xref: /netbsd-src/usr.bin/make/arch.c (revision 82d56013d7b633d116a93943de88e08335357a7c)
1 /*	$NetBSD: arch.c,v 1.199 2021/04/03 11:08:40 rillig Exp $	*/
2 
3 /*
4  * Copyright (c) 1988, 1989, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Adam de Boor.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 1989 by Berkeley Softworks
37  * All rights reserved.
38  *
39  * This code is derived from software contributed to Berkeley by
40  * Adam de Boor.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  */
70 
71 /*
72  * Manipulate libraries, archives and their members.
73  *
74  * The first time an archive is referenced, all of its members' headers are
75  * read and cached and the archive closed again.  All cached archives are kept
76  * on a list which is searched each time an archive member is referenced.
77  *
78  * The interface to this module is:
79  *
80  *	Arch_Init	Initialize this module.
81  *
82  *	Arch_End	Clean up this module.
83  *
84  *	Arch_ParseArchive
85  *			Parse an archive specification such as
86  *			"archive.a(member1 member2)".
87  *
88  *	Arch_Touch	Alter the modification time of the archive
89  *			member described by the given node to be
90  *			the time when make was started.
91  *
92  *	Arch_TouchLib	Update the modification time of the library
93  *			described by the given node. This is special
94  *			because it also updates the modification time
95  *			of the library's table of contents.
96  *
97  *	Arch_UpdateMTime
98  *			Find the modification time of a member of
99  *			an archive *in the archive* and place it in the
100  *			member's GNode.
101  *
102  *	Arch_UpdateMemberMTime
103  *			Find the modification time of a member of
104  *			an archive. Called when the member doesn't
105  *			already exist. Looks in the archive for the
106  *			modification time. Returns the modification
107  *			time.
108  *
109  *	Arch_FindLib	Search for a library along a path. The
110  *			library name in the GNode should be in
111  *			-l<name> format.
112  *
113  *	Arch_LibOODate	Decide if a library node is out-of-date.
114  */
115 
116 #include <sys/types.h>
117 #include <sys/stat.h>
118 #include <sys/time.h>
119 #include <sys/param.h>
120 
121 #include <ar.h>
122 #include <utime.h>
123 
124 #include "make.h"
125 #include "dir.h"
126 #include "config.h"
127 
128 /*	"@(#)arch.c	8.2 (Berkeley) 1/2/94"	*/
129 MAKE_RCSID("$NetBSD: arch.c,v 1.199 2021/04/03 11:08:40 rillig Exp $");
130 
131 typedef struct List ArchList;
132 typedef struct ListNode ArchListNode;
133 
134 static ArchList archives;	/* The archives we've already examined */
135 
136 typedef struct Arch {
137 	char *name;		/* Name of archive */
138 	HashTable members;	/* All the members of the archive described
139 				 * by <name, struct ar_hdr *> key/value pairs */
140 	char *fnametab;		/* Extended name table strings */
141 	size_t fnamesize;	/* Size of the string table */
142 } Arch;
143 
144 static FILE *ArchFindMember(const char *, const char *,
145 			    struct ar_hdr *, const char *);
146 #if defined(__svr4__) || defined(__SVR4) || defined(__ELF__)
147 #define SVR4ARCHIVES
148 static int ArchSVR4Entry(Arch *, char *, size_t, FILE *);
149 #endif
150 
151 #ifdef CLEANUP
152 static void
153 ArchFree(void *ap)
154 {
155 	Arch *a = ap;
156 	HashIter hi;
157 
158 	/* Free memory from hash entries */
159 	HashIter_Init(&hi, &a->members);
160 	while (HashIter_Next(&hi) != NULL)
161 		free(hi.entry->value);
162 
163 	free(a->name);
164 	free(a->fnametab);
165 	HashTable_Done(&a->members);
166 	free(a);
167 }
168 #endif
169 
170 
171 /*
172  * Parse an archive specification such as "archive.a(member1 member2.${EXT})",
173  * adding nodes for the expanded members to gns.  Nodes are created as
174  * necessary.
175  *
176  * Input:
177  *	pp		The start of the specification.
178  *	gns		The list on which to place the nodes.
179  *	scope		The scope in which to expand variables.
180  *
181  * Output:
182  *	return		True if it was a valid specification.
183  *	*pp		Points to the first non-space after the archive spec.
