libarchive/archive_read_support_format_warc.c

/*-
 * Copyright (c) 2014 Sebastian Freundt
 * All rights reserved.
 *
 * 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. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * 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(S) 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, WHETHER 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.
 */

#include "archive_platform.h"
__FBSDID("$FreeBSD$");

/**
 * WARC is standardised by ISO TC46/SC4/WG12 and currently available as
 * ISO 28500:2009.
 * For the purposes of this file we used the final draft from:
 * http://bibnum.bnf.fr/warc/WARC_ISO_28500_version1_latestdraft.pdf
 *
 * Todo:
 * [ ] real-world warcs can contain resources at endpoints ending in /
 *     e.g. http://bibnum.bnf.fr/warc/
 *     if you're lucky their response contains a Content-Location: header
 *     pointing to a unix-compliant filename, in the example above it's
 *     Content-Location: http://bibnum.bnf.fr/warc/index.html
 *     however, that's not mandated and github for example doesn't follow
 *     this convention.
 *     We need a set of archive options to control what to do with
 *     entries like these, at the moment care is taken to skip them.
 *
 **/

#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#ifdef HAVE_LIMITS_H
#include <limits.h>
#endif
#ifdef HAVE_CTYPE_H
#include <ctype.h>
#endif
#ifdef HAVE_TIME_H
#include <time.h>
#endif

#include "archive.h"
#include "archive_entry.h"
#include "archive_private.h"
#include "archive_read_private.h"

typedef enum {
	WT_NONE,
	/* warcinfo */
	WT_INFO,
	/* metadata */
	WT_META,
	/* resource */
	WT_RSRC,
	/* request, unsupported */
	WT_REQ,
	/* response, unsupported */
	WT_RSP,
	/* revisit, unsupported */
	WT_RVIS,
	/* conversion, unsupported */
	WT_CONV,
	/* continuation, unsupported at the moment */
	WT_CONT,
	/* invalid type */
	LAST_WT
} warc_type_t;

typedef struct {
	size_t len;
	const char *str;
} warc_string_t;

typedef struct {
	size_t len;
	char *str;
} warc_strbuf_t;

struct warc_s {
	/* content length ahead */
	size_t cntlen;
	/* and how much we've processed so far */
	size_t cntoff;
	/* and how much we need to consume between calls */
	size_t unconsumed;

	/* string pool */
	warc_strbuf_t pool;
	/* previous version */
	unsigned int pver;
	/* stringified format name */
	struct archive_string sver;
};

static int _warc_bid(struct archive_read *a, int);
static int _warc_cleanup(struct archive_read *a);
static int _warc_read(struct archive_read*, const void**, size_t*, int64_t*);
static int _warc_skip(struct archive_read *a);
static int _warc_rdhdr(struct archive_read *a, struct archive_entry *e);

/* private routines */
static unsigned int _warc_rdver(const char buf[10], size_t bsz);
static unsigned int _warc_rdtyp(const char *buf, size_t bsz);
static warc_string_t _warc_rduri(const char *buf, size_t bsz);
static ssize_t _warc_rdlen(const char *buf, size_t bsz);
static time_t _warc_rdrtm(const char *buf, size_t bsz);
static time_t _warc_rdmtm(const char *buf, size_t bsz);
static const char *_warc_find_eoh(const char *buf, size_t bsz);
static const char *_warc_find_eol(const char *buf, size_t bsz);

int
archive_read_support_format_warc(struct archive *_a)
{
	struct archive_read *a = (struct archive_read *)_a;
	struct warc_s *w;
	int r;

	archive_check_magic(_a, ARCHIVE_READ_MAGIC,
	    ARCHIVE_STATE_NEW, "archive_read_support_format_warc");

	if ((w = calloc(1, sizeof(*w))) == NULL) {
		archive_set_error(&a->archive, ENOMEM,
		    "Can't allocate warc data");
		return (ARCHIVE_FATAL);
	}

	r = __archive_read_register_format(
		a, w, "warc",
		_warc_bid, NULL, _warc_rdhdr, _warc_read,
		_warc_skip, NULL, _warc_cleanup, NULL, NULL);

	if (r != ARCHIVE_OK) {
		free(w);
		return (r);
	}
	return (ARCHIVE_OK);
}

static int
_warc_cleanup(struct archive_read *a)
{
	struct warc_s *w = a->format->data;

	if (w->pool.len > 0U) {
		free(w->pool.str);
	}
	archive_string_free(&w->sver);
	free(w);
	a->format->data = NULL;
	return (ARCHIVE_OK);
}

static int
_warc_bid(struct archive_read *a, int best_bid)
{
	const char *hdr;
	ssize_t nrd;
	unsigned int ver;

