Re: [linux-lvm] pvscan: /dev/sdc: open failed: No medium found

[David Teigland <teigland@redhat.com>]

On Wed, Apr 24, 2019 at 10:37:26AM +0200, Zdenek Kabelac wrote:
> Dne 23. 04. 19 v 16:27 David Teigland napsal(a):
> > On Mon, Apr 22, 2019 at 11:15:53PM -0600, Gang He wrote:
> > > Hello List,
> > > 
> > > One user complained this error message.
> > > The user has a usb sd card reader with no media present.  When they issue a pvscan under lvm2-2.02.180 the device is opened which results in 'No medium found' being reported.
> > > But lvm2-2.02.120 did not do this (the device appears to get filtered out earlier). The customer views the 'No medium found' message as an issue/bug.
> > > Any suggest/comments for this error message?
> > > 
> > > The detailed information is as below,
> > > lvm2 2.02.180-9.4.2
> > > OS: SLES12 SP4
> > > Kernel 4.12.14-95.3-default
> > > Hardware: HP ProLiant DL380 Gen10
> > > 
> > > After upgrade from sles12SP3 to SP4, customer is reporting the following error message:
> > > 
> > >   # pvscan
> > >   /dev/sdc: open failed: No medium found
> > >   PV /dev/sdb   VG Q11vg10         lvm2 [5.24 TiB / 2.00 TiB free]
> > >   Total: 1 [5.24 TiB] / in use: 1 [5.24 TiB] / in no VG: 0 [0   ]
> > 
> > I've heard this a few times now, I guess we should drop it, it's probably
> > more trouble than help.
> > 
> 
> This hotfix patch:
> 
> https://www.redhat.com/archives/lvm-devel/2019-April/msg00067.html
> 
> is unfortunately faulty and should be reverted.
> 
> The core problem is - the filtering layer is somehow bypassed where cdrom
> used be recognized as unsuitable device for opening.
>
> So the error reported by lvm2 is really an error where lvm2 should not be
> trying to even open such device.

I don't believe opening is the problem; most filters need the device open.
I suspect the difference is that filter-usable once used
dev_open_readonly_quiet itself, which suppressed the warning, to use with
dev_get_size.  Now the filter relies on dev_get_size to open the dev if
necessary (it's often already open), and that does not use the quiet flag.

This might be an effective alternative to suppressing all open warnings
(I can't say for certain in this case until seeing which open triggered
the warning.)

diff --git a/lib/device/dev-io.c b/lib/device/dev-io.c
index 3fe264755443..bd6cbcb04a1b 100644
--- a/lib/device/dev-io.c
+++ b/lib/device/dev-io.c
@@ -165,7 +165,7 @@ static int _dev_get_size_dev(struct device *dev, uint64_t *size)
        }
 
        if (fd <= 0) {
-               if (!dev_open_readonly(dev))
+               if (!dev_open_readonly_quiet(dev))
                        return_0;
                fd = dev_fd(dev);
                do_close = 1;