Bug 1013: Don't assume errno is between 0 and 100000

[elinks.git] / src / protocol / http / http.c

/* Internal "http" protocol implementation */

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_FCNTL_H
#include <fcntl.h> /* OS/2 needs this after sys/types.h */
#endif
#ifdef HAVE_LIMITS_H
#include <limits.h>
#endif

#include "elinks.h"

#include "cache/cache.h"
#include "config/options.h"
#include "cookies/cookies.h"
#include "intl/charsets.h"
#include "intl/gettext/libintl.h"
#include "main/module.h"
#include "network/connection.h"
#include "network/progress.h"
#include "network/socket.h"
#include "osdep/ascii.h"
#include "osdep/osdep.h"
#include "osdep/sysname.h"
#include "protocol/auth/auth.h"
#include "protocol/auth/digest.h"
#include "protocol/date.h"
#include "protocol/header.h"
#include "protocol/http/blacklist.h"
#include "protocol/http/codes.h"
#include "protocol/http/http.h"
#include "protocol/uri.h"
#include "session/session.h"
#include "terminal/terminal.h"
#include "util/base64.h"
#include "util/conv.h"
#include "util/memory.h"
#include "util/string.h"

#ifdef CONFIG_GSSAPI
#include "http_negotiate.h"
#endif

struct http_version {
        int major;
        int minor;
};

#define HTTP_0_9(x)      ((x).major == 0 && (x).minor == 9)
#define HTTP_1_0(x)      ((x).major == 1 && (x).minor == 0)
#define HTTP_1_1(x)      ((x).major == 1 && (x).minor == 1)
#define PRE_HTTP_1_0(x)  ((x).major < 1)
#define PRE_HTTP_1_1(x)  (PRE_HTTP_1_0(x) || HTTP_1_0(x))
#define POST_HTTP_1_0(x) ((x).major > 1 || ((x).major == 1 && (x).minor > 0))
#define POST_HTTP_1_1(x) ((x).major > 1 || ((x).major == 1 && (x).minor > 1))


struct http_connection_info {
        enum blacklist_flags bl_flags;
        struct http_version recv_version;
        struct http_version sent_version;

        int close;

#define LEN_CHUNKED -2 /* == we get data in unknown number of chunks */
#define LEN_FINISHED 0
        int length;

        /* Either bytes coming in this chunk yet or "parser state". */
#define CHUNK_DATA_END  -3
#define CHUNK_ZERO_SIZE -2
#define CHUNK_SIZE      -1
        int chunk_remaining;

        int code;
};


static struct auth_entry proxy_auth;

static unsigned char *accept_charset = NULL;


static struct option_info http_options[] = {
        INIT_OPT_TREE("protocol", N_("HTTP"),
                "http", 0,
                N_("HTTP-specific options.")),


        INIT_OPT_TREE("protocol.http", N_("Server bug workarounds"),
                "bugs", 0,
                N_("Server-side HTTP bugs workarounds.")),

        INIT_OPT_BOOL("protocol.http.bugs", N_("Do not send Accept-Charset"),
                "accept_charset", 0, 1,
                N_("The Accept-Charset header is quite long and sending it can trigger\n"
                "bugs in some rarely found servers.")),

        INIT_OPT_BOOL("protocol.http.bugs", N_("Allow blacklisting"),
                "allow_blacklist", 0, 1,
                N_("Allow blacklisting of buggy servers.")),

        INIT_OPT_BOOL("protocol.http.bugs", N_("Broken 302 redirects"),
                "broken_302_redirect", 0, 1,
                N_("Broken 302 redirect (violates RFC but compatible with Netscape).\n"
                "This is a problem for a lot of web discussion boards and the like.\n"
                "If they will do strange things to you, try to play with this.")),

        INIT_OPT_BOOL("protocol.http.bugs", N_("No keepalive after POST requests"),
                "post_no_keepalive", 0, 0,
                N_("Disable keepalive connection after POST request.")),

        INIT_OPT_BOOL("protocol.http.bugs", N_("Use HTTP/1.0"),
                "http10", 0, 0,
                N_("Use HTTP/1.0 protocol instead of HTTP/1.1.")),

        INIT_OPT_TREE("protocol.http", N_("Proxy configuration"),
                "proxy", 0,
                N_("HTTP proxy configuration.")),

        INIT_OPT_STRING("protocol.http.proxy", N_("Host and port-number"),
                "host", 0, "",
                N_("Host and port-number (host:port) of the HTTP proxy, or blank.\n"
                "If it's blank, HTTP_PROXY environment variable is checked as well.")),

        INIT_OPT_STRING("protocol.http.proxy", N_("Username"),
                "user", 0, "",
                N_("Proxy authentication username.")),

        INIT_OPT_STRING("protocol.http.proxy", N_("Password"),
                "passwd", 0, "",
                N_("Proxy authentication password.")),


        INIT_OPT_TREE("protocol.http", N_("Referer sending"),
                "referer", 0,
                N_("HTTP referer sending options. HTTP referer is a special header\n"
                "sent in the HTTP requests, which is supposed to contain the previous\n"
                "page visited by the browser. This way, the server can know what link\n"
                "did you follow when accessing that page. However, this behaviour\n"
                "can unfortunately considerably affect privacy and can lead even to a\n"
                "security problem on some badly designed web pages.")),

        INIT_OPT_INT("protocol.http.referer", N_("Policy"),
                "policy", 0,
                REFERER_NONE, REFERER_TRUE, REFERER_TRUE,
                N_("Mode of sending HTTP referer:\n"
                "0 is send no referer\n"
                "1 is send current URL as referer\n"
                "2 is send fixed fake referer\n"
                "3 is send previous URL as referer (correct, but insecure)")),

        INIT_OPT_STRING("protocol.http.referer", N_("Fake referer URL"),
                "fake", 0, "",
                N_("Fake referer to be sent when policy is 2.")),


        INIT_OPT_STRING("protocol.http", N_("Send Accept-Language header"),
                "accept_language", 0, "",
                N_("Send Accept-Language header.")),

        INIT_OPT_BOOL("protocol.http", N_("Use UI language as Accept-Language"),
                "accept_ui_language", 0, 1,
                N_("Request localised versions of documents from web-servers (using the\n"
                "Accept-Language header) using the language you have configured for\n"
                "ELinks' user-interface (this also affects navigator.language ECMAScript\n"
                "value available to scripts). Note that some see this as a potential\n"
                "security risk because it tells web-masters and the FBI sniffers about\n"
                "your language preference.")),

        INIT_OPT_BOOL("protocol.http", N_("Activate HTTP TRACE debugging"),
                "trace", 0, 0,
                N_("If active, all HTTP requests are sent with TRACE as their method\n"
                "rather than GET or POST. This is useful for debugging of both ELinks\n"
                "and various server-side scripts --- the server only returns the client's\n"
                "request back to the client verbatim. Note that this type of request may\n"
                "not be enabled on all servers.")),

        /* OSNews.com is supposed to be relying on the textmode token, at least. */
        INIT_OPT_STRING("protocol.http", N_("User-agent identification"),
                "user_agent", 0, "ELinks/%v (textmode; %s; %t-%b)",
                N_("Change the User Agent ID. That means identification string, which\n"
                "is sent to HTTP server when a document is requested. The 'textmode'\n"
                "token in the first field is our silent attempt to establish this as\n"
                "a standard for new textmode user agents, so that the webmasters can\n"
                "have just a single uniform test for these if they are e.g. pushing\n"
                "some lite version to them automagically.\n"
                "Use \" \" if you don't want any User-Agent header to be sent at all.\n"
                "%v in the string means ELinks version,\n"
                "%s in the string means system identification,\n"
                "%t in the string means size of the terminal,\n"
                "%b in the string means number of bars displayed by ELinks.")),


