On Thu, May 21, 2009 at 02:54:19PM +0200, Jan Rekorajski wrote:
> On Mon, 18 May 2009, Jan Blunck wrote:
>
> > Here is another post of the VFS based union mount implementation.
>
> Is there any chance this will support NFS? I can union-mount tmpfs over
NFS as the read-only layer ought to work. NFS as the read-write layer
is still up in the air.
> nfs mounted fs, but if I try to mount --union two NFS filesystems I
> always get -EBUSY on second mount on the same mountpoint.
>
> Something along these lines:
>
> doesn't matter if I use --union on first mount, the result is always the
> same.
>
> mount <--union> -t nfs server:/export/system /mnt
> OK
> mount --union -t nfs server:/export/profile /mnt
> mount.nfs: /mnt is busy or already mounted
>
> I patched mount.nfs so it knows about MS_UNION, and strace shows me that
> it passes that flag to kernel.
FYI, using --union on the first mount will make it union with the
local directory below it. The --union option is not needed when you
mount the lower read-only layer.
You'll get -EBUSY on the second mount of any NFS file system over
another - try it again with the --union flag. Support for NFS on NFS
union mount would have to change this.
-VAL
On Mon, 08 Jun 2009, Valerie Aurora wrote:
> On Thu, May 21, 2009 at 02:54:19PM +0200, Jan Rekorajski wrote:
> > On Mon, 18 May 2009, Jan Blunck wrote:
> >
> > > Here is another post of the VFS based union mount implementation.
> >
> > Is there any chance this will support NFS? I can union-mount tmpfs over
>
> NFS as the read-only layer ought to work. NFS as the read-write layer
> is still up in the air.
As I don't need rw NFS, i didn't even try that :)
> > nfs mounted fs, but if I try to mount --union two NFS filesystems I
> > always get -EBUSY on second mount on the same mountpoint.
> >
> > Something along these lines:
> >
> > doesn't matter if I use --union on first mount, the result is always the
> > same.
> >
> > mount <--union> -t nfs server:/export/system /mnt
> > OK
> > mount --union -t nfs server:/export/profile /mnt
> > mount.nfs: /mnt is busy or already mounted
> >
> > I patched mount.nfs so it knows about MS_UNION, and strace shows me that
> > it passes that flag to kernel.
>
> FYI, using --union on the first mount will make it union with the
> local directory below it. The --union option is not needed when you
> mount the lower read-only layer.
Thanks for clarification.
> You'll get -EBUSY on the second mount of any NFS file system over
> another - try it again with the --union flag. Support for NFS on NFS
> union mount would have to change this.
I did just that, --union didn't change standard NFS behaviour.
mount -t nfs server:/export/system /mnt
mount --union -t nfs server:/export/profile /mnt
mount.nfs: /mnt is busy or already mounted
I did an experiment by using different IP of the server (same machine)
when mounting the second fs, mount worked then, but 'ls -1 /mnt' oopsed.
I can reproduce this and send you the oops next week.
--
Jan Rekorajski | ALL SUSPECTS ARE GUILTY. PERIOD!
baggins<at>mimuw.edu.pl | OTHERWISE THEY WOULDN'T BE SUSPECTS, WOULD THEY?
BOFH, MANIAC | -- TROOPS by Kevin Rubio
On Tue, Jun 09, 2009 at 12:44:06AM +0200, Jan Rekorajski wrote:
> On Mon, 08 Jun 2009, Valerie Aurora wrote:
>
> > On Thu, May 21, 2009 at 02:54:19PM +0200, Jan Rekorajski wrote:
> > > On Mon, 18 May 2009, Jan Blunck wrote:
> > >
> > > > Here is another post of the VFS based union mount implementation.
> > >
> > > Is there any chance this will support NFS? I can union-mount tmpfs over
> >
> > NFS as the read-only layer ought to work. NFS as the read-write layer
> > is still up in the air.
>
> As I don't need rw NFS, i didn't even try that :)
>
> > > nfs mounted fs, but if I try to mount --union two NFS filesystems I
> > > always get -EBUSY on second mount on the same mountpoint.
> > >
> > > Something along these lines:
> > >
> > > doesn't matter if I use --union on first mount, the result is always the
> > > same.
> > >
> > > mount <--union> -t nfs server:/export/system /mnt
> > > OK
> > > mount --union -t nfs server:/export/profile /mnt
> > > mount.nfs: /mnt is busy or already mounted
> > >
> > > I patched mount.nfs so it knows about MS_UNION, and strace shows me that
> > > it passes that flag to kernel.
> >
> > FYI, using --union on the first mount will make it union with the
> > local directory below it. The --union option is not needed when you
> > mount the lower read-only layer.
