From 9879db1195172e94e18f5054da6680c2db35c83c Mon Sep 17 00:00:00 2001
From: JustAnotherArchivist <JustAnotherArchivist@users.noreply.github.com>
Date: Wed, 3 May 2023 04:18:11 +0000
Subject: [PATCH] Proper HTTP/1.0 support

HTTP/1.0 does not mandate a Content-Length header in responses since keep-alive connections aren't a thing; the connection closure then signals the end of the response.
This change requires the URL metadata line for processing HTTP/1.0 data, so it plays well with `warc-dump-responses --meta`.
---
 http-response-bodies.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/http-response-bodies.c b/http-response-bodies.c
index 1ac302d..f7184eb 100644
--- a/http-response-bodies.c
+++ b/http-response-bodies.c
@@ -68,6 +68,8 @@ int main(int argc, char* argv[]) {
 	bool html_fake_base = false;
 	char* url = NULL; // Warning, pointer is only valid within the STATE_HEADERS block below.
 	size_t urllen;
+	bool have_response_length = false;
+	size_t response_length;
 
 	if (argc == 2 && strcmp(argv[1], "--html-fake-base") == 0) {
 		html_fake_base = true;
@@ -89,9 +91,10 @@ checkstate:
 				fprintf(stderr, "Error: too little data before HTTP headers\n");
 				return 1;
 			}
-			// Handle optional URL + Length line
+			// Handle URL + Length line; optional for HTTP/1.1 but required for HTTP/1.0
 			url = NULL;
 			urllen = 0;
+			response_length = 0;
 			if (memcmp(bufp, "HTTP/1.1 ", 9) != 0 && memcmp(bufp, "HTTP/1.0 ", 9) != 0) {
 				DEBUG_PRINTF("No HTTP header, looking for URL line\n");
 				m0 = memmem(bufp, n, "\n", 1);
@@ -100,13 +103,24 @@ checkstate:
 					return 1;
 				}
 				m1 = m0;
-				// Skip over length field, which we don't need.
+				// Skip back over length field
 				--m0;
 				while (m0 > bufp && '0' <= *m0 && *m0 <= '9') --m0;
 				if (*m0 != ' ') {
 					fprintf(stderr, "Error: URL line has unexpected format\n");
 					return 1;
 				}
+				// Read length
+				if (sscanf(m0, " %zu%ln", &response_length, &nscan) <= 0) {
+					fprintf(stderr, "Error: URL line contains no length\n");
+					return 1;
+				}
+				if (nscan != m1 - m0) {
+					fprintf(stderr, "Error: URL line length read mismatch\n");
+					return 1;
+				}
+				have_response_length = true;
+				DEBUG_PRINTF("Response length: %zu\n", response_length);
 				// Rest must now be the URL; check that there is a scheme and no CR, LF, or whitespace.
 				url = bufp;
 				urllen = m0 - bufp;
@@ -133,6 +147,10 @@ checkstate:
 				fprintf(stderr, "Error: --html-fake-base requires URL lines\n");
 				return 1;
 			}
+			if (memcmp(bufp, "HTTP/1.0 ", 9) == 0 && !have_response_length) {
+				fprintf(stderr, "Error: HTTP/1.0 requires URL metadata lines\n");
+				return 1;
+			}
 
 			if (n < 9) {
 				fprintf(stderr, "Error: too little data before HTTP headers\n");
@@ -148,8 +166,15 @@ checkstate:
 				eoh += 4;
 				DEBUG_PRINTF("Response body begins at %p (offset %zu)\n", (void*)eoh, eoh - bufp);
 
-				m0 = memcasemem(bufp, eoh - bufp, "\r\ncontent-length:", 17);
-				if (m0 && m0 < eoh) {
+				if (memcmp(bufp, "HTTP/1.0 ", 9) == 0) {
+					// HTTP 1.0 doesn't have TE, so just use the response_length for the content length and skip the other header parsing
+					if (bufp + response_length < eoh) {
+						fprintf(stderr, "Error: end of headers occurs after alleged response length\n");
+						return 1;
+					}
+					length = response_length - (eoh - bufp);
+					state = STATE_BODY;
+				} else if ((m0 = memcasemem(bufp, eoh - bufp, "\r\ncontent-length:", 17)) && m0 < eoh) {
 					DEBUG_PRINTF("Found Content-Length header at %p (offset %zu)\n", (void*)(m0 + 2), m0 + 2 - bufp);
 					m1 = memmem(m0 + 1, eoh - (m0 + 1), "\r\n", 2);
 					if (!m1) {