        INIT_OPT_TREE("protocol", N_("HTTPS"),
                "https", 0,
                N_("HTTPS-specific options.")),

        INIT_OPT_TREE("protocol.https", N_("Proxy configuration"),
                "proxy", 0,
                N_("HTTPS proxy configuration.")),

        INIT_OPT_STRING("protocol.https.proxy", N_("Host and port-number"),
                "host", 0, "",
                N_("Host and port-number (host:port) of the HTTPS CONNECT proxy, or blank.\n"
                "If it's blank, HTTPS_PROXY environment variable is checked as well.")),
        NULL_OPTION_INFO,
};

static void done_http();

struct module http_protocol_module = struct_module(
        /* name: */             N_("HTTP"),
        /* options: */          http_options,
        /* hooks: */            NULL,
        /* submodules: */       NULL,
        /* data: */             NULL,
        /* init: */             NULL,
        /* done: */             done_http
);


static void
done_http(void)
{
        mem_free_if(proxy_auth.realm);
        mem_free_if(proxy_auth.nonce);
        mem_free_if(proxy_auth.opaque);

        free_blacklist();

        if (accept_charset)
                mem_free(accept_charset);
}

static void
init_accept_charset(void)
{
        struct string ac;
        unsigned char *cs;
        int i;

        if (!init_string(&ac)) return;

        for (i = 0; (cs = get_cp_mime_name(i)); i++) {
                if (ac.length) {
                        add_to_string(&ac, ", ");
                } else {
                        add_to_string(&ac, "Accept-Charset: ");
                }
                add_to_string(&ac, cs);
        }

        if (ac.length) {
                add_crlf_to_string(&ac);
        }

        accept_charset = squeezastring(&ac);

        done_string(&ac);
}


unsigned char *
subst_user_agent(unsigned char *fmt, unsigned char *version,
                 unsigned char *sysname, unsigned char *termsize)
{
        struct string agent;

        if (!init_string(&agent)) return NULL;

        while (*fmt) {
                int p;

                for (p = 0; fmt[p] && fmt[p] != '%'; p++);

                add_bytes_to_string(&agent, fmt, p);
                fmt += p;

                if (*fmt != '%') continue;

                fmt++;
                switch (*fmt) {
                        case 'b':
                                if (!list_empty(sessions)) {
                                        unsigned char bs[4] = "";
                                        int blen = 0;
                                        struct session *ses = sessions.prev;
                                        int bars = ses->status.show_status_bar
                                                + ses->status.show_tabs_bar
                                                + ses->status.show_title_bar;

                                        ulongcat(bs, &blen, bars, 2, 0);
                                        add_to_string(&agent, bs);
                                }
                                break;
                        case 'v':
                                add_to_string(&agent, version);
                                break;
                        case 's':
                                add_to_string(&agent, sysname);
                                break;
                        case 't':
                                if (termsize)
                                        add_to_string(&agent, termsize);
                                break;
                        default:
                                add_bytes_to_string(&agent, fmt - 1, 2);
                                break;
                }
                if (*fmt) fmt++;
        }

        return agent.source;
}

static void
add_url_to_http_string(struct string *header, struct uri *uri, int components)
{
        /* This block substitues spaces in URL by %20s. This is
         * certainly not the right place where to do it, but now the
         * behaviour is at least improved compared to what we had
         * before. We should probably encode all URLs as early as
         * possible, and possibly decode them back in protocol
         * backends. --pasky */
        unsigned char *string = get_uri_string(uri, components);
        unsigned char *data = string;

        if (!string) return;

        while (*data) {
                int len = strcspn(data, " \t\r\n\\");

                add_bytes_to_string(header, data, len);

                if (!data[len]) break;

                if (data[len++] == '\\')
                        add_char_to_string(header, '/');
                else
                        add_to_string(header, "%20");

                data    += len;
        }

        mem_free(string);
}

/* Parse from @end - 1 to @start and set *@value to integer found.
 * It returns -1 if not a number, 0 otherwise.
 * @end should be > @start. */
static int
revstr2num(unsigned char *start, unsigned char *end, int *value)
{
        int q = 1, val = 0;

        do {
                --end;
                if (!isdigit(*end)) return -1; /* NaN */
                val += (*end - '0') * q;
                q *= 10;
        } while (end > start);

        *value = val;
        return 0;
}

/* This function extracts code, major and minor version from string
 * "\s*HTTP/\d+.\d+\s+\d\d\d..."
 * It returns a negative value on error, 0 on success.
 */
static int
get_http_code(struct read_buffer *rb, int *code, struct http_version *version)
{
        unsigned char *head = rb->data;
        unsigned char *start;

        *code = 0;
        version->major = 0;
        version->minor = 0;

        /* Ignore spaces. */
        while (*head == ' ') head++;

        /* HTTP/ */
        if (toupper(*head) != 'H' || toupper(*++head) != 'T' ||
            toupper(*++head) != 'T' || toupper(*++head) != 'P'
            || *++head != '/')
                return -1;

        /* Version */
        start = ++head;
        /* Find next '.' */
        while (*head && *head != '.') head++;
        /* Sanity check. */
        if (!*head || !(head - start)
            || (head - start) > 4
            || !isdigit(*(head + 1)))
                return -2;

        /* Extract major version number. */
        if (revstr2num(start, head, &version->major)) return -3; /* NaN */

        start = head + 1;

        /* Find next ' '. */
        while (*head && *head != ' ') head++;
        /* Sanity check. */
        if (!*head || !(head - start) || (head - start) > 4) return -4;

        /* Extract minor version number. */
        if (revstr2num(start, head, &version->minor)) return -5; /* NaN */

        /* Ignore spaces. */
        while (*head == ' ') head++;

        /* Sanity check for code. */
        if (head[0] < '1' || head[0] > '9' ||
            !isdigit(head[1]) ||
            !isdigit(head[2]))
                return -6; /* Invalid code. */

        /* Extract code. */
        *code = (head[0] - '0') * 100 + (head[1] - '0') * 10 + head[2] - '0';

        return 0;
}

static int
check_http_server_bugs(struct uri *uri, struct http_connection_info *http,
                       unsigned char *head)
{
        unsigned char *server;
        const unsigned char *const *s;
        static const unsigned char *const buggy_servers[] = {
                "mod_czech/3.1.0",
                "Purveyor",
                "Netscape-Enterprise",
                NULL
        };

        if (!get_opt_bool("protocol.http.bugs.allow_blacklist")
            || HTTP_1_0(http->sent_version))
                return 0;

        server = parse_header(head, "Server", NULL);
        if (!server)
                return 0;

        for (s = buggy_servers; *s; s++) {
                if (strstr(server, *s)) {
                        add_blacklist_entry(uri, SERVER_BLACKLIST_HTTP10);
                        break;
                }
        }

        mem_free(server);
        return (*s != NULL);
}

static void
http_end_request(struct connection *conn, struct connection_state state,
                 int notrunc)
{
        shutdown_connection_stream(conn);

        if (conn->info && !((struct http_connection_info *) conn->info)->close
            && (!conn->socket->ssl) /* We won't keep alive ssl connections */
            && (!get_opt_bool("protocol.http.bugs.post_no_keepalive")
                || !conn->uri->post)) {
                if (is_in_state(state, S_OK) && conn->cached)
                        normalize_cache_entry(conn->cached, !notrunc ? conn->from : -1);
                set_connection_state(conn, state);
                add_keepalive_connection(conn, HTTP_KEEPALIVE_TIMEOUT, NULL);
        } else {
                abort_connection(conn, state);
        }
}

static void http_send_header(struct socket *);

void
http_protocol_handler(struct connection *conn)
{
        /* setcstate(conn, S_CONN); */

        if (!has_keepalive_connection(conn)) {
                make_connection(conn->socket, conn->uri, http_send_header,
                                conn->cache_mode >= CACHE_MODE_FORCE_RELOAD);
        } else {
                http_send_header(conn->socket);
        }
}

void
proxy_protocol_handler(struct connection *conn)
{
        http_protocol_handler(conn);
}

