http.c revision 241840 
1/*-
2 * Copyright (c) 2000-2011 Dag-Erling Sm��rgrav
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer
10 *    in this position and unchanged.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote products
15 *    derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/lib/libfetch/http.c 241840 2012-10-22 03:00:10Z eadler $");
31
32/*
33 * The following copyright applies to the base64 code:
34 *
35 *-
36 * Copyright 1997 Massachusetts Institute of Technology
37 *
38 * Permission to use, copy, modify, and distribute this software and
39 * its documentation for any purpose and without fee is hereby
40 * granted, provided that both the above copyright notice and this
41 * permission notice appear in all copies, that both the above
42 * copyright notice and this permission notice appear in all
43 * supporting documentation, and that the name of M.I.T. not be used
44 * in advertising or publicity pertaining to distribution of the
45 * software without specific, written prior permission.  M.I.T. makes
46 * no representations about the suitability of this software for any
47 * purpose.  It is provided "as is" without express or implied
48 * warranty.
49 *
50 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
51 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
52 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
53 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
54 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
55 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
56 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
57 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
58 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
59 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
60 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61 * SUCH DAMAGE.
62 */
63
64#include <sys/param.h>
65#include <sys/socket.h>
66#include <sys/time.h>
67
68#include <ctype.h>
69#include <err.h>
70#include <errno.h>
71#include <locale.h>
72#include <netdb.h>
73#include <stdarg.h>
74#include <stdio.h>
75#include <stdlib.h>
76#include <string.h>
77#include <time.h>
78#include <unistd.h>
79
80#ifdef WITH_SSL
81#include <openssl/md5.h>
82#define MD5Init(c) MD5_Init(c)
83#define MD5Update(c, data, len) MD5_Update(c, data, len)
84#define MD5Final(md, c) MD5_Final(md, c)
85#else
86#include <md5.h>
87#endif
88
89#include <netinet/in.h>
90#include <netinet/tcp.h>
91
92#include "fetch.h"
93#include "common.h"
94#include "httperr.h"
95
96/* Maximum number of redirects to follow */
97#define MAX_REDIRECT 20
98
99/* Symbolic names for reply codes we care about */
100#define HTTP_OK			200
101#define HTTP_PARTIAL		206
102#define HTTP_MOVED_PERM		301
103#define HTTP_MOVED_TEMP		302
104#define HTTP_SEE_OTHER		303
105#define HTTP_NOT_MODIFIED	304
106#define HTTP_TEMP_REDIRECT	307
107#define HTTP_PERM_REDIRECT	308
108#define HTTP_NEED_AUTH		401
109#define HTTP_NEED_PROXY_AUTH	407
110#define HTTP_BAD_RANGE		416
111#define HTTP_PROTOCOL_ERROR	999
112
113#define HTTP_REDIRECT(xyz) ((xyz) == HTTP_MOVED_PERM \
114			    || (xyz) == HTTP_MOVED_TEMP \
115			    || (xyz) == HTTP_TEMP_REDIRECT \
116			    || (xyz) == HTTP_SEE_OTHER)
117
118#define HTTP_ERROR(xyz) ((xyz) > 400 && (xyz) < 599)
119
120
121/*****************************************************************************
122 * I/O functions for decoding chunked streams
123 */
124
125struct httpio
126{
127	conn_t		*conn;		/* connection */
128	int		 chunked;	/* chunked mode */
129	char		*buf;		/* chunk buffer */
130	size_t		 bufsize;	/* size of chunk buffer */
131	ssize_t		 buflen;	/* amount of data currently in buffer */
132	int		 bufpos;	/* current read offset in buffer */
133	int		 eof;		/* end-of-file flag */
134	int		 error;		/* error flag */
135	size_t		 chunksize;	/* remaining size of current chunk */
136#ifndef NDEBUG
137	size_t		 total;
138#endif
139};
140
141/*
142 * Get next chunk header
143 */
144static int
145http_new_chunk(struct httpio *io)
146{
147	char *p;
148
149	if (fetch_getln(io->conn) == -1)
150		return (-1);
151
152	if (io->conn->buflen < 2 || !isxdigit((unsigned char)*io->conn->buf))
153		return (-1);
154
155	for (p = io->conn->buf; *p && !isspace((unsigned char)*p); ++p) {
156		if (*p == ';')
157			break;
158		if (!isxdigit((unsigned char)*p))
159			return (-1);
160		if (isdigit((unsigned char)*p)) {
161			io->chunksize = io->chunksize * 16 +
162			    *p - '0';
163		} else {
164			io->chunksize = io->chunksize * 16 +
165			    10 + tolower((unsigned char)*p) - 'a';
166		}
167	}
168
169#ifndef NDEBUG
170	if (fetchDebug) {
171		io->total += io->chunksize;
172		if (io->chunksize == 0)
173			fprintf(stderr, "%s(): end of last chunk\n", __func__);
174		else
175			fprintf(stderr, "%s(): new chunk: %lu (%lu)\n",
176			    __func__, (unsigned long)io->chunksize,
177			    (unsigned long)io->total);
178	}
179#endif
180
181	return (io->chunksize);
182}
183
184/*
185 * Grow the input buffer to at least len bytes
186 */
187static inline int
188http_growbuf(struct httpio *io, size_t len)
189{
190	char *tmp;
191
192	if (io->bufsize >= len)
193		return (0);
194
195	if ((tmp = realloc(io->buf, len)) == NULL)
196		return (-1);
197	io->buf = tmp;
198	io->bufsize = len;
199	return (0);
200}
201
202/*
203 * Fill the input buffer, do chunk decoding on the fly
204 */
205static int
206http_fillbuf(struct httpio *io, size_t len)
207{
208	ssize_t nbytes;
209
210	if (io->error)
211		return (-1);
212	if (io->eof)
213		return (0);
214
215	if (io->chunked == 0) {
216		if (http_growbuf(io, len) == -1)
217			return (-1);
218		if ((nbytes = fetch_read(io->conn, io->buf, len)) == -1) {
219			io->error = errno;
220			return (-1);
221		}
222		io->buflen = nbytes;
223		io->bufpos = 0;
224		return (io->buflen);
225	}
226
227	if (io->chunksize == 0) {
228		switch (http_new_chunk(io)) {
229		case -1:
230			io->error = 1;
231			return (-1);
232		case 0:
233			io->eof = 1;
234			return (0);
235		}
236	}
237
238	if (len > io->chunksize)
239		len = io->chunksize;
240	if (http_growbuf(io, len) == -1)
241		return (-1);
242	if ((nbytes = fetch_read(io->conn, io->buf, len)) == -1) {
243		io->error = errno;
244		return (-1);
245	}
246	io->buflen = nbytes;
247	io->chunksize -= io->buflen;
248
249	if (io->chunksize == 0) {
250		char endl[2];
251
252		if (fetch_read(io->conn, endl, 2) != 2 ||
253		    endl[0] != '\r' || endl[1] != '\n')
254			return (-1);
255	}
256
257	io->bufpos = 0;
258
259	return (io->buflen);
260}
261
262/*
263 * Read function
264 */
265static int
266http_readfn(void *v, char *buf, int len)
267{
268	struct httpio *io = (struct httpio *)v;
269	int l, pos;
270
271	if (io->error)
272		return (-1);
273	if (io->eof)
274		return (0);
275
276	for (pos = 0; len > 0; pos += l, len -= l) {
277		/* empty buffer */
278		if (!io->buf || io->bufpos == io->buflen)
279			if (http_fillbuf(io, len) < 1)
280				break;
281		l = io->buflen - io->bufpos;
282		if (len < l)
283			l = len;
284		memcpy(buf + pos, io->buf + io->bufpos, l);
285		io->bufpos += l;
286	}
287
288	if (!pos && io->error) {
289		if (io->error == EINTR)
290			io->error = 0;
291		return (-1);
292	}
293	return (pos);
294}
295
296/*
297 * Write function
298 */
299static int
300http_writefn(void *v, const char *buf, int len)
301{
302	struct httpio *io = (struct httpio *)v;
303
304	return (fetch_write(io->conn, buf, len));
305}
306
307/*
308 * Close function
309 */
310static int
311http_closefn(void *v)
312{
313	struct httpio *io = (struct httpio *)v;
314	int r;
315
316	r = fetch_close(io->conn);
317	if (io->buf)
318		free(io->buf);
319	free(io);
320	return (r);
321}
322
323/*
324 * Wrap a file descriptor up
325 */
326static FILE *
327http_funopen(conn_t *conn, int chunked)
328{
329	struct httpio *io;
330	FILE *f;
331
332	if ((io = calloc(1, sizeof(*io))) == NULL) {
333		fetch_syserr();
334		return (NULL);
335	}
336	io->conn = conn;
337	io->chunked = chunked;
338	f = funopen(io, http_readfn, http_writefn, NULL, http_closefn);
339	if (f == NULL) {
340		fetch_syserr();
341		free(io);
342		return (NULL);
343	}
344	return (f);
345}
346
347
348/*****************************************************************************
349 * Helper functions for talking to the server and parsing its replies
350 */
351
352/* Header types */
353typedef enum {
354	hdr_syserror = -2,
355	hdr_error = -1,
356	hdr_end = 0,
357	hdr_unknown = 1,
358	hdr_content_length,
359	hdr_content_range,
360	hdr_last_modified,
361	hdr_location,
362	hdr_transfer_encoding,
363	hdr_www_authenticate,
364	hdr_proxy_authenticate,
365} hdr_t;
366
367/* Names of interesting headers */
368static struct {
369	hdr_t		 num;
370	const char	*name;
371} hdr_names[] = {
372	{ hdr_content_length,		"Content-Length" },
373	{ hdr_content_range,		"Content-Range" },
374	{ hdr_last_modified,		"Last-Modified" },
375	{ hdr_location,			"Location" },
376	{ hdr_transfer_encoding,	"Transfer-Encoding" },
377	{ hdr_www_authenticate,		"WWW-Authenticate" },
378	{ hdr_proxy_authenticate,	"Proxy-Authenticate" },
379	{ hdr_unknown,			NULL },
380};
381
382/*
383 * Send a formatted line; optionally echo to terminal
384 */
385static int
386http_cmd(conn_t *conn, const char *fmt, ...)