>
> Thanks for clarification.
>
> > You'll get -EBUSY on the second mount of any NFS file system over
> > another - try it again with the --union flag. Support for NFS on NFS
> > union mount would have to change this.
>
> I did just that, --union didn't change standard NFS behaviour.
Er, excuse me - I mean to type "try it again WITHOUT the --union
flag." My apologies!
>
> mount -t nfs server:/export/system /mnt
> mount --union -t nfs server:/export/profile /mnt
> mount.nfs: /mnt is busy or already mounted
>
> I did an experiment by using different IP of the server (same machine)
> when mounting the second fs, mount worked then, but 'ls -1 /mnt' oopsed.
> I can reproduce this and send you the oops next week.
Interesting! Does this happen without the --union flag?
-VAL
On Mon, 08 Jun 2009, Valerie Aurora wrote:
> On Tue, Jun 09, 2009 at 12:44:06AM +0200, Jan Rekorajski wrote:
> > On Mon, 08 Jun 2009, Valerie Aurora wrote:
> >
> > > You'll get -EBUSY on the second mount of any NFS file system over
> > > another - try it again with the --union flag. Support for NFS on NFS
> > > union mount would have to change this.
> >
> > I did just that, --union didn't change standard NFS behaviour.
>
> Er, excuse me - I mean to type "try it again WITHOUT the --union
> flag." My apologies!
I did, see below.
> >
> > mount -t nfs server:/export/system /mnt
> > mount --union -t nfs server:/export/profile /mnt
> > mount.nfs: /mnt is busy or already mounted
> >
> > I did an experiment by using different IP of the server (same machine)
> > when mounting the second fs, mount worked then, but 'ls -1 /mnt' oopsed.
> > I can reproduce this and send you the oops next week.
>
> Interesting! Does this happen without the --union flag?
Filesystems are exported ro, all mounts nfs3,ro,tcp.
Without --union:
mount -t nfs 10.1.0.4:/nfs/system /mnt -oro,nolock,vers=3,tcp
OK
mount -t nfs 10.1.0.4:/nfs/profile /mnt -oro,nolock,vers=3,tcp
-EBUSY (as expected)
mount -t nfs 10.1.0.3:/nfs/profile /mnt -oro,nolock,vers=3,tcp
(notice different IP - but it's the same machine)
Works, just overmounts /mnt, ls shown contents of /nfs/profile
Now, --union:
mount -t nfs 10.1.0.4:/nfs/system /mnt -oro,nolock,vers=3,tcp
OK
mount --union -t nfs 10.1.0.4:/nfs/profile /mnt -oro,nolock,vers=3,tcp
-EBUSY (/mnt busy or already mounted)
mount --union -t nfs 10.1.0.3:/nfs/profile /mnt -oro,nolock,vers=3,tcp
(notice different IP - but it's the same machine)
mount command works, ls Oopses:
[ 61.766392] creating fallthru for opt
[ 61.766417] BUG: unable to handle kernel NULL pointer dereference at (null)
[ 61.766433] IP: [<(null)>] (null)
[ 61.766482] *pdpt = 000000001e0e9001 *pde = 0000000000000000
[ 61.767324] Oops: 0010 [#1] PREEMPT SMP
[ 61.767324] last sysfs file: /sys/kernel/uevent_seqnum
[ 61.767324] Modules linked in: nfs nfsd lockd nfs_acl auth_rpcgss sunrpc sch_sfq thermal processor thermal_sys rtc_cmos rtc_core i2c_piix4 e1000 ac psmouse rtc_lib hwmon pcspkr button i2c_core sg sr_mod serio_raw evdev cdrom
[ 61.767324]
[ 61.767324] Pid: 2460, comm: ls Not tainted (2.6.29.3 #1) VirtualBox
[ 61.767324] EIP: 0060:[<00000000>] EFLAGS: 00010286 CPU: 0
[ 61.767324] EIP is at 0x0
[ 61.767324] EAX: df6d23c0 EBX: df69d1b0 ECX: e105d3e0 EDX: df69d630
[ 61.767324] ESI: df69d630 EDI: de16d300 EBP: de1b1d18 ESP: de1b1d04
[ 61.767324] DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0069
[ 61.767324] Process ls (pid: 2460, ti=de1b0000 task=de0ee3f0 task.ti=de1b0000)
[ 61.767324] Stack:
