bug#21142: Sshfs-fuse requires fuse

bug#21142: Sshfs-fuse requires fuse

[Andreas Enge]

Hello,

when trying to execute sshfs from the sshfs-fuse package, I obtain
   fuse: failed to exec fusermount: No such file or directory

I think that fuse should be a propagated input of sshfs-fuse.

Andreas

bug#21142: Sshfs-fuse requires fuse

[Claes Wallin]

[-- Attachment #1: Type: text/plain, Size: 358 bytes --]

On 27-Jul-2015 5:06 pm, "Andreas Enge" <andreas@enge.fr> wrote:
>
> Hello,
>
> when trying to execute sshfs from the sshfs-fuse package, I obtain
>    fuse: failed to exec fusermount: No such file or directory
>
> I think that fuse should be a propagated input of sshfs-fuse.

Maybe it's less intrusive toward the user to wrap sshfs to add fuse to its
PATH?

[-- Attachment #2: Type: text/html, Size: 508 bytes --]

bug#21142: Sshfs-fuse requires fuse

[Mark H Weaver]

Claes Wallin (韋嘉誠) <gnu@clacke.user.lysator.liu.se> writes:

> On 27-Jul-2015 5:06 pm, "Andreas Enge" <andreas@enge.fr> wrote:
>>
>> Hello,
>>
>> when trying to execute sshfs from the sshfs-fuse package, I obtain
>> fuse: failed to exec fusermount: No such file or directory
>>
>> I think that fuse should be a propagated input of sshfs-fuse.
>
> Maybe it's less intrusive toward the user to wrap sshfs to add fuse to
> its PATH?

Or patch the reference to 'fusermount' to be an absolute path?

I haven't looked into this, so I don't know the feasibility of these
approaches, but in general I think that 'propagated-inputs' should be
avoided whenever there is a reasonable alternative.

What do you think?

      Mark

bug#21142: Sshfs-fuse requires fuse

[Claes Wallin]

[-- Attachment #1: Type: text/plain, Size: 931 bytes --]

On Jul 28, 2015 4:21 PM, "Mark H Weaver" <mhw@netris.org> wrote:
> Claes Wallin (韋嘉誠) <gnu@clacke.user.lysator.liu.se> writes:
> > On 27-Jul-2015 5:06 pm, "Andreas Enge" <andreas@enge.fr> wrote:
> >>
> >> Hello,
> >>
> >> when trying to execute sshfs from the sshfs-fuse package, I obtain
> >> fuse: failed to exec fusermount: No such file or directory
> >>
> >> I think that fuse should be a propagated input of sshfs-fuse.
> >
> > Maybe it's less intrusive toward the user to wrap sshfs to add fuse to
> > its PATH?
>
> Or patch the reference to 'fusermount' to be an absolute path?

That's even better!

> I haven't looked into this, so I don't know the feasibility of these
> approaches, but in general I think that 'propagated-inputs' should be
> avoided whenever there is a reasonable alternative.
>
> What do you think?

Yeah, installing a package should cause the minimum amount of surprise.

[-- Attachment #2: Type: text/html, Size: 1375 bytes --]

bug#21142: Sshfs-fuse requires fuse

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> Claes Wallin (韋嘉誠) <gnu@clacke.user.lysator.liu.se> writes:
>
>> On 27-Jul-2015 5:06 pm, "Andreas Enge" <andreas@enge.fr> wrote:
>>>
>>> Hello,
>>>
>>> when trying to execute sshfs from the sshfs-fuse package, I obtain
>>> fuse: failed to exec fusermount: No such file or directory
>>>
>>> I think that fuse should be a propagated input of sshfs-fuse.
>>
>> Maybe it's less intrusive toward the user to wrap sshfs to add fuse to
>> its PATH?
>
> Or patch the reference to 'fusermount' to be an absolute path?

This what libfuse does by default in lib/mount.c:

--8<---------------cut here---------------start------------->8---
static void exec_fusermount(const char *argv[])
{
	execv(FUSERMOUNT_DIR "/" FUSERMOUNT_PROG, (char **) argv);
	execvp(FUSERMOUNT_PROG, (char **) argv);
}
--8<---------------cut here---------------end--------------->8---

However, we explicitly patch it because in practice the ‘fusermount’
binary must be at least set-user-gid to fuse, per mount.fuse(8):

--8<---------------cut here---------------start------------->8---
    ;; This hack leads libfuse to search for 'fusermount' in
    ;; $PATH, where it may find a setuid-root binary, instead of
    ;; trying solely $out/sbin/fusermount and failing because
    ;; it's not setuid.
    (substitute* "lib/Makefile"
      (("-DFUSERMOUNT_DIR=[[:graph:]]+")
       "-DFUSERMOUNT_DIR=\\\"/var/empty\\\""))
--8<---------------cut here---------------end--------------->8---

On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
/run/setuid-programs, so things just work.

On foreign distros, there’s not much we can do: Users have to set add a
setuid ‘fusermount’ in their PATH.

Closing this bug as “wontfix.”

Ludo’.

bug#21142: Sshfs-fuse requires fuse

[Claes Wallin]

[-- Attachment #1: Type: text/plain, Size: 641 bytes --]

On 29-Nov-2015 12:19 pm, "Ludovic Courtès" <ludo@gnu.org> wrote:

> However, we explicitly patch it because in practice the ‘fusermount’
> binary must be at least set-user-gid to fuse, per mount.fuse(8):
[ . . . ]
> On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
> /run/setuid-programs, so things just work.
>
> On foreign distros, there’s not much we can do: Users have to set add a
> setuid ‘fusermount’ in their PATH.
>
> Closing this bug as “wontfix.”

