xref: /illumos-gate/usr/src/cmd/xargs/xargs.c (revision 03270635d68df6a0392fb8f4b7c04acad764648b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2014 Garrett D'Amore <garrett@damore.org>
23  * Copyright 2012 DEY Storage Systems, Inc.
24  * Copyright (c) 2017, Joyent, Inc.
25  *
26  * Portions of this file developed by DEY Storage Systems, Inc. are licensed
27  * under the terms of the Common Development and Distribution License (CDDL)
28  * version 1.0 only.  The use of subsequent versions of the License are
29  * is specifically prohibited unless those terms are not in conflict with
30  * version 1.0 of the License.  You can find this license on-line at
31  * http://www.illumos.org/license/CDDL
32  */
33 /*
34  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
35  * Use is subject to license terms.
36  */
37 
38 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
39 /*	  All Rights Reserved  	*/
40 
41 
42 #include <stdio.h>
43 #include <sys/types.h>
44 #include <sys/wait.h>
45 #include <unistd.h>
46 #include <fcntl.h>
47 #include <string.h>
48 #include <stdarg.h>
49 #include <stdlib.h>
50 #include <limits.h>
51 #include <wchar.h>
52 #include <locale.h>
53 #include <langinfo.h>
54 #include <stropts.h>
55 #include <poll.h>
56 #include <errno.h>
57 #include <stdarg.h>
58 #include <sys/fork.h>
59 #include "getresponse.h"
60 
61 #define	HEAD	0
62 #define	TAIL	1
63 #define	FALSE 0
64 #define	TRUE 1
65 #define	MAXSBUF 255
66 #define	MAXIBUF 512
67 #define	MAXINSERTS 5
68 #define	BUFSIZE LINE_MAX
69 #define	MAXARGS 255
70 #define	INSPAT_STR	"{}"	/* default replstr string for -[Ii]	*/
71 #define	FORK_RETRY	5
72 
73 #define	QBUF_STARTLEN 255  /* start size of growable string buffer */
74 #define	QBUF_INC 100	   /* how much to grow a growable string by */
75 
76 /* We use these macros to help make formatting look "consistent" */
77 #define	EMSG(s)		ermsg(gettext(s "\n"))
78 #define	EMSG2(s, a)	ermsg(gettext(s "\n"), a)
79 #define	PERR(s)		perror(gettext("xargs: " s))
80 
81 /* Some common error messages */
82 
83 #define	LIST2LONG	"Argument list too long"
84 #define	ARG2LONG	"A single argument was greater than %d bytes"
85 #define	MALLOCFAIL	"Memory allocation failure"
86 #define	CORRUPTFILE	"Corrupt input file"
87 #define	WAITFAIL	"Wait failure"
88 #define	CHILDSIG	"Child killed with signal %d"
89 #define	CHILDFAIL	"Command could not continue processing data"
90 #define	FORKFAIL	"Could not fork child"
91 #define	EXECFAIL	"Could not exec command"
92 #define	MISSQUOTE	"Missing quote"
93 #define	BADESCAPE	"Incomplete escape"
94 #define	IBUFOVERFLOW	"Insert buffer overflow"
95 #define	NOCHILDSLOT	"No free child slot available"
96 
97 #define	_(x)	gettext(x)
98 
99 static wctype_t	blank;
100 static char	*arglist[MAXARGS+1];
101 static char	argbuf[BUFSIZE * 2 + 1];
102 static char	lastarg[BUFSIZE + 1];
103 static char	**ARGV = arglist;
104 static char	*LEOF = "_";
105 static char	*INSPAT = INSPAT_STR;
106 static char	ins_buf[MAXIBUF];
107 static char	*p_ibuf;
108 
109 static struct inserts {
110 	char	**p_ARGV;	/* where to put newarg ptr in arg list */
111 	char	*p_skel;	/* ptr to arg template */
112 } saveargv[MAXINSERTS];
113 
114 static int	PROMPT = -1;
115 static int	BUFLIM = BUFSIZE;
116 static int	MAXPROCS = 1;
117 static int	N_ARGS = 0;
118 static int	N_args = 0;
119 static int	N_lines = 0;
120 static int	DASHX = FALSE;
121 static int	MORE = TRUE;
122 static int	PER_LINE = FALSE;
123 static int	LINE_CONT = FALSE;
124 static int	EAT_LEAD = FALSE;
125 static int	ERR = FALSE;
126 static int	OK = TRUE;
127 static int	LEGAL = FALSE;
128 static int	TRACE = FALSE;
129 static int	INSERT = FALSE;
130 static int	ZERO = FALSE;
131 static int	linesize = 0;
132 static int	ibufsize = 0;
133 static int	exitstat = 0;	/* our exit status			*/
134 static int	mac;		/* modified argc, after parsing		*/
