suffix of `.0' indicates a catpage, `.1-9' is a manpage.
907 */
908 p = strrchr(start, '\0');
909 while (p-- > start && '/' != *p && '.' != *p)
910 /* Loop. */ ;
911
912 if ('.' == *p) {
913 *p++ = '\0';
914 mlink->fsec = p;
915 }
916
917 /*
918 * Now try to parse the name.
919 * Use the filename portion of the path.
920 */
921 mlink->name = start;
922 if (NULL != (p = strrchr(start, '/'))) {
923 mlink->name = p + 1;
924 *p = '\0';
925 }
926 mlink_add(mlink, &st);
927 }
928
929 static void
930 mlink_add(struct mlink *mlink, const struct stat *st)
931 {
932 struct inodev inodev;
933 struct mpage *mpage;
934 unsigned int slot;
935
936 assert(NULL != mlink->file);
937
938 mlink->dsec = mandoc_strdup(mlink->dsec ? mlink->dsec : "");
939 mlink->arch = mandoc_strdup(mlink->arch ? mlink->arch : "");
940 mlink->name = mandoc_strdup(mlink->name ? mlink->name : "");
941 mlink->fsec = mandoc_strdup(mlink->fsec ? mlink->fsec : "");
942
943 if ('0' == *mlink->fsec) {
944 free(mlink->fsec);
945 mlink->fsec = mandoc_strdup(mlink->dsec);
946 mlink->fform = FORM_CAT;
947 } else if ('1' <= *mlink->fsec && '9' >= *mlink->fsec)
948 mlink->fform = FORM_SRC;
949 else
950 mlink->fform = FORM_NONE;
951
952 slot = ohash_qlookup(&mlinks, mlink->file);
953 assert(NULL == ohash_find(&mlinks, slot));
954 ohash_insert(&mlinks, slot, mlink);
955
956 memset(&inodev, 0, sizeof(inodev)); /* Clear padding. */
957 inodev.st_ino = st->st_ino;
958 inodev.st_dev = st->st_dev;
959 slot = ohash_lookup_memory(&mpages, (char *)&inodev,
960 sizeof(struct inodev), inodev.st_ino);
961 mpage = ohash_find(&mpages, slot);
962 if (NULL == mpage) {
963 mpage = mandoc_calloc(1, sizeof(struct mpage));
964 mpage->inodev.st_ino = inodev.st_ino;
965 mpage->inodev.st_dev = inodev.st_dev;
966 mpage->form = FORM_NONE;
967 mpage->next = mpage_head;
968 mpage_head = mpage;
969 ohash_insert(&mpages, slot, mpage);
970 } else
971 mlink->next = mpage->mlinks;
972 mpage->mlinks = mlink;
973 mlink->mpage = mpage;
974 }
975
976 static void
977 mlink_free(struct mlink *mlink)
978 {
979
980 free(mlink->dsec);
981 free(mlink->arch);
982 free(mlink->name);
983 free(mlink->fsec);
984 free(mlink);
985 }
986
987 static void
988 mpages_free(void)
989 {
990 struct mpage *mpage;
991 struct mlink *mlink;
992
993 while ((mpage = mpage_head) != NULL) {
994 while ((mlink = mpage->mlinks) != NULL) {
995 mpage->mlinks = mlink->next;
996 mlink_free(mlink);
997 }
998 mpage_head = mpage->next;
999 free(mpage->sec);
1000 free(mpage->arch);
1001 free(mpage->title);
1002 free(mpage->desc);
1003 free(mpage);
1004 }
1005 }
1006
1007 /*
1008 * For each mlink to the mpage, check whether the path looks like
1009 * it is formatted, and if it does, check whether a source manual
1010 * exists by the same name, ignoring the suffix.
1011 * If both conditions hold, drop the mlink.
1012 */
1013 static void
1014 mlinks_undupe(struct mpage *mpage)
1015 {
1016 char buf[PATH_MAX];
1017 struct mlink **prev;
1018 struct mlink *mlink;
1019 char *bufp;
1020
1021 mpage->form = FORM_CAT;
1022 prev = &mpage->mlinks;
1023 while (NULL != (mlink = *prev)) {
1024 if (FORM_CAT != mlink->dform) {
1025 mpage->form = FORM_NONE;
1026 goto nextlink;
1027 }
1028 (void)strlcpy(buf, mlink->file, sizeof(buf));
1029 bufp = strstr(buf, "cat");
1030 assert(NULL != bufp);
1031 memcpy(bufp, "man", 3);
1032 if (NULL != (bufp = strrchr(buf, '.')))
1033 *++bufp = '\0';
1034 (void)strlcat(buf, mlink->dsec, sizeof(buf));
1035 if (NULL == ohash_find(&mlinks,
1036 ohash_qlookup(&mlinks, buf)))
1037 goto nextlink;
1038 if (warnings)
1039 say(mlink->file, "Man source exists: %s", buf);
1040 if (use_all)
1041 goto nextlink;
1042 *prev = mlink->next;
1043 mlink_free(mlink);
1044 continue;
1045 nextlink:
1046 prev = &(*prev)->next;
1047 }
1048 }
1049
1050 static void
1051 mlink_check(struct mpage *mpage, struct mlink *mlink)
1052 {
1053 struct str *str;
1054 unsigned int slot;
1055
1056 /*
1057 * Check whether the manual section given in a file
1058 * agrees with the directory where the file is located.
1059 * Some manuals have suffixes like (3p) on their
1060 * section number either inside the file or in the
1061 * directory name, some are linked into more than one
1062 * section, like encrypt(1) = makekey(8).
1063 */
1064
1065 if (FORM_SRC == mpage->form &&
1066 strcasecmp(mpage->sec, mlink->dsec))
1067 say(mlink->file, "Section \"%s\" manual in %s directory",
1068 mpage->sec, mlink->dsec);
1069
1070 /*
1071 * Manual page directories exist for each kernel
1072 * architecture as returned by machine(1).
1073 * However, many manuals only depend on the
1074 * application architecture as returned by arch(1).
1075 * For example, some (2/ARM) manuals are shared
1076 * across the "armish" and "zaurus" kernel
1077 * architectures.
