Re: [uml-devel] hostfs as root

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Henrik" == Henrik Nordstrom <hno@marasystems.com> writes:
    Henrik> The rw mount option works fine here, but I am not using a full
    Henrik> system  
    Henrik> image but an image primarily designed for read-only root so mount
    Henrik> is happy  
    Henrik> in either case here..

  I went back to 2.4.19 + patch 47, with the same boot scripts (i.e. rootfs=,
not ubd0=)  No problem there.

    Henrik> Verify the permissions again.. also try booting in a emergency
    Henrik> shell  
    Henrik> (init=/bin/sh) to verify that it is not your system image which
    Henrik> remounts  
    Henrik> the root read-only while booting..

  Hmm. that's possible, but why would 2.4.19 not care?

  Let me try this hypothesis.... you are partly right.

  It looks like the only real complaint is that it can't open /etc/mtab~
at boot time, and it complains, and the /etc/mtab file isn't created
properly:

Mounted devfs on /dev
INIT: version 2.78 booting
Activating swap...
can't open lock file /etc/mtab~: Permission denied (use -n flag to override)
can't open lock file /etc/mtab~: Permission denied (use -n flag to override)
...

  /etc/mtab~ is created with mode 0, and this shows up in the host that way
as well. Perhaps this is the real problem. 

  Actually, I don't seem to be able to reproduce the read-only root now!
I swear it was complaining that it couldn't write, because the file system
was read only before.

cassidy-[nightly/testing/klips/east-icmp-01] mcr 1141 %ls -lta /c2/freeswan/sandboxes/nightly/UMLPOOL/east/root/etc/mtab*  
- -rw-r--r--    1 mcr      mcr             0 Sep  2 12:24 /c2/freeswan/sandboxes/nightly/UMLPOOL/east/root/etc/mtab
- ----------    1 mcr      mcr             0 Sep  2 12:24 /c2/freeswan/sandboxes/nightly/UMLPOOL/east/root/etc/mtab~

  I tried removing the file in my boot script first, but that doesn't help.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [




-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1TFRIqHRg3pndX9AQEZjAQAxTM87CRMtT3dLR9XWK8kTruB4bITRONS
tuPlUVq9Pp1cNUzwycKK2SIJqBfzOvlkNCI9TQ4B3mzsfS8zNQPdBXJk/Iyd6X7F
VTmM91KMoPCmCwLihcxxT6Dx86rPo1vu9xc338wnkhjQ9/BkCuhQFZ5YgGKcGqjf
/N7s/nDoB+U=
=L/3s
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Henrik Nordstrom]

tis 2003-09-02 klockan 18.28 skrev Michael Richardson:

> can't open lock file /etc/mtab~: Permission denied (use -n flag to override)
> can't open lock file /etc/mtab~: Permission denied (use -n flag to override)
> ...
> 
>   /etc/mtab~ is created with mode 0, and this shows up in the host that way
> as well. Perhaps this is the real problem.

Right.. hostfs will almost certainly not work well with file mode 0
until the virtual permissions layer is implemented.

Today hostfs directly maps permissions to the host files, which breaks
root semantics for UML. Normally root expects that mode does not matter
when reading/writing files, but when using hostfs it does matter when
UML accesses the host.

Regards
Henrik

-- 
Henrik Nordstrom <hno@marasystems.com>
MARA Systems AB





-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Wed, Sep 03, 2003 at 11:32:18AM +0200, Henrik Nordstrom wrote:

> Right.. hostfs will almost certainly not work well with file mode 0
> until the virtual permissions layer is implemented.
> 
> Today hostfs directly maps permissions to the host files, which breaks
> root semantics for UML. Normally root expects that mode does not matter
> when reading/writing files, but when using hostfs it does matter when
> UML accesses the host.

If your UML is dynamically linked, you can run it inside fakeroot to get
these semantics.

It would be very handy to have this implemented in UML itself, though.  I
work around it in various annoying ways at the moment.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Henrik" == Henrik Nordstrom <hno@marasystems.com> writes:
    >> /etc/mtab~ is created with mode 0, and this shows up in the host that
    >> way 
    >> as well. Perhaps this is the real problem.

    Henrik> Right.. hostfs will almost certainly not work well with file mode 0
    Henrik> until the virtual permissions layer is implemented.

