[Qemu-devel] 9p broken?

[Qemu-devel] 9p broken?

[Avi Kivity]

Having an annoying bug on i386 kvm I decided to debug it buy running an
i386 guest on my x86_64 host, use 9p to access a guest image, and run it
using nested kvm.

However, 9p appears to be broken: first, the configure test fails (patch
sent).  Second, while mount works, ls on the mount point causes qemu to
crash with

(gdb) bt
#0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
/home/tlv/akivity/qemu/error.c:32
#1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
/home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
#2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
#3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
#4  0x00007fffffffce00 in ?? ()
#5  0x0000000000000000 in ?? (

**errp already points to a VirtFSFeatureBlocksMigration error;
v9fs_attach() has been called a second time (the first time,
understandably, on mount; the second on ls).

Command line:

qemu-system-x86_64 -m 4G -smp 4 -drive
file=/images/Fedora-i386.img,if=virtio,cache=none -cdrom
/images/iso/bfo.iso -device virtio-9p-pci,fsdev=root,mount_tag=root
-fsdev local,id=root,path=/,security_model=passthrough -enable-kvm -net
nic,model=virtio,netdev=net -netdev user,id=net -monitor stdio -cpu host

-- 
error compiling committee.c: too many arguments to function

Re: [Qemu-devel] 9p broken?

[Richard W.M. Jones]

On Mon, Jul 30, 2012 at 03:35:39PM +0300, Avi Kivity wrote:
> Having an annoying bug on i386 kvm I decided to debug it buy running an
> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
> using nested kvm.
> 
> However, 9p appears to be broken: first, the configure test fails (patch
> sent).  Second, while mount works, ls on the mount point causes qemu to
> crash with
> 
> (gdb) bt
> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
> /home/tlv/akivity/qemu/error.c:32
> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
> #4  0x00007fffffffce00 in ?? ()
> #5  0x0000000000000000 in ?? (
> 
> **errp already points to a VirtFSFeatureBlocksMigration error;
> v9fs_attach() has been called a second time (the first time,
> understandably, on mount; the second on ls).

Yes, I can reproduce this too.

LIBGUESTFS_QEMU=~/d/qemu/qemu.wrapper \
guestfish -v -- \
  sparse /tmp/unused 100M : \
  config -device 'virtio-9p-pci,fsdev=root,mount_tag=root' : \
  config -fsdev 'local,id=root,path=/tmp,security_model=passthrough' : \
  run : \
  mount-9p root / : \
  ls /

Stack trace:

#0  0x00007fb1d4d19ba5 in __GI_raise (sig=sig@entry=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:63
#1  0x00007fb1d4d1b358 in __GI_abort () at abort.c:90
#2  0x00007fb1d4d12972 in __assert_fail_base (fmt=
    0x7fb1d4e5c8e8 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", 
    assertion=assertion@entry=0x7fb1d8e2e87e "*errp == ((void *)0)", 
    file=file@entry=0x7fb1d8e56217 "error.c", line=line@entry=35, 
    function=function@entry=
    0x7fb1d8e2e8ca <__PRETTY_FUNCTION__.13983> "error_set") at assert.c:92
#3  0x00007fb1d4d12a22 in __GI___assert_fail (assertion=assertion@entry=
    0x7fb1d8e2e87e "*errp == ((void *)0)", file=file@entry=
    0x7fb1d8e56217 "error.c", line=line@entry=35, function=function@entry=
    0x7fb1d8e2e8ca <__PRETTY_FUNCTION__.13983> "error_set") at assert.c:101
#4  0x00007fb1d8c1147f in error_set (errp=errp@entry=0x7fb1ce36a128, 
    fmt=fmt@entry=
    0x7fb1d8e39a78 "{ 'class': 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at error.c:35
#5  0x00007fb1d8c57f9b in v9fs_attach (opaque=0x7fb1ce352020)
    at /home/rjones/d/qemu/hw/9pfs/virtio-9p.c:988
#6  0x00007fb1d8c0fcfa in coroutine_trampoline (i0=<optimized out>, 
    i1=<optimized out>) at coroutine-ucontext.c:138
#7  0x00007fb1d4d2a2f0 in ?? () from /lib64/libc.so.6
#8  0x00007fff51061aa0 in ?? ()
#9  0xb5b5b5b5b5b5b5b5 in ?? ()
#10 0x0000000000000000 in ?? ()

