2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
|
2009-01-05 20:05:12 -07:00
|
|
|
* Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
|
2005-04-16 15:20:36 -07:00
|
|
|
*
|
|
|
|
* This file is released under the GPL.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "dm.h"
|
|
|
|
#include "dm-bio-list.h"
|
2007-10-19 14:48:00 -07:00
|
|
|
#include "dm-uevent.h"
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/module.h>
|
2006-03-27 02:18:20 -07:00
|
|
|
#include <linux/mutex.h>
|
2005-04-16 15:20:36 -07:00
|
|
|
#include <linux/moduleparam.h>
|
|
|
|
#include <linux/blkpg.h>
|
|
|
|
#include <linux/bio.h>
|
|
|
|
#include <linux/buffer_head.h>
|
|
|
|
#include <linux/mempool.h>
|
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/idr.h>
|
2006-03-27 02:17:54 -07:00
|
|
|
#include <linux/hdreg.h>
|
2006-03-23 12:00:26 -07:00
|
|
|
#include <linux/blktrace_api.h>
|
2008-10-30 00:34:33 -07:00
|
|
|
#include <trace/block.h>
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-26 00:27:35 -07:00
|
|
|
#define DM_MSG_PREFIX "core"
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
static const char *_name = DM_NAME;
|
|
|
|
|
|
|
|
static unsigned int major = 0;
|
|
|
|
static unsigned int _major = 0;
|
|
|
|
|
2006-06-26 00:27:22 -07:00
|
|
|
static DEFINE_SPINLOCK(_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
2009-01-05 20:05:06 -07:00
|
|
|
* For bio-based dm.
|
2005-04-16 15:20:36 -07:00
|
|
|
* One of these is allocated per bio.
|
|
|
|
*/
|
|
|
|
struct dm_io {
|
|
|
|
struct mapped_device *md;
|
|
|
|
int error;
|
|
|
|
atomic_t io_count;
|
2008-07-21 04:00:28 -07:00
|
|
|
struct bio *bio;
|
2006-02-01 04:04:53 -07:00
|
|
|
unsigned long start_time;
|
2005-04-16 15:20:36 -07:00
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
2009-01-05 20:05:06 -07:00
|
|
|
* For bio-based dm.
|
2005-04-16 15:20:36 -07:00
|
|
|
* One of these is allocated per target within a bio. Hopefully
|
|
|
|
* this will be simplified out one day.
|
|
|
|
*/
|
2007-07-12 09:26:32 -07:00
|
|
|
struct dm_target_io {
|
2005-04-16 15:20:36 -07:00
|
|
|
struct dm_io *io;
|
|
|
|
struct dm_target *ti;
|
|
|
|
union map_info info;
|
|
|
|
};
|
|
|
|
|
2008-11-26 03:59:56 -07:00
|
|
|
DEFINE_TRACE(block_bio_complete);
|
|
|
|
|
2009-01-05 20:05:06 -07:00
|
|
|
/*
|
|
|
|
* For request-based dm.
|
|
|
|
* One of these is allocated per request.
|
|
|
|
*/
|
|
|
|
struct dm_rq_target_io {
|
|
|
|
struct mapped_device *md;
|
|
|
|
struct dm_target *ti;
|
|
|
|
struct request *orig, clone;
|
|
|
|
int error;
|
|
|
|
union map_info info;
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For request-based dm.
|
|
|
|
* One of these is allocated per bio.
|
|
|
|
*/
|
|
|
|
struct dm_rq_clone_bio_info {
|
|
|
|
struct bio *orig;
|
|
|
|
struct request *rq;
|
|
|
|
};
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
union map_info *dm_get_mapinfo(struct bio *bio)
|
|
|
|
{
|
2006-06-26 00:27:33 -07:00
|
|
|
if (bio && bio->bi_private)
|
2007-07-12 09:26:32 -07:00
|
|
|
return &((struct dm_target_io *)bio->bi_private)->info;
|
2006-06-26 00:27:33 -07:00
|
|
|
return NULL;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
2006-06-26 00:27:21 -07:00
|
|
|
#define MINOR_ALLOCED ((void *)-1)
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Bits for the md->flags field.
|
|
|
|
*/
|
|
|
|
#define DMF_BLOCK_IO 0
|
|
|
|
#define DMF_SUSPENDED 1
|
2006-01-06 01:20:06 -07:00
|
|
|
#define DMF_FROZEN 2
|
2006-06-26 00:27:23 -07:00
|
|
|
#define DMF_FREEING 3
|
2006-06-26 00:27:34 -07:00
|
|
|
#define DMF_DELETING 4
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
#define DMF_NOFLUSH_SUSPENDING 5
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
/*
|
|
|
|
* Work processed by per-device workqueue.
|
|
|
|
*/
|
|
|
|
struct dm_wq_req {
|
|
|
|
enum {
|
|
|
|
DM_WQ_FLUSH_DEFERRED,
|
|
|
|
} type;
|
|
|
|
struct work_struct work;
|
|
|
|
struct mapped_device *md;
|
|
|
|
void *context;
|
|
|
|
};
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
struct mapped_device {
|
2005-07-28 21:16:00 -07:00
|
|
|
struct rw_semaphore io_lock;
|
2008-02-07 19:10:08 -07:00
|
|
|
struct mutex suspend_lock;
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
spinlock_t pushback_lock;
|
2005-04-16 15:20:36 -07:00
|
|
|
rwlock_t map_lock;
|
|
|
|
atomic_t holders;
|
2006-06-26 00:27:34 -07:00
|
|
|
atomic_t open_count;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
unsigned long flags;
|
|
|
|
|
2007-07-24 00:28:11 -07:00
|
|
|
struct request_queue *queue;
|
2005-04-16 15:20:36 -07:00
|
|
|
struct gendisk *disk;
|
2006-03-27 02:17:52 -07:00
|
|
|
char name[16];
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
void *interface_ptr;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A list of ios that arrived while we were suspended.
|
|
|
|
*/
|
|
|
|
atomic_t pending;
|
|
|
|
wait_queue_head_t wait;
|
2006-12-08 03:41:02 -07:00
|
|
|
struct bio_list deferred;
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
struct bio_list pushback;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
/*
|
|
|
|
* Processing queue (flush/barriers)
|
|
|
|
*/
|
|
|
|
struct workqueue_struct *wq;
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* The current mapping.
|
|
|
|
*/
|
|
|
|
struct dm_table *map;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* io objects are allocated from here.
|
|
|
|
*/
|
|
|
|
mempool_t *io_pool;
|
|
|
|
mempool_t *tio_pool;
|
|
|
|
|
2006-10-03 01:15:41 -07:00
|
|
|
struct bio_set *bs;
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Event handling.
|
|
|
|
*/
|
|
|
|
atomic_t event_nr;
|
|
|
|
wait_queue_head_t eventq;
|
2007-10-19 14:48:01 -07:00
|
|
|
atomic_t uevent_seq;
|
|
|
|
struct list_head uevent_list;
|
|
|
|
spinlock_t uevent_lock; /* Protect access to uevent_list */
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* freeze/thaw support require holding onto a super block
|
|
|
|
*/
|
|
|
|
struct super_block *frozen_sb;
|
2006-01-06 01:20:05 -07:00
|
|
|
struct block_device *suspended_bdev;
|
2006-03-27 02:17:54 -07:00
|
|
|
|
|
|
|
/* forced geometry settings */
|
|
|
|
struct hd_geometry geometry;
|
2009-01-05 20:05:12 -07:00
|
|
|
|
|
|
|
/* sysfs handle */
|
|
|
|
struct kobject kobj;
|
2005-04-16 15:20:36 -07:00
|
|
|
};
|
|
|
|
|
|
|
|
#define MIN_IOS 256
|
2006-12-06 21:33:20 -07:00
|
|
|
static struct kmem_cache *_io_cache;
|
|
|
|
static struct kmem_cache *_tio_cache;
|
2009-01-05 20:05:06 -07:00
|
|
|
static struct kmem_cache *_rq_tio_cache;
|
|
|
|
static struct kmem_cache *_rq_bio_info_cache;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
static int __init local_init(void)
|
|
|
|
{
|
2008-10-21 09:45:08 -07:00
|
|
|
int r = -ENOMEM;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/* allocate a slab for the dm_ios */
|
2007-07-12 09:26:32 -07:00
|
|
|
_io_cache = KMEM_CACHE(dm_io, 0);
|
2005-04-16 15:20:36 -07:00
|
|
|
if (!_io_cache)
|
2008-10-21 09:45:08 -07:00
|
|
|
return r;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/* allocate a slab for the target ios */
|
2007-07-12 09:26:32 -07:00
|
|
|
_tio_cache = KMEM_CACHE(dm_target_io, 0);
|
2008-10-21 09:45:08 -07:00
|
|
|
if (!_tio_cache)
|
|
|
|
goto out_free_io_cache;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-01-05 20:05:06 -07:00
|
|
|
_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
|
|
|
|
if (!_rq_tio_cache)
|
|
|
|
goto out_free_tio_cache;
|
|
|
|
|
|
|
|
_rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
|
|
|
|
if (!_rq_bio_info_cache)
|
|
|
|
goto out_free_rq_tio_cache;
|
|
|
|
|
2007-10-19 14:48:00 -07:00
|
|
|
r = dm_uevent_init();
|
2008-10-21 09:45:08 -07:00
|
|
|
if (r)
|
2009-01-05 20:05:06 -07:00
|
|
|
goto out_free_rq_bio_info_cache;
|
2007-10-19 14:48:00 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
_major = major;
|
|
|
|
r = register_blkdev(_major, _name);
|
2008-10-21 09:45:08 -07:00
|
|
|
if (r < 0)
|
|
|
|
goto out_uevent_exit;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
if (!_major)
|
|
|
|
_major = r;
|
|
|
|
|
|
|
|
return 0;
|
2008-10-21 09:45:08 -07:00
|
|
|
|
|
|
|
out_uevent_exit:
|
|
|
|
dm_uevent_exit();
|
2009-01-05 20:05:06 -07:00
|
|
|
out_free_rq_bio_info_cache:
|
|
|
|
kmem_cache_destroy(_rq_bio_info_cache);
|
|
|
|
out_free_rq_tio_cache:
|
|
|
|
kmem_cache_destroy(_rq_tio_cache);
|
2008-10-21 09:45:08 -07:00
|
|
|
out_free_tio_cache:
|
|
|
|
kmem_cache_destroy(_tio_cache);
|
|
|
|
out_free_io_cache:
|
|
|
|
kmem_cache_destroy(_io_cache);
|
|
|
|
|
|
|
|
return r;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static void local_exit(void)
|
|
|
|
{
|
2009-01-05 20:05:06 -07:00
|
|
|
kmem_cache_destroy(_rq_bio_info_cache);
|
|
|
|
kmem_cache_destroy(_rq_tio_cache);
|
2005-04-16 15:20:36 -07:00
|
|
|
kmem_cache_destroy(_tio_cache);
|
|
|
|
kmem_cache_destroy(_io_cache);
|
2007-07-17 04:03:46 -07:00
|
|
|
unregister_blkdev(_major, _name);
|
2007-10-19 14:48:00 -07:00
|
|
|
dm_uevent_exit();
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
_major = 0;
|
|
|
|
|
|
|
|
DMINFO("cleaned up");
|
|
|
|
}
|
|
|
|
|