135 static char	**mav;		/* modified argv, after parsing		*/
136 static int	n_inserts;	/* # of insertions.			*/
137 static pid_t	*procs;		/* pids of children			*/
138 static int	n_procs;	/* # of child processes.		*/
139 
140 /* our usage message:							*/
141 #define	USAGEMSG "Usage: xargs: [-t] [-p] [-0] [-e[eofstr]] [-E eofstr] "\
142 	"[-I replstr] [-i[replstr]] [-L #] [-l[#]] [-n # [-x]] [-P maxprocs] "\
143 	"[-s size] [cmd [args ...]]\n"
144 
145 static int	echoargs();
146 static wint_t	getwchr(char *, size_t *);
147 static void	lcall(char *sub, char **subargs);
148 static void	addibuf(struct inserts *p);
149 static void	ermsg(char *messages, ...);
150 static char	*addarg(char *arg);
151 static void	store_str(char **, char *, size_t);
152 static char	*getarg(char *);
153 static char	*insert(char *pattern, char *subst);
154 static void	usage();
155 static void	parseargs();
156 static int	procs_find(pid_t child);
157 static void	procs_store(pid_t child);
158 static boolean_t procs_delete(pid_t child);
159 static pid_t	procs_waitpid(boolean_t blocking, int *stat_loc);
160 static void	procs_wait(boolean_t blocking);
161 
162 int
163 main(int argc, char **argv)
164 {
165 	int	j;
166 	unsigned long	l;
167 	struct inserts *psave;
168 	int c;
169 	int	initsize;
170 	char	*cmdname, **initlist;
171 	char	*arg;
172 	char	*next;
173 	char	*eptr;
174 
175 	/* initialization */
176 	blank = wctype("blank");
177 	n_inserts = 0;
178 	psave = saveargv;
179 	(void) setlocale(LC_ALL, "");
180 #if !defined(TEXT_DOMAIN)	/* Should be defined by cc -D 		*/
181 #define	TEXT_DOMAIN "SYS_TEST"	/* Use this only if it weren't 		*/
182 #endif
183 	(void) textdomain(TEXT_DOMAIN);
184 	if (init_yes() < 0) {
185 		ermsg(_(ERR_MSG_INIT_YES), strerror(errno));
186 		exit(1);
187 	}
188 
189 	parseargs(argc, argv);
190 
191 	/* handling all of xargs arguments:				*/
192 	while ((c = getopt(mac, mav, "0tpe:E:I:i:L:l:n:P:s:x")) != EOF) {
193 		switch (c) {
194 		case '0':
195 			ZERO = TRUE;
196 			break;
197 
198 		case 't':	/* -t: turn trace mode on		*/
199 			TRACE = TRUE;
200 			break;
201 
202 		case 'p':	/* -p: turn on prompt mode.		*/
203 			if ((PROMPT = open("/dev/tty", O_RDONLY)) == -1) {
204 				PERR("can't read from tty for -p");
205 			} else {
206 				TRACE = TRUE;
207 			}
208 			break;
209 
210 		case 'e':
211 			/*
212 			 * -e[eofstr]: set/disable end-of-file.
213 			 * N.B. that an argument *isn't* required here; but
214 			 * parseargs forced an argument if not was given.  The
215 			 * forced argument is the default...
216 			 */
217 			LEOF = optarg; /* can be empty */
218 			break;
219 
220 		case 'E':
221 			/*
222 			 * -E eofstr: change end-of-file string.
223 			 * eofstr *is* required here, but can be empty:
224 			 */
225 			LEOF = optarg;
226 			break;
227 
228 		case 'I':
229 			/* -I replstr: Insert mode. replstr *is* required. */
230 			INSERT = PER_LINE = LEGAL = EAT_LEAD = TRUE;
231 			LINE_CONT = FALSE;
232 			N_ARGS = 0;
233 			INSPAT = optarg;
234 			if (*optarg == '\0') {
235 				ermsg(_("Option requires an argument: -%c\n"),
236 				    c);
237 			}
238 			break;
239 
240 		case 'i':
241 			/*
242 			 * -i [replstr]: insert mode, with *optional* replstr.
243 			 * N.B. that an argument *isn't* required here; if
244 			 * it's not given, then the string INSPAT_STR will
245 			 * be assumed.
246 			 *
247 			 * Since getopts(3C) doesn't handle the case of an
248 			 * optional variable argument at all, we have to
249 			 * parse this by hand:
250 			 */
251 
252 			INSERT = PER_LINE = LEGAL = EAT_LEAD = TRUE;
253 			LINE_CONT = FALSE;
254 			N_ARGS = 0;
255 			if ((optarg != NULL) && (*optarg != '\0')) {
256 				INSPAT = optarg;
257 			} else {
258 				/*
259 				 * here, there is no next argument. so
260 				 * we reset INSPAT to the INSPAT_STR.
