dm: do not initialise full request queue when bio based
authorMike Snitzer <snitzer@redhat.com>
Thu, 12 Aug 2010 03:14:02 +0000 (04:14 +0100)
committerAlasdair G Kergon <agk@redhat.com>
Thu, 12 Aug 2010 03:14:02 +0000 (04:14 +0100)
commit4a0b4ddf261fc89c050fe0a10ec57a61251d7ac0
tree23854bb6d42e7cedec9100fcdd63c307ee5d1332
parenta5664dad7e1a278d2915c2bf79cf42250e12d7db
dm: do not initialise full request queue when bio based

Change bio-based mapped devices no longer to have a fully initialized
request_queue (request_fn, elevator, etc).  This means bio-based DM
devices no longer register elevator sysfs attributes ('iosched/' tree
or 'scheduler' other than "none").

In contrast, a request-based DM device will continue to have a full
request_queue and will register elevator sysfs attributes.  Therefore
a user can determine a DM device's type by checking if elevator sysfs
attributes exist.

First allocate a minimalist request_queue structure for a DM device
(needed for both bio and request-based DM).

Initialization of a full request_queue is deferred until it is known
that the DM device is request-based, at the end of the table load
sequence.

Factor DM device's request_queue initialization:
- common to both request-based and bio-based into dm_init_md_queue().
- specific to request-based into dm_init_request_based_queue().

The md->type_lock mutex is used to protect md->queue, in addition to
md->type, during table_load().

A DM device's first table_load will establish the immutable md->type.
But md->queue initialization, based on md->type, may fail at that time
(because blk_init_allocated_queue cannot allocate memory).  Therefore
any subsequent table_load must (re)try dm_setup_md_queue independently of
establishing md->type.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
drivers/md/dm-ioctl.c
drivers/md/dm.c
drivers/md/dm.h