I'll add a regression test for 9p to libguestfs so at least we will
catch this in future during Fedora builds.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
virt-df lists disk usage of guests without needing to install any
software inside the virtual machine.  Supports Linux and Windows.
http://et.redhat.com/~rjones/virt-df/

Re: [Qemu-devel] 9p broken?

[Aneesh Kumar K.V]

"Richard W.M. Jones" <rjones@redhat.com> writes:

> On Mon, Jul 30, 2012 at 03:35:39PM +0300, Avi Kivity wrote:
>> Having an annoying bug on i386 kvm I decided to debug it buy running an
>> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
>> using nested kvm.
>> 
>> However, 9p appears to be broken: first, the configure test fails (patch
>> sent).  Second, while mount works, ls on the mount point causes qemu to
>> crash with
>> 
>> (gdb) bt
>> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
>> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
>> /home/tlv/akivity/qemu/error.c:32
>> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
>> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
>> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
>> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
>> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
>> #4  0x00007fffffffce00 in ?? ()
>> #5  0x0000000000000000 in ?? (
>> 
>> **errp already points to a VirtFSFeatureBlocksMigration error;
>> v9fs_attach() has been called a second time (the first time,
>> understandably, on mount; the second on ls).
>
> Yes, I can reproduce this too.
>
> LIBGUESTFS_QEMU=~/d/qemu/qemu.wrapper \
> guestfish -v -- \
>   sparse /tmp/unused 100M : \
>   config -device 'virtio-9p-pci,fsdev=root,mount_tag=root' : \
>   config -fsdev 'local,id=root,path=/tmp,security_model=passthrough' : \
>   run : \
>   mount-9p root / : \
>   ls /
>
> Stack trace:
>
> #0  0x00007fb1d4d19ba5 in __GI_raise (sig=sig@entry=6)
>     at ../nptl/sysdeps/unix/sysv/linux/raise.c:63
> #1  0x00007fb1d4d1b358 in __GI_abort () at abort.c:90
> #2  0x00007fb1d4d12972 in __assert_fail_base (fmt=
>     0x7fb1d4e5c8e8 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", 
>     assertion=assertion@entry=0x7fb1d8e2e87e "*errp == ((void *)0)", 
>     file=file@entry=0x7fb1d8e56217 "error.c", line=line@entry=35, 
>     function=function@entry=
>     0x7fb1d8e2e8ca <__PRETTY_FUNCTION__.13983> "error_set") at assert.c:92
> #3  0x00007fb1d4d12a22 in __GI___assert_fail (assertion=assertion@entry=
>     0x7fb1d8e2e87e "*errp == ((void *)0)", file=file@entry=
>     0x7fb1d8e56217 "error.c", line=line@entry=35, function=function@entry=
>     0x7fb1d8e2e8ca <__PRETTY_FUNCTION__.13983> "error_set") at assert.c:101
> #4  0x00007fb1d8c1147f in error_set (errp=errp@entry=0x7fb1ce36a128, 
>     fmt=fmt@entry=
>     0x7fb1d8e39a78 "{ 'class': 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at error.c:35
> #5  0x00007fb1d8c57f9b in v9fs_attach (opaque=0x7fb1ce352020)
>     at /home/rjones/d/qemu/hw/9pfs/virtio-9p.c:988
> #6  0x00007fb1d8c0fcfa in coroutine_trampoline (i0=<optimized out>, 
>     i1=<optimized out>) at coroutine-ucontext.c:138
> #7  0x00007fb1d4d2a2f0 in ?? () from /lib64/libc.so.6
> #8  0x00007fff51061aa0 in ?? ()
> #9  0xb5b5b5b5b5b5b5b5 in ?? ()
> #10 0x0000000000000000 in ?? ()
>
> I'll add a regression test for 9p to libguestfs so at least we will
> catch this in future during Fedora builds.
>

