Project

General

Profile

Bug #14972

Net::HTTP inconsistently raises EOFError when peer closes the connection

Added by joshc (Josh C) 4 months ago. Updated 3 months ago.

Status:
Open
Priority:
Normal
Assignee:
-
Target version:
-
ruby -v:
ruby 2.4.3p205 (2017-12-14 revision 61247) [x86_64-darwin15]
[ruby-core:88324]

Description

If chunked transfer encoding is used, and the peer closes the connection while the caller is reading data, then the Net::HTTP::Response#read_body method will raise EOFError. If chunked transfer encoding is not used (and an explicit Content-Length is used instead), the read_body method swallows the EOFError exception. I would expect read_body to raise EOFError if it reads fewer than Content-Length bytes.

The current behavior is explained by the ignore_eof parameter in https://github.com/ruby/ruby/blob/v2_4_3/lib/net/http/response.rb#L284-L301. However, RFC 7230 section 3.3.3 https://tools.ietf.org/html/rfc7230#section-3.3.3 says:

   5.  If a valid Content-Length header field is present without
       Transfer-Encoding, its decimal value defines the expected message
       body length in octets.  If the sender closes the connection or
       the recipient times out before the indicated number of octets are
       received, the recipient MUST consider the message to be
       incomplete and close the connection.

As it is now, if chunked encoding is not used, then the caller is unaware when the response body is truncated. In order to detect it, the caller must count the number of bytes read until Content-Length is reached. However, that means you can't use ruby's automatic decompression, because Content-Length is the number of compressed bytes, while read_body yields chunks of uncompressed data.

Here's sample code to reproduce. Run the following http server. Note chunked is currently false, but can be toggled.

require 'webrick'

server = WEBrick::HTTPServer.new :Port => 8000
trap 'INT' do
  server.shutdown
end

# toggle this
chunked = false

server.mount_proc '/' do |req, res|
  res.status = 200
  res['Content-Type'] = 'text/plain'

  str = "0123456789" * 10000
  res.body = str
  if chunked
    res.chunked = true
  else
    res['Content-Length'] = str.length
  end
end

server.start

Run the following http client code. In order to simulate a closed connection, the block raises EOFError.

require 'net/http'
require 'uri'

uri = URI("http://localhost:8000/")
Net::HTTP.start(uri.host, uri.port) do |http|
  http.request_get(uri.path) do |response|
    response.read_body do |chunk|
      puts "Read #{chunk.length} bytes"
      raise EOFError.new("whoops")
    end
  end
end
puts "EOF was silently caught"

When chunked encoding is used, the exception is properly raised. I believe ruby is retrying the request because GET is idempotent:

$ ruby --version
ruby 2.4.3p205 (2017-12-14 revision 61247) [x86_64-darwin15]
$ ruby client.rb
Content-Length:
Transfer-Encoding: chunked
Read 16377 bytes
Content-Length:
Transfer-Encoding: chunked
Read 16377 bytes
client.rb:11:in `block (3 levels) in <main>': whoops (EOFError)
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/protocol.rb:429:in `call_block'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/protocol.rb:420:in `<<'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/protocol.rb:122:in `read'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:322:in `read_chunked'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:286:in `block in read_body_0'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:278:in `inflater'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:283:in `read_body_0'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:204:in `read_body'
    from client.rb:9:in `block (2 levels) in <main>'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:1455:in `block in transport_request'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http/response.rb:165:in `reading_body'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:1454:in `transport_request'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:1416:in `request'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:1317:in `request_get'
    from client.rb:6:in `block in <main>'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:877:in `start'
    from /usr/local/opt/rbenv/versions/2.4.3/lib/ruby/2.4.0/net/http.rb:608:in `start'
    from client.rb:5:in `<main>'

When chunked encoding is not used, the exception is not raised:

ruby client.rb
Content-Length: 100000
Transfer-Encoding:
Read 0 bytes
EOF was silently caught

I verified the behavior exists as far back as ruby 1.9.3p551. It was introduced in https://github.com/ruby/ruby/commit/cdc7602379c9d911983db2c044d69ac417869266#diff-8c2ab8e0fb4f052e1d95ab6334e192c1R949.

History

#1 Updated by joshc (Josh C) 4 months ago

  • Description updated (diff)

#2 [ruby-core:88512] Updated by naruse (Yui NARUSE) 4 months ago

What is the ideal behavior you think? Just below?

diff --git a/lib/net/http/response.rb b/lib/net/http/response.rb
index 66132985d9..7c744d02f4 100644
--- a/lib/net/http/response.rb
+++ b/lib/net/http/response.rb
@@ -290,7 +290,7 @@ def read_body_0(dest)

       clen = content_length()
       if clen
-        @socket.read clen, dest, true   # ignore EOF
+        @socket.read clen, dest
         return
       end
       clen = range_length()

#3 [ruby-core:89111] Updated by joshc (Josh C) 3 months ago

If @socket.read clen, dest reads fully clen bytes then that seems ok. But if it can read fewer than clen bytes, then we should keep reading until we read clen bytes or reach EOF.

Also available in: Atom PDF