linux.git/drivers/block/drbd, branch v2.6.35-rc4

drbd: Fixed a race between disk-attach and unexpected state changes

2010-06-14T10:19:41+00:00

This was a very hard to trigger race condition.

If we got a state packet from the peer, after drbd_nl_disk() has
already changed the disk state to D_NEGOTIATING but
after_state_ch() was not yet run by the worker, then receive_state()
might called drbd_sync_handshake(), which in turn crashed
when accessing p_uuid.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg

drbd: Reduce verbosity

2010-06-01T09:12:27+00:00

The "Local READ/WRITE failed" messages are too verbose.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: use drbd specific ratelimit instead of global printk_ratelimit

2010-06-01T09:12:27+00:00

using the global printk_ratelimit() may mask other messages.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: fix hang on local read errors while disconnected

2010-06-01T09:12:27+00:00

"canceled" w_read_retry_remote never completed, if they have been
canceled after drbd_disconnect connection teardown cleanup has already
run (or we are currently not connected anyways).

Fixed by not queueing a remote retry if we already know it won't work
(pdsk not uptodate), and cleanup ourselves on "cancel", in case we hit a
race with drbd_disconnect.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: Removed the now empty w_io_error() function

2010-06-01T09:12:27+00:00

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: removed duplicated #includes

2010-06-01T09:12:27+00:00

drbd/drbd_receiver.c: linux/mm.h is included more than once.

Signed-off-by: Andrea Gelmini 

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: improve usage of MSG_MORE

2010-06-01T09:12:27+00:00

It seems to improve performance if we allow the "p_data" header in its
own frame (no MSG_MORE), but sendpage all but the last page with MSG_MORE.
This is also in preparation of a later zero copy receive implementation.

Suggested by Eduard.Guzovsky@stratus.com on drbd-dev.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: need to set socket bufsize early to take effect

2010-06-01T09:12:27+00:00

quoting tcp(7):
    On individual connections, the socket buffer size must be set prior to the
    listen(2) or connect(2) calls in order to have it take effect.

This adds a wrapper to do so, and uses it appropriately.
Improves performance in certain situations.

Note that because we cannot easily determine which socket will be
"meta" and wich "data" (bulk) socket, we adjust both sockets.
Previously, DRBD only adjusted the bufsizes of the "data" socket.

Thanks again to Eduard.Guzovsky@stratus.com.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: improve network latency, TCP_QUICKACK

2010-06-01T09:12:27+00:00

On Thu, Apr 29, 2010 at 04:00:50PM -0400, Eduard.Guzovsky@stratus.com
 wrote on drbd-dev@lists.linbit.com
 Subject: [Drbd-dev] DRBD small synchronous writes performance improvements

> 1. TCP_QUICKACK option is set incorrectly. The goal was force TCP to
> send and ACK as a  "one time" event.  Instead the code permanently sets
> connection in the QUICKACK mode.

He is right, we actually want to use an even val with TCP_QUICKACK.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe

drbd: Revert "drbd: Create new current UUID as late as possible"

2010-06-01T09:12:26+00:00

The late-UUID writing is delayed until the next release.

Signed-off-by: Philipp Reisner 
Signed-off-by: Lars Ellenberg 
Signed-off-by: Jens Axboe