Improve Transfer-Encoding handling (#702)

Reject messages containing Transfer-Encoding header with coding other
than chunked or identity. Squid does not support other codings.

For simplicity and security sake, also reject messages where
Transfer-Encoding contains unnecessary complex values that are
technically equivalent to "chunked" or "identity" (e.g., ",,chunked" or
"identity, chunked").

RFC 7230 formally deprecated and removed identity coding, but it is
still used by some agents.

diff --git a/src/HttpHeader.cc b/src/HttpHeader.cc
index 52f7ae8ab0b5f09387cb6f001facccdfdfb8e56e..786a8dc1440352541ded084521606f674579b227 100644 (file)
--- a/src/HttpHeader.cc
+++ b/src/HttpHeader.cc
@@ -181,6 +181,7 @@ HttpHeader::operator =(const HttpHeader &other)
         update(&other); // will update the mask as well
         len = other.len;
         conflictingContentLength_ = other.conflictingContentLength_;
+        teUnsupported_ = other.teUnsupported_;
     }
     return *this;
 }
@@ -229,6 +230,7 @@ HttpHeader::clean()
     httpHeaderMaskInit(&mask, 0);
     len = 0;
     conflictingContentLength_ = false;
+    teUnsupported_ = false;
     PROF_stop(HttpHeaderClean);
 }
 
@@ -525,18 +527,40 @@ HttpHeader::parse(const char *header_start, size_t hdrLen, Http::ContentLengthIn
                Raw("header", header_start, hdrLen));
     }
 
+    String rawTe;
     if (clen.prohibitedAndIgnored()) {
+        // prohibitedAndIgnored() includes trailer header blocks
+        // being parsed as a case to forbid/ignore these headers.
+
         // RFC 7230 section 3.3.2: A server MUST NOT send a Content-Length
         // header field in any response with a status code of 1xx (Informational)
         // or 204 (No Content). And RFC 7230 3.3.3#1 tells recipients to ignore
         // such Content-Lengths.
         if (delById(Http::HdrType::CONTENT_LENGTH))
             debugs(55, 3, "Content-Length is " << clen.prohibitedAndIgnored());
-    } else if (chunked()) {
+
+        // RFC 7230 section 3.3.1 has the same criteria forbid Transfer-Encoding
+        if (delById(Http::HdrType::TRANSFER_ENCODING)) {
+            debugs(55, 3, "Transfer-Encoding is " << clen.prohibitedAndIgnored());
+            teUnsupported_ = true;
+        }
+
+    } else if (getByIdIfPresent(Http::HdrType::TRANSFER_ENCODING, &rawTe)) {
         // RFC 2616 section 4.4: ignore Content-Length with Transfer-Encoding
         // RFC 7230 section 3.3.3 #3: Transfer-Encoding overwrites Content-Length
         delById(Http::HdrType::CONTENT_LENGTH);
         // and clen state becomes irrelevant
+
+        if (rawTe == "chunked") {
+            ; // leave header present for chunked() method
+        } else if (rawTe == "identity") { // deprecated. no coding
+            delById(Http::HdrType::TRANSFER_ENCODING);
+        } else {
+            // This also rejects multiple encodings until we support them properly.
+            debugs(55, warnOnError, "WARNING: unsupported Transfer-Encoding used by client: " << rawTe);
+            teUnsupported_ = true;
+        }
+
     } else if (clen.sawBad) {
         // ensure our callers do not accidentally see bad Content-Length values
         delById(Http::HdrType::CONTENT_LENGTH);

diff --git a/src/HttpHeader.h b/src/HttpHeader.h
index 8ffe7ba55af1035f0944d742dba0b1f33376a8b9..7af47847589b7140284182c5eeddc4631c92340a 100644 (file)
--- a/src/HttpHeader.h
+++ b/src/HttpHeader.h
@@ -159,7 +159,13 @@ public:
     int hasListMember(Http::HdrType id, const char *member, const char separator) const;
     int hasByNameListMember(const char *name, const char *member, const char separator) const;
     void removeHopByHopEntries();
-    inline bool chunked() const; ///< whether message uses chunked Transfer-Encoding
+
+    /// whether the message uses chunked Transfer-Encoding
+    /// optimized implementation relies on us rejecting/removing other codings
+    bool chunked() const { return has(Http::HdrType::TRANSFER_ENCODING); }
+
+    /// whether message used an unsupported and/or invalid Transfer-Encoding
+    bool unsupportedTe() const { return teUnsupported_; }
 
