123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585 |
- /***************************************************************************
- * _ _ ____ _
- * Project ___| | | | _ \| |
- * / __| | | | |_) | |
- * | (__| |_| | _ <| |___
- * \___|\___/|_| \_\_____|
- *
- * Copyright (C) 1998 - 2016, Daniel Stenberg, <[email protected]>, et al.
- *
- * This software is licensed as described in the file COPYING, which
- * you should have received as part of this distribution. The terms
- * are also available at https://curl.haxx.se/docs/copyright.html.
- *
- * You may opt to use, copy, modify, merge, publish, distribute and/or sell
- * copies of the Software, and permit persons to whom the Software is
- * furnished to do so, under the terms of the COPYING file.
- *
- * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
- * KIND, either express or implied.
- *
- ***************************************************************************/
- /*
- A brief summary of the date string formats this parser groks:
- RFC 2616 3.3.1
- Sun, 06 Nov 1994 08:49:37 GMT ; RFC 822, updated by RFC 1123
- Sunday, 06-Nov-94 08:49:37 GMT ; RFC 850, obsoleted by RFC 1036
- Sun Nov 6 08:49:37 1994 ; ANSI C's asctime() format
- we support dates without week day name:
- 06 Nov 1994 08:49:37 GMT
- 06-Nov-94 08:49:37 GMT
- Nov 6 08:49:37 1994
- without the time zone:
- 06 Nov 1994 08:49:37
- 06-Nov-94 08:49:37
- weird order:
- 1994 Nov 6 08:49:37 (GNU date fails)
- GMT 08:49:37 06-Nov-94 Sunday
- 94 6 Nov 08:49:37 (GNU date fails)
- time left out:
- 1994 Nov 6
- 06-Nov-94
- Sun Nov 6 94
- unusual separators:
- 1994.Nov.6
- Sun/Nov/6/94/GMT
- commonly used time zone names:
- Sun, 06 Nov 1994 08:49:37 CET
- 06 Nov 1994 08:49:37 EST
- time zones specified using RFC822 style:
- Sun, 12 Sep 2004 15:05:58 -0700
- Sat, 11 Sep 2004 21:32:11 +0200
- compact numerical date strings:
- 20040912 15:05:58 -0700
- 20040911 +0200
- */
- #include "curl_setup.h"
- #ifdef HAVE_LIMITS_H
- #include <limits.h>
- #endif
- #include <curl/curl.h>
- #include "strcase.h"
- #include "warnless.h"
- #include "parsedate.h"
- const char * const Curl_wkday[] =
- {"Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"};
- static const char * const weekday[] =
- { "Monday", "Tuesday", "Wednesday", "Thursday",
- "Friday", "Saturday", "Sunday" };
- const char * const Curl_month[]=
- { "Jan", "Feb", "Mar", "Apr", "May", "Jun",
- "Jul", "Aug", "Sep", "Oct", "Nov", "Dec" };
- struct tzinfo {
- char name[5];
- int offset; /* +/- in minutes */
- };
- /*
- * parsedate()
- *
- * Returns:
- *
- * PARSEDATE_OK - a fine conversion
- * PARSEDATE_FAIL - failed to convert
- * PARSEDATE_LATER - time overflow at the far end of time_t
- * PARSEDATE_SOONER - time underflow at the low end of time_t
- */
- static int parsedate(const char *date, time_t *output);
- #define PARSEDATE_OK 0
- #define PARSEDATE_FAIL -1
- #define PARSEDATE_LATER 1
- #define PARSEDATE_SOONER 2
- /* Here's a bunch of frequently used time zone names. These were supported
- by the old getdate parser. */
- #define tDAYZONE -60 /* offset for daylight savings time */
- static const struct tzinfo tz[]= {
- {"GMT", 0}, /* Greenwich Mean */
- {"UTC", 0}, /* Universal (Coordinated) */
- {"WET", 0}, /* Western European */
- {"BST", 0 tDAYZONE}, /* British Summer */
- {"WAT", 60}, /* West Africa */
- {"AST", 240}, /* Atlantic Standard */
- {"ADT", 240 tDAYZONE}, /* Atlantic Daylight */
- {"EST", 300}, /* Eastern Standard */
- {"EDT", 300 tDAYZONE}, /* Eastern Daylight */
- {"CST", 360}, /* Central Standard */
- {"CDT", 360 tDAYZONE}, /* Central Daylight */
- {"MST", 420}, /* Mountain Standard */
- {"MDT", 420 tDAYZONE}, /* Mountain Daylight */
- {"PST", 480}, /* Pacific Standard */
- {"PDT", 480 tDAYZONE}, /* Pacific Daylight */
- {"YST", 540}, /* Yukon Standard */
- {"YDT", 540 tDAYZONE}, /* Yukon Daylight */
