Svcrdma currently advertises 1MB, which is too large. The correct value
is the minimum of RPCSVC_MAXPAYLOAD and the max scatter-gather allowed
in an NFSRDMA IO chunk * the host page size. This bug is usually benign
because the Linux X64 NFSRDMA client correctly limits the payload size to
the correct value (64*4096 = 256KB). But if the Linux client is PPC64
with a 64KB page size, then the client will indeed use a payload size
that will overflow the server.
Signed-off-by: Steve Wise <[email protected]>
---
net/sunrpc/xprtrdma/svc_rdma_transport.c | 2 +-
net/sunrpc/xprtrdma/xprt_rdma.h | 7 +++++++
2 files changed, 8 insertions(+), 1 deletions(-)
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 374feb4..4e61880 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -91,7 +91,7 @@ struct svc_xprt_class svc_rdma_class = {
.xcl_name = "rdma",
.xcl_owner = THIS_MODULE,
.xcl_ops = &svc_rdma_ops,
- .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
+ .xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
.xcl_ident = XPRT_TRANSPORT_RDMA,
};
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index c419498..ac7fc9a 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -51,6 +51,7 @@
#include <linux/sunrpc/clnt.h> /* rpc_xprt */
#include <linux/sunrpc/rpc_rdma.h> /* RPC/RDMA protocol */
#include <linux/sunrpc/xprtrdma.h> /* xprt parameters */
+#include <linux/sunrpc/svc.h> /* RPCSVC_MAXPAYLOAD */
#define RDMA_RESOLVE_TIMEOUT (5000) /* 5 seconds */
#define RDMA_CONNECT_RETRY_MAX (2) /* retries if no listener backlog */
@@ -392,4 +393,10 @@ extern struct kmem_cache *svc_rdma_ctxt_cachep;
/* Workqueue created in svc_rdma.c */
extern struct workqueue_struct *svc_rdma_wq;
+#if RPCSVC_MAXPAYLOAD < (RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT)
+#define RPCSVC_MAXPAYLOAD_RDMA RPCSVC_MAXPAYLOAD
+#else
+#define RPCSVC_MAXPAYLOAD_RDMA (RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT)
+#endif
+
#endif /* _LINUX_SUNRPC_XPRT_RDMA_H */
On 9/24/2014 1:11 AM, Steve Wise wrote:
> Svcrdma currently advertises 1MB, which is too large. The correct value
> is the minimum of RPCSVC_MAXPAYLOAD and the max scatter-gather allowed
> in an NFSRDMA IO chunk * the host page size. This bug is usually benign
> because the Linux X64 NFSRDMA client correctly limits the payload size to
> the correct value (64*4096 = 256KB). But if the Linux client is PPC64
> with a 64KB page size, then the client will indeed use a payload size
> that will overflow the server.
>
Maybe I'm a bit late with this, but can you explain why SG table is
limited to 64 (fastreg MR and page_list probably bound it). Where is it
coming from? Naturally it needs to be bound by some figure, but why 64?
Sorry if I'm nit-picking on this pretty straight-forward bug fix...
Sagi.
On Mon, Sep 29, 2014 at 11:07:25AM -0500, Steve Wise wrote:
>
> Hey Bruce, is this version acceptable for 3.18?
Yes, applying, thanks for the reminder.--b.
On Sep 24, 2014, at 4:55 AM, Sagi Grimberg <[email protected]> wrote:
> On 9/24/2014 1:11 AM, Steve Wise wrote:
>> Svcrdma currently advertises 1MB, which is too large. The correct value
>> is the minimum of RPCSVC_MAXPAYLOAD and the max scatter-gather allowed
>> in an NFSRDMA IO chunk * the host page size. This bug is usually benign
>> because the Linux X64 NFSRDMA client correctly limits the payload size to
>> the correct value (64*4096 = 256KB). But if the Linux client is PPC64
>> with a 64KB page size, then the client will indeed use a payload size
>> that will overflow the server.
>>
>
> Maybe I'm a bit late with this, but can you explain why SG table is
> limited to 64 (fastreg MR and page_list probably bound it). Where is it
> coming from? Naturally it needs to be bound by some figure, but why 64?
It?s arbitrary, AFAIK. It has been used to allocate some data structures
on the stack, for example, so it can?t be too large.
It?s something we need to look at.
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
Hey Bruce, is this version acceptable for 3.18?
Thanks,
Steve.