#define IS_PROXY_URI(x) ((x)->protocol == PROTOCOL_PROXY)

#define connection_is_https_proxy(conn) \
        (IS_PROXY_URI((conn)->uri) && (conn)->proxied_uri->protocol == PROTOCOL_HTTPS)

struct http_connection_info *
init_http_connection_info(struct connection *conn, int major, int minor, int close)
{
        struct http_connection_info *http;

        http = mem_calloc(1, sizeof(*http));
        if (!http) {
                http_end_request(conn, connection_state(S_OUT_OF_MEM), 0);
                return NULL;
        }

        http->sent_version.major = major;
        http->sent_version.minor = minor;
        http->close = close;

        /* The CGI code uses this too and blacklisting expects a host name. */
        if (conn->proxied_uri->protocol != PROTOCOL_FILE)
                http->bl_flags = get_blacklist_flags(conn->proxied_uri);

        if (http->bl_flags & SERVER_BLACKLIST_HTTP10
            || get_opt_bool("protocol.http.bugs.http10")) {
                http->sent_version.major = 1;
                http->sent_version.minor = 0;
        }

        /* If called from HTTPS proxy connection the connection info might have
         * already been allocated. */
        mem_free_set(&conn->info, http);

        return http;
}

static void
accept_encoding_header(struct string *header)
{
#if defined(CONFIG_GZIP) || defined(CONFIG_BZIP2) || defined(CONFIG_LZMA)
        int comma = 0;

        add_to_string(header, "Accept-Encoding: ");

#ifdef CONFIG_BZIP2
        add_to_string(header, "bzip2");
        comma = 1;
#endif

#ifdef CONFIG_GZIP
        if (comma) add_to_string(header, ", ");
        add_to_string(header, "deflate, gzip");
        comma = 1;
#endif

#ifdef CONFIG_LZMA
        if (comma) add_to_string(header, ", ");
        add_to_string(header, "lzma");
#endif
        add_crlf_to_string(header);
#endif
}

static void
http_send_header(struct socket *socket)
{
        struct connection *conn = socket->conn;
        struct http_connection_info *http;
        int trace = get_opt_bool("protocol.http.trace");
        struct string header;
        unsigned char *post_data = NULL;
        struct auth_entry *entry = NULL;
        struct uri *uri = conn->proxied_uri; /* Set to the real uri */
        unsigned char *optstr;
        int use_connect, talking_to_proxy;

        /* Sanity check for a host */
        if (!uri || !uri->host || !*uri->host || !uri->hostlen) {
                http_end_request(conn, connection_state(S_BAD_URL), 0);
                return;
        }

        http = init_http_connection_info(conn, 1, 1, 0);
        if (!http) return;

        if (!init_string(&header)) {
                http_end_request(conn, connection_state(S_OUT_OF_MEM), 0);
                return;
        }

        if (!conn->cached) conn->cached = find_in_cache(uri);

        talking_to_proxy = IS_PROXY_URI(conn->uri) && !conn->socket->ssl;
        use_connect = connection_is_https_proxy(conn) && !conn->socket->ssl;

        if (trace) {
                add_to_string(&header, "TRACE ");
        } else if (use_connect) {
                add_to_string(&header, "CONNECT ");
                /* In CONNECT requests, we send only a subset of the
                 * headers to the proxy.  See the "CONNECT:" comments
                 * below.  After the CONNECT request succeeds, we
                 * negotiate TLS with the real server and make a new
                 * HTTP request that includes all the headers.  */
        } else if (uri->post) {
                add_to_string(&header, "POST ");
                conn->unrestartable = 1;
        } else {
                add_to_string(&header, "GET ");
        }

        if (!talking_to_proxy) {
                add_char_to_string(&header, '/');
        }

        if (use_connect) {
                /* Add port if it was specified or the default port */
                add_uri_to_string(&header, uri, URI_HTTP_CONNECT);
        } else {
                if (connection_is_https_proxy(conn) && conn->socket->ssl) {
                        add_url_to_http_string(&header, uri, URI_DATA);

                } else if (talking_to_proxy) {
                        add_url_to_http_string(&header, uri, URI_PROXY);

                } else {
                        add_url_to_http_string(&header, conn->uri, URI_DATA);
                }
        }

        add_to_string(&header, " HTTP/");
        add_long_to_string(&header, http->sent_version.major);
        add_char_to_string(&header, '.');
        add_long_to_string(&header, http->sent_version.minor);
        add_crlf_to_string(&header);

        /* CONNECT: Sending a Host header seems pointless as the same
         * information is already in the CONNECT line.  It's harmless
         * though and Mozilla does it too.  */
        add_to_string(&header, "Host: ");
        add_uri_to_string(&header, uri, URI_HTTP_HOST);
        add_crlf_to_string(&header);

        /* CONNECT: Proxy-Authorization is intended to be seen by the proxy.  */
        if (talking_to_proxy) {
                unsigned char *user = get_opt_str("protocol.http.proxy.user");
                unsigned char *passwd = get_opt_str("protocol.http.proxy.passwd");

                if (proxy_auth.digest) {
                        unsigned char *response;
                        int userlen = int_min(strlen(user), AUTH_USER_MAXLEN - 1);
                        int passwordlen = int_min(strlen(passwd), AUTH_PASSWORD_MAXLEN - 1);

                        if (userlen)
                                memcpy(proxy_auth.user, user, userlen);
                        proxy_auth.user[userlen] = '\0';
                        if (passwordlen)
                                memcpy(proxy_auth.password, passwd, passwordlen);
                        proxy_auth.password[passwordlen] = '\0';

                        /* FIXME: @uri is the proxied URI. Maybe the passed URI
                         * should be the proxy URI aka conn->uri. --jonas */
                        response = get_http_auth_digest_response(&proxy_auth, uri);
                        if (response) {
                                add_to_string(&header, "Proxy-Authorization: Digest ");
                                add_to_string(&header, response);
                                add_crlf_to_string(&header);

                                mem_free(response);
                        }

                } else {
                        if (user[0]) {
                                unsigned char *proxy_data;

                                proxy_data = straconcat(user, ":", passwd, (unsigned char *) NULL);
                                if (proxy_data) {
                                        unsigned char *proxy_64 = base64_encode(proxy_data);

                                        if (proxy_64) {
                                                add_to_string(&header, "Proxy-Authorization: Basic ");
                                                add_to_string(&header, proxy_64);
                                                add_crlf_to_string(&header);
                                                mem_free(proxy_64);
                                        }
                                        mem_free(proxy_data);
                                }
                        }
                }
        }

        /* CONNECT: User-Agent does not reveal anything about the
         * resource we're fetching, and it may help the proxy return
         * better error messages.  */
        optstr = get_opt_str("protocol.http.user_agent");
        if (*optstr && strcmp(optstr, " ")) {
                unsigned char *ustr, ts[64] = "";

                add_to_string(&header, "User-Agent: ");

                if (!list_empty(terminals)) {
                        unsigned int tslen = 0;
                        struct terminal *term = terminals.prev;