2008-02-07 19:09:51 -07:00
|
|
|
static int (*_inits[])(void) __initdata = {
|
2005-04-16 15:20:36 -07:00
|
|
|
local_init,
|
|
|
|
dm_target_init,
|
|
|
|
dm_linear_init,
|
|
|
|
dm_stripe_init,
|
2008-04-24 13:43:49 -07:00
|
|
|
dm_kcopyd_init,
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_interface_init,
|
|
|
|
};
|
|
|
|
|
2008-02-07 19:09:51 -07:00
|
|
|
static void (*_exits[])(void) = {
|
2005-04-16 15:20:36 -07:00
|
|
|
local_exit,
|
|
|
|
dm_target_exit,
|
|
|
|
dm_linear_exit,
|
|
|
|
dm_stripe_exit,
|
2008-04-24 13:43:49 -07:00
|
|
|
dm_kcopyd_exit,
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_interface_exit,
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __init dm_init(void)
|
|
|
|
{
|
|
|
|
const int count = ARRAY_SIZE(_inits);
|
|
|
|
|
|
|
|
int r, i;
|
|
|
|
|
|
|
|
for (i = 0; i < count; i++) {
|
|
|
|
r = _inits[i]();
|
|
|
|
if (r)
|
|
|
|
goto bad;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
bad:
|
|
|
|
while (i--)
|
|
|
|
_exits[i]();
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __exit dm_exit(void)
|
|
|
|
{
|
|
|
|
int i = ARRAY_SIZE(_exits);
|
|
|
|
|
|
|
|
while (i--)
|
|
|
|
_exits[i]();
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Block device functions
|
|
|
|
*/
|
2008-03-02 08:29:31 -07:00
|
|
|
static int dm_blk_open(struct block_device *bdev, fmode_t mode)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct mapped_device *md;
|
|
|
|
|
2006-06-26 00:27:23 -07:00
|
|
|
spin_lock(&_minor_lock);
|
|
|
|
|
2008-03-02 08:29:31 -07:00
|
|
|
md = bdev->bd_disk->private_data;
|
2006-06-26 00:27:23 -07:00
|
|
|
if (!md)
|
|
|
|
goto out;
|
|
|
|
|
2006-06-26 00:27:34 -07:00
|
|
|
if (test_bit(DMF_FREEING, &md->flags) ||
|
|
|
|
test_bit(DMF_DELETING, &md->flags)) {
|
2006-06-26 00:27:23 -07:00
|
|
|
md = NULL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_get(md);
|
2006-06-26 00:27:34 -07:00
|
|
|
atomic_inc(&md->open_count);
|
2006-06-26 00:27:23 -07:00
|
|
|
|
|
|
|
out:
|
|
|
|
spin_unlock(&_minor_lock);
|
|
|
|
|
|
|
|
return md ? 0 : -ENXIO;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
2008-03-02 08:29:31 -07:00
|
|
|
static int dm_blk_close(struct gendisk *disk, fmode_t mode)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2008-03-02 08:29:31 -07:00
|
|
|
struct mapped_device *md = disk->private_data;
|
2006-06-26 00:27:34 -07:00
|
|
|
atomic_dec(&md->open_count);
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_put(md);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-06-26 00:27:34 -07:00
|
|
|
int dm_open_count(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return atomic_read(&md->open_count);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Guarantees nothing is using the device before it's deleted.
|
|
|
|
*/
|
|
|
|
int dm_lock_for_deletion(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
int r = 0;
|
|
|
|
|
|
|
|
spin_lock(&_minor_lock);
|
|
|
|
|
|
|
|
if (dm_open_count(md))
|
|
|
|
r = -EBUSY;
|
|
|
|
else
|
|
|
|
set_bit(DMF_DELETING, &md->flags);
|
|
|
|
|
|
|
|
spin_unlock(&_minor_lock);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2006-03-27 02:17:54 -07:00
|
|
|
static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
|
|
|
|
{
|
|
|
|
struct mapped_device *md = bdev->bd_disk->private_data;
|
|
|
|
|
|
|
|
return dm_get_geometry(md, geo);
|
|
|
|
}
|
|
|
|
|
2008-03-02 08:29:31 -07:00
|
|
|
static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
|
2006-10-03 01:15:15 -07:00
|
|
|
unsigned int cmd, unsigned long arg)
|
|
|
|
{
|
2008-03-02 08:29:31 -07:00
|
|
|
struct mapped_device *md = bdev->bd_disk->private_data;
|
|
|
|
struct dm_table *map = dm_get_table(md);
|
2006-10-03 01:15:15 -07:00
|
|
|
struct dm_target *tgt;
|
|
|
|
int r = -ENOTTY;
|
|
|
|
|
|
|
|
if (!map || !dm_table_get_size(map))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* We only support devices that have a single target */
|
|
|
|
if (dm_table_get_num_targets(map) != 1)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
tgt = dm_table_get_target(map, 0);
|
|
|
|
|
|
|
|
if (dm_suspended(md)) {
|
|
|
|
r = -EAGAIN;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (tgt->type->ioctl)
|
2007-08-28 19:15:59 -07:00
|
|
|
r = tgt->type->ioctl(tgt, cmd, arg);
|
2006-10-03 01:15:15 -07:00
|
|
|
|
|
|
|
out:
|
|
|
|
dm_table_put(map);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2007-07-12 09:26:32 -07:00
|
|
|
static struct dm_io *alloc_io(struct mapped_device *md)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
return mempool_alloc(md->io_pool, GFP_NOIO);
|
|
|
|
}
|
|
|
|
|
2007-07-12 09:26:32 -07:00
|
|
|
static void free_io(struct mapped_device *md, struct dm_io *io)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
mempool_free(io, md->io_pool);
|
|
|
|
}
|
|
|
|
|
2007-07-12 09:26:32 -07:00
|
|
|
static struct dm_target_io *alloc_tio(struct mapped_device *md)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
return mempool_alloc(md->tio_pool, GFP_NOIO);
|
|
|
|
}
|
|
|
|
|
2007-07-12 09:26:32 -07:00
|
|
|
static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
mempool_free(tio, md->tio_pool);
|
|
|
|
}
|
|
|
|
|
2006-02-01 04:04:53 -07:00
|
|
|
static void start_io_acct(struct dm_io *io)
|
|
|
|
{
|
|
|
|
struct mapped_device *md = io->md;
|
2008-08-25 03:47:21 -07:00
|
|
|
int cpu;
|
2006-02-01 04:04:53 -07:00
|
|
|
|
|
|
|
io->start_time = jiffies;
|
|
|
|
|
2008-08-25 03:56:14 -07:00
|
|
|
cpu = part_stat_lock();
|
|
|
|
part_round_stats(cpu, &dm_disk(md)->part0);
|
|
|
|
part_stat_unlock();
|
|
|
|
dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
|
2006-02-01 04:04:53 -07:00
|
|
|
}
|
|
|
|
|
2008-11-13 16:39:10 -07:00
|
|
|
static void end_io_acct(struct dm_io *io)
|
2006-02-01 04:04:53 -07:00
|
|
|
{
|
|
|
|
struct mapped_device *md = io->md;
|
|
|
|
struct bio *bio = io->bio;
|
|
|
|
unsigned long duration = jiffies - io->start_time;
|
2008-08-25 03:47:21 -07:00
|
|
|
int pending, cpu;
|
2006-02-01 04:04:53 -07:00
|
|
|
int rw = bio_data_dir(bio);
|
|
|
|
|
2008-08-25 03:56:14 -07:00
|
|
|
cpu = part_stat_lock();
|
|
|
|
part_round_stats(cpu, &dm_disk(md)->part0);
|
|
|
|
part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
|
|
|
|
part_stat_unlock();
|
2006-02-01 04:04:53 -07:00
|
|
|
|
2008-08-25 03:56:14 -07:00
|
|
|
dm_disk(md)->part0.in_flight = pending =
|
|
|
|
atomic_dec_return(&md->pending);
|
2006-02-01 04:04:53 -07:00
|
|
|
|
2008-11-13 16:39:10 -07:00
|
|
|
/* nudge anyone waiting on suspend queue */
|
|
|
|
if (!pending)
|
|
|
|
wake_up(&md->wait);
|
2006-02-01 04:04:53 -07:00
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Add the bio to the list of deferred io.
|
|
|
|
*/
|
|
|
|
static int queue_io(struct mapped_device *md, struct bio *bio)
|
|
|
|
{
|
2005-07-28 21:16:00 -07:00
|
|
|
down_write(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
|
2005-07-28 21:16:00 -07:00
|
|
|
up_write(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
bio_list_add(&md->deferred, bio);
|
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
up_write(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
return 0; /* deferred successfully */
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Everyone (including functions in this file), should use this
|
|
|
|
* function to access the md->map field, and make sure they call
|
|
|
|
* dm_table_put() when finished.
|
|
|
|
*/
|
|
|
|
struct dm_table *dm_get_table(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
struct dm_table *t;
|
|
|
|
|
|
|
|
read_lock(&md->map_lock);
|
|
|
|
t = md->map;
|
|
|
|
if (t)
|
|
|
|
dm_table_get(t);
|
|
|
|
read_unlock(&md->map_lock);
|
|
|
|
|
|
|
|
return t;
|
|
|
|
}
|
|
|
|
|
2006-03-27 02:17:54 -07:00
|
|
|
/*
|
|
|
|
* Get the geometry associated with a dm device
|
|
|
|
*/
|
|
|
|
int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
|
|
|
|
{
|
|
|
|
*geo = md->geometry;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Set the geometry of a device.
|
|
|
|
*/
|
|
|
|
int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
|
|
|
|
{
|
|
|
|
sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
|
|
|
|
|
|
|
|
if (geo->start > sz) {
|
|
|
|
DMWARN("Start sector is beyond the geometry limits.");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
md->geometry = *geo;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*-----------------------------------------------------------------
|
|
|
|
* CRUD START:
|
|
|
|
* A more elegant soln is in the works that uses the queue
|
|
|
|
* merge fn, unfortunately there are a couple of changes to
|
|
|
|
* the block layer that I want to make for this. So in the
|
|
|
|
* interests of getting something for people to use I give
|
|
|
|
* you this clearly demarcated crap.
|
|
|
|
*---------------------------------------------------------------*/
|
|
|
|
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
static int __noflush_suspending(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Decrements the number of outstanding ios that a bio has been
|
|
|
|
* cloned into, completing the original io if necc.
