mirror of
https://git.hardenedbsd.org/hardenedbsd/HardenedBSD.git
synced 2024-12-24 19:21:38 +01:00
29eaad36e5
type instead of int all over the place. (Cosmetic, enhances debugging.) Point out that a date specification _must_ follow the time of day spec, in the man page. This clarifies the last point PR # of bin/483: "at doesn't seem to ..." (the remainder has already been fixed with version 1.3 of parsetime.c).
616 lines
14 KiB
C
616 lines
14 KiB
C
/*
|
|
* parsetime.c - parse time for at(1)
|
|
* Copyright (C) 1993, 1994 Thomas Koenig
|
|
*
|
|
* modifications for english-language times
|
|
* Copyright (C) 1993 David Parsons
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* 2. The name of the author(s) may not be used to endorse or promote
|
|
* products derived from this software without specific prior written
|
|
* permission.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
|
|
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
|
|
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
|
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
* THEORY OF LIABILITY, WETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
|
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
*
|
|
* at [NOW] PLUS NUMBER MINUTES|HOURS|DAYS|WEEKS
|
|
* /NUMBER [DOT NUMBER] [AM|PM]\ /[MONTH NUMBER [NUMBER]] \
|
|
* |NOON | |[TOMORROW] |
|
|
* |MIDNIGHT | |[DAY OF WEEK] |
|
|
* \TEATIME / |NUMBER [SLASH NUMBER [SLASH NUMBER]]|
|
|
* \PLUS NUMBER MINUTES|HOURS|DAYS|WEEKS/
|
|
*/
|
|
|
|
/* System Headers */
|
|
|
|
|
|
#include <sys/types.h>
|
|
#include <errno.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <time.h>
|
|
#include <unistd.h>
|
|
#include <ctype.h>
|
|
#ifndef __FreeBSD__
|
|
#include <getopt.h>
|
|
#endif
|
|
|
|
/* Local headers */
|
|
|
|
#include "at.h"
|
|
#include "panic.h"
|
|
|
|
|
|
/* Structures and unions */
|
|
|
|
enum tok { /* symbols */
|
|
MIDNIGHT, NOON, TEATIME,
|
|
PM, AM, TOMORROW, TODAY, NOW,
|
|
MINUTES, HOURS, DAYS, WEEKS,
|
|
NUMBER, PLUS, DOT, SLASH, ID, JUNK,
|
|
JAN, FEB, MAR, APR, MAY, JUN,
|
|
JUL, AUG, SEP, OCT, NOV, DEC,
|
|
SUN, MON, TUE, WED, THU, FRI, SAT
|
|
};
|
|
|
|
/* parse translation table - table driven parsers can be your FRIEND!
|
|
*/
|
|
struct {
|
|
const char *name; /* token name */
|
|
enum tok value; /* token id */
|
|
int plural; /* is this plural? */
|
|
} Specials[] = {
|
|
{ "midnight", MIDNIGHT,0 }, /* 00:00:00 of today or tomorrow */
|
|
{ "noon", NOON,0 }, /* 12:00:00 of today or tomorrow */
|
|
{ "teatime", TEATIME,0 }, /* 16:00:00 of today or tomorrow */
|
|
{ "am", AM,0 }, /* morning times for 0-12 clock */
|
|
{ "pm", PM,0 }, /* evening times for 0-12 clock */
|
|
{ "tomorrow", TOMORROW,0 }, /* execute 24 hours from time */
|
|
{ "today", TODAY, 0 }, /* execute today - don't advance time */
|
|
{ "now", NOW,0 }, /* opt prefix for PLUS */
|
|
|
|
{ "minute", MINUTES,0 }, /* minutes multiplier */
|
|
{ "minutes", MINUTES,1 }, /* (pluralized) */
|
|
{ "hour", HOURS,0 }, /* hours ... */
|
|
{ "hours", HOURS,1 }, /* (pluralized) */
|
|
{ "day", DAYS,0 }, /* days ... */
|
|
{ "days", DAYS,1 }, /* (pluralized) */
|
|
{ "week", WEEKS,0 }, /* week ... */
|
|
{ "weeks", WEEKS,1 }, /* (pluralized) */
|
|
{ "jan", JAN,0 },
|
|
{ "feb", FEB,0 },
