Some NFSv4.1 OPEN requests were hanging waiting for the NFS server
to finish recalling delegations. Turns out that each NFSv4.1 CB
request on RDMA gets a GARBAGE_ARGS reply from the Linux client.
Commit 756b9b37cfb2e3dc added a line in bc_svc_process that
overwrites the incoming rq_rcv_buf's length with the value in
rq_private_buf.len. But rpcrdma_bc_receive_call() does not invoke
xprt_complete_bc_request(), thus rq_private_buf.len is not
initialized. svc_process_common() is invoked with a zero-length
RPC message, and fails.
Fixes: 756b9b37cfb2e3dc ('SUNRPC: Fix callback channel')
Signed-off-by: Chuck Lever <[email protected]>
---
Hi Anna-
Commit 756b9b37cfb2e3dc in v4.5-rc1 introduced a behavior
regression for NFSv4.1-over-RDMA. Would it be possible to merge
this bug fix into 4.5-rc ?
net/sunrpc/xprtrdma/backchannel.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c
index cc1251d..2dcd764 100644
--- a/net/sunrpc/xprtrdma/backchannel.c
+++ b/net/sunrpc/xprtrdma/backchannel.c
@@ -341,6 +341,8 @@ void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
rqst->rq_reply_bytes_recvd = 0;
rqst->rq_bytes_sent = 0;
rqst->rq_xid = headerp->rm_xid;
+
+ rqst->rq_private_buf.len = size;
set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
buf = &rqst->rq_rcv_buf;
Hi Chuck,
On 02/15/2016 10:23 AM, Chuck Lever wrote:
> Some NFSv4.1 OPEN requests were hanging waiting for the NFS server
> to finish recalling delegations. Turns out that each NFSv4.1 CB
> request on RDMA gets a GARBAGE_ARGS reply from the Linux client.
>
> Commit 756b9b37cfb2e3dc added a line in bc_svc_process that
> overwrites the incoming rq_rcv_buf's length with the value in
> rq_private_buf.len. But rpcrdma_bc_receive_call() does not invoke
> xprt_complete_bc_request(), thus rq_private_buf.len is not
> initialized. svc_process_common() is invoked with a zero-length
> RPC message, and fails.
>
> Fixes: 756b9b37cfb2e3dc ('SUNRPC: Fix callback channel')
> Signed-off-by: Chuck Lever <[email protected]>
> ---
> Hi Anna-
>
> Commit 756b9b37cfb2e3dc in v4.5-rc1 introduced a behavior
> regression for NFSv4.1-over-RDMA. Would it be possible to merge
> this bug fix into 4.5-rc ?
>
>
> net/sunrpc/xprtrdma/backchannel.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c
> index cc1251d..2dcd764 100644
> --- a/net/sunrpc/xprtrdma/backchannel.c
> +++ b/net/sunrpc/xprtrdma/backchannel.c
> @@ -341,6 +341,8 @@ void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
> rqst->rq_reply_bytes_recvd = 0;
> rqst->rq_bytes_sent = 0;
> rqst->rq_xid = headerp->rm_xid;
> +
nit: Any reason for the extra newline?
> + rqst->rq_private_buf.len = size;
This patch looks okay to me, and I'll make sure it's included in the next rc pull request Trond sends.
Thanks,
Anna
> set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
>
> buf = &rqst->rq_rcv_buf;
>
> On Feb 17, 2016, at 10:16 AM, Anna Schumaker <[email protected]> wrote:
>
> Hi Chuck,
>
> On 02/15/2016 10:23 AM, Chuck Lever wrote:
>> Some NFSv4.1 OPEN requests were hanging waiting for the NFS server
>> to finish recalling delegations. Turns out that each NFSv4.1 CB
>> request on RDMA gets a GARBAGE_ARGS reply from the Linux client.
>>
>> Commit 756b9b37cfb2e3dc added a line in bc_svc_process that
>> overwrites the incoming rq_rcv_buf's length with the value in
>> rq_private_buf.len. But rpcrdma_bc_receive_call() does not invoke
>> xprt_complete_bc_request(), thus rq_private_buf.len is not
>> initialized. svc_process_common() is invoked with a zero-length
>> RPC message, and fails.
