[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] questions about dm-thin and discard

On Mon, Jul 16 2012 at  1:14pm -0400,
Mikulas Patocka <mpatocka redhat com> wrote:

> Hi Joe
> I would like to ask you about this code path: In process_discard, there is 
> a branch with a comment "This path is hit if people are ignoring 
> limits->discard_granularity." It trims the discard request so that it 
> doesn't span a block boundary and submits it.
> The question is: what if the block is shared? In this case, we can't 
> submit discard to the block, because it would damage the other snapshot 
> that is sharing this block. Shouldn't there be shomething like this?
> if ((!lookup_result.shared) & pool->pf.discard_passdown) {
> 	remap_and_issue(tc, bio, lookup_result.block);
> } else {
> 	bio_endio(bio, 0) 
> } 
> ... or is it tested elsewhere and am I missing something?

in process_discard:

 m->pass_discard = (!lookup_result.shared) && pool->pf.disard_passdown;

then in process_prepared_discard:

 if (m->pass_discard)
        remap_and_issue(tc, m->bio, m->data_block);
        bio_endio(m->bio, 0);

> Another question is about setting "ti->discards_supported = 1" in 
> pool_ctr. ti->discards_supported means that the target supports discards 
> even if the underlying disk doesn't. Since the pool device is passing 
> anyth I/O unchanged to the underlying disk, ti->discards_supported 
> shouldn't be set. Or is there any other reason why is it set?

The thin device's bios will be remapped to the pool device.

process_prepared_discard's remap_and_issue() will send the bio to the
pool device via generic_make_request().

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]