|
|
{ "mar", MAR,0 },
|
|
{ "apr", APR,0 },
|
|
{ "may", MAY,0 },
|
|
{ "jun", JUN,0 },
|
|
{ "jul", JUL,0 },
|
|
{ "aug", AUG,0 },
|
|
{ "sep", SEP,0 },
|
|
{ "oct", OCT,0 },
|
|
{ "nov", NOV,0 },
|
|
{ "dec", DEC,0 },
|
|
{ "sunday", SUN, 0 },
|
|
{ "sun", SUN, 0 },
|
|
{ "monday", MON, 0 },
|
|
{ "mon", MON, 0 },
|
|
{ "tuesday", TUE, 0 },
|
|
{ "tue", TUE, 0 },
|
|
{ "wednesday", WED, 0 },
|
|
{ "wed", WED, 0 },
|
|
{ "thursday", THU, 0 },
|
|
{ "thu", THU, 0 },
|
|
{ "friday", FRI, 0 },
|
|
{ "fri", FRI, 0 },
|
|
{ "saturday", SAT, 0 },
|
|
{ "sat", SAT, 0 },
|
|
} ;
|
|
|
|
/* File scope variables */
|
|
|
|
static char **scp; /* scanner - pointer at arglist */
|
|
static char scc; /* scanner - count of remaining arguments */
|
|
static char *sct; /* scanner - next char pointer in current argument */
|
|
static int need; /* scanner - need to advance to next argument */
|
|
|
|
static char *sc_token; /* scanner - token buffer */
|
|
static size_t sc_len; /* scanner - lenght of token buffer */
|
|
static enum tok sc_tokid; /* scanner - token id */
|
|
static int sc_tokplur; /* scanner - is token plural? */
|
|
|
|
static char rcsid[] = "$Id: parsetime.c,v 1.4 1995/05/30 06:29:25 rgrimes Exp $";
|
|
|
|
/* Local functions */
|
|
|
|
/*
|
|
* parse a token, checking if it's something special to us
|
|
*/
|
|
static enum tok
|
|
parse_token(const char *arg)
|
|
{
|
|
int i;
|
|
|
|
for (i=0; i<(sizeof Specials/sizeof Specials[0]); i++)
|
|
if (strcasecmp(Specials[i].name, arg) == 0) {
|
|
sc_tokplur = Specials[i].plural;
|
|
return sc_tokid = Specials[i].value;
|
|
}
|
|
|
|
/* not special - must be some random id */
|
|
return ID;
|
|
} /* parse_token */
|
|
|
|
|
|
/*
|
|
* init_scanner() sets up the scanner to eat arguments
|
|
*/
|
|
static void
|
|
init_scanner(int argc, char **argv)
|
|
{
|
|
scp = argv;
|
|
scc = argc;
|
|
need = 1;
|
|
sc_len = 1;
|
|
while (--argc > 0)
|
|
sc_len += strlen(*++argv);
|
|
|
|
sc_token = (char *) mymalloc(sc_len);
|
|
} /* init_scanner */
|
|
|
|
/*
|
|
* token() fetches a token from the input stream
|
|
*/
|
|
static enum tok
|
|
token()
|
|
{
|
|
int idx;
|
|
|
|
while (1) {
|
|
memset(sc_token, 0, sc_len);
|
|
sc_tokid = EOF;
|
|
sc_tokplur = 0;
|
|
idx = 0;
|
|
|
|
/* if we need to read another argument, walk along the argument list;
|
|
* when we fall off the arglist, we'll just return EOF forever
|
|
*/
|
|
if (need) {
|
|
if (scc < 1)
|
|
return sc_tokid;
|
|
sct = *scp;
|
|
scp++;
|
|
scc--;
|
|
need = 0;
|
|
}
|
|
/* eat whitespace now - if we walk off the end of the argument,
|
|
* we'll continue, which puts us up at the top of the while loop
|
|
* to fetch the next argument in
|
|
*/
|
|
while (isspace(*sct))
|
|
++sct;
|
|
if (!*sct) {
|
|
need = 1;
|
|
continue;
|
|
}
|
|
|
|
/* preserve the first character of the new token
|
|
*/
|
|
sc_token[0] = *sct++;
|
|
|
|
/* then see what it is
|
|
*/
|
|
if (isdigit(sc_token[0])) {
|
|
while (isdigit(*sct))
|
|
sc_token[++idx] = *sct++;
|
|
sc_token[++idx] = 0;
|
|
return sc_tokid = NUMBER;
|
|
}
|
|
else if (isalpha(sc_token[0])) {
|
|
while (isalpha(*sct))
|
|
sc_token[++idx] = *sct++;
|
|
sc_token[++idx] = 0;
|
|
return parse_token(sc_token);
|
|
}
|
|
else if (sc_token[0] == ':' || sc_token[0] == '.')