	(void)best_bid; /* UNUSED */

	/* check first line of file, it should be a record already */
	if ((hdr = __archive_read_ahead(a, 12U, &nrd)) == NULL) {
		/* no idea what to do */
		return -1;
	} else if (nrd < 12) {
		/* nah, not for us, our magic cookie is at least 12 bytes */
		return -1;
	}

	/* otherwise snarf the record's version number */
	ver = _warc_rdver(hdr, nrd);
	if (ver < 1200U || ver > 10000U) {
		/* we only support WARC 0.12 to 1.0 */
		return -1;
	}

	/* otherwise be confident */
	return (64);
}

static int
_warc_rdhdr(struct archive_read *a, struct archive_entry *entry)
{
#define HDR_PROBE_LEN		(12U)
	struct warc_s *w = a->format->data;
	unsigned int ver;
	const char *buf;
	ssize_t nrd;
	const char *eoh;
	/* for the file name, saves some strndup()'ing */
	warc_string_t fnam;
	/* warc record type, not that we really use it a lot */
	warc_type_t ftyp;
	/* content-length+error monad */
	ssize_t cntlen;
	/* record time is the WARC-Date time we reinterpret it as ctime */
	time_t rtime;
	/* mtime is the Last-Modified time which will be the entry's mtime */
	time_t mtime;

start_over:
	/* just use read_ahead() they keep track of unconsumed
	 * bits and bobs for us; no need to put an extra shift in
	 * and reproduce that functionality here */
	buf = __archive_read_ahead(a, HDR_PROBE_LEN, &nrd);

	if (nrd < 0) {
		/* no good */
		archive_set_error(
			&a->archive, ARCHIVE_ERRNO_MISC,
			"Bad record header");
		return (ARCHIVE_FATAL);
	} else if (buf == NULL) {
		/* there should be room for at least WARC/bla\r\n
		 * must be EOF therefore */
		return (ARCHIVE_EOF);
	}
 	/* looks good so far, try and find the end of the header now */
	eoh = _warc_find_eoh(buf, nrd);
	if (eoh == NULL) {
		/* still no good, the header end might be beyond the
		 * probe we've requested, but then again who'd cram
		 * so much stuff into the header *and* be 28500-compliant */
		archive_set_error(
			&a->archive, ARCHIVE_ERRNO_MISC,
			"Bad record header");
		return (ARCHIVE_FATAL);
	}
	ver = _warc_rdver(buf, eoh - buf);
	/* we currently support WARC 0.12 to 1.0 */
	if (ver == 0U) {
		archive_set_error(
			&a->archive, ARCHIVE_ERRNO_MISC,
			"Invalid record version");
		return (ARCHIVE_FATAL);
	} else if (ver < 1200U || ver > 10000U) {
		archive_set_error(
			&a->archive, ARCHIVE_ERRNO_MISC,
			"Unsupported record version: %u.%u",
			ver / 10000, (ver % 10000) / 100);
		return (ARCHIVE_FATAL);
	}
	cntlen = _warc_rdlen(buf, eoh - buf);
	if (cntlen < 0) {
		/* nightmare!  the specs say content-length is mandatory
		 * so I don't feel overly bad stopping the reader here */
		archive_set_error(
			&a->archive, EINVAL,
			"Bad content length");
		return (ARCHIVE_FATAL);
	}
	rtime = _warc_rdrtm(buf, eoh - buf);
	if (rtime == (time_t)-1) {
		/* record time is mandatory as per WARC/1.0,
		 * so just barf here, fast and loud */
		archive_set_error(
			&a->archive, EINVAL,
			"Bad record time");
		return (ARCHIVE_FATAL);
	}