|
|
|
|
*/
|
2006-01-14 14:20:43 -07:00
|
|
|
static void dec_pending(struct dm_io *io, int error)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
unsigned long flags;
|
2009-03-16 10:44:36 -07:00
|
|
|
int io_error;
|
|
|
|
struct bio *bio;
|
|
|
|
struct mapped_device *md = io->md;
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
|
|
|
|
/* Push-back supersedes any I/O errors */
|
2009-03-16 10:44:36 -07:00
|
|
|
if (error && !(io->error > 0 && __noflush_suspending(md)))
|
2005-04-16 15:20:36 -07:00
|
|
|
io->error = error;
|
|
|
|
|
|
|
|
if (atomic_dec_and_test(&io->io_count)) {
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
if (io->error == DM_ENDIO_REQUEUE) {
|
|
|
|
/*
|
|
|
|
* Target requested pushing back the I/O.
|
|
|
|
* This must be handled before the sleeper on
|
|
|
|
* suspend queue merges the pushback list.
|
|
|
|
*/
|
2009-03-16 10:44:36 -07:00
|
|
|
spin_lock_irqsave(&md->pushback_lock, flags);
|
|
|
|
if (__noflush_suspending(md))
|
|
|
|
bio_list_add(&md->pushback, io->bio);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
else
|
|
|
|
/* noflush suspend was interrupted. */
|
|
|
|
io->error = -EIO;
|
2009-03-16 10:44:36 -07:00
|
|
|
spin_unlock_irqrestore(&md->pushback_lock, flags);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
}
|
|
|
|
|
2008-11-13 16:39:10 -07:00
|
|
|
end_io_acct(io);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-03-16 10:44:36 -07:00
|
|
|
io_error = io->error;
|
|
|
|
bio = io->bio;
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
|
2009-03-16 10:44:36 -07:00
|
|
|
free_io(md, io);
|
|
|
|
|
|
|
|
if (io_error != DM_ENDIO_REQUEUE) {
|
|
|
|
trace_block_bio_complete(md->queue, bio);
|
2006-03-23 12:00:26 -07:00
|
|
|
|
2009-03-16 10:44:36 -07:00
|
|
|
bio_endio(bio, io_error);
|
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-09-27 03:47:43 -07:00
|
|
|
static void clone_endio(struct bio *bio, int error)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
int r = 0;
|
2007-07-12 09:26:32 -07:00
|
|
|
struct dm_target_io *tio = bio->bi_private;
|
2009-03-16 10:44:36 -07:00
|
|
|
struct dm_io *io = tio->io;
|
2006-10-03 01:15:41 -07:00
|
|
|
struct mapped_device *md = tio->io->md;
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_endio_fn endio = tio->ti->type->end_io;
|
|
|
|
|
|
|
|
if (!bio_flagged(bio, BIO_UPTODATE) && !error)
|
|
|
|
error = -EIO;
|
|
|
|
|
|
|
|
if (endio) {
|
|
|
|
r = endio(tio->ti, bio, error, &tio->info);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
if (r < 0 || r == DM_ENDIO_REQUEUE)
|
|
|
|
/*
|
|
|
|
* error and requeue request are handled
|
|
|
|
* in dec_pending().
|
|
|
|
*/
|
2005-04-16 15:20:36 -07:00
|
|
|
error = r;
|
2006-12-08 03:41:05 -07:00
|
|
|
else if (r == DM_ENDIO_INCOMPLETE)
|
|
|
|
/* The target will handle the io */
|
2007-09-27 03:47:43 -07:00
|
|
|
return;
|
2006-12-08 03:41:05 -07:00
|
|
|
else if (r) {
|
|
|
|
DMWARN("unimplemented target endio return value: %d", r);
|
|
|
|
BUG();
|
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
2006-10-03 01:15:41 -07:00
|
|
|
/*
|
|
|
|
* Store md for cleanup instead of tio which is about to get freed.
|
|
|
|
*/
|
|
|
|
bio->bi_private = md->bs;
|
|
|
|
|
|
|
|
free_tio(md, tio);
|
2009-03-16 10:44:36 -07:00
|
|
|
bio_put(bio);
|
|
|
|
dec_pending(io, error);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static sector_t max_io_len(struct mapped_device *md,
|
|
|
|
sector_t sector, struct dm_target *ti)
|
|
|
|
{
|
|
|
|
sector_t offset = sector - ti->begin;
|
|
|
|
sector_t len = ti->len - offset;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Does the target need to split even further ?
|
|
|
|
*/
|
|
|
|
if (ti->split_io) {
|
|
|
|
sector_t boundary;
|
|
|
|
boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
|
|
|
|
- offset;
|
|
|
|
if (len > boundary)
|
|
|
|
len = boundary;
|
|
|
|
}
|
|
|
|
|
|
|
|
return len;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __map_bio(struct dm_target *ti, struct bio *clone,
|
2007-07-12 09:26:32 -07:00
|
|
|
struct dm_target_io *tio)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
int r;
|
2006-03-23 12:00:26 -07:00
|
|
|
sector_t sector;
|
2006-10-03 01:15:41 -07:00
|
|
|
struct mapped_device *md;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Sanity checks.
|
|
|
|
*/
|
|
|
|
BUG_ON(!clone->bi_size);
|
|
|
|
|
|
|
|
clone->bi_end_io = clone_endio;
|
|
|
|
clone->bi_private = tio;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Map the clone. If r == 0 we don't need to do
|
|
|
|
* anything, the target has assumed ownership of
|
|
|
|
* this io.
|
|
|
|
*/
|
|
|
|
atomic_inc(&tio->io->io_count);
|
2006-03-23 12:00:26 -07:00
|
|
|
sector = clone->bi_sector;
|
2005-04-16 15:20:36 -07:00
|
|
|
r = ti->type->map(ti, clone, &tio->info);
|
2006-12-08 03:41:05 -07:00
|
|
|
if (r == DM_MAPIO_REMAPPED) {
|
2005-04-16 15:20:36 -07:00
|
|
|
/* the bio has been remapped so dispatch it */
|
2006-03-23 12:00:26 -07:00
|
|
|
|
2008-10-30 00:34:33 -07:00
|
|
|
trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
|
2007-08-07 06:30:23 -07:00
|
|
|
tio->io->bio->bi_bdev->bd_dev,
|
|
|
|
clone->bi_sector, sector);
|
2006-03-23 12:00:26 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
generic_make_request(clone);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
|
|
|
|
/* error the io and bail out, or requeue it if needed */
|
2006-10-03 01:15:41 -07:00
|
|
|
md = tio->io->md;
|
|
|
|
dec_pending(tio->io, r);
|
|
|
|
/*
|
|
|
|
* Store bio_set for cleanup.
|
|
|
|
*/
|
|
|
|
clone->bi_private = md->bs;
|
2005-04-16 15:20:36 -07:00
|
|
|
bio_put(clone);
|
2006-10-03 01:15:41 -07:00
|
|
|
free_tio(md, tio);
|
2006-12-08 03:41:05 -07:00
|
|
|
} else if (r) {
|
|
|
|
DMWARN("unimplemented target map return value: %d", r);
|
|
|
|
BUG();
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
struct clone_info {
|
|
|
|
struct mapped_device *md;
|
|
|
|
struct dm_table *map;
|
|
|
|
struct bio *bio;
|
|
|
|
struct dm_io *io;
|
|
|
|
sector_t sector;
|
|
|
|
sector_t sector_count;
|
|
|
|
unsigned short idx;
|
|
|
|
};
|
|
|
|
|
2005-09-06 15:16:42 -07:00
|
|
|
static void dm_bio_destructor(struct bio *bio)
|
|
|
|
{
|
2006-10-03 01:15:41 -07:00
|
|
|
struct bio_set *bs = bio->bi_private;
|
|
|
|
|
|
|
|
bio_free(bio, bs);
|
2005-09-06 15:16:42 -07:00
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Creates a little bio that is just does part of a bvec.
|
|
|
|
*/
|
|
|
|
static struct bio *split_bvec(struct bio *bio, sector_t sector,
|
|
|
|
unsigned short idx, unsigned int offset,
|
2006-10-03 01:15:41 -07:00
|
|
|
unsigned int len, struct bio_set *bs)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct bio *clone;
|
|
|
|
struct bio_vec *bv = bio->bi_io_vec + idx;
|
|
|
|
|
2006-10-03 01:15:41 -07:00
|
|
|
clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
|
2005-09-06 15:16:42 -07:00
|
|
|
clone->bi_destructor = dm_bio_destructor;
|
2005-04-16 15:20:36 -07:00
|
|
|
*clone->bi_io_vec = *bv;
|
|
|
|
|
|
|
|
clone->bi_sector = sector;
|
|
|
|
clone->bi_bdev = bio->bi_bdev;
|
|
|
|
clone->bi_rw = bio->bi_rw;
|
|
|
|
clone->bi_vcnt = 1;
|
|
|
|
clone->bi_size = to_bytes(len);
|
|
|
|
clone->bi_io_vec->bv_offset = offset;
|
|
|
|
clone->bi_io_vec->bv_len = clone->bi_size;
|
2008-10-21 09:45:04 -07:00
|
|
|
clone->bi_flags |= 1 << BIO_CLONED;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
return clone;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Creates a bio that consists of range of complete bvecs.
|
|
|
|
*/
|
|
|
|
static struct bio *clone_bio(struct bio *bio, sector_t sector,
|
|
|
|
unsigned short idx, unsigned short bv_count,
|
2006-10-03 01:15:41 -07:00
|
|
|
unsigned int len, struct bio_set *bs)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct bio *clone;
|
|
|
|
|
2006-10-03 01:15:41 -07:00
|
|
|
clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
|
|
|
|
__bio_clone(clone, bio);
|
|
|
|
clone->bi_destructor = dm_bio_destructor;
|
2005-04-16 15:20:36 -07:00
|
|
|
clone->bi_sector = sector;
|
|
|
|
clone->bi_idx = idx;
|
|
|
|
clone->bi_vcnt = idx + bv_count;
|
|
|
|
clone->bi_size = to_bytes(len);
|
|
|
|
clone->bi_flags &= ~(1 << BIO_SEG_VALID);
|
|
|
|
|
|
|
|
return clone;
|
|
|
|
}
|
|
|
|
|
2007-12-13 07:15:25 -07:00
|
|
|
static int __clone_and_map(struct clone_info *ci)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct bio *clone, *bio = ci->bio;
|
2007-12-13 07:15:25 -07:00
|
|
|
struct dm_target *ti;
|
|
|
|
sector_t len = 0, max;
|
2007-07-12 09:26:32 -07:00
|
|
|
struct dm_target_io *tio;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-12-13 07:15:25 -07:00
|
|
|
ti = dm_table_find_target(ci->map, ci->sector);
|
|
|
|
if (!dm_target_is_valid(ti))
|
|
|
|
return -EIO;
|
|
|
|
|
|
|
|
max = max_io_len(ci->md, ci->sector, ti);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Allocate a target io object.