387{
388	va_list ap;
389	size_t len;
390	char *msg;
391	int r;
392
393	va_start(ap, fmt);
394	len = vasprintf(&msg, fmt, ap);
395	va_end(ap);
396
397	if (msg == NULL) {
398		errno = ENOMEM;
399		fetch_syserr();
400		return (-1);
401	}
402
403	r = fetch_putln(conn, msg, len);
404	free(msg);
405
406	if (r == -1) {
407		fetch_syserr();
408		return (-1);
409	}
410
411	return (0);
412}
413
414/*
415 * Get and parse status line
416 */
417static int
418http_get_reply(conn_t *conn)
419{
420	char *p;
421
422	if (fetch_getln(conn) == -1)
423		return (-1);
424	/*
425	 * A valid status line looks like "HTTP/m.n xyz reason" where m
426	 * and n are the major and minor protocol version numbers and xyz
427	 * is the reply code.
428	 * Unfortunately, there are servers out there (NCSA 1.5.1, to name
429	 * just one) that do not send a version number, so we can't rely
430	 * on finding one, but if we do, insist on it being 1.0 or 1.1.
431	 * We don't care about the reason phrase.
432	 */
433	if (strncmp(conn->buf, "HTTP", 4) != 0)
434		return (HTTP_PROTOCOL_ERROR);
435	p = conn->buf + 4;
436	if (*p == '/') {
437		if (p[1] != '1' || p[2] != '.' || (p[3] != '0' && p[3] != '1'))
438			return (HTTP_PROTOCOL_ERROR);
439		p += 4;
440	}
441	if (*p != ' ' ||
442	    !isdigit((unsigned char)p[1]) ||
443	    !isdigit((unsigned char)p[2]) ||
444	    !isdigit((unsigned char)p[3]))
445		return (HTTP_PROTOCOL_ERROR);
446
447	conn->err = (p[1] - '0') * 100 + (p[2] - '0') * 10 + (p[3] - '0');
448	return (conn->err);
449}
450
451/*
452 * Check a header; if the type matches the given string, return a pointer
453 * to the beginning of the value.
454 */
455static const char *
456http_match(const char *str, const char *hdr)
457{
458	while (*str && *hdr &&
459	    tolower((unsigned char)*str++) == tolower((unsigned char)*hdr++))
460		/* nothing */;
461	if (*str || *hdr != ':')
462		return (NULL);
463	while (*hdr && isspace((unsigned char)*++hdr))
464		/* nothing */;
465	return (hdr);
466}
467
468
469/*
470 * Get the next header and return the appropriate symbolic code.  We
471 * need to read one line ahead for checking for a continuation line
472 * belonging to the current header (continuation lines start with
473 * white space).
474 *
475 * We get called with a fresh line already in the conn buffer, either
476 * from the previous http_next_header() invocation, or, the first
477 * time, from a fetch_getln() performed by our caller.
478 *
479 * This stops when we encounter an empty line (we dont read beyond the header
480 * area).
481 *
482 * Note that the "headerbuf" is just a place to return the result. Its
483 * contents are not used for the next call. This means that no cleanup
484 * is needed when ie doing another connection, just call the cleanup when
485 * fully done to deallocate memory.
486 */
487
488/* Limit the max number of continuation lines to some reasonable value */
489#define HTTP_MAX_CONT_LINES 10
490
491/* Place into which to build a header from one or several lines */
492typedef struct {
493	char	*buf;		/* buffer */
494	size_t	 bufsize;	/* buffer size */
495	size_t	 buflen;	/* length of buffer contents */
496} http_headerbuf_t;
497
498static void
499init_http_headerbuf(http_headerbuf_t *buf)
500{
501	buf->buf = NULL;
502	buf->bufsize = 0;
503	buf->buflen = 0;
504}
505
506static void
507clean_http_headerbuf(http_headerbuf_t *buf)
508{
509	if (buf->buf)
510		free(buf->buf);
511	init_http_headerbuf(buf);
512}
513
514/* Remove whitespace at the end of the buffer */
515static void
516http_conn_trimright(conn_t *conn)
517{
518	while (conn->buflen &&
519	       isspace((unsigned char)conn->buf[conn->buflen - 1]))
520		conn->buflen--;
521	conn->buf[conn->buflen] = '\0';
522}
523
524static hdr_t
525http_next_header(conn_t *conn, http_headerbuf_t *hbuf, const char **p)
526{
527	unsigned int i, len;
528
529	/*
530	 * Have to do the stripping here because of the first line. So
531	 * it's done twice for the subsequent lines. No big deal
532	 */
533	http_conn_trimright(conn);
534	if (conn->buflen == 0)
535		return (hdr_end);
536
537	/* Copy the line to the headerbuf */
538	if (hbuf->bufsize < conn->buflen + 1) {
539		if ((hbuf->buf = realloc(hbuf->buf, conn->buflen + 1)) == NULL)
540			return (hdr_syserror);
541		hbuf->bufsize = conn->buflen + 1;
542	}
543	strcpy(hbuf->buf, conn->buf);
544	hbuf->buflen = conn->buflen;
545
546	/*
547	 * Fetch possible continuation lines. Stop at 1st non-continuation
548	 * and leave it in the conn buffer
549	 */
550	for (i = 0; i < HTTP_MAX_CONT_LINES; i++) {
551		if (fetch_getln(conn) == -1)
552			return (hdr_syserror);
553
554		/*
555		 * Note: we carry on the idea from the previous version
556		 * that a pure whitespace line is equivalent to an empty
557		 * one (so it's not continuation and will be handled when
558		 * we are called next)
559		 */
560		http_conn_trimright(conn);
561		if (conn->buf[0] != ' ' && conn->buf[0] != "\t"[0])
562			break;
563
564		/* Got a continuation line. Concatenate to previous */
565		len = hbuf->buflen + conn->buflen;
566		if (hbuf->bufsize < len + 1) {
567			len *= 2;
568			if ((hbuf->buf = realloc(hbuf->buf, len + 1)) == NULL)
569				return (hdr_syserror);
570			hbuf->bufsize = len + 1;
571		}
572		strcpy(hbuf->buf + hbuf->buflen, conn->buf);
573		hbuf->buflen += conn->buflen;
574	}
575
576	/*
577	 * We could check for malformed headers but we don't really care.
578	 * A valid header starts with a token immediately followed by a
579	 * colon; a token is any sequence of non-control, non-whitespace
580	 * characters except "()<>@,;:\\\"{}".