184  */
185 bool
186 Arch_ParseArchive(char **pp, GNodeList *gns, GNode *scope)
187 {
188 	char *cp;		/* Pointer into line */
189 	GNode *gn;		/* New node */
190 	MFStr libName;		/* Library-part of specification */
191 	char *memName;		/* Member-part of specification */
192 	char saveChar;		/* Ending delimiter of member-name */
193 	bool expandLibName;	/* Whether the parsed libName contains
194 				 * variable expressions that need to be
195 				 * expanded */
196 
197 	libName = MFStr_InitRefer(*pp);
198 	expandLibName = false;
199 
200 	for (cp = libName.str; *cp != '(' && *cp != '\0';) {
201 		if (*cp == '$') {
202 			/* Expand nested variable expressions. */
203 			/* XXX: This code can probably be shortened. */
204 			const char *nested_p = cp;
205 			FStr result;
206 			bool isError;
207 
208 			/* XXX: is expanded twice: once here and once below */
209 			(void)Var_Parse(&nested_p, scope,
210 			    VARE_UNDEFERR, &result);
211 			/* TODO: handle errors */
212 			isError = result.str == var_Error;
213 			FStr_Done(&result);
214 			if (isError)
215 				return false;
216 
217 			expandLibName = true;
218 			cp += nested_p - cp;
219 		} else
220 			cp++;
221 	}
222 
223 	*cp++ = '\0';
224 	if (expandLibName) {
225 		char *expanded;
226 		(void)Var_Subst(libName.str, scope, VARE_UNDEFERR, &expanded);
227 		/* TODO: handle errors */
228 		libName = MFStr_InitOwn(expanded);
229 	}
230 
231 
232 	for (;;) {
233 		/*
234 		 * First skip to the start of the member's name, mark that
235 		 * place and skip to the end of it (either white-space or
236 		 * a close paren).
237 		 */
238 		bool doSubst = false;
239 
240 		pp_skip_whitespace(&cp);
241 
242 		memName = cp;
243 		while (*cp != '\0' && *cp != ')' && !ch_isspace(*cp)) {
244 			if (*cp == '$') {
245 				/* Expand nested variable expressions. */
246 				/* XXX: This code can probably be shortened. */
247 				FStr result;
248 				bool isError;
249 				const char *nested_p = cp;
250 
251 				(void)Var_Parse(&nested_p, scope,
252 				    VARE_UNDEFERR, &result);
253 				/* TODO: handle errors */
254 				isError = result.str == var_Error;
255 				FStr_Done(&result);
256 
257 				if (isError)
258 					return false;
259 
260 				doSubst = true;
261 				cp += nested_p - cp;
262 			} else {
263 				cp++;
264 			}
265 		}
266 
267 		/*
268 		 * If the specification ends without a closing parenthesis,
269 		 * chances are there's something wrong (like a missing
270 		 * backslash), so it's better to return failure than allow
271 		 * such things to happen
272 		 */
273 		if (*cp == '\0') {
274 			Parse_Error(PARSE_FATAL,
275 				    "No closing parenthesis "
276 				    "in archive specification");
277 			return false;
278 		}
279 
280 		/*
281 		 * If we didn't move anywhere, we must be done
282 		 */
283 		if (cp == memName)
284 			break;
285 
286 		saveChar = *cp;
287 		*cp = '\0';
288 
289 		/*
290 		 * XXX: This should be taken care of intelligently by
291 		 * SuffExpandChildren, both for the archive and the member
292 		 * portions.
293 		 */
294 		/*
295 		 * If member contains variables, try and substitute for them.
296 		 * This will slow down archive specs with dynamic sources, of
297 		 * course, since we'll be (non-)substituting them three
298 		 * times, but them's the breaks -- we need to do this since
299 		 * SuffExpandChildren calls us, otherwise we could assume the
300 		 * thing would be taken care of later.
301 		 */
302 		if (doSubst) {
303 			char *fullName;
304 			char *p;
305 			char *unexpandedMemName = memName;
306 
307 			(void)Var_Subst(memName, scope, VARE_UNDEFERR,
308 			    &memName);
309 			/* TODO: handle errors */
310 
311 			/*
312 			 * Now form an archive spec and recurse to deal with
313 			 * nested variables and multi-word variable values.
314 			 */
315 			fullName = str_concat4(libName.str, "(", memName, ")");
316 			p = fullName;
317 
318 			if (strchr(memName, '$') != NULL &&
319 			    strcmp(memName, unexpandedMemName) == 0) {
320 				/*
321 				 * Must contain dynamic sources, so we can't
322 				 * deal with it now. Just create an ARCHV node
323 				 * for the thing and let SuffExpandChildren
324 				 * handle it.
