Re: [RFC] [PATCH] VM & VFS changes

On  2 Jun, Alexander Leidinger wrote:
> Don Lewis <truckman_at_FreeBSD.org> wrote:
> 
>>>>>> Wouldn't a loop like the following be enough?
>>>>>> while swap
>>>>>>     umount unbusy-FS
>>>>>>     swap-off swap
>>>>>>
>>>>>> This assumes that swap-off doesn't turns off the swap if it isn't 
>>>>>> able to put
>>>>>> everything back into other swap or physical RAM areas.
> 
>> I think this can be unwound in one pass if a list of the dependency
>> pairs is kept and then properly sorted before processing.  The types of
>> dependencies are:
>> 	md depends on file system (vnode backed md)
>> 	md depends on swap (swap backed file system)
>> 	file system depends on md (md backed file system)
>> 	swap depends on md (swap on an md)
>> 	file system depends on file system (mount relationship)
>> First undo any dependencies that ultimately depend on swap, unconfigure
>> the swap devices, and finally undo any dependencies that swap depended
>> on.
> 
> I still don't understand why my approach above doesn't solve this problem.
> 
> A FS is busy when something is still open. So if the FS is used as a
> container for swap, the FS is busy and it isn't supposed to be umounted. If
> a md is configured on the swap area you want to disable with swap-off, my
> above description allows the call to fail. Since the md/swap/FS part is
> cycle-free, we have an upper bound of sum(#md)+sum(#swap)+sum(#FS)
> iterations (actually it's less than that, but my point is: a linear number
> of iterations with an upper bound) of this loop. When the loop finishes, no
> swap is enabled anymore. -> Goal reached.
> 
> What am I overlooking?

Create a large, but nearly empty file system, /a
Mount a file system backed by a physical disk on /a/b
Create the file /a/b/c and configure it to be used as swap
Write a large amount of data to the file /a/d, which will overflow RAM
and be paged out to /a/b/c

It won't be possible to disable swapping to /a/b/c because there is not
sufficient RAM to page in the data stored there.  It won't be possible
to unmount /a/b because /a/b/c is busy.  It won't be possible to unmount
/a because it is busy because /a/b is mounted on it.

If the dependencies are tracked so that this configuration (swapping to
anything that is directly or indirectly dependent on a swap-backed file
system) can be forbidden, then either the algorithm that I suggested, or
your iterative algorithm should work. Your algorithm could even be
simplified by pulling the swap-off out of the loop.

unconfigure md's that are not busy
do {
	foreach filesystem in reverse(mountlist) {
		unmount filesystem
		if (success and backed by md)
			unconfigure md /* either swap or vnode backed */
	}
} while (progress)
swap-off
do {
	foreach filesystem in reverse(mountlist) {
		unmount filesystem
		if (success and backed by md)
			unconfigure md /* only vnode backed remain */
	}
} while (progress)

Removing the swap-off from the loop may avoid a lot of paging activity
because all the paged-out data from swap backed md's will be discarded
before swap is disabled.