581	 */
582	for (i = 0; hdr_names[i].num != hdr_unknown; i++)
583		if ((*p = http_match(hdr_names[i].name, hbuf->buf)) != NULL)
584			return (hdr_names[i].num);
585
586	return (hdr_unknown);
587}
588
589/**************************
590 * [Proxy-]Authenticate header parsing
591 */
592
593/*
594 * Read doublequote-delimited string into output buffer obuf (allocated
595 * by caller, whose responsibility it is to ensure that it's big enough)
596 * cp points to the first char after the initial '"'
597 * Handles \ quoting
598 * Returns pointer to the first char after the terminating double quote, or
599 * NULL for error.
600 */
601static const char *
602http_parse_headerstring(const char *cp, char *obuf)
603{
604	for (;;) {
605		switch (*cp) {
606		case 0: /* Unterminated string */
607			*obuf = 0;
608			return (NULL);
609		case '"': /* Ending quote */
610			*obuf = 0;
611			return (++cp);
612		case '\\':
613			if (*++cp == 0) {
614				*obuf = 0;
615				return (NULL);
616			}
617			/* FALLTHROUGH */
618		default:
619			*obuf++ = *cp++;
620		}
621	}
622}
623
624/* Http auth challenge schemes */
625typedef enum {HTTPAS_UNKNOWN, HTTPAS_BASIC,HTTPAS_DIGEST} http_auth_schemes_t;
626
627/* Data holder for a Basic or Digest challenge. */
628typedef struct {
629	http_auth_schemes_t scheme;
630	char	*realm;
631	char	*qop;
632	char	*nonce;
633	char	*opaque;
634	char	*algo;
635	int	 stale;
636	int	 nc; /* Nonce count */
637} http_auth_challenge_t;
638
639static void
640init_http_auth_challenge(http_auth_challenge_t *b)
641{
642	b->scheme = HTTPAS_UNKNOWN;
643	b->realm = b->qop = b->nonce = b->opaque = b->algo = NULL;
644	b->stale = b->nc = 0;
645}
646
647static void
648clean_http_auth_challenge(http_auth_challenge_t *b)
649{
650	if (b->realm)
651		free(b->realm);
652	if (b->qop)
653		free(b->qop);
654	if (b->nonce)
655		free(b->nonce);
656	if (b->opaque)
657		free(b->opaque);
658	if (b->algo)
659		free(b->algo);
660	init_http_auth_challenge(b);
661}
662
663/* Data holder for an array of challenges offered in an http response. */
664#define MAX_CHALLENGES 10
665typedef struct {
666	http_auth_challenge_t *challenges[MAX_CHALLENGES];
667	int	count; /* Number of parsed challenges in the array */
668	int	valid; /* We did parse an authenticate header */
669} http_auth_challenges_t;
670
671static void
672init_http_auth_challenges(http_auth_challenges_t *cs)
673{
674	int i;
675	for (i = 0; i < MAX_CHALLENGES; i++)
676		cs->challenges[i] = NULL;
677	cs->count = cs->valid = 0;
678}
679
680static void
681clean_http_auth_challenges(http_auth_challenges_t *cs)
682{
683	int i;
684	/* We rely on non-zero pointers being allocated, not on the count */
685	for (i = 0; i < MAX_CHALLENGES; i++) {
686		if (cs->challenges[i] != NULL) {
687			clean_http_auth_challenge(cs->challenges[i]);
688			free(cs->challenges[i]);
689		}
690	}
691	init_http_auth_challenges(cs);
692}
693
694/*
695 * Enumeration for lexical elements. Separators will be returned as their own
696 * ascii value
697 */
698typedef enum {HTTPHL_WORD=256, HTTPHL_STRING=257, HTTPHL_END=258,
699	      HTTPHL_ERROR = 259} http_header_lex_t;
700
701/*
702 * Determine what kind of token comes next and return possible value
703 * in buf, which is supposed to have been allocated big enough by
704 * caller. Advance input pointer and return element type.
705 */
706static int
707http_header_lex(const char **cpp, char *buf)
708{
709	size_t l;
710	/* Eat initial whitespace */
711	*cpp += strspn(*cpp, " \t");
712	if (**cpp == 0)
713		return (HTTPHL_END);
714
715	/* Separator ? */
716	if (**cpp == ',' || **cpp == '=')
717		return (*((*cpp)++));
718
719	/* String ? */
720	if (**cpp == '"') {
721		*cpp = http_parse_headerstring(++*cpp, buf);
722		if (*cpp == NULL)
723			return (HTTPHL_ERROR);
724		return (HTTPHL_STRING);
725	}
726
727	/* Read other token, until separator or whitespace */
728	l = strcspn(*cpp, " \t,=");
729	memcpy(buf, *cpp, l);
730	buf[l] = 0;
731	*cpp += l;
732	return (HTTPHL_WORD);
733}
734
735/*
736 * Read challenges from http xxx-authenticate header and accumulate them
737 * in the challenges list structure.
738 *
739 * Headers with multiple challenges are specified by rfc2617, but
740 * servers (ie: squid) often send them in separate headers instead,
741 * which in turn is forbidden by the http spec (multiple headers with
742 * the same name are only allowed for pure comma-separated lists, see
743 * rfc2616 sec 4.2).
744 *
745 * We support both approaches anyway
746 */
747static int
748http_parse_authenticate(const char *cp, http_auth_challenges_t *cs)
749{
750	int ret = -1;
751	http_header_lex_t lex;
752	char *key = malloc(strlen(cp) + 1);
753	char *value = malloc(strlen(cp) + 1);
754	char *buf = malloc(strlen(cp) + 1);
755
756	if (key == NULL || value == NULL || buf == NULL) {
757		fetch_syserr();
758		goto out;
759	}
760
761	/* In any case we've seen the header and we set the valid bit */
762	cs->valid = 1;
763
764	/* Need word first */
765	lex = http_header_lex(&cp, key);
766	if (lex != HTTPHL_WORD)
767		goto out;
768
769	/* Loop on challenges */
770	for (; cs->count < MAX_CHALLENGES; cs->count++) {
771		cs->challenges[cs->count] =
772			malloc(sizeof(http_auth_challenge_t));
773		if (cs->challenges[cs->count] == NULL) {
774			fetch_syserr();
775			goto out;
776		}
777		init_http_auth_challenge(cs->challenges[cs->count]);
778		if (!strcasecmp(key, "basic")) {
779			cs->challenges[cs->count]->scheme = HTTPAS_BASIC;
780		} else if (!strcasecmp(key, "digest")) {
781			cs->challenges[cs->count]->scheme = HTTPAS_DIGEST;
782		} else {
783			cs->challenges[cs->count]->scheme = HTTPAS_UNKNOWN;
784			/*
785			 * Continue parsing as basic or digest may
786			 * follow, and the syntax is the same for
787			 * all. We'll just ignore this one when
788			 * looking at the list
789			 */
790		}
791
792		/* Loop on attributes */
793		for (;;) {
794			/* Key */
795			lex = http_header_lex(&cp, key);
796			if (lex != HTTPHL_WORD)
797				goto out;
798
799			/* Equal sign */
800			lex = http_header_lex(&cp, buf);
801			if (lex != '=')
802				goto out;
803
804			/* Value */
805			lex = http_header_lex(&cp, value);
806			if (lex != HTTPHL_WORD && lex != HTTPHL_STRING)
807				goto out;
808
809			if (!strcasecmp(key, "realm"))
810				cs->challenges[cs->count]->realm =
811					strdup(value);
812			else if (!strcasecmp(key, "qop"))
813				cs->challenges[cs->count]->qop =
814					strdup(value);
815			else if (!strcasecmp(key, "nonce"))
816				cs->challenges[cs->count]->nonce =
817					strdup(value);
818			else if (!strcasecmp(key, "opaque"))
819				cs->challenges[cs->count]->opaque =
820					strdup(value);
821			else if (!strcasecmp(key, "algorithm"))