1078 * A few manuals are even shared across completely
1079 * different architectures, for example fdformat(1)
1080 * on amd64, i386, and sparc64.
1081 */
1082
1083 if (strcasecmp(mpage->arch, mlink->arch))
1084 say(mlink->file, "Architecture \"%s\" manual in "
1085 "\"%s\" directory", mpage->arch, mlink->arch);
1086
1087 /*
1088 * XXX
1089 * parse_cat() doesn't set NAME_TITLE yet.
1090 */
1091
1092 if (FORM_CAT == mpage->form)
1093 return;
1094
1095 /*
1096 * Check whether this mlink
1097 * appears as a name in the NAME section.
1098 */
1099
1100 slot = ohash_qlookup(&names, mlink->name);
1101 str = ohash_find(&names, slot);
1102 assert(NULL != str);
1103 if ( ! (NAME_TITLE & str->mask))
1104 say(mlink->file, "Name missing in NAME section");
1105 }
1106
1107 /*
1108 * Run through the files in the global vector "mpages"
1109 * and add them to the database specified in "basedir".
1110 *
1111 * This handles the parsing scheme itself, using the cues of directory
1112 * and filename to determine whether the file is parsable or not.
1113 */
1114 static void
1115 mpages_merge(struct dba *dba, struct mparse *mp)
1116 {
1117 struct mpage *mpage, *mpage_dest;
1118 struct mlink *mlink, *mlink_dest;
1119 struct roff_man *man;
1120 char *sodest;
1121 char *cp;
1122 int fd;
1123
1124 for (mpage = mpage_head; mpage != NULL; mpage = mpage->next) {
1125 mlinks_undupe(mpage);
1126 if ((mlink = mpage->mlinks) == NULL)
1127 continue;
1128
1129 name_mask = NAME_MASK;
1130 mandoc_ohash_init(&names, 4, offsetof(struct str, key));
1131 mandoc_ohash_init(&strings, 6, offsetof(struct str, key));
1132 mparse_reset(mp);
1133 man = NULL;
1134 sodest = NULL;
1135
1136 if ((fd = mparse_open(mp, mlink->file)) == -1) {
1137 say(mlink->file, "&open");
1138 goto nextpage;
1139 }
1140
1141 /*
1142 * Interpret the file as mdoc(7) or man(7) source
1143 * code, unless it is known to be formatted.
1144 */
1145 if (mlink->dform != FORM_CAT || mlink->fform != FORM_CAT) {
1146 mparse_readfd(mp, fd, mlink->file);
1147 close(fd);
1148 fd = -1;
1149 mparse_result(mp, &man, &sodest);
1150 }
1151
1152 if (sodest != NULL) {
1153 mlink_dest = ohash_find(&mlinks,
1154 ohash_qlookup(&mlinks, sodest));
1155 if (mlink_dest == NULL) {
1156 mandoc_asprintf(&cp, "%s.gz", sodest);
1157 mlink_dest = ohash_find(&mlinks,
1158 ohash_qlookup(&mlinks, cp));
1159 free(cp);
1160 }
1161 if (mlink_dest != NULL) {
1162
1163 /* The .so target exists. */
1164
1165 mpage_dest = mlink_dest->mpage;
1166 while (1) {
1167 mlink->mpage = mpage_dest;
1168
1169 /*
1170 * If the target was already
1171 * processed, add the links
1172 * to the database now.
1173 * Otherwise, this will
1174 * happen when we come
1175 * to the target.
1176 */
1177
1178 if (mpage_dest->dba != NULL)
1179 dbadd_mlink(mlink);
1180
1181 if (mlink->next == NULL)
1182 break;
1183 mlink = mlink->next;
1184 }
1185
1186 /* Move all links to the target. */
1187
1188 mlink->next = mlink_dest->next;
1189 mlink_dest->next = mpage->mlinks;
1190 mpage->mlinks = NULL;
1191 }
1192 goto nextpage;
1193 } else if (man != NULL && man->macroset == MACROSET_MDOC) {
1194 mdoc_validate(man);
1195 mpage->form = FORM_SRC;
1196 mpage->sec = man->meta.msec;
1197 mpage->sec = mandoc_strdup(
1198 mpage->sec == NULL ? "" : mpage->sec);
1199 mpage->arch = man->meta.arch;
1200 mpage->arch = mandoc_strdup(
1201 mpage->arch == NULL ? "" : mpage->arch);
1202 mpage->title = mandoc_strdup(man->meta.title);
1203 } else if (man != NULL && man->macroset == MACROSET_MAN) {
1204 man_validate(man);
1205 if (*man->meta.msec != '\0' ||
1206 *man->meta.title != '\0') {
1207 mpage->form = FORM_SRC;
1208 mpage->sec = mandoc_strdup(man->meta.msec);
1209 mpage->arch = mandoc_strdup(mlink->arch);
1210 mpage->title = mandoc_strdup(man->meta.title);
1211 } else
1212 man = NULL;
1213 }
1214
1215 assert(mpage->desc == NULL);
1216 if (man == NULL) {
1217 mpage->form = FORM_CAT;
1218 mpage->sec = mandoc_strdup(mlink->dsec);
1219 mpage->arch = mandoc_strdup(mlink->arch);
1220 mpage->title = mandoc_strdup(mlink->name);
1221 parse_cat(mpage, fd);
1222 } else if (man->macroset == MACROSET_MDOC)
1223 parse_mdoc(mpage, &man->meta, man->first);
1224 else
1225 parse_man(mpage, &man->meta, man->first);
1226 if (mpage->desc == NULL) {
1227 mpage->desc = mandoc_strdup(mlink->name);
1228 if (warnings)
1229 say(mlink->file, "No one-line description, "
1230 "using filename \"%s\"", mlink->name);
1231 }
1232
1233 for (mlink = mpage->mlinks;
1234 mlink != NULL;
1235 mlink = mlink->next) {
1236 putkey(mpage, mlink->name, NAME_FILE);
1237 if (warnings && !use_all)
1238 mlink_check(mpage, mlink);
1239 }
1240
1241 dbadd(dba, mpage);
1242
1243 nextpage:
1244 ohash_delete(&strings);
1245 ohash_delete(&names);
1246 }
1247 }
1248
1249 static void
1250 parse_cat(struct mpage *mpage, int fd)
1251 {
1252 FILE *stream;
1253 struct mlink *mlink;
1254 char *line, *p, *title, *sec;
1255 size_t linesz, plen, titlesz;
1256 ssize_t len;
1257 int offs;
1258
1259 mlink = mpage->mlinks;
1260 stream = fd == -1 ? fopen(mlink->file, "r") : fdopen(fd, "r");
1261 if (stream == NULL) {
1262 if (fd != -1)
1263 close(fd);
1264 if (warnings)
1265 say(mlink->file, "&fopen");
1266 return;
1267 }
1268
1269 line = NULL;
1270 linesz = 0;
1271
1272 /* Parse the section number from the header line. */
1273
1274 while (getline(&line, &linesz, stream) != -1) {
1275 if (*line == '\n')
1276 continue;
1277 if ((sec = strchr(line, '(')) == NULL)
1278 break;
1279 if ((p = strchr(++sec, ')')) == NULL)
1280 break;
1281 free(mpage->sec);
1282 mpage->sec = mandoc_strndup(sec, p - sec);
1283 if (warnings && *mlink->dsec != '\0' &&
1284 strcasecmp(mpage->sec, mlink->dsec))
1285 say(mlink->file,
1286 "Section \"%s\" manual in %s directory",
1287 mpage->sec, mlink->dsec);
1288 break;
1289 }
1290
1291 /* Skip to first blank line. */
1292
1293 while (line == NULL || *line != '\n')
1294 if (getline(&line, &linesz, stream) == -1)
1295 break;
1296
1297 /*
1298 * Assume the first line that is not indented
1299 * is the first section header. Skip to it.