325 				 */
326 				gn = Targ_GetNode(fullName);
327 				gn->type |= OP_ARCHV;
328 				Lst_Append(gns, gn);
329 
330 			} else if (!Arch_ParseArchive(&p, gns, scope)) {
331 				/* Error in nested call. */
332 				free(fullName);
333 				/* XXX: does unexpandedMemName leak? */
334 				return false;
335 			}
336 			free(fullName);
337 			/* XXX: does unexpandedMemName leak? */
338 
339 		} else if (Dir_HasWildcards(memName)) {
340 			StringList members = LST_INIT;
341 			SearchPath_Expand(&dirSearchPath, memName, &members);
342 
343 			while (!Lst_IsEmpty(&members)) {
344 				char *member = Lst_Dequeue(&members);
345 				char *fullname = str_concat4(libName.str, "(",
346 							     member, ")");
347 				free(member);
348 
349 				gn = Targ_GetNode(fullname);
350 				free(fullname);
351 
352 				gn->type |= OP_ARCHV;
353 				Lst_Append(gns, gn);
354 			}
355 			Lst_Done(&members);
356 
357 		} else {
358 			char *fullname = str_concat4(libName.str, "(", memName,
359 						     ")");
360 			gn = Targ_GetNode(fullname);
361 			free(fullname);
362 
363 			/*
364 			 * We've found the node, but have to make sure the
365 			 * rest of the world knows it's an archive member,
366 			 * without having to constantly check for parentheses,
367 			 * so we type the thing with the OP_ARCHV bit before
368 			 * we place it on the end of the provided list.
369 			 */
370 			gn->type |= OP_ARCHV;
371 			Lst_Append(gns, gn);
372 		}
373 		if (doSubst)
374 			free(memName);
375 
376 		*cp = saveChar;
377 	}
378 
379 	MFStr_Done(&libName);
380 
381 	cp++;			/* skip the ')' */
382 	/* We promised that pp would be set up at the next non-space. */
383 	pp_skip_whitespace(&cp);
384 	*pp = cp;
385 	return true;
386 }
387 
388 /*
389  * Locate a member of an archive, given the path of the archive and the path
390  * of the desired member.
391  *
392  * Input:
393  *	archive		Path to the archive
394  *	member		Name of member; only its basename is used.
395  *	addToCache	True if archive should be cached if not already so.
396  *
397  * Results:
398  *	The ar_hdr for the member, or NULL.
399  *
400  * See ArchFindMember for an almost identical copy of this code.
401  */
402 static struct ar_hdr *
403 ArchStatMember(const char *archive, const char *member, bool addToCache)
404 {
405 #define AR_MAX_NAME_LEN (sizeof arh.ar_name - 1)
406 	FILE *arch;
407 	size_t size;		/* Size of archive member */
408 	char magic[SARMAG];
409 	ArchListNode *ln;
410 	Arch *ar;		/* Archive descriptor */
411 	struct ar_hdr arh;	/* archive-member header for reading archive */
412 	char memName[MAXPATHLEN + 1];
413 	/* Current member name while hashing. */
414 
415 	/*
416 	 * Because of space constraints and similar things, files are archived
417 	 * using their basename, not the entire path.
418 	 */
419 	member = str_basename(member);
420 
421 	for (ln = archives.first; ln != NULL; ln = ln->next) {
422 		const Arch *a = ln->datum;
423 		if (strcmp(a->name, archive) == 0)
424 			break;
425 	}
426 
427 	if (ln != NULL) {
428 		struct ar_hdr *hdr;
429 
430 		ar = ln->datum;
431 		hdr = HashTable_FindValue(&ar->members, member);
432 		if (hdr != NULL)
433 			return hdr;
434 
435 		{
436 			/* Try truncated name */
437 			char copy[AR_MAX_NAME_LEN + 1];
438 			size_t len = strlen(member);
439 
440 			if (len > AR_MAX_NAME_LEN) {
441 				snprintf(copy, sizeof copy, "%s", member);
442 				hdr = HashTable_FindValue(&ar->members, copy);
443 			}
444 			return hdr;
445 		}
446 	}
447 
448 	if (!addToCache) {
449 		/*
450 		 * Caller doesn't want the thing cached, just use
451 		 * ArchFindMember to read the header for the member out and
452 		 * close down the stream again. Since the archive is not to be
453 		 * cached, we assume there's no need to allocate extra room
454 		 * for the header we're returning, so just declare it static.
455 		 */
456 		static struct ar_hdr sarh;
457 
458 		arch = ArchFindMember(archive, member, &sarh, "r");
459 		if (arch == NULL)
460 			return NULL;
461 
462 		fclose(arch);
463 		return &sarh;
464 	}
465 
466 	/*
467 	 * We don't have this archive on the list yet, so we want to find out
468 	 * everything that's in it and cache it so we can get at it quickly.