|
|
|
|
*/
|
|
|
|
tio = alloc_tio(ci->md);
|
|
|
|
tio->io = ci->io;
|
|
|
|
tio->ti = ti;
|
|
|
|
memset(&tio->info, 0, sizeof(tio->info));
|
|
|
|
|
|
|
|
if (ci->sector_count <= max) {
|
|
|
|
/*
|
|
|
|
* Optimise for the simple case where we can do all of
|
|
|
|
* the remaining io with a single clone.
|
|
|
|
*/
|
|
|
|
clone = clone_bio(bio, ci->sector, ci->idx,
|
2006-10-03 01:15:41 -07:00
|
|
|
bio->bi_vcnt - ci->idx, ci->sector_count,
|
|
|
|
ci->md->bs);
|
2005-04-16 15:20:36 -07:00
|
|
|
__map_bio(ti, clone, tio);
|
|
|
|
ci->sector_count = 0;
|
|
|
|
|
|
|
|
} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
|
|
|
|
/*
|
|
|
|
* There are some bvecs that don't span targets.
|
|
|
|
* Do as many of these as possible.
|
|
|
|
*/
|
|
|
|
int i;
|
|
|
|
sector_t remaining = max;
|
|
|
|
sector_t bv_len;
|
|
|
|
|
|
|
|
for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
|
|
|
|
bv_len = to_sector(bio->bi_io_vec[i].bv_len);
|
|
|
|
|
|
|
|
if (bv_len > remaining)
|
|
|
|
break;
|
|
|
|
|
|
|
|
remaining -= bv_len;
|
|
|
|
len += bv_len;
|
|
|
|
}
|
|
|
|
|
2006-10-03 01:15:41 -07:00
|
|
|
clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
|
|
|
|
ci->md->bs);
|
2005-04-16 15:20:36 -07:00
|
|
|
__map_bio(ti, clone, tio);
|
|
|
|
|
|
|
|
ci->sector += len;
|
|
|
|
ci->sector_count -= len;
|
|
|
|
ci->idx = i;
|
|
|
|
|
|
|
|
} else {
|
|
|
|
/*
|
2006-03-22 01:07:42 -07:00
|
|
|
* Handle a bvec that must be split between two or more targets.
|
2005-04-16 15:20:36 -07:00
|
|
|
*/
|
|
|
|
struct bio_vec *bv = bio->bi_io_vec + ci->idx;
|
2006-03-22 01:07:42 -07:00
|
|
|
sector_t remaining = to_sector(bv->bv_len);
|
|
|
|
unsigned int offset = 0;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-03-22 01:07:42 -07:00
|
|
|
do {
|
|
|
|
if (offset) {
|
|
|
|
ti = dm_table_find_target(ci->map, ci->sector);
|
2007-12-13 07:15:25 -07:00
|
|
|
if (!dm_target_is_valid(ti))
|
|
|
|
return -EIO;
|
|
|
|
|
2006-03-22 01:07:42 -07:00
|
|
|
max = max_io_len(ci->md, ci->sector, ti);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-03-22 01:07:42 -07:00
|
|
|
tio = alloc_tio(ci->md);
|
|
|
|
tio->io = ci->io;
|
|
|
|
tio->ti = ti;
|
|
|
|
memset(&tio->info, 0, sizeof(tio->info));
|
|
|
|
}
|
|
|
|
|
|
|
|
len = min(remaining, max);
|
|
|
|
|
|
|
|
clone = split_bvec(bio, ci->sector, ci->idx,
|
2006-10-03 01:15:41 -07:00
|
|
|
bv->bv_offset + offset, len,
|
|
|
|
ci->md->bs);
|
2006-03-22 01:07:42 -07:00
|
|
|
|
|
|
|
__map_bio(ti, clone, tio);
|
|
|
|
|
|
|
|
ci->sector += len;
|
|
|
|
ci->sector_count -= len;
|
|
|
|
offset += to_bytes(len);
|
|
|
|
} while (remaining -= len);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
ci->idx++;
|
|
|
|
}
|
2007-12-13 07:15:25 -07:00
|
|
|
|
|
|
|
return 0;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Split the bio into several clones.
|
|
|
|
*/
|
2007-10-19 14:38:53 -07:00
|
|
|
static int __split_bio(struct mapped_device *md, struct bio *bio)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct clone_info ci;
|
2007-12-13 07:15:25 -07:00
|
|
|
int error = 0;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
ci.map = dm_get_table(md);
|
2007-10-19 14:38:53 -07:00
|
|
|
if (unlikely(!ci.map))
|
|
|
|
return -EIO;
|
2009-01-05 20:05:09 -07:00
|
|
|
if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) {
|
|
|
|
dm_table_put(ci.map);
|
|
|
|
bio_endio(bio, -EOPNOTSUPP);
|
|
|
|
return 0;
|
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
ci.md = md;
|
|
|
|
ci.bio = bio;
|
|
|
|
ci.io = alloc_io(md);
|
|
|
|
ci.io->error = 0;
|
|
|
|
atomic_set(&ci.io->io_count, 1);
|
|
|
|
ci.io->bio = bio;
|
|
|
|
ci.io->md = md;
|
|
|
|
ci.sector = bio->bi_sector;
|
|
|
|
ci.sector_count = bio_sectors(bio);
|
|
|
|
ci.idx = bio->bi_idx;
|
|
|
|
|
2006-02-01 04:04:53 -07:00
|
|
|
start_io_acct(ci.io);
|
2007-12-13 07:15:25 -07:00
|
|
|
while (ci.sector_count && !error)
|
|
|
|
error = __clone_and_map(&ci);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/* drop the extra reference count */
|
2007-12-13 07:15:25 -07:00
|
|
|
dec_pending(ci.io, error);
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_table_put(ci.map);
|
2007-10-19 14:38:53 -07:00
|
|
|
|
|
|
|
return 0;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
/*-----------------------------------------------------------------
|
|
|
|
* CRUD END
|
|
|
|
*---------------------------------------------------------------*/
|
|
|
|
|
2008-07-21 04:00:37 -07:00
|
|
|
static int dm_merge_bvec(struct request_queue *q,
|
|
|
|
struct bvec_merge_data *bvm,
|
|
|
|
struct bio_vec *biovec)
|
|
|
|
{
|
|
|
|
struct mapped_device *md = q->queuedata;
|
|
|
|
struct dm_table *map = dm_get_table(md);
|
|
|
|
struct dm_target *ti;
|
|
|
|
sector_t max_sectors;
|
2008-10-01 06:39:17 -07:00
|
|
|
int max_size = 0;
|
2008-07-21 04:00:37 -07:00
|
|
|
|
|
|
|
if (unlikely(!map))
|
2008-10-01 06:39:17 -07:00
|
|
|
goto out;
|
2008-07-21 04:00:37 -07:00
|
|
|
|
|
|
|
ti = dm_table_find_target(map, bvm->bi_sector);
|
2008-10-01 06:39:24 -07:00
|
|
|
if (!dm_target_is_valid(ti))
|
|
|
|
goto out_table;
|
2008-07-21 04:00:37 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Find maximum amount of I/O that won't need splitting
|
|
|
|
*/
|
|
|
|
max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
|
|
|
|
(sector_t) BIO_MAX_SECTORS);
|
|
|
|
max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
|
|
|
|
if (max_size < 0)
|
|
|
|
max_size = 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* merge_bvec_fn() returns number of bytes
|
|
|
|
* it can accept at this offset
|
|
|
|
* max is precomputed maximal io size
|
|
|
|
*/
|
|
|
|
if (max_size && ti->type->merge)
|
|
|
|
max_size = ti->type->merge(ti, bvm, biovec, max_size);
|
|
|
|
|
2008-10-01 06:39:24 -07:00
|
|
|
out_table:
|
2008-10-01 06:39:17 -07:00
|
|
|
dm_table_put(map);
|
|
|
|
|
|
|
|
out:
|
2008-07-21 04:00:37 -07:00
|
|
|
/*
|
|
|
|
* Always allow an entire first page
|
|
|
|
*/
|
|
|
|
if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
|
|
|
|
max_size = biovec->bv_len;
|
|
|
|
|
|
|
|
return max_size;
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* The request function that just remaps the bio built up by
|
|
|
|
* dm_merge_bvec.
|
|
|
|
*/
|
2007-07-24 00:28:11 -07:00
|
|
|
static int dm_request(struct request_queue *q, struct bio *bio)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2007-10-19 14:38:53 -07:00
|
|
|
int r = -EIO;
|
2006-02-01 04:04:52 -07:00
|
|
|
int rw = bio_data_dir(bio);
|
2005-04-16 15:20:36 -07:00
|
|
|
struct mapped_device *md = q->queuedata;
|
2008-08-25 03:47:21 -07:00
|
|
|
int cpu;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
down_read(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-08-25 03:56:14 -07:00
|
|
|
cpu = part_stat_lock();
|
|
|
|
part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
|
|
|
|
part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
|
|
|
|
part_stat_unlock();
|
2006-02-01 04:04:52 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* If we're suspended we have to queue
|
|
|
|
* this io for later.
|
|
|
|
*/
|
|
|
|
while (test_bit(DMF_BLOCK_IO, &md->flags)) {
|
2005-07-28 21:16:00 -07:00
|
|
|
up_read(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-10-19 14:38:53 -07:00
|
|
|
if (bio_rw(bio) != READA)
|
|
|
|
r = queue_io(md, bio);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-10-19 14:38:53 -07:00
|
|
|
if (r <= 0)
|
|
|
|
goto out_req;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We're in a while loop, because someone could suspend
|
|
|
|
* before we get to the following read lock.
|
|
|
|
*/
|
2005-07-28 21:16:00 -07:00
|
|
|
down_read(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
2007-10-19 14:38:53 -07:00
|
|
|
r = __split_bio(md, bio);
|
2005-07-28 21:16:00 -07:00
|
|
|
up_read(&md->io_lock);
|
2007-10-19 14:38:53 -07:00
|
|
|
|
|
|
|
out_req:
|
|
|
|
if (r < 0)
|
|
|
|
bio_io_error(bio);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-07-24 00:28:11 -07:00
|
|
|
static void dm_unplug_all(struct request_queue *q)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct mapped_device *md = q->queuedata;
|
|
|
|
struct dm_table *map = dm_get_table(md);
|
|
|
|
|
|
|
|
if (map) {
|
|
|
|
dm_table_unplug_all(map);
|
|
|
|
dm_table_put(map);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static int dm_any_congested(void *congested_data, int bdi_bits)
|
|
|
|
{
|
2008-11-13 16:39:14 -07:00
|
|
|
int r = bdi_bits;
|
|
|
|
struct mapped_device *md = congested_data;
|
|
|
|
struct dm_table *map;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-11-13 16:39:14 -07:00
|
|
|
if (!test_bit(DMF_BLOCK_IO, &md->flags)) {
|
|
|
|
map = dm_get_table(md);
|
|
|
|
if (map) {
|
|
|
|
r = dm_table_any_congested(map, bdi_bits);
|
|
|
|
dm_table_put(map);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*-----------------------------------------------------------------
|
|
|
|
* An IDR is used to keep track of allocated minor numbers.
