Possible busy-looping in curl_multi_poll

[3dyd]

Take a look at this piece of code at the end of Curl_multi_wait:

curl/lib/multi.c

Lines 1150 to 1161 in 7a46aeb

	if(!extrawait || extra_fds || curlfds)
	/* if any socket was checked */
	;
	else {
	long sleep_ms = 0;
	/* Avoid busy-looping when there's nothing particular to wait for */
	if(!curl_multi_timeout(multi, &sleep_ms) && sleep_ms) {
	if(sleep_ms > timeout_ms)
	sleep_ms = timeout_ms;
	Curl_wait_ms((int)sleep_ms);
	}

If curl_multi_poll is given valid empty multi handle (no easy handles were added to it [yet]) without extra curl_waitfd structures then sleep_ms receives value -1 from curl_multi_timeout effectively avoiding waiting for given timeout_ms period. So function returns immediately.

Is this intended behavior, and such edge case still has to be accounted by the caller to avoid busy-looping?

[bagder]

I think it should wait for the timeout period in this case. Something like this?

diff --git a/lib/multi.c b/lib/multi.c
index 7e8e38dc9..db08ac636 100755
--- a/lib/multi.c
+++ b/lib/multi.c
@@ -1155,10 +1155,12 @@ static CURLMcode Curl_multi_wait(struct Curl_multi *multi,
 
     /* Avoid busy-looping when there's nothing particular to wait for */
     if(!curl_multi_timeout(multi, &sleep_ms) && sleep_ms) {
       if(sleep_ms > timeout_ms)
         sleep_ms = timeout_ms;
+      else if((sleep_ms < 0) && extrawait)
+        sleep_ms = timeout_ms;
       Curl_wait_ms((int)sleep_ms);
     }
   }
 
   return CURLM_OK;

[3dyd]

Not sure if notification is fired if one comments diff directly, so I am duplicating it here:

else if((sleep_ms < 0) && extrawait)

extrawait is always true in this block.

[3dyd]

Confirmed the change has fixed described issue.