Deprecated mandatory file locking

Deprecated mandatory file locking

[Jan Kara]

Hello Jeff,

we've got a report from user
(https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
entry (for root filesystem!) using 'mand' mount option stopped working.
Now I understand your rationale in 9e8925b67a "locks: Allow disabling
mandatory locking at compile time" but I guess there's some work to do wrt
documentation. At least mount(8) manpage could mention that mandatory
locking is broken and may be disabled referencing the rationale in fcntl
manpage? Or the kernel could mention something in the log about failing
mount because of 'mand' mount option?  What do you think? Because it took
me some code searching to understand why the mount is actually failing
which we can hardly expect from a normal sysadmin...

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: Deprecated mandatory file locking

[Jan Kara]

Resending to proper Jeff's address...

On Wed 14-08-19 19:33:45, Jan Kara wrote:
> Hello Jeff,
> 
> we've got a report from user
> (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> entry (for root filesystem!) using 'mand' mount option stopped working.
> Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> mandatory locking at compile time" but I guess there's some work to do wrt
> documentation. At least mount(8) manpage could mention that mandatory
> locking is broken and may be disabled referencing the rationale in fcntl
> manpage? Or the kernel could mention something in the log about failing
> mount because of 'mand' mount option?  What do you think? Because it took
> me some code searching to understand why the mount is actually failing
> which we can hardly expect from a normal sysadmin...
> 
> 								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: Deprecated mandatory file locking

[Jeff Layton]

On Wed, 2019-08-14 at 19:46 +0200, Jan Kara wrote:
> Resending to proper Jeff's address...
> 
> On Wed 14-08-19 19:33:45, Jan Kara wrote:
> > Hello Jeff,
> > 
> > we've got a report from user
> > (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> > entry (for root filesystem!) using 'mand' mount option stopped working.
> > Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> > mandatory locking at compile time" but I guess there's some work to do wrt
> > documentation. At least mount(8) manpage could mention that mandatory
> > locking is broken and may be disabled referencing the rationale in fcntl
> > manpage? Or the kernel could mention something in the log about failing
> > mount because of 'mand' mount option?  What do you think? Because it took
> > me some code searching to understand why the mount is actually failing
> > which we can hardly expect from a normal sysadmin...
> > 
> > 								Honza

Wow, I think this is the first actual user fallout we've ever had from
that change! Why was he setting that option? Does he actually use
mandatory locking?

I think a pr_notice() or pr_warn() at mount time when someone tries to
use it sounds like a very reasonable thing to do. Perhaps we can just
stick one in may_mandlock()?

I'll draft up a patch, and also update
Documentation/filesystems/mandatory-locking.txt with the current
situation.

Thanks,
-- 
Jeff Layton <jlayton@kernel.org>

Re: Deprecated mandatory file locking

[Jan Kara]

On Thu 15-08-19 15:18:45, Jeff Layton wrote:
> On Wed, 2019-08-14 at 19:46 +0200, Jan Kara wrote:
> > Resending to proper Jeff's address...
> > 
> > On Wed 14-08-19 19:33:45, Jan Kara wrote:
> > > Hello Jeff,
> > > 
> > > we've got a report from user
> > > (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> > > entry (for root filesystem!) using 'mand' mount option stopped working.
> > > Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> > > mandatory locking at compile time" but I guess there's some work to do wrt
> > > documentation. At least mount(8) manpage could mention that mandatory
> > > locking is broken and may be disabled referencing the rationale in fcntl
> > > manpage? Or the kernel could mention something in the log about failing
> > > mount because of 'mand' mount option?  What do you think? Because it took
> > > me some code searching to understand why the mount is actually failing
> > > which we can hardly expect from a normal sysadmin...
> > > 
> > > 								Honza
> 
> Wow, I think this is the first actual user fallout we've ever had from
> that change! Why was he setting that option? Does he actually use
> mandatory locking?

Yeah, reportedly they had an application that required mandatory locking.
But they don't use it anymore so they just removed the mount option.

> I think a pr_notice() or pr_warn() at mount time when someone tries to
> use it sounds like a very reasonable thing to do. Perhaps we can just
> stick one in may_mandlock()?

Yeah, sounds reasonable to me.

> I'll draft up a patch, and also update
> Documentation/filesystems/mandatory-locking.txt with the current
> situation.

Thanks!

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: Deprecated mandatory file locking

[Jan Kara]

Hi Jeff,

On Fri 16-08-19 17:31:49, Jan Kara wrote:
> On Thu 15-08-19 15:18:45, Jeff Layton wrote:
> > On Wed, 2019-08-14 at 19:46 +0200, Jan Kara wrote:
> > > Resending to proper Jeff's address...
> > > 
> > > On Wed 14-08-19 19:33:45, Jan Kara wrote:
> > > > Hello Jeff,
> > > > 
> > > > we've got a report from user
> > > > (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> > > > entry (for root filesystem!) using 'mand' mount option stopped working.
> > > > Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> > > > mandatory locking at compile time" but I guess there's some work to do wrt
> > > > documentation. At least mount(8) manpage could mention that mandatory
> > > > locking is broken and may be disabled referencing the rationale in fcntl
> > > > manpage? Or the kernel could mention something in the log about failing
> > > > mount because of 'mand' mount option?  What do you think? Because it took
> > > > me some code searching to understand why the mount is actually failing
> > > > which we can hardly expect from a normal sysadmin...
> > > > 
> > > > 								Honza
> > 
> > Wow, I think this is the first actual user fallout we've ever had from
> > that change! Why was he setting that option? Does he actually use
> > mandatory locking?
> 
> Yeah, reportedly they had an application that required mandatory locking.
> But they don't use it anymore so they just removed the mount option.
> 
> > I think a pr_notice() or pr_warn() at mount time when someone tries to
> > use it sounds like a very reasonable thing to do. Perhaps we can just
> > stick one in may_mandlock()?
> 
> Yeah, sounds reasonable to me.
> 
> > I'll draft up a patch, and also update
> > Documentation/filesystems/mandatory-locking.txt with the current
> > situation.
> 
> Thanks!