    Henrik> Today hostfs directly maps permissions to the host files, which
    Henrik> breaks 

  But, did this change between 2.4.19-47 and 2.4.20-8?

  2.4.19 works just fine.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1YxfYqHRg3pndX9AQGFYQP+L9zsftdJvN22ca1Ty10VlDdqGJxnQ+2d
7ntU0LcXZVvnnQEZE8tH0HqDK4NrHVZIucFbDJahJmeSMZ+i+cuGAkjx8Qt3K/Wi
Mf927+F2YblgCunTByDatZlcfuangH44kcrIcljpJuZdYS/sljsKRFLT76urEYpK
E5Pc8YHpjfc=
=YQW5
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Henrik Nordstrom]

On Wed, 3 Sep 2003, Michael Richardson wrote:

>   But, did this change between 2.4.19-47 and 2.4.20-8?

Not that I know of, but some timing may have changed making the problem of 
the dual permission layers more apparent compared to earlier..

As I said I mostly use hostfs in read-only mode, and sometimes in 
read-write for plain filetransfer purposes, so I barely if ever run into 
this kind of issues other than when playing around testing what would 
happen..

Regards
Henrik



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Wed, Sep 03, 2003 at 02:22:54PM -0400, Michael Richardson wrote:

> >>>>> "Henrik" == Henrik Nordstrom <hno@marasystems.com> writes:
>     >> /etc/mtab~ is created with mode 0, and this shows up in the host that
>     >> way 
>     >> as well. Perhaps this is the real problem.
> 
>     Henrik> Right.. hostfs will almost certainly not work well with file mode 0
>     Henrik> until the virtual permissions layer is implemented.
> 
>     Henrik> Today hostfs directly maps permissions to the host files, which
>     Henrik> breaks 
> 
>   But, did this change between 2.4.19-47 and 2.4.20-8?
> 
>   2.4.19 works just fine.

Yes, it did.  Things were changed so that files could be created read-only
with O_RDWR.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
    >> >>>>> "Henrik" == Henrik Nordstrom <hno@marasystems.com> writes:
    >> >> /etc/mtab~ is created with mode 0, and this shows up in the host that
    >> >> way 
    >> >> as well. Perhaps this is the real problem.
    >> 
    Henrik> Right.. hostfs will almost certainly not work well with file mode 0
    Henrik> until the virtual permissions layer is implemented.
    >> 
    Henrik> Today hostfs directly maps permissions to the host files, which
    Henrik> breaks 
    >> 
    >> But, did this change between 2.4.19-47 and 2.4.20-8?
    >> 
    >> 2.4.19 works just fine.

    Matt> Yes, it did.  Things were changed so that files could be created
    Matt> read-only 
    Matt> with O_RDWR.

  okay.. Makes sense. But, since they are mode 0, we can't open them as
the user, which root inside could do. Is there some compromise that we can
do until we have the permissions layer in place?
  Like, forcing u+r on the underlying file maybe? 

  Or, if the mode is 0, and uid=0 inside, that we chmod u+r, then open?
  I'm happy to code this, but I don't want to waste time here :-)

  I could stay with 2.4.19-47, but we'd like to run on RH9 ;-)

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1fe5YqHRg3pndX9AQHR2wP+JaOp4GcXqIdz1ewR3Zh6+tcu5mi5EOZW
aWbya1Xq2m+EsG0d8hs3o3Jsuq2oVrLtncy+OYs7An+GDOL0/y7QmrSqnXS38grC
NCyYi5C25grnjNq17JWLtOzsE9c8VqTLujMhkF8mV88IHtd9sapOXuiXs+NkZDuS
WytpNbf9oW4=
=dbdy
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Thu, Sep 04, 2003 at 08:55:03PM -0400, Michael Richardson wrote:

>   okay.. Makes sense. But, since they are mode 0, we can't open them as
> the user, which root inside could do. Is there some compromise that we can
> do until we have the permissions layer in place?
>   Like, forcing u+r on the underlying file maybe? 

The problem I ran into which led me to this bug was extracting tar archives.
When they contain read-only files, tar opens them for writing with read-only
permissions.  Since UML was creating the file and then, subsequently,
attempting to open it for writing, the second open failed.