1300 */
1301
1302 while (getline(&line, &linesz, stream) != -1)
1303 if (*line != '\n' && *line != ' ')
1304 break;
1305
1306 /*
1307 * Read up until the next section into a buffer.
1308 * Strip the leading and trailing newline from each read line,
1309 * appending a trailing space.
1310 * Ignore empty (whitespace-only) lines.
1311 */
1312
1313 titlesz = 0;
1314 title = NULL;
1315
1316 while ((len = getline(&line, &linesz, stream)) != -1) {
1317 if (*line != ' ')
1318 break;
1319 offs = 0;
1320 while (isspace((unsigned char)line[offs]))
1321 offs++;
1322 if (line[offs] == '\0')
1323 continue;
1324 title = mandoc_realloc(title, titlesz + len - offs);
1325 memcpy(title + titlesz, line + offs, len - offs);
1326 titlesz += len - offs;
1327 title[titlesz - 1] = ' ';
1328 }
1329 free(line);
1330
1331 /*
1332 * If no page content can be found, or the input line
1333 * is already the next section header, or there is no
1334 * trailing newline, reuse the page title as the page
1335 * description.
1336 */
1337
1338 if (NULL == title || '\0' == *title) {
1339 if (warnings)
1340 say(mlink->file, "Cannot find NAME section");
1341 fclose(stream);
1342 free(title);
1343 return;
1344 }
1345
1346 title[titlesz - 1] = '\0';
1347
1348 /*
1349 * Skip to the first dash.
1350 * Use the remaining line as the description (no more than 70
1351 * bytes).
1352 */
1353
1354 if (NULL != (p = strstr(title, "- "))) {
1355 for (p += 2; ' ' == *p || '\b' == *p; p++)
1356 /* Skip to next word. */ ;
1357 } else {
1358 if (warnings)
1359 say(mlink->file, "No dash in title line, "
1360 "reusing \"%s\" as one-line description", title);
1361 p = title;
1362 }
1363
1364 plen = strlen(p);
1365
1366 /* Strip backspace-encoding from line. */
1367
1368 while (NULL != (line = memchr(p, '\b', plen))) {
1369 len = line - p;
1370 if (0 == len) {
1371 memmove(line, line + 1, plen--);
1372 continue;
1373 }
1374 memmove(line - 1, line + 1, plen - len);
1375 plen -= 2;
1376 }
1377
1378 /*
1379 * Cut off excessive one-line descriptions.
1380 * Bad pages are not worth better heuristics.
1381 */
1382
1383 mpage->desc = mandoc_strndup(p, 150);
1384 fclose(stream);
1385 free(title);
1386 }
1387
1388 /*
1389 * Put a type/word pair into the word database for this particular file.
1390 */
1391 static void
1392 putkey(const struct mpage *mpage, char *value, uint64_t type)
1393 {
1394 putkeys(mpage, value, strlen(value), type);
1395 }
1396
1397 /*
1398 * Grok all nodes at or below a certain mdoc node into putkey().
1399 */
1400 static void
1401 putmdockey(const struct mpage *mpage,
1402 const struct roff_node *n, uint64_t m, int taboo)
1403 {
1404
1405 for ( ; NULL != n; n = n->next) {
1406 if (n->flags & taboo)
1407 continue;
1408 if (NULL != n->child)
1409 putmdockey(mpage, n->child, m, taboo);
1410 if (n->type == ROFFT_TEXT)
1411 putkey(mpage, n->string, m);
1412 }
1413 }
1414
1415 static void
1416 parse_man(struct mpage *mpage, const struct roff_meta *meta,
1417 const struct roff_node *n)
1418 {
1419 const struct roff_node *head, *body;
1420 char *start, *title;
1421 char byte;
1422 size_t sz;
1423
1424 if (n == NULL)
1425 return;
1426
1427 /*
1428 * We're only searching for one thing: the first text child in
1429 * the BODY of a NAME section. Since we don't keep track of
1430 * sections in -man, run some hoops to find out whether we're in
1431 * the correct section or not.
1432 */
1433
1434 if (n->type == ROFFT_BODY && n->tok == MAN_SH) {
1435 body = n;
1436 if ((head = body->parent->head) != NULL &&
1437 (head = head->child) != NULL &&
1438 head->next == NULL &&
1439 head->type == ROFFT_TEXT &&
1440 strcmp(head->string, "NAME") == 0 &&
1441 body->child != NULL) {
1442
1443 /*
1444 * Suck the entire NAME section into memory.