|
|
|
|
*---------------------------------------------------------------*/
|
|
|
|
static DEFINE_IDR(_minor_idr);
|
|
|
|
|
2006-06-26 00:27:32 -07:00
|
|
|
static void free_minor(int minor)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_lock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
idr_remove(&_minor_idr, minor);
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* See if the device with a specific minor # is free.
|
|
|
|
*/
|
2008-04-24 14:10:59 -07:00
|
|
|
static int specific_minor(int minor)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
int r, m;
|
|
|
|
|
|
|
|
if (minor >= (1 << MINORBITS))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-06-26 00:27:21 -07:00
|
|
|
r = idr_pre_get(&_minor_idr, GFP_KERNEL);
|
|
|
|
if (!r)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_lock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
if (idr_find(&_minor_idr, minor)) {
|
|
|
|
r = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2006-06-26 00:27:21 -07:00
|
|
|
r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
|
2006-06-26 00:27:21 -07:00
|
|
|
if (r)
|
2005-04-16 15:20:36 -07:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (m != minor) {
|
|
|
|
idr_remove(&_minor_idr, m);
|
|
|
|
r = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2008-04-24 14:10:59 -07:00
|
|
|
static int next_free_minor(int *minor)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2006-06-26 00:27:32 -07:00
|
|
|
int r, m;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
r = idr_pre_get(&_minor_idr, GFP_KERNEL);
|
2006-06-26 00:27:21 -07:00
|
|
|
if (!r)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_lock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-26 00:27:21 -07:00
|
|
|
r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
|
2008-04-24 14:10:59 -07:00
|
|
|
if (r)
|
2005-04-16 15:20:36 -07:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (m >= (1 << MINORBITS)) {
|
|
|
|
idr_remove(&_minor_idr, m);
|
|
|
|
r = -ENOSPC;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
*minor = m;
|
|
|
|
|
|
|
|
out:
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct block_device_operations dm_blk_dops;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Allocate and initialise a blank device with a given minor.
|
|
|
|
*/
|
2006-06-26 00:27:32 -07:00
|
|
|
static struct mapped_device *alloc_dev(int minor)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
int r;
|
2008-04-24 14:10:59 -07:00
|
|
|
struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
|
2006-06-26 00:27:21 -07:00
|
|
|
void *old_md;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
if (!md) {
|
|
|
|
DMWARN("unable to allocate device, out of memory.");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2006-06-26 00:27:25 -07:00
|
|
|
if (!try_module_get(THIS_MODULE))
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_module_get;
|
2006-06-26 00:27:25 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/* get a minor number for the dev */
|
2006-06-26 00:27:32 -07:00
|
|
|
if (minor == DM_ANY_MINOR)
|
2008-04-24 14:10:59 -07:00
|
|
|
r = next_free_minor(&minor);
|
2006-06-26 00:27:32 -07:00
|
|
|
else
|
2008-04-24 14:10:59 -07:00
|
|
|
r = specific_minor(minor);
|
2005-04-16 15:20:36 -07:00
|
|
|
if (r < 0)
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_minor;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
init_rwsem(&md->io_lock);
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_init(&md->suspend_lock);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
spin_lock_init(&md->pushback_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
rwlock_init(&md->map_lock);
|
|
|
|
atomic_set(&md->holders, 1);
|
2006-06-26 00:27:34 -07:00
|
|
|
atomic_set(&md->open_count, 0);
|
2005-04-16 15:20:36 -07:00
|
|
|
atomic_set(&md->event_nr, 0);
|
2007-10-19 14:48:01 -07:00
|
|
|
atomic_set(&md->uevent_seq, 0);
|
|
|
|
INIT_LIST_HEAD(&md->uevent_list);
|
|
|
|
spin_lock_init(&md->uevent_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
md->queue = blk_alloc_queue(GFP_KERNEL);
|
|
|
|
if (!md->queue)
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_queue;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
md->queue->queuedata = md;
|
|
|
|
md->queue->backing_dev_info.congested_fn = dm_any_congested;
|
|
|
|
md->queue->backing_dev_info.congested_data = md;
|
|
|
|
blk_queue_make_request(md->queue, dm_request);
|
2006-01-10 02:48:02 -07:00
|
|
|
blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
|
2005-04-16 15:20:36 -07:00
|
|
|
md->queue->unplug_fn = dm_unplug_all;
|
2008-07-21 04:00:37 -07:00
|
|
|
blk_queue_merge_bvec(md->queue, dm_merge_bvec);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-03-26 02:37:50 -07:00
|
|
|
md->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache);
|
2006-12-08 03:41:02 -07:00
|
|
|
if (!md->io_pool)
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_io_pool;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-03-26 02:37:50 -07:00
|
|
|
md->tio_pool = mempool_create_slab_pool(MIN_IOS, _tio_cache);
|
2005-04-16 15:20:36 -07:00
|
|
|
if (!md->tio_pool)
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_tio_pool;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-12-10 07:35:05 -07:00
|
|
|
md->bs = bioset_create(16, 0);
|
2006-10-03 01:15:41 -07:00
|
|
|
if (!md->bs)
|
|
|
|
goto bad_no_bioset;
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
md->disk = alloc_disk(1);
|
|
|
|
if (!md->disk)
|
2008-02-07 19:10:19 -07:00
|
|
|
goto bad_disk;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-26 00:27:25 -07:00
|
|
|
atomic_set(&md->pending, 0);
|
|
|
|
init_waitqueue_head(&md->wait);
|
|
|
|
init_waitqueue_head(&md->eventq);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
md->disk->major = _major;
|
|
|
|
md->disk->first_minor = minor;
|
|
|
|
md->disk->fops = &dm_blk_dops;
|
|
|
|
md->disk->queue = md->queue;
|
|
|
|
md->disk->private_data = md;
|
|
|
|
sprintf(md->disk->disk_name, "dm-%d", minor);
|
|
|
|
add_disk(md->disk);
|
2006-03-27 02:17:52 -07:00
|
|
|
format_dev_t(md->name, MKDEV(_major, minor));
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
md->wq = create_singlethread_workqueue("kdmflush");
|
|
|
|
if (!md->wq)
|
|
|
|
goto bad_thread;
|
|
|
|
|
2006-06-26 00:27:21 -07:00
|
|
|
/* Populate the mapping, nobody knows we exist yet */
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_lock(&_minor_lock);
|
2006-06-26 00:27:21 -07:00
|
|
|
old_md = idr_replace(&_minor_idr, md, minor);
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2006-06-26 00:27:21 -07:00
|
|
|
|
|
|
|
BUG_ON(old_md != MINOR_ALLOCED);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
return md;
|
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
bad_thread:
|
|
|
|
put_disk(md->disk);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_disk:
|
2006-10-03 01:15:41 -07:00
|
|
|
bioset_free(md->bs);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_no_bioset:
|
2005-04-16 15:20:36 -07:00
|
|
|
mempool_destroy(md->tio_pool);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_tio_pool:
|
2005-04-16 15:20:36 -07:00
|
|
|
mempool_destroy(md->io_pool);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_io_pool:
|
2006-03-12 09:02:03 -07:00
|
|
|
blk_cleanup_queue(md->queue);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_queue:
|
2005-04-16 15:20:36 -07:00
|
|
|
free_minor(minor);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_minor:
|
2006-06-26 00:27:25 -07:00
|
|
|
module_put(THIS_MODULE);
|
2008-02-07 19:10:19 -07:00
|
|
|
bad_module_get:
|
2005-04-16 15:20:36 -07:00
|
|
|
kfree(md);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2007-10-19 14:38:43 -07:00
|
|
|
static void unlock_fs(struct mapped_device *md);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
static void free_dev(struct mapped_device *md)
|
|
|
|
{
|
2008-09-03 00:01:48 -07:00
|
|
|
int minor = MINOR(disk_devt(md->disk));
|
2006-02-24 14:04:25 -07:00
|
|
|
|
2006-02-24 14:04:24 -07:00
|
|
|
if (md->suspended_bdev) {
|
2007-10-19 14:38:43 -07:00
|
|
|
unlock_fs(md);
|
2006-02-24 14:04:24 -07:00
|
|
|
bdput(md->suspended_bdev);
|
|
|
|
}
|
2008-02-07 19:11:17 -07:00
|
|
|
destroy_workqueue(md->wq);
|
2005-04-16 15:20:36 -07:00
|
|
|
mempool_destroy(md->tio_pool);
|
|
|
|
mempool_destroy(md->io_pool);
|
2006-10-03 01:15:41 -07:00
|
|
|
bioset_free(md->bs);
|
2005-04-16 15:20:36 -07:00
|
|
|
del_gendisk(md->disk);
|
2006-02-24 14:04:25 -07:00
|
|
|
free_minor(minor);
|
2006-06-26 00:27:23 -07:00
|
|
|
|
|
|
|
spin_lock(&_minor_lock);
|
|
|
|
md->disk->private_data = NULL;
|
|
|
|
spin_unlock(&_minor_lock);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
put_disk(md->disk);
|
2006-03-12 09:02:03 -07:00
|
|
|
blk_cleanup_queue(md->queue);
|
2006-06-26 00:27:25 -07:00
|
|
|
module_put(THIS_MODULE);
|
2005-04-16 15:20:36 -07:00
|
|
|
kfree(md);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Bind a table to the device.
|
|
|
|
*/
|
|
|
|
static void event_callback(void *context)
|
|
|
|
{
|
2007-10-19 14:48:01 -07:00
|
|
|
unsigned long flags;
|
|
|
|
LIST_HEAD(uevents);
|
2005-04-16 15:20:36 -07:00
|
|
|
struct mapped_device *md = (struct mapped_device *) context;
|
|
|
|
|
2007-10-19 14:48:01 -07:00
|
|
|
spin_lock_irqsave(&md->uevent_lock, flags);
|
|
|
|
list_splice_init(&md->uevent_list, &uevents);
|
|
|
|
spin_unlock_irqrestore(&md->uevent_lock, flags);
|
|
|
|
|
2008-08-25 03:56:05 -07:00
|
|
|
dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
|
2007-10-19 14:48:01 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
atomic_inc(&md->event_nr);
|
|
|
|
wake_up(&md->eventq);
|
|
|
|
}
|
|
|
|
|
2005-07-28 21:15:59 -07:00
|
|
|
static void __set_size(struct mapped_device *md, sector_t size)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2005-07-28 21:15:59 -07:00
|
|
|
set_capacity(md->disk, size);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-01-09 16:59:24 -07:00
|
|
|
mutex_lock(&md->suspended_bdev->bd_inode->i_mutex);
|
2006-01-06 01:20:05 -07:00
|
|
|
i_size_write(md->suspended_bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
|
2006-01-09 16:59:24 -07:00
|
|
|
mutex_unlock(&md->suspended_bdev->bd_inode->i_mutex);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static int __bind(struct mapped_device *md, struct dm_table *t)
|
|
|
|
{
|
2007-07-24 00:28:11 -07:00
|
|
|
struct request_queue *q = md->queue;
|
2005-04-16 15:20:36 -07:00
|
|
|
sector_t size;
|
|
|
|
|
|
|
|
size = dm_table_get_size(t);
|
2006-03-27 02:17:54 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Wipe any geometry if the size of the table changed.
|
|
|
|
*/
|
|
|
|
if (size != get_capacity(md->disk))
|
|
|
|
memset(&md->geometry, 0, sizeof(md->geometry));
|
|
|
|
|
2007-01-26 01:57:07 -07:00
|
|
|
if (md->suspended_bdev)
|
|
|
|
__set_size(md, size);
|
dm table: rework reference counting
Rework table reference counting.
