Chuck Lever wrote:
Hey Ricardo-
On Nov 4, 2008, at 9:40 PM, Labiaga, Ricardo wrote:
Hi Tom,
The thinking is that NFSRDMA could initially use a TCP callback channel.
We'll implement BIND_CONN_TO_SESSION so that the backchannel does not
need to be tied to the forechannel connection. This should address the
case where you have NFSRDMA for the forechannel and TCP for the
backchannel. BIND_CONN_TO_SESSION is also required to reestablish
dropped connections effectively (to avoid losing the reply cache).
We're first trying to get enough of the backchannel functionality
implemented to make it useful. We can then build on that. Tom Talpey
and I had a brief conversation about what would be needed to implement
the backchannel over NFSRDMA some time ago. We didn't see anything
inherently wrong with the current approach. Tom mentioned there needs
to be some RDMA work during the pre-allocation. It will also need to
inspect the RPC direction to determine if it needs to queue the request
on the callback service, or proceed as usual with the processing of the
request.
Your approach is fine for a prototype. However...
Now we have to have this conversation again when adding any new
transport capability, such as SCTP.
Plus, to support NFSv4.1 over RDMA, you now _require_ the existence of
an IP network between client and server. This may be as minor as
ensuring IPoIB is configured on both ends, but it's still a significant
extra step for a pure SAN over RDMA environment.
FWIW, We already require IPoIB because the NFSRDMA implementation uses
the RDMA CM which is IP address based.
This obviates the
whole point of combining the forward and back channels onto one transport.
In my opinion you are making the developer's life simpler now and
complicating the lives of current users and admins, and future developers.
-----Original Message-----
From: Tom Tucker [mailto:tom@xxxxxxxxxxxxxxxxxxxxx]
Sent: Tuesday, November 04, 2008 4:44 PM
To: Labiaga, Ricardo
Cc: Trond Myklebust; Benny Halevy; NFS list; pNFS Mailing List
Subject: Re: [pnfs] nfs41: sunrpc: handle clnt==NULL in call_status
Ricardo:
I don't mean to throw this whole discussion down a rat-hole, but the
current back channel architecture isn't transport independent, so
NFSRDMA on NFS41 is basically broken because all the back-channel
logic
is in the xprtsock.c (transport specific) file. It also assumes that
the
same socket (not transport, socket) is used for both fore and back
channel communication.
What are the plans long-term for making this consistent with the
existing architecture? Does it have to be the same socket/connection
for
some reason?
Thanks,
Tom
Labiaga, Ricardo wrote:
-----Original Message-----
From: Trond Myklebust [mailto:trond.myklebust@xxxxxxxxxx]
Sent: Tuesday, November 04, 2008 10:19 AM
To: Benny Halevy
Cc: pNFS Mailing List; NFS list
Subject: Re: nfs41: sunrpc: handle clnt==NULL in call_status
On Tue, 2008-11-04 at 17:52 +0200, Benny Halevy wrote:
On the following error path task->tk_client == NULL causing NULL
deref:
Nov 4 14:14:32 tl1 kernel: Callback slot table overflowed