1445 * Yes, we might run away.
1446 * But too many manuals have big, spread-out
1447 * NAME sections over many lines.
1448 */
1449
1450 title = NULL;
1451 deroff(&title, body);
1452 if (NULL == title)
1453 return;
1454
1455 /*
1456 * Go through a special heuristic dance here.
1457 * Conventionally, one or more manual names are
1458 * comma-specified prior to a whitespace, then a
1459 * dash, then a description. Try to puzzle out
1460 * the name parts here.
1461 */
1462
1463 start = title;
1464 for ( ;; ) {
1465 sz = strcspn(start, " ,");
1466 if ('\0' == start[sz])
1467 break;
1468
1469 byte = start[sz];
1470 start[sz] = '\0';
1471
1472 /*
1473 * Assume a stray trailing comma in the
1474 * name list if a name begins with a dash.
1475 */
1476
1477 if ('-' == start[0] ||
1478 ('\\' == start[0] && '-' == start[1]))
1479 break;
1480
1481 putkey(mpage, start, NAME_TITLE);
1482 if ( ! (mpage->name_head_done ||
1483 strcasecmp(start, meta->title))) {
1484 putkey(mpage, start, NAME_HEAD);
1485 mpage->name_head_done = 1;
1486 }
1487
1488 if (' ' == byte) {
1489 start += sz + 1;
1490 break;
1491 }
1492
1493 assert(',' == byte);
1494 start += sz + 1;
1495 while (' ' == *start)
1496 start++;
1497 }
1498
1499 if (start == title) {
1500 putkey(mpage, start, NAME_TITLE);
1501 if ( ! (mpage->name_head_done ||
1502 strcasecmp(start, meta->title))) {
1503 putkey(mpage, start, NAME_HEAD);
1504 mpage->name_head_done = 1;
1505 }
1506 free(title);
1507 return;
1508 }
1509
1510 while (isspace((unsigned char)*start))
1511 start++;
1512
1513 if (0 == strncmp(start, "-", 1))
1514 start += 1;
1515 else if (0 == strncmp(start, "\\-\\-", 4))
1516 start += 4;
1517 else if (0 == strncmp(start, "\\-", 2))
1518 start += 2;
1519 else if (0 == strncmp(start, "\\(en", 4))
1520 start += 4;
1521 else if (0 == strncmp(start, "\\(em", 4))
1522 start += 4;
1523
1524 while (' ' == *start)
1525 start++;
1526
1527 /*
1528 * Cut off excessive one-line descriptions.
1529 * Bad pages are not worth better heuristics.
1530 */
1531
1532 mpage->desc = mandoc_strndup(start, 150);
1533 free(title);
1534 return;
1535 }
1536 }
1537
1538 for (n = n->child; n; n = n->next) {
1539 if (NULL != mpage->desc)
1540 break;
1541 parse_man(mpage, meta, n);
1542 }
1543 }
1544
1545 static void
1546 parse_mdoc(struct mpage *mpage, const struct roff_meta *meta,
1547 const struct roff_node *n)
1548 {
1549
1550 for (n = n->child; n != NULL; n = n->next) {
1551 if (n->tok == TOKEN_NONE ||
1552 n->tok < ROFF_MAX ||
1553 n->flags & mdocs[n->tok].taboo)
1554 continue;
1555 assert(n->tok >= MDOC_Dd && n->tok < MDOC_MAX);
1556 switch (n->type) {
1557 case ROFFT_ELEM:
1558 case ROFFT_BLOCK:
1559 case ROFFT_HEAD:
1560 case ROFFT_BODY:
1561 case ROFFT_TAIL:
1562 if (mdocs[n->tok].fp != NULL &&
1563 (*mdocs[n->tok].fp)(mpage, meta, n) == 0)
1564 break;
1565 if (mdocs[n->tok].mask)
1566 putmdockey(mpage, n->child,
1567 mdocs[n->tok].mask, mdocs[n->tok].taboo);
1568 break;
1569 default:
1570 continue;
1571 }
1572 if (NULL != n->child)
1573 parse_mdoc(mpage, meta, n);
1574 }
1575 }
1576
1577 static int
1578 parse_mdoc_Fa(struct mpage *mpage, const struct roff_meta *meta,
1579 const struct roff_node *n)
1580 {
1581 uint64_t mask;
1582
1583 mask = TYPE_Fa;
1584 if (n->sec == SEC_SYNOPSIS)
1585 mask |= TYPE_Vt;
1586
1587 putmdockey(mpage, n->child, mask, 0);
1588 return 0;
1589 }
1590
1591 static int
1592 parse_mdoc_Fd(struct mpage *mpage, const struct roff_meta *meta,
1593 const struct roff_node *n)
1594 {
1595 char *start, *end;
1596 size_t sz;
1597
1598 if (SEC_SYNOPSIS != n->sec ||
1599 NULL == (n = n->child) ||
1600 n->type != ROFFT_TEXT)
1601 return 0;
1602
1603 /*
1604 * Only consider those `Fd' macro fields that begin with an
1605 * "inclusion" token (versus, e.g., #define).
1606 */
1607
1608 if (strcmp("#include", n->string))
1609 return 0;
1610
1611 if ((n = n->next) == NULL || n->type != ROFFT_TEXT)
1612 return 0;
1613
1614 /*
1615 * Strip away the enclosing angle brackets and make sure we're
1616 * not zero-length.