822				cs->challenges[cs->count]->algo =
823					strdup(value);
824			else if (!strcasecmp(key, "stale"))
825				cs->challenges[cs->count]->stale =
826					strcasecmp(value, "no");
827			/* Else ignore unknown attributes */
828
829			/* Comma or Next challenge or End */
830			lex = http_header_lex(&cp, key);
831			/*
832			 * If we get a word here, this is the beginning of the
833			 * next challenge. Break the attributes loop
834			 */
835			if (lex == HTTPHL_WORD)
836				break;
837
838			if (lex == HTTPHL_END) {
839				/* End while looking for ',' is normal exit */
840				cs->count++;
841				ret = 0;
842				goto out;
843			}
844			/* Anything else is an error */
845			if (lex != ',')
846				goto out;
847
848		} /* End attributes loop */
849	} /* End challenge loop */
850
851	/*
852	 * Challenges max count exceeded. This really can't happen
853	 * with normal data, something's fishy -> error
854	 */
855
856out:
857	if (key)
858		free(key);
859	if (value)
860		free(value);
861	if (buf)
862		free(buf);
863	return (ret);
864}
865
866
867/*
868 * Parse a last-modified header
869 */
870static int
871http_parse_mtime(const char *p, time_t *mtime)
872{
873	char locale[64], *r;
874	struct tm tm;
875
876	strncpy(locale, setlocale(LC_TIME, NULL), sizeof(locale));
877	setlocale(LC_TIME, "C");
878	r = strptime(p, "%a, %d %b %Y %H:%M:%S GMT", &tm);
879	/* XXX should add support for date-2 and date-3 */
880	setlocale(LC_TIME, locale);
881	if (r == NULL)
882		return (-1);
883	DEBUG(fprintf(stderr, "last modified: [%04d-%02d-%02d "
884		  "%02d:%02d:%02d]\n",
885		  tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday,
886		  tm.tm_hour, tm.tm_min, tm.tm_sec));
887	*mtime = timegm(&tm);
888	return (0);
889}
890
891/*
892 * Parse a content-length header
893 */
894static int
895http_parse_length(const char *p, off_t *length)
896{
897	off_t len;
898
899	for (len = 0; *p && isdigit((unsigned char)*p); ++p)
900		len = len * 10 + (*p - '0');
901	if (*p)
902		return (-1);
903	DEBUG(fprintf(stderr, "content length: [%lld]\n",
904	    (long long)len));
905	*length = len;
906	return (0);
907}
908
909/*
910 * Parse a content-range header
911 */
912static int
913http_parse_range(const char *p, off_t *offset, off_t *length, off_t *size)
914{
915	off_t first, last, len;
916
917	if (strncasecmp(p, "bytes ", 6) != 0)
918		return (-1);
919	p += 6;
920	if (*p == '*') {
921		first = last = -1;
922		++p;
923	} else {
924		for (first = 0; *p && isdigit((unsigned char)*p); ++p)
925			first = first * 10 + *p - '0';
926		if (*p != '-')
927			return (-1);
928		for (last = 0, ++p; *p && isdigit((unsigned char)*p); ++p)
929			last = last * 10 + *p - '0';
930	}
931	if (first > last || *p != '/')
932		return (-1);
933	for (len = 0, ++p; *p && isdigit((unsigned char)*p); ++p)
934		len = len * 10 + *p - '0';
935	if (*p || len < last - first + 1)
936		return (-1);
937	if (first == -1) {
938		DEBUG(fprintf(stderr, "content range: [*/%lld]\n",
939		    (long long)len));
940		*length = 0;
941	} else {
942		DEBUG(fprintf(stderr, "content range: [%lld-%lld/%lld]\n",
943		    (long long)first, (long long)last, (long long)len));
944		*length = last - first + 1;
945	}
946	*offset = first;
947	*size = len;
948	return (0);
949}
950
951
952/*****************************************************************************
953 * Helper functions for authorization
954 */
955
956/*
957 * Base64 encoding
958 */
959static char *
960http_base64(const char *src)
961{
962	static const char base64[] =
963	    "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
964	    "abcdefghijklmnopqrstuvwxyz"
965	    "0123456789+/";
966	char *str, *dst;
967	size_t l;
968	int t, r;
969
970	l = strlen(src);
971	if ((str = malloc(((l + 2) / 3) * 4 + 1)) == NULL)
972		return (NULL);
973	dst = str;
974	r = 0;
975
976	while (l >= 3) {
977		t = (src[0] << 16) | (src[1] << 8) | src[2];
978		dst[0] = base64[(t >> 18) & 0x3f];
979		dst[1] = base64[(t >> 12) & 0x3f];
980		dst[2] = base64[(t >> 6) & 0x3f];
981		dst[3] = base64[(t >> 0) & 0x3f];
982		src += 3; l -= 3;
983		dst += 4; r += 4;
984	}
985
986	switch (l) {
987	case 2:
988		t = (src[0] << 16) | (src[1] << 8);
989		dst[0] = base64[(t >> 18) & 0x3f];
990		dst[1] = base64[(t >> 12) & 0x3f];
991		dst[2] = base64[(t >> 6) & 0x3f];
992		dst[3] = '=';
993		dst += 4;
994		r += 4;
995		break;
996	case 1:
997		t = src[0] << 16;
998		dst[0] = base64[(t >> 18) & 0x3f];
999		dst[1] = base64[(t >> 12) & 0x3f];
1000		dst[2] = dst[3] = '=';
1001		dst += 4;
1002		r += 4;
1003		break;
1004	case 0:
1005		break;
1006	}
1007
1008	*dst = 0;
1009	return (str);
1010}
1011
1012
1013/*
1014 * Extract authorization parameters from environment value.
1015 * The value is like scheme:realm:user:pass
1016 */
1017typedef struct {
1018	char	*scheme;
1019	char	*realm;
1020	char	*user;
1021	char	*password;
1022} http_auth_params_t;
1023
1024static void
1025init_http_auth_params(http_auth_params_t *s)
1026{
1027	s->scheme = s->realm = s->user = s->password = 0;
1028}
1029
1030static void
1031clean_http_auth_params(http_auth_params_t *s)
1032{
1033	if (s->scheme)
1034		free(s->scheme);
1035	if (s->realm)
1036		free(s->realm);
1037	if (s->user)
1038		free(s->user);
1039	if (s->password)
1040		free(s->password);
1041	init_http_auth_params(s);
1042}
1043
1044static int
1045http_authfromenv(const char *p, http_auth_params_t *parms)
1046{
1047	int ret = -1;
1048	char *v, *ve;
1049	char *str = strdup(p);
1050
1051	if (str == NULL) {
1052		fetch_syserr();
1053		return (-1);
1054	}
1055	v = str;
1056
1057	if ((ve = strchr(v, ':')) == NULL)
1058		goto out;
1059
1060	*ve = 0;
1061	if ((parms->scheme = strdup(v)) == NULL) {
1062		fetch_syserr();
1063		goto out;
1064	}
1065	v = ve + 1;
1066
1067	if ((ve = strchr(v, ':')) == NULL)
1068		goto out;
1069
1070	*ve = 0;
1071	if ((parms->realm = strdup(v)) == NULL) {
1072		fetch_syserr();
1073		goto out;
1074	}
1075	v = ve + 1;
1076
1077	if ((ve = strchr(v, ':')) == NULL)
1078		goto out;
1079
1080	*ve = 0;
1081	if ((parms->user = strdup(v)) == NULL) {
1082		fetch_syserr();
1083		goto out;
1084	}
1085	v = ve + 1;
1086
1087
1088	if ((parms->password = strdup(v)) == NULL) {
1089		fetch_syserr();
1090		goto out;
1091	}
1092	ret = 0;
1093out:
1094	if (ret == -1)
1095		clean_http_auth_params(parms);
1096	if (str)
1097		free(str);
1098	return (ret);
1099}
1100
1101
1102/*
1103 * Digest response: the code to compute the digest is taken from the
1104 * sample implementation in RFC2616
1105 */
1106#define IN const
1107#define OUT
1108
1109#define HASHLEN 16
1110typedef char HASH[HASHLEN];
1111#define HASHHEXLEN 32
1112typedef char HASHHEX[HASHHEXLEN+1];
1113
1114static const char *hexchars = "0123456789abcdef";