|
|
return sc_tokid = DOT;
|
|
else if (sc_token[0] == '+')
|
|
return sc_tokid = PLUS;
|
|
else if (sc_token[0] == '/')
|
|
return sc_tokid = SLASH;
|
|
else
|
|
return sc_tokid = JUNK;
|
|
} /* while (1) */
|
|
} /* token */
|
|
|
|
|
|
/*
|
|
* plonk() gives an appropriate error message if a token is incorrect
|
|
*/
|
|
static void
|
|
plonk(int tok)
|
|
{
|
|
panic((tok == EOF) ? "incomplete time"
|
|
: "garbled time");
|
|
} /* plonk */
|
|
|
|
|
|
/*
|
|
* expect() gets a token and dies most horribly if it's not the token we want
|
|
*/
|
|
static void
|
|
expect(enum tok desired)
|
|
{
|
|
if (token() != desired)
|
|
plonk(sc_tokid); /* and we die here... */
|
|
} /* expect */
|
|
|
|
|
|
/*
|
|
* dateadd() adds a number of minutes to a date. It is extraordinarily
|
|
* stupid regarding day-of-month overflow, and will most likely not
|
|
* work properly
|
|
*/
|
|
static void
|
|
dateadd(int minutes, struct tm *tm)
|
|
{
|
|
/* increment days */
|
|
|
|
while (minutes > 24*60) {
|
|
minutes -= 24*60;
|
|
tm->tm_mday++;
|
|
}
|
|
|
|
/* increment hours */
|
|
while (minutes > 60) {
|
|
minutes -= 60;
|
|
tm->tm_hour++;
|
|
if (tm->tm_hour > 23) {
|
|
tm->tm_mday++;
|
|
tm->tm_hour = 0;
|
|
}
|
|
}
|
|
|
|
/* increment minutes */
|
|
tm->tm_min += minutes;
|
|
|
|
if (tm->tm_min > 59) {
|
|
tm->tm_hour++;
|
|
tm->tm_min -= 60;
|
|
|
|
if (tm->tm_hour > 23) {
|
|
tm->tm_mday++;
|
|
tm->tm_hour = 0;
|
|
}
|
|
}
|
|
} /* dateadd */
|
|
|
|
|
|
/*
|
|
* plus() parses a now + time
|
|
*
|
|
* at [NOW] PLUS NUMBER [MINUTES|HOURS|DAYS|WEEKS]
|
|
*
|
|
*/
|
|
static void
|
|
plus(struct tm *tm)
|
|
{
|
|
int delay;
|
|
int expectplur;
|
|
|
|
expect(NUMBER);
|
|
|
|
delay = atoi(sc_token);
|
|
expectplur = (delay != 1) ? 1 : 0;
|
|
|
|
switch (token()) {
|
|
case WEEKS:
|
|
delay *= 7;
|
|
case DAYS:
|
|
delay *= 24;
|
|
case HOURS:
|
|
delay *= 60;
|
|
case MINUTES:
|
|
if (expectplur != sc_tokplur)
|
|
fprintf(stderr, "at: pluralization is wrong\n");
|
|
dateadd(delay, tm);
|
|
return;
|
|
}
|
|
plonk(sc_tokid);
|
|
} /* plus */
|
|
|
|
|
|
/*
|
|
* tod() computes the time of day
|
|
* [NUMBER [DOT NUMBER] [AM|PM]]
|
|
*/
|
|
static void
|
|
tod(struct tm *tm)
|
|
{
|
|
int hour, minute = 0;
|
|
int tlen;
|
|
|
|
hour = atoi(sc_token);
|
|
tlen = strlen(sc_token);
|
|
|
|
/* first pick out the time of day - if it's 4 digits, we assume
|
|
* a HHMM time, otherwise it's HH DOT MM time
|
|
*/
|
|
if (token() == DOT) {
|
|
expect(NUMBER);
|
|
minute = atoi(sc_token);
|
|
if (minute > 59)
|
|
panic("garbled time");
|
|
token();
|
|
}
|
|
else if (tlen == 4) {
|
|
minute = hour%100;
|
|
if (minute > 59)
|
|
panic("garbeld time");
|
|
hour = hour/100;
|
|
}
|
|
|
|
/* check if an AM or PM specifier was given
