lib: Properly handle POLLERR from poll()

There are situations where POLLERR will be returned.  But
since we were not handling it.  Thread processing effectively
is turned into an infinite loop, which is bad.

Modify the code so that if we receive a POLLERR we turn it
into a read event to be handled as an error from the handler
function.

This was discovered in pim:

Thread statistics for pimd:
Showing poll FD's for main
--------------------------
Count: 14/1024
	     0 fd:     9 events: 1 revents: 0		mroute_read
	     1 fd:    12 events: 1 revents: 0		vty_accept
	     2 fd:    13 events: 1 revents: 0		vtysh_accept
	     3 fd:    11 events: 1 revents: 0		zclient_read
	     4 fd:    15 events: 1 revents: 0		mroute_read
	     5 fd:    16 events: 1 revents: 0		mroute_read
	     6 fd:    17 events: 1 revents: 0		pim_sock_read
	     7 fd:    19 events: 1 revents: 0		pim_sock_read
	     8 fd:    21 events: 1 revents: 0		pim_igmp_read
	     9 fd:    22 events: 1 revents: 0		pim_sock_read
	    10 fd:    23 events: 1 revents: 0		pim_sock_read
	    11 fd:    20 events: 1 revents: 0		vtysh_read
	    12 fd:    18 events: 1 revents: 0		pim_sock_read
	    13 fd:    24 events: 0 revents: 0

strace was showing this line over and over and over:

poll([{fd=9, events=POLLIN}, {fd=12, events=POLLIN}, {fd=13, events=POLLIN}, {fd=11, events=POLLIN}, {fd=15, events=POLLIN}, {fd=16, events=POLLIN}, {fd=17, events=POLLIN}, {fd=19, events=POLLIN}, {fd=21, events=POLLIN}, {fd=22, events=POLLIN}, {fd=23, events=POLLIN}, {fd=20, events=POLLIN}, {fd=18, events=POLLIN}, {fd=6, events=POLLIN}], 14, 20) = 1 ([{fd=21, revents=POLLERR}])

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

1 files changed, 5 insertions, 1 deletions

diff --git a/lib/thread.c b/lib/thread.c
index 48183cc9a..19e482728 100644
--- a/lib/thread.c
+++ b/lib/thread.c
@@ -1314,8 +1314,12 @@ static void thread_process_io(struct thread_master *m, unsigned int num)
 		 * copy, so there's no need to update it. Similarily,
 		 * barring deletion, the fd should still be a valid index
 		 * into the master's pfds.
+		 *
+		 * We are including POLLERR here to do a READ event
+		 * this is because the read should fail and the
+		 * read function should handle it appropriately
 		 */
-		if (pfds[i].revents & (POLLIN | POLLHUP)) {
+		if (pfds[i].revents & (POLLIN | POLLHUP | POLLERR)) {
 			thread_process_io_helper(m, m->read[pfds[i].fd], POLLIN,
 						 pfds[i].revents, i);
 		}