[ 61.767324] c02d03d9 c054f57d df69d698 00000002 00000000 de1b1d7c e10357f0 00000002
[ 61.767324] 00000000 0180677d 00000000 0000000a c02d0340 df69d1b0 00000000 00000002
[ 61.767324] 0180677d 00000000 df6d2128 de16d300 de1b1f08 de1b1f08 df69d5a0 de1b1ed4
[ 61.767324] Call Trace:
[ 61.767324] [<c02d03d9>] union_copyup_dir_one+0x99/0xc0
[ 61.767324] [<e10357f0>] nfs_do_filldir+0x210/0x570 [nfs]
[ 61.767324] [<c02d0340>] union_copyup_dir_one+0x0/0xc0
[ 61.767324] [<e1036072>] nfs_readdir+0x522/0xa10 [nfs]
[ 61.767324] [<c02d0340>] union_copyup_dir_one+0x0/0xc0
[ 61.767324] [<c0293502>] __mem_cgroup_commit_charge+0x42/0x100
[ 61.767324] [<c0293c07>] mem_cgroup_charge_common+0x57/0x70
[ 61.767324] [<e1039f18>] __put_nfs_open_context+0x28/0xb0 [nfs]
[ 61.767324] [<e0a6fbe2>] rpcauth_lookup_credcache+0x152/0x1f0 [sunrpc]
[ 61.767324] [<e0a6f92d>] rpcauth_lookupcred+0x5d/0xb0 [sunrpc]
[ 61.767324] [<e1049c10>] nfs3_decode_dirent+0x0/0x220 [nfs]
[ 61.767324] [<c02d0ad1>] union_copyup_dir+0x111/0x160
[ 61.767324] [<c02a6190>] filldir64+0x0/0x110
[ 61.767324] [<c02a6575>] vfs_readdir+0xd5/0xf0
[ 61.767324] [<c02a65fd>] sys_getdents64+0x6d/0xc0
[ 61.767324] [<c02033da>] syscall_call+0x7/0xb
[ 61.767324] Code: Bad EIP value.
[ 61.767324] EIP: [<00000000>] 0x0 SS:ESP 0068:de1b1d04
[ 61.816597] ---[ end trace 60fb13bae2f23426 ]---
--
Jan Rekorajski | ALL SUSPECTS ARE GUILTY. PERIOD!
baggins<at>mimuw.edu.pl | OTHERWISE THEY WOULDN'T BE SUSPECTS, WOULD THEY?
BOFH, MANIAC | -- TROOPS by Kevin Rubio
On Mon, Jun 15, 2009 at 11:55:28AM +0200, Jan Rekorajski wrote:
> On Mon, 08 Jun 2009, Valerie Aurora wrote:
>
> > On Tue, Jun 09, 2009 at 12:44:06AM +0200, Jan Rekorajski wrote:
> > > On Mon, 08 Jun 2009, Valerie Aurora wrote:
> > >
> > > > You'll get -EBUSY on the second mount of any NFS file system over
> > > > another - try it again with the --union flag. Support for NFS on NFS
> > > > union mount would have to change this.
> > >
> > > I did just that, --union didn't change standard NFS behaviour.
> >
> > Er, excuse me - I mean to type "try it again WITHOUT the --union
> > flag." My apologies!
>
> I did, see below.
>
> > >
> > > mount -t nfs server:/export/system /mnt
> > > mount --union -t nfs server:/export/profile /mnt
> > > mount.nfs: /mnt is busy or already mounted
> > >
> > > I did an experiment by using different IP of the server (same machine)
> > > when mounting the second fs, mount worked then, but 'ls -1 /mnt' oopsed.
> > > I can reproduce this and send you the oops next week.
> >
> > Interesting! Does this happen without the --union flag?
>
> Filesystems are exported ro, all mounts nfs3,ro,tcp.
>
> Without --union:
>
> mount -t nfs 10.1.0.4:/nfs/system /mnt -oro,nolock,vers=3,tcp
> OK
>
> mount -t nfs 10.1.0.4:/nfs/profile /mnt -oro,nolock,vers=3,tcp
> -EBUSY (as expected)
>
> mount -t nfs 10.1.0.3:/nfs/profile /mnt -oro,nolock,vers=3,tcp
> (notice different IP - but it's the same machine)
> Works, just overmounts /mnt, ls shown contents of /nfs/profile
>
> Now, --union:
>
> mount -t nfs 10.1.0.4:/nfs/system /mnt -oro,nolock,vers=3,tcp
> OK
>
> mount --union -t nfs 10.1.0.4:/nfs/profile /mnt -oro,nolock,vers=3,tcp
> -EBUSY (/mnt busy or already mounted)
>
> mount --union -t nfs 10.1.0.3:/nfs/profile /mnt -oro,nolock,vers=3,tcp
> (notice different IP - but it's the same machine)
> mount command works, ls Oopses:
Sorry for the delay in reply! This is on our to-do list, thanks for
doing a little more testing.
-VAL