|
|
*/
|
|
if (sc_tokid == AM || sc_tokid == PM) {
|
|
if (hour > 12)
|
|
panic("garbled time");
|
|
|
|
if (sc_tokid == PM)
|
|
hour += 12;
|
|
token();
|
|
}
|
|
else if (hour > 23)
|
|
panic("garbled time");
|
|
|
|
/* if we specify an absolute time, we don't want to bump the day even
|
|
* if we've gone past that time - but if we're specifying a time plus
|
|
* a relative offset, it's okay to bump things
|
|
*/
|
|
if ((sc_tokid == EOF || sc_tokid == PLUS) && tm->tm_hour > hour) {
|
|
tm->tm_mday++;
|
|
tm->tm_wday++;
|
|
}
|
|
|
|
tm->tm_hour = hour;
|
|
tm->tm_min = minute;
|
|
if (tm->tm_hour == 24) {
|
|
tm->tm_hour = 0;
|
|
tm->tm_mday++;
|
|
}
|
|
} /* tod */
|
|
|
|
|
|
/*
|
|
* assign_date() assigns a date, wrapping to next year if needed
|
|
*/
|
|
static void
|
|
assign_date(struct tm *tm, long mday, long mon, long year)
|
|
{
|
|
if (year > 99) {
|
|
if (year > 1899)
|
|
year -= 1900;
|
|
else
|
|
panic("garbled time");
|
|
}
|
|
|
|
if (year < 0 &&
|
|
(tm->tm_mon > mon ||(tm->tm_mon == mon && tm->tm_mday > mday)))
|
|
year = tm->tm_year + 1;
|
|
|
|
tm->tm_mday = mday;
|
|
tm->tm_mon = mon;
|
|
|
|
if (year >= 0)
|
|
tm->tm_year = year;
|
|
} /* assign_date */
|
|
|
|
|
|
/*
|
|
* month() picks apart a month specification
|
|
*
|
|
* /[<month> NUMBER [NUMBER]] \
|
|
* |[TOMORROW] |
|
|
* |[DAY OF WEEK] |
|
|
* |NUMBER [SLASH NUMBER [SLASH NUMBER]]|
|
|
* \PLUS NUMBER MINUTES|HOURS|DAYS|WEEKS/
|
|
*/
|
|
static void
|
|
month(struct tm *tm)
|
|
{
|
|
long year= (-1);
|
|
long mday, wday, mon;
|
|
int tlen;
|
|
|
|
switch (sc_tokid) {
|
|
case PLUS:
|
|
plus(tm);
|
|
break;
|
|
|
|
case TOMORROW:
|
|
/* do something tomorrow */
|
|
tm->tm_mday ++;
|
|
tm->tm_wday ++;
|
|
case TODAY: /* force ourselves to stay in today - no further processing */
|
|
token();
|
|
break;
|
|
|
|
case JAN: case FEB: case MAR: case APR: case MAY: case JUN:
|
|
case JUL: case AUG: case SEP: case OCT: case NOV: case DEC:
|
|
/* do month mday [year]
|
|
*/
|
|
mon = (sc_tokid-JAN);
|
|
expect(NUMBER);
|
|
mday = atol(sc_token);
|
|
if (token() == NUMBER) {
|
|
year = atol(sc_token);
|
|
token();
|
|
}
|
|
assign_date(tm, mday, mon, year);
|
|
break;
|
|
|
|
case SUN: case MON: case TUE:
|
|
case WED: case THU: case FRI:
|
|
case SAT:
|
|
/* do a particular day of the week
|
|
*/
|
|
wday = (sc_tokid-SUN);
|
|
|
|
mday = tm->tm_mday;
|
|
|
|
/* if this day is < today, then roll to next week
|
|
*/
|
|
if (wday < tm->tm_wday)
|
|
mday += 7 - (tm->tm_wday - wday);
|
|
else
|
|
mday += (wday - tm->tm_wday);
|
|
|
|
tm->tm_wday = wday;
|
|
|
|
assign_date(tm, mday, tm->tm_mon, tm->tm_year);
|
|
break;
|
|
|
|
case NUMBER:
|
|
/* get numeric MMDDYY, mm/dd/yy, or dd.mm.yy
|
|
*/
|
|
tlen = strlen(sc_token);