	/* let the world know we're a WARC archive */
	a->archive.archive_format = ARCHIVE_FORMAT_WARC;
	if (ver != w->pver) {
		/* stringify this entry's version */
		archive_string_sprintf(&w->sver,
			"WARC/%u.%u", ver / 10000, (ver % 10000) / 100);
		/* remember the version */
		w->pver = ver;
	}
	/* start off with the type */
	ftyp = _warc_rdtyp(buf, eoh - buf);
	/* and let future calls know about the content */
	w->cntlen = cntlen;
	w->cntoff = 0U;
	mtime = 0;/* Avoid compiling error on some platform. */

	switch (ftyp) {
	case WT_RSRC:
	case WT_RSP:
		/* only try and read the filename in the cases that are
		 * guaranteed to have one */
		fnam = _warc_rduri(buf, eoh - buf);
		/* check the last character in the URI to avoid creating
		 * directory endpoints as files, see Todo above */
		if (fnam.len == 0 || fnam.str[fnam.len - 1] == '/') {
			/* break here for now */
			fnam.len = 0U;
			fnam.str = NULL;
			break;
		}
		/* bang to our string pool, so we save a
		 * malloc()+free() roundtrip */
		if (fnam.len + 1U > w->pool.len) {
			w->pool.len = ((fnam.len + 64U) / 64U) * 64U;
			w->pool.str = realloc(w->pool.str, w->pool.len);
		}
		memcpy(w->pool.str, fnam.str, fnam.len);
		w->pool.str[fnam.len] = '\0';
		/* let no one else know about the pool, it's a secret, shhh */
		fnam.str = w->pool.str;

		/* snarf mtime or deduce from rtime
		 * this is a custom header added by our writer, it's quite
		 * hard to believe anyone else would go through with it
		 * (apart from being part of some http responses of course) */
		if ((mtime = _warc_rdmtm(buf, eoh - buf)) == (time_t)-1) {
			mtime = rtime;
		}
		break;
	default:
		fnam.len = 0U;
		fnam.str = NULL;
		break;
	}

	/* now eat some of those delicious buffer bits */
	__archive_read_consume(a, eoh - buf);

	switch (ftyp) {
	case WT_RSRC:
	case WT_RSP:
		if (fnam.len > 0U) {
			/* populate entry object */
			archive_entry_set_filetype(entry, AE_IFREG);
			archive_entry_copy_pathname(entry, fnam.str);
			archive_entry_set_size(entry, cntlen);
			archive_entry_set_perm(entry, 0644);
			/* rtime is the new ctime, mtime stays mtime */
			archive_entry_set_ctime(entry, rtime, 0L);
			archive_entry_set_mtime(entry, mtime, 0L);
			break;
		}
		/* FALLTHROUGH */
	default:
		/* consume the content and start over */
		_warc_skip(a);
		goto start_over;
	}
	return (ARCHIVE_OK);
}

static int
_warc_read(struct archive_read *a, const void **buf, size_t *bsz, int64_t *off)
{
	struct warc_s *w = a->format->data;
	const char *rab;
	ssize_t nrd;

	if (w->cntoff >= w->cntlen) {
	eof:
		/* it's our lucky day, no work, we can leave early */
		*buf = NULL;
		*bsz = 0U;
		*off = w->cntoff + 4U/*for \r\n\r\n separator*/;
		w->unconsumed = 0U;
		return (ARCHIVE_EOF);
	}

	rab = __archive_read_ahead(a, 1U, &nrd);
	if (nrd < 0) {
		*bsz = 0U;
		/* big catastrophe */
		return (int)nrd;
	} else if (nrd == 0) {
		goto eof;
	} else if ((size_t)nrd > w->cntlen - w->cntoff) {
		/* clamp to content-length */
		nrd = w->cntlen - w->cntoff;
	}
	*off = w->cntoff;
	*bsz = nrd;
	*buf = rab;

	w->cntoff += nrd;
	w->unconsumed = (size_t)nrd;
	return (ARCHIVE_OK);
}

static int
_warc_skip(struct archive_read *a)
{
	struct warc_s *w = a->format->data;