1617 */
1618
1619 start = n->string;
1620 if ('<' == *start || '"' == *start)
1621 start++;
1622
1623 if (0 == (sz = strlen(start)))
1624 return 0;
1625
1626 end = &start[(int)sz - 1];
1627 if ('>' == *end || '"' == *end)
1628 end--;
1629
1630 if (end > start)
1631 putkeys(mpage, start, end - start + 1, TYPE_In);
1632 return 0;
1633 }
1634
1635 static void
1636 parse_mdoc_fname(struct mpage *mpage, const struct roff_node *n)
1637 {
1638 char *cp;
1639 size_t sz;
1640
1641 if (n->type != ROFFT_TEXT)
1642 return;
1643
1644 /* Skip function pointer punctuation. */
1645
1646 cp = n->string;
1647 while (*cp == '(' || *cp == '*')
1648 cp++;
1649 sz = strcspn(cp, "()");
1650
1651 putkeys(mpage, cp, sz, TYPE_Fn);
1652 if (n->sec == SEC_SYNOPSIS)
1653 putkeys(mpage, cp, sz, NAME_SYN);
1654 }
1655
1656 static int
1657 parse_mdoc_Fn(struct mpage *mpage, const struct roff_meta *meta,
1658 const struct roff_node *n)
1659 {
1660 uint64_t mask;
1661
1662 if (n->child == NULL)
1663 return 0;
1664
1665 parse_mdoc_fname(mpage, n->child);
1666
1667 n = n->child->next;
1668 if (n != NULL && n->type == ROFFT_TEXT) {
1669 mask = TYPE_Fa;
1670 if (n->sec == SEC_SYNOPSIS)
1671 mask |= TYPE_Vt;
1672 putmdockey(mpage, n, mask, 0);
1673 }
1674
1675 return 0;
1676 }
1677
1678 static int
1679 parse_mdoc_Fo(struct mpage *mpage, const struct roff_meta *meta,
1680 const struct roff_node *n)
1681 {
1682
1683 if (n->type != ROFFT_HEAD)
1684 return 1;
1685
1686 if (n->child != NULL)
1687 parse_mdoc_fname(mpage, n->child);
1688
1689 return 0;
1690 }
1691
1692 static int
1693 parse_mdoc_Va(struct mpage *mpage, const struct roff_meta *meta,
1694 const struct roff_node *n)
1695 {
1696 char *cp;
1697
1698 if (n->type != ROFFT_ELEM && n->type != ROFFT_BODY)
1699 return 0;
1700
1701 if (n->child != NULL &&
1702 n->child->next == NULL &&
1703 n->child->type == ROFFT_TEXT)
1704 return 1;
1705
1706 cp = NULL;
1707 deroff(&cp, n);
1708 if (cp != NULL) {
1709 putkey(mpage, cp, TYPE_Vt | (n->tok == MDOC_Va ||
1710 n->type == ROFFT_BODY ? TYPE_Va : 0));
1711 free(cp);
1712 }
1713
1714 return 0;
1715 }
1716
1717 static int
1718 parse_mdoc_Xr(struct mpage *mpage, const struct roff_meta *meta,
1719 const struct roff_node *n)
1720 {
1721 char *cp;
1722
1723 if (NULL == (n = n->child))
1724 return 0;
1725
1726 if (NULL == n->next) {
1727 putkey(mpage, n->string, TYPE_Xr);
1728 return 0;
1729 }
1730
1731 mandoc_asprintf(&cp, "%s(%s)", n->string, n->next->string);
1732 putkey(mpage, cp, TYPE_Xr);
1733 free(cp);
1734 return 0;
1735 }
1736
1737 static int
1738 parse_mdoc_Nd(struct mpage *mpage, const struct roff_meta *meta,
1739 const struct roff_node *n)
1740 {
1741
1742 if (n->type == ROFFT_BODY)
1743 deroff(&mpage->desc, n);
1744 return 0;
1745 }
1746
1747 static int
1748 parse_mdoc_Nm(struct mpage *mpage, const struct roff_meta *meta,
1749 const struct roff_node *n)
1750 {
1751
1752 if (SEC_NAME == n->sec)
1753 putmdockey(mpage, n->child, NAME_TITLE, 0);
1754 else if (n->sec == SEC_SYNOPSIS && n->type == ROFFT_HEAD) {
1755 if (n->child == NULL)
1756 putkey(mpage, meta->name, NAME_SYN);
1757 else
1758 putmdockey(mpage, n->child, NAME_SYN, 0);
1759 }
1760 if ( ! (mpage->name_head_done ||
1761 n->child == NULL || n->child->string == NULL ||
1762 strcasecmp(n->child->string, meta->title))) {
1763 putkey(mpage, n->child->string, NAME_HEAD);
1764 mpage->name_head_done = 1;
1765 }
1766 return 0;
1767 }
1768
1769 static int
1770 parse_mdoc_Sh(struct mpage *mpage, const struct roff_meta *meta,
1771 const struct roff_node *n)
1772 {
1773
1774 return n->sec == SEC_CUSTOM && n->type == ROFFT_HEAD;
1775 }
1776
1777 static int
1778 parse_mdoc_head(struct mpage *mpage, const struct roff_meta *meta,
1779 const struct roff_node *n)
1780 {
1781
1782 return n->type == ROFFT_HEAD;
1783 }
1784
1785 /*
1786 * Add a string to the hash table for the current manual.
1787 * Each string has a bitmask telling which macros it belongs to.
1788 * When we finish the manual, we'll dump the table.
1789 */
1790 static void
1791 putkeys(const struct mpage *mpage, char *cp, size_t sz, uint64_t v)
1792 {
1793 struct ohash *htab;
1794 struct str *s;
1795 const char *end;
1796 unsigned int slot;
1797 int i, mustfree;
1798
1799 if (0 == sz)
1800 return;
1801
1802 mustfree = render_string(&cp, &sz);
1803
1804 if (TYPE_Nm & v) {
1805 htab = &names;
1806 v &= name_mask;
1807 if (v & NAME_FIRST)
1808 name_mask &= ~NAME_FIRST;
1809 if (debug > 1)
1810 say(mpage->mlinks->file,
1811 "Adding name %*s, bits=0x%llx", (int)sz, cp,
1812 (unsigned long long)v);
1813 } else {
1814 htab = &strings;
1815 if (debug > 1)
1816 for (i = 0; i < KEY_MAX; i++)
1817 if ((uint64_t)1 << i & v)
1818 say(mpage->mlinks->file,
1819 "Adding key %s=%*s",
1820 mansearch_keynames[i], (int)sz, cp);
1821 }
1822
1823 end = cp + sz;
1824 slot = ohash_qlookupi(htab, cp, &end);
1825 s = ohash_find(htab, slot);
1826
1827 if (NULL != s && mpage == s->mpage) {
1828 s->mask |= v;
1829 return;
1830 } else if (NULL == s) {
1831 s = mandoc_calloc(1, sizeof(struct str) + sz + 1);
1832 memcpy(s->key, cp, sz);
1833 ohash_insert(htab, slot, s);
1834 }
1835 s->mpage = mpage;
1836 s->mask = v;
1837
1838 if (mustfree)
1839 free(cp);
1840 }
1841
1842 /*
1843 * Take a Unicode codepoint and produce its UTF-8 encoding.
