lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Tue, 10 Oct 2023 15:42:39 -0700
From:   Sarthak Kukreti <>
To:     Dave Chinner <>
Cc:,,,,, Jens Axboe <>,
        Alasdair Kergon <>,
        Mike Snitzer <>,
        Christoph Hellwig <>,
        Brian Foster <>,
        "Theodore Ts'o" <>,
        Andreas Dilger <>,
        Bart Van Assche <>,
        "Darrick J. Wong" <>
Subject: Re: [PATCH v8 5/5] block: Pass unshare intent via REQ_OP_PROVISION

On Sun, Oct 8, 2023 at 4:27 PM Dave Chinner <> wrote:
> On Fri, Oct 06, 2023 at 06:28:17PM -0700, Sarthak Kukreti wrote:
> > Allow REQ_OP_PROVISION to pass in an extra REQ_UNSHARE bit to
> > annotate unshare requests to underlying layers. Layers that support
> > FALLOC_FL_UNSHARE will be able to use this as an indicator of which
> > fallocate() mode to use.
> >
> > Suggested-by: Darrick J. Wong <>
> > Signed-off-by: Sarthak Kukreti <>
> > ---
> >  block/blk-lib.c           |  6 +++++-
> >  block/fops.c              |  6 ++++--
> >  drivers/block/loop.c      | 35 +++++++++++++++++++++++++++++------
> >  include/linux/blk_types.h |  3 +++
> >  include/linux/blkdev.h    |  3 ++-
> >  5 files changed, 43 insertions(+), 10 deletions(-)
> I have no idea how filesystems (or even userspace applications, for
> that matter) are supposed to use this - they have no idea if the
> underlying block device has shared blocks for LBA ranges it already
> has allocated and provisioned. IOWs, I don't know waht the semantics
> of this function is, it is not documented anywhere, and there is no
> use case present that tells me how it might get used.
> Yes, unshare at the file level means the filesystem tries to break
> internal data extent sharing, but if the block layers or backing
> devices are doing deduplication and sharing unknown to the
> application or filesystem, how do they ever know that this operation
> might need to be performed? In what cases do we need to be able to
> unshare block device ranges, and how is that different to the
> guarantees that REQ_PROVISION is already supposed to give for
> provisioned ranges that are then subsequently shared by the block
> device (e.g. by snapshots)?
> Also, from an API perspective, this is an "unshare" data operation,
> not a "provision" operation. Hence I'd suggest that the API should
> be blkdev_issue_unshare() rather than optional behaviour to
> _provision() which - before this patch - had clear and well defined
> meaning....
Fair points, the intent from the conversation with Darrick was the
addition of support for FALLOC_FL_UNSHARE_RANGE in patch 2 of v4
(originally suggested by Brian Forster in [1]): if we allow
fallocate(UNSHARE_RANGE) on a loop device (ex. for creating a
snapshot, similar in nature to the FICLONE example you mentioned on
the loop patch), we'd (ideally) want to pass it through to the
underlying layers and let them figure out what to do with it. But it
is only for situations where we are explicitly know what the
underlying layers are and what's the mecha

I agree though that it clouds the API a bit and I don't think it
necessarily needs to be a part of the initial patch series: for now, I
propose keeping just mode zero (and FALLOC_FL_KEEP_SIZE) handling in
the block series patch and drop this patch for now. WDYT?



> Cheers,
> Dave.
> --
> Dave Chinner

Powered by blists - more mailing lists