Arnon Yaari (1):
libmultipath: update INFINIDAT builtin config
Benjamin Marzinski (5):
libmultipath: pull functions into util.c
libmultipath: change reservation_key to a be64
libmpathpersist: fix update_prflag code
multipath: add alternate reservation_key method
Oops, the shortlog was truncated.
Here it is again:
Arnon Yaari (1):
libmultipath: update INFINIDAT builtin config
Benjamin Marzinski (5):
libmultipath: pull functions into util.c
libmultipath: change reservation_key to a be64
libmpathpersist: fix update_prflag code
Instead of retrying request allocation after a delay if a path is
busy, wait until the underlying path has completed a request. This
patch avoids that submission of requests to busy paths is delayed and
hence creates more opportunities for merging sequential I/O requests.
Reported-by: Ming Lei
This patch just pulls safe_write out of rbd. and the persistent
reservation key parsing code out of dict.c and puts them in util.c,
so that other functions can make use of them.
Signed-off-by: Benjamin Marzinski
---
libmultipath/checkers/rbd.c | 16 +---
There are multiple problems with the prflag code. First, it doesn't do
anything useful at all currently. update_prflags is called with "set"
and "unset" instead of "setprstatus" and "unsetprstatus", so it doesn't
actually enable persistent reservation tracking in multipathd when a key
is
The reservation key is currently being stored as any array of 8 unsigned
chars. This is exactly the same in-memory representation as a big
endian 64 bit integer. However, the code for dealing with a big endian
64 bit integer is much simpler, so switch to use that instead. Instead
of directly
Update the man pages to list the new configuration options and
multipathd commands.
Signed-off-by: Benjamin Marzinski
---
multipath/multipath.conf.5 | 17 +
multipathd/multipathd.8| 16
2 files changed, 33 insertions(+)
diff --git
The scsi persistent reservation API doesn't force devices to implement
any method to display the mapping from a reservation key to an I_T Nexus
(the READ_FULL_STATUS action is an optional later addition, and a number
of devices don't support it). To allow multipathd to determine the
correct
On Wed, 2017-09-20 at 14:56 -0400, Mike Snitzer wrote:
> On Wed, Sep 20 2017 at 2:12pm -0400,
> Bart Van Assche wrote:
> > bdev = pgpath->path.dev->bdev;
> > q = bdev_get_queue(bdev);
> > - clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC);
> >
On Wed, Sep 20 2017 at 2:12pm -0400,
Bart Van Assche wrote:
> Instead of retrying request allocation after a delay if a path is
> busy, wait until the underlying path has completed a request. This
> patch avoids that submission of requests to busy paths is delayed and
>
On Wed, 2017-09-20 at 13:32 +0200, Martin Wilck wrote:
> On Fri, 2017-09-15 at 08:15 +0200, Hannes Reinecke wrote:
> >
> > Signed-off-by: Hannes Reinecke
> > ---
> > libmultipath/checkers.c | 4 ++--
> > libmultipath/print.c| 2 +-
> > 2 files changed, 3 insertions(+), 3
When the reservation_key is set to "file", the reservation key needs to
be added to the prkeys file for multipathd to be able to register new
paths as they are added to the device. This patch adds support to the
mpathpersist command to message multipathd when a key is registered or
unregistered
The scsi persistent reservation API doesn't force devices to implement any
method to display the mapping from a reservation key to an I_T Nexus (the
READ_FULL_STATUS action is an optional later addition, and a number of devices
don't support it). To allow multipathd to determine the correct
do_sd_notify contains a call to sd_notify which is unknown if USE_SYSTEMD is
undefined. In this case, do_sd_notify is never called anyway, so embed the
entire function into an #ifdef USE_SYSTEMD.
This fixes a regression introduced in 88ddca5.
Signed-off-by: Michael Lass
---
On Wed, Sep 20, 2017 at 11:26:09PM +, Bart Van Assche wrote:
> On Thu, 2017-09-21 at 06:36 +0800, Ming Lei wrote:
> > Actually with GFP_ATOMIC, dispatch in dm-rq can't move on and no request
> > will be dequeued from IO scheduler queue if this allocation fails, that
> > means IO merge is still
On Fri, 2017-09-15 at 08:15 +0200, Hannes Reinecke wrote:
> One should remember that '0' _is_ a valid fd, so we need to set
> the fd to '-1' upon allocating the path structure and ensure we're
> checking for a value _smaller_ than 0 to detect an invalid fd.
>
> > diff --git
On Thu, 14 Sep 2017, Brassow Jonathan wrote:
> Mikulas,
>
> Do you have any tests for the new dm event code you added a while back?
> (https://bugzilla.redhat.com/show_bug.cgi?id=1475380#c3) I?d like to
> give QA instructions for how to perform regression testing on it so we
> can get
On Fri, 2017-09-15 at 08:14 +0200, Hannes Reinecke wrote:
> The udev monitor from uevent_listen() needs to be cleaned up from
> a pthread cleanup hook.
> Valgrind complained here.
>
> Signed-off-by: Hannes Reinecke
Reviewed-by: Martin Wilck
--
Dr. Martin Wilck
Hi Guan,
>>>Shall we use existing PATH_SHAKY ?
As the path_shaky Indicates path not available for "normal" operations we can
use this state. That's a good idea.
Regarding the marginal paths below is my explanation. And brocade is publishing
couple of white papers regarding the same to educate
19 matches
Mail list logo