                        ulongcat(ts, &tslen, term->width, 3, 0);
                        ts[tslen++] = 'x';
                        ulongcat(ts, &tslen, term->height, 3, 0);
                }
                ustr = subst_user_agent(optstr, VERSION_STRING, system_name,
                                        ts);

                if (ustr) {
                        add_to_string(&header, ustr);
                        mem_free(ustr);
                }

                add_crlf_to_string(&header);
        }

        /* CONNECT: Referer probably is a secret page in the HTTPS
         * server, so don't reveal it to the proxy.  */
        if (!use_connect) {
                switch (get_opt_int("protocol.http.referer.policy")) {
                        case REFERER_NONE:
                                /* oh well */
                                break;

                        case REFERER_FAKE:
                                optstr = get_opt_str("protocol.http.referer.fake");
                                if (!optstr[0]) break;
                                add_to_string(&header, "Referer: ");
                                add_to_string(&header, optstr);
                                add_crlf_to_string(&header);
                                break;

                        case REFERER_TRUE:
                                if (!conn->referrer) break;
                                add_to_string(&header, "Referer: ");
                                add_url_to_http_string(&header, conn->referrer, URI_HTTP_REFERRER);
                                add_crlf_to_string(&header);
                                break;

                        case REFERER_SAME_URL:
                                add_to_string(&header, "Referer: ");
                                add_url_to_http_string(&header, uri, URI_HTTP_REFERRER);
                                add_crlf_to_string(&header);
                                break;
                }
        }

        /* CONNECT: Do send all Accept* headers to the CONNECT proxy,
         * because they do not reveal anything about the resource
         * we're going to request via TLS, and they may affect the
         * error message if the CONNECT request fails.
         *
         * If ELinks is ever changed to vary its Accept headers based
         * on what it intends to do with the returned resource, e.g.
         * sending "Accept: text/css" when it wants an external
         * stylesheet, then it should do that only in the inner GET
         * and not in the outer CONNECT.  */
        add_to_string(&header, "Accept: */*");
        add_crlf_to_string(&header);

        accept_encoding_header(&header);

        if (!accept_charset) {
                init_accept_charset();
        }

        if (!(http->bl_flags & SERVER_BLACKLIST_NO_CHARSET)
            && !get_opt_bool("protocol.http.bugs.accept_charset")
            && accept_charset) {
                add_to_string(&header, accept_charset);
        }

        optstr = get_opt_str("protocol.http.accept_language");
        if (optstr[0]) {
                add_to_string(&header, "Accept-Language: ");
                add_to_string(&header, optstr);
                add_crlf_to_string(&header);
        }
#ifdef CONFIG_NLS
        else if (get_opt_bool("protocol.http.accept_ui_language")) {
                unsigned char *code = language_to_iso639(current_language);

                if (code) {
                        add_to_string(&header, "Accept-Language: ");
                        add_to_string(&header, code);
                        add_crlf_to_string(&header);
                }
        }
#endif

        /* CONNECT: Proxy-Connection is intended to be seen by the
         * proxy.  If the CONNECT request succeeds, then the proxy
         * will forward the remainder of the TCP connection to the
         * origin server, and Proxy-Connection does not matter; but
         * if the request fails, then Proxy-Connection may matter.  */
        /* FIXME: What about post-HTTP/1.1?? --Zas */
        if (HTTP_1_1(http->sent_version)) {
                if (!IS_PROXY_URI(conn->uri)) {
                        add_to_string(&header, "Connection: ");
                } else {
                        add_to_string(&header, "Proxy-Connection: ");
                }

                if (!uri->post || !get_opt_bool("protocol.http.bugs.post_no_keepalive")) {
                        add_to_string(&header, "Keep-Alive");
                } else {
                        add_to_string(&header, "close");
                }
                add_crlf_to_string(&header);
        }

        /* CONNECT: Do not tell the proxy anything we have cached
         * about the resource.  */
        if (!use_connect && conn->cached) {
                if (!conn->cached->incomplete && conn->cached->head
                    && conn->cache_mode <= CACHE_MODE_CHECK_IF_MODIFIED) {
                        if (conn->cached->last_modified) {
                                add_to_string(&header, "If-Modified-Since: ");
                                add_to_string(&header, conn->cached->last_modified);
                                add_crlf_to_string(&header);
                        }
                        if (conn->cached->etag) {
                                add_to_string(&header, "If-None-Match: ");
                                add_to_string(&header, conn->cached->etag);
                                add_crlf_to_string(&header);
                        }
                }
        }

        /* CONNECT: Let's send cache control headers to the proxy too;
         * they may affect DNS caching.  */
        if (conn->cache_mode >= CACHE_MODE_FORCE_RELOAD) {
                add_to_string(&header, "Pragma: no-cache");
                add_crlf_to_string(&header);
                add_to_string(&header, "Cache-Control: no-cache");
                add_crlf_to_string(&header);
        }

        /* CONNECT: Do not reveal byte ranges to the proxy.  It can't
         * do anything good with that information anyway.  */
        if (!use_connect && (conn->from || conn->progress->start > 0)) {
                /* conn->from takes precedence. conn->progress.start is set only the first
                 * time, then conn->from gets updated and in case of any retries
                 * etc we have everything interesting in conn->from already. */
                add_to_string(&header, "Range: bytes=");
                add_long_to_string(&header, conn->from ? conn->from : conn->progress->start);
                add_char_to_string(&header, '-');
                add_crlf_to_string(&header);
        }

        /* CONNECT: The Authorization header is for the origin server only.  */
        if (!use_connect) {
#ifdef CONFIG_GSSAPI
                if (http_negotiate_output(uri, &header) != 0)
#endif
                        entry = find_auth(uri);
        }

        if (entry) {
                if (entry->digest) {
                        unsigned char *response;

                        response = get_http_auth_digest_response(entry, uri);
                        if (response) {
                                add_to_string(&header, "Authorization: Digest ");
                                add_to_string(&header, response);
                                add_crlf_to_string(&header);

                                mem_free(response);
                        }

                } else {
                        /* RFC2617 section 2 [Basic Authentication Scheme]
                         *
                         * To receive authorization, the client sends the userid
                         * and password, separated by a single colon (":")
                         * character, within a base64 [7] encoded string in the
                         * credentials. */
                        unsigned char *id;

                        /* Create base64 encoded string. */
                        id = straconcat(entry->user, ":", entry->password,
                                        (unsigned char *) NULL);
                        if (id) {
                                unsigned char *base64 = base64_encode(id);

                                mem_free_set(&id, base64);
                        }

                        if (id) {
                                add_to_string(&header, "Authorization: Basic ");
                                add_to_string(&header, id);
                                add_crlf_to_string(&header);
                                mem_free(id);
                        }
                }
        }

        /* CONNECT: Any POST data is for the origin server only.  */
        if (!use_connect && uri->post) {
                /* We search for first '\n' in uri->post to get content type
                 * as set by get_form_uri(). This '\n' is dropped if any
                 * and replaced by correct '\r\n' termination here. */
                unsigned char *postend = strchr(uri->post, '\n');

                if (postend) {
                        add_to_string(&header, "Content-Type: ");
                        add_bytes_to_string(&header, uri->post, postend - uri->post);
                        add_crlf_to_string(&header);
                }

                post_data = postend ? postend + 1 : uri->post;
                add_to_string(&header, "Content-Length: ");
                add_long_to_string(&header, strlen(post_data) / 2);
                add_crlf_to_string(&header);
        }