The existing code uses a reference counter. When the last reference is
dropped and the counter reaches zero, the table destructor is called.
Table reference counters are acquired/released from upcalls from other
kernel code (dm_any_congested, dm_merge_bvec, dm_unplug_all).
If the reference counter reaches zero in one of the upcalls, the table
destructor is called from almost random kernel code.
This leads to various problems:
* dm_any_congested being called under a spinlock, which calls the
destructor, which calls some sleeping function.
* the destructor attempting to take a lock that is already taken by the
same process.
* stale reference from some other kernel code keeps the table
constructed, which keeps some devices open, even after successful
return from "dmsetup remove". This can confuse lvm and prevent closing
of underlying devices or reusing device minor numbers.
The patch changes reference counting so that the table destructor can be
called only at predetermined places.
The table has always exactly one reference from either mapped_device->map
or hash_cell->new_map. After this patch, this reference is not counted
in table->holders. A pair of dm_create_table/dm_destroy_table functions
is used for table creation/destruction.
Temporary references from the other code increase table->holders. A pair
of dm_table_get/dm_table_put functions is used to manipulate it.
When the table is about to be destroyed, we wait for table->holders to
reach 0. Then, we call the table destructor. We use active waiting with
msleep(1), because the situation happens rarely (to one user in 5 years)
and removing the device isn't performance-critical task: the user doesn't
care if it takes one tick more or not.
This way, the destructor is called only at specific points
(dm_table_destroy function) and the above problems associated with lazy
destruction can't happen.
Finally remove the temporary protection added to dm_any_congested().
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2009-01-05 20:05:10 -07:00
|
|
|
|
|
|
|
if (!size) {
|
|
|
|
dm_table_destroy(t);
|
2005-04-16 15:20:36 -07:00
|
|
|
return 0;
|
dm table: rework reference counting
Rework table reference counting.
The existing code uses a reference counter. When the last reference is
dropped and the counter reaches zero, the table destructor is called.
Table reference counters are acquired/released from upcalls from other
kernel code (dm_any_congested, dm_merge_bvec, dm_unplug_all).
If the reference counter reaches zero in one of the upcalls, the table
destructor is called from almost random kernel code.
This leads to various problems:
* dm_any_congested being called under a spinlock, which calls the
destructor, which calls some sleeping function.
* the destructor attempting to take a lock that is already taken by the
same process.
* stale reference from some other kernel code keeps the table
constructed, which keeps some devices open, even after successful
return from "dmsetup remove". This can confuse lvm and prevent closing
of underlying devices or reusing device minor numbers.
The patch changes reference counting so that the table destructor can be
called only at predetermined places.
The table has always exactly one reference from either mapped_device->map
or hash_cell->new_map. After this patch, this reference is not counted
in table->holders. A pair of dm_create_table/dm_destroy_table functions
is used for table creation/destruction.
Temporary references from the other code increase table->holders. A pair
of dm_table_get/dm_table_put functions is used to manipulate it.
When the table is about to be destroyed, we wait for table->holders to
reach 0. Then, we call the table destructor. We use active waiting with
msleep(1), because the situation happens rarely (to one user in 5 years)
and removing the device isn't performance-critical task: the user doesn't
care if it takes one tick more or not.
This way, the destructor is called only at specific points
(dm_table_destroy function) and the above problems associated with lazy
destruction can't happen.
Finally remove the temporary protection added to dm_any_congested().
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2009-01-05 20:05:10 -07:00
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
dm_table_event_callback(t, event_callback, md);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
write_lock(&md->map_lock);
|
|
|
|
md->map = t;
|
2005-07-28 21:16:00 -07:00
|
|
|
dm_table_set_restrictions(t, q);
|
2005-04-16 15:20:36 -07:00
|
|
|
write_unlock(&md->map_lock);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __unbind(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
struct dm_table *map = md->map;
|
|
|
|
|
|
|
|
if (!map)
|
|
|
|
return;
|
|
|
|
|
|
|
|
dm_table_event_callback(map, NULL, NULL);
|
|
|
|
write_lock(&md->map_lock);
|
|
|
|
md->map = NULL;
|
|
|
|
write_unlock(&md->map_lock);
|
dm table: rework reference counting
Rework table reference counting.
The existing code uses a reference counter. When the last reference is
dropped and the counter reaches zero, the table destructor is called.
Table reference counters are acquired/released from upcalls from other
kernel code (dm_any_congested, dm_merge_bvec, dm_unplug_all).
If the reference counter reaches zero in one of the upcalls, the table
destructor is called from almost random kernel code.
This leads to various problems:
* dm_any_congested being called under a spinlock, which calls the
destructor, which calls some sleeping function.
* the destructor attempting to take a lock that is already taken by the
same process.
* stale reference from some other kernel code keeps the table
constructed, which keeps some devices open, even after successful
return from "dmsetup remove". This can confuse lvm and prevent closing
of underlying devices or reusing device minor numbers.
The patch changes reference counting so that the table destructor can be
called only at predetermined places.
The table has always exactly one reference from either mapped_device->map
or hash_cell->new_map. After this patch, this reference is not counted
in table->holders. A pair of dm_create_table/dm_destroy_table functions
is used for table creation/destruction.
Temporary references from the other code increase table->holders. A pair
of dm_table_get/dm_table_put functions is used to manipulate it.
When the table is about to be destroyed, we wait for table->holders to
reach 0. Then, we call the table destructor. We use active waiting with
msleep(1), because the situation happens rarely (to one user in 5 years)
and removing the device isn't performance-critical task: the user doesn't
care if it takes one tick more or not.
This way, the destructor is called only at specific points
(dm_table_destroy function) and the above problems associated with lazy
destruction can't happen.
Finally remove the temporary protection added to dm_any_congested().
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
2009-01-05 20:05:10 -07:00
|
|
|
dm_table_destroy(map);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Constructor for a new device.
|
|
|
|
*/
|
2006-06-26 00:27:32 -07:00
|
|
|
int dm_create(int minor, struct mapped_device **result)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct mapped_device *md;
|
|
|
|
|
2006-06-26 00:27:32 -07:00
|
|
|
md = alloc_dev(minor);
|
2005-04-16 15:20:36 -07:00
|
|
|
if (!md)
|
|
|
|
return -ENXIO;
|
|
|
|
|
2009-01-05 20:05:12 -07:00
|
|
|
dm_sysfs_init(md);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
*result = md;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-01-06 01:20:00 -07:00
|
|
|
static struct mapped_device *dm_find_md(dev_t dev)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
|
|
|
struct mapped_device *md;
|
|
|
|
unsigned minor = MINOR(dev);
|
|
|
|
|
|
|
|
if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
|
|
|
|
return NULL;
|
|
|
|
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_lock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
md = idr_find(&_minor_idr, minor);
|
2006-06-26 00:27:23 -07:00
|
|
|
if (md && (md == MINOR_ALLOCED ||
|
2008-09-03 00:01:48 -07:00
|
|
|
(MINOR(disk_devt(dm_disk(md))) != minor) ||
|
2006-06-26 00:27:33 -07:00
|
|
|
test_bit(DMF_FREEING, &md->flags))) {
|
2006-01-06 01:20:00 -07:00
|
|
|
md = NULL;
|
2006-06-26 00:27:23 -07:00
|
|
|
goto out;
|
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-26 00:27:23 -07:00
|
|
|
out:
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-01-06 01:20:00 -07:00
|
|
|
return md;
|
|
|
|
}
|
|
|
|
|
2006-01-06 01:20:01 -07:00
|
|
|
struct mapped_device *dm_get_md(dev_t dev)
|
|
|
|
{
|
|
|
|
struct mapped_device *md = dm_find_md(dev);
|
|
|
|
|
|
|
|
if (md)
|
|
|
|
dm_get(md);
|
|
|
|
|
|
|
|
return md;
|
|
|
|
}
|
|
|
|
|
2006-03-27 02:17:53 -07:00
|
|
|
void *dm_get_mdptr(struct mapped_device *md)
|
2006-01-06 01:20:00 -07:00
|
|
|
{
|
2006-03-27 02:17:53 -07:00
|
|
|
return md->interface_ptr;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
void dm_set_mdptr(struct mapped_device *md, void *ptr)
|
|
|
|
{
|
|
|
|
md->interface_ptr = ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
void dm_get(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
atomic_inc(&md->holders);
|
|
|
|
}
|
|
|
|
|
2006-06-26 00:27:35 -07:00
|
|
|
const char *dm_device_name(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return md->name;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(dm_device_name);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
void dm_put(struct mapped_device *md)
|
|
|
|
{
|
2006-03-27 02:17:54 -07:00
|
|
|
struct dm_table *map;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-26 00:27:23 -07:00
|
|
|
BUG_ON(test_bit(DMF_FREEING, &md->flags));
|
|
|
|
|
2006-06-26 00:27:22 -07:00
|
|
|
if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
|
2006-03-27 02:17:54 -07:00
|
|
|
map = dm_get_table(md);
|
2008-09-03 00:01:48 -07:00
|
|
|
idr_replace(&_minor_idr, MINOR_ALLOCED,
|
|
|
|
MINOR(disk_devt(dm_disk(md))));
|
2006-06-26 00:27:23 -07:00
|
|
|
set_bit(DMF_FREEING, &md->flags);
|
2006-06-26 00:27:22 -07:00
|
|
|
spin_unlock(&_minor_lock);
|
2005-07-28 21:15:57 -07:00
|
|
|
if (!dm_suspended(md)) {
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_table_presuspend_targets(map);
|
|
|
|
dm_table_postsuspend_targets(map);
|
|
|
|
}
|
2009-01-05 20:05:12 -07:00
|
|
|
dm_sysfs_exit(md);
|
2006-03-27 02:17:54 -07:00
|
|
|
dm_table_put(map);
|
2009-01-05 20:04:53 -07:00
|
|
|
__unbind(md);
|
2005-04-16 15:20:36 -07:00
|
|
|
free_dev(md);
|
|
|
|
}
|
|
|
|
}
|
2007-05-09 02:32:56 -07:00
|
|
|
EXPORT_SYMBOL_GPL(dm_put);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:10:30 -07:00
|
|
|
static int dm_wait_for_completion(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
int r = 0;
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
set_current_state(TASK_INTERRUPTIBLE);
|
|
|
|
|
|
|
|
smp_mb();
|
|
|
|
if (!atomic_read(&md->pending))
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (signal_pending(current)) {
|
|
|
|
r = -EINTR;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
io_schedule();
|
|
|
|
}
|
|
|
|
set_current_state(TASK_RUNNING);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Process the deferred bios
|
|
|
|
*/
|
2008-02-07 19:10:22 -07:00
|
|
|
static void __flush_deferred_io(struct mapped_device *md)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2008-02-07 19:10:22 -07:00
|
|
|
struct bio *c;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:10:22 -07:00
|
|
|
while ((c = bio_list_pop(&md->deferred))) {
|
2007-10-19 14:38:53 -07:00
|
|
|
if (__split_bio(md, c))
|
|
|
|
bio_io_error(c);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
2008-02-07 19:10:25 -07:00
|
|
|
|
|
|
|
clear_bit(DMF_BLOCK_IO, &md->flags);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
2008-02-07 19:10:22 -07:00
|
|
|
static void __merge_pushback_list(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&md->pushback_lock, flags);
|
|
|
|
clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
|
|
|
|
bio_list_merge_head(&md->deferred, &md->pushback);
|
|
|
|
bio_list_init(&md->pushback);
|
|
|
|
spin_unlock_irqrestore(&md->pushback_lock, flags);
|
|
|
|
}
|
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
static void dm_wq_work(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
|
|
|
|
struct mapped_device *md = req->md;
|
|
|
|
|
|
|
|
down_write(&md->io_lock);
|
|
|
|
switch (req->type) {
|
|
|
|
case DM_WQ_FLUSH_DEFERRED:
|
|
|
|
__flush_deferred_io(md);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DMERR("dm_wq_work: unrecognised work type %d", req->type);
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
up_write(&md->io_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void dm_wq_queue(struct mapped_device *md, int type, void *context,
|
|
|
|
struct dm_wq_req *req)
|
|
|
|
{
|
|
|
|
req->type = type;
|
|
|
|
req->md = md;
|
|
|
|
req->context = context;
|
|
|
|
INIT_WORK(&req->work, dm_wq_work);
|
|
|
|
queue_work(md->wq, &req->work);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void dm_queue_flush(struct mapped_device *md, int type, void *context)
|
|
|
|
{
|
|
|
|
struct dm_wq_req req;
|
|
|
|
|
|
|
|
dm_wq_queue(md, type, context, &req);
|
|
|
|
flush_workqueue(md->wq);
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* Swap in a new table (destroying old one).