1115static void
1116CvtHex(IN HASH Bin, OUT HASHHEX Hex)
1117{
1118	unsigned short i;
1119	unsigned char j;
1120
1121	for (i = 0; i < HASHLEN; i++) {
1122		j = (Bin[i] >> 4) & 0xf;
1123		Hex[i*2] = hexchars[j];
1124		j = Bin[i] & 0xf;
1125		Hex[i*2+1] = hexchars[j];
1126	};
1127	Hex[HASHHEXLEN] = '\0';
1128};
1129
1130/* calculate H(A1) as per spec */
1131static void
1132DigestCalcHA1(
1133	IN char * pszAlg,
1134	IN char * pszUserName,
1135	IN char * pszRealm,
1136	IN char * pszPassword,
1137	IN char * pszNonce,
1138	IN char * pszCNonce,
1139	OUT HASHHEX SessionKey
1140	)
1141{
1142	MD5_CTX Md5Ctx;
1143	HASH HA1;
1144
1145	MD5Init(&Md5Ctx);
1146	MD5Update(&Md5Ctx, pszUserName, strlen(pszUserName));
1147	MD5Update(&Md5Ctx, ":", 1);
1148	MD5Update(&Md5Ctx, pszRealm, strlen(pszRealm));
1149	MD5Update(&Md5Ctx, ":", 1);
1150	MD5Update(&Md5Ctx, pszPassword, strlen(pszPassword));
1151	MD5Final(HA1, &Md5Ctx);
1152	if (strcasecmp(pszAlg, "md5-sess") == 0) {
1153
1154		MD5Init(&Md5Ctx);
1155		MD5Update(&Md5Ctx, HA1, HASHLEN);
1156		MD5Update(&Md5Ctx, ":", 1);
1157		MD5Update(&Md5Ctx, pszNonce, strlen(pszNonce));
1158		MD5Update(&Md5Ctx, ":", 1);
1159		MD5Update(&Md5Ctx, pszCNonce, strlen(pszCNonce));
1160		MD5Final(HA1, &Md5Ctx);
1161	};
1162	CvtHex(HA1, SessionKey);
1163}
1164
1165/* calculate request-digest/response-digest as per HTTP Digest spec */
1166static void
1167DigestCalcResponse(
1168	IN HASHHEX HA1,           /* H(A1) */
1169	IN char * pszNonce,       /* nonce from server */
1170	IN char * pszNonceCount,  /* 8 hex digits */
1171	IN char * pszCNonce,      /* client nonce */
1172	IN char * pszQop,         /* qop-value: "", "auth", "auth-int" */
1173	IN char * pszMethod,      /* method from the request */
1174	IN char * pszDigestUri,   /* requested URL */
1175	IN HASHHEX HEntity,       /* H(entity body) if qop="auth-int" */
1176	OUT HASHHEX Response      /* request-digest or response-digest */
1177	)
1178{
1179/*	DEBUG(fprintf(stderr,
1180		      "Calc: HA1[%s] Nonce[%s] qop[%s] method[%s] URI[%s]\n",
1181		      HA1, pszNonce, pszQop, pszMethod, pszDigestUri));*/
1182	MD5_CTX Md5Ctx;
1183	HASH HA2;
1184	HASH RespHash;
1185	HASHHEX HA2Hex;
1186
1187	// calculate H(A2)
1188	MD5Init(&Md5Ctx);
1189	MD5Update(&Md5Ctx, pszMethod, strlen(pszMethod));
1190	MD5Update(&Md5Ctx, ":", 1);
1191	MD5Update(&Md5Ctx, pszDigestUri, strlen(pszDigestUri));
1192	if (strcasecmp(pszQop, "auth-int") == 0) {
1193		MD5Update(&Md5Ctx, ":", 1);
1194		MD5Update(&Md5Ctx, HEntity, HASHHEXLEN);
1195	};
1196	MD5Final(HA2, &Md5Ctx);
1197	CvtHex(HA2, HA2Hex);
1198
1199	// calculate response
1200	MD5Init(&Md5Ctx);
1201	MD5Update(&Md5Ctx, HA1, HASHHEXLEN);
1202	MD5Update(&Md5Ctx, ":", 1);
1203	MD5Update(&Md5Ctx, pszNonce, strlen(pszNonce));
1204	MD5Update(&Md5Ctx, ":", 1);
1205	if (*pszQop) {
1206		MD5Update(&Md5Ctx, pszNonceCount, strlen(pszNonceCount));
1207		MD5Update(&Md5Ctx, ":", 1);
1208		MD5Update(&Md5Ctx, pszCNonce, strlen(pszCNonce));
1209		MD5Update(&Md5Ctx, ":", 1);
1210		MD5Update(&Md5Ctx, pszQop, strlen(pszQop));
1211		MD5Update(&Md5Ctx, ":", 1);
1212	};
1213	MD5Update(&Md5Ctx, HA2Hex, HASHHEXLEN);
1214	MD5Final(RespHash, &Md5Ctx);
1215	CvtHex(RespHash, Response);
1216}
1217
1218/*
1219 * Generate/Send a Digest authorization header
1220 * This looks like: [Proxy-]Authorization: credentials
1221 *
1222 *  credentials      = "Digest" digest-response
1223 *  digest-response  = 1#( username | realm | nonce | digest-uri
1224 *                      | response | [ algorithm ] | [cnonce] |
1225 *                      [opaque] | [message-qop] |
1226 *                          [nonce-count]  | [auth-param] )
1227 *  username         = "username" "=" username-value
1228 *  username-value   = quoted-string
1229 *  digest-uri       = "uri" "=" digest-uri-value
1230 *  digest-uri-value = request-uri   ; As specified by HTTP/1.1
1231 *  message-qop      = "qop" "=" qop-value
1232 *  cnonce           = "cnonce" "=" cnonce-value
1233 *  cnonce-value     = nonce-value
1234 *  nonce-count      = "nc" "=" nc-value
1235 *  nc-value         = 8LHEX
1236 *  response         = "response" "=" request-digest
1237 *  request-digest = <"> 32LHEX <">
1238 */
1239static int
1240http_digest_auth(conn_t *conn, const char *hdr, http_auth_challenge_t *c,
1241		 http_auth_params_t *parms, struct url *url)
1242{
1243	int r;
1244	char noncecount[10];
1245	char cnonce[40];
1246	char *options = 0;
1247
1248	if (!c->realm || !c->nonce) {
1249		DEBUG(fprintf(stderr, "realm/nonce not set in challenge\n"));
1250		return(-1);
1251	}
1252	if (!c->algo)
1253		c->algo = strdup("");
1254
1255	if (asprintf(&options, "%s%s%s%s",
1256		     *c->algo? ",algorithm=" : "", c->algo,
1257		     c->opaque? ",opaque=" : "", c->opaque?c->opaque:"")== -1)
1258		return (-1);
1259
1260	if (!c->qop) {
1261		c->qop = strdup("");
1262		*noncecount = 0;
1263		*cnonce = 0;
1264	} else {
1265		c->nc++;
1266		sprintf(noncecount, "%08x", c->nc);
1267		/* We don't try very hard with the cnonce ... */
1268		sprintf(cnonce, "%x%lx", getpid(), (unsigned long)time(0));
1269	}
1270
1271	HASHHEX HA1;
1272	DigestCalcHA1(c->algo, parms->user, c->realm,
1273		      parms->password, c->nonce, cnonce, HA1);
1274	DEBUG(fprintf(stderr, "HA1: [%s]\n", HA1));
1275	HASHHEX digest;
1276	DigestCalcResponse(HA1, c->nonce, noncecount, cnonce, c->qop,
1277			   "GET", url->doc, "", digest);
1278
1279	if (c->qop[0]) {
1280		r = http_cmd(conn, "%s: Digest username=\"%s\",realm=\"%s\","
1281			     "nonce=\"%s\",uri=\"%s\",response=\"%s\","
1282			     "qop=\"auth\", cnonce=\"%s\", nc=%s%s",
1283			     hdr, parms->user, c->realm,
1284			     c->nonce, url->doc, digest,
1285			     cnonce, noncecount, options);
1286	} else {
1287		r = http_cmd(conn, "%s: Digest username=\"%s\",realm=\"%s\","
1288			     "nonce=\"%s\",uri=\"%s\",response=\"%s\"%s",
1289			     hdr, parms->user, c->realm,
1290			     c->nonce, url->doc, digest, options);
1291	}
1292	if (options)
1293		free(options);
1294	return (r);
1295}
1296
1297/*
1298 * Encode username and password
1299 */
1300static int
1301http_basic_auth(conn_t *conn, const char *hdr, const char *usr, const char *pwd)
1302{
1303	char *upw, *auth;
1304	int r;
1305
1306	DEBUG(fprintf(stderr, "basic: usr: [%s]\n", usr));
1307	DEBUG(fprintf(stderr, "basic: pwd: [%s]\n", pwd));
1308	if (asprintf(&upw, "%s:%s", usr, pwd) == -1)
1309		return (-1);
1310	auth = http_base64(upw);
1311	free(upw);
1312	if (auth == NULL)
1313		return (-1);
1314	r = http_cmd(conn, "%s: Basic %s", hdr, auth);
1315	free(auth);
1316	return (r);
1317}
1318
1319/*
1320 * Chose the challenge to answer and call the appropriate routine to
1321 * produce the header.