	__archive_read_consume(a, w->cntlen + 4U/*\r\n\r\n separator*/);
	w->cntlen = 0U;
	w->cntoff = 0U;
	return (ARCHIVE_OK);
}


/* private routines */
static void*
deconst(const void *c)
{
	return (char *)0x1 + (((const char *)c) - (const char *)0x1);
}

static char*
xmemmem(const char *hay, const size_t haysize,
	const char *needle, const size_t needlesize)
{
	const char *const eoh = hay + haysize;
	const char *const eon = needle + needlesize;
	const char *hp;
	const char *np;
	const char *cand;
	unsigned int hsum;
	unsigned int nsum;
	unsigned int eqp;

	/* trivial checks first
         * a 0-sized needle is defined to be found anywhere in haystack
         * then run strchr() to find a candidate in HAYSTACK (i.e. a portion
         * that happens to begin with *NEEDLE) */
	if (needlesize == 0UL) {
		return deconst(hay);
	} else if ((hay = memchr(hay, *needle, haysize)) == NULL) {
		/* trivial */
		return NULL;
	}

	/* First characters of haystack and needle are the same now. Both are
	 * guaranteed to be at least one character long.  Now computes the sum
	 * of characters values of needle together with the sum of the first
	 * needle_len characters of haystack. */
	for (hp = hay + 1U, np = needle + 1U, hsum = *hay, nsum = *hay, eqp = 1U;
	     hp < eoh && np < eon;
	     hsum ^= *hp, nsum ^= *np, eqp &= *hp == *np, hp++, np++);

	/* HP now references the (NEEDLESIZE + 1)-th character. */
	if (np < eon) {
		/* haystack is smaller than needle, :O */
		return NULL;
	} else if (eqp) {
		/* found a match */
		return deconst(hay);
	}

	/* now loop through the rest of haystack,
	 * updating the sum iteratively */
	for (cand = hay; hp < eoh; hp++) {
		hsum ^= *cand++;
		hsum ^= *hp;

		/* Since the sum of the characters is already known to be
		 * equal at that point, it is enough to check just NEEDLESIZE - 1
		 * characters for equality,
		 * also CAND is by design < HP, so no need for range checks */
		if (hsum == nsum && memcmp(cand, needle, needlesize - 1U) == 0) {
			return deconst(cand);
		}
	}
	return NULL;
}

static int
strtoi_lim(const char *str, const char **ep, int llim, int ulim)
{
	int res = 0;
	const char *sp;
	/* we keep track of the number of digits via rulim */
	int rulim;

	for (sp = str, rulim = ulim > 10 ? ulim : 10;
	     res * 10 <= ulim && rulim && *sp >= '0' && *sp <= '9';
	     sp++, rulim /= 10) {
		res *= 10;
		res += *sp - '0';
	}
	if (sp == str) {
		res = -1;
	} else if (res < llim || res > ulim) {
		res = -2;
	}
	*ep = (const char*)sp;
	return res;
}

static time_t
time_from_tm(struct tm *t)
{
#if HAVE_TIMEGM
        /* Use platform timegm() if available. */
        return (timegm(t));
#elif HAVE__MKGMTIME64
        return (_mkgmtime64(t));
#else
        /* Else use direct calculation using POSIX assumptions. */
        /* First, fix up tm_yday based on the year/month/day. */
        if (mktime(t) == (time_t)-1)
                return ((time_t)-1);
        /* Then we can compute timegm() from first principles. */
        return (t->tm_sec
            + t->tm_min * 60
            + t->tm_hour * 3600
            + t->tm_yday * 86400
            + (t->tm_year - 70) * 31536000
            + ((t->tm_year - 69) / 4) * 86400
            - ((t->tm_year - 1) / 100) * 86400
            + ((t->tm_year + 299) / 400) * 86400);
#endif
}

static time_t
xstrpisotime(const char *s, char **endptr)
{
/** like strptime() but strictly for ISO 8601 Zulu strings */
	struct tm tm;
	time_t res = (time_t)-1;

	/* make sure tm is clean */
	memset(&tm, 0, sizeof(tm));

	/* as a courtesy to our callers, and since this is a non-standard
	 * routine, we skip leading whitespace */
	while (*s == ' ' || *s == '\t')
		++s;

