[Devel] Re: [PATCH v2] SUNRPC: check current nsproxy before set of node name on client creation

[Stanislav Kinsbursky]

10.09.2012 19:41, Myklebust, Trond пишет:
> On Mon, 2012-09-10 at 19:37 +0400, Stanislav Kinsbursky wrote:
>> Hi, Trond.
>> So, if I understand you right, we can create rpc client (or increase usage
>> counter) on NSMPROC_MON call and destroy (or decrease usage counter) on
>> NSMPROC_UNMON call.
>> Will this solution works?
>
> The rpc client(s) will need to be per-net-namespace, which complicates
> matters a little bit, but yes, creation at NSMPROC_MON, and destruction
> at NSMPROC_UNMON should work.
>

Hi, Trond.
With reference-counted NSM client I got this:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
IP: [<ffffffffa00c0d7f>] encode_mon_id+0x2e/0x64 [lockd]
PGD 0
Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
Modules linked in: nfsv3 nfs_acl nfs lockd veth sunrpc bridge stp llc i2c_piix4 
i2c_core virtio_blk virtio_net floppy virtio_pci virtio_ring virtio
CPU 1
Pid: 1174, comm: bash Not tainted 3.5.0-kitchensink+ #44 Bochs Bochs
RIP: 0010:[<ffffffffa00c0d7f>]  [<ffffffffa00c0d7f>] encode_mon_id+0x2e/0x64 [lockd]
RSP: 0018:ffff88001d0f1a08  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff88001d0f1c38 RCX: 0000000000000000
RDX: ffff88001c85803f RSI: ffff88001d23b204 RDI: ffff88001d0f1a48
RBP: ffff88001d0f1a18 R08: ffff88001c858040 R09: 0000000000000003
R10: ffff88001a5aba10 R11: ffff88001d0f1ac8 R12: ffff88001d0f1a48
R13: ffff88001a8a3138 R14: ffffffffa008d580 R15: ffffffffa00c0db5
FS:  00007f0d60eea700(0000) GS:ffff88001f700000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
CR2: 0000000000000008 CR3: 000000001db3d000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process bash (pid: 1174, threadinfo ffff88001d0f0000, task ffff88001d1f4160)
Stack:
  ffff88001d0f1a48 ffff88001c858030 ffff88001d0f1a28 ffffffffa00c0dc9
  ffff88001d0f1ab8 ffffffffa00731a0 ffff88001a5aba10 ffff88001d0f1c38
  ffff88001c858040 ffff88001a8a3140 ffff88001c858854 ffff88001a8a3140
Call Trace:
  [<ffffffffa00c0dc9>] nsm_xdr_enc_unmon+0x14/0x16 [lockd]
  [<ffffffffa00731a0>] rpcauth_wrap_req+0xa1/0xb2 [sunrpc]
  [<ffffffffa006b83f>] ? xprt_prepare_transmit+0x83/0x93 [sunrpc]
  [<ffffffffa0068e54>] call_transmit+0x1e4/0x263 [sunrpc]
  [<ffffffffa00728e2>] __rpc_execute+0xe7/0x313 [sunrpc]
  [<ffffffffa0068c70>] ? call_transmit_status+0xb8/0xb8 [sunrpc]
  [<ffffffff81055ed9>] ? wake_up_bit+0x25/0x2a
  [<ffffffffa0072be0>] rpc_execute+0x9d/0xa5 [sunrpc]
  [<ffffffffa006a6ae>] rpc_run_task+0x7e/0x86 [sunrpc]
  [<ffffffffa006a7a3>] rpc_call_sync+0x44/0x65 [sunrpc]
  [<ffffffffa00c0883>] nsm_mon_unmon+0x81/0xad [lockd]
  [<ffffffffa00c0931>] nsm_unmonitor+0x82/0x13a [lockd]
  [<ffffffffa00bc251>] nlm_destroy_host_locked+0x93/0xc9 [lockd]
  [<ffffffffa00bc33a>] nlmclnt_release_host+0xb3/0xc3 [lockd]
  [<ffffffffa00ba09f>] nlmclnt_done+0x1a/0x26 [lockd]
  [<ffffffffa00d583e>] nfs_destroy_server+0x24/0x26 [nfs]
  [<ffffffffa00d5d85>] nfs_free_server+0xad/0x134 [nfs]
  [<ffffffffa00dd5ff>] nfs_kill_super+0x22/0x26 [nfs]
  [<ffffffff8112b278>] deactivate_locked_super+0x26/0x57
  [<ffffffff8112bd89>] deactivate_super+0x45/0x4c
  [<ffffffff811423ec>] mntput_no_expire+0x110/0x119
  [<ffffffff8114241f>] mntput+0x2a/0x2c
  [<ffffffff81142605>] release_mounts+0x72/0x84
  [<ffffffff811427cf>] put_mnt_ns+0x6f/0x7e
  [<ffffffff8105a3db>] free_nsproxy+0x1f/0x87
  [<ffffffff8105a49f>] switch_task_namespaces+0x5c/0x65
  [<ffffffff8105a4b8>] exit_task_namespaces+0x10/0x12
  [<ffffffff8103c232>] do_exit+0x69b/0x84f
  [<ffffffff8103c469>] do_group_exit+0x83/0xae
  [<ffffffff8103c4ab>] sys_exit_group+0x17/0x1b
  [<ffffffff814657e9>] system_call_fastpath+0x16/0x1b
Code: e5 41 54 53 66 66 66 66 90 48 89 f3 49 89 fc 48 8b 76 18 e8 93 ff ff ff 4c 
89 e7 65 48 8b 04 25 c0 b9 00 00 48 8b 80 00 05 00 00 <48> 8b 70 08 48 83 c6 45 
e8 73 ff ff ff 4c 89 e7 be 0c 00 00 00
RIP  [<ffffffffa00c0d7f>] encode_mon_id+0x2e/0x64 [lockd]


There are more places, where NFS and Lockd code dereference utsname().
In XDR layer, for instance.

So, probably, we have to tie NFS to utsns as well as to netns.
Add one more ns to xprt? What do you think?

-- 
Best regards,
Stanislav Kinsbursky