drbd-0.7.22.tar.gz

Hi,

While working on drbd-8.0 and rewriting a lot of stuff there, we noticed a bug in the previous handling of protocol A and B.

In case you are using protocol A or B you should upgrade to this version !

0.7.22 (api:79/proto:74)
-----
* Fixed bugs in the implementation of protocol A and B. we noticed that in protocol A and B, on connection loss, we could "forget" to set certain areas out of sync. so resync in those protocol has been broken all along!
* Fix races between failure in drbd_send_dblock
and concurrently running tl_clear
* Fix potential access-afer-free in drbd_dio_end
* Fix possible list corruption respective resulting deadlock in receive_DataRequest
* Improved the drbd_thread_stop / start code
* Gracefull removal of devfs and other updates.
* Implemented the "freeze_io" option for "on-disconnect"

http://oss.linbit.com/drbd/0.7/drbd-0.7.22.tar.gz
http://www.linbit.com/support/drbd-0.7.22/
http://svn.drbd.org/drbd/tags/drbd-0.7.22/

-Phil

: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Schönbrunnerstr 244, 1120 Vienna, Austria    http://www.linbit.com :

Am Dienstag, 31. Oktober 2006 05:00 schrieb Jacobson:

If I remeber correctly, pages that are controlled by slab have only a use count of one (Slab holds that single reference) [maybe it was zero, because slab does not use the reference counting on pages]. When zero copy send is done, the network code simply decreases the reference count to that page by one.

I think it was the slab code that does not like it if someone modifies the reference counts of the pages with are owned by slab.

PS: As noted in the code, the only FS at that time, was XFS, which
was using such pages...

-Phil

: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Schönbrunnerstr 244, 1120 Vienna, Austria    http://www.linbit.com :