469 	 */
470 	arch = fopen(archive, "r");
471 	if (arch == NULL)
472 		return NULL;
473 
474 	/*
475 	 * We use the ARMAG string to make sure this is an archive we
476 	 * can handle...
477 	 */
478 	if (fread(magic, SARMAG, 1, arch) != 1 ||
479 	    strncmp(magic, ARMAG, SARMAG) != 0) {
480 		(void)fclose(arch);
481 		return NULL;
482 	}
483 
484 	ar = bmake_malloc(sizeof *ar);
485 	ar->name = bmake_strdup(archive);
486 	ar->fnametab = NULL;
487 	ar->fnamesize = 0;
488 	HashTable_Init(&ar->members);
489 	memName[AR_MAX_NAME_LEN] = '\0';
490 
491 	while (fread(&arh, sizeof arh, 1, arch) == 1) {
492 		char *nameend;
493 
494 		/* If the header is bogus, there's no way we can recover. */
495 		if (strncmp(arh.ar_fmag, ARFMAG, sizeof arh.ar_fmag) != 0)
496 			goto badarch;
497 
498 		/*
499 		 * We need to advance the stream's pointer to the start of the
500 		 * next header. Files are padded with newlines to an even-byte
501 		 * boundary, so we need to extract the size of the file from
502 		 * the 'size' field of the header and round it up during the
503 		 * seek.
504 		 */
505 		arh.ar_size[sizeof arh.ar_size - 1] = '\0';
506 		size = (size_t)strtol(arh.ar_size, NULL, 10);
507 
508 		memcpy(memName, arh.ar_name, sizeof arh.ar_name);
509 		nameend = memName + AR_MAX_NAME_LEN;
510 		while (nameend > memName && *nameend == ' ')
511 			nameend--;
512 		nameend[1] = '\0';
513 
514 #ifdef SVR4ARCHIVES
515 		/*
516 		 * svr4 names are slash-terminated.
517 		 * Also svr4 extended the AR format.
518 		 */
519 		if (memName[0] == '/') {
520 			/* svr4 magic mode; handle it */
521 			switch (ArchSVR4Entry(ar, memName, size, arch)) {
522 			case -1:	/* Invalid data */
523 				goto badarch;
524 			case 0:		/* List of files entry */
525 				continue;
526 			default:	/* Got the entry */
527 				break;
528 			}
529 		} else {
530 			if (nameend[0] == '/')
531 				nameend[0] = '\0';
532 		}
533 #endif
534 
535 #ifdef AR_EFMT1
536 		/*
537 		 * BSD 4.4 extended AR format: #1/<namelen>, with name as the
538 		 * first <namelen> bytes of the file
539 		 */
540 		if (strncmp(memName, AR_EFMT1, sizeof AR_EFMT1 - 1) == 0 &&
541 		    ch_isdigit(memName[sizeof AR_EFMT1 - 1])) {
542 
543 			int elen = atoi(memName + sizeof AR_EFMT1 - 1);
544 
545 			if ((unsigned int)elen > MAXPATHLEN)
546 				goto badarch;
547 			if (fread(memName, (size_t)elen, 1, arch) != 1)
548 				goto badarch;
549 			memName[elen] = '\0';
550 			if (fseek(arch, -elen, SEEK_CUR) != 0)
551 				goto badarch;
552 			if (DEBUG(ARCH) || DEBUG(MAKE))
553 				debug_printf(
554 				    "ArchStatMember: "
555 				    "Extended format entry for %s\n",
556 				    memName);
557 		}
558 #endif
559 
560 		{
561 			struct ar_hdr *cached_hdr = bmake_malloc(
562 			    sizeof *cached_hdr);
563 			memcpy(cached_hdr, &arh, sizeof arh);
564 			HashTable_Set(&ar->members, memName, cached_hdr);
565 		}
566 
567 		if (fseek(arch, ((long)size + 1) & ~1, SEEK_CUR) != 0)
568 			goto badarch;
569 	}
570 
571 	fclose(arch);
572 
573 	Lst_Append(&archives, ar);
574 
575 	/*
576 	 * Now that the archive has been read and cached, we can look into
577 	 * the addToCache table to find the desired member's header.
578 	 */
579 	return HashTable_FindValue(&ar->members, member);
580 
581 badarch:
582 	fclose(arch);
583 	HashTable_Done(&ar->members);
584 	free(ar->fnametab);
585 	free(ar);
586 	return NULL;
587 }
588 
589 #ifdef SVR4ARCHIVES
590 /*
591  * Parse an SVR4 style entry that begins with a slash.