	/* read year */
	if ((tm.tm_year = strtoi_lim(s, &s, 1583, 4095)) < 0 || *s++ != '-') {
		goto out;
	}
	/* read month */
	if ((tm.tm_mon = strtoi_lim(s, &s, 1, 12)) < 0 || *s++ != '-') {
		goto out;
	}
	/* read day-of-month */
	if ((tm.tm_mday = strtoi_lim(s, &s, 1, 31)) < 0 || *s++ != 'T') {
		goto out;
	}
	/* read hour */
	if ((tm.tm_hour = strtoi_lim(s, &s, 0, 23)) < 0 || *s++ != ':') {
		goto out;
	}
	/* read minute */
	if ((tm.tm_min = strtoi_lim(s, &s, 0, 59)) < 0 || *s++ != ':') {
		goto out;
	}
	/* read second */
	if ((tm.tm_sec = strtoi_lim(s, &s, 0, 60)) < 0 || *s++ != 'Z') {
		goto out;
	}

	/* massage TM to fulfill some of POSIX' constraints */
	tm.tm_year -= 1900;
	tm.tm_mon--;

	/* now convert our custom tm struct to a unix stamp using UTC */
	res = time_from_tm(&tm);

out:
	if (endptr != NULL) {
		*endptr = deconst(s);
	}
	return res;
}

static unsigned int
_warc_rdver(const char *buf, size_t bsz)
{
	static const char magic[] = "WARC/";
	const char *c;
	unsigned int ver = 0U;
	unsigned int end = 0U;

	if (bsz < 12 || memcmp(buf, magic, sizeof(magic) - 1U) != 0) {
		/* buffer too small or invalid magic */
		return ver;
	}
	/* looks good so far, read the version number for a laugh */
	buf += sizeof(magic) - 1U;

	if (isdigit((unsigned char)buf[0U]) && (buf[1U] == '.') &&
	    isdigit((unsigned char)buf[2U])) {
		/* we support a maximum of 2 digits in the minor version */
		if (isdigit((unsigned char)buf[3U]))
			end = 1U;
		/* set up major version */
		ver = (buf[0U] - '0') * 10000U;
		/* set up minor version */
		if (end == 1U) {
			ver += (buf[2U] - '0') * 1000U;
			ver += (buf[3U] - '0') * 100U;
		} else
			ver += (buf[2U] - '0') * 100U;
		/*
		 * WARC below version 0.12 has a space-separated header
		 * WARC 0.12 and above terminates the version with a CRLF
		 */
		c = buf + 3U + end;
		if (ver >= 1200U) {
			if (memcmp(c, "\r\n", 2U) != 0)
				ver = 0U;
		} else if (ver < 1200U) {
			if (*c != ' ' && *c != '\t')
				ver = 0U;
		}
	}
	return ver;
}

static unsigned int
_warc_rdtyp(const char *buf, size_t bsz)
{
	static const char _key[] = "\r\nWARC-Type:";
	const char *val, *eol;

	if ((val = xmemmem(buf, bsz, _key, sizeof(_key) - 1U)) == NULL) {
		/* no bother */
		return WT_NONE;
	}
	val += sizeof(_key) - 1U;
	if ((eol = _warc_find_eol(val, buf + bsz - val)) == NULL) {
		/* no end of line */
		return WT_NONE;
	}

	/* overread whitespace */
	while (val < eol && (*val == ' ' || *val == '\t'))
		++val;

	if (val + 8U == eol) {
		if (memcmp(val, "resource", 8U) == 0)
			return WT_RSRC;
		else if (memcmp(val, "response", 8U) == 0)
			return WT_RSP;
	}
	return WT_NONE;
}

static warc_string_t
_warc_rduri(const char *buf, size_t bsz)
{
	static const char _key[] = "\r\nWARC-Target-URI:";
	const char *val, *uri, *eol, *p;
	warc_string_t res = {0U, NULL};

	if ((val = xmemmem(buf, bsz, _key, sizeof(_key) - 1U)) == NULL) {
		/* no bother */
		return res;
	}
	/* overread whitespace */
	val += sizeof(_key) - 1U;
	if ((eol = _warc_find_eol(val, buf + bsz - val)) == NULL) {
		/* no end of line */
		return res;
	}

	while (val < eol && (*val == ' ' || *val == '\t'))
		++val;

