client_side.c

-

#endif
    request->cache_control = httpHeaderGetCc(req_hdr);
    if (request->method == METHOD_TRACE) {
	request->max_forwards = httpHeaderGetInt(req_hdr, HDR_MAX_FORWARDS);
    }
    if (clientCachable(http))
	request->flags.cachable = 1;
    if (clientHierarchical(http))
	request->flags.hierarchical = 1;
    debug(33, 5) ("clientInterpretRequestHeaders: REQ_NOCACHE = %s\n",
	request->flags.nocache ? "SET" : "NOT SET");
    debug(33, 5) ("clientInterpretRequestHeaders: REQ_CACHABLE = %s\n",
	request->flags.cachable ? "SET" : "NOT SET");
    debug(33, 5) ("clientInterpretRequestHeaders: REQ_HIERARCHICAL = %s\n",
	request->flags.hierarchical ? "SET" : "NOT SET");
}

/*
 * clientSetKeepaliveFlag() sets request->flags.proxy_keepalive.
 * This is the client-side persistent connection flag.  We need
 * to set this relatively early in the request processing
 * to handle hacks for broken servers and clients.
 */
static void
clientSetKeepaliveFlag(clientHttpRequest * http)
{
    request_t *request = http->request;
    const HttpHeader *req_hdr = &request->header;
    debug(33, 3) ("clientSetKeepaliveFlag: http_ver = %3.1f\n",
	request->http_ver);
    debug(33, 3) ("clientSetKeepaliveFlag: method = %s\n",
	RequestMethodStr[request->method]);
    if (httpMsgIsPersistent(request->http_ver, req_hdr))
	request->flags.proxy_keepalive = 1;
}

static int
clientCheckContentLength(request_t * r)
{
    /* We only require a content-length for "upload" methods */
    if (!pumpMethod(r->method))
	return 1;
    if (httpHeaderGetInt(&r->header, HDR_CONTENT_LENGTH) < 0)
	return 0;
    return 1;
}

static int
clientCachable(clientHttpRequest * http)
{
    const char *url = http->uri;
    request_t *req = http->request;
    method_t method = req->method;
    aclCheck_t ch;
    memset(&ch, '\0', sizeof(ch));
    /*
     * Hopefully, nobody really wants 'no_cache' by client's IP
     * address, but if they do, this should work if they use IP
     * addresses in their ACLs, or if the client's address is in
     * the FQDN cache.
     *
     * This may not work yet for 'dst' and 'dst_domain' ACLs.
     */
    ch.src_addr = http->conn->peer.sin_addr;
    ch.my_addr = http->conn->me.sin_addr;
    ch.request = http->request;
    /*
     * aclCheckFast returns 1 for ALLOW and 0 for DENY.  The default
     * is ALLOW, so we require 'no_cache DENY foo' in squid.conf
     * to indicate uncachable objects.
     */
    if (!aclCheckFast(Config.accessList.noCache, &ch))
	return 0;
    if (req->protocol == PROTO_HTTP)
	return httpCachable(method);
    /* FTP is always cachable */
    if (req->protocol == PROTO_GOPHER)
	return gopherCachable(url);
    if (req->protocol == PROTO_WAIS)
	return 0;
    if (method == METHOD_CONNECT)
	return 0;
    if (method == METHOD_TRACE)
	return 0;
    if (req->protocol == PROTO_CACHEOBJ)
	return 0;
    return 1;
}

/* Return true if we can query our neighbors for this object */
static int
clientHierarchical(clientHttpRequest * http)
{
    const char *url = http->uri;
    request_t *request = http->request;
    method_t method = request->method;
    const wordlist *p = NULL;