#ifdef CONFIG_COOKIES
        /* CONNECT: Cookies are for the origin server only.  */
        if (!use_connect) {
                struct string *cookies = send_cookies(uri);

                if (cookies) {
                        add_to_string(&header, "Cookie: ");
                        add_string_to_string(&header, cookies);
                        add_crlf_to_string(&header);
                        done_string(cookies);
                }
        }
#endif

        add_crlf_to_string(&header);

        /* CONNECT: Any POST data is for the origin server only.
         * This was already checked above and post_data is NULL
         * in that case.  Verified with an assertion below.  */
        if (post_data) {
#define POST_BUFFER_SIZE 4096
                unsigned char *post = post_data;
                unsigned char buffer[POST_BUFFER_SIZE];
                int n = 0;

                assert(!use_connect); /* see comment above */

                while (post[0] && post[1]) {
                        int h1, h2;

                        h1 = unhx(post[0]);
                        assertm(h1 >= 0 && h1 < 16, "h1 in the POST buffer is %d (%d/%c)", h1, post[0], post[0]);
                        if_assert_failed h1 = 0;

                        h2 = unhx(post[1]);
                        assertm(h2 >= 0 && h2 < 16, "h2 in the POST buffer is %d (%d/%c)", h2, post[1], post[1]);
                        if_assert_failed h2 = 0;

                        buffer[n++] = (h1<<4) + h2;
                        post += 2;
                        if (n == POST_BUFFER_SIZE) {
                                add_bytes_to_string(&header, buffer, n);
                                n = 0;
                        }
                }

                if (n)
                        add_bytes_to_string(&header, buffer, n);
#undef POST_BUFFER_SIZE
        }

        request_from_socket(socket, header.source, header.length,
                            connection_state(S_SENT),
                            SOCKET_END_ONCLOSE, http_got_header);
        done_string(&header);
}


/* This function decompresses the data block given in @data (if it was
 * compressed), which is long @len bytes. The decompressed data block is given
 * back to the world as the return value and its length is stored into
 * @new_len. After this function returns, the caller will discard all the @len
 * input bytes, so this function must use all of them unless an error occurs.
 *
 * In this function, value of either http->chunk_remaining or http->length is
 * being changed (it depends on if chunked mode is used or not).
 *
 * Note that the function is still a little esotheric for me. Don't take it
 * lightly and don't mess with it without grave reason! If you dare to touch
 * this without testing the changes on slashdot, freshmeat and cvsweb
 * (including revision history), don't dare to send me any patches! ;) --pasky
 *
 * This function gotta die. */
static unsigned char *
decompress_data(struct connection *conn, unsigned char *data, int len,
                int *new_len)
{
        struct http_connection_info *http = conn->info;
        enum { NORMAL, FINISHING } state = NORMAL;
        int did_read = 0;
        int *length_of_block;
        unsigned char *output = NULL;

#define BIG_READ 65536

        if (http->length == LEN_CHUNKED) {
                if (http->chunk_remaining == CHUNK_ZERO_SIZE)
                        state = FINISHING;
                length_of_block = &http->chunk_remaining;
        } else {
                length_of_block = &http->length;
                if (!*length_of_block) {
                        /* Going to finish this decoding bussiness. */
                        state = FINISHING;
                }
        }

        if (conn->content_encoding == ENCODING_NONE) {
                *new_len = len;
                if (*length_of_block > 0) *length_of_block -= len;
                return data;
        }

        *new_len = 0; /* new_len must be zero if we would ever return NULL */

        if (conn->stream_pipes[0] == -1
            && (c_pipe(conn->stream_pipes) < 0
                || set_nonblocking_fd(conn->stream_pipes[0]) < 0
                || set_nonblocking_fd(conn->stream_pipes[1]) < 0)) {
                return NULL;
        }

        do {
                unsigned char *tmp;

                if (state == NORMAL) {
                        /* ... we aren't finishing yet. */
                        int written = safe_write(conn->stream_pipes[1], data, len);

                        if (written >= 0) {
                                data += written;
                                len -= written;

                                /* In non-keep-alive connections http->length == -1, so the test below */
                                if (*length_of_block > 0)
                                        *length_of_block -= written;
                                /* http->length is 0 at the end of block for all modes: keep-alive,
                                 * non-keep-alive and chunked */
                                if (!http->length) {
                                        /* That's all, folks - let's finish this. */
                                        state = FINISHING;
                                } else if (!len) {
                                        /* We've done for this round (but not done
                                         * completely). Thus we will get out with
                                         * what we have and leave what we wrote to
                                         * the next round - we have to do that since
                                         * we MUST NOT ever empty the pipe completely
                                         * - this would cause a disaster for
                                         * read_encoded(), which would simply not
                                         * work right then. */
                                        return output;
                                }
                        }
                }

                if (!conn->stream) {
                        conn->stream = open_encoded(conn->stream_pipes[0],
                                        conn->content_encoding);
                        if (!conn->stream) return NULL;
                }

                tmp = mem_realloc(output, *new_len + BIG_READ);
                if (!tmp) break;
                output = tmp;

                did_read = read_encoded(conn->stream, output + *new_len, BIG_READ);

                /* Do not break from the loop if did_read == 0.  It
                 * means no decoded data is available yet, but some may
                 * become available later.  This happens especially with
                 * the bzip2 decoder, which needs an entire compressed
                 * block as input before it generates any output.  */
                if (did_read < 0) {
                        state = FINISHING;
                        break;
                }
                *new_len += did_read;
        } while (len || (did_read == BIG_READ));

        if (state == FINISHING) shutdown_connection_stream(conn);
        return output;
}

static int
is_line_in_buffer(struct read_buffer *rb)
{
        int l;

        for (l = 0; l < rb->length; l++) {
                unsigned char a0 = rb->data[l];

                if (a0 == ASCII_LF)
                        return l + 1;
                if (a0 == ASCII_CR) {
                        if (rb->data[l + 1] == ASCII_LF
                            && l < rb->length - 1)
                                return l + 2;
                        if (l == rb->length - 1)
                                return 0;
                }
                if (a0 < ' ')
                        return -1;
        }
        return 0;
}

static void read_http_data(struct socket *socket, struct read_buffer *rb);

static void
read_more_http_data(struct connection *conn, struct read_buffer *rb,
                    int already_got_anything)
{
        struct connection_state state = already_got_anything
                ? connection_state(S_TRANS) : conn->state;

        read_from_socket(conn->socket, rb, state, read_http_data);
}

static void
read_http_data_done(struct connection *conn)
{
        struct http_connection_info *http = conn->info;

        /* There's no content but an error so just print
         * that instead of nothing. */
        if (!conn->from) {
                if (http->code >= 400) {
                        http_error_document(conn, http->code);

                } else {
                        /* This is not an error, thus fine. No need generate any
                         * document, as this may be empty and it's not a problem.
                         * In case of 3xx, we're probably just getting kicked to
                         * another page anyway. And in case of 2xx, the document
                         * may indeed be empty and thus the user should see it so. */
                }
        }

        http_end_request(conn, connection_state(S_OK), 0);
}

/* Returns:
 * -1 on error
 * 0 if more to read
 * 1 if done
 */
static int
read_chunked_http_data(struct connection *conn, struct read_buffer *rb)
{
        struct http_connection_info *http = conn->info;
        int total_data_len = 0;

        while (1) {
                /* Chunked. Good luck! */
                /* See RFC2616, section 3.6.1. Basically, it looks like:
                 * 1234 ; a = b ; c = d\r\n
                 * aklkjadslkfjalkfjlkajkljfdkljdsfkljdf*1234\r\n
                 * 0\r\n
                 * \r\n */
                if (http->chunk_remaining == CHUNK_DATA_END) {
                        int l = is_line_in_buffer(rb);