- {"HST", 600}, /* Hawaii Standard */
- {"HDT", 600 tDAYZONE}, /* Hawaii Daylight */
- {"CAT", 600}, /* Central Alaska */
- {"AHST", 600}, /* Alaska-Hawaii Standard */
- {"NT", 660}, /* Nome */
- {"IDLW", 720}, /* International Date Line West */
- {"CET", -60}, /* Central European */
- {"MET", -60}, /* Middle European */
- {"MEWT", -60}, /* Middle European Winter */
- {"MEST", -60 tDAYZONE}, /* Middle European Summer */
- {"CEST", -60 tDAYZONE}, /* Central European Summer */
- {"MESZ", -60 tDAYZONE}, /* Middle European Summer */
- {"FWT", -60}, /* French Winter */
- {"FST", -60 tDAYZONE}, /* French Summer */
- {"EET", -120}, /* Eastern Europe, USSR Zone 1 */
- {"WAST", -420}, /* West Australian Standard */
- {"WADT", -420 tDAYZONE}, /* West Australian Daylight */
- {"CCT", -480}, /* China Coast, USSR Zone 7 */
- {"JST", -540}, /* Japan Standard, USSR Zone 8 */
- {"EAST", -600}, /* Eastern Australian Standard */
- {"EADT", -600 tDAYZONE}, /* Eastern Australian Daylight */
- {"GST", -600}, /* Guam Standard, USSR Zone 9 */
- {"NZT", -720}, /* New Zealand */
- {"NZST", -720}, /* New Zealand Standard */
- {"NZDT", -720 tDAYZONE}, /* New Zealand Daylight */
- {"IDLE", -720}, /* International Date Line East */
- /* Next up: Military timezone names. RFC822 allowed these, but (as noted in
- RFC 1123) had their signs wrong. Here we use the correct signs to match
- actual military usage.
- */
- {"A", +1 * 60}, /* Alpha */
- {"B", +2 * 60}, /* Bravo */
- {"C", +3 * 60}, /* Charlie */
- {"D", +4 * 60}, /* Delta */
- {"E", +5 * 60}, /* Echo */
- {"F", +6 * 60}, /* Foxtrot */
- {"G", +7 * 60}, /* Golf */
- {"H", +8 * 60}, /* Hotel */
- {"I", +9 * 60}, /* India */
- /* "J", Juliet is not used as a timezone, to indicate the observer's local
- time */
- {"K", +10 * 60}, /* Kilo */
- {"L", +11 * 60}, /* Lima */
- {"M", +12 * 60}, /* Mike */
- {"N", -1 * 60}, /* November */
- {"O", -2 * 60}, /* Oscar */
- {"P", -3 * 60}, /* Papa */
- {"Q", -4 * 60}, /* Quebec */
- {"R", -5 * 60}, /* Romeo */
- {"S", -6 * 60}, /* Sierra */
- {"T", -7 * 60}, /* Tango */
- {"U", -8 * 60}, /* Uniform */
- {"V", -9 * 60}, /* Victor */
- {"W", -10 * 60}, /* Whiskey */
- {"X", -11 * 60}, /* X-ray */
- {"Y", -12 * 60}, /* Yankee */
- {"Z", 0}, /* Zulu, zero meridian, a.k.a. UTC */
- };
- /* returns:
- -1 no day
- 0 monday - 6 sunday
- */
- static int checkday(const char *check, size_t len)
- {
- int i;
- const char * const *what;
- bool found= FALSE;
- if(len > 3)
- what = &weekday[0];
- else
- what = &Curl_wkday[0];
- for(i=0; i<7; i++) {
- if(strcasecompare(check, what[0])) {
- found=TRUE;
- break;
- }
- what++;
- }
- return found?i:-1;
- }
- static int checkmonth(const char *check)
- {
- int i;
- const char * const *what;
- bool found= FALSE;
- what = &Curl_month[0];
- for(i=0; i<12; i++) {
- if(strcasecompare(check, what[0])) {
- found=TRUE;
- break;
- }
- what++;
- }
- return found?i:-1; /* return the offset or -1, no real offset is -1 */
- }
- /* return the time zone offset between GMT and the input one, in number
- of seconds or -1 if the timezone wasn't found/legal */
- static int checktz(const char *check)
- {
- unsigned int i;
- const struct tzinfo *what;
- bool found= FALSE;
- what = tz;
- for(i=0; i< sizeof(tz)/sizeof(tz[0]); i++) {
- if(strcasecompare(check, what->name)) {
- found=TRUE;
- break;
- }
- what++;
- }
- return found?what->offset*60:-1;
- }
- static void skip(const char **date)
- {
- /* skip everything that aren't letters or digits */
- while(**date && !ISALNUM(**date))
- (*date)++;
- }
- enum assume {
- DATE_MDAY,
- DATE_YEAR,
- DATE_TIME
- };
- /* this is a clone of 'struct tm' but with all fields we don't need or use
- cut out */
- struct my_tm {
- int tm_sec;
- int tm_min;
- int tm_hour;
- int tm_mday;
- int tm_mon;
- int tm_year;
- };
- /* struct tm to time since epoch in GMT time zone.