Did you ever get to this?

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: Deprecated mandatory file locking

[Jeff Layton]

On Thu, 2019-11-07 at 17:15 +0100, Jan Kara wrote:
> Hi Jeff,
> 
> On Fri 16-08-19 17:31:49, Jan Kara wrote:
> > On Thu 15-08-19 15:18:45, Jeff Layton wrote:
> > > On Wed, 2019-08-14 at 19:46 +0200, Jan Kara wrote:
> > > > Resending to proper Jeff's address...
> > > > 
> > > > On Wed 14-08-19 19:33:45, Jan Kara wrote:
> > > > > Hello Jeff,
> > > > > 
> > > > > we've got a report from user
> > > > > (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> > > > > entry (for root filesystem!) using 'mand' mount option stopped working.
> > > > > Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> > > > > mandatory locking at compile time" but I guess there's some work to do wrt
> > > > > documentation. At least mount(8) manpage could mention that mandatory
> > > > > locking is broken and may be disabled referencing the rationale in fcntl
> > > > > manpage? Or the kernel could mention something in the log about failing
> > > > > mount because of 'mand' mount option?  What do you think? Because it took
> > > > > me some code searching to understand why the mount is actually failing
> > > > > which we can hardly expect from a normal sysadmin...
> > > > > 
> > > > > 								Honza
> > > 
> > > Wow, I think this is the first actual user fallout we've ever had from
> > > that change! Why was he setting that option? Does he actually use
> > > mandatory locking?
> > 
> > Yeah, reportedly they had an application that required mandatory locking.
> > But they don't use it anymore so they just removed the mount option.
> > 
> > > I think a pr_notice() or pr_warn() at mount time when someone tries to
> > > use it sounds like a very reasonable thing to do. Perhaps we can just
> > > stick one in may_mandlock()?
> > 
> > Yeah, sounds reasonable to me.
> > 
> > > I'll draft up a patch, and also update
> > > Documentation/filesystems/mandatory-locking.txt with the current
> > > situation.
> > 
> > Thanks!
> 
> Did you ever get to this?
> 
> 								Honza

Yes. It went into v5.4-rc1. You even reviewed it! ;)

See:

commit df2474a22c42ce419b67067c52d71da06c385501
Author: Jeff Layton <jlayton@kernel.org>
Date:   Thu Aug 15 15:21:17 2019 -0400

    locks: print a warning when mount fails due to lack of "mand" support

Cheers,
-- 
Jeff Layton <jlayton@kernel.org>

Re: Deprecated mandatory file locking

[Jan Kara]

On Thu 07-11-19 13:34:21, Jeff Layton wrote:
> On Thu, 2019-11-07 at 17:15 +0100, Jan Kara wrote:
> > Hi Jeff,
> > 
> > On Fri 16-08-19 17:31:49, Jan Kara wrote:
> > > On Thu 15-08-19 15:18:45, Jeff Layton wrote:
> > > > On Wed, 2019-08-14 at 19:46 +0200, Jan Kara wrote:
> > > > > Resending to proper Jeff's address...
> > > > > 
> > > > > On Wed 14-08-19 19:33:45, Jan Kara wrote:
> > > > > > Hello Jeff,
> > > > > > 
> > > > > > we've got a report from user
> > > > > > (https://bugzilla.suse.com/show_bug.cgi?id=1145007) wondering why his fstab
> > > > > > entry (for root filesystem!) using 'mand' mount option stopped working.
> > > > > > Now I understand your rationale in 9e8925b67a "locks: Allow disabling
> > > > > > mandatory locking at compile time" but I guess there's some work to do wrt
> > > > > > documentation. At least mount(8) manpage could mention that mandatory
> > > > > > locking is broken and may be disabled referencing the rationale in fcntl
> > > > > > manpage? Or the kernel could mention something in the log about failing
> > > > > > mount because of 'mand' mount option?  What do you think? Because it took
> > > > > > me some code searching to understand why the mount is actually failing
> > > > > > which we can hardly expect from a normal sysadmin...
> > > > > > 
> > > > > > 								Honza
> > > > 
> > > > Wow, I think this is the first actual user fallout we've ever had from
> > > > that change! Why was he setting that option? Does he actually use
> > > > mandatory locking?
> > > 
> > > Yeah, reportedly they had an application that required mandatory locking.
> > > But they don't use it anymore so they just removed the mount option.
> > > 
> > > > I think a pr_notice() or pr_warn() at mount time when someone tries to
> > > > use it sounds like a very reasonable thing to do. Perhaps we can just
> > > > stick one in may_mandlock()?
> > > 
> > > Yeah, sounds reasonable to me.
> > > 
> > > > I'll draft up a patch, and also update
> > > > Documentation/filesystems/mandatory-locking.txt with the current
> > > > situation.
> > > 
> > > Thanks!
> > 
> > Did you ever get to this?
> > 
> > 								Honza
> 
> Yes. It went into v5.4-rc1. You even reviewed it! ;)

Bah ;) I completely forgot and somehow didn't realize the section 7. in
Documentation/filesystems/mandatory-locking.txt is describing what I
wanted when I looked at it yesterday. Sorry for the noise!

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR