From: Willy Tarreau <w@1wt.eu>
Date: Wed, 19 Mar 2014 11:07:52 +0000 (+0100)
Subject: MEDIUM: compression: consider the "q=" attribute in Accept-Encoding
X-Git-Tag: v1.5-dev23~76
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=0e9b1b4d1f0efc5e46a10371d9be21e97581faab;p=thirdparty%2Fhaproxy.git

MEDIUM: compression: consider the "q=" attribute in Accept-Encoding

Till now we didn't consider "q=". It's problematic because the first
effect is that compression tokens were not even matched if it was
present.

It is important to parse it correctly because we still want to allow
a user-agent to send "q=0" to explicitly disable a compressor, or to
specify its preferences.

Now, q-values are respected in order of precedence, and when several
q-values are equal, the first occurrence is used.
---

diff --git a/src/proto_http.c b/src/proto_http.c
index 89e75b7172..0bca45cfe3 100644
--- a/src/proto_http.c
+++ b/src/proto_http.c
@@ -2144,6 +2144,38 @@ static inline int http_skip_chunk_crlf(struct http_msg *msg)
 	return 1;
 }
 
+/* Parses a qvalue and returns it multipled by 1000, from 0 to 1000. If the
+ * value is larger than 1000, it is bound to 1000. The parser consumes up to
+ * 1 digit, one dot and 3 digits and stops on the first invalid character.
+ * Unparsable qvalues return 1000 as "q=1.000".
+ */
+int parse_qvalue(const char *qvalue)
+{
+	int q = 1000;
+
+	if (!isdigit(*qvalue))
+		goto out;
+	q = (*qvalue++ - '0') * 1000;
+
+	if (*qvalue++ != '.')
+		goto out;
+
+	if (!isdigit(*qvalue))
+		goto out;
+	q += (*qvalue++ - '0') * 100;
+
+	if (!isdigit(*qvalue))
+		goto out;
+	q += (*qvalue++ - '0') * 10;
+
+	if (!isdigit(*qvalue))
+		goto out;
+	q += (*qvalue++ - '0') * 1;
+ out:
+	if (q > 1000)
+		q = 1000;
+	return q;
+}
 
 /*
  * Selects a compression algorithm depending on the client request.
@@ -2175,26 +2207,71 @@ int select_compression_request_header(struct session *s, struct buffer *req)
 
 	/* search for the algo in the backend in priority or the frontend */
 	if ((s->be->comp && (comp_algo_back = s->be->comp->algos)) || (s->fe->comp && (comp_algo_back = s->fe->comp->algos))) {
+		int best_q = 0;
+
 		ctx.idx = 0;
 		while (http_find_header2("Accept-Encoding", 15, req->p, &txn->hdr_idx, &ctx)) {
-			for (comp_algo = comp_algo_back; comp_algo; comp_algo = comp_algo->next) {
-				if (word_match(ctx.line + ctx.val, ctx.vlen, comp_algo->name, comp_algo->name_len)) {
-					s->comp_algo = comp_algo;
+			const char *qval;
+			int q;
+			int toklen;
+
+			/* try to isolate the token from the optional q-value */
+			toklen = 0;
+			while (toklen < ctx.vlen && http_is_token[(unsigned char)*(ctx.line + ctx.val + toklen)])
+				toklen++;
+
+			qval = ctx.line + ctx.val + toklen;
+			while (1) {
+				while (qval < ctx.line + ctx.val + ctx.vlen && http_is_lws[(unsigned char)*qval])
+					qval++;
+
+				if (qval >= ctx.line + ctx.val + ctx.vlen || *qval != ';') {
+					qval = NULL;
+					break;
+				}
+				qval++;
 
-					/* remove all occurrences of the header when "compression offload" is set */
+				while (qval < ctx.line + ctx.val + ctx.vlen && http_is_lws[(unsigned char)*qval])
+					qval++;
 
-					if ((s->be->comp && s->be->comp->offload) ||
-					    (s->fe->comp && s->fe->comp->offload)) {
-						http_remove_header2(msg, &txn->hdr_idx, &ctx);
-						ctx.idx = 0;
-						while (http_find_header2("Accept-Encoding", 15, req->p, &txn->hdr_idx, &ctx)) {
-							http_remove_header2(msg, &txn->hdr_idx, &ctx);
-						}
-					}
-					return 1;
+				if (qval >= ctx.line + ctx.val + ctx.vlen) {
+					qval = NULL;
+					break;
 				}
+				if (strncmp(qval, "q=", MIN(ctx.line + ctx.val + ctx.vlen - qval, 2)) == 0)
+					break;
+
+				while (qval < ctx.line + ctx.val + ctx.vlen && *qval != ';')
+					qval++;
+			}
+
+			/* here we have qval pointing to the first "q=" attribute or NULL if not found */
+			q = qval ? parse_qvalue(qval + 2) : 1000;
+
+			if (q <= best_q)
+				continue;
+
+			for (comp_algo = comp_algo_back; comp_algo; comp_algo = comp_algo->next) {
+				if (*(ctx.line + ctx.val) == '*' ||
+				    word_match(ctx.line + ctx.val, toklen, comp_algo->name, comp_algo->name_len)) {
+					s->comp_algo = comp_algo;
+					best_q = q;
+					break;
+				}
+			}
+		}
+	}
+
+	/* remove all occurrences of the header when "compression offload" is set */
+	if (s->comp_algo) {
+		if ((s->be->comp && s->be->comp->offload) || (s->fe->comp && s->fe->comp->offload)) {
+			http_remove_header2(msg, &txn->hdr_idx, &ctx);
+			ctx.idx = 0;
+			while (http_find_header2("Accept-Encoding", 15, req->p, &txn->hdr_idx, &ctx)) {
+				http_remove_header2(msg, &txn->hdr_idx, &ctx);
 			}
 		}
+		return 1;
 	}
 
 	/* identity is implicit does not require headers */