ne_request.c

linux subdivision ying gai ke yi le ba

/* 
   HTTP request/response handling
   Copyright (C) 1999-2004, Joe Orton <[e-mail]>

   This library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public
   License as published by the Free Software Foundation; either
   version 2 of the License, or (at your option) any later version.
   
   This library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public
   License along with this library; if not, write to the Free
   Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
   MA 02111-1307, USA

*/

/* This is the HTTP client request/response implementation.
 * The goal of this code is to be modular and simple.
 */

#include "config.h"

#include <sys/types.h>
#include <sys/stat.h>
#ifdef __EMX__
#include <sys/select.h>
#endif

#ifdef HAVE_LIMITS_H
#include <limits.h> /* for UINT_MAX etc */
#endif

#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif 
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif

#include "ne_i18n.h"

#include "ne_alloc.h"
#include "ne_request.h"
#include "ne_string.h" /* for ne_buffer */
#include "ne_utils.h"
#include "ne_socket.h"
#include "ne_uri.h"

#include "ne_private.h"

#define HTTP_EXPECT_TIMEOUT 15
/* 100-continue only used if size > HTTP_EXPECT_MINSIZ */
#define HTTP_EXPECT_MINSIZE 1024

/* with thanks to Jim Blandy; this macro simplified loads of code. */
#define HTTP_ERR(x) do { int _ret = (x); if (_ret != NE_OK) return _ret; } while (0)

#define SOCK_ERR(req, op, msg) do { ssize_t sret = (op); \
if (sret < 0) return aborted(req, msg, sret); } while (0)

/* This is called with each of the headers in the response */
struct header_handler {
    char *name;
    ne_header_handler handler;
    void *userdata;
    struct header_handler *next;
};

/* TODO: could unify these all into a generic callback list */

struct body_reader {
    ne_block_reader handler;
    ne_accept_response accept_response;
    unsigned int use:1;
    void *userdata;
    struct body_reader *next;
};

struct ne_request_s {
    char *method, *uri; /* method and Request-URI */

    ne_buffer *headers; /* request headers */

    /* Request body. */
    ne_provide_body body_cb;
    void *body_ud;

    /* Comes from either an fd or a buffer. */
    union {
	int fd;
	struct {
	    const char *buffer, *pnt;
	    size_t left;
	} buf;
    } body;
	    
    size_t body_size, body_progress;

    /* temporary store for response lines. */
    char respbuf[BUFSIZ];

    /**** Response ***/

    /* The transfer encoding types */
    struct ne_response {
	int length;            /* Response entity-body content-length */
	size_t left;              /* Bytes left to read */
	size_t chunk_left;        /* Bytes of chunk left to read */
	size_t total;             /* total bytes read so far. */
	/* how the message length is detemined: */
	enum {
	    R_TILLEOF = 0, /* read till eof */
	    R_NO_BODY, /* implicitly no body (HEAD, 204, 205, 304) */
	    R_CHUNKED, /* using chunked transfer-encoding */
	    R_CLENGTH  /* using given content-length */
	} mode;
    } resp;

    /* List of callbacks which are passed response headers */
    struct header_handler *header_catchers;
    
    struct hook *private;

    /* We store response header handlers in a hash table.  The hash is
     * derived from the header name in lower case. */

    /* FIXME: this comment is WAY out of date, and no longer in any
     * way true. */
	    
    /* 53 is magic, of course.  For a standard ne_get() (with
     * redirects), 9 header handlers are defined.  Two of these are
     * for Content-Length (which is a bug, and should be fixed
     * really).  Ignoring that hash clash, the 8 *different* handlers
     * all hash uniquely into the hash table of size 53.  */
#define HH_HASHSIZE 53
    
    struct header_handler *header_handlers[HH_HASHSIZE];
    /* List of callbacks which are passed response body blocks */
    struct body_reader *body_readers;

    /*** Miscellaneous ***/
    unsigned int method_is_head:1;
    unsigned int use_expect100:1;
    unsigned int can_persist:1;

    ne_session *session;
    ne_status status;
};

static int open_connection(ne_request *req);

/* The iterative step used to produce the hash value.  This is DJB's
 * magic "*33" hash function.  Ralf Engelschall has done some amazing
 * statistical analysis to show that *33 really is a good hash
 * function: check the new-httpd list archives, or his 'str' library
 * source code, for the details.
 *
 * TODO: due to limited range of characters used in header names,
 * could maybe get a better hash function to use? */
 
#define HH_ITERATE(hash, ch) (((hash)*33 + (ch)) % HH_HASHSIZE);