>>
>> Fixes: 756b9b37cfb2e3dc ('SUNRPC: Fix callback channel')
>> Signed-off-by: Chuck Lever <[email protected]>
>> ---
>> Hi Anna-
>>
>> Commit 756b9b37cfb2e3dc in v4.5-rc1 introduced a behavior
>> regression for NFSv4.1-over-RDMA. Would it be possible to merge
>> this bug fix into 4.5-rc ?
>>
>>
>> net/sunrpc/xprtrdma/backchannel.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c
>> index cc1251d..2dcd764 100644
>> --- a/net/sunrpc/xprtrdma/backchannel.c
>> +++ b/net/sunrpc/xprtrdma/backchannel.c
>> @@ -341,6 +341,8 @@ void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
>> rqst->rq_reply_bytes_recvd = 0;
>> rqst->rq_bytes_sent = 0;
>> rqst->rq_xid = headerp->rm_xid;
>> +
>
> nit: Any reason for the extra newline?
Yes: xprt_complete_bc_request sets the private buffer length
and sets RPC_BC_PA_IN_USE. So I want to group these two
lines in their own paragraph just to "document" that they
belong together and are not related to the lines above
them.
>> + rqst->rq_private_buf.len = size;
>
> This patch looks okay to me, and I'll make sure it's included in the next rc pull request Trond sends.
Thanks!
> Thanks,
> Anna
>
>> set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
>>
>> buf = &rqst->rq_rcv_buf;
--
Chuck Lever
On 02/17/2016 10:19 AM, Chuck Lever wrote:
>
>> On Feb 17, 2016, at 10:16 AM, Anna Schumaker <[email protected]> wrote:
>>
>> Hi Chuck,
>>
>> On 02/15/2016 10:23 AM, Chuck Lever wrote:
>>> Some NFSv4.1 OPEN requests were hanging waiting for the NFS server
>>> to finish recalling delegations. Turns out that each NFSv4.1 CB
>>> request on RDMA gets a GARBAGE_ARGS reply from the Linux client.
>>>
>>> Commit 756b9b37cfb2e3dc added a line in bc_svc_process that
>>> overwrites the incoming rq_rcv_buf's length with the value in
>>> rq_private_buf.len. But rpcrdma_bc_receive_call() does not invoke
>>> xprt_complete_bc_request(), thus rq_private_buf.len is not
>>> initialized. svc_process_common() is invoked with a zero-length
>>> RPC message, and fails.
>>>
>>> Fixes: 756b9b37cfb2e3dc ('SUNRPC: Fix callback channel')
>>> Signed-off-by: Chuck Lever <[email protected]>
>>> ---
>>> Hi Anna-
>>>
>>> Commit 756b9b37cfb2e3dc in v4.5-rc1 introduced a behavior
>>> regression for NFSv4.1-over-RDMA. Would it be possible to merge
>>> this bug fix into 4.5-rc ?
>>>
>>>
>>> net/sunrpc/xprtrdma/backchannel.c | 2 ++
>>> 1 file changed, 2 insertions(+)
>>>
>>> diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c
>>> index cc1251d..2dcd764 100644
>>> --- a/net/sunrpc/xprtrdma/backchannel.c
>>> +++ b/net/sunrpc/xprtrdma/backchannel.c
>>> @@ -341,6 +341,8 @@ void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt,
>>> rqst->rq_reply_bytes_recvd = 0;
>>> rqst->rq_bytes_sent = 0;
>>> rqst->rq_xid = headerp->rm_xid;
>>> +
>>
>> nit: Any reason for the extra newline?
>
> Yes: xprt_complete_bc_request sets the private buffer length
> and sets RPC_BC_PA_IN_USE. So I want to group these two
> lines in their own paragraph just to "document" that they
> belong together and are not related to the lines above
> them.
Makes sense. Thanks!
Anna
>
>
>>> + rqst->rq_private_buf.len = size;
>>
>> This patch looks okay to me, and I'll make sure it's included in the next rc pull request Trond sends.
>
> Thanks!
>
>
>> Thanks,
>> Anna
>>
>>> set_bit(RPC_BC_PA_IN_USE, &rqst->rq_bc_pa_state);
>>>
>>> buf = &rqst->rq_rcv_buf;
>
> --
> Chuck Lever
>
>
>
>