I am not able to reproduce this, I had to patch configure to fix some
compile errors, but then I am not hitting the crash. I am using the
latest qemu. We do the below in 9p

    error_set(&s->migration_blocker, QERR_VIRTFS_FEATURE_BLOCKS_MIGRATION,
              s->ctx.fs_root ? s->ctx.fs_root : "NULL", s->tag);

I am not sure how we can hit that assert() in error_set() ?. 
We allocate s via 

    s = (V9fsState *)virtio_common_init() which does

    vdev = g_malloc0(struct_size);

-aneesh

Re: [Qemu-devel] 9p broken?

[Aneesh Kumar K.V]

Avi Kivity <avi@redhat.com> writes:

> Having an annoying bug on i386 kvm I decided to debug it buy running an
> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
> using nested kvm.
>
> However, 9p appears to be broken: first, the configure test fails (patch
> sent).  Second, while mount works, ls on the mount point causes qemu to
> crash with

I missed that you have already sent a patch for configure fix. That
looks better that what i sent. I will ack that patch 

>
> (gdb) bt
> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
> /home/tlv/akivity/qemu/error.c:32
> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
> #4  0x00007fffffffce00 in ?? ()
> #5  0x0000000000000000 in ?? (
>
> **errp already points to a VirtFSFeatureBlocksMigration error;
> v9fs_attach() has been called a second time (the first time,
> understandably, on mount; the second on ls).
>

Why are we calling attach a second time ?. I am also not able to reproduce this

root@qemu-img-64:~# mount -t 9p -otrans=virtio,version=9p2000.L v_tmp /mnt
root@qemu-img-64:~# ls /mnt/a.c
/mnt/a.c


> Command line:
>
> qemu-system-x86_64 -m 4G -smp 4 -drive
> file=/images/Fedora-i386.img,if=virtio,cache=none -cdrom
> /images/iso/bfo.iso -device virtio-9p-pci,fsdev=root,mount_tag=root
> -fsdev local,id=root,path=/,security_model=passthrough -enable-kvm -net
> nic,model=virtio,netdev=net -netdev user,id=net -monitor stdio -cpu host
>

-aneesh

Re: [Qemu-devel] 9p broken?

[Avi Kivity]

On 07/31/2012 09:51 AM, Aneesh Kumar K.V wrote:
> Avi Kivity <avi@redhat.com> writes:
> 
>> Having an annoying bug on i386 kvm I decided to debug it buy running an
>> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
>> using nested kvm.
>>
>> However, 9p appears to be broken: first, the configure test fails (patch
>> sent).  Second, while mount works, ls on the mount point causes qemu to
>> crash with
> 
> I missed that you have already sent a patch for configure fix. That
> looks better that what i sent. I will ack that patch 
> 
>>
>> (gdb) bt
>> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
>> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
>> /home/tlv/akivity/qemu/error.c:32
>> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
>> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
>> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
>> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
>> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
>> #4  0x00007fffffffce00 in ?? ()
>> #5  0x0000000000000000 in ?? (
>>
>> **errp already points to a VirtFSFeatureBlocksMigration error;
>> v9fs_attach() has been called a second time (the first time,
>> understandably, on mount; the second on ls).
>>
> 
> Why are we calling attach a second time ?. I am also not able to reproduce this
> 
> root@qemu-img-64:~# mount -t 9p -otrans=virtio,version=9p2000.L v_tmp /mnt
> root@qemu-img-64:~# ls /mnt/a.c
> /mnt/a.c
> 

I'm just doing ls /mnt (even tab completion: ls /mn<TAB> crashes qemu).