Nov 4 14:14:32 tl1 kernel: BUG: unable to handle kernel NULL
pointer
dereference at 0000000000000078
Nov 4 14:14:32 tl1 kernel: IP: [<ffffffffa02b4dd2>]
call_status+0xe1/0x145 [sunrpc]
Nov 4 14:14:32 tl1 kernel: PGD 2f973067 PUD 2f974067 PMD 0
Nov 4 14:14:32 tl1 kernel: Oops: 0000 [1] SMP
Nov 4 14:14:32 tl1 kernel: CPU 0
Nov 4 14:14:32 tl1 kernel: Modules linked in: panfs(P)
panlayoutdriver
nfslayoutdriver nfsd auth_rpcgss exportfs nfs lockd nfs_acl fuse
sunrpc
ipv6 cpufreq_ondemand powernow_k8 freq_table dm_mirror dm_log
dm_multipath
dm_mod snd_hda_intel snd_seq_dummy sr_mod snd_seq_oss cdrom
snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss
snd_pcm snd_timer snd_page_alloc snd_hwdep snd pcspkr serio_raw
k8temp
hwmon forcedeth soundcore pata_amd sg jedec_probe cfi_probe
gen_probe
mtd
i2c_nforce2 chipreg i2c_core button map_funcs sata_nv ata_generic
pata_acpi libata sd_mod scsi_mod ext3 jbd mbcache uhci_hcd ohci_hcd
ehci_hcd [last unloaded: scsi_wait_scan]
Nov 4 14:14:32 tl1 kernel: Pid: 3405, comm: nfsv4.1-svc Tainted:
P
2.6.27-pnfs #1
Nov 4 14:14:32 tl1 kernel: RIP: 0010:[<ffffffffa02b4dd2>]
[<ffffffffa02b4dd2>] call_status+0xe1/0x145 [sunrpc]
Nov 4 14:14:32 tl1 kernel: RSP: 0018:ffff88002a70bd60 EFLAGS:
00010246
Nov 4 14:14:32 tl1 kernel: RAX: 0000000000000000 RBX:
00000000ffffff95
RCX: 0000000000000000
Nov 4 14:14:32 tl1 kernel: RDX: ffff88002a70bd10 RSI:
ffff88002fc52dc0
RDI: ffff88002fc52dc0
Nov 4 14:14:32 tl1 kernel: RBP: ffff88002a70bd80 R08:
ffff88002f9dae40
R09: ffffffff81489e60
Nov 4 14:14:32 tl1 kernel: R10: ffff880038497800 R11:
ffff88002a70bd30
R12: ffff88002fc52dc0
Nov 4 14:14:32 tl1 kernel: R13: 0000000000000000 R14:
ffff88002f57fa00
R15: ffff88002fc52e50
Nov 4 14:14:32 tl1 kernel: FS: 00007fe78ffac6f0(0000)
GS:ffffffff8148da80(0000) knlGS:0000000000000000
Nov 4 14:14:32 tl1 kernel: CS: 0010 DS: 0018 ES: 0018 CR0:
000000008005003b
Nov 4 14:14:32 tl1 kernel: CR2: 0000000000000078 CR3:
000000002f971000
CR4: 00000000000006e0
Nov 4 14:14:32 tl1 kernel: DR0: 0000000000000000 DR1:
0000000000000000
DR2: 0000000000000000
Nov 4 14:14:32 tl1 kernel: DR3: 0000000000000000 DR6:
00000000ffff0ff0
DR7: 0000000000000400
Nov 4 14:14:32 tl1 kernel: Process nfsv4.1-svc (pid: 3405,
threadinfo
ffff88002a70a000, task ffff88003694d900)
Nov 4 14:14:32 tl1 kernel: Stack: 0000000000000000
ffff88002f57fa00
ffff88002fc52dc0 ffff880030450ce8
Nov 4 14:14:32 tl1 kernel: ffff88002a70bdc0 ffffffffa02bb9db
ffff88002a70bdf0 ffff88002fc52dc0
Nov 4 14:14:32 tl1 kernel: ffff88002f57fa00 ffff88002f57fa00
ffff880030450ce8 ffff88002a734000
Nov 4 14:14:32 tl1 kernel: Call Trace:
Nov 4 14:14:32 tl1 kernel: [<ffffffffa02bb9db>]
__rpc_execute+0x7e/0x21f [sunrpc]
Nov 4 14:14:32 tl1 kernel: [<ffffffffa02bbba4>]
rpc_execute+0x28/0x2c
[sunrpc]
Nov 4 14:14:32 tl1 kernel: [<ffffffffa02b5596>]
rpc_run_bc_task+0x88/0xae [sunrpc]
Nov 4 14:14:32 tl1 kernel: [<ffffffffa02c923d>] bc_send+0x35/0x5a
[sunrpc]
Nov 4 14:14:32 tl1 kernel: [<ffffffffa02bdc92>]
bc_svc_process+0xf6/0x101 [sunrpc]
Nov 4 14:14:32 tl1 kernel: [<ffffffff81048a6c>] ?
