From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 From: Benjamin Marzinski Date: Thu, 25 Apr 2024 19:35:16 -0400 Subject: [PATCH] libmultipath: pad dev_loss_tmo to avoid race with no_path_retry Currently multipath makes sure that dev_loss_tmo is at least as long as the configured no path queueing time. The goal is to make sure that path devices aren't removed while the multipath device is still queueing in hopes that they will become usable again. This is racy. Multipathd may take longer to check the paths than configured. If strict_timing isn't set, it will definitely take longer. To account for this, pad the minimum dev_loss_tmo value by five seconds (one default checker interval) plus one second per minute of no path queueing time. Signed-off-by: Benjamin Marzinski Reviewed-by: Martin Wilck Reviewed-by: Benjamin Marzinski --- libmultipath/discovery.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/libmultipath/discovery.c b/libmultipath/discovery.c index ae7eb7e6..b24594cd 100644 --- a/libmultipath/discovery.c +++ b/libmultipath/discovery.c @@ -884,6 +884,11 @@ sysfs_set_scsi_tmo (struct config *conf, struct multipath *mpp) uint64_t no_path_retry_tmo = (uint64_t)mpp->no_path_retry * conf->checkint; + /* pad no_path_retry_tmo by one standard check interval + * plus one second per minute to account for timing + * issues with the rechecks */ + no_path_retry_tmo += no_path_retry_tmo / 60 + DEFAULT_CHECKINT; + if (no_path_retry_tmo > MAX_DEV_LOSS_TMO) min_dev_loss = MAX_DEV_LOSS_TMO; else