261 				 * we *have* to do this, as -i/I may have
262 				 * been given previously, and XCU4 requires
263 				 * that only "the last one specified takes
264 				 * effect".
265 				 */
266 				INSPAT = INSPAT_STR;
267 			}
268 			break;
269 
270 		case 'L':
271 			/*
272 			 * -L number: # of times cmd is executed
273 			 * number *is* required here:
274 			 */
275 			PER_LINE = LINE_CONT = TRUE;
276 			N_ARGS = 0;
277 			INSERT = EAT_LEAD = FALSE;
278 			if ((PER_LINE = atoi(optarg)) <= 0) {
279 				ermsg(_("#lines must be positive int: %s\n"),
280 				    optarg);
281 			}
282 			break;
283 
284 		case 'l':
285 			/*
286 			 * -l [number]: # of times cmd is executed
287 			 * N.B. that an argument *isn't* required here; if
288 			 * it's not given, then 1 is assumed.
289 			 *
290 			 * parseargs handles the optional arg processing.
291 			 */
292 
293 			PER_LINE = LINE_CONT = LEGAL = TRUE;
294 			N_ARGS = 0;
295 			INSERT = EAT_LEAD = FALSE;
296 
297 			if ((optarg != NULL) && (*optarg != '\0')) {
298 				if ((PER_LINE = atoi(optarg)) <= 0)
299 					PER_LINE = 1;
300 			}
301 			break;
302 
303 		case 'n':	/* -n number: # stdin args		*/
304 			/*
305 			 * -n number: # stdin args.
306 			 * number *is* required here:
307 			 */
308 			if ((N_ARGS = atoi(optarg)) <= 0) {
309 				ermsg(_("#args must be positive int: %s\n"),
310 				    optarg);
311 			} else {
312 				LEGAL = DASHX || N_ARGS == 1;
313 				INSERT = PER_LINE = LINE_CONT = FALSE;
314 			}
315 			break;
316 
317 		case 'P':	/* -P maxprocs: # of child processses	*/
318 			errno = 0;
319 			l = strtoul(optarg, &eptr, 10);
320 			if (*eptr != '\0' || errno != 0) {
321 				ermsg(_("failed to parse maxprocs (-P): %s\n"),
322 				    optarg);
323 				break;
324 			}
325 
326 			/*
327 			 * Come up with an upper bound that'll probably fit in
328 			 * memory.
329 			 */
330 			if (l == 0 || l > ((INT_MAX / sizeof (pid_t) >> 1))) {
331 				l = INT_MAX / sizeof (pid_t) >> 1;
332 			}
333 			MAXPROCS = (int)l;
334 			break;
335 
336 		case 's':	/* -s size: set max size of each arg list */
337 			BUFLIM = atoi(optarg);
338 			if (BUFLIM > BUFSIZE || BUFLIM <= 0) {
339 				ermsg(_("0 < max-cmd-line-size <= %d: %s\n"),
340 				    BUFSIZE, optarg);
341 			}
342 			break;
343 
344 		case 'x':	/* -x: terminate if args > size limit	*/
345 			DASHX = LEGAL = TRUE;
346 			break;
347 
348 		default:
349 			/*
350 			 * bad argument. complain and get ready to die.
351 			 */
352 			usage();
353 			exit(2);
354 			break;
355 		}
356 	}
357 
358 	/*
359 	 * if anything called ermsg(), something screwed up, so
360 	 * we exit early.
361 	 */
362 	if (OK == FALSE) {
363 		usage();
364 		exit(2);
365 	}
366 
367 	/*
368 	 * we're finished handling xargs's options, so now pick up
369 	 * the command name (if any), and it's options.
370 	 */
371 
372 
373 	mac -= optind;	/* dec arg count by what we've processed 	*/
374 	mav += optind;	/* inc to current mav				*/
375 
376 	procs = calloc(MAXPROCS, sizeof (pid_t));
377 	if (procs == NULL) {
378 		PERR(MALLOCFAIL);
379 		exit(1);
380 	}
381 
382 	if (mac <= 0) {	/* if there're no more args to process,	*/
383 		cmdname = "/usr/bin/echo";	/* our default command	*/
384 		*ARGV++ = addarg(cmdname);	/* use the default cmd.	*/
385 	} else {	/* otherwise keep parsing rest of the string.	*/
386 		/*
387 		 * note that we can't use getopts(3C), and *must* parse
388 		 * this by hand, as we don't know apriori what options the
389 		 * command will take.
