When breaking early out of the upstream response body loop, we
must ensure we know we've hit the parser body EOF and preserve
the buffer for trailer parsing. Otherwise, reentering
proxy_read_body will put us in a bad state and corrupt
responses.
This is a critical bugfix which only affects users of
the soon-to-be-released "proxy_buffering: false" feature
of proxy_pass.
---
Fwiw, I discovered this using dumb git HTTP clones in a dumb
configuration:
yahns (proxy_pass) -> varnish -> dumb HTTP server
GIT_SMART_HTTP=0 git -c http.maxRequests=100 clone $URL_OF_YAHNS
But it's a decent HTTP server test for finding bugs like these
since git checksums object responses to detect corruption.
And damn, dumb HTTP in git is really slow at cloning a 750M
repo (I guess one problem is the tree layout is really bad
for this...) :<
lib/yahns/proxy_http_response.rb | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/lib/yahns/proxy_http_response.rb b/lib/yahns/proxy_http_response.rb
index 316c310..0ca2c34 100644
--- a/lib/yahns/proxy_http_response.rb
+++ b/lib/yahns/proxy_http_response.rb
@@ -167,7 +167,10 @@ def proxy_read_body(tip, kcar, req_res)
end
wbuf = proxy_write(wbuf, tmp, req_res)
chunk.clear if chunk
- return proxy_unbuffer(wbuf) if Yahns::WbufLite === wbuf
+ if Yahns::WbufLite === wbuf
+ req_res.proxy_trailers = [ rbuf.dup, tip ] if chunk && kcar.body_eof?
+ return proxy_unbuffer(wbuf)
+ end
when nil # EOF
# HTTP/1.1 upstream, unexpected premature EOF:
msg = "upstream EOF (#{len} bytes left)" if len
--
unsubscribe: [email protected]
archive: https://yhbt.net/yahns-public/