	/* overread URL designators */
	if ((uri = xmemmem(val, eol - val, "://", 3U)) == NULL) {
		/* not touching that! */
		return res;
	}

	/* spaces inside uri are not allowed, CRLF should follow */
	for (p = val; p < eol; p++) {
		if (isspace((unsigned char)*p))
			return res;
	}

	/* there must be at least space for ftp */
	if (uri < (val + 3U))
		return res;

	/* move uri to point to after :// */
	uri += 3U;

	/* now then, inspect the URI */
	if (memcmp(val, "file", 4U) == 0) {
		/* perfect, nothing left to do here */

	} else if (memcmp(val, "http", 4U) == 0 ||
		   memcmp(val, "ftp", 3U) == 0) {
		/* overread domain, and the first / */
		while (uri < eol && *uri++ != '/');
	} else {
		/* not sure what to do? best to bugger off */
		return res;
	}
	res.str = uri;
	res.len = eol - uri;
	return res;
}

static ssize_t
_warc_rdlen(const char *buf, size_t bsz)
{
	static const char _key[] = "\r\nContent-Length:";
	const char *val, *eol;
	char *on = NULL;
	long int len;

	if ((val = xmemmem(buf, bsz, _key, sizeof(_key) - 1U)) == NULL) {
		/* no bother */
		return -1;
	}
	val += sizeof(_key) - 1U;
	if ((eol = _warc_find_eol(val, buf + bsz - val)) == NULL) {
		/* no end of line */
		return -1;
	}

	/* skip leading whitespace */
	while (val < eol && (*val == ' ' || *val == '\t'))
		val++;
	/* there must be at least one digit */
	if (!isdigit((unsigned char)*val))
		return -1;
	len = strtol(val, &on, 10);
	if (on != eol) {
		/* line must end here */
		return -1;
	}

	return (size_t)len;
}

static time_t
_warc_rdrtm(const char *buf, size_t bsz)
{
	static const char _key[] = "\r\nWARC-Date:";
	const char *val, *eol;
	char *on = NULL;
	time_t res;

	if ((val = xmemmem(buf, bsz, _key, sizeof(_key) - 1U)) == NULL) {
		/* no bother */
		return (time_t)-1;
	}
	val += sizeof(_key) - 1U;
	if ((eol = _warc_find_eol(val, buf + bsz - val)) == NULL ) {
		/* no end of line */
		return -1;
	}

	/* xstrpisotime() kindly overreads whitespace for us, so use that */
	res = xstrpisotime(val, &on);
	if (on != eol) {
		/* line must end here */
		return -1;
	}
	return res;
}

static time_t
_warc_rdmtm(const char *buf, size_t bsz)
{
	static const char _key[] = "\r\nLast-Modified:";
	const char *val, *eol;
	char *on = NULL;
	time_t res;

	if ((val = xmemmem(buf, bsz, _key, sizeof(_key) - 1U)) == NULL) {
		/* no bother */
		return (time_t)-1;
	}
	val += sizeof(_key) - 1U;
	if ((eol = _warc_find_eol(val, buf + bsz - val)) == NULL ) {
		/* no end of line */
		return -1;
	}

	/* xstrpisotime() kindly overreads whitespace for us, so use that */
	res = xstrpisotime(val, &on);
	if (on != eol) {
		/* line must end here */
		return -1;
	}
	return res;
}

static const char*
_warc_find_eoh(const char *buf, size_t bsz)
{
	static const char _marker[] = "\r\n\r\n";
	const char *hit = xmemmem(buf, bsz, _marker, sizeof(_marker) - 1U);

	if (hit != NULL) {
		hit += sizeof(_marker) - 1U;
	}
	return hit;
}

static const char*
_warc_find_eol(const char *buf, size_t bsz)
{
	static const char _marker[] = "\r\n";
	const char *hit = xmemmem(buf, bsz, _marker, sizeof(_marker) - 1U);

	return hit;
}
/* archive_read_support_format_warc.c ends here */