prepare_to_wait+0x5c/0x65
Nov 4 14:14:32 tl1 kernel: [<ffffffffa033a4c9>]
nfs41_callback_svc+0xda/0x131 [nfs]
Nov 4 14:14:32 tl1 kernel: [<ffffffff810488a0>] ?
autoremove_wake_function+0x0/0x38
Nov 4 14:14:32 tl1 kernel: [<ffffffffa033a3ef>] ?
nfs41_callback_svc+0x0/0x131 [nfs]
Nov 4 14:14:32 tl1 kernel: [<ffffffff81048561>] kthread+0x49/0x76
Nov 4 14:14:32 tl1 kernel: [<ffffffff8100cf49>]
child_rip+0xa/0x11
Nov 4 14:14:32 tl1 kernel: [<ffffffff8102a401>] ?
pick_next_task_fair+0x93/0xa3
Nov 4 14:14:32 tl1 kernel: [<ffffffff81048518>] ?
kthread+0x0/0x76
Nov 4 14:14:32 tl1 kernel: [<ffffffff8100cf3f>] ?
child_rip+0x0/0x11
Nov 4 14:14:32 tl1 kernel:
Nov 4 14:14:32 tl1 kernel:
Nov 4 14:14:32 tl1 kernel: Code: e8 7c 70 00 00 49 8b 44 24 20 49
c7 44
24 68 dc 48 2b a0 f6 40 78 02 74 6d 41 8b b6 34 01 00 00 48 8b 78
30
e8 b9
2a 00 00 eb 5b <41> f6 45 78 04 74 13 49 8b 55 30 48 8d 82 80 04 00
00
f0
80 a2
Nov 4 14:14:32 tl1 kernel: RIP [<ffffffffa02b4dd2>]
call_status+0xe1/0x145 [sunrpc]
Nov 4 14:14:32 tl1 kernel: RSP <ffff88002a70bd60>
Nov 4 14:14:32 tl1 kernel: CR2: 0000000000000078
Nov 4 14:14:32 tl1 kernel: ---[ end trace 1584c489f2573aaf ]---
Signed-off-by: Benny Halevy <bhalevy@xxxxxxxxxxx>
---
Trond, I'm not sure if this can happen without nfs41.
However, please consider this patch for upstream since
it is safe to do in any case.
Benny
net/sunrpc/clnt.c | 8 +++++---
1 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 78fc483..b555d9f 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1206,7 +1206,8 @@ call_status(struct rpc_task *task)
break;
case -ECONNREFUSED:
case -ENOTCONN:
- rpc_force_rebind(clnt);
+ if (clnt)
+ rpc_force_rebind(clnt);
task->tk_action = call_bind;
break;
case -EAGAIN:
@@ -1217,9 +1218,10 @@ call_status(struct rpc_task *task)
rpc_exit(task, status);
break;
default:
- if (clnt->cl_chatty)
+ if (!clnt || clnt->cl_chatty)
printk("%s: RPC call returned error %d\n",
- clnt->cl_protname, -status);
+ clnt ? clnt->cl_protname : "<unknown
protocol>",
+ -status);
rpc_exit(task, status);
}
}
BIG NACK!
How does even it make sense for a task to get past call_transmit
and
call_status without having task->tk_client set? This sounds like
serious
borkenness in the nfsv4.1 patches...
The "Callback slot table overflowed" message means we couldn't
obtain a
pre-allocated rpc_rqst to process the callback. When this occurs,
the
client sets the XPRT_CLOSE_WAIT bit to close the connection. (As an
aside, should we be using xprt_force_disconnect() instead?).
This leads me to believe the server exceeded the number of
outstanding
requests allowed on the backchannel (slot count of one at the
moment),
the new request caused us to close the connection and pulled the rug
from under the callback service. I'll investigate further.
Is this reproducible?
- ricardo
Trond
...
--
To unsubscribe from this list: send the line "unsubscribe
linux-nfs"
in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
_______________________________________________
pNFS mailing list
pNFS@xxxxxxxxxxxxx
http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
