The choice was to either delay creation of the new UUID until
IO got thawed or to delay it until the first IO request.

Both are correct, the later is more friendly to users of
dual-primary setups, that actually only write on one side.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

If we detect late (= after grabing mdev->req_lock) that IO got frozen, we
return 1 to generic_make_request(), which simply will retry to make a
request for that bio.

In the subsequent call of generic_make_request() into drbd_make_request_26()
we sleep in inc_ap_bio().
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Now that the peer may handle multi-bio EEs,
we can ignore the peer's limit,
and concentrate on the limits of the local IO stack.

This is safe accross drbd protocol versions,
as our queue_max_sectors() will be adjusted accordingly.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

this should allow for better background resync performance.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

This should allow for better performance if the lower level IO stack
of the peers differs in limits exposed either via the queue,
or via some merge_bvec_fn.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

* Only send delay_probes with protocol 93 or newer
* drbd_send_delay_probes() is called only from worker context,
  no atomic_t needed for delay_seq
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

* Mention P_DELAY_PROBE in the packet naming array
* Do not corrupt the mdev->data.work list in case the timer goes
  off before delay_probe_work got handled by the worker
* Do not mod_timer() twice for a single delay_probe pair
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

In a setup with a high bandwidth and high latency network, eventually
involving deep queues in routers, it is beneficial to only fill those
queues up to an limited extend with resync data.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

To reasonably control resync speed over drbd-proxy connections,
drbd has to measure the current delay of packets transmitted over
the (possibly congested) data socket vs the meta-data socket.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Delay_probes are new packets in the DRBD protocol, which allow
DRBD to know the current delay packets have on the data socket.
(relative to the meta data socket)
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

The "surplus" bits of the old (smaller) bitmap must be clean
in case of online-grow without resync.

Note: Reverted 67ae8b80d4a116ab3b7094eb3723506b20c06dff as
well, since the lines added by this patch are redundant. The
bits get set by the bm_set_surplus(b) call before that.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Some wish to be notified of all instances of split brain, not just those that
go unresolved.  The initial-split-brain handler is called to notify someone
upon  detection of all split brain conditions even if auto-recovery policies
are configured.
Signed-off-by: NAdam Gandelman <adam.gandelman@linbit.com>
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

The condition does not fit the commend (I may well be Primary,
even if I lost the disk earlier and now the connection).

And this is catched below anyways, where it also gets logged.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Even if it should never happen if the peer does behave, we need to
double check, and not even attempt access beyond end of device.
It usually would be caught by lower layers, resulting in "IO error",
but may also end up in the internal meta data area.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

In case both nodes are "inconsistent", invalidate would
have started a resync anyways, without a chance to ever
succeed, just filling the logs with warning messages.

Simply disallow that state change,
re-using the SS_NO_UP_TO_DATE_DISK return value.

This also changes the corresponding error string to
"Need access to UpToDate Data" -- I found the
"Refusing to be Primary without at least one UpToDate disk"
answer misleading in some situations anyways.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Don't forget to drain the digest in case we cannot satisfy a
checksum based resync or online-verify request.

It would additionally cause a protocoll error,
dropping the connection.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

block_id may be ID_SYNCER,
as well as checksum based resync request magic, or online verify magic.

Let's just drop that ASSERT.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

commit e4f925e1
Author: Philipp Reisner <philipp.reisner@linbit.com>
Date:   Wed Mar 17 14:18:41 2010 +0100

    drbd: Do not upgrade state to Outdated if already Inconsistent

prevented the necessary state transition for attaching while connected
(Diskless -> Consistent respectively Outdated).
This is the fix for the fix.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

There was a race condition:
  In a situation with a SyncSource+Primary and a SyncTarget+Secondary node,
  and a resync dependency to some other device. After both nodes decided
  to do the resync, the other device finishes its resync process.
  At that time SyncSource already sent the P_SYNC_UUID packet, and
  already updated its peer disk state to Inconsistent.
  The SyncTarget node waits for the P_SYNC_UUID and sends a state packet
  to report the resync dependency change. That packet still carries
  a disk state of Outdated.

Impact:
  If application writes come in, during that time on the Primary node,
  those do not get replicated, and the out-of-sync counter gets increased.
  => The completion of resync is not detected on the primary node.
  => stalled.
  Those blocks get resync'ed with the next resync, since the are get
  marked as out-of-sync in the bitmap.

In order to fix this, we filter out that wrong state change in the
sanitize_state() function.
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

To document that we know about deprecation of proc_create,
even though we are not affected, as we don't use the ->data member,
open code proc_create_data(..., NULL);
Signed-off-by: NPhilipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: NLars Ellenberg <lars.ellenberg@linbit.com>

Filesystems with delalloc support may dirty inode during writepages.
As result inode will have dirty metadata flags even after write_inode.
In fact we have two dedicated functions for proper data and metadata
writeback. It is reasonable to separate flags updates in two stages.

https://bugzilla.kernel.org/show_bug.cgi?id=15906Signed-off-by: NDmitry Monakhov <dmonakhov@openvz.org>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

When umount calls sync_filesystem(), we first do a WB_SYNC_NONE
writeback to kick off writeback of pending dirty inodes, then follow
that up with a WB_SYNC_ALL to wait for it. Since umount already holds
the sb s_umount mutex, WB_SYNC_NONE ends up doing nothing and all
writeback happens as WB_SYNC_ALL. This can greatly slow down umount,
since WB_SYNC_ALL writeback is a data integrity operation and thus
a bigger hammer than simple WB_SYNC_NONE. For barrier aware file systems
it's a lot slower.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Prior to 2.6.32, setting /proc/sys/vm/dirty_writeback_centisecs disabled
periodic dirty writeback from kupdate. This got broken and now causes
excessive sys CPU usage if set to zero, as we'll keep beating on
schedule().

Cc: stable@kernel.org
Reported-by: NJustin Maggard <jmaggard10@gmail.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Make the PARIDE menu be displayed correctly, with proper/expected
indentation, by moving the GDROM kconfig symbol, which was
splitting the PARIDE kconfig symbol from its dependent symbols.
Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

blk_init_queue() allocates the request_queue structure and then
initializes it as needed (request_fn, elevator, etc).

Split initialization out to blk_init_allocated_queue_node.
Introduce blk_init_allocated_queue wrapper function to model existing
blk_init_queue and blk_init_queue_node interfaces.

Export elv_register_queue to allow a newly added elevator to be
registered with sysfs.  Export elv_unregister_queue for symmetry.

These changes allow DM to initialize a device's request_queue with more
precision.  In particular, DM no longer unconditionally initializes a
full request_queue (elevator et al).  It only does so for a
request-based DM device.
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

goto has its place, but lets cut back on some of the more
frivolous uses of it.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Conflicts:
	fs/block_dev.c
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>