390 		 */
391 		cmdname = *mav;	/* get the command name	*/
392 
393 
394 		/* pick up the remaining args from the command line:	*/
395 		while ((OK == TRUE) && (mac-- > 0)) {
396 			/*
397 			 * while we haven't crapped out, and there's
398 			 * work to do:
399 			 */
400 			if (INSERT && ! ERR) {
401 				if (strstr(*mav, INSPAT) != NULL) {
402 					if (++n_inserts > MAXINSERTS) {
403 						ermsg(_("too many args "
404 						    "with %s\n"), INSPAT);
405 						ERR = TRUE;
406 					}
407 					psave->p_ARGV = ARGV;
408 					(psave++)->p_skel = *mav;
409 				}
410 			}
411 			*ARGV++ = addarg(*mav++);
412 		}
413 	}
414 
415 	/* pick up args from standard input */
416 
417 	initlist = ARGV;
418 	initsize = linesize;
419 	lastarg[0] = '\0';
420 
421 	while (OK) {
422 		N_args = 0;
423 		N_lines = 0;
424 		ARGV = initlist;
425 		linesize = initsize;
426 		next = argbuf;
427 
428 		while (MORE || (lastarg[0] != '\0')) {
429 			int l;
430 
431 			if (*lastarg != '\0') {
432 				arg = strcpy(next, lastarg);
433 				*lastarg = '\0';
434 			} else if ((arg = getarg(next)) == NULL) {
435 				break;
436 			}
437 
438 			l = strlen(arg) + 1;
439 			linesize += l;
440 			next += l;
441 
442 			/* Inserts are handled specially later. */
443 			if ((n_inserts == 0) && (linesize >= BUFLIM)) {
444 				/*
445 				 * Legal indicates hard fail if the list is
446 				 * truncated due to size.  So fail, or if we
447 				 * cannot create any list because it would be
448 				 * too big.
449 				 */
450 				if (LEGAL || N_args == 0) {
451 					EMSG(LIST2LONG);
452 					procs_wait(B_TRUE);
453 					exit(2);
454 					/* NOTREACHED */
455 				}
456 
457 				/*
458 				 * Otherwise just save argument for later.
459 				 */
460 				(void) strcpy(lastarg, arg);
461 				break;
462 			}
463 
464 			*ARGV++ = arg;
465 
466 			N_args++;
467 
468 			if ((PER_LINE && (N_lines >= PER_LINE)) ||
469 			    (N_ARGS && (N_args >= N_ARGS))) {
470 				break;
471 			}
472 
473 
474 			if ((ARGV - arglist) == MAXARGS) {
475 				break;
476 			}
477 		}
478 
479 		*ARGV = NULL;
480 		if (N_args == 0) {
481 			/* Reached the end with no more work. */
482 			break;
483 		}
484 
485 		/* insert arg if requested */
486 
487 		if (!ERR && INSERT) {
488 
489 			p_ibuf = ins_buf;
490 			ARGV--;
491 			j = ibufsize = 0;
492 			for (psave = saveargv; ++j <= n_inserts; ++psave) {
493 				addibuf(psave);
494 				if (ERR)
495 					break;
496 			}
497 		}
498 		*ARGV = NULL;
499 
500 		if (n_inserts > 0) {
501 			/*
502 			 * if we've done any insertions, re-calculate the
503 			 * linesize. bomb out if we've exceeded our length.
504 			 */
505 			linesize = 0;
506 			for (ARGV = arglist; *ARGV != NULL; ARGV++) {
507 				linesize += strlen(*ARGV) + 1;
508 			}
509 			if (linesize >= BUFLIM) {
510 				EMSG(LIST2LONG);
511 				procs_wait(B_TRUE);
512 				exit(2);
513 				/* NOTREACHED */
514 			}
515 		}
516 
517 		/* exec command */
518 
519 		if (!ERR) {
520 			if (!MORE &&
521 			    (PER_LINE && N_lines == 0 || N_ARGS && N_args == 0))
522 				exit(exitstat);
523 			OK = TRUE;
524 			j = TRACE ? echoargs() : TRUE;
525 			if (j) {
526 				/*
527 				 * for xcu4, all invocations of cmdname must
528 				 * return 0, in order for us to return 0.
529 				 * so if we have a non-zero status here,
530 				 * quit immediately.
531 				 */
532 				(void) lcall(cmdname, arglist);
533 			}
534 		}
535 	}
536 
537 	procs_wait(B_TRUE);
538 
539 	if (OK)
540 		return (exitstat);
541 
542 	/*
543 	 * if exitstat was set, to match XCU4 complience,
544 	 * return that value, otherwise, return 1.