Ok, that's pretty clear then. Is this clear somewhere in the documentation?
Otherwise I suppose we should fix that.

-- 
   /c

[-- Attachment #2: Type: text/html, Size: 835 bytes --]

bug#21142: Sshfs-fuse requires fuse

[Ludovic Courtès]

"Claes Wallin (韋嘉誠)" <gnu@clacke.user.lysator.liu.se> skribis:

> On 29-Nov-2015 12:19 pm, "Ludovic Courtès" <ludo@gnu.org> wrote:
>
>> However, we explicitly patch it because in practice the ‘fusermount’
>> binary must be at least set-user-gid to fuse, per mount.fuse(8):
> [ . . . ]
>> On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
>> /run/setuid-programs, so things just work.
>>
>> On foreign distros, there’s not much we can do: Users have to set add a
>> setuid ‘fusermount’ in their PATH.
>>
>> Closing this bug as “wontfix.”
>
> Ok, that's pretty clear then. Is this clear somewhere in the documentation?
> Otherwise I suppose we should fix that.

Regarding foreign distros, I suppose we could add a paragraph about FUSE
under “Application Setup”?  Would you like to submit a patch against
guix.texi?

Thanks,
Ludo’.

bug#21142: Sshfs-fuse requires fuse

[Claes Wallin]

[-- Attachment #1: Type: text/plain, Size: 957 bytes --]

On Nov 29, 2015 3:58 PM, "Ludovic Courtès" <ludo@gnu.org> wrote:

> >> However, we explicitly patch it because in practice the ‘fusermount’
> >> binary must be at least set-user-gid to fuse, per mount.fuse(8):
> > [ . . . ]
> >> On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
> >> /run/setuid-programs, so things just work.

You mean under GuixSD this works automagically? Or only if fuse is
installed globally and indicated as setuid? (I forget how that works, but
I'll look into it, no need to describe in detail here)

> > Ok, that's pretty clear then. Is this clear somewhere in the
documentation?
> > Otherwise I suppose we should fix that.
>
> Regarding foreign distros, I suppose we could add a paragraph about FUSE
> under “Application Setup”?  Would you like to submit a patch against
> guix.texi?

I would like to. I'll see when I get around to it, hopefully around New
Year's.

-- 
   /c

[-- Attachment #2: Type: text/html, Size: 1220 bytes --]

bug#21142: Sshfs-fuse requires fuse

[Ludovic Courtès]

"Claes Wallin (韋嘉誠)" <gnu@clacke.user.lysator.liu.se> skribis:

> On Nov 29, 2015 3:58 PM, "Ludovic Courtès" <ludo@gnu.org> wrote:
>
>> >> However, we explicitly patch it because in practice the ‘fusermount’
>> >> binary must be at least set-user-gid to fuse, per mount.fuse(8):
>> > [ . . . ]
>> >> On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
>> >> /run/setuid-programs, so things just work.
>
> You mean under GuixSD this works automagically?

Yes.  ‘fusermount’ is part of ‘%setuid-programs’:

  http://www.gnu.org/software/guix/manual/html_node/Setuid-Programs.html

Ludo’.

bug#21142: Sshfs-fuse requires fuse

[Claes Wallin]

[-- Attachment #1: Type: text/plain, Size: 483 bytes --]

On Nov 29, 2015 6:01 PM, "Ludovic Courtès" <ludo@gnu.org> wrote:

> >> >> On GuixSD, this is addressed by having ‘fusermount’ setuid-root in
> >> >> /run/setuid-programs, so things just work.
> >
> > You mean under GuixSD this works automagically?
>
> Yes.  ‘fusermount’ is part of ‘%setuid-programs’:

Still, fuse needs to be installed on the OS if a user wants to install and
run sshfs. When I take a look at the docs, I'll keep this in mind.

-- 
   /c

[-- Attachment #2: Type: text/html, Size: 677 bytes --]

bug#21142: Sshfs-fuse requires fuse

[outlook user]

That workaround worked in 2014, but now "libfuse" seems to search under "/usr/bin" and not "/run/setuid-programs". The `DFUSERMOUNT_DIR` modification doesn't seem to work anymore. I think to reproduce all what is needed is `strace` with follow forks to see the software seeking "/usr/bin." And personally, I use an "AppImage" to invoke fuse that will seek for `fusermount3`

The author of libfuse will push an update in the future for libfuse to give up upon searching binaries by itself and only relying on $PATH simply, but that's not for now...

bug#21142: Sshfs-fuse requires fuse

[outlook user]

[-- Attachment #1: Type: text/plain, Size: 551 bytes --]

That workaround worked in 2014, but now "libfuse" seems to search under "/usr/bin" and not "/run/setuid-programs". The `DFUSERMOUNT_DIR` modification doesn't seem to work anymore. I think to reproduce all what is needed is `strace` with follow forks to see the software seeking "/usr/bin." And personally, I use an "AppImage" to invoke fuse that will seek for `fusermount3`

The author of libfuse will push an update in the future for libfuse to give up upon searching binaries by itself and only relying on $PATH simply, but that's not for now...

[-- Attachment #2: Type: text/html, Size: 963 bytes --]