mount nofail: what failures should we allow ?

mount nofail: what failures should we allow ?

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 930 bytes --]

i've received two requests for the "nofail" option.  the doc for the
option is a bit ... terse ... so it's hard to guess at the overall
intention.

(1) ignore ENOMEDIUM like:
sys-utils/mount.c could be updated to do:
    case ENOMEDIUM:
        if (uflags & MNT_MS_NOFAIL)
            return MOUNT_EX_SUCCESS;
        warnx(_("no medium found on %s"), src);
        break;
this is for cases like cd/dvd drives which happens to have no disk loaded.

(2) ignore unknown fs types.  e.g. when a kernel config/module is missing
support for the requested filesystem type.  so a fstab entry like:
	..src..  /mnt/foo  somefs defaults,nofail
rather than error out with:
	mount: unknown filesystem type 'somefs'
it would just issue a warning like it does for other nofail options.

if we go this route though, how far should we take it ?  should we make
all failures when nofail is active the same ?  i'm kind of leaning that
way ...
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: mount nofail: what failures should we allow ?

[Karel Zak]

On Tue, Jan 19, 2016 at 06:24:58PM -0500, Mike Frysinger wrote:
> i've received two requests for the "nofail" option.  the doc for the
> option is a bit ... terse ... so it's hard to guess at the overall
> intention.

man mount:
   nofail Do not report errors for this device if it does not exist.


from my point of view this description is pretty explicit :-)

> (1) ignore ENOMEDIUM like:
> sys-utils/mount.c could be updated to do:
>     case ENOMEDIUM:
>         if (uflags & MNT_MS_NOFAIL)
>             return MOUNT_EX_SUCCESS;
>         warnx(_("no medium found on %s"), src);
>         break;
> this is for cases like cd/dvd drives which happens to have no disk loaded.

It seems like a good idea, send patch :-)

> (2) ignore unknown fs types.  e.g. when a kernel config/module is missing
> support for the requested filesystem type.  so a fstab entry like:
> 	..src..  /mnt/foo  somefs defaults,nofail
> rather than error out with:
> 	mount: unknown filesystem type 'somefs'
> it would just issue a warning like it does for other nofail options.

I'm not sure with this. It's unusual situation when any filesystem is unknown
for libblkid, but it's pretty common that kernel returns EINVAL. This
happen when a kernel config/module is missing, but also if you specify
wrong mount options and in some another situations. I don't think we
want to hide such problems. It's too generic...

    Karel


-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

Re: mount nofail: what failures should we allow ?

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 1870 bytes --]

On 20 Jan 2016 11:28, Karel Zak wrote:
> On Tue, Jan 19, 2016 at 06:24:58PM -0500, Mike Frysinger wrote:
> > i've received two requests for the "nofail" option.  the doc for the
> > option is a bit ... terse ... so it's hard to guess at the overall
> > intention.
> 
> man mount:
>    nofail Do not report errors for this device if it does not exist.
> 
> from my point of view this description is pretty explicit :-)

does it mean the device node doesn't exist (ENOENT) ?  or does it also
accept the node being there, but returning other errors like ENXIO (the
driver isn't loaded) or ENOTDIR (bad path) or ENOTBLK (used a bad path
like /dev/zero) or ENOMEDIUM (the node & hardware exists, but is not
loaded) ?  there's probably other errno values you could catch here.

surely you agree that "does not exist" does not cover all these cases.
or at the very least, it's pretty ambiguous/fuzzy.

> > (2) ignore unknown fs types.  e.g. when a kernel config/module is missing
> > support for the requested filesystem type.  so a fstab entry like:
> > 	..src..  /mnt/foo  somefs defaults,nofail
> > rather than error out with:
> > 	mount: unknown filesystem type 'somefs'
> > it would just issue a warning like it does for other nofail options.
> 
> I'm not sure with this. It's unusual situation when any filesystem is unknown
> for libblkid, but it's pretty common that kernel returns EINVAL. This
> happen when a kernel config/module is missing, but also if you specify
> wrong mount options and in some another situations. I don't think we
> want to hide such problems. It's too generic...

i think there's a lot going on in this response.  let me distill it a
bit.  if i have a reiserfs that is usable, but i forgot to enable or
load the reiserfs kernel driver, should nofail be allowed to skip ?
or is this a hard failure ?
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: mount nofail: what failures should we allow ?