- * This is similar to the standard mktime function but for GMT only, and
- * doesn't suffer from the various bugs and portability problems that
- * some systems' implementations have.
- */
- static time_t my_timegm(struct my_tm *tm)
- {
- static const int month_days_cumulative [12] =
- { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
- int month, year, leap_days;
- if(tm->tm_year < 70)
- /* we don't support years before 1970 as they will cause this function
- to return a negative value */
- return -1;
- year = tm->tm_year + 1900;
- month = tm->tm_mon;
- if(month < 0) {
- year += (11 - month) / 12;
- month = 11 - (11 - month) % 12;
- }
- else if(month >= 12) {
- year -= month / 12;
- month = month % 12;
- }
- leap_days = year - (tm->tm_mon <= 1);
- leap_days = ((leap_days / 4) - (leap_days / 100) + (leap_days / 400)
- - (1969 / 4) + (1969 / 100) - (1969 / 400));
- return ((((time_t) (year - 1970) * 365
- + leap_days + month_days_cumulative [month] + tm->tm_mday - 1) * 24
- + tm->tm_hour) * 60 + tm->tm_min) * 60 + tm->tm_sec;
- }
- /*
- * parsedate()
- *
- * Returns:
- *
- * PARSEDATE_OK - a fine conversion
- * PARSEDATE_FAIL - failed to convert
- * PARSEDATE_LATER - time overflow at the far end of time_t
- * PARSEDATE_SOONER - time underflow at the low end of time_t
- */
- static int parsedate(const char *date, time_t *output)
- {
- time_t t = 0;
- int wdaynum=-1; /* day of the week number, 0-6 (mon-sun) */
- int monnum=-1; /* month of the year number, 0-11 */
- int mdaynum=-1; /* day of month, 1 - 31 */
- int hournum=-1;
- int minnum=-1;
- int secnum=-1;
- int yearnum=-1;
- int tzoff=-1;
- struct my_tm tm;
- enum assume dignext = DATE_MDAY;
- const char *indate = date; /* save the original pointer */
- int part = 0; /* max 6 parts */
- while(*date && (part < 6)) {
- bool found=FALSE;
- skip(&date);
- if(ISALPHA(*date)) {
- /* a name coming up */
- char buf[32]="";
- size_t len;
- if(sscanf(date, "%31[ABCDEFGHIJKLMNOPQRSTUVWXYZ"
- "abcdefghijklmnopqrstuvwxyz]", buf))
- len = strlen(buf);
- else
- len = 0;
- if(wdaynum == -1) {
- wdaynum = checkday(buf, len);
- if(wdaynum != -1)
- found = TRUE;
- }
- if(!found && (monnum == -1)) {
- monnum = checkmonth(buf);
- if(monnum != -1)
- found = TRUE;
- }
- if(!found && (tzoff == -1)) {
- /* this just must be a time zone string */
- tzoff = checktz(buf);
- if(tzoff != -1)
- found = TRUE;
- }
- if(!found)
- return PARSEDATE_FAIL; /* bad string */
- date += len;
- }
- else if(ISDIGIT(*date)) {
- /* a digit */
- int val;
- char *end;
- int len=0;
- if((secnum == -1) &&
- (3 == sscanf(date, "%02d:%02d:%02d%n",
- &hournum, &minnum, &secnum, &len))) {
- /* time stamp! */
- date += len;
- }
- else if((secnum == -1) &&
- (2 == sscanf(date, "%02d:%02d%n", &hournum, &minnum, &len))) {
- /* time stamp without seconds */
- date += len;
- secnum = 0;
- }
- else {
- long lval;
- int error;
- int old_errno;
- old_errno = ERRNO;
- SET_ERRNO(0);
- lval = strtol(date, &end, 10);
- error = ERRNO;
- if(error != old_errno)
- SET_ERRNO(old_errno);
- if(error)
- return PARSEDATE_FAIL;
- #if LONG_MAX != INT_MAX
- if((lval > (long)INT_MAX) || (lval < (long)INT_MIN))
- return PARSEDATE_FAIL;
- #endif
- val = curlx_sltosi(lval);
- if((tzoff == -1) &&
- ((end - date) == 4) &&
- (val <= 1400) &&
- (indate< date) &&
- ((date[-1] == '+' || date[-1] == '-'))) {
- /* four digits and a value less than or equal to 1400 (to take into
- account all sorts of funny time zone diffs) and it is preceded
- with a plus or minus. This is a time zone indication. 1400 is