1322 */
1323static int
1324http_authorize(conn_t *conn, const char *hdr, http_auth_challenges_t *cs,
1325	       http_auth_params_t *parms, struct url *url)
1326{
1327	http_auth_challenge_t *basic = NULL;
1328	http_auth_challenge_t *digest = NULL;
1329	int i;
1330
1331	/* If user or pass are null we're not happy */
1332	if (!parms->user || !parms->password) {
1333		DEBUG(fprintf(stderr, "NULL usr or pass\n"));
1334		return (-1);
1335	}
1336
1337	/* Look for a Digest and a Basic challenge */
1338	for (i = 0; i < cs->count; i++) {
1339		if (cs->challenges[i]->scheme == HTTPAS_BASIC)
1340			basic = cs->challenges[i];
1341		if (cs->challenges[i]->scheme == HTTPAS_DIGEST)
1342			digest = cs->challenges[i];
1343	}
1344
1345	/* Error if "Digest" was specified and there is no Digest challenge */
1346	if (!digest && (parms->scheme &&
1347			!strcasecmp(parms->scheme, "digest"))) {
1348		DEBUG(fprintf(stderr,
1349			      "Digest auth in env, not supported by peer\n"));
1350		return (-1);
1351	}
1352	/*
1353	 * If "basic" was specified in the environment, or there is no Digest
1354	 * challenge, do the basic thing. Don't need a challenge for this,
1355	 * so no need to check basic!=NULL
1356	 */
1357	if (!digest || (parms->scheme && !strcasecmp(parms->scheme,"basic")))
1358		return (http_basic_auth(conn,hdr,parms->user,parms->password));
1359
1360	/* Else, prefer digest. We just checked that it's not NULL */
1361	return (http_digest_auth(conn, hdr, digest, parms, url));
1362}
1363
1364/*****************************************************************************
1365 * Helper functions for connecting to a server or proxy
1366 */
1367
1368/*
1369 * Connect to the correct HTTP server or proxy.
1370 */
1371static conn_t *
1372http_connect(struct url *URL, struct url *purl, const char *flags)
1373{
1374	conn_t *conn;
1375	int verbose;
1376	int af, val;
1377
1378#ifdef INET6
1379	af = AF_UNSPEC;
1380#else
1381	af = AF_INET;
1382#endif
1383
1384	verbose = CHECK_FLAG('v');
1385	if (CHECK_FLAG('4'))
1386		af = AF_INET;
1387#ifdef INET6
1388	else if (CHECK_FLAG('6'))
1389		af = AF_INET6;
1390#endif
1391
1392	if (purl && strcasecmp(URL->scheme, SCHEME_HTTPS) != 0) {
1393		URL = purl;
1394	} else if (strcasecmp(URL->scheme, SCHEME_FTP) == 0) {
1395		/* can't talk http to an ftp server */
1396		/* XXX should set an error code */
1397		return (NULL);
1398	}
1399
1400	if ((conn = fetch_connect(URL->host, URL->port, af, verbose)) == NULL)
1401		/* fetch_connect() has already set an error code */
1402		return (NULL);
1403	if (strcasecmp(URL->scheme, SCHEME_HTTPS) == 0 &&
1404	    fetch_ssl(conn, verbose) == -1) {
1405		fetch_close(conn);
1406		/* grrr */
1407		errno = EAUTH;
1408		fetch_syserr();
1409		return (NULL);
1410	}
1411
1412	val = 1;
1413	setsockopt(conn->sd, IPPROTO_TCP, TCP_NOPUSH, &val, sizeof(val));
1414
1415	return (conn);
1416}
1417
1418static struct url *
1419http_get_proxy(struct url * url, const char *flags)
1420{
1421	struct url *purl;
1422	char *p;
1423
1424	if (flags != NULL && strchr(flags, 'd') != NULL)
1425		return (NULL);
1426	if (fetch_no_proxy_match(url->host))
1427		return (NULL);
1428	if (((p = getenv("HTTP_PROXY")) || (p = getenv("http_proxy"))) &&
1429	    *p && (purl = fetchParseURL(p))) {
1430		if (!*purl->scheme)
1431			strcpy(purl->scheme, SCHEME_HTTP);
1432		if (!purl->port)
1433			purl->port = fetch_default_proxy_port(purl->scheme);
1434		if (strcasecmp(purl->scheme, SCHEME_HTTP) == 0)
1435			return (purl);
1436		fetchFreeURL(purl);
1437	}
1438	return (NULL);
1439}
1440
1441static void
1442http_print_html(FILE *out, FILE *in)
1443{
1444	size_t len;
1445	char *line, *p, *q;
1446	int comment, tag;
1447
1448	comment = tag = 0;
1449	while ((line = fgetln(in, &len)) != NULL) {
1450		while (len && isspace((unsigned char)line[len - 1]))
1451			--len;
1452		for (p = q = line; q < line + len; ++q) {
1453			if (comment && *q == '-') {
1454				if (q + 2 < line + len &&
1455				    strcmp(q, "-->") == 0) {
1456					tag = comment = 0;
1457					q += 2;
1458				}
1459			} else if (tag && !comment && *q == '>') {
1460				p = q + 1;
1461				tag = 0;
1462			} else if (!tag && *q == '<') {
1463				if (q > p)
1464					fwrite(p, q - p, 1, out);
1465				tag = 1;
1466				if (q + 3 < line + len &&
1467				    strcmp(q, "<!--") == 0) {
1468					comment = 1;
1469					q += 3;
1470				}
1471			}
1472		}
1473		if (!tag && q > p)
1474			fwrite(p, q - p, 1, out);
1475		fputc('\n', out);
1476	}
1477}
1478
1479
1480/*****************************************************************************
1481 * Core
1482 */
1483
1484/*
1485 * Send a request and process the reply
1486 *
1487 * XXX This function is way too long, the do..while loop should be split
1488 * XXX off into a separate function.