545 	 */
546 	return (exitstat ? exitstat : 1);
547 }
548 
549 static char *
550 addarg(char *arg)
551 {
552 	linesize += (strlen(arg) + 1);
553 	return (arg);
554 }
555 
556 
557 static void
558 store_str(char **buffer, char *str, size_t len)
559 {
560 	(void) memcpy(*buffer, str, len);
561 	(*buffer)[len] = '\0';
562 	*buffer += len;
563 }
564 
565 
566 static char *
567 getarg(char *arg)
568 {
569 	char	*xarg = arg;
570 	wchar_t	c = 0;
571 	char	mbc[MB_LEN_MAX];
572 	size_t	len;
573 	int	escape = 0;
574 	int	inquote = 0;
575 	int	last = 0;
576 
577 	arg[0] = '\0';
578 
579 	while (MORE) {
580 
581 		len = 0;
582 		last = c;
583 		c = getwchr(mbc, &len);
584 
585 		if (((arg - xarg) + len) > BUFLIM) {
586 			EMSG2(ARG2LONG, BUFLIM);
587 			exit(2);
588 			ERR = TRUE;
589 			return (NULL);
590 		}
591 
592 		switch (c) {
593 		case '\n':
594 			if (ZERO) {
595 				store_str(&arg, mbc, len);
596 				continue;
597 			}
598 			/*
599 			 * NB: Some other versions rip off all of the trailing
600 			 * blanks.  The spec only claims that this should
601 			 * be done for a single blank.  We follow the spec.
602 			 */
603 			if (LINE_CONT && iswctype(last, blank)) {
604 				len = 0;
605 				*arg = 0;
606 				continue;
607 			}
608 			/* FALLTHRU */
609 
610 		case '\0':
611 		case WEOF:	/* Note WEOF == EOF */
612 
613 			if (escape) {
614 				EMSG(BADESCAPE);
615 				ERR = TRUE;
616 				return (NULL);
617 			}
618 			if (inquote) {
619 				EMSG(MISSQUOTE);
620 				ERR = TRUE;
621 				return (NULL);
622 			}
623 
624 			N_lines++;
625 			break;
626 
627 		case '"':
628 			if (ZERO || escape || (inquote == 1)) {
629 				/* treat it literally */
630 				escape = 0;
631 				store_str(&arg, mbc, len);
632 
633 			} else if (inquote == 2) {
634 				/* terminating double quote */
635 				inquote = 0;
636 
637 			} else {
638 				/* starting quoted string */
639 				inquote = 2;
640 			}
641 			continue;
642 
643 		case '\'':
644 			if (ZERO || escape || (inquote == 2)) {
645 				/* treat it literally */
646 				escape = 0;
647 				store_str(&arg, mbc, len);
648 
649 			} else if (inquote == 1) {
650 				/* terminating single quote */
651 				inquote = 0;
652 
653 			} else {
654 				/* starting quoted string */
655 				inquote = 1;
656 			}
657 			continue;
658 
659 		case '\\':
660 			/*
661 			 * Any unquoted character can be escaped by
662 			 * preceding it with a backslash.
663 			 */
664 			if (ZERO || inquote || escape) {
665 				escape = 0;
666 				store_str(&arg, mbc, len);
667 			} else {
668 				escape = 1;
669 			}
670 			continue;
671 
672 		default:
673 			/* most times we will just want to store it */
674 			if (inquote || escape || ZERO || !iswctype(c, blank)) {
675 				escape = 0;
676 				store_str(&arg, mbc, len);
677 				continue;
678 			}
679 			if (EAT_LEAD && last == 0) {
680 				c = 0;		/* Roll it back */
681 				continue;
682 			}
683 			if (PER_LINE) {
684 				store_str(&arg, mbc, len);
685 				continue;
686 			}
687 
688 			/* unquoted blank without special handling */
689 			break;
690 		}
691 
692 		/*
693 		 * At this point we are processing a complete argument.
694 		 */
695 		if (strcmp(xarg, LEOF) == 0 && *LEOF != '\0') {
696 			MORE = FALSE;
697 			return (NULL);
698 		}
699 		if (c == WEOF) {
700 			MORE = FALSE;
701 		}
702 		if (xarg[0] == '\0')
703 			continue;
704 		break;
705 	}
706 
707 	return (xarg[0] == '\0' ? NULL : xarg);
708 }
709 
710 /*
711  * ermsg():	print out an error message, and indicate failure globally.
712  *
713  *	Assumes that message has already been gettext()'d. It would be
714  *	nice if we could just do the gettext() here, but we can't, since
715  *	since xgettext(1M) wouldn't be able to pick up our error message.
716  */
717 /* PRINTFLIKE1 */
718 static void
719 ermsg(char *messages, ...)
720 {
721 	va_list	ap;
722 
723 	va_start(ap, messages);
724 
725 	(void) fprintf(stderr, "xargs: ");
726 	(void) vfprintf(stderr, messages, ap);
727 
728 	va_end(ap);
729 	OK = FALSE;
730 }
731 
732 static int
733 echoargs()
734 {
735 	char	**anarg;
736 	char	**tanarg;	/* tmp ptr			*/
737 	int		i;
738 	char		reply[LINE_MAX];
739 
740 	tanarg = anarg = arglist-1;
741 
742 	/*
743 	 * write out each argument, separated by a space. the tanarg
744 	 * nonsense is for xcu4 testsuite compliance - so that an
745 	 * extra space isn't echoed after the last argument.