|
|
|
|
*/
|
|
|
|
int dm_swap_table(struct mapped_device *md, struct dm_table *table)
|
|
|
|
{
|
2005-07-12 15:53:05 -07:00
|
|
|
int r = -EINVAL;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_lock(&md->suspend_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/* device must be suspended */
|
2005-07-28 21:15:57 -07:00
|
|
|
if (!dm_suspended(md))
|
2005-07-12 15:53:05 -07:00
|
|
|
goto out;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-01-26 01:57:07 -07:00
|
|
|
/* without bdev, the device size cannot be changed */
|
|
|
|
if (!md->suspended_bdev)
|
|
|
|
if (get_capacity(md->disk) != dm_table_get_size(table))
|
|
|
|
goto out;
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
__unbind(md);
|
|
|
|
r = __bind(md, table);
|
|
|
|
|
2005-07-12 15:53:05 -07:00
|
|
|
out:
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_unlock(&md->suspend_lock);
|
2005-07-12 15:53:05 -07:00
|
|
|
return r;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Functions to lock and unlock any filesystem running on the
|
|
|
|
* device.
|
|
|
|
*/
|
2005-07-28 21:16:00 -07:00
|
|
|
static int lock_fs(struct mapped_device *md)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2006-01-06 01:20:05 -07:00
|
|
|
int r;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
WARN_ON(md->frozen_sb);
|
2005-05-05 16:16:04 -07:00
|
|
|
|
2006-01-06 01:20:05 -07:00
|
|
|
md->frozen_sb = freeze_bdev(md->suspended_bdev);
|
2005-05-05 16:16:04 -07:00
|
|
|
if (IS_ERR(md->frozen_sb)) {
|
2005-07-28 21:15:57 -07:00
|
|
|
r = PTR_ERR(md->frozen_sb);
|
2006-01-06 01:20:05 -07:00
|
|
|
md->frozen_sb = NULL;
|
|
|
|
return r;
|
2005-05-05 16:16:04 -07:00
|
|
|
}
|
|
|
|
|
2006-01-06 01:20:06 -07:00
|
|
|
set_bit(DMF_FROZEN, &md->flags);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/* don't bdput right now, we don't want the bdev
|
2006-01-06 01:20:05 -07:00
|
|
|
* to go away while it is locked.
|
2005-04-16 15:20:36 -07:00
|
|
|
*/
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
static void unlock_fs(struct mapped_device *md)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2006-01-06 01:20:06 -07:00
|
|
|
if (!test_bit(DMF_FROZEN, &md->flags))
|
|
|
|
return;
|
|
|
|
|
2006-01-06 01:20:05 -07:00
|
|
|
thaw_bdev(md->suspended_bdev, md->frozen_sb);
|
2005-04-16 15:20:36 -07:00
|
|
|
md->frozen_sb = NULL;
|
2006-01-06 01:20:06 -07:00
|
|
|
clear_bit(DMF_FROZEN, &md->flags);
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We need to be able to change a mapping table under a mounted
|
|
|
|
* filesystem. For example we might want to move some data in
|
|
|
|
* the background. Before the table can be swapped with
|
|
|
|
* dm_bind_table, dm_suspend must be called to flush any in
|
|
|
|
* flight bios and ensure that any further io gets deferred.
|
|
|
|
*/
|
2006-12-08 03:41:04 -07:00
|
|
|
int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
|
2005-04-16 15:20:36 -07:00
|
|
|
{
|
2005-07-28 21:16:00 -07:00
|
|
|
struct dm_table *map = NULL;
|
2005-04-16 15:20:36 -07:00
|
|
|
DECLARE_WAITQUEUE(wait, current);
|
2008-02-07 19:10:30 -07:00
|
|
|
int r = 0;
|
2006-12-08 03:41:04 -07:00
|
|
|
int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_lock(&md->suspend_lock);
|
2005-07-28 21:16:00 -07:00
|
|
|
|
2008-02-07 19:10:25 -07:00
|
|
|
if (dm_suspended(md)) {
|
|
|
|
r = -EINVAL;
|
2006-11-08 18:44:43 -07:00
|
|
|
goto out_unlock;
|
2008-02-07 19:10:25 -07:00
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
map = dm_get_table(md);
|
|
|
|
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
/*
|
|
|
|
* DMF_NOFLUSH_SUSPENDING must be set before presuspend.
|
|
|
|
* This flag is cleared before dm_suspend returns.
|
|
|
|
*/
|
|
|
|
if (noflush)
|
|
|
|
set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
|
|
|
|
|
2005-07-28 21:15:57 -07:00
|
|
|
/* This does not get reverted if there's an error later. */
|
|
|
|
dm_table_presuspend_targets(map);
|
|
|
|
|
2007-01-26 01:57:07 -07:00
|
|
|
/* bdget() can stall if the pending I/Os are not flushed */
|
|
|
|
if (!noflush) {
|
|
|
|
md->suspended_bdev = bdget_disk(md->disk, 0);
|
|
|
|
if (!md->suspended_bdev) {
|
|
|
|
DMWARN("bdget failed in dm_suspend");
|
|
|
|
r = -ENOMEM;
|
2008-10-21 09:45:07 -07:00
|
|
|
goto out;
|
2007-01-26 01:57:07 -07:00
|
|
|
}
|
2006-01-06 01:20:05 -07:00
|
|
|
|
2008-02-07 19:10:22 -07:00
|
|
|
/*
|
|
|
|
* Flush I/O to the device. noflush supersedes do_lockfs,
|
|
|
|
* because lock_fs() needs to flush I/Os.
|
|
|
|
*/
|
|
|
|
if (do_lockfs) {
|
|
|
|
r = lock_fs(md);
|
|
|
|
if (r)
|
|
|
|
goto out;
|
|
|
|
}
|
2006-01-06 01:20:06 -07:00
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/*
|
2005-05-05 16:16:05 -07:00
|
|
|
* First we set the BLOCK_IO flag so no more ios will be mapped.
|
2005-04-16 15:20:36 -07:00
|
|
|
*/
|
2005-07-28 21:16:00 -07:00
|
|
|
down_write(&md->io_lock);
|
|
|
|
set_bit(DMF_BLOCK_IO, &md->flags);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
add_wait_queue(&md->wait, &wait);
|
2005-07-28 21:16:00 -07:00
|
|
|
up_write(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
/* unplug */
|
2005-07-28 21:16:00 -07:00
|
|
|
if (map)
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_table_unplug_all(map);
|
|
|
|
|
|
|
|
/*
|
2008-02-07 19:10:30 -07:00
|
|
|
* Wait for the already-mapped ios to complete.
