Improve compliance with RFC 9112 for status line parsing

This patch improves compliance with RFC 9112 Section 4 for parsing the
status-line of the HTTP response.

- Require the HTTP version to be in the format "HTTP/x.y"
  - Disallow cases like "http/1.1" (lowercase "http"), "HTTP" (missing
    version), "HTTP/10.23" (multidigit versions)
- Require a space between the status-code and the reason-phrase, even
  when the reason-phrase is empty
  - Disallow cases like "HTTP/1.1 200" (missing space after status code)
- Make clear that SP parsing operates on the lenient behavior, which
  allows 'HTAB, VT (%x0B), FF (%x0C), or bare CR' in addition to space

Author: osyoyu

lib/net/http/response.rb

@@ -157,7 +157,7 @@ def read_new(sock)   #:nodoc: internal use only
 
     def read_status_line(sock)
       str = sock.readline
-      m = /\AHTTP(?:\/(\d+\.\d+))?\s+(\d\d\d)(?:\s+(.*))?\z/in.match(str) or
+      m = /\AHTTP\/(\d\.\d)[ \t\v\f\r]+(\d\d\d)[ \t\v\f\r]+(.*)\z/n.match(str) or
         raise Net::HTTPBadResponse, "wrong status line: #{str.dump}"
       m.captures
     end

test/net/http/test_httpresponse.rb

@@ -639,9 +639,24 @@ def test_uri_equals
     assert_not_same  uri, response.uri
   end
 
-  def test_ensure_zero_space_does_not_regress
+  def test_normal_status_line
+    io = dummy_io(<<EOS)
+HTTP/1.1 200 OK
+Content-Length: 5
+Connection: close
+
+hello
+EOS
+
+    res = Net::HTTPResponse.read_new(io)
+    assert_equal('1.1', res.http_version)
+    assert_equal('200', res.code)
+    assert_equal('OK', res.message)
+  end
+
+  def test_reject_status_line_without_version
     io = dummy_io(<<EOS)
-HTTP/1.1 200OK
+HTTP 200 OK
 Content-Length: 5
 Connection: close
 
@@ -653,24 +668,67 @@ def test_ensure_zero_space_does_not_regress
     end
   end
 
-  def test_allow_trailing_space_after_status
+  def test_reject_status_line_with_multidigit_version
     io = dummy_io(<<EOS)
-HTTP/1.1 200\s
+HTTP/1.10 200 OK
 Content-Length: 5
 Connection: close
 
 hello
 EOS
 
-    res = Net::HTTPResponse.read_new(io)
-    assert_equal('1.1', res.http_version)
-    assert_equal('200', res.code)
-    assert_equal('', res.message)
+    assert_raise Net::HTTPBadResponse do
+      Net::HTTPResponse.read_new(io)
+    end
   end
 
-  def test_normal_status_line
+  def test_reject_lowercase_http_name
     io = dummy_io(<<EOS)
-HTTP/1.1 200 OK
+http/1.1 200 OK
+Content-Length: 5
+Connection: close
+
+hello
+EOS
+
+    assert_raise Net::HTTPBadResponse do
+      Net::HTTPResponse.read_new(io)
+    end
+  end
+
+  def test_reject_status_line_with_four_digit_code
+    io = dummy_io(<<EOS)
+HTTP/1.1 2000 OK
+Content-Length: 5
+Connection: close
+
+hello
+EOS
+
+    assert_raise Net::HTTPBadResponse do
+      Net::HTTPResponse.read_new(io)
+    end
+  end
+
+  def test_reject_status_line_with_non_numeric_code
+    io = dummy_io(<<EOS)
+HTTP/1.1 2oo OK
+Content-Length: 5
+Connection: close
+
+hello
+EOS
+
+    assert_raise Net::HTTPBadResponse do
+      Net::HTTPResponse.read_new(io)
+    end
+  end
+
+  def test_allow_empty_reason_phrase
+    # RFC 9112 allows the reason-phrase itself to be empty, but the SP preceding it is required
+    # Note the trailing space after the status code (200)
+    io = dummy_io(<<EOS)
+HTTP/1.1 200 
 Content-Length: 5
 Connection: close
 
@@ -680,38 +738,51 @@ def test_normal_status_line
     res = Net::HTTPResponse.read_new(io)
     assert_equal('1.1', res.http_version)
     assert_equal('200', res.code)
-    assert_equal('OK', res.message)
+    assert_equal('', res.message)
   end
 
-  def test_allow_empty_reason_code
+  def test_reject_empty_reason_phrase_without_preceding_space
     io = dummy_io(<<EOS)
 HTTP/1.1 200
 Content-Length: 5
 Connection: close
 
+hello
+EOS
+
+    assert_raise Net::HTTPBadResponse do
+      Net::HTTPResponse.read_new(io)
+    end
+  end
+
+  def test_allow_whitespace_chars_as_separator
+    io = dummy_io(<<EOS)
+HTTP/1.1\t200\r OK
+Content-Length: 5
+Connection: close
+
 hello
 EOS
 
     res = Net::HTTPResponse.read_new(io)
     assert_equal('1.1', res.http_version)
     assert_equal('200', res.code)
-    assert_equal(nil, res.message)
+    assert_equal('OK', res.message)
   end
 
-  def test_raises_exception_with_missing_reason
+  def test_allow_multiple_space_between_version_and_code
     io = dummy_io(<<EOS)
-HTTP/1.1 404
+HTTP/1.1  200 OK
 Content-Length: 5
 Connection: close
 
 hello
 EOS
 
     res = Net::HTTPResponse.read_new(io)
-    assert_equal(nil, res.message)
-    assert_raise Net::HTTPClientException do
-      res.error!
-    end
+    assert_equal('1.1', res.http_version)
+    assert_equal('200', res.code)
+    assert_equal('OK', res.message)
   end
 
   def test_read_code_type