/* Returns hash value for header 'name', converting it to lower-case
 * in-place. */
static inline unsigned int hash_and_lower(char *name)
{
    char *pnt;
    unsigned int hash = 0;

    for (pnt = name; *pnt != '\0'; pnt++) {
	*pnt = tolower(*pnt);
	hash = HH_ITERATE(hash,*pnt);
    }

    return hash;
}

/* Abort a request due to an non-recoverable HTTP protocol error,
 * whilst doing 'doing'.  'code', if non-zero, is the socket error
 * code, NE_SOCK_*, or if zero, is ignored. */
static int aborted(ne_request *req, const char *doing, ssize_t code)
{
    ne_session *sess = req->session;
    int ret = NE_ERROR;

    NE_DEBUG(NE_DBG_HTTP, "Aborted request (%" NE_FMT_SSIZE_T "): %s\n",
	     code, doing);

    switch(code) {
    case NE_SOCK_CLOSED:
	if (sess->use_proxy) {
	    ne_set_error(sess, _("%s: connection was closed by proxy server."),
			 doing);
	} else {
	    ne_set_error(sess, _("%s: connection was closed by server."),
			 doing);
	}
	break;
    case NE_SOCK_TIMEOUT:
	ne_set_error(sess, _("%s: connection timed out."), doing);
	ret = NE_TIMEOUT;
	break;
    case NE_SOCK_ERROR:
    case NE_SOCK_RESET:
    case NE_SOCK_TRUNC:
        ne_set_error(sess, "%s: %s", doing, ne_sock_error(sess->socket));
        break;
    case 0:
	ne_set_error(sess, "%s", doing);
	break;
    }

    ne_close_connection(sess);
    return ret;
}

static void notify_status(ne_session *sess, ne_conn_status status,
			  const char *info)
{
    if (sess->notify_cb) {
	sess->notify_cb(sess->notify_ud, status, info);
    }
}

void ne_duplicate_header(void *userdata, const char *value)
{
    char **location = userdata;
    *location = ne_strdup(value);
}

void ne_handle_numeric_header(void *userdata, const char *value)
{
    int *location = userdata;
    *location = atoi(value);
}

static void *get_private(const struct hook *hk, const char *id)
{
    for (; hk != NULL; hk = hk->next)
	if (strcmp(hk->id, id) == 0)
	    return hk->userdata;
    return NULL;
}

void *ne_get_request_private(ne_request *req, const char *id)
{
    return get_private(req->private, id);
}

void *ne_get_session_private(ne_session *sess, const char *id)
{
    return get_private(sess->private, id);
}

typedef void (*void_fn)(void);

#define ADD_HOOK(hooks, fn, ud) add_hook(&(hooks), NULL, (void_fn)(fn), (ud))

static void add_hook(struct hook **hooks, const char *id, void_fn fn, void *ud)
{
    struct hook *hk = ne_malloc(sizeof (struct hook)), *pos;

    if (*hooks != NULL) {
	for (pos = *hooks; pos->next != NULL; pos = pos->next)
	    /* nullop */;
	pos->next = hk;
    } else {
	*hooks = hk;
    }

    hk->id = id;
    hk->fn = fn;
    hk->userdata = ud;
    hk->next = NULL;
}

void ne_hook_create_request(ne_session *sess, 
			    ne_create_request_fn fn, void *userdata)
{
    ADD_HOOK(sess->create_req_hooks, fn, userdata);
}

void ne_hook_pre_send(ne_session *sess, ne_pre_send_fn fn, void *userdata)
{
    ADD_HOOK(sess->pre_send_hooks, fn, userdata);
}

void ne_hook_post_send(ne_session *sess, ne_post_send_fn fn, void *userdata)
{
    ADD_HOOK(sess->post_send_hooks, fn, userdata);
}

void ne_hook_destroy_request(ne_session *sess,
			     ne_destroy_req_fn fn, void *userdata)
{
    ADD_HOOK(sess->destroy_req_hooks, fn, userdata);    
}

void ne_hook_destroy_session(ne_session *sess,
			     ne_destroy_sess_fn fn, void *userdata)
{
    ADD_HOOK(sess->destroy_sess_hooks, fn, userdata);
}