- picked since +1300 is frequently used and +1400 is mentioned as
- an edge number in the document "ISO C 200X Proposal: Timezone
- Functions" at http://david.tribble.com/text/c0xtimezone.html If
- anyone has a more authoritative source for the exact maximum time
- zone offsets, please speak up! */
- found = TRUE;
- tzoff = (val/100 * 60 + val%100)*60;
- /* the + and - prefix indicates the local time compared to GMT,
- this we need ther reversed math to get what we want */
- tzoff = date[-1]=='+'?-tzoff:tzoff;
- }
- if(((end - date) == 8) &&
- (yearnum == -1) &&
- (monnum == -1) &&
- (mdaynum == -1)) {
- /* 8 digits, no year, month or day yet. This is YYYYMMDD */
- found = TRUE;
- yearnum = val/10000;
- monnum = (val%10000)/100-1; /* month is 0 - 11 */
- mdaynum = val%100;
- }
- if(!found && (dignext == DATE_MDAY) && (mdaynum == -1)) {
- if((val > 0) && (val<32)) {
- mdaynum = val;
- found = TRUE;
- }
- dignext = DATE_YEAR;
- }
- if(!found && (dignext == DATE_YEAR) && (yearnum == -1)) {
- yearnum = val;
- found = TRUE;
- if(yearnum < 1900) {
- if(yearnum > 70)
- yearnum += 1900;
- else
- yearnum += 2000;
- }
- if(mdaynum == -1)
- dignext = DATE_MDAY;
- }
- if(!found)
- return PARSEDATE_FAIL;
- date = end;
- }
- }
- part++;
- }
- if(-1 == secnum)
- secnum = minnum = hournum = 0; /* no time, make it zero */
- if((-1 == mdaynum) ||
- (-1 == monnum) ||
- (-1 == yearnum))
- /* lacks vital info, fail */
- return PARSEDATE_FAIL;
- #if SIZEOF_TIME_T < 5
- /* 32 bit time_t can only hold dates to the beginning of 2038 */
- if(yearnum > 2037) {
- *output = 0x7fffffff;
- return PARSEDATE_LATER;
- }
- #endif
- if(yearnum < 1970) {
- *output = 0;
- return PARSEDATE_SOONER;
- }
- if((mdaynum > 31) || (monnum > 11) ||
- (hournum > 23) || (minnum > 59) || (secnum > 60))
- return PARSEDATE_FAIL; /* clearly an illegal date */
- tm.tm_sec = secnum;
- tm.tm_min = minnum;
- tm.tm_hour = hournum;
- tm.tm_mday = mdaynum;
- tm.tm_mon = monnum;
- tm.tm_year = yearnum - 1900;
- /* my_timegm() returns a time_t. time_t is often 32 bits, even on many
- architectures that feature 64 bit 'long'.
- Some systems have 64 bit time_t and deal with years beyond 2038. However,
- even on some of the systems with 64 bit time_t mktime() returns -1 for
- dates beyond 03:14:07 UTC, January 19, 2038. (Such as AIX 5100-06)
- */
- t = my_timegm(&tm);
- /* time zone adjust (cast t to int to compare to negative one) */
- if(-1 != (int)t) {
- /* Add the time zone diff between local time zone and GMT. */
- long delta = (long)(tzoff!=-1?tzoff:0);
- if((delta>0) && (t > LONG_MAX - delta)) {
- *output = 0x7fffffff;
- return PARSEDATE_LATER; /* time_t overflow */
- }
- t += delta;
- }
- *output = t;
- return PARSEDATE_OK;
- }
- time_t curl_getdate(const char *p, const time_t *now)
- {
- time_t parsed = -1;
- int rc = parsedate(p, &parsed);
- (void)now; /* legacy argument from the past that we ignore */
- switch(rc) {
- case PARSEDATE_OK:
- case PARSEDATE_LATER:
- case PARSEDATE_SOONER:
- return parsed;
- }
- /* everything else is fail */
- return -1;
- }
- /*
- * Curl_gmtime() is a gmtime() replacement for portability. Do not use the
- * gmtime_r() or gmtime() functions anywhere else but here.
- *
- */
- CURLcode Curl_gmtime(time_t intime, struct tm *store)
- {
- const struct tm *tm;
- #ifdef HAVE_GMTIME_R
- /* thread-safe version */
- tm = (struct tm *)gmtime_r(&intime, store);
- #else
- tm = gmtime(&intime);
- if(tm)
- *store = *tm; /* copy the pointed struct to the local copy */
- #endif
- if(!tm)
- return CURLE_BAD_FUNCTION_ARGUMENT;
- return CURLE_OK;
- }
|