Hi all,
This is a proof-of-concept of a block reservation allocation model
between XFS and dm-thin. The purpose is to create a mechanism by which
the filesystem can determine an underlying thin volume is out of space
and opt to return -ENOSPC to userspace rather than waiting until the
volume is dr
Hi Linus,
The following changes since commit 4328daa2e79ed904a42ce00a9f38b9c36b44b21a:
dm: fix dm_rq_target_io leak on faults with .request_fn DM w/ blk-mq paths
(2016-02-21 20:27:50 -0500)
are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/
This is a hacky interface added to support the filesystem experiment for
reserving blocks out of thin pools. The XFS POC uses this method
provision space in the pool when blocks are allocated. The number of
actual blocks allocated is returned, which is used to update the
filesystem accounting.
Sig
uevents both let us attach additional info to the event, and also
allow interested user processes to receive them without having to
create and manage threads to sit in DEV_WAIT ioctls for each target
we're interested in.
dm-mpath already supports generating uevents. This patchset
generalizes this
On Tue, Mar 15, 2016 at 10:59:15AM +, Thanos Makatos wrote:
> On 15 March 2016 at 01:45, Eric Wheeler wrote:
> > Hi Joe,
> >
> > Please review the patch below when you have a moment. I am interested in
> > your feedback, and also interested in having this functionality merged
> > upstream. T
Document the current dm uevent API, as modified by this patchset.
Signed-off-by: Andy Grover
---
Documentation/device-mapper/dm-uevent.txt | 48 ++-
1 file changed, 41 insertions(+), 7 deletions(-)
diff --git a/Documentation/device-mapper/dm-uevent.txt
b/Documentati
2016-03-18 6:45 GMT+08:00 Eric Wheeler :
>
> On Thu, 17 Mar 2016, Joe Thornber wrote:
>
>> If you're skipping the mappings does the new thin_ls provide enough
>> information for you?
>
> Our primary need is the ability to specify --device-id in the o:L:d output
> format (old_format.cc) with mapping
From: Mike Snitzer
Signed-off-by: Mike Snitzer
---
drivers/md/dm.c | 75 +++
include/linux/device-mapper.h | 5 +++
2 files changed, 80 insertions(+)
diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index be49057..4da5d3e 100644
--- a/driver
Adopt a reservation-based block allocation model when XFS runs on top of
a dm-thin device with accompanying support. As of today, the filesystem
has no indication of available space in the underlying device. If the
thin pool is depleted, the filesystem has no recourse but to handle the
read-only st
Signed-off-by: Brian Foster
---
drivers/md/dm.c | 35 +++
include/linux/device-mapper.h | 2 ++
2 files changed, 37 insertions(+)
diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 4da5d3e..e1aec28 100644
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.
The dm-thin set_reserve_count() function is designed to control the
current block reservation for a thin pool. It currently only provides
the ability to increase the reservation.
Clients such as XFS will rely on over-reservation and thus require the
ability to release excess reservation back to th
2016-03-18 6:45 GMT+08:00 Eric Wheeler :
>
> On Thu, 17 Mar 2016, Joe Thornber wrote:
>
>> If you're skipping the mappings does the new thin_ls provide enough
>> information for you?
>
> Our primary need is the ability to specify --device-id in the o:L:d output
> format (old_format.cc) with mapping
Add block device reservation infrastructure to XFS. This primarily
consists of wrappers around the associated block device functions. This
mechanism provides the ability to reserve, release and provision a set
of blocks in the underlying block device.
The mechanism enables the filesystem to adopt
From: Joe Thornber
Experimental reserve interface for XFS guys to play with.
I have big reservations (no pun intended) about this patch.
Not-Signed-off-by: Joe Thornber
Not-Signed-off-by: Mike Snitzer
---
drivers/md/dm-thin.c | 142 +++
1 file
Signed-off-by: Brian Foster
---
fs/block_dev.c | 10 ++
include/linux/blkdev.h | 2 ++
2 files changed, 12 insertions(+)
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 375a2e4..73e57b9 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -517,6 +517,16 @@ int blk_get_reserved
Since dm-uevent.c (if CONFIG_DM_UEVENT) is part of the same module as where
dm_sent_uevents is called, it does not need to be exported.
Signed-off-by: Andy Grover
---
drivers/md/dm-uevent.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/drivers/md/dm-uevent.c b/drivers/md/dm-uevent.c
index 8
Dne 18.3.2016 v 12:06 Werner Koch napsal(a):
On Fri, 26 Feb 2016 12:42, w...@gnupg.org said:
I am playing with a new crypto container format and propose to enhance
"dmsetup message" to accept the actual message from stdin instead of
taking it only from the command line. This is useful to set a
On Wed, 16 Mar 2016, Ming-Hung Tsai wrote:
> 2016-03-16 1:51 GMT+08:00 Eric Wheeler :
> > On Tue, 15 Mar 2016, Thanos Makatos wrote:
> >
> >> On 15 March 2016 at 01:45, Eric Wheeler
> >> wrote:
> >> > Hi Joe,
> >> >
> >> > We use thin_dump on live dm-thin metadata snapshots all the time. In our
On Thu, 17 Mar 2016, Joe Thornber wrote:
> If you're skipping the mappings does the new thin_ls provide enough
> information for you?
Our primary need is the ability to specify --device-id in the o:L:d output
format (old_format.cc) with mappings. Our nightly dumps now take 50
seconds instead
On Fri, 26 Feb 2016 12:42, w...@gnupg.org said:
> I am playing with a new crypto container format and propose to enhance
> "dmsetup message" to accept the actual message from stdin instead of
> taking it only from the command line. This is useful to set a key and
Is there anything I can do to he
If you're skipping the mappings does the new thin_ls provide enough
information for you?
- Joe
On Tue, Mar 15, 2016 at 01:45:15AM +, Eric Wheeler wrote:
> Hi Joe,
>
> Please review the patch below when you have a moment. I am interested in
> your feedback, and also interested in having thi
21 matches
Mail list logo