/* 0.24.x hack to fix ne_compress layer problems */
void ne_kill_pre_send(ne_session *sess, ne_pre_send_fn fn, void *userdata)
{
    struct hook **last, *hk;

    last = &sess->pre_send_hooks;
    hk = *last;

    while (hk) {
        if (hk->fn == (void_fn)fn && hk->userdata == userdata) {
            *last = hk->next;
            ne_free(hk);
            return;
        }
        last = &hk->next;
        hk = *last;
    }
}

void ne_set_session_private(ne_session *sess, const char *id, void *userdata)
{
    add_hook(&sess->private, id, NULL, userdata);
}

void ne_set_request_private(ne_request *req, const char *id, void *userdata)
{
    add_hook(&req->private, id, NULL, userdata);
}

static ssize_t body_string_send(void *userdata, char *buffer, size_t count)
{
    ne_request *req = userdata;
    
    if (count == 0) {
	req->body.buf.left = req->body_size;
	req->body.buf.pnt = req->body.buf.buffer;
    } else {
	/* if body_left == 0 we fall through and return 0. */
	if (req->body.buf.left < count)
	    count = req->body.buf.left;

	memcpy(buffer, req->body.buf.pnt, count);
	req->body.buf.pnt += count;
	req->body.buf.left -= count;
    }

    return count;
}    

static ssize_t body_fd_send(void *userdata, char *buffer, size_t count)
{
    ne_request *req = userdata;

    if (count) {
	return read(req->body.fd, buffer, count);
    } else {
	/* rewind since we may have to send it again */
	return lseek(req->body.fd, SEEK_SET, 0);
    }
}

/* Pulls the request body from the source and pushes it to the given
 * callback.  Returns 0 on success, or NE_* code */
int ne_pull_request_body(ne_request *req, ne_push_fn fn, void *ud)
{
    int ret = 0;
    char buffer[BUFSIZ];
    ssize_t bytes;
    
    /* tell the source to start again from the beginning. */
    (void) req->body_cb(req->body_ud, NULL, 0);
    
    /* TODO: should this attempt to pull exactly the number of bytes
     * they specified were in the body? Currently it just pulls until
     * they return zero. That makes it possible to extend this to do
     * chunked request bodies (i.e. indefinitely long, no C-L), so
     * this is probably a better long-term interface. */
    while ((bytes = req->body_cb(req->body_ud, buffer, sizeof buffer)) > 0) {
	ret = fn(ud, buffer, bytes);
	if (ret < 0)
	    break;
	NE_DEBUG(NE_DBG_HTTPBODY, 
		 "Body block (%" NE_FMT_SSIZE_T " bytes):\n[%.*s]\n",
		 bytes, (int)bytes, buffer);
    }

    if (bytes < 0) {
	ne_set_error(req->session, _("Error reading request body."));
	ret = NE_ERROR;
    }

    return ret;
}

static int send_with_progress(void *userdata, const char *data, size_t n)
{
    ne_request *req = userdata;
    int ret;
    
    ret = ne_sock_fullwrite(req->session->socket, data, n);
    if (ret == 0) {
	req->body_progress += n;
	req->session->progress_cb(req->session->progress_ud,
				  req->body_progress, req->body_size);
    }
    
    return ret;    
}

/* Sends the request body down the socket.
 * Returns 0 on success, or NE_* code */
static int send_request_body(ne_request *req)
{
    int ret; 
    NE_DEBUG(NE_DBG_HTTP, "Sending request body...\n");
    if (req->session->progress_cb) {
	/* with progress callbacks. */
	req->body_progress = 0;
	ret = ne_pull_request_body(req, send_with_progress, req);
    } else {
	/* without progress callbacks. */
	ret = ne_pull_request_body(req, (ne_push_fn)ne_sock_fullwrite,
				   req->session->socket);
    }

    NE_DEBUG(NE_DBG_HTTP, "Request body sent: %s.\n", ret?"failed":"okay");
    return ret;
}

/* Lob the User-Agent, connection and host headers in to the request
 * headers */
static void add_fixed_headers(ne_request *req) 
{
    if (req->session->user_agent) {
        ne_buffer_zappend(req->headers, req->session->user_agent);
    }

    /* Send Connection: Keep-Alive to pre-1.1 origin servers to try
     * harder to get a persistent connection, except if using a proxy
     * as per 2068 sec 19.7.1.  Always add TE: trailers since those
     * are understood. */
    if (!req->session->is_http11 && !req->session->use_proxy) {
	ne_buffer_zappend(req->headers, 
                          "Keep-Alive: " EOL
                          "Connection: TE, Keep-Alive" EOL
                          "TE: trailers" EOL);
    } else {
	ne_buffer_zappend(req->headers, 
                          "Connection: TE" EOL
                          "TE: trailers" EOL);
    }

}