    /* IMS needs a private key, so we can use the hierarchy for IMS only
     * if our neighbors support private keys */
    if (request->flags.ims && !neighbors_do_private_keys)
	return 0;
    if (request->flags.auth)
	return 0;
    if (method == METHOD_TRACE)
	return 1;
    if (method != METHOD_GET)
	return 0;
    /* scan hierarchy_stoplist */
    for (p = Config.hierarchy_stoplist; p; p = p->next)
	if (strstr(url, p->key))
	    return 0;
    if (request->flags.loopdetect)
	return 0;
    if (request->protocol == PROTO_HTTP)
	return httpCachable(method);
    if (request->protocol == PROTO_GOPHER)
	return gopherCachable(url);
    if (request->protocol == PROTO_WAIS)
	return 0;
    if (request->protocol == PROTO_CACHEOBJ)
	return 0;
    return 1;
}

int
isTcpHit(log_type code)
{
    /* this should be a bitmap for better optimization */
    if (code == LOG_TCP_HIT)
	return 1;
    if (code == LOG_TCP_IMS_HIT)
	return 1;
    if (code == LOG_TCP_REFRESH_FAIL_HIT)
	return 1;
    if (code == LOG_TCP_REFRESH_HIT)
	return 1;
    if (code == LOG_TCP_NEGATIVE_HIT)
	return 1;
    if (code == LOG_TCP_MEM_HIT)
	return 1;
    if (code == LOG_TCP_OFFLINE_HIT)
	return 1;
    return 0;
}

/*
 * returns true if If-Range specs match reply, false otherwise
 */
static int
clientIfRangeMatch(clientHttpRequest * http, HttpReply * rep)
{
    const TimeOrTag spec = httpHeaderGetTimeOrTag(&http->request->header, HDR_IF_RANGE);
    /* check for parsing falure */
    if (!spec.valid)
	return 0;
    /* got an ETag? */
    if (spec.tag.str) {
	ETag rep_tag = httpHeaderGetETag(&rep->header, HDR_ETAG);
	debug(33, 3) ("clientIfRangeMatch: ETags: %s and %s\n",
	    spec.tag.str, rep_tag.str ? rep_tag.str : "<none>");
	if (!rep_tag.str)
	    return 0;		/* entity has no etag to compare with! */
	if (spec.tag.weak || rep_tag.weak) {
	    debug(33, 1) ("clientIfRangeMatch: Weak ETags are not allowed in If-Range: %s ? %s\n",
		spec.tag.str, rep_tag.str);
	    return 0;		/* must use strong validator for sub-range requests */
	}
	return etagIsEqual(&rep_tag, &spec.tag);
    }
    /* got modification time? */
    if (spec.time >= 0) {
	return http->entry->lastmod <= spec.time;
    }
    assert(0);			/* should not happen */
    return 0;
}

/* adds appropriate Range headers if needed */
static void
clientBuildRangeHeader(clientHttpRequest * http, HttpReply * rep)
{
    HttpHeader *hdr = rep ? &rep->header : 0;
    const char *range_err = NULL;
    assert(http->request->range);
    /* check if we still want to do ranges */
    if (!rep)
	range_err = "no [parse-able] reply";
    else if (rep->sline.status != HTTP_OK)
	range_err = "wrong status code";
    else if (httpHeaderHas(hdr, HDR_CONTENT_RANGE))
	range_err = "origin server does ranges";
    else if (rep->content_length < 0)
	range_err = "unknown length";
    else if (rep->content_length != http->entry->mem_obj->reply->content_length)
	range_err = "INCONSISTENT length";	/* a bug? */
    else if (httpHeaderHas(&http->request->header, HDR_IF_RANGE) && !clientIfRangeMatch(http, rep))
	range_err = "If-Range match failed";
    else if (!httpHdrRangeCanonize(http->request->range, rep->content_length))
	range_err = "canonization failed";
    else if (httpHdrRangeIsComplex(http->request->range))
	range_err = "too complex range header";
    /* get rid of our range specs on error */
    if (range_err) {
	debug(33, 2) ("clientBuildRangeHeader: will not do ranges: %s.\n", range_err);
	httpHdrRangeDestroy(http->request->range);
	http->request->range = NULL;
    } else {
	const int spec_count = http->request->range->specs.count;
	debug(33, 2) ("clientBuildRangeHeader: range spec count: %d clen: %d\n",
	    spec_count, rep->content_length);
	assert(spec_count > 0);
	/* ETags should not be returned with Partial Content replies? */
	httpHeaderDelById(hdr, HDR_ETAG);
	/* append appropriate header(s) */
	if (spec_count == 1) {
	    HttpHdrRangePos pos = HttpHdrRangeInitPos;
	    const HttpHdrRangeSpec *spec = httpHdrRangeGetSpec(http->request->range, &pos);
	    assert(spec);
	    /* append Content-Range */
	    httpHeaderAddContRange(hdr, *spec, rep->content_length);
	    /* set new Content-Length to the actual number of OCTETs
	     * transmitted in the message-body */
	    httpHeaderDelById(hdr, HDR_CONTENT_LENGTH);
	    httpHeaderPutInt(hdr, HDR_CONTENT_LENGTH, spec->length);
	    debug(33, 2) ("clientBuildRangeHeader: actual content length: %d\n", spec->length);
	} else {
	    /* multipart! */
	    /* generate boundary string */
	    http->range_iter.boundary = httpHdrRangeBoundaryStr(http);
	    /* delete old Content-Type, add ours */
	    httpHeaderDelById(hdr, HDR_CONTENT_TYPE);
	    httpHeaderPutStrf(hdr, HDR_CONTENT_TYPE,
		"multipart/byteranges; boundary=\"%s\"",
		strBuf(http->range_iter.boundary));
	    /* no need for Content-Length in multipart responses */
	    /* but we must delete the original one if we cannot (yet)
	     * calculate the actual length */
	    httpHeaderDelById(hdr, HDR_CONTENT_LENGTH);
	}
    }
}