                        if (l) {
                                if (l == -1) {
                                        /* Invalid character in buffer. */
                                        return -1;
                                }

                                /* Remove everything to the EOLN. */
                                kill_buffer_data(rb, l);
                                if (l <= 2) {
                                        /* Empty line. */
                                        return 2;
                                }
                                continue;
                        }

                } else if (http->chunk_remaining == CHUNK_SIZE) {
                        int l = is_line_in_buffer(rb);

                        if (l) {
                                unsigned char *de;
                                int n = 0;

                                if (l != -1) {
                                        errno = 0;
                                        n = strtol(rb->data, (char **) &de, 16);
                                        if (errno || !*de) {
                                                return -1;
                                        }
                                }

                                if (l == -1 || de == rb->data) {
                                        return -1;
                                }

                                /* Remove everything to the EOLN. */
                                kill_buffer_data(rb, l);
                                http->chunk_remaining = n;
                                if (!http->chunk_remaining)
                                        http->chunk_remaining = CHUNK_ZERO_SIZE;
                                continue;
                        }

                } else {
                        unsigned char *data;
                        int data_len;
                        int zero = (http->chunk_remaining == CHUNK_ZERO_SIZE);
                        int len = zero ? 0 : http->chunk_remaining;

                        /* Maybe everything necessary didn't come yet.. */
                        int_upper_bound(&len, rb->length);
                        conn->received += len;

                        data = decompress_data(conn, rb->data, len, &data_len);

                        if (add_fragment(conn->cached, conn->from,
                                         data, data_len) == 1)
                                conn->tries = 0;

                        if (data && data != rb->data) mem_free(data);

                        conn->from += data_len;
                        total_data_len += data_len;

                        kill_buffer_data(rb, len);

                        if (zero) {
                                /* Last chunk has zero length, so this is last
                                 * chunk, we finished decompression just now
                                 * and now we can happily finish reading this
                                 * stuff. */
                                http->chunk_remaining = CHUNK_DATA_END;
                                continue;
                        }

                        if (!http->chunk_remaining && rb->length > 0) {
                                /* Eat newline succeeding each chunk. */
                                if (rb->data[0] == ASCII_LF) {
                                        kill_buffer_data(rb, 1);
                                } else {
                                        if (rb->data[0] != ASCII_CR
                                            || (rb->length >= 2
                                                && rb->data[1] != ASCII_LF)) {
                                                return -1;
                                        }
                                        if (rb->length < 2) break;
                                        kill_buffer_data(rb, 2);
                                }
                                http->chunk_remaining = CHUNK_SIZE;
                                continue;
                        }
                }
                break;
        }

        /* More to read. */
        return !!total_data_len;
}

/* Returns 0 if more data, 1 if done. */
static int
read_normal_http_data(struct connection *conn, struct read_buffer *rb)
{
        struct http_connection_info *http = conn->info;
        unsigned char *data;
        int data_len;
        int len = rb->length;

        if (http->length >= 0 && http->length < len) {
                /* We won't read more than we have to go. */
                len = http->length;
        }

        conn->received += len;

        data = decompress_data(conn, rb->data, len, &data_len);

        if (add_fragment(conn->cached, conn->from, data, data_len) == 1)
                conn->tries = 0;

        if (data && data != rb->data) mem_free(data);

        conn->from += data_len;

        kill_buffer_data(rb, len);

        if (!http->length && conn->socket->state == SOCKET_RETRY_ONCLOSE) {
                return 2;
        }

        return !!data_len;
}

static void
read_http_data(struct socket *socket, struct read_buffer *rb)
{
        struct connection *conn = socket->conn;
        struct http_connection_info *http = conn->info;
        int ret;

        if (socket->state == SOCKET_CLOSED) {
                if (conn->content_encoding && http->length == -1) {
                        /* Flush decompression first. */
                        http->length = 0;
                } else {
                        read_http_data_done(conn);
                        return;
                }
        }

        if (http->length != LEN_CHUNKED) {
                ret = read_normal_http_data(conn, rb);

        } else {
                ret = read_chunked_http_data(conn, rb);
        }

        switch (ret) {
        case 0:
                read_more_http_data(conn, rb, 0);
                break;
        case 1:
                read_more_http_data(conn, rb, 1);
                break;
        case 2:
                read_http_data_done(conn);
                break;
        default:
                assertm(ret == -1, "Unexpected return value: %d", ret);
                abort_connection(conn, connection_state(S_HTTP_ERROR));
        }
}

/* Returns offset of the header end, zero if more data is needed, -1 when
 * incorrect data was received, -2 if this is HTTP/0.9 and no header is to
 * come. */
static int
get_header(struct read_buffer *rb)
{
        int i;

        /* XXX: We will have to do some guess about whether an HTTP header is
         * coming or not, in order to support HTTP/0.9 reply correctly. This
         * means a little code duplication with get_http_code(). --pasky */
        if (rb->length > 4 && strncasecmp(rb->data, "HTTP/", 5))
                return -2;

        for (i = 0; i < rb->length; i++) {
                unsigned char a0 = rb->data[i];
                unsigned char a1 = rb->data[i + 1];

                if (a0 == 0) {
                        rb->data[i] = ' ';
                        continue;
                }
                if (a0 == ASCII_LF && a1 == ASCII_LF
                    && i < rb->length - 1)
                        return i + 2;
                if (a0 == ASCII_CR && i < rb->length - 3) {
                        if (a1 == ASCII_CR) continue;
                        if (a1 != ASCII_LF) return -1;
                        if (rb->data[i + 2] == ASCII_CR) {
                                if (rb->data[i + 3] != ASCII_LF) return -1;
                                return i + 4;
                        }
                }
        }

        return 0;
}

/* returns 1 if we need retry the connection (for negotiate-auth only) */
static int
check_http_authentication(struct connection *conn, struct uri *uri,
                unsigned char *header, unsigned char *header_field)
{
        unsigned char *str, *d;
        int ret = 0;

        d = parse_header(header, header_field, &str);
        while (d) {
                if (!strncasecmp(d, "Basic", 5)) {
                        unsigned char *realm = get_header_param(d, "realm");

                        if (realm) {
                                add_auth_entry(uri, realm, NULL, NULL, 0);
                                mem_free(realm);
                                mem_free(d);
                                break;
                        }
                } else if (!strncasecmp(d, "Digest", 6)) {
                        unsigned char *realm = get_header_param(d, "realm");
                        unsigned char *nonce = get_header_param(d, "nonce");
                        unsigned char *opaque = get_header_param(d, "opaque");

                        add_auth_entry(uri, realm, nonce, opaque, 1);