1489 */
1490FILE *
1491http_request(struct url *URL, const char *op, struct url_stat *us,
1492	struct url *purl, const char *flags)
1493{
1494	char timebuf[80];
1495	char hbuf[MAXHOSTNAMELEN + 7], *host;
1496	conn_t *conn;
1497	struct url *url, *new;
1498	int chunked, direct, ims, noredirect, verbose;
1499	int e, i, n, val;
1500	off_t offset, clength, length, size;
1501	time_t mtime;
1502	const char *p;
1503	FILE *f;
1504	hdr_t h;
1505	struct tm *timestruct;
1506	http_headerbuf_t headerbuf;
1507	http_auth_challenges_t server_challenges;
1508	http_auth_challenges_t proxy_challenges;
1509
1510	/* The following calls don't allocate anything */
1511	init_http_headerbuf(&headerbuf);
1512	init_http_auth_challenges(&server_challenges);
1513	init_http_auth_challenges(&proxy_challenges);
1514
1515	direct = CHECK_FLAG('d');
1516	noredirect = CHECK_FLAG('A');
1517	verbose = CHECK_FLAG('v');
1518	ims = CHECK_FLAG('i');
1519
1520	if (direct && purl) {
1521		fetchFreeURL(purl);
1522		purl = NULL;
1523	}
1524
1525	/* try the provided URL first */
1526	url = URL;
1527
1528	n = MAX_REDIRECT;
1529	i = 0;
1530
1531	e = HTTP_PROTOCOL_ERROR;
1532	do {
1533		new = NULL;
1534		chunked = 0;
1535		offset = 0;
1536		clength = -1;
1537		length = -1;
1538		size = -1;
1539		mtime = 0;
1540
1541		/* check port */
1542		if (!url->port)
1543			url->port = fetch_default_port(url->scheme);
1544
1545		/* were we redirected to an FTP URL? */
1546		if (purl == NULL && strcmp(url->scheme, SCHEME_FTP) == 0) {
1547			if (strcmp(op, "GET") == 0)
1548				return (ftp_request(url, "RETR", us, purl, flags));
1549			else if (strcmp(op, "HEAD") == 0)
1550				return (ftp_request(url, "STAT", us, purl, flags));
1551		}
1552
1553		/* connect to server or proxy */
1554		if ((conn = http_connect(url, purl, flags)) == NULL)
1555			goto ouch;
1556
1557		host = url->host;
1558#ifdef INET6
1559		if (strchr(url->host, ':')) {
1560			snprintf(hbuf, sizeof(hbuf), "[%s]", url->host);
1561			host = hbuf;
1562		}
1563#endif
1564		if (url->port != fetch_default_port(url->scheme)) {
1565			if (host != hbuf) {
1566				strcpy(hbuf, host);
1567				host = hbuf;
1568			}
1569			snprintf(hbuf + strlen(hbuf),
1570			    sizeof(hbuf) - strlen(hbuf), ":%d", url->port);
1571		}
1572
1573		/* send request */
1574		if (verbose)
1575			fetch_info("requesting %s://%s%s",
1576			    url->scheme, host, url->doc);
1577		if (purl) {
1578			http_cmd(conn, "%s %s://%s%s HTTP/1.1",
1579			    op, url->scheme, host, url->doc);
1580		} else {
1581			http_cmd(conn, "%s %s HTTP/1.1",
1582			    op, url->doc);
1583		}
1584
1585		if (ims && url->ims_time) {
1586			timestruct = gmtime((time_t *)&url->ims_time);
1587			(void)strftime(timebuf, 80, "%a, %d %b %Y %T GMT",
1588			    timestruct);
1589			if (verbose)
1590				fetch_info("If-Modified-Since: %s", timebuf);
1591			http_cmd(conn, "If-Modified-Since: %s", timebuf);
1592		}
1593		/* virtual host */
1594		http_cmd(conn, "Host: %s", host);
1595
1596		/*
1597		 * Proxy authorization: we only send auth after we received
1598		 * a 407 error. We do not first try basic anyway (changed
1599		 * when support was added for digest-auth)
1600		 */
1601		if (purl && proxy_challenges.valid) {
1602			http_auth_params_t aparams;
1603			init_http_auth_params(&aparams);
1604			if (*purl->user || *purl->pwd) {
1605				aparams.user = purl->user ?
1606					strdup(purl->user) : strdup("");
1607				aparams.password = purl->pwd?
1608					strdup(purl->pwd) : strdup("");
1609			} else if ((p = getenv("HTTP_PROXY_AUTH")) != NULL &&
1610				   *p != '\0') {
1611				if (http_authfromenv(p, &aparams) < 0) {
1612					http_seterr(HTTP_NEED_PROXY_AUTH);
1613					goto ouch;
1614				}
1615			}
1616			http_authorize(conn, "Proxy-Authorization",
1617				       &proxy_challenges, &aparams, url);
1618			clean_http_auth_params(&aparams);
1619		}
1620
1621		/*
1622		 * Server authorization: we never send "a priori"
1623		 * Basic auth, which used to be done if user/pass were
1624		 * set in the url. This would be weird because we'd send the
1625		 * password in the clear even if Digest is finally to be
1626		 * used (it would have made more sense for the
1627		 * pre-digest version to do this when Basic was specified
1628		 * in the environment)
1629		 */
1630		if (server_challenges.valid) {
1631			http_auth_params_t aparams;
1632			init_http_auth_params(&aparams);
1633			if (*url->user || *url->pwd) {
1634				aparams.user = url->user ?
1635					strdup(url->user) : strdup("");
1636				aparams.password = url->pwd ?
1637					strdup(url->pwd) : strdup("");
1638			} else if ((p = getenv("HTTP_AUTH")) != NULL &&
1639				   *p != '\0') {
1640				if (http_authfromenv(p, &aparams) < 0) {
1641					http_seterr(HTTP_NEED_AUTH);
1642					goto ouch;
1643				}
1644			} else if (fetchAuthMethod &&
1645				   fetchAuthMethod(url) == 0) {
1646				aparams.user = url->user ?
1647					strdup(url->user) : strdup("");
1648				aparams.password = url->pwd ?
1649					strdup(url->pwd) : strdup("");
1650			} else {
1651				http_seterr(HTTP_NEED_AUTH);
1652				goto ouch;
1653			}
1654			http_authorize(conn, "Authorization",
1655				       &server_challenges, &aparams, url);
1656			clean_http_auth_params(&aparams);
1657		}
1658
1659		/* other headers */
1660		if ((p = getenv("HTTP_REFERER")) != NULL && *p != '\0') {
1661			if (strcasecmp(p, "auto") == 0)
1662				http_cmd(conn, "Referer: %s://%s%s",
1663				    url->scheme, host, url->doc);
1664			else
1665				http_cmd(conn, "Referer: %s", p);
1666		}
1667		if ((p = getenv("HTTP_USER_AGENT")) != NULL && *p != '\0')
1668			http_cmd(conn, "User-Agent: %s", p);
1669		else
1670			http_cmd(conn, "User-Agent: %s " _LIBFETCH_VER, getprogname());
1671		if (url->offset > 0)
1672			http_cmd(conn, "Range: bytes=%lld-", (long long)url->offset);
1673		http_cmd(conn, "Connection: close");
1674		http_cmd(conn, "");
1675
1676		/*
1677		 * Force the queued request to be dispatched.  Normally, one
1678		 * would do this with shutdown(2) but squid proxies can be
1679		 * configured to disallow such half-closed connections.  To
1680		 * be compatible with such configurations, fiddle with socket
1681		 * options to force the pending data to be written.
1682		 */
1683		val = 0;
1684		setsockopt(conn->sd, IPPROTO_TCP, TCP_NOPUSH, &val,
1685			   sizeof(val));
1686		val = 1;
1687		setsockopt(conn->sd, IPPROTO_TCP, TCP_NODELAY, &val,
1688			   sizeof(val));
1689
1690		/* get reply */
1691		switch (http_get_reply(conn)) {
1692		case HTTP_OK:
1693		case HTTP_PARTIAL:
1694		case HTTP_NOT_MODIFIED:
1695			/* fine */
1696			break;
1697		case HTTP_MOVED_PERM:
1698		case HTTP_MOVED_TEMP:
1699		case HTTP_SEE_OTHER:
1700			/*
1701			 * Not so fine, but we still have to read the
1702			 * headers to get the new location.
1703			 */
1704			break;
1705		case HTTP_NEED_AUTH:
1706			if (server_challenges.valid) {
1707				/*
1708				 * We already sent out authorization code,
1709				 * so there's nothing more we can do.
1710				 */
1711				http_seterr(conn->err);
1712				goto ouch;
1713			}
1714			/* try again, but send the password this time */
1715			if (verbose)
1716				fetch_info("server requires authorization");
1717			break;
1718		case HTTP_NEED_PROXY_AUTH:
1719			if (proxy_challenges.valid) {
1720				/*
1721				 * We already sent our proxy
1722				 * authorization code, so there's
1723				 * nothing more we can do. */
1724				http_seterr(conn->err);
1725				goto ouch;
1726			}
1727			/* try again, but send the password this time */
1728			if (verbose)
1729				fetch_info("proxy requires authorization");
1730			break;
1731		case HTTP_BAD_RANGE:
1732			/*
1733			 * This can happen if we ask for 0 bytes because
1734			 * we already have the whole file.  Consider this
1735			 * a success for now, and check sizes later.
1736			 */
1737			break;
1738		case HTTP_PROTOCOL_ERROR:
1739			/* fall through */
1740		case -1:
1741			fetch_syserr();
1742			goto ouch;
1743		default:
1744			http_seterr(conn->err);
1745			if (!verbose)
1746				goto ouch;
1747			/* fall through so we can get the full error message */
1748		}
1749
1750		/* get headers. http_next_header expects one line readahead */
1751		if (fetch_getln(conn) == -1) {
1752		    fetch_syserr();
1753		    goto ouch;
1754		}
1755		do {
1756		    switch ((h = http_next_header(conn, &headerbuf, &p))) {
1757			case hdr_syserror:
1758				fetch_syserr();
1759				goto ouch;
1760			case hdr_error:
1761				http_seterr(HTTP_PROTOCOL_ERROR);
1762				goto ouch;
1763			case hdr_content_length:
1764				http_parse_length(p, &clength);
1765				break;
1766			case hdr_content_range:
1767				http_parse_range(p, &offset, &length, &size);
1768				break;
1769			case hdr_last_modified:
1770				http_parse_mtime(p, &mtime);
1771				break;
1772			case hdr_location:
1773				if (!HTTP_REDIRECT(conn->err))
1774					break;
1775				/*
1776				 * if the A flag is set, we don't follow
1777				 * temporary redirects.