/* filters out unwanted entries from original reply header
 * adds extra entries if we have more info than origin server
 * adds Squid specific entries */
static void
clientBuildReplyHeader(clientHttpRequest * http, HttpReply * rep)
{
    HttpHeader *hdr = &rep->header;
    int is_hit = isTcpHit(http->log_type);
    request_t *request = http->request;
#if DONT_FILTER_THESE
    /* but you might want to if you run Squid as an HTTP accelerator */
    /* httpHeaderDelById(hdr, HDR_ACCEPT_RANGES); */
    httpHeaderDelById(hdr, HDR_ETAG);
#endif
    httpHeaderDelById(hdr, HDR_PROXY_CONNECTION);
    /* here: Keep-Alive is a field-name, not a connection directive! */
    httpHeaderDelByName(hdr, "Keep-Alive");
    /* remove Set-Cookie if a hit */
    if (is_hit)
	httpHeaderDelById(hdr, HDR_SET_COOKIE);
    /* handle Connection header */
    if (httpHeaderHas(hdr, HDR_CONNECTION)) {
	/* anything that matches Connection list member will be deleted */
	String strConnection = httpHeaderGetList(hdr, HDR_CONNECTION);
	const HttpHeaderEntry *e;
	HttpHeaderPos pos = HttpHeaderInitPos;
	/*
	 * think: on-average-best nesting of the two loops (hdrEntry
	 * and strListItem) @?@
	 */
	/*
	 * maybe we should delete standard stuff ("keep-alive","close")
	 * from strConnection first?
	 */
	while ((e = httpHeaderGetEntry(hdr, &pos))) {
	    if (strListIsMember(&strConnection, strBuf(e->name), ','))
		httpHeaderDelAt(hdr, pos);
	}
	httpHeaderDelById(hdr, HDR_CONNECTION);
	stringClean(&strConnection);
    }
    /* Handle Ranges */
    if (request->range)
	clientBuildRangeHeader(http, rep);
    /*
     * Add Age header, not that our header must replace Age headers
     * from other caches if any
     */
    if (http->entry->timestamp > 0) {
	httpHeaderDelById(hdr, HDR_AGE);
	/*
	 * we do not follow HTTP/1.1 precisely here becuase we rely
	 * on Date header when computing entry->timestamp; we should
	 * be using _request_ time if Date header is not available
	 * or if it is out of sync
	 */
	httpHeaderPutInt(hdr, HDR_AGE,
	    http->entry->timestamp <= squid_curtime ?
	    squid_curtime - http->entry->timestamp : 0);
    }
    /* Append X-Cache */
    httpHeaderPutStrf(hdr, HDR_X_CACHE, "%s from %s",
	is_hit ? "HIT" : "MISS", getMyHostname());
#if USE_CACHE_DIGESTS
    /* Append X-Cache-Lookup: -- temporary hack, to be removed @?@ @?@ */
    httpHeaderPutStrf(hdr, HDR_X_CACHE_LOOKUP, "%s from %s:%d",
	http->lookup_type ? http->lookup_type : "NONE",
	getMyHostname(), Config.Port.http->i);
#endif
    /*
     * Clear keepalive for NON-HEAD requests with invalid content length
     */
    if (request->method != METHOD_HEAD)
	if (http->entry->mem_obj->reply->content_length < 0)
	    request->flags.proxy_keepalive = 0;
    /* Signal keep-alive if needed */
    httpHeaderPutStr(hdr,
	http->flags.accel ? HDR_CONNECTION : HDR_PROXY_CONNECTION,
	request->flags.proxy_keepalive ? "keep-alive" : "close");
#if ADD_X_REQUEST_URI
    /*
     * Knowing the URI of the request is useful when debugging persistent
     * connections in a client; we cannot guarantee the order of http headers,
     * but X-Request-URI is likely to be the very last header to ease use from a
     * debugger [hdr->entries.count-1].
     */
    httpHeaderPutStr(hdr, HDR_X_REQUEST_URI,
	http->entry->mem_obj->url ? http->entry->mem_obj->url : http->uri);
#endif
}