                        mem_free_if(realm);
                        mem_free_if(nonce);
                        mem_free_if(opaque);
                        mem_free(d);
                        break;
                }
#ifdef CONFIG_GSSAPI
                else if (!strncasecmp(d, HTTPNEG_GSS_STR, HTTPNEG_GSS_STRLEN)) {
                        if (http_negotiate_input(conn, uri, HTTPNEG_GSS, str)==0)
                                ret = 1;
                        mem_free(d);
                        break;
                }
                else if (!strncasecmp(d, HTTPNEG_NEG_STR, HTTPNEG_NEG_STRLEN)) {
                        if (http_negotiate_input(conn, uri, HTTPNEG_NEG, str)==0)
                                ret = 1;
                        mem_free(d);
                        break;
                }
#endif
                mem_free(d);
                d = parse_header(str, header_field, &str);
        }
        return ret;
}


void
http_got_header(struct socket *socket, struct read_buffer *rb)
{
        struct connection *conn = socket->conn;
        struct http_connection_info *http = conn->info;
        unsigned char *head;
#ifdef CONFIG_COOKIES
        unsigned char *cookie, *ch;
#endif
        unsigned char *d;
        struct uri *uri = conn->proxied_uri; /* Set to the real uri */
        struct http_version version = { 0, 9 };
        struct connection_state state = (!is_in_state(conn->state, S_PROC)
                                         ? connection_state(S_GETH)
                                         : connection_state(S_PROC));
        int a, h = 200;
        int cf;

        if (socket->state == SOCKET_CLOSED) {
                if (!conn->tries && uri->host) {
                        if (http->bl_flags & SERVER_BLACKLIST_NO_CHARSET) {
                                del_blacklist_entry(uri, SERVER_BLACKLIST_NO_CHARSET);
                        } else {
                                add_blacklist_entry(uri, SERVER_BLACKLIST_NO_CHARSET);
                                conn->tries = -1;
                        }
                }
                retry_connection(conn, connection_state(S_CANT_READ));
                return;
        }
        socket->state = SOCKET_RETRY_ONCLOSE;

again:
        a = get_header(rb);
        if (a == -1) {
                abort_connection(conn, connection_state(S_HTTP_ERROR));
                return;
        }
        if (!a) {
                read_from_socket(conn->socket, rb, state, http_got_header);
                return;
        }
        /* a == -2 from get_header means HTTP/0.9.  In that case, skip
         * the get_http_code call; @h and @version have already been
         * initialized with the right values.  */
        if (a == -2) a = 0;
        if ((a && get_http_code(rb, &h, &version))
            || h == 101) {
                abort_connection(conn, connection_state(S_HTTP_ERROR));
                return;
        }

        /* When no header, HTTP/0.9 document. That's always text/html,
         * according to
         * http://www.w3.org/Protocols/HTTP/AsImplemented.html. */
        /* FIXME: This usage of fake protocol headers for setting up the
         * content type has been obsoleted by the @content_type member of
         * {struct cache_entry}. */
        head = (a ? memacpy(rb->data, a)
                  : stracpy("\r\nContent-Type: text/html\r\n"));
        if (!head) {
                abort_connection(conn, connection_state(S_OUT_OF_MEM));
                return;
        }

        if (check_http_server_bugs(uri, http, head)) {
                mem_free(head);
                retry_connection(conn, connection_state(S_RESTART));
                return;
        }

#ifdef CONFIG_CGI
        if (uri->protocol == PROTOCOL_FILE) {
                /* ``Status'' is not a standard HTTP header field although some
                 * HTTP servers like www.php.net uses it for some reason. It should
                 * only be used for CGI scripts so that it does not interfere
                 * with status code depended handling for ``normal'' HTTP like
                 * redirects. */
                d = parse_header(head, "Status", NULL);
                if (d) {
                        int h2 = atoi(d);

                        mem_free(d);
                        if (h2 >= 100 && h2 < 600) h = h2;
                        if (h == 101) {
                                mem_free(head);
                                abort_connection(conn, connection_state(S_HTTP_ERROR));
                                return;
                        }
                }
        }
#endif

#ifdef CONFIG_COOKIES
        ch = head;
        while ((cookie = parse_header(ch, "Set-Cookie", &ch))) {
                set_cookie(uri, cookie);
                mem_free(cookie);
        }
#endif
        http->code = h;

        if (h == 100) {
                mem_free(head);
                state = connection_state(S_PROC);
                kill_buffer_data(rb, a);
                goto again;
        }
        if (h < 200) {
                mem_free(head);
                abort_connection(conn, connection_state(S_HTTP_ERROR));
                return;
        }
        if (h == 304) {
                mem_free(head);
                http_end_request(conn, connection_state(S_OK), 1);
                return;
        }
        if (h == 204) {
                mem_free(head);
                http_end_request(conn, connection_state(S_HTTP_204), 0);
                return;
        }
        if (h == 200 && connection_is_https_proxy(conn) && !conn->socket->ssl) {
#ifdef CONFIG_SSL
                mem_free(head);
                socket->need_ssl = 1;
                complete_connect_socket(socket, uri, http_send_header);
#else
                abort_connection(conn, connection_state(S_SSL_ERROR));
#endif
                return;
        }

        conn->cached = get_cache_entry(conn->uri);
        if (!conn->cached) {
                mem_free(head);
                abort_connection(conn, connection_state(S_OUT_OF_MEM));
                return;
        }
        conn->cached->cgi = conn->cgi;
        mem_free_set(&conn->cached->head, head);

        if (!get_opt_bool("document.cache.ignore_cache_control")) {
                struct cache_entry *cached = conn->cached;

                /* I am not entirely sure in what order we should process these
                 * headers and if we should still process Cache-Control max-age
                 * if we already set max age to date mentioned in Expires.
                 * --jonas */
                /* Ensure that when ever cached->max_age is set, cached->expired
                 * is also set, so the cache management knows max_age contains a
                 * valid time. If on the other hand no caching is requested
                 * cached->expire should be set to zero.  */
                if ((d = parse_header(cached->head, "Expires", NULL))) {
                        /* Convert date to seconds. */
                        time_t expires = parse_date(&d, NULL, 0, 1);

                        mem_free(d);

                        if (expires && cached->cache_mode != CACHE_MODE_NEVER) {
                                timeval_from_seconds(&cached->max_age, expires);
                                cached->expire = 1;
                        }
                }

                if ((d = parse_header(cached->head, "Pragma", NULL))) {
                        if (strstr(d, "no-cache")) {
                                cached->cache_mode = CACHE_MODE_NEVER;
                                cached->expire = 0;
                        }
                        mem_free(d);
                }

                if (cached->cache_mode != CACHE_MODE_NEVER
                    && (d = parse_header(cached->head, "Cache-Control", NULL))) {
                        if (strstr(d, "no-cache") || strstr(d, "must-revalidate")) {
                                cached->cache_mode = CACHE_MODE_NEVER;
                                cached->expire = 0;

                        } else  {
                                unsigned char *pos = strstr(d, "max-age=");

                                assert(cached->cache_mode != CACHE_MODE_NEVER);

                                if (pos) {
                                        /* Grab the number of seconds. */
                                        timeval_T max_age;

                                        timeval_from_seconds(&max_age, atol(pos + 8));
                                        timeval_now(&cached->max_age);
                                        timeval_add_interval(&cached->max_age, &max_age);

                                        cached->expire = 1;
                                }
                        }

                        mem_free(d);
                }
        }

        /* XXX: Is there some reason why NOT to follow the Location header
         * for any status? If the server didn't mean it, it wouldn't send
         * it, after all...? --pasky */
        if (h == 201 || h == 301 || h == 302 || h == 303 || h == 307) {
                d = parse_header(conn->cached->head, "Location", NULL);
                if (d) {
                        int use_get_method = (h == 303);

                        /* A note from RFC 2616 section 10.3.3:
                         * RFC 1945 and RFC 2068 specify that the client is not
                         * allowed to change the method on the redirected
                         * request. However, most existing user agent
                         * implementations treat 302 as if it were a 303
                         * response, performing a GET on the Location
                         * field-value regardless of the original request
                         * method. */
                        /* So POST must not be redirected to GET, but some
                         * BUGGY message boards rely on it :-( */
                        if (h == 302
                            && get_opt_bool("protocol.http.bugs.broken_302_redirect"))
                                use_get_method = 1;