|
2005-04-16 15:20:36 -07:00
|
|
|
*/
|
2008-02-07 19:10:30 -07:00
|
|
|
r = dm_wait_for_completion(md);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
down_write(&md->io_lock);
|
2005-04-16 15:20:36 -07:00
|
|
|
remove_wait_queue(&md->wait, &wait);
|
|
|
|
|
2008-02-07 19:10:22 -07:00
|
|
|
if (noflush)
|
|
|
|
__merge_pushback_list(md);
|
2008-02-07 19:10:27 -07:00
|
|
|
up_write(&md->io_lock);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/* were we interrupted ? */
|
2008-02-07 19:10:30 -07:00
|
|
|
if (r < 0) {
|
2008-02-07 19:11:17 -07:00
|
|
|
dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
|
2008-02-07 19:10:25 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
unlock_fs(md);
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
goto out; /* pushback list is already flushed, so skip flush */
|
2005-07-28 21:16:00 -07:00
|
|
|
}
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:15:57 -07:00
|
|
|
dm_table_postsuspend_targets(map);
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
set_bit(DMF_SUSPENDED, &md->flags);
|
2005-05-05 16:16:06 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
out:
|
2006-01-06 01:20:05 -07:00
|
|
|
if (r && md->suspended_bdev) {
|
|
|
|
bdput(md->suspended_bdev);
|
|
|
|
md->suspended_bdev = NULL;
|
|
|
|
}
|
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
dm_table_put(map);
|
2006-11-08 18:44:43 -07:00
|
|
|
|
|
|
|
out_unlock:
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_unlock(&md->suspend_lock);
|
2005-07-28 21:15:57 -07:00
|
|
|
return r;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
int dm_resume(struct mapped_device *md)
|
|
|
|
{
|
2005-07-28 21:15:57 -07:00
|
|
|
int r = -EINVAL;
|
|
|
|
struct dm_table *map = NULL;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_lock(&md->suspend_lock);
|
2005-07-28 21:16:00 -07:00
|
|
|
if (!dm_suspended(md))
|
2005-07-28 21:15:57 -07:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
map = dm_get_table(md);
|
2005-07-28 21:16:00 -07:00
|
|
|
if (!map || !dm_table_get_size(map))
|
2005-07-28 21:15:57 -07:00
|
|
|
goto out;
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-10-03 01:15:36 -07:00
|
|
|
r = dm_table_resume_targets(map);
|
|
|
|
if (r)
|
|
|
|
goto out;
|
2005-07-28 21:16:00 -07:00
|
|
|
|
2008-02-07 19:11:17 -07:00
|
|
|
dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
|
2005-07-28 21:16:00 -07:00
|
|
|
|
|
|
|
unlock_fs(md);
|
|
|
|
|
2007-01-26 01:57:07 -07:00
|
|
|
if (md->suspended_bdev) {
|
|
|
|
bdput(md->suspended_bdev);
|
|
|
|
md->suspended_bdev = NULL;
|
|
|
|
}
|
2006-01-06 01:20:05 -07:00
|
|
|
|
2005-07-28 21:16:00 -07:00
|
|
|
clear_bit(DMF_SUSPENDED, &md->flags);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
dm_table_unplug_all(map);
|
|
|
|
|
2007-12-13 07:15:57 -07:00
|
|
|
dm_kobject_uevent(md);
|
2006-10-03 01:15:35 -07:00
|
|
|
|
2005-07-28 21:15:57 -07:00
|
|
|
r = 0;
|
2005-07-28 21:16:00 -07:00
|
|
|
|
2005-07-28 21:15:57 -07:00
|
|
|
out:
|
|
|
|
dm_table_put(map);
|
2008-02-07 19:10:08 -07:00
|
|
|
mutex_unlock(&md->suspend_lock);
|
2005-07-28 21:16:00 -07:00
|
|
|
|
2005-07-28 21:15:57 -07:00
|
|
|
return r;
|
2005-04-16 15:20:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*-----------------------------------------------------------------
|
|
|
|
* Event notification.
|
|
|
|
*---------------------------------------------------------------*/
|
2007-12-13 07:15:57 -07:00
|
|
|
void dm_kobject_uevent(struct mapped_device *md)
|
|
|
|
{
|
2008-08-25 03:56:05 -07:00
|
|
|
kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
|
2007-12-13 07:15:57 -07:00
|
|
|
}
|
|
|
|
|
2007-10-19 14:48:01 -07:00
|
|
|
uint32_t dm_next_uevent_seq(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return atomic_add_return(1, &md->uevent_seq);
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
uint32_t dm_get_event_nr(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return atomic_read(&md->event_nr);
|
|
|
|
}
|
|
|
|
|
|
|
|
int dm_wait_event(struct mapped_device *md, int event_nr)
|
|
|
|
{
|
|
|
|
return wait_event_interruptible(md->eventq,
|
|
|
|
(event_nr != atomic_read(&md->event_nr)));
|
|
|
|
}
|
|
|
|
|
2007-10-19 14:48:01 -07:00
|
|
|
void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&md->uevent_lock, flags);
|
|
|
|
list_add(elist, &md->uevent_list);
|
|
|
|
spin_unlock_irqrestore(&md->uevent_lock, flags);
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
/*
|
|
|
|
* The gendisk is only valid as long as you have a reference
|
|
|
|
* count on 'md'.
|
|
|
|
*/
|
|
|
|
struct gendisk *dm_disk(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return md->disk;
|
|
|
|
}
|
|
|
|
|
2009-01-05 20:05:12 -07:00
|
|
|
struct kobject *dm_kobject(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return &md->kobj;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* struct mapped_device should not be exported outside of dm.c
|
|
|
|
* so use this check to verify that kobj is part of md structure
|
|
|
|
*/
|
|
|
|
struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
|
|
|
|
{
|
|
|
|
struct mapped_device *md;
|
|
|
|
|
|
|
|
md = container_of(kobj, struct mapped_device, kobj);
|
|
|
|
if (&md->kobj != kobj)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
dm_get(md);
|
|
|
|
return md;
|
|
|
|
}
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
int dm_suspended(struct mapped_device *md)
|
|
|
|
{
|
|
|
|
return test_bit(DMF_SUSPENDED, &md->flags);
|
|
|
|
}
|
|
|
|
|
[PATCH] dm: suspend: add noflush pushback
In device-mapper I/O is sometimes queued within targets for later processing.
For example the multipath target can be configured to store I/O when no paths
are available instead of returning it -EIO.
This patch allows the device-mapper core to instruct a target to transfer the
contents of any such in-target queue back into the core. This frees up the
resources used by the target so the core can replace that target with an
alternative one and then resend the I/O to it. Without this patch the only
way to change the target in such circumstances involves returning the I/O with
an error back to the filesystem/application. In the multipath case, this
patch will let us add new paths for existing I/O to try after all the existing
paths have failed.
DMF_NOFLUSH_SUSPENDING
----------------------
If the DM_NOFLUSH_FLAG ioctl option is specified at suspend time, the
DMF_NOFLUSH_SUSPENDING flag is set in md->flags during dm_suspend(). It
is always cleared before dm_suspend() returns.
The flag must be visible while the target is flushing pending I/Os so it
is set before presuspend where the flush starts and unset after the wait
for md->pending where the flush ends.
Target drivers can check this flag by calling dm_noflush_suspending().
DM_MAPIO_REQUEUE / DM_ENDIO_REQUEUE
-----------------------------------
A target's map() function can now return DM_MAPIO_REQUEUE to request the
device mapper core queue the bio.
Similarly, a target's end_io() function can return DM_ENDIO_REQUEUE to request
the same. This has been labelled 'pushback'.
The __map_bio() and clone_endio() functions in the core treat these return
values as errors and call dec_pending() to end the I/O.
dec_pending
-----------
dec_pending() saves the pushback request in struct dm_io->error. Once all
the split clones have ended, dec_pending() will put the original bio on
the md->pushback list. Note that this supercedes any I/O errors.
It is possible for the suspend with DM_NOFLUSH_FLAG to be aborted while
in progress (e.g. by user interrupt). dec_pending() checks for this and
returns -EIO if it happened.
pushdback list and pushback_lock
--------------------------------
The bio is queued on md->pushback temporarily in dec_pending(), and after
all pending I/Os return, md->pushback is merged into md->deferred in
dm_suspend() for re-issuing at resume time.
md->pushback_lock protects md->pushback.
The lock should be held with irq disabled because dec_pending() can be
called from interrupt context.
Queueing bios to md->pushback in dec_pending() must be done atomically
with the check for DMF_NOFLUSH_SUSPENDING flag. So md->pushback_lock is
held when checking the flag. Otherwise dec_pending() may queue a bio to
md->pushback after the interrupted dm_suspend() flushes md->pushback.
Then the bio would be left in md->pushback.
Flag setting in dm_suspend() can be done without md->pushback_lock because
the flag is checked only after presuspend and the set value is already
made visible via the target's presuspend function.
The flag can be checked without md->pushback_lock (e.g. the first part of
the dec_pending() or target drivers), because the flag is checked again
with md->pushback_lock held when the bio is really queued to md->pushback
as described above. So even if the flag is cleared after the lockless
checkings, the bio isn't left in md->pushback but returned to applications
with -EIO.
Other notes on the current patch
--------------------------------
- md->pushback is added to the struct mapped_device instead of using
md->deferred directly because md->io_lock which protects md->deferred is
rw_semaphore and can't be used in interrupt context like dec_pending(),
and md->io_lock protects the DMF_BLOCK_IO flag of md->flags too.
- Don't issue lock_fs() in dm_suspend() if the DM_NOFLUSH_FLAG
ioctl option is specified, because I/Os generated by lock_fs() would be
pushed back and never return if there were no valid devices.
- If an error occurs in dm_suspend() after the DMF_NOFLUSH_SUSPENDING
flag is set, md->pushback must be flushed because I/Os may be queued to
the list already. (flush_and_out label in dm_suspend())
Test results
------------
I have tested using multipath target with the next patch.
The following tests are for regression/compatibility:
- I/Os succeed when valid paths exist;
- I/Os fail when there are no valid paths and queue_if_no_path is not
set;
- I/Os are queued in the multipath target when there are no valid paths and
queue_if_no_path is set;
- The queued I/Os above fail when suspend is issued without the
DM_NOFLUSH_FLAG ioctl option. I/Os spanning 2 multipath targets also
fail.
The following tests are for the normal code path of new pushback feature:
- Queued I/Os in the multipath target are flushed from the target
but don't return when suspend is issued with the DM_NOFLUSH_FLAG
ioctl option;
- The I/Os above are queued in the multipath target again when
resume is issued without path recovery;
- The I/Os above succeed when resume is issued after path recovery
or table load;
- Queued I/Os in the multipath target succeed when resume is issued
with the DM_NOFLUSH_FLAG ioctl option after table load. I/Os
spanning 2 multipath targets also succeed.
The following tests are for the error paths of the new pushback feature:
- When the bdget_disk() fails in dm_suspend(), the
DMF_NOFLUSH_SUSPENDING flag is cleared and I/Os already queued to the
pushback list are flushed properly.
- When suspend with the DM_NOFLUSH_FLAG ioctl option is interrupted,
o I/Os which had already been queued to the pushback list
at the time don't return, and are re-issued at resume time;
o I/Os which hadn't been returned at the time return with EIO.
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-08 03:41:09 -07:00
|
|
|
int dm_noflush_suspending(struct dm_target *ti)
|
|
|
|
{
|
|
|
|
struct mapped_device *md = dm_table_get_md(ti->table);
|
|
|
|
int r = __noflush_suspending(md);
|
|
|
|
|
|
|
|
dm_put(md);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(dm_noflush_suspending);
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
static struct block_device_operations dm_blk_dops = {
|
|
|
|
.open = dm_blk_open,
|
|
|
|
.release = dm_blk_close,
|
2006-10-03 01:15:15 -07:00
|
|
|
.ioctl = dm_blk_ioctl,
|
2006-03-27 02:17:54 -07:00
|
|
|
.getgeo = dm_blk_getgeo,
|
2005-04-16 15:20:36 -07:00
|
|
|
.owner = THIS_MODULE
|
|
|
|
};
|
|
|
|
|
|
|
|
EXPORT_SYMBOL(dm_get_mapinfo);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* module hooks
|
|
|
|
*/
|
|
|
|
module_init(dm_init);
|
|
|
|
module_exit(dm_exit);
|
|
|
|
|
|
|
|
module_param(major, uint, 0);
|
|
|
|
MODULE_PARM_DESC(major, "The major number of the device mapper");
|
|
|
|
MODULE_DESCRIPTION(DM_NAME " driver");
|
|
|
|
MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
|
|
|
|
MODULE_LICENSE("GPL");
|