746 	 */
747 	while (*++anarg) {		/* while there's an argument	*/
748 		++tanarg;		/* follow anarg			*/
749 		(void) write(2, *anarg, strlen(*anarg));
750 
751 		if (*++tanarg) {	/* if there's another argument:	*/
752 			(void) write(2, " ", 1); /* add a space		*/
753 			--tanarg;	/* reset back to anarg		*/
754 		}
755 	}
756 	if (PROMPT == -1) {
757 		(void) write(2, "\n", 1);
758 		return (TRUE);
759 	}
760 
761 	(void) write(2, "?...", 4);	/* ask the user for input	*/
762 
763 	for (i = 0; i < LINE_MAX && read(PROMPT, &reply[i], 1) > 0; i++) {
764 		if (reply[i] == '\n') {
765 			if (i == 0)
766 				return (FALSE);
767 			break;
768 		}
769 	}
770 	reply[i] = 0;
771 
772 	/* flush remainder of line if necessary */
773 	if (i == LINE_MAX) {
774 		char	bitbucket;
775 
776 		while ((read(PROMPT, &bitbucket, 1) > 0) && (bitbucket != '\n'))
777 			;
778 	}
779 
780 	return (yes_check(reply));
781 }
782 
783 
784 static char *
785 insert(char *pattern, char *subst)
786 {
787 	static char	buffer[MAXSBUF+1];
788 	int		len, ipatlen;
789 	char	*pat;
790 	char	*bufend;
791 	char	*pbuf;
792 
793 	len = strlen(subst);
794 	ipatlen = strlen(INSPAT) - 1;
795 	pat = pattern - 1;
796 	pbuf = buffer;
797 	bufend = &buffer[MAXSBUF];
798 
799 	while (*++pat) {
800 		if (strncmp(pat, INSPAT, ipatlen + 1) == 0) {
801 			if (pbuf + len >= bufend) {
802 				break;
803 			} else {
804 				(void) strcpy(pbuf, subst);
805 				pat += ipatlen;
806 				pbuf += len;
807 			}
808 		} else {
809 			*pbuf++ = *pat;
810 			if (pbuf >= bufend)
811 				break;
812 		}
813 	}
814 
815 	if (!*pat) {
816 		*pbuf = '\0';
817 		return (buffer);
818 	} else {
819 		ermsg(gettext("Maximum argument size with insertion via %s's "
820 		    "exceeded\n"), INSPAT);
821 		ERR = TRUE;
822 		return (NULL);
823 	}
824 }
825 
826 
827 static void
828 addibuf(struct inserts	*p)
829 {
830 	char	*newarg, *skel, *sub;
831 	int		l;
832 
833 	skel = p->p_skel;
834 	sub = *ARGV;
835 	newarg = insert(skel, sub);
836 	if (ERR)
837 		return;
838 
839 	l = strlen(newarg) + 1;
840 	if ((ibufsize += l) > MAXIBUF) {
841 		EMSG(IBUFOVERFLOW);
842 		ERR = TRUE;
843 	}
844 	(void) strcpy(p_ibuf, newarg);
845 	*(p->p_ARGV) = p_ibuf;
846 	p_ibuf += l;
847 }
848 
849 
850 /*
851  * getwchr():	get the next wide character.
852  * description:
853  *	we get the next character from stdin.  This returns WEOF if no
854  *	character is present.  If ZERO is set, it gets a single byte instead
855  *	a wide character.
856  */
857 static wint_t
858 getwchr(char *mbc, size_t *sz)
859 {
860 	size_t		i;
861 	int		c;
862 	wchar_t		wch;
863 
864 	i = 0;
865 	while (i < MB_CUR_MAX) {
866 
867 		if ((c = fgetc(stdin)) == EOF) {
868 
869 			if (i == 0) {
870 				/* TRUE EOF has been reached */
871 				return (WEOF);
872 			}
873 
874 			/*
875 			 * We have some characters in our buffer still so it
876 			 * must be an invalid character right before EOF.
877 			 */
878 			break;
879 		}
880 		mbc[i++] = (char)c;
881 
882 		/* If this succeeds then we are done */
883 		if (ZERO) {
884 			*sz = i;
885 			return ((char)c);
886 		}
887 		if (mbtowc(&wch, mbc, i) != -1) {
888 			*sz = i;
889 			return ((wint_t)wch);
890 		}
891 	}
892 
893 	/*
894 	 * We have now encountered an illegal character sequence.
895 	 * There is nothing much we can do at this point but
896 	 * return an error.  If we attempt to recover we may in fact
897 	 * return garbage as arguments, from the customer's point
898 	 * of view.  After all what if they are feeding us a file
899 	 * generated in another locale?