I think that in general, having the permissions silently changed would be
unwise, though I admit that u-r permissions are weird at best.  It would
seem better to have mount fixed to stop doing this strange thing, and
eventually to emulate full permission control in hostfs.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
    >> okay.. Makes sense. But, since they are mode 0, we can't open them as
    >> the user, which root inside could do. Is there some compromise that we
    >> can 
    >> do until we have the permissions layer in place?
    >> Like, forcing u+r on the underlying file maybe? 

    Matt> The problem I ran into which led me to this bug was extracting tar
    Matt> archives. 
    Matt> When they contain read-only files, tar opens them for writing with
    Matt> read-only 
    Matt> permissions.  Since UML was creating the file and then, subsequently,
    Matt> attempting to open it for writing, the second open failed.

  How is this the correct thing to do for the UML then?
  What bug did you uncover?

    Matt> I think that in general, having the permissions silently changed
    Matt> would be 
    Matt> unwise, though I admit that u-r permissions are weird at best.  It

  Well, if the opener is root, you get to ignore the permissions. That means
that something has to break.

  I propose:
    stat(), save permissions
    chmod(u+rw) (as appropriate for the open)
    open()
    chmod(oldperms)

    Matt> would 
    Matt> seem better to have mount fixed to stop doing this strange thing, and
    Matt> eventually to emulate full permission control in hostfs.

  Well, leaving us in a lurch is decidely not friendly.  
  Your fix has broken debian potato's mount. Being able to run other distros,
versions, etc. inside of UML is one of the major features of UML.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1iTG4qHRg3pndX9AQFsbQP9FP28wQ6dkztxuLL+No3+3sc/nmh7ZSyC
YKmwdPdXBWJ8bMdQUdW2RAnGOXKHUtZ70S4pUOMljQe2pWkRs4aCspAdKpaQwG2c
iLmqMabpoH5eACRgxfeuqgHVdDu0ognFJELkT61pAQrN66nAfGIYvN59cRIPjR56
PXZdGyk1/G4=
=lErV
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Fri, Sep 05, 2003 at 09:43:56AM -0400, Michael Richardson wrote:

> 
> >>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
>     Matt> The problem I ran into which led me to this bug was extracting tar
>     Matt> archives. 
>     Matt> When they contain read-only files, tar opens them for writing with
>     Matt> read-only 
>     Matt> permissions.  Since UML was creating the file and then, subsequently,
>     Matt> attempting to open it for writing, the second open failed.
> 
>   How is this the correct thing to do for the UML then?
>   What bug did you uncover?

I described it above.  tar was completely unable to extract files with
read-only permissions.

>     Matt> I think that in general, having the permissions silently changed
>     Matt> would be 
>     Matt> unwise, though I admit that u-r permissions are weird at best.  It
> 
>   Well, if the opener is root, you get to ignore the permissions. That means
> that something has to break.
> 
>   I propose:
>     stat(), save permissions
>     chmod(u+rw) (as appropriate for the open)
>     open()
>     chmod(oldperms)

Sounds like a recipe for lots of race conditions.

>     Matt> would 
>     Matt> seem better to have mount fixed to stop doing this strange thing, and
>     Matt> eventually to emulate full permission control in hostfs.
> 
>   Well, leaving us in a lurch is decidely not friendly.  
>   Your fix has broken debian potato's mount. Being able to run other distros,
> versions, etc. inside of UML is one of the major features of UML.

OK, so mount is already fixed.  So the only problem is that you can't use an
old, mildly buggy version of mount with root-hostfs as a non-root user on
the host.  I'd rather be able to extract tar files than have that scenario.

You must have already worked around some other problems in order to run
potato in this configuration; it wouldn't work out of the box with a 2.4
kernel.  How are you managing device nodes?  Loading any kernel modules?

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
    >> Well, if the opener is root, you get to ignore the permissions. That
    >> means that something has to break.
    >> 
    >> I propose: stat(), save permissions chmod(u+rw) (as appropriate for
    >> the open) open() chmod(oldperms)

    Matt> Sounds like a recipe for lots of race conditions.

  Uh, how?
  Remember, this is in the kernel. We have locks.

  If two UMLs come along, using the same rootfs, well, that's a different
problem. 