|
|
mon = atol(sc_token);
|
|
token();
|
|
|
|
if (sc_tokid == SLASH || sc_tokid == DOT) {
|
|
enum tok sep;
|
|
|
|
sep = sc_tokid;
|
|
expect(NUMBER);
|
|
mday = atol(sc_token);
|
|
if (token() == sep) {
|
|
expect(NUMBER);
|
|
year = atol(sc_token);
|
|
token();
|
|
}
|
|
|
|
/* flip months and days for european timing
|
|
*/
|
|
if (sep == DOT) {
|
|
int x = mday;
|
|
mday = mon;
|
|
mon = x;
|
|
}
|
|
}
|
|
else if (tlen == 6 || tlen == 8) {
|
|
if (tlen == 8) {
|
|
year = (mon % 10000) - 1900;
|
|
mon /= 10000;
|
|
}
|
|
else {
|
|
year = mon % 100;
|
|
mon /= 100;
|
|
}
|
|
mday = mon % 100;
|
|
mon /= 100;
|
|
}
|
|
else
|
|
panic("garbled time");
|
|
|
|
mon--;
|
|
if (mon < 0 || mon > 11 || mday < 1 || mday > 31)
|
|
panic("garbled time");
|
|
|
|
assign_date(tm, mday, mon, year);
|
|
break;
|
|
} /* case */
|
|
} /* month */
|
|
|
|
|
|
/* Global functions */
|
|
|
|
time_t
|
|
parsetime(int argc, char **argv)
|
|
{
|
|
/* Do the argument parsing, die if necessary, and return the time the job
|
|
* should be run.
|
|
*/
|
|
time_t nowtimer, runtimer;
|
|
struct tm nowtime, runtime;
|
|
int hr = 0;
|
|
/* this MUST be initialized to zero for midnight/noon/teatime */
|
|
|
|
nowtimer = time(NULL);
|
|
nowtime = *localtime(&nowtimer);
|
|
|
|
runtime = nowtime;
|
|
runtime.tm_sec = 0;
|
|
runtime.tm_isdst = 0;
|
|
|
|
if (argc <= optind)
|
|
usage();
|
|
|
|
init_scanner(argc-optind, argv+optind);
|
|
|
|
switch (token()) {
|
|
case NOW: /* now is optional prefix for PLUS tree */
|
|
expect(PLUS);
|
|
case PLUS:
|
|
plus(&runtime);
|
|
break;
|
|
|
|
case NUMBER:
|
|
tod(&runtime);
|
|
month(&runtime);
|
|
break;
|
|
|
|
/* evil coding for TEATIME|NOON|MIDNIGHT - we've initialised
|
|
* hr to zero up above, then fall into this case in such a
|
|
* way so we add +12 +4 hours to it for teatime, +12 hours
|
|
* to it for noon, and nothing at all for midnight, then
|
|
* set our runtime to that hour before leaping into the
|
|
* month scanner
|
|
*/
|
|
case TEATIME:
|
|
hr += 4;
|
|
case NOON:
|
|
hr += 12;
|
|
case MIDNIGHT:
|
|
if (runtime.tm_hour >= hr) {
|
|
runtime.tm_mday++;
|
|
runtime.tm_wday++;
|
|
}
|
|
runtime.tm_hour = hr;
|
|
runtime.tm_min = 0;
|
|
token();
|
|
/* fall through to month setting */
|
|
default:
|
|
month(&runtime);
|
|
break;
|
|
} /* ugly case statement */
|
|
expect(EOF);
|
|
|
|
/* adjust for daylight savings time
|
|
*/
|
|
runtime.tm_isdst = -1;
|
|
runtimer = mktime(&runtime);
|
|
if (runtime.tm_isdst > 0) {
|
|
runtimer -= 3600;
|
|
runtimer = mktime(&runtime);
|
|
}
|
|
|
|
if (runtimer < 0)
|
|
panic("garbled time");
|
|
|
|
if (nowtimer > runtimer)
|
|
panic("Trying to travel back in time");
|
|
|
|
return runtimer;
|
|
} /* parsetime */
|