900 	 */
901 	errno = EILSEQ;
902 	PERR(CORRUPTFILE);
903 	exit(1);
904 	/* NOTREACHED */
905 }
906 
907 
908 static void
909 lcall(char *sub, char **subargs)
910 {
911 	int	retry = 0;
912 	pid_t	child;
913 
914 	for (;;) {
915 		switch (child = forkx(FORK_NOSIGCHLD)) {
916 		default:
917 			procs_store(child);
918 			/*
919 			 * Note, if we have used up all of our slots, then this
920 			 * call may end up blocking.
921 			 */
922 			procs_wait(B_FALSE);
923 			return;
924 		case 0:
925 			(void) execvp(sub, subargs);
926 			PERR(EXECFAIL);
927 			if (errno == EACCES)
928 				exit(126);
929 			exit(127);
930 			/* NOTREACHED */
931 		case -1:
932 			if (errno != EAGAIN && retry++ < FORK_RETRY) {
933 				PERR(FORKFAIL);
934 				exit(123);
935 			}
936 			(void) sleep(1);
937 		}
938 	}
939 }
940 
941 /*
942  * Return the index of child in the procs array.
943  */
944 static int
945 procs_find(pid_t child)
946 {
947 	int	i;
948 
949 	for (i = 0; i < MAXPROCS; i++) {
950 		if (procs[i] == child) {
951 			return (i);
952 		}
953 	}
954 
955 	return (-1);
956 }
957 
958 static void
959 procs_store(pid_t child)
960 {
961 	int	i;
962 
963 	i = procs_find(0);
964 	if (i < 0) {
965 		EMSG(NOCHILDSLOT);
966 		exit(1);
967 	}
968 	procs[i] = child;
969 	n_procs++;
970 }
971 
972 static boolean_t
973 procs_delete(pid_t child)
974 {
975 	int	i;
976 
977 	i = procs_find(child);
978 	if (i < 0) {
979 		return (B_FALSE);
980 	}
981 
982 	procs[i] = (pid_t)0;
983 	n_procs--;
984 
985 	return (B_TRUE);
986 }
987 
988 static pid_t
989 procs_waitpid(boolean_t blocking, int *stat_loc)
990 {
991 	pid_t	child;
992 	int	options;
993 
994 	if (n_procs == 0) {
995 		errno = ECHILD;
996 		return (-1);
997 	}
998 
999 	options = 0;
1000 	if (!blocking) {
1001 		options |= WNOHANG;
1002 	}
1003 
1004 	while ((child = waitpid((pid_t)-1, stat_loc, options)) > 0) {
1005 		if (procs_delete(child)) {
1006 			break;
1007 		}
1008 	}
1009 
1010 	return (child);
1011 }
1012 
1013 static void
1014 procs_wait(boolean_t blocking)
1015 {
1016 	pid_t	child;
1017 	int	stat_loc;
1018 
1019 	/*
1020 	 * If we currently have filled all of our slots, then we need to block
1021 	 * further execution.
1022 	 */
1023 	if (n_procs >= MAXPROCS)
1024 		blocking = B_TRUE;
1025 	while ((child = procs_waitpid(blocking, &stat_loc)) > 0) {
1026 		if (WIFSIGNALED(stat_loc)) {
1027 			EMSG2(CHILDSIG, WTERMSIG(stat_loc));
1028 			exit(125);
1029 			/* NOTREACHED */
1030 		} else if ((WEXITSTATUS(stat_loc) & 0377) == 0377) {
1031 			EMSG(CHILDFAIL);
1032 			exit(124);
1033 			/* NOTREACHED */
1034 		} else {
1035 			exitstat |= WEXITSTATUS(stat_loc);
1036 		}
1037 	}
1038 
1039 	if (child == (pid_t)(-1) && errno != ECHILD) {
1040 		EMSG(WAITFAIL);
1041 		exit(122);
1042 		/* NOTREACHED */
1043 	}
1044 }
1045 
1046 static void
1047 usage()
1048 {
1049 	ermsg(_(USAGEMSG));
1050 	OK = FALSE;
1051 }
1052 
1053 
1054 
1055 /*
1056  * parseargs():		modify the args
1057  *	since the -e, -i and -l flags all take optional subarguments,
1058  *	and getopts(3C) is clueless about this nonsense, we change the
1059  *	our local argument count and strings to separate this out,
1060  *	and make it easier to handle via getopts(3c).
