threads / discuss / 31865

Fix potential hang in https handshake (v2).

Subject: Fix potential hang in https handshake (v2).

## tl;dr

3 messages between Oct 18, 2012 and Oct 19, 2012.

replies: 2people: 2as markdown or json

szager@google.com· Oct 18, 2012, 23:27 UTC · lore

>From aa77ab3dd5b98a5786ac158528f45355fc0ddbc3 Mon Sep 17 00:00:00 2001

From: Stefan Zager <szager@google.com>
Date: Thu, 18 Oct 2012 16:23:53 -0700
Subject: [PATCH] Fix potential hang in https handshake.

It will sometimes happen that curl_multi_fdset() doesn't return any file descriptors. In that case, it's recommended that the application sleep for a short time before running curl_multi_perform() again.

http://curl.haxx.se/libcurl/c/curl_multi_fdset.html

Signed-off-by: Stefan Zager <szager@google.com>
---
 http.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/http.c b/http.c
index df9bb71..e8aba7f 100644
--- a/http.c
+++ b/http.c
@@ -630,6 +630,10 @@ void run_active_slot(struct active_request_slot *slot)
 			FD_ZERO(&writefds);
 			FD_ZERO(&excfds);
 			curl_multi_fdset(curlm, &readfds, &writefds, &excfds, &max_fd);
+			if (max_fd < 0) {
+				select_timeout.tv_sec  = 0;
+				select_timeout.tv_usec = 50000;
+			}
 
 			select(max_fd+1, &readfds, &writefds, &excfds, &select_timeout);
 		}

-- 
1.7.7.3

Junio C Hamano· Oct 19, 2012, 00:33 UTC · re: szager@google.com · lore

Re: Fix potential hang in https handshake (v2).

szager@google.com writes:

> From aa77ab3dd5b98a5786ac158528f45355fc0ddbc3 Mon Sep 17 00:00:00 2001
> From: Stefan Zager <szager@google.com>
> Date: Thu, 18 Oct 2012 16:23:53 -0700
> Subject: [PATCH] Fix potential hang in https handshake.

Please don't do the above (as usual ;-)

Show 9 quoted lines

> It will sometimes happen that curl_multi_fdset() doesn't
> return any file descriptors.  In that case, it's recommended
> that the application sleep for a short time before running
> curl_multi_perform() again.
>
> http://curl.haxx.se/libcurl/c/curl_multi_fdset.html
>
> Signed-off-by: Stefan Zager <szager@google.com>
> ---

The above sounds like "the code is doing something against a recommended practice", but is there a user-visible symptom due to this?

We end up calling select() without any bit set in fds, so it would become micro-sleep of select_timeout in such a case, but as far as I can see, the existing code either

 * does not select() and keeps polling step_active_slots() without
   delay, when curl_timeout gives a 0 return value; or

 * sets 50ms timeout or whatever negative value derived from
   curl_timeout when the returned value is not 0 nor -1.

Is the symptom that select(), when given a negative timeout and no fd to wake it, sleeps forever (or "loooong time, taking that negative value as if it is a large unsigned long") or something?

Thanks.

Show 18 quoted lines

>  http.c |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)
>
> diff --git a/http.c b/http.c
> index df9bb71..e8aba7f 100644
> --- a/http.c
> +++ b/http.c
> @@ -630,6 +630,10 @@ void run_active_slot(struct active_request_slot *slot)
>  			FD_ZERO(&writefds);
>  			FD_ZERO(&excfds);
>  			curl_multi_fdset(curlm, &readfds, &writefds, &excfds, &max_fd);
> +			if (max_fd < 0) {
> +				select_timeout.tv_sec  = 0;
> +				select_timeout.tv_usec = 50000;
> +			}
>  
>  			select(max_fd+1, &readfds, &writefds, &excfds, &select_timeout);
>  		}

Junio C Hamano· Oct 19, 2012, 00:37 UTC · re: Junio C Hamano · lore

Re: Fix potential hang in https handshake (v2).

Junio C Hamano <gitster@pobox.com> writes:

Show 13 quoted lines

> We end up calling select() without any bit set in fds, so it would
> become micro-sleep of select_timeout in such a case, but as far as I
> can see, the existing code either
>
>  * does not select() and keeps polling step_active_slots() without
>    delay, when curl_timeout gives a 0 return value; or
>
>  * sets 50ms timeout or whatever negative value derived from
>    curl_timeout when the returned value is not 0 nor -1.
>
> Is the symptom that select(), when given a negative timeout and no
> fd to wake it, sleeps forever (or "loooong time, taking that negative
> value as if it is a large unsigned long") or something?

Addendum.

What I am trying to get at are (1) three line description I can put in the release notes for this fix when it is merged to the maintenance track, and (2) a feel of how often this happens and how bad the damage is.

The latter helps us judge if this "do the normal thing, but if in a rare case where we do not find any fds, patch it up to proceed" is a better approach over your original.

Thanks.