-- 
error compiling committee.c: too many arguments to function

Re: [Qemu-devel] 9p broken?

[Aneesh Kumar K.V]

Avi Kivity <avi@redhat.com> writes:

> On 07/31/2012 09:51 AM, Aneesh Kumar K.V wrote:
>> Avi Kivity <avi@redhat.com> writes:
>> 
>>> Having an annoying bug on i386 kvm I decided to debug it buy running an
>>> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
>>> using nested kvm.
>>>
>>> However, 9p appears to be broken: first, the configure test fails (patch
>>> sent).  Second, while mount works, ls on the mount point causes qemu to
>>> crash with
>> 
>> I missed that you have already sent a patch for configure fix. That
>> looks better that what i sent. I will ack that patch 
>> 
>>>
>>> (gdb) bt
>>> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
>>> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
>>> /home/tlv/akivity/qemu/error.c:32
>>> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
>>> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
>>> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
>>> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
>>> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
>>> #4  0x00007fffffffce00 in ?? ()
>>> #5  0x0000000000000000 in ?? (
>>>
>>> **errp already points to a VirtFSFeatureBlocksMigration error;
>>> v9fs_attach() has been called a second time (the first time,
>>> understandably, on mount; the second on ls).
>>>
>> 
>> Why are we calling attach a second time ?. I am also not able to reproduce this
>> 
>> root@qemu-img-64:~# mount -t 9p -otrans=virtio,version=9p2000.L v_tmp /mnt
>> root@qemu-img-64:~# ls /mnt/a.c
>> /mnt/a.c
>> 
>
> I'm just doing ls /mnt (even tab completion: ls /mn<TAB> crashes qemu).
>

Did this help ?

http://mid.gmane.org/1343719453-26768-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com

-aneesh

Re: [Qemu-devel] 9p broken?

[Avi Kivity]

On 07/31/2012 04:30 PM, Aneesh Kumar K.V wrote:

> 
> Did this help ?
> 
> http://mid.gmane.org/1343719453-26768-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com
> 

It did: thanks.

-- 
error compiling committee.c: too many arguments to function

Re: [Qemu-devel] 9p broken?

[Aneesh Kumar K.V]

Avi Kivity <avi@redhat.com> writes:

> Having an annoying bug on i386 kvm I decided to debug it buy running an
> i386 guest on my x86_64 host, use 9p to access a guest image, and run it
> using nested kvm.
>
> However, 9p appears to be broken: first, the configure test fails (patch
> sent).  Second, while mount works, ls on the mount point causes qemu to
> crash with
>
> (gdb) bt
> #0  error_set (errp=0x7fffe95fb128, fmt=0x5555558d4568 "{ 'class':
> 'VirtFSFeatureBlocksMigration', 'data': { 'path': %s, 'tag': %s } }") at
> /home/tlv/akivity/qemu/error.c:32
> #1  0x000055555567cb06 in v9fs_attach (opaque=0x7fffe95e3020) at
> /home/tlv/akivity/qemu/hw/9pfs/virtio-9p.c:988
> #2  0x000055555561d19f in coroutine_trampoline (i0=1449767888, i1=21845)
> at /home/tlv/akivity/qemu/coroutine-ucontext.c:138
> #3  0x00007ffff5a93ef0 in ?? ()	from /lib64/libc.so.6
> #4  0x00007fffffffce00 in ?? ()
> #5  0x0000000000000000 in ?? (
>
> **errp already points to a VirtFSFeatureBlocksMigration error;
> v9fs_attach() has been called a second time (the first time,
> understandably, on mount; the second on ls).
>

Ok we mount with fid uid = (unsigned)~0. So we need to attach again when
we do ls as root. I guess we can do that migration block in attach, or
we have to do it conditionally. We also can hit the same if we do a
path lookup as a different user.

-aneesh