592  * If it is "//", then load the table of filenames.
593  * If it is "/<offset>", then try to substitute the long file name
594  * from offset of a table previously read.
595  * If a table is read, the file pointer is moved to the next archive member.
596  *
597  * Results:
598  *	-1: Bad data in archive
599  *	 0: A table was loaded from the file
600  *	 1: Name was successfully substituted from table
601  *	 2: Name was not successfully substituted from table
602  */
603 static int
604 ArchSVR4Entry(Arch *ar, char *inout_name, size_t size, FILE *arch)
605 {
606 #define ARLONGNAMES1 "//"
607 #define ARLONGNAMES2 "/ARFILENAMES"
608 	size_t entry;
609 	char *ptr, *eptr;
610 
611 	if (strncmp(inout_name, ARLONGNAMES1, sizeof ARLONGNAMES1 - 1) == 0 ||
612 	    strncmp(inout_name, ARLONGNAMES2, sizeof ARLONGNAMES2 - 1) == 0) {
613 
614 		if (ar->fnametab != NULL) {
615 			DEBUG0(ARCH,
616 			       "Attempted to redefine an SVR4 name table\n");
617 			return -1;
618 		}
619 
620 		/*
621 		 * This is a table of archive names, so we build one for
622 		 * ourselves
623 		 */
624 		ar->fnametab = bmake_malloc(size);
625 		ar->fnamesize = size;
626 
627 		if (fread(ar->fnametab, size, 1, arch) != 1) {
628 			DEBUG0(ARCH, "Reading an SVR4 name table failed\n");
629 			return -1;
630 		}
631 		eptr = ar->fnametab + size;
632 		for (entry = 0, ptr = ar->fnametab; ptr < eptr; ptr++)
633 			if (*ptr == '/') {
634 				entry++;
635 				*ptr = '\0';
636 			}
637 		DEBUG1(ARCH, "Found svr4 archive name table with %lu entries\n",
638 		       (unsigned long)entry);
639 		return 0;
640 	}
641 
642 	if (inout_name[1] == ' ' || inout_name[1] == '\0')
643 		return 2;
644 
645 	entry = (size_t)strtol(&inout_name[1], &eptr, 0);
646 	if ((*eptr != ' ' && *eptr != '\0') || eptr == &inout_name[1]) {
647 		DEBUG1(ARCH, "Could not parse SVR4 name %s\n", inout_name);
648 		return 2;
649 	}
650 	if (entry >= ar->fnamesize) {
651 		DEBUG2(ARCH, "SVR4 entry offset %s is greater than %lu\n",
652 		       inout_name, (unsigned long)ar->fnamesize);
653 		return 2;
654 	}
655 
656 	DEBUG2(ARCH, "Replaced %s with %s\n", inout_name, &ar->fnametab[entry]);
657 
658 	snprintf(inout_name, MAXPATHLEN + 1, "%s", &ar->fnametab[entry]);
659 	return 1;
660 }
661 #endif
662 
663 
664 static bool
665 ArchiveMember_HasName(const struct ar_hdr *hdr,
666 		      const char *name, size_t namelen)
667 {
668 	const size_t ar_name_len = sizeof hdr->ar_name;
669 	const char *ar_name = hdr->ar_name;
670 
671 	if (strncmp(ar_name, name, namelen) != 0)
672 		return false;
673 
674 	if (namelen >= ar_name_len)
675 		return namelen == ar_name_len;
676 
677 	/* hdr->ar_name is space-padded to the right. */
678 	if (ar_name[namelen] == ' ')
679 		return true;
680 
681 	/* In archives created by GNU binutils 2.27, the member names end with
682 	 * a slash. */
683 	if (ar_name[namelen] == '/' &&
684 	    (namelen == ar_name_len || ar_name[namelen + 1] == ' '))
685 		return true;
686 
687 	return false;
688 }
689 
690 /*
691  * Locate a member of an archive, given the path of the archive and the path
692  * of the desired member.
693  *
694  * Input:
695  *	archive		Path to the archive
696  *	member		Name of member. If it is a path, only the last
697  *			component is used.
698  *	out_arh		Archive header to be filled in
699  *	mode		"r" for read-only access, "r+" for read-write access
700  *
701  * Output:
702  *	return		The archive file, positioned at the start of the
703  *			member's struct ar_hdr, or NULL if the member doesn't
704  *			exist.
705  *	*out_arh	The current struct ar_hdr for member.
706  *
707  * See ArchStatMember for an almost identical copy of this code.