1778				 */
1779				if (noredirect &&
1780				    conn->err != HTTP_MOVED_PERM &&
1781				    conn->err != HTTP_PERM_REDIRECT) {
1782					n = 1;
1783					break;
1784                                }
1785				if (new)
1786					free(new);
1787				if (verbose)
1788					fetch_info("%d redirect to %s", conn->err, p);
1789				if (*p == '/')
1790					/* absolute path */
1791					new = fetchMakeURL(url->scheme, url->host, url->port, p,
1792					    url->user, url->pwd);
1793				else
1794					new = fetchParseURL(p);
1795				if (new == NULL) {
1796					/* XXX should set an error code */
1797					DEBUG(fprintf(stderr, "failed to parse new URL\n"));
1798					goto ouch;
1799				}
1800
1801				/* Only copy credentials if the host matches */
1802				if (!strcmp(new->host, url->host) && !*new->user && !*new->pwd) {
1803					strcpy(new->user, url->user);
1804					strcpy(new->pwd, url->pwd);
1805				}
1806				new->offset = url->offset;
1807				new->length = url->length;
1808				break;
1809			case hdr_transfer_encoding:
1810				/* XXX weak test*/
1811				chunked = (strcasecmp(p, "chunked") == 0);
1812				break;
1813			case hdr_www_authenticate:
1814				if (conn->err != HTTP_NEED_AUTH)
1815					break;
1816				if (http_parse_authenticate(p, &server_challenges) == 0)
1817					++n;
1818				break;
1819			case hdr_proxy_authenticate:
1820				if (conn->err != HTTP_NEED_PROXY_AUTH)
1821					break;
1822				if (http_parse_authenticate(p, &proxy_challenges) == 0)
1823					++n;
1824				break;
1825			case hdr_end:
1826				/* fall through */
1827			case hdr_unknown:
1828				/* ignore */
1829				break;
1830			}
1831		} while (h > hdr_end);
1832
1833		/* we need to provide authentication */
1834		if (conn->err == HTTP_NEED_AUTH ||
1835		    conn->err == HTTP_NEED_PROXY_AUTH) {
1836			e = conn->err;
1837			if ((conn->err == HTTP_NEED_AUTH &&
1838			     !server_challenges.valid) ||
1839			    (conn->err == HTTP_NEED_PROXY_AUTH &&
1840			     !proxy_challenges.valid)) {
1841				/* 401/7 but no www/proxy-authenticate ?? */
1842				DEBUG(fprintf(stderr, "401/7 and no auth header\n"));
1843				goto ouch;
1844			}
1845			fetch_close(conn);
1846			conn = NULL;
1847			continue;
1848		}
1849
1850		/* requested range not satisfiable */
1851		if (conn->err == HTTP_BAD_RANGE) {
1852			if (url->offset == size && url->length == 0) {
1853				/* asked for 0 bytes; fake it */
1854				offset = url->offset;
1855				clength = -1;
1856				conn->err = HTTP_OK;
1857				break;
1858			} else {
1859				http_seterr(conn->err);
1860				goto ouch;
1861			}
1862		}
1863
1864		/* we have a hit or an error */
1865		if (conn->err == HTTP_OK
1866		    || conn->err == HTTP_NOT_MODIFIED
1867		    || conn->err == HTTP_PARTIAL
1868		    || HTTP_ERROR(conn->err))
1869			break;
1870
1871		/* all other cases: we got a redirect */
1872		e = conn->err;
1873		clean_http_auth_challenges(&server_challenges);
1874		fetch_close(conn);
1875		conn = NULL;
1876		if (!new) {
1877			DEBUG(fprintf(stderr, "redirect with no new location\n"));
1878			break;
1879		}
1880		if (url != URL)
1881			fetchFreeURL(url);
1882		url = new;
1883	} while (++i < n);
1884
1885	/* we failed, or ran out of retries */
1886	if (conn == NULL) {
1887		http_seterr(e);
1888		goto ouch;
1889	}
1890
1891	DEBUG(fprintf(stderr, "offset %lld, length %lld,"
1892		  " size %lld, clength %lld\n",
1893		  (long long)offset, (long long)length,
1894		  (long long)size, (long long)clength));
1895
1896	if (conn->err == HTTP_NOT_MODIFIED) {
1897		http_seterr(HTTP_NOT_MODIFIED);
1898		return (NULL);
1899	}
1900
1901	/* check for inconsistencies */
1902	if (clength != -1 && length != -1 && clength != length) {
1903		http_seterr(HTTP_PROTOCOL_ERROR);
1904		goto ouch;
1905	}
1906	if (clength == -1)
1907		clength = length;
1908	if (clength != -1)
1909		length = offset + clength;
1910	if (length != -1 && size != -1 && length != size) {
1911		http_seterr(HTTP_PROTOCOL_ERROR);
1912		goto ouch;
1913	}
1914	if (size == -1)
1915		size = length;
1916
1917	/* fill in stats */
1918	if (us) {
1919		us->size = size;
1920		us->atime = us->mtime = mtime;
1921	}
1922
1923	/* too far? */
1924	if (URL->offset > 0 && offset > URL->offset) {
1925		http_seterr(HTTP_PROTOCOL_ERROR);
1926		goto ouch;
1927	}
1928
1929	/* report back real offset and size */
1930	URL->offset = offset;
1931	URL->length = clength;
1932
1933	/* wrap it up in a FILE */
1934	if ((f = http_funopen(conn, chunked)) == NULL) {
1935		fetch_syserr();
1936		goto ouch;
1937	}
1938
1939	if (url != URL)
1940		fetchFreeURL(url);
1941	if (purl)
1942		fetchFreeURL(purl);
1943
1944	if (HTTP_ERROR(conn->err)) {
1945		http_print_html(stderr, f);
1946		fclose(f);
1947		f = NULL;
1948	}
1949	clean_http_headerbuf(&headerbuf);
1950	clean_http_auth_challenges(&server_challenges);
1951	clean_http_auth_challenges(&proxy_challenges);
1952	return (f);
1953
1954ouch:
1955	if (url != URL)
1956		fetchFreeURL(url);
1957	if (purl)
1958		fetchFreeURL(purl);
1959	if (conn != NULL)
1960		fetch_close(conn);
1961	clean_http_headerbuf(&headerbuf);
1962	clean_http_auth_challenges(&server_challenges);
1963	clean_http_auth_challenges(&proxy_challenges);
1964	return (NULL);
1965}
1966
1967
1968/*****************************************************************************
1969 * Entry points
1970 */
1971
1972/*
1973 * Retrieve and stat a file by HTTP
1974 */
1975FILE *
1976fetchXGetHTTP(struct url *URL, struct url_stat *us, const char *flags)
1977{
1978	return (http_request(URL, "GET", us, http_get_proxy(URL, flags), flags));
1979}
1980
1981/*
1982 * Retrieve a file by HTTP
1983 */
1984FILE *
1985fetchGetHTTP(struct url *URL, const char *flags)
1986{
1987	return (fetchXGetHTTP(URL, NULL, flags));
1988}
1989
1990/*
1991 * Store a file by HTTP
1992 */
1993FILE *
1994fetchPutHTTP(struct url *URL __unused, const char *flags __unused)
1995{
1996	warnx("fetchPutHTTP(): not implemented");
1997	return (NULL);
1998}
1999
2000/*
2001 * Get an HTTP document's metadata
2002 */
2003int
2004fetchStatHTTP(struct url *URL, struct url_stat *us, const char *flags)
2005{
2006	FILE *f;
2007
2008	f = http_request(URL, "HEAD", us, http_get_proxy(URL, flags), flags);
2009	if (f == NULL)
2010		return (-1);
2011	fclose(f);
2012	return (0);
2013}
2014
2015/*
2016 * List a directory
2017 */
2018struct url_ent *
2019fetchListHTTP(struct url *url __unused, const char *flags __unused)
2020{
2021	warnx("fetchListHTTP(): not implemented");
2022	return (NULL);
2023}
2024