Return-Path: Received: from aserp1040.oracle.com ([141.146.126.69]:45538 "EHLO aserp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S2993341AbbEEPPG convert rfc822-to-8bit (ORCPT ); Tue, 5 May 2015 11:15:06 -0400 Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 (Mac OS X Mail 7.3 \(1878.6\)) Subject: Re: [PATCH v1 01/14] xprtrdma: Transport fault injection From: Chuck Lever In-Reply-To: <5548D741.7090803@Netapp.com> Date: Tue, 5 May 2015 11:15:18 -0400 Cc: linux-rdma@vger.kernel.org, Linux NFS Mailing List Message-Id: <2098B4A5-48C7-4458-BAC6-10F64359C405@oracle.com> References: <20150504174626.3483.97639.stgit@manet.1015granger.net> <20150504175651.3483.35554.stgit@manet.1015granger.net> <5548CA82.9060903@Netapp.com> <06C797AA-4E45-49DA-8CA8-2B8D22B818DB@oracle.com> <5548D741.7090803@Netapp.com> To: Anna Schumaker Sender: linux-nfs-owner@vger.kernel.org List-ID: On May 5, 2015, at 10:44 AM, Anna Schumaker wrote: > On 05/05/2015 09:53 AM, Chuck Lever wrote: >> >> On May 5, 2015, at 9:49 AM, Anna Schumaker wrote: >> >>> Hi Chuck, >>> >>> Neat idea! Are servers able to handle client recovery without getting too confused? >> >> So far I have encountered only issues on the client side. I think this >> is because the client is the active part of re-establishing transport >> connections. In addition, RPC/RDMA clients have a bunch of resources >> that need to be reset after a transport disconnect. >> >> I think this idea can be translated into something that can be done >> in the generic layer (ie, xprt.c) if people think that would be of >> benefit for testing TCP also. > > It might, and now is the time to discuss it before we're stuck maintaining multiple interfaces to the same thing. > > Another thought: can you move this under debugfs instead of proc? That's where the other kernel fault injection controls are, and it might give us a little more flexibility if we need to change the interface later. Something like /sys/kernel/debug/sunrpc/inject_transport_fault ? > > Anna >> >> >>> Anna >>> >>> On 05/04/2015 01:56 PM, Chuck Lever wrote: >>>> It has been exceptionally useful to exercise the logic that handles >>>> local immediate errors and RDMA connection loss. To enable >>>> developers to test this regularly and repeatably, add logic to >>>> simulate connection loss every so often. >>>> >>>> Fault injection is disabled by default. It is enabled with >>>> >>>> $ sudo echo xxx > /proc/sys/sunrpc/rdma_inject_transport_fault >>>> >>>> where "xxx" is a large positive number of transport method calls >>>> before a disconnect. A value of several thousand is usually a good >>>> number that allows reasonable forward progress while still causing a >>>> lot of connection drops. >>>> >>>> Signed-off-by: Chuck Lever >>>> --- >>>> net/sunrpc/Kconfig | 12 ++++++++++++ >>>> net/sunrpc/xprtrdma/transport.c | 34 ++++++++++++++++++++++++++++++++++ >>>> net/sunrpc/xprtrdma/xprt_rdma.h | 1 + >>>> 3 files changed, 47 insertions(+) >>>> >>>> diff --git a/net/sunrpc/Kconfig b/net/sunrpc/Kconfig >>>> index 9068e72..329f82c 100644 >>>> --- a/net/sunrpc/Kconfig >>>> +++ b/net/sunrpc/Kconfig >>>> @@ -61,6 +61,18 @@ config SUNRPC_XPRT_RDMA_CLIENT >>>> >>>> If unsure, say N. >>>> >>>> +config SUNRPC_XPRT_RDMA_FAULT_INJECTION >>>> + bool "RPC over RDMA client fault injection" >>>> + depends on SUNRPC_XPRT_RDMA_CLIENT >>>> + default N >>>> + help >>>> + This option enables fault injection in the xprtrdma module. >>>> + Fault injection is disabled by default. It is enabled with: >>>> + >>>> + $ sudo echo xxx > /proc/sys/sunrpc/rdma_inject_fault >>>> + >>>> + If unsure, say N. >>>> + >>>> config SUNRPC_XPRT_RDMA_SERVER >>>> tristate "RPC over RDMA Server Support" >>>> depends on SUNRPC && INFINIBAND && INFINIBAND_ADDR_TRANS >>>> diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c >>>> index 54f23b1..fdcb2c7 100644 >>>> --- a/net/sunrpc/xprtrdma/transport.c >>>> +++ b/net/sunrpc/xprtrdma/transport.c >>>> @@ -74,6 +74,7 @@ static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE; >>>> static unsigned int xprt_rdma_inline_write_padding; >>>> static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR; >>>> int xprt_rdma_pad_optimize = 1; >>>> +static unsigned int xprt_rdma_inject_transport_fault; >>>> >>>> #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) >>>> >>>> @@ -135,6 +136,13 @@ static struct ctl_table xr_tunables_table[] = { >>>> .mode = 0644, >>>> .proc_handler = proc_dointvec, >>>> }, >>>> + { >>>> + .procname = "rdma_inject_transport_fault", >>>> + .data = &xprt_rdma_inject_transport_fault, >>>> + .maxlen = sizeof(unsigned int), >>>> + .mode = 0644, >>>> + .proc_handler = proc_dointvec, >>>> + }, >>>> { }, >>>> }; >>>> >>>> @@ -246,6 +254,27 @@ xprt_rdma_connect_worker(struct work_struct *work) >>>> xprt_clear_connecting(xprt); >>>> } >>>> >>>> +#if defined CONFIG_SUNRPC_XPRT_RDMA_FAULT_INJECTION >>>> +static void >>>> +xprt_rdma_inject_disconnect(struct rpcrdma_xprt *r_xprt) >>>> +{ >>>> + if (!xprt_rdma_inject_transport_fault) >>>> + return; >>>> + >>>> + if (atomic_dec_return(&r_xprt->rx_inject_count) == 0) { >>>> + atomic_set(&r_xprt->rx_inject_count, >>>> + xprt_rdma_inject_transport_fault); >>>> + pr_info("rpcrdma: injecting transport disconnect\n"); >>>> + (void)rdma_disconnect(r_xprt->rx_ia.ri_id); >>>> + } >>>> +} >>>> +#else >>>> +static void >>>> +xprt_rdma_inject_disconnect(struct rpcrdma_xprt *r_xprt) >>>> +{ >>>> +} >>>> +#endif >>>> + >>>> /* >>>> * xprt_rdma_destroy >>>> * >>>> @@ -405,6 +434,8 @@ xprt_setup_rdma(struct xprt_create *args) >>>> INIT_DELAYED_WORK(&new_xprt->rx_connect_worker, >>>> xprt_rdma_connect_worker); >>>> >>>> + atomic_set(&new_xprt->rx_inject_count, >>>> + xprt_rdma_inject_transport_fault); >>>> xprt_rdma_format_addresses(xprt); >>>> xprt->max_payload = new_xprt->rx_ia.ri_ops->ro_maxpages(new_xprt); >>>> if (xprt->max_payload == 0) >>>> @@ -515,6 +546,7 @@ xprt_rdma_allocate(struct rpc_task *task, size_t size) >>>> out: >>>> dprintk("RPC: %s: size %zd, request 0x%p\n", __func__, size, req); >>>> req->rl_connect_cookie = 0; /* our reserved value */ >>>> + xprt_rdma_inject_disconnect(r_xprt); >>>> return req->rl_sendbuf->rg_base; >>>> >>>> out_rdmabuf: >>>> @@ -589,6 +621,7 @@ xprt_rdma_free(void *buffer) >>>> } >>>> >>>> rpcrdma_buffer_put(req); >>>> + xprt_rdma_inject_disconnect(r_xprt); >>>> } >>>> >>>> /* >>>> @@ -634,6 +667,7 @@ xprt_rdma_send_request(struct rpc_task *task) >>>> >>>> rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len; >>>> rqst->rq_bytes_sent = 0; >>>> + xprt_rdma_inject_disconnect(r_xprt); >>>> return 0; >>>> >>>> failed_marshal: >>>> diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h >>>> index 78e0b8b..08aee53 100644 >>>> --- a/net/sunrpc/xprtrdma/xprt_rdma.h >>>> +++ b/net/sunrpc/xprtrdma/xprt_rdma.h >>>> @@ -377,6 +377,7 @@ struct rpcrdma_xprt { >>>> struct rpcrdma_create_data_internal rx_data; >>>> struct delayed_work rx_connect_worker; >>>> struct rpcrdma_stats rx_stats; >>>> + atomic_t rx_inject_count; >>>> }; >>>> >>>> #define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt) >>>> >>>> -- >>>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in >>>> the body of a message to majordomo@vger.kernel.org >>>> More majordomo info at http://vger.kernel.org/majordomo-info.html >>>> >>> >>> -- >>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in >>> the body of a message to majordomo@vger.kernel.org >>> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> -- >> Chuck Lever >> chuck[dot]lever[at]oracle[dot]com >> >> >> > -- Chuck Lever chuck[dot]lever[at]oracle[dot]com