    >> Well, leaving us in a lurch is decidely not friendly.  Your fix has
    >> broken debian potato's mount. Being able to run other distros,
    >> versions, etc. inside of UML is one of the major features of UML.

    Matt> OK, so mount is already fixed.  So the only problem is that you
    Matt> can't use an old, mildly buggy version of mount with root-hostfs as
    Matt> a non-root user on the host.  I'd rather be able to extract tar
    Matt> files than have that scenario.

  Okay. I see.

    Matt> You must have already worked around some other problems in order to
    Matt> run potato in this configuration; it wouldn't work out of the box
    Matt> with a 2.4 kernel.  How are you managing device nodes?  Loading any
    Matt> kernel modules?

  Works fine for me.
  It has been working fine for two years now.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1iw2YqHRg3pndX9AQEzJwP/T0G/+pKEc91jtkOOaVOcHEhGA7kyt+u2
fum+nFEkyeI7P/B+gn56j7ilGflQF/9PJUAZliQByXSh3kT+t+K6QfANV7b9MRSI
V6tk23VZ9aiUJffqo0pirs0l2ePFfgR09kw7U0se8IGM0c/+OWvPYZSXClEt8yGb
ptqs9Hzamms=
=Biha
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Adam Heath]

On Fri, 5 Sep 2003, Michael Richardson wrote:

>     >> I propose: stat(), save permissions chmod(u+rw) (as appropriate for
>     >> the open) open() chmod(oldperms)
>
>     Matt> Sounds like a recipe for lots of race conditions.
>
>   Uh, how?
>   Remember, this is in the kernel. We have locks.
>
>   If two UMLs come along, using the same rootfs, well, that's a different
> problem.

No, it's a problem.  Just because there's no race in the emulated environment,
doesn't mean there isn't a race in the host.



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Adam" == Adam Heath <adam@doogie.org> writes:
    Adam> On Fri, 5 Sep 2003, Michael Richardson wrote:

    >> >> I propose: stat(), save permissions chmod(u+rw) (as appropriate for
    >> >> the open) open() chmod(oldperms)
    >> 
    Matt> Sounds like a recipe for lots of race conditions.
    >> Uh, how?  Remember, this is in the kernel. We have locks.
    >> 
    >> If two UMLs come along, using the same rootfs, well, that's a
    >> different problem.

    Adam> No, it's a problem.  Just because there's no race in the emulated
    Adam> environment, doesn't mean there isn't a race in the host.

  I don't think that hostfs provides any inter-UML locking guarantees.
  What do we do with flock(2)?

  If you like, we can flock(2) the file too while we play with it.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [
  
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1i5x4qHRg3pndX9AQErKAQAjb02jUIC6x+RT3ojY8Ta9G3wuyqJFs35
O/ySSn/73X5kaAc3Vh96zSkaLI3+hE3LTR+CWhtbfSGFZCXDjcahVu9/2wuUm8Ud
7p6/1C7ZwNOZk5EJk/8rGpHP2oJAl4qY/bvcSoZ9f1JequzCilhjyxuua/4t09kW
ZxqtZdP9nEM=
=TF2P
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Adam Heath]

On Fri, 5 Sep 2003, Michael Richardson wrote:

>     Adam> No, it's a problem.  Just because there's no race in the emulated
>     Adam> environment, doesn't mean there isn't a race in the host.
>
>   I don't think that hostfs provides any inter-UML locking guarantees.
>   What do we do with flock(2)?
>
>   If you like, we can flock(2) the file too while we play with it.

This has not hing to do with inter-uml.  This has everything to do with uml
behaving like a sane, properly written application.

If 2 normal applications did something like this, they'd be buggy just as
well.



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Fri, Sep 05, 2003 at 11:50:50AM -0400, Michael Richardson wrote:

> >>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
>     Matt> Sounds like a recipe for lots of race conditions.
> 
>   Uh, how?
>   Remember, this is in the kernel. We have locks.

There is not currently a lock held by hostfs_file_open that I see; this does
not seem like sufficient reason to make it impossible to do two opens
simultaneously.

>   If two UMLs come along, using the same rootfs, well, that's a different
> problem. 

It's the same problem, and it affects any shared hostfs, not only root
(unless you were planning to add a test for that).

