[dm-devel] [PATCH] dm table: do not allow queue limits that will exceed hardware limits

[Mike Snitzer]

DM recalculates queue limits based only on devices which currently exist
in the table.  This creates a problem in the event all devices are
temporarily removed such as all fibre channel paths being lost in
multipath.  DM will reset the limits to the maximum permissible, which
can then assemble requests which exceed the limits of the paths when the
paths are restored.  The request will fail the blk_rq_check_limits()
test when sent to a path with lower limits, and will be retried without
end by multipath.

This becomes a much bigger issue after commit fe86cdcef ("block: do not
artificially constrain max_sectors for stacking drivers").  Previously,
most storage had max_sector limits which exceeded the default value
used.  This meant most setups wouldn't trigger this issue as the default
values used when there were no paths were still less than the limits of
the underlying devices.  Now that the default stacking values are no
longer constrained, any hardware setup can potentially hit this issue.

So add a safety net that will establish safe default limits, via
blk_set_default_limits, in the event that a table temporarily doesn't
have any component devices.

Reported-by: David Jeffery <djeffery at redhat.com>
Signed-off-by: Mike Snitzer <snitzer at redhat.com>
---
 drivers/md/dm-table.c |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index f6979ad..9b931b4 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -1264,6 +1264,15 @@ combine_limits:
 			       (unsigned long long) ti->len);
 	}
 
+	/*
+	 * If a table doesn't have any component devices (e.g. multipath
+	 * loses all paths) don't allow the queue_limits to be left at
+	 * their maximum (as established by blk_set_stacking_limits() so
+	 * limits could be inherited from component devices).
+	 */
+	if (limits->max_sectors == UINT_MAX)
+		blk_set_default_limits(limits);
+
 	return validate_hardware_logical_block_alignment(table, limits);
 }
 
-- 
1.7.1