static HttpReply *
clientBuildReply(clientHttpRequest * http, const char *buf, size_t size)
{
    HttpReply *rep = httpReplyCreate();
    if (httpReplyParse(rep, buf)) {
	/* enforce 1.0 reply version */
	rep->sline.version = 1.0;
	/* do header conversions */
	clientBuildReplyHeader(http, rep);
	/* if we do ranges, change status to "Partial Content" */
	if (http->request->range)
	    httpStatusLineSet(&rep->sline, rep->sline.version, HTTP_PARTIAL_CONTENT, NULL);
    } else {
	/* parsing failure, get rid of the invalid reply */
	httpReplyDestroy(rep);
	rep = NULL;
	/* if we were going to do ranges, backoff */
	if (http->request->range)
	    clientBuildRangeHeader(http, rep);	/* will fail and destroy request->range */
    }
    return rep;
}

/*
 * clientCacheHit should only be called until the HTTP reply headers
 * have been parsed.  Normally this should be a single call, but
 * it might take more than one.  As soon as we have the headers,
 * we hand off to clientSendMoreData, clientProcessExpired, or
 * clientProcessMiss.
 */
static void
clientCacheHit(void *data, char *buf, ssize_t size)
{
    clientHttpRequest *http = data;
    StoreEntry *e = http->entry;
    MemObject *mem;
    request_t *r = http->request;
    debug(33, 3) ("clientCacheHit: %s, %d bytes\n", http->uri, (int) size);
    if (http->entry == NULL) {
	memFree(buf, MEM_CLIENT_SOCK_BUF);
	debug(33, 3) ("clientCacheHit: request aborted\n");
	return;
    } else if (size < 0) {
	/* swap in failure */
	memFree(buf, MEM_CLIENT_SOCK_BUF);
	debug(33, 3) ("clientCacheHit: swapin failure for %s\n", http->uri);
	http->log_type = LOG_TCP_SWAPFAIL_MISS;
	if ((e = http->entry)) {
	    http->entry = NULL;
	    storeUnregister(e, http);
	    storeUnlockObject(e);
	}
	clientProcessMiss(http);
	return;
    }
    assert(size > 0);
    mem = e->mem_obj;
    assert(!EBIT_TEST(e->flags, ENTRY_ABORTED));
    if (mem->reply->sline.status == 0) {
	/*
	 * we don't have full reply headers yet; either wait for more or
	 * punt to clientProcessMiss.
	 */
	if (e->mem_status == IN_MEMORY || e->store_status == STORE_OK) {
	    memFree(buf, MEM_CLIENT_SOCK_BUF);
	    clientProcessMiss(http);
	} else if (size == CLIENT_SOCK_SZ && http->out.offset == 0) {
	    memFree(buf, MEM_CLIENT_SOCK_BUF);