                        redirect_cache(conn->cached, d, use_get_method, -1);
                        mem_free(d);
                }
        }

        if (h == 401) {
                if (check_http_authentication(conn, uri,
                                conn->cached->head, "WWW-Authenticate")) {
                        retry_connection(conn, connection_state(S_RESTART));
                        return;
                }

        }
        if (h == 407) {
                unsigned char *str;

                d = parse_header(conn->cached->head, "Proxy-Authenticate", &str);
                while (d) {
                        if (!strncasecmp(d, "Basic", 5)) {
                                unsigned char *realm = get_header_param(d, "realm");

                                if (realm) {
                                        mem_free_set(&proxy_auth.realm, realm);
                                        proxy_auth.digest = 0;
                                        mem_free(d);
                                        break;
                                }

                        } else if (!strncasecmp(d, "Digest", 6)) {
                                unsigned char *realm = get_header_param(d, "realm");
                                unsigned char *nonce = get_header_param(d, "nonce");
                                unsigned char *opaque = get_header_param(d, "opaque");

                                mem_free_set(&proxy_auth.realm, realm);
                                mem_free_set(&proxy_auth.nonce, nonce);
                                mem_free_set(&proxy_auth.opaque, opaque);
                                proxy_auth.digest = 1;

                                mem_free(d);
                                break;
                        }

                        mem_free(d);
                        d = parse_header(str, "Proxy-Authenticate", &str);
                }
        }

        kill_buffer_data(rb, a);
        http->close = 0;
        http->length = -1;
        http->recv_version = version;

        if ((d = parse_header(conn->cached->head, "Connection", NULL))
             || (d = parse_header(conn->cached->head, "Proxy-Connection", NULL))) {
                if (!strcasecmp(d, "close")) http->close = 1;
                mem_free(d);
        } else if (PRE_HTTP_1_1(version)) {
                http->close = 1;
        }

        cf = conn->from;
        conn->from = 0;
        d = parse_header(conn->cached->head, "Content-Range", NULL);
        if (d) {
                if (strlen(d) > 6) {
                        d[5] = 0;
                        if (isdigit(d[6]) && !strcasecmp(d, "bytes")) {
                                int f;

                                errno = 0;
                                f = strtol(d + 6, NULL, 10);

                                if (!errno && f >= 0) conn->from = f;
                        }
                }
                mem_free(d);
        }
        if (cf && !conn->from && !conn->unrestartable) conn->unrestartable = 1;
        if ((conn->progress->start <= 0 && conn->from > cf) || conn->from < 0) {
                /* We don't want this if conn->progress.start because then conn->from will
                 * be probably value of conn->progress.start, while cf is 0. */
                abort_connection(conn, connection_state(S_HTTP_ERROR));
                return;
        }

#if 0
        {
                struct status *s;
                foreach (s, conn->downloads) {
                        fprintf(stderr, "conn %p status %p pri %d st %d er %d :: ce %s",
                                conn, s, s->pri, s->state, s->prev_error,
                                s->cached ? s->cached->url : (unsigned char *) "N-U-L-L");
                }
        }
#endif

        if (conn->progress->start >= 0) {
                /* Update to the real value which we've got from Content-Range. */
                conn->progress->seek = conn->from;
        }
        conn->progress->start = conn->from;

        d = parse_header(conn->cached->head, "Content-Length", NULL);
        if (d) {
                unsigned char *ep;
                int l;

                errno = 0;
                l = strtol(d, (char **) &ep, 10);

                if (!errno && !*ep && l >= 0) {
                        if (!http->close || POST_HTTP_1_0(version))
                                http->length = l;
                        conn->est_length = conn->from + l;
                }
                mem_free(d);
        }

        if (!conn->unrestartable) {
                d = parse_header(conn->cached->head, "Accept-Ranges", NULL);

                if (d) {
                        if (!strcasecmp(d, "none"))
                                conn->unrestartable = 1;
                        mem_free(d);
                } else {
                        if (!conn->from)
                                conn->unrestartable = 1;
                }
        }

        d = parse_header(conn->cached->head, "Transfer-Encoding", NULL);
        if (d) {
                if (!strcasecmp(d, "chunked")) {
                        http->length = LEN_CHUNKED;
                        http->chunk_remaining = CHUNK_SIZE;
                }
                mem_free(d);
        }
        if (!http->close && http->length == -1) http->close = 1;

        d = parse_header(conn->cached->head, "Last-Modified", NULL);
        if (d) {
                if (conn->cached->last_modified && strcasecmp(conn->cached->last_modified, d)) {
                        delete_entry_content(conn->cached);
                        if (conn->from) {
                                conn->from = 0;
                                mem_free(d);
                                retry_connection(conn, connection_state(S_MODIFIED));
                                return;
                        }
                }
                if (!conn->cached->last_modified) conn->cached->last_modified = d;
                else mem_free(d);
        }
        if (!conn->cached->last_modified) {
                d = parse_header(conn->cached->head, "Date", NULL);
                if (d) conn->cached->last_modified = d;
        }

        /* FIXME: Parse only if HTTP/1.1 or later? --Zas */
        d = parse_header(conn->cached->head, "ETag", NULL);
        if (d) {
                if (conn->cached->etag) {
                        unsigned char *old_tag = conn->cached->etag;
                        unsigned char *new_tag = d;

                        /* http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.19 */

                        if (new_tag[0] == 'W' && new_tag[1] == '/')
                                new_tag += 2;

                        if (old_tag[0] == 'W' && old_tag[1] == '/')
                                old_tag += 2;

                        if (strcmp(new_tag, old_tag)) {
                                delete_entry_content(conn->cached);
                                if (conn->from) {
                                        conn->from = 0;
                                        mem_free(d);
                                        retry_connection(conn, connection_state(S_MODIFIED));
                                        return;
                                }
                        }
                }

                if (!conn->cached->etag)
                        conn->cached->etag = d;
                else
                        mem_free(d);
        }

        d = parse_header(conn->cached->head, "Content-Encoding", NULL);
        if (d) {
                unsigned char *extension = get_extension_from_uri(uri);
                enum stream_encoding file_encoding;

                file_encoding = extension ? guess_encoding(extension) : ENCODING_NONE;
                mem_free_if(extension);

                /* If the content is encoded, we want to preserve the encoding
                 * if it is implied by the extension, so that saving the URI
                 * will leave the saved file with the correct encoding. */
#ifdef CONFIG_GZIP
                if (file_encoding != ENCODING_GZIP
                    && (!strcasecmp(d, "gzip") || !strcasecmp(d, "x-gzip")))
                        conn->content_encoding = ENCODING_GZIP;
                if (!strcasecmp(d, "deflate") || !strcasecmp(d, "x-deflate"))
                        conn->content_encoding = ENCODING_DEFLATE;
#endif

#ifdef CONFIG_BZIP2
                if (file_encoding != ENCODING_BZIP2
                    && (!strcasecmp(d, "bzip2") || !strcasecmp(d, "x-bzip2")))
                        conn->content_encoding = ENCODING_BZIP2;
#endif

#ifdef CONFIG_LZMA
                if (file_encoding != ENCODING_LZMA
                    && (!strcasecmp(d, "lzma") || !strcasecmp(d, "x-lzma")))
                        conn->content_encoding = ENCODING_LZMA;
#endif
                mem_free(d);
        }

        if (conn->content_encoding != ENCODING_NONE) {
                mem_free_if(conn->cached->encoding_info);
                conn->cached->encoding_info = stracpy(get_encoding_name(conn->content_encoding));
        }

        if (http->length == -1 || http->close)
                socket->state = SOCKET_END_ONCLOSE;

        read_http_data(socket, rb);
}