1061  *
1062  *	-e	-> "-e ""
1063  *	-e3	-> "-e "3"
1064  *	-Estr	-> "-E "str"
1065  *	-i	-> "-i "{}"
1066  *	-irep	-> "-i "rep"
1067  *	-l	-> "-l "1"
1068  *	-l10	-> "-l "10"
1069  *
1070  *	since the -e, -i and -l flags all take optional subarguments,
1071  */
1072 static void
1073 parseargs(int ac, char **av)
1074 {
1075 	int i;			/* current argument			*/
1076 	int cflag;		/* 0 = not processing cmd arg		*/
1077 
1078 	if ((mav = malloc((ac * 2 + 1) * sizeof (char *))) == NULL) {
1079 		PERR(MALLOCFAIL);
1080 		exit(1);
1081 	}
1082 
1083 	/* for each argument, see if we need to change things:		*/
1084 	for (i = mac = cflag = 0; (av[i] != NULL) && i < ac; i++, mac++) {
1085 		if ((mav[mac] = strdup(av[i])) == NULL) {
1086 			PERR(MALLOCFAIL);
1087 			exit(1);
1088 		}
1089 
1090 		/* -- has been found or argument list is fully processes */
1091 		if (cflag)
1092 			continue;
1093 
1094 		/*
1095 		 * if we're doing special processing, and we've got a flag
1096 		 */
1097 		else if ((av[i][0] == '-') && (av[i][1] != NULL)) {
1098 			char	*def;
1099 
1100 			switch (av[i][1]) {
1101 			case	'e':
1102 				def = ""; /* -e with no arg turns off eof */
1103 				goto process_special;
1104 			case	'i':
1105 				def = INSPAT_STR;
1106 				goto process_special;
1107 			case	'l':
1108 				def = "1";
1109 process_special:
1110 				/*
1111 				 * if there's no sub-option, we *must* add
1112 				 * a default one. this is because xargs must
1113 				 * be able to distinguish between a valid
1114 				 * suboption, and a command name.
1115 				 */
1116 				if (av[i][2] == NULL) {
1117 					mav[++mac] = strdup(def);
1118 				} else {
1119 					/* clear out our version: */
1120 					mav[mac][2] = NULL;
1121 					mav[++mac] = strdup(&av[i][2]);
1122 				}
1123 				if (mav[mac] == NULL) {
1124 					PERR(MALLOCFAIL);
1125 					exit(1);
1126 				}
1127 				break;
1128 
1129 			/* flags with required subarguments:		*/
1130 
1131 			/*
1132 			 * there are two separate cases here. either the
1133 			 * flag can have the normal XCU4 handling
1134 			 * (of the form: -X subargument); or it can have
1135 			 * the old solaris 2.[0-4] handling (of the
1136 			 * form: -Xsubargument). in order to maintain
1137 			 * backwards compatibility, we must support the
1138 			 * latter case. we handle the latter possibility
1139 			 * first so both the old solaris way of handling
1140 			 * and the new XCU4 way of handling things are allowed.
1141 			 */
1142 			case	'n':	/* FALLTHROUGH			*/
1143 			case	'P':	/* FALLTHROUGH			*/
1144 			case	's':	/* FALLTHROUGH			*/
1145 			case	'E':	/* FALLTHROUGH			*/
1146 			case	'I':	/* FALLTHROUGH			*/
1147 			case	'L':
1148 				/*
1149 				 * if the second character isn't null, then
1150 				 * the user has specified the old syntax.
1151 				 * we move the subargument into our
1152 				 * mod'd argument list.
1153 				 */
1154 				if (av[i][2] != NULL) {
1155 					/* first clean things up:	*/
1156 					mav[mac][2] = NULL;
1157 
1158 					/* now add the separation:	*/
1159 					++mac;	/* inc to next mod'd arg */
1160 					if ((mav[mac] = strdup(&av[i][2])) ==
1161 					    NULL) {
1162 						PERR(MALLOCFAIL);
1163 						exit(1);
1164 					}
1165 					break;
1166 				}
1167 				i++;
1168 				mac++;
1169 
1170 				if (av[i] == NULL) {
1171 					mav[mac] = NULL;
1172 					return;
1173 				}
1174 				if ((mav[mac] = strdup(av[i])) == NULL) {
1175 					PERR(MALLOCFAIL);
1176 					exit(1);
1177 				}
1178 				break;
1179 
1180 			/* flags */
1181 			case 'p' :
1182 			case 't' :
1183 			case 'x' :
1184 			case '0' :
1185 				break;
1186 
1187 			case '-' :
1188 			default:
1189 				/*
1190 				 * here we've hit the cmd argument. so
1191 				 * we'll stop special processing, as the
1192 				 * cmd may have a "-i" etc., argument,
1193 				 * and we don't want to add a "" to it.
1194 				 */
1195 				cflag = 1;
1196 				break;
1197 			}
1198 		} else if (i > 0) {	/* if we're not the 1st arg	*/
1199 			/*
1200 			 * if it's not a flag, then it *must* be the cmd.
1201 			 * set cflag, so we don't mishandle the -[eil] flags.
1202 			 */
1203 			cflag = 1;
1204 		}
1205 	}
1206 
1207 	mav[mac] = NULL;
1208 }
1209