708  */
709 static FILE *
710 ArchFindMember(const char *archive, const char *member, struct ar_hdr *out_arh,
711 	       const char *mode)
712 {
713 	FILE *arch;		/* Stream to archive */
714 	int size;		/* Size of archive member */
715 	char magic[SARMAG];
716 	size_t len;
717 
718 	arch = fopen(archive, mode);
719 	if (arch == NULL)
720 		return NULL;
721 
722 	/*
723 	 * We use the ARMAG string to make sure this is an archive we
724 	 * can handle...
725 	 */
726 	if (fread(magic, SARMAG, 1, arch) != 1 ||
727 	    strncmp(magic, ARMAG, SARMAG) != 0) {
728 		fclose(arch);
729 		return NULL;
730 	}
731 
732 	/*
733 	 * Because of space constraints and similar things, files are archived
734 	 * using their basename, not the entire path.
735 	 */
736 	member = str_basename(member);
737 
738 	len = strlen(member);
739 
740 	while (fread(out_arh, sizeof *out_arh, 1, arch) == 1) {
741 
742 		if (strncmp(out_arh->ar_fmag, ARFMAG,
743 			    sizeof out_arh->ar_fmag) != 0) {
744 			/*
745 			 * The header is bogus, so the archive is bad
746 			 * and there's no way we can recover...
747 			 */
748 			fclose(arch);
749 			return NULL;
750 		}
751 
752 		DEBUG5(ARCH, "Reading archive %s member %.*s mtime %.*s\n",
753 		       archive,
754 		       (int)sizeof out_arh->ar_name, out_arh->ar_name,
755 		       (int)sizeof out_arh->ar_date, out_arh->ar_date);
756 
757 		if (ArchiveMember_HasName(out_arh, member, len)) {
758 			/*
759 			 * To make life easier for callers that want to update
760 			 * the archive, we reposition the file at the start of
761 			 * the header we just read before we return the
762 			 * stream. In a more general situation, it might be
763 			 * better to leave the file at the actual member,
764 			 * rather than its header, but not here.
765 			 */
766 			if (fseek(arch, -(long)sizeof *out_arh, SEEK_CUR) !=
767 			    0) {
768 				fclose(arch);
769 				return NULL;
770 			}
771 			return arch;
772 		}
773 
774 #ifdef AR_EFMT1
775 		/*
776 		 * BSD 4.4 extended AR format: #1/<namelen>, with name as the
777 		 * first <namelen> bytes of the file
778 		 */
779 		if (strncmp(out_arh->ar_name, AR_EFMT1, sizeof AR_EFMT1 - 1) ==
780 		    0 &&
781 		    (ch_isdigit(out_arh->ar_name[sizeof AR_EFMT1 - 1]))) {
782 			int elen = atoi(&out_arh->ar_name[sizeof AR_EFMT1 - 1]);
783 			char ename[MAXPATHLEN + 1];
784 
785 			if ((unsigned int)elen > MAXPATHLEN) {
786 				fclose(arch);
787 				return NULL;
788 			}
789 			if (fread(ename, (size_t)elen, 1, arch) != 1) {
790 				fclose(arch);
791 				return NULL;
792 			}
793 			ename[elen] = '\0';
794 			if (DEBUG(ARCH) || DEBUG(MAKE))
795 				debug_printf(
796 				    "ArchFindMember: "
797 				    "Extended format entry for %s\n",
798 				    ename);
799 			if (strncmp(ename, member, len) == 0) {
800 				/* Found as extended name */
801 				if (fseek(arch,
802 					  -(long)sizeof(struct ar_hdr) - elen,
803 					  SEEK_CUR) != 0) {
804 					fclose(arch);
805 					return NULL;
806 				}
807 				return arch;
808 			}
809 			if (fseek(arch, -elen, SEEK_CUR) != 0) {
810 				fclose(arch);
811 				return NULL;
812 			}
813 		}
814 #endif
815 
816 		/*
817 		 * This isn't the member we're after, so we need to advance the
818 		 * stream's pointer to the start of the next header. Files are
819 		 * padded with newlines to an even-byte boundary, so we need to
820 		 * extract the size of the file from the 'size' field of the
821 		 * header and round it up during the seek.
822 		 */
823 		out_arh->ar_size[sizeof out_arh->ar_size - 1] = '\0';
824 		size = (int)strtol(out_arh->ar_size, NULL, 10);
825 		if (fseek(arch, (size + 1) & ~1, SEEK_CUR) != 0) {
826 			fclose(arch);
827 			return NULL;
828 		}
829 	}
830 
831 	fclose(arch);
832 	return NULL;
833 }
834 
835 /*
836  * Touch a member of an archive, on disk.