1844 * This isn't the best way to do this, but it works.
1845 * The magic numbers are from the UTF-8 packaging.
1846 * They're not as scary as they seem: read the UTF-8 spec for details.
1847 */
1848 static size_t
1849 utf8(unsigned int cp, char out[7])
1850 {
1851 size_t rc;
1852
1853 rc = 0;
1854 if (cp <= 0x0000007F) {
1855 rc = 1;
1856 out[0] = (char)cp;
1857 } else if (cp <= 0x000007FF) {
1858 rc = 2;
1859 out[0] = (cp >> 6 & 31) | 192;
1860 out[1] = (cp & 63) | 128;
1861 } else if (cp <= 0x0000FFFF) {
1862 rc = 3;
1863 out[0] = (cp >> 12 & 15) | 224;
1864 out[1] = (cp >> 6 & 63) | 128;
1865 out[2] = (cp & 63) | 128;
1866 } else if (cp <= 0x001FFFFF) {
1867 rc = 4;
1868 out[0] = (cp >> 18 & 7) | 240;
1869 out[1] = (cp >> 12 & 63) | 128;
1870 out[2] = (cp >> 6 & 63) | 128;
1871 out[3] = (cp & 63) | 128;
1872 } else if (cp <= 0x03FFFFFF) {
1873 rc = 5;
1874 out[0] = (cp >> 24 & 3) | 248;
1875 out[1] = (cp >> 18 & 63) | 128;
1876 out[2] = (cp >> 12 & 63) | 128;
1877 out[3] = (cp >> 6 & 63) | 128;
1878 out[4] = (cp & 63) | 128;
1879 } else if (cp <= 0x7FFFFFFF) {
1880 rc = 6;
1881 out[0] = (cp >> 30 & 1) | 252;
1882 out[1] = (cp >> 24 & 63) | 128;
1883 out[2] = (cp >> 18 & 63) | 128;
1884 out[3] = (cp >> 12 & 63) | 128;
1885 out[4] = (cp >> 6 & 63) | 128;
1886 out[5] = (cp & 63) | 128;
1887 } else
1888 return 0;
1889
1890 out[rc] = '\0';
1891 return rc;
1892 }
1893
1894 /*
1895 * If the string contains escape sequences,
1896 * replace it with an allocated rendering and return 1,
1897 * such that the caller can free it after use.
1898 * Otherwise, do nothing and return 0.
1899 */
1900 static int
1901 render_string(char **public, size_t *psz)
1902 {
1903 const char *src, *scp, *addcp, *seq;
1904 char *dst;
1905 size_t ssz, dsz, addsz;
1906 char utfbuf[7], res[6];
1907 int seqlen, unicode;
1908
1909 res[0] = '\\';
1910 res[1] = '\t';
1911 res[2] = ASCII_NBRSP;
1912 res[3] = ASCII_HYPH;
1913 res[4] = ASCII_BREAK;
1914 res[5] = '\0';
1915
1916 src = scp = *public;
1917 ssz = *psz;
1918 dst = NULL;
1919 dsz = 0;
1920
1921 while (scp < src + *psz) {
1922
1923 /* Leave normal characters unchanged. */
1924
1925 if (strchr(res, *scp) == NULL) {
1926 if (dst != NULL)
1927 dst[dsz++] = *scp;
1928 scp++;
1929 continue;
1930 }
1931
1932 /*
1933 * Found something that requires replacing,
1934 * make sure we have a destination buffer.
1935 */
1936
1937 if (dst == NULL) {
1938 dst = mandoc_malloc(ssz + 1);
1939 dsz = scp - src;
1940 memcpy(dst, src, dsz);
1941 }
1942
1943 /* Handle single-char special characters. */
1944
1945 switch (*scp) {
1946 case '\\':
1947 break;
1948 case '\t':
1949 case ASCII_NBRSP:
1950 dst[dsz++] = ' ';
1951 scp++;
1952 continue;
1953 case ASCII_HYPH:
1954 dst[dsz++] = '-';
1955 /* FALLTHROUGH */
1956 case ASCII_BREAK:
1957 scp++;
1958 continue;
1959 default:
1960 abort();
1961 }
1962
1963 /*
1964 * Found an escape sequence.
1965 * Read past the slash, then parse it.
1966 * Ignore everything except characters.
1967 */
1968
1969 scp++;
1970 if (mandoc_escape(&scp, &seq, &seqlen) != ESCAPE_SPECIAL)
1971 continue;
1972
1973 /*
1974 * Render the special character
1975 * as either UTF-8 or ASCII.
1976 */
1977
1978 if (write_utf8) {
1979 unicode = mchars_spec2cp(seq, seqlen);
1980 if (unicode <= 0)
1981 continue;
1982 addsz = utf8(unicode, utfbuf);
1983 if (addsz == 0)
1984 continue;
1985 addcp = utfbuf;
1986 } else {
1987 addcp = mchars_spec2str(seq, seqlen, &addsz);
1988 if (addcp == NULL)
1989 continue;
1990 if (*addcp == ASCII_NBRSP) {
1991 addcp = " ";
1992 addsz = 1;
1993 }
1994 }
1995
1996 /* Copy the rendered glyph into the stream. */
1997
1998 ssz += addsz;
1999 dst = mandoc_realloc(dst, ssz + 1);
2000 memcpy(dst + dsz, addcp, addsz);
2001 dsz += addsz;
2002 }
2003 if (dst != NULL) {
2004 *public = dst;
2005 *psz = dsz;
2006 }
2007
2008 /* Trim trailing whitespace and NUL-terminate. */
2009
2010 while (*psz > 0 && (*public)[*psz - 1] == ' ')
2011 --*psz;
2012 if (dst != NULL) {
2013 (*public)[*psz] = '\0';
2014 return 1;
2015 } else
2016 return 0;
2017 }
2018
2019 static void
2020 dbadd_mlink(const struct mlink *mlink)
2021 {
2022 dba_page_alias(mlink->mpage->dba, mlink->name, NAME_FILE);
2023 dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->dsec);
2024 dba_page_add(mlink->mpage->dba, DBP_SECT, mlink->fsec);
2025 dba_page_add(mlink->mpage->dba, DBP_ARCH, mlink->arch);
2026 dba_page_add(mlink->mpage->dba, DBP_FILE, mlink->file);
2027 }
2028
2029 /*
2030 * Flush the current page's terms (and their bits) into the database.