     /* protected, do not use these, use interface functions instead */
     std::vector<HttpHeaderEntry*, PoolingAllocator<HttpHeaderEntry*> > entries; /**< parsed fields in raw format */
@@ -183,6 +189,9 @@ protected:
 private:
     HttpHeaderEntry *findLastEntry(Http::HdrType id) const;
     bool conflictingContentLength_; ///< found different Content-Length fields
+    /// unsupported encoding, unnecessary syntax characters, and/or
+    /// invalid field-value found in Transfer-Encoding header
+    bool teUnsupported_ = false;
 };
 
 int httpHeaderParseQuotedString(const char *start, const int len, String *val);
@@ -192,13 +201,6 @@ SBuf httpHeaderQuoteString(const char *raw);
 
 void httpHeaderCalcMask(HttpHeaderMask * mask, Http::HdrType http_hdr_type_enums[], size_t count);
 
-inline bool
-HttpHeader::chunked() const
-{
-    return has(Http::HdrType::TRANSFER_ENCODING) &&
-           hasListMember(Http::HdrType::TRANSFER_ENCODING, "chunked", ',');
-}
-
 void httpHeaderInitModule(void);
 
 #endif /* SQUID_HTTPHEADER_H */

diff --git a/src/client_side.cc b/src/client_side.cc
index a026a114a052067e4ba3a8b77d05533178e380fc..e4f0d825af9cdff1052793bfad892ea97f1d9320 100644 (file)
--- a/src/client_side.cc
+++ b/src/client_side.cc
@@ -1607,9 +1607,7 @@ void
 clientProcessRequest(ConnStateData *conn, const Http1::RequestParserPointer &hp, Http::Stream *context)
 {
     ClientHttpRequest *http = context->http;
-    bool chunked = false;
     bool mustReplyToOptions = false;
-    bool unsupportedTe = false;
     bool expectBody = false;
 
     // We already have the request parsed and checked, so we
@@ -1666,13 +1664,7 @@ clientProcessRequest(ConnStateData *conn, const Http1::RequestParserPointer &hp,
         request->http_ver.minor = http_ver.minor;
     }
 
-    if (request->header.chunked()) {
-        chunked = true;
-    } else if (request->header.has(Http::HdrType::TRANSFER_ENCODING)) {
-        const String te = request->header.getList(Http::HdrType::TRANSFER_ENCODING);
-        // HTTP/1.1 requires chunking to be the last encoding if there is one
-        unsupportedTe = te.size() && te != "identity";
-    } // else implied identity coding
+    const auto unsupportedTe = request->header.unsupportedTe();
 
     mustReplyToOptions = (request->method == Http::METHOD_OPTIONS) &&
                          (request->header.getInt64(Http::HdrType::MAX_FORWARDS) == 0);
@@ -1689,6 +1681,7 @@ clientProcessRequest(ConnStateData *conn, const Http1::RequestParserPointer &hp,
         return;
     }
 
+    const auto chunked = request->header.chunked();
     if (!chunked && !clientIsContentLengthValid(request.getRaw())) {
         clientStreamNode *node = context->getClientReplyContext();
         clientReplyContext *repContext = dynamic_cast<clientReplyContext *>(node->data.getRaw());

diff --git a/src/http.cc b/src/http.cc
index 4c0f337810f6947f50db146e477bc0fc16ff719e..14907fff8e95d446fb2b0f9ace13bc62d71a0965 100644 (file)
--- a/src/http.cc
+++ b/src/http.cc
@@ -1380,6 +1380,9 @@ HttpStateData::continueAfterParsingHeader()
             } else if (vrep->header.conflictingContentLength()) {
                 fwd->dontRetry(true);
                 error = ERR_INVALID_RESP;
+            } else if (vrep->header.unsupportedTe()) {
+                fwd->dontRetry(true);
+                error = ERR_INVALID_RESP;
             } else {
                 return true; // done parsing, got reply, and no error
             }