837  * The GNode's modification time is left as-is.
838  *
839  * The st_mtime of the entire archive is also changed.
840  * For a library, it may be required to run ranlib after this.
841  *
842  * Input:
843  *	gn		Node of member to touch
844  *
845  * Results:
846  *	The 'time' field of the member's header is updated.
847  */
848 void
849 Arch_Touch(GNode *gn)
850 {
851 	FILE *f;
852 	struct ar_hdr arh;
853 
854 	f = ArchFindMember(GNode_VarArchive(gn), GNode_VarMember(gn), &arh,
855 			   "r+");
856 	if (f == NULL)
857 		return;
858 
859 	snprintf(arh.ar_date, sizeof arh.ar_date, "%-ld", (unsigned long)now);
860 	(void)fwrite(&arh, sizeof arh, 1, f);
861 	fclose(f);		/* TODO: handle errors */
862 }
863 
864 /*
865  * Given a node which represents a library, touch the thing, making sure that
866  * the table of contents is also touched.
867  *
868  * Both the modification time of the library and of the RANLIBMAG member are
869  * set to 'now'.
870  */
871 /*ARGSUSED*/
872 void
873 Arch_TouchLib(GNode *gn MAKE_ATTR_UNUSED)
874 {
875 #ifdef RANLIBMAG
876 	FILE *f;
877 	struct ar_hdr arh;	/* Header describing table of contents */
878 	struct utimbuf times;
879 
880 	f = ArchFindMember(gn->path, RANLIBMAG, &arh, "r+");
881 	if (f == NULL)
882 		return;
883 
884 	snprintf(arh.ar_date, sizeof arh.ar_date, "%-ld", (unsigned long)now);
885 	(void)fwrite(&arh, sizeof arh, 1, f);
886 	fclose(f);		/* TODO: handle errors */
887 
888 	times.actime = times.modtime = now;
889 	utime(gn->path, &times);	/* TODO: handle errors */
890 #endif
891 }
892 
893 /*
894  * Update the mtime of the GNode with the mtime from the archive member on
895  * disk (or in the cache).
896  */
897 void
898 Arch_UpdateMTime(GNode *gn)
899 {
900 	struct ar_hdr *arh;
901 
902 	arh = ArchStatMember(GNode_VarArchive(gn), GNode_VarMember(gn), true);
903 	if (arh != NULL)
904 		gn->mtime = (time_t)strtol(arh->ar_date, NULL, 10);
905 	else
906 		gn->mtime = 0;
907 }
908 
909 /*
910  * Given a nonexistent archive member's node, update gn->mtime from its
911  * archived form, if it exists.
912  */
913 void
914 Arch_UpdateMemberMTime(GNode *gn)
915 {
916 	GNodeListNode *ln;
917 
918 	for (ln = gn->parents.first; ln != NULL; ln = ln->next) {
919 		GNode *pgn = ln->datum;
920 
921 		if (pgn->type & OP_ARCHV) {
922 			/*
923 			 * If the parent is an archive specification and is
924 			 * being made and its member's name matches the name
925 			 * of the node we were given, record the modification
926 			 * time of the parent in the child. We keep searching
927 			 * its parents in case some other parent requires this
928 			 * child to exist.
929 			 */
930 			const char *nameStart = strchr(pgn->name, '(') + 1;
931 			const char *nameEnd = strchr(nameStart, ')');
932 			size_t nameLen = (size_t)(nameEnd - nameStart);
933 
934 			if ((pgn->flags & REMAKE) &&
935 			    strncmp(nameStart, gn->name, nameLen) == 0) {
936 				Arch_UpdateMTime(pgn);
937 				gn->mtime = pgn->mtime;
938 			}
939 		} else if (pgn->flags & REMAKE) {
940 			/*
941 			 * Something which isn't a library depends on the
942 			 * existence of this target, so it needs to exist.
943 			 */
944 			gn->mtime = 0;
945 			break;
946 		}
947 	}
948 }
949 
950 /*
951  * Search for a library along the given search path.
952  *
953  * The node's 'path' field is set to the found path (including the
954  * actual file name, not -l...). If the system can handle the -L
955  * flag when linking (or we cannot find the library), we assume that
956  * the user has placed the .LIBS variable in the final linking
957  * command (or the linker will know where to find it) and set the
958  * TARGET variable for this node to be the node's name. Otherwise,
959  * we set the TARGET variable to be the full path of the library,
960  * as returned by Dir_FindFile.