2031 * Also, handle escape sequences at the last possible moment.
2032 */
2033 static void
2034 dbadd(struct dba *dba, struct mpage *mpage)
2035 {
2036 struct mlink *mlink;
2037 struct str *key;
2038 char *cp;
2039 uint64_t mask;
2040 size_t i;
2041 unsigned int slot;
2042 int mustfree;
2043
2044 mlink = mpage->mlinks;
2045
2046 if (nodb) {
2047 for (key = ohash_first(&names, &slot); NULL != key;
2048 key = ohash_next(&names, &slot))
2049 free(key);
2050 for (key = ohash_first(&strings, &slot); NULL != key;
2051 key = ohash_next(&strings, &slot))
2052 free(key);
2053 if (0 == debug)
2054 return;
2055 while (NULL != mlink) {
2056 fputs(mlink->name, stdout);
2057 if (NULL == mlink->next ||
2058 strcmp(mlink->dsec, mlink->next->dsec) ||
2059 strcmp(mlink->fsec, mlink->next->fsec) ||
2060 strcmp(mlink->arch, mlink->next->arch)) {
2061 putchar('(');
2062 if ('\0' == *mlink->dsec)
2063 fputs(mlink->fsec, stdout);
2064 else
2065 fputs(mlink->dsec, stdout);
2066 if ('\0' != *mlink->arch)
2067 printf("/%s", mlink->arch);
2068 putchar(')');
2069 }
2070 mlink = mlink->next;
2071 if (NULL != mlink)
2072 fputs(", ", stdout);
2073 }
2074 printf(" - %s\n", mpage->desc);
2075 return;
2076 }
2077
2078 if (debug)
2079 say(mlink->file, "Adding to database");
2080
2081 cp = mpage->desc;
2082 i = strlen(cp);
2083 mustfree = render_string(&cp, &i);
2084 mpage->dba = dba_page_new(dba->pages,
2085 *mpage->arch == '\0' ? mlink->arch : mpage->arch,
2086 cp, mlink->file, mpage->form);
2087 if (mustfree)
2088 free(cp);
2089 dba_page_add(mpage->dba, DBP_SECT, mpage->sec);
2090
2091 while (mlink != NULL) {
2092 dbadd_mlink(mlink);
2093 mlink = mlink->next;
2094 }
2095
2096 for (key = ohash_first(&names, &slot); NULL != key;
2097 key = ohash_next(&names, &slot)) {
2098 assert(key->mpage == mpage);
2099 dba_page_alias(mpage->dba, key->key, key->mask);
2100 free(key);
2101 }
2102 for (key = ohash_first(&strings, &slot); NULL != key;
2103 key = ohash_next(&strings, &slot)) {
2104 assert(key->mpage == mpage);
2105 i = 0;
2106 for (mask = TYPE_Xr; mask <= TYPE_Lb; mask *= 2) {
2107 if (key->mask & mask)
2108 dba_macro_add(dba->macros, i,
2109 key->key, mpage->dba);
2110 i++;
2111 }
2112 free(key);
2113 }
2114 }
2115
2116 static void
2117 dbprune(struct dba *dba)
2118 {
2119 struct dba_array *page, *files;
2120 char *file;
2121
2122 dba_array_FOREACH(dba->pages, page) {
2123 files = dba_array_get(page, DBP_FILE);
2124 dba_array_FOREACH(files, file) {
2125 if (*file < ' ')
2126 file++;
2127 if (ohash_find(&mlinks, ohash_qlookup(&mlinks,
2128 file)) != NULL) {
2129 if (debug)
2130 say(file, "Deleting from database");
2131 dba_array_del(dba->pages);
2132 break;
2133 }
2134 }
2135 }
2136 }
2137
2138 /*
2139 * Write the database from memory to disk.
2140 */
2141 static void
2142 dbwrite(struct dba *dba)
2143 {
2144 struct stat sb1, sb2;
2145 char tfn[33], *cp1, *cp2;
2146 off_t i;
2147 int fd1, fd2;
2148
2149 /*
2150 * Do not write empty databases, and delete existing ones
2151 * when makewhatis -u causes them to become empty.
2152 */
2153
2154 dba_array_start(dba->pages);
2155 if (dba_array_next(dba->pages) == NULL) {
2156 if (unlink(MANDOC_DB) == -1 && errno != ENOENT)
2157 say(MANDOC_DB, "&unlink");
2158 return;
2159 }
2160
2161 /*
2162 * Build the database in a temporary file,
2163 * then atomically move it into place.
2164 */
2165
2166 if (dba_write(MANDOC_DB "~", dba) != -1) {
2167 if (rename(MANDOC_DB "~", MANDOC_DB) == -1) {
2168 exitcode = (int)MANDOCLEVEL_SYSERR;
2169 say(MANDOC_DB, "&rename");
2170 unlink(MANDOC_DB "~");
2171 }
2172 return;
2173 }
2174
2175 /*
2176 * We lack write permission and cannot replace the database
2177 * file, but let's at least check whether the data changed.
