Block hotplug is accidentally quadratic

[nwf]

The cache_load call in mount_action ignores the device name parameter:

fstools/block.c

Line 1256 in bfe882d

cache_load(1);

As such, each hotplug event (specifically, each /sbin/block hotplug invoked via /etc/hotplug.d/block/10-mount via /sbin/hotplug-call through the machinations of /etc/hotplug.json) probes all the block devices named here:

fstools/block.c

Lines 580 to 596 in bfe882d

	static void cache_load(int mtd)
	{
	if (mtd) {
	_cache_load("/dev/mtdblock*");
	_cache_load("/dev/ubiblock*");
	_cache_load("/dev/ubi[0-9]*");
	}
	_cache_load("/dev/loop*");
	_cache_load("/dev/mmcblk*");
	_cache_load("/dev/sd*");
	_cache_load("/dev/hd*");
	_cache_load("/dev/md*");
	_cache_load("/dev/nvme*");
	_cache_load("/dev/vd*");
	_cache_load("/dev/xvd*");
	_cache_load("/dev/dm-*");
	}

That's rather a lot, if not all, of the block devices on the system, making the total block hotplug sequence accidentally quadratic.

I hope this is the right place to raise this issue, and I've also documented a bit about how I found this over in this forum thread, but if it belongs somewhere else, please let me know.

[eric-j-ason]

In addition to your optimization, isn't it the case that one could break the loop just after cache_load, once a match has been found, as there should only ever be one match, right?

fstools/block.c

Lines 1258 to 1260 in bfe882d

	list_for_each_entry(pr, &devices, list)
	if (!strcmp(basename(pr->dev), device))
	path = pr->dev;

[nwf]

Apologies for dropping the ball. Yes, I suspect it is safe to break out of that loop as you describe. Should I amend the PR?