>     Matt> You must have already worked around some other problems in order to
>     Matt> run potato in this configuration; it wouldn't work out of the box
>     Matt> with a 2.4 kernel.  How are you managing device nodes?  Loading any
>     Matt> kernel modules?
> 
>   Works fine for me.
>   It has been working fine for two years now.

What works fine?  The modutils in potato doesn't support 2.4 kernels,
only root is allowed to create device nodes in the host filesystem, and the
paths in /etc/inittab and everywhere else won't exist with devfs unless
devfsd is running.  You're obviously working around these problems.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
    >> Works fine for me.  It has been working fine for two years now.

    Matt> What works fine?  The modutils in potato doesn't support 2.4
    Matt> kernels, only root is allowed to create device nodes in the host
  
  I don't use modutils from potato, since I build static kernels.

    Matt> filesystem, and the paths in /etc/inittab and everywhere else won't
    Matt> exist with devfs unless devfsd is running.  You're obviously
    Matt> working around these problems.

  wasn't me.
  *someone* did this when they created the root system.

  If you have a simple /sbin/mount that you can recommend that solves my
problem, I will consider using it.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [

  
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1jJ8IqHRg3pndX9AQHU2wQAqUYSRG2Sq7juITMIoGRQn0RyNnT2tG90
CvUVozOCCuayBdVnZj0b3X/biqYjZ/es/5dH4xQWtskE266mE91iq7aJrykM/eF/
ai8+1yesR8MWxNr17oF49A5TzgWGnClNyHXRkocJPooTF0Se0zwv87AoG08IOIrT
sAziFFX0tjM=
=TBUB
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Fri, Sep 05, 2003 at 01:37:53PM -0400, Michael Richardson wrote:

>   If you have a simple /sbin/mount that you can recommend that solves my
> problem, I will consider using it.

The fact that you said it broke _potato_'s mount, specifically, led me to
believe that newer versions did not do this (they do).

However, the fix to mount is trivial, so you can easily create an
/sbin/mount which should work.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
    >> If you have a simple /sbin/mount that you can recommend that solves my
    >> problem, I will consider using it.

    Matt> The fact that you said it broke _potato_'s mount, specifically, led
    Matt> me to believe that newer versions did not do this (they do).

  Ah, so your fix basically breaks all distros? I have no idea.
  Then clearly it needs to be fixed in the kernel, don't you think?

    Matt> However, the fix to mount is trivial, so you can easily create an
    Matt> /sbin/mount which should work.

  I think that removing your patch is a better idea. Find another way to make
tar happy. 

  Booting the system is pretty important.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1jYQ4qHRg3pndX9AQEtSwQA7KjbmS24RfWnvO1z2bRBzjmmr/u2prXP
q3bDmsEiS49oFA+00d/Q+pp2z9CIC6NWE2GOzK4VnMDoOXu48mDURUiZ+BhNdv0Y
rwzqW0tJrBTrBeceLNZYtAU+hTUoEKrLDLcL44ihPEKxLqDkWjWBL2kEDyAgFMiJ
P9n+/XjBWjo=
=34L/
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Fri, Sep 05, 2003 at 02:39:01PM -0400, Michael Richardson wrote:

> >>>>> "Matt" == Matt Zimmerman <mdz@debian.org> writes:
>     Matt> The fact that you said it broke _potato_'s mount, specifically, led
>     Matt> me to believe that newer versions did not do this (they do).
> 
>   Ah, so your fix basically breaks all distros?

No.

>   I think that removing your patch is a better idea. Find another way to make
> tar happy. 

I'm not motivated to put any more effort into helping you fix your problem.
I already filed a patch for mount in the BTS.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[BlaisorBlade]

Alle 20:39, venerdì 5 settembre 2003, Michael Richardson ha scritto:

