HTTP/2: curl hangs with upload if server sends response early and resets stream

[tatsuhiro-t]

I did this

Run h2 server. With --early-response option, this server returns HTTP response without waiting for upload completion. Additionally, it sends RST_STREAM if it expects request body from client:

$ nghttpd 8443 --no-tls -v --early-response

Then do some large upload from curl:

$ curl http://localhost:8443 --http2-prior-knowledge -T 100m  -v
*   Trying ::1...
* TCP_NODELAY set
* Connected to localhost (::1) port 8443 (#0)
* Using HTTP2, server supports multi-use
* Connection state changed (HTTP/2 confirmed)
* Copying HTTP/2 data in stream buffer to connection buffer after upgrade: len=0
* Using Stream ID: 1 (easy handle 0xf91f30)
> PUT /100m HTTP/1.1
> Host: localhost:8443
> User-Agent: curl/7.50.2-DEV
> Accept: */*
> Content-Length: 104857600
> 
* Connection state changed (MAX_CONCURRENT_STREAMS updated)!
< HTTP/2 200 
< server: nghttpd nghttp2/1.15.0-DEV
< cache-control: max-age=3600
< date: Sun, 28 Aug 2016 08:49:30 GMT
< content-length: 0
< last-modified: Sun, 28 Aug 2016 08:32:37 GMT
< 

And it stuck.

Background info: nghttp2/nghttp2#669

I expected the following

curl should detect stream closure, and exit. Since upload was not completed, with some error code (not sure which one we should use here).

curl/libcurl version

I build curl from git master

curl 7.50.2-DEV (x86_64-pc-linux-gnu) libcurl/7.50.2-DEV OpenSSL/1.0.2h zlib/1.2.8 libidn/1.32 libssh2/1.5.0 nghttp2/1.8.0-DEV
Protocols: dict file ftp ftps gopher http https imap imaps pop3 pop3s rtsp scp sftp smb smbs smtp smtps telnet tftp 
Features: IDN IPv6 Largefile NTLM NTLM_WB SSL libz TLS-SRP HTTP2 UnixSockets 

operating system

Linux

[tatsuhiro-t]

I attempted to fix this. The following patch seems to work, but I have not checked side effects.

diff --git a/lib/http.h b/lib/http.h
index 6529005..9fb669c 100644
--- a/lib/http.h
+++ b/lib/http.h
@@ -168,6 +168,7 @@ struct HTTP {
   const uint8_t *pausedata; /* pointer to data received in on_data_chunk */
   size_t pauselen; /* the number of bytes left in data */
   bool closed; /* TRUE on HTTP2 stream close */
+  bool close_handled; /* TRUE if stream closure is handled by libcurl */
   uint32_t error_code; /* HTTP/2 error code */

   char *mem;     /* points to a buffer in memory to store received data */
diff --git a/lib/http2.c b/lib/http2.c
index a14f75e..e611058 100644
--- a/lib/http2.c
+++ b/lib/http2.c
@@ -151,6 +151,7 @@ void Curl_http2_setup_req(struct Curl_easy *data)
   http->pauselen = 0;
   http->error_code = NGHTTP2_NO_ERROR;
   http->closed = FALSE;
+  http->close_handled = FALSE;
   http->mem = data->state.buffer;
   http->len = BUFSIZE;
   http->memlen = 0;
@@ -1216,6 +1217,8 @@ static ssize_t http2_handle_stream_close(struct connectdata *conn,
     }
   }

+  stream->close_handled = TRUE;
+
   DEBUGF(infof(data, "http2_recv returns 0, http2_handle_stream_close\n"));
   return 0;
 }
@@ -1520,6 +1523,14 @@ static ssize_t http2_send(struct connectdata *conn, int sockindex,
   DEBUGF(infof(conn->data, "http2_send len=%zu\n", len));

   if(stream->stream_id != -1) {
+    if(stream->close_handled) {
+      infof(conn->data, "stream %d closed\n", stream->stream_id);
+      *err = CURLE_HTTP2;
+      return -1;
+    }
+    else if(stream->closed) {
+      return http2_handle_stream_close(conn, conn->data, stream, err);
+    }
     /* If stream_id != -1, we have dispatched request HEADERS, and now
        are going to send or sending request body in DATA frame */
     stream->upload_mem = mem;

Here, we use error code CURLE_HTTP2. Not sure it is the best code in this situation.

[bagder]

Thanks, I'll give this a closer look and test out a little before I merge. Regarding the error code, I think it is time for us to go through and check if we shouldn't add a few more error codes for h2 specific situations that would be helpful if we exposed better to users instead of shoveling everything under a generic "http2 error" code.

[jay]

There is a new error code we added CURLE_HTTP2_STREAM a while ago, but I know you are aware of this, does it not apply?

[bagder]

Oh right. forgot about that one. Yes it seems suitable for this case. Thanks, will fix!