2178 */
2179
2180 (void)strlcpy(tfn, "/tmp/mandocdb.XXXXXXXX", sizeof(tfn));
2181 if (mkdtemp(tfn) == NULL) {
2182 exitcode = (int)MANDOCLEVEL_SYSERR;
2183 say("", "&%s", tfn);
2184 return;
2185 }
2186 cp1 = cp2 = MAP_FAILED;
2187 fd1 = fd2 = -1;
2188 (void)strlcat(tfn, "/" MANDOC_DB, sizeof(tfn));
2189 if (dba_write(tfn, dba) == -1) {
2190 say(tfn, "&dba_write");
2191 goto err;
2192 }
2193 if ((fd1 = open(MANDOC_DB, O_RDONLY, 0)) == -1) {
2194 say(MANDOC_DB, "&open");
2195 goto err;
2196 }
2197 if ((fd2 = open(tfn, O_RDONLY, 0)) == -1) {
2198 say(tfn, "&open");
2199 goto err;
2200 }
2201 if (fstat(fd1, &sb1) == -1) {
2202 say(MANDOC_DB, "&fstat");
2203 goto err;
2204 }
2205 if (fstat(fd2, &sb2) == -1) {
2206 say(tfn, "&fstat");
2207 goto err;
2208 }
2209 if (sb1.st_size != sb2.st_size)
2210 goto err;
2211 if ((cp1 = mmap(NULL, sb1.st_size, PROT_READ, MAP_PRIVATE,
2212 fd1, 0)) == MAP_FAILED) {
2213 say(MANDOC_DB, "&mmap");
2214 goto err;
2215 }
2216 if ((cp2 = mmap(NULL, sb2.st_size, PROT_READ, MAP_PRIVATE,
2217 fd2, 0)) == MAP_FAILED) {
2218 say(tfn, "&mmap");
2219 goto err;
2220 }
2221 for (i = 0; i < sb1.st_size; i++)
2222 if (cp1[i] != cp2[i])
2223 goto err;
2224 goto out;
2225
2226 err:
2227 exitcode = (int)MANDOCLEVEL_SYSERR;
2228 say(MANDOC_DB, "Data changed, but cannot replace database");
2229
2230 out:
2231 if (cp1 != MAP_FAILED)
2232 munmap(cp1, sb1.st_size);
2233 if (cp2 != MAP_FAILED)
2234 munmap(cp2, sb2.st_size);
2235 if (fd1 != -1)
2236 close(fd1);
2237 if (fd2 != -1)
2238 close(fd2);
2239 unlink(tfn);
2240 *strrchr(tfn, '/') = '\0';
2241 rmdir(tfn);
2242 }
2243
2244 static int
2245 set_basedir(const char *targetdir, int report_baddir)
2246 {
2247 static char startdir[PATH_MAX];
2248 static int getcwd_status; /* 1 = ok, 2 = failure */
2249 static int chdir_status; /* 1 = changed directory */
2250 char *cp;
2251
2252 /*
2253 * Remember the original working directory, if possible.
2254 * This will be needed if the second or a later directory
2255 * on the command line is given as a relative path.
2256 * Do not error out if the current directory is not
2257 * searchable: Maybe it won't be needed after all.
2258 */
2259 if (0 == getcwd_status) {
2260 if (NULL == getcwd(startdir, sizeof(startdir))) {
2261 getcwd_status = 2;
2262 (void)strlcpy(startdir, strerror(errno),
2263 sizeof(startdir));
2264 } else
2265 getcwd_status = 1;
2266 }
2267
2268 /*
2269 * We are leaving the old base directory.
2270 * Do not use it any longer, not even for messages.
2271 */
2272 *basedir = '\0';
2273
2274 /*
2275 * If and only if the directory was changed earlier and
2276 * the next directory to process is given as a relative path,
2277 * first go back, or bail out if that is impossible.
2278 */
2279 if (chdir_status && '/' != *targetdir) {
2280 if (2 == getcwd_status) {
2281 exitcode = (int)MANDOCLEVEL_SYSERR;
2282 say("", "getcwd: %s", startdir);
2283 return 0;
2284 }
2285 if (-1 == chdir(startdir)) {
2286 exitcode = (int)MANDOCLEVEL_SYSERR;
2287 say("", "&chdir %s", startdir);
2288 return 0;
2289 }
2290 }
2291
2292 /*
2293 * Always resolve basedir to the canonicalized absolute
2294 * pathname and append a trailing slash, such that
2295 * we can reliably check whether files are inside.
2296 */
2297 if (NULL == realpath(targetdir, basedir)) {
2298 if (report_baddir || errno != ENOENT) {
2299 exitcode = (int)MANDOCLEVEL_BADARG;
2300 say("", "&%s: realpath", targetdir);
2301 }
2302 return 0;
2303 } else if (-1 == chdir(basedir)) {
2304 if (report_baddir || errno != ENOENT) {
2305 exitcode = (int)MANDOCLEVEL_BADARG;
2306 say("", "&chdir");
2307 }
2308 return 0;
2309 }
2310 chdir_status = 1;
2311 cp = strchr(basedir, '\0');
2312 if ('/' != cp[-1]) {
2313 if (cp - basedir >= PATH_MAX - 1) {
2314 exitcode = (int)MANDOCLEVEL_SYSERR;
2315 say("", "Filename too long");
2316 return 0;
2317 }
2318 *cp++ = '/';
2319 *cp = '\0';
2320 }
2321 return 1;
2322 }
2323
2324 static void
2325 say(const char *file, const char *format, ...)
2326 {
2327 va_list ap;
2328 int use_errno;
2329
2330 if ('\0' != *basedir)
2331 fprintf(stderr, "%s", basedir);
2332 if ('\0' != *basedir && '\0' != *file)
2333 fputc('/', stderr);
2334 if ('\0' != *file)
2335 fprintf(stderr, "%s", file);
2336
2337 use_errno = 1;
2338 if (NULL != format) {
2339 switch (*format) {
2340 case '&':
2341 format++;
2342 break;
2343 case '\0':
2344 format = NULL;
2345 break;
2346 default:
2347 use_errno = 0;
2348 break;
2349 }
2350 }
2351 if (NULL != format) {
2352 if ('\0' != *basedir || '\0' != *file)
2353 fputs(": ", stderr);
2354 va_start(ap, format);
2355 vfprintf(stderr, format, ap);
2356 va_end(ap);
2357 }
2358 if (use_errno) {
2359 if ('\0' != *basedir || '\0' != *file || NULL != format)
2360 fputs(": ", stderr);
2361 perror(NULL);
2362 } else
2363 fputc('\n', stderr);
2364 }