961  *
962  * Input:
963  *	gn		Node of library to find
964  */
965 void
966 Arch_FindLib(GNode *gn, SearchPath *path)
967 {
968 	char *libName = str_concat3("lib", gn->name + 2, ".a");
969 	gn->path = Dir_FindFile(libName, path);
970 	free(libName);
971 
972 #ifdef LIBRARIES
973 	Var_Set(gn, TARGET, gn->name);
974 #else
975 	Var_Set(gn, TARGET, GNode_Path(gn));
976 #endif
977 }
978 
979 /*
980  * Decide if a node with the OP_LIB attribute is out-of-date. Called from
981  * GNode_IsOODate to make its life easier.
982  * The library is cached if it hasn't been already.
983  *
984  * There are several ways for a library to be out-of-date that are
985  * not available to ordinary files. In addition, there are ways
986  * that are open to regular files that are not available to
987  * libraries.
988  *
989  * A library that is only used as a source is never
990  * considered out-of-date by itself. This does not preclude the
991  * library's modification time from making its parent be out-of-date.
992  * A library will be considered out-of-date for any of these reasons,
993  * given that it is a target on a dependency line somewhere:
994  *
995  *	Its modification time is less than that of one of its sources
996  *	(gn->mtime < gn->youngestChild->mtime).
997  *
998  *	Its modification time is greater than the time at which the make
999  *	began (i.e. it's been modified in the course of the make, probably
1000  *	by archiving).
1001  *
1002  *	The modification time of one of its sources is greater than the one
1003  *	of its RANLIBMAG member (i.e. its table of contents is out-of-date).
1004  *	We don't compare the archive time vs. TOC time because they can be
1005  *	too close. In my opinion we should not bother with the TOC at all
1006  *	since this is used by 'ar' rules that affect the data contents of the
1007  *	archive, not by ranlib rules, which affect the TOC.
1008  */
1009 bool
1010 Arch_LibOODate(GNode *gn)
1011 {
1012 	bool oodate;
1013 
1014 	if (gn->type & OP_PHONY) {
1015 		oodate = true;
1016 	} else if (!GNode_IsTarget(gn) && Lst_IsEmpty(&gn->children)) {
1017 		oodate = false;
1018 	} else if ((!Lst_IsEmpty(&gn->children) && gn->youngestChild == NULL) ||
1019 		   (gn->mtime > now) ||
1020 		   (gn->youngestChild != NULL &&
1021 		    gn->mtime < gn->youngestChild->mtime)) {
1022 		oodate = true;
1023 	} else {
1024 #ifdef RANLIBMAG
1025 		struct ar_hdr *arh;	/* Header for __.SYMDEF */
1026 		int modTimeTOC;		/* The table-of-contents' mod time */
1027 
1028 		arh = ArchStatMember(gn->path, RANLIBMAG, false);
1029 
1030 		if (arh != NULL) {
1031 			modTimeTOC = (int)strtol(arh->ar_date, NULL, 10);
1032 
1033 			if (DEBUG(ARCH) || DEBUG(MAKE))
1034 				debug_printf("%s modified %s...",
1035 					     RANLIBMAG,
1036 					     Targ_FmtTime(modTimeTOC));
1037 			oodate = gn->youngestChild == NULL ||
1038 				 gn->youngestChild->mtime > modTimeTOC;
1039 		} else {
1040 			/*
1041 			 * A library without a table of contents is out-of-date.
1042 			 */
1043 			if (DEBUG(ARCH) || DEBUG(MAKE))
1044 				debug_printf("no toc...");
1045 			oodate = true;
1046 		}
1047 #else
1048 		oodate = false;
1049 #endif
1050 	}
1051 	return oodate;
1052 }
1053 
1054 /* Initialize the archives module. */
1055 void
1056 Arch_Init(void)
1057 {
1058 	Lst_Init(&archives);
1059 }
1060 
1061 /* Clean up the archives module. */
1062 void
1063 Arch_End(void)
1064 {
1065 #ifdef CLEANUP
1066 	Lst_DoneCall(&archives, ArchFree);
1067 #endif
1068 }
1069 
1070 bool
1071 Arch_IsLib(GNode *gn)
1072 {
1073 	static const char armag[] = "!<arch>\n";
1074 	char buf[sizeof armag - 1];
1075 	int fd;
1076 
1077 	if ((fd = open(gn->path, O_RDONLY)) == -1)
1078 		return false;
1079 
1080 	if (read(fd, buf, sizeof buf) != sizeof buf) {
1081 		(void)close(fd);
1082 		return false;
1083 	}
1084 
1085 	(void)close(fd);
1086 
1087 	return memcmp(buf, armag, sizeof buf) == 0;
1088 }
1089