[Karel Zak]

On Wed, Jan 20, 2016 at 03:00:28PM -0500, Mike Frysinger wrote:
> On 20 Jan 2016 11:28, Karel Zak wrote:
> > On Tue, Jan 19, 2016 at 06:24:58PM -0500, Mike Frysinger wrote:
> > > i've received two requests for the "nofail" option.  the doc for the
> > > option is a bit ... terse ... so it's hard to guess at the overall
> > > intention.
> > 
> > man mount:
> >    nofail Do not report errors for this device if it does not exist.
> > 
> > from my point of view this description is pretty explicit :-)
> 
> does it mean the device node doesn't exist (ENOENT) ?  or does it also
> accept the node being there, but returning other errors like ENXIO (the
> driver isn't loaded) or ENOTDIR (bad path) or ENOTBLK (used a bad path
> like /dev/zero) or ENOMEDIUM (the node & hardware exists, but is not
> loaded) ?  there's probably other errno values you could catch here.
> 
> surely you agree that "does not exist" does not cover all these cases.
> or at the very least, it's pretty ambiguous/fuzzy.

now we use "nofail" if:

 - impossible to translate UUID/LABEL/etc to device name
 - mount(2) returns ENOENT and stat() on source fails
 - mount(2) returns ENOTDIR and stat() on source fails with ENOTDIR
 - mount(2) returns ENOTBLK
 - mount(2) returns ENXIO
 - and ENOMEDIUM now

... well, maybe we can improve the description in the man page to make
it more user friendly.

> > > (2) ignore unknown fs types.  e.g. when a kernel config/module is missing
> > > support for the requested filesystem type.  so a fstab entry like:
> > > 	..src..  /mnt/foo  somefs defaults,nofail
> > > rather than error out with:
> > > 	mount: unknown filesystem type 'somefs'
> > > it would just issue a warning like it does for other nofail options.
> > 
> > I'm not sure with this. It's unusual situation when any filesystem is unknown
> > for libblkid, but it's pretty common that kernel returns EINVAL. This
> > happen when a kernel config/module is missing, but also if you specify
> > wrong mount options and in some another situations. I don't think we
> > want to hide such problems. It's too generic...
> 
> i think there's a lot going on in this response.  let me distill it a
> bit.  if i have a reiserfs that is usable, but i forgot to enable or
> load the reiserfs kernel driver, should nofail be allowed to skip ?
> or is this a hard failure ?

 Now it's hard failure and I'm fine with it ;)

 In this case kernel returns EINVAL -- the problem is that kernel uses
 EINVAL it in another situations too. It makes "nofail" very
 problematic for this use-case, because we're not able to detect that
 you have no reiserfs driver.

 Frankly, people will ask for "nofail" extension all time...

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

[PATCH] mount: allow nofail to silence ENOMEDIUM cases

[Mike Frysinger]

Allow people to use nofail to ignore empty cd/dvd drive errors.

URL: https://bugs.gentoo.org/559356
Reported-by: William Hubbs <williamh@gentoo.org>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
---
 sys-utils/mount.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/sys-utils/mount.c b/sys-utils/mount.c
index 73f9d0b..01eaccb 100644
--- a/sys-utils/mount.c
+++ b/sys-utils/mount.c
@@ -628,6 +628,9 @@ try_readonly:
 		break;
 
 	case ENOMEDIUM:
+		if (uflags & MNT_MS_NOFAIL)
+			return MOUNT_EX_SUCCESS;
+
 		warnx(_("no medium found on %s"), src);
 		break;
 
-- 
2.6.2

Re: [PATCH] mount: allow nofail to silence ENOMEDIUM cases

[Karel Zak]

On Wed, Jan 20, 2016 at 03:20:08PM -0500, Mike Frysinger wrote:
>  sys-utils/mount.c | 3 +++
>  1 file changed, 3 insertions(+)

Applied, thanks.

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com