>     >> If you have a simple /sbin/mount that you can recommend that solves
>     >> my problem, I will consider using it.
>
>     Matt> The fact that you said it broke _potato_'s mount, specifically,
> led Matt> me to believe that newer versions did not do this (they do).
>
>   Ah, so your fix basically breaks all distros? I have no idea.
>   Then clearly it needs to be fixed in the kernel, don't you think?
>
>     Matt> However, the fix to mount is trivial, so you can easily create an
>     Matt> /sbin/mount which should work.
>
>   I think that removing your patch is a better idea. Find another way to
> make tar happy.
>
>   Booting the system is pretty important.
I think that since anyway there is not a permission layer, you shouldn't boot 
off an hostfs filesystem. Mount with -o loop a new file, then copy all the 
filesystem content. Or maybe(but ask to other developers/check docs) if you 
have a whole partition, say ubd0=/dev/<your part> to UML. With hostfs, 
anyway, if you don't run the guest with root privileges, files created by 
root will be owned by the user running the Uml(i.e. it can happen that 
everything works, especially if you login in the Uml only with a uid=<uid of 
Uml on the host>, but not everything will work), while root won't have access 
to certain files(you can avoid also this, by not putting in that hierarchy 
any file of root from the host).

-- 
cat <<EOSIGN
Paolo Giarrusso, aka Blaisorblade
Linux Kernel 2.4.21/2.6.0-test on an i686; Linux registered user n. 292729
EOSIGN



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[stian]

The problem occurs when you create a file you are later not allowed to
open if I'm seeing this right, so, the best fix would be not to close the
file when we create them in the first place. Fix the behavour that we

create file
close file
open file again
do action

If that is redone till

create file
do action


The mode issue goes pretty much away, since trying to open a file you are
not allowed to read will normally fail both in host and uml. Only problem
is when creating new files as far as I can see.


Might be more job, but the result is more cleaner than than chmod stuff on
the fly.

Stian




This mail has been scanned for known viruses on an open, Linux-based mailsystem
developed by http://Knowledge-Network.no and http://Nixia.no.


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Michael Richardson]

-----BEGIN PGP SIGNED MESSAGE-----


>>>>> "stian" == stian  <stian@nixia.no> writes:
    stian> The problem occurs when you create a file you are later not
    stian> allowed to 
    stian> open if I'm seeing this right, so, the best fix would be not to
    stian> close the 
    stian> file when we create them in the first place. Fix the behavour that
    stian> we

  I believe that we do this because in TT mode at least, we may have to open
it in multiple processes.

  So the fix would work for SKAS, but not TT mode.

    stian> The mode issue goes pretty much away, since trying to open a file
    stian> you are  not allowed to read will normally fail both in host and
    stian> uml. Only problem is when creating new files as far as I can see.

  Root can open a file for write that doesn't have write permissions.

]      Out and about in Ottawa.    hmmm... beer.                |  firewalls  [
]   Michael Richardson, Sandelman Software Works, Ottawa, ON    |net architect[
] mcr@sandelman.ottawa.on.ca http://www.sandelman.ottawa.on.ca/ |device driver[
] panic("Just another Debian/notebook using, kernel hacking, security guy");  [
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)
Comment: Finger me for keys - custom hacks make this fully PGP2 compat

iQCVAwUBP1jOyoqHRg3pndX9AQG4cAP/codBpW4FkhBKFrLpDmFpp5+tqy0hPVlt
VFwfdJ3nTKJzWJBkt82FVH9J5/2MkVTN0w+V4/LWw/DskOz22YdIczyoFZIXTChK
IGHncjsx8FMhhgxwBZ8s9toHluQGG8t88ZRETDLQQmm3BIBGfUmlsVUdoLRjKhvg
dyi7xT4sT3s=
=mfEh
-----END PGP SIGNATURE-----


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [uml-devel] hostfs as root

[Matt Zimmerman]

On Fri, Sep 05, 2003 at 05:14:57PM +0200, stian@nixia.no wrote:

> The problem occurs when you create a file you are later not allowed to
> open if I'm seeing this right, so, the best fix would be not to close the
> file when we create them in the first place. Fix the behavour that we
> 
> create file
> close file
> open file again
> do action
> 
> If that is redone till
> 
> create file
> do action
> 
> 
> The mode issue goes pretty much away, since trying to open a file you are
> not allowed to read will normally fail both in host and uml. Only problem
> is when creating new files as far as I can see.
> 
> 
> Might be more job, but the result is more cleaner than than chmod stuff on
> the fly.

That is what the original bug was, and that is how it was fixed.  Michael's
problem is that mount(8) creates a file with mode 0, closes it, and then
tries to open it.

-- 
 - mdz


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel