md device is read only mode

md device is read only mode

[Piergiorgio Sartor]

Hi all,

following the comment of Neil about the fact that
an md device, in r/o mode, will BUG() in case of
write attempt, I would like to ask for some more
information on the topic.

The problem is the following, mounting a filesystem
in r/o does not, correctly, prevent writes to the
underlying device.

Now, assuming a filesystem is mounted r/w over an
md device, being r/w too, it is pretty clear that
switching the md device to r/o will create problems.

What if the md device is started in r/o mode and
than, later, the filesystem is mounted *without*
specifying r/o mode?

The filesystem might try to write to the device,
but it will not work, of course.
Does this will cause a BUG() in md?

Because using a r/o memory card, there are no
problems, the mount recognize the device is r/o
and it sets the filesystems r/o too.
I do not know if it anyway tries to write to the
r/o device, ignoring the errors reported.

I would say that, if md BUG() in any case, it
could be a bug... :-)

Any comments?

Thanks,

bye,

-- 

piergiorgio

Re: md device is read only mode

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 1911 bytes --]

On Thu, 5 Jul 2012 19:27:48 +0200 Piergiorgio Sartor
<piergiorgio.sartor@nexgo.de> wrote:

> Hi all,
> 
> following the comment of Neil about the fact that
> an md device, in r/o mode, will BUG() in case of
> write attempt, I would like to ask for some more
> information on the topic.
> 
> The problem is the following, mounting a filesystem
> in r/o does not, correctly, prevent writes to the
> underlying device.

This is only half the story.
If the mount command requests read-only, the filesystem may still write to
the device.
However if the device is marked read-only then the filesystem should not
write to the disk - if it finds that it needs to, it should fail the
mount request.
I believe that filesystems behave correctly in this manner, though there
could still be bugs lurking.

> 
> Now, assuming a filesystem is mounted r/w over an
> md device, being r/w too, it is pretty clear that
> switching the md device to r/o will create problems.

Not quite.  When the filesystem is mounted, md will refuse to switch the
device to read-only, so that scenario is not possible.

> 
> What if the md device is started in r/o mode and
> than, later, the filesystem is mounted *without*
> specifying r/o mode?

The mount system call will notice that the device is read-only, and will make
assume the read-only mount option.

> 
> The filesystem might try to write to the device,
> but it will not work, of course.
> Does this will cause a BUG() in md?

Only if there are bugs.

> 
> Because using a r/o memory card, there are no
> problems, the mount recognize the device is r/o
> and it sets the filesystems r/o too.
> I do not know if it anyway tries to write to the
> r/o device, ignoring the errors reported.
> 
> I would say that, if md BUG() in any case, it
> could be a bug... :-)
> 
> Any comments?
> 
> Thanks,
> 
> bye,
> 

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: md device is read only mode

[Piergiorgio Sartor]

Hi Neil,

please see one comment below.

On Mon, Jul 09, 2012 at 11:40:22AM +1000, NeilBrown wrote:
[...]
> The mount system call will notice that the device is read-only, and will make
> assume the read-only mount option.

OK, so it should work as expected.

The story was this.
After fixing the RAID-5 superblock problem, we
wanted to see if the filesystem was still OK.
So we started the RAID in r/o mode (so to avoid
any possible further damage from mount) and then
mounted the filesystem (actually, now that I
think of it, there was an LVM layer in the middle).
This mount caused a kernel error report (I do
not know if BUG() or else).

Is there any explanation for that?

Thanks,

bye,

-- 

piergiorgio

Re: md device is read only mode

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 1010 bytes --]

On Mon, 9 Jul 2012 19:34:07 +0200 Piergiorgio Sartor
<piergiorgio.sartor@nexgo.de> wrote:

> Hi Neil,
> 
> please see one comment below.
> 
> On Mon, Jul 09, 2012 at 11:40:22AM +1000, NeilBrown wrote:
> [...]
> > The mount system call will notice that the device is read-only, and will make
> > assume the read-only mount option.
> 
> OK, so it should work as expected.
> 
> The story was this.
> After fixing the RAID-5 superblock problem, we
> wanted to see if the filesystem was still OK.
> So we started the RAID in r/o mode (so to avoid
> any possible further damage from mount) and then
> mounted the filesystem (actually, now that I
> think of it, there was an LVM layer in the middle).
> This mount caused a kernel error report (I do
> not know if BUG() or else).
> 
> Is there any explanation for that?

Probably LVM isn't detecting that the underlying device is readonly and so
forcing the upper device to be readonly.
md doesn't do this either - it should.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]