Message ID | 20140421220214.12569.23157.stgit@manet.1015granger.net (mailing list archive) |
---|---|
State | Not Applicable, archived |
Headers | show |
On Mon, Apr 21, 2014 at 06:02:14PM -0400, Chuck Lever wrote: > ALLPHYSICAL is not a safe memory registration mode because it > permits NFS servers to write anywhere in a client's memory. NFS > server bugs could result in client memory being overwritten. > > This can be useful for embedded systems which do not support more > surgical RDMA memory registration and protection methods, or for > bring-up of new HCA hardware. > > However, enterprise Linux distributions have expressed a desire to > disable it in production environments. It's just as unsafe in embedded devices. I think it should go -- 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
On 4/22/2014 9:23 AM, Christoph Hellwig wrote: > On Mon, Apr 21, 2014 at 06:02:14PM -0400, Chuck Lever wrote: >> ALLPHYSICAL is not a safe memory registration mode because it >> permits NFS servers to write anywhere in a client's memory. NFS >> server bugs could result in client memory being overwritten. >> >> This can be useful for embedded systems which do not support more >> surgical RDMA memory registration and protection methods, or for >> bring-up of new HCA hardware. >> >> However, enterprise Linux distributions have expressed a desire to >> disable it in production environments. > It's just as unsafe in embedded devices. I think it should go For small IOs pattern, ALLPHYSICAL should outperform any registration method in terms of IOP rate (simply because it doesn't do it). Generally speaking, deployments that may prefer higher IOP rate in the cost of a security do exist out there... Sagi. -- 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
diff --git a/include/linux/sunrpc/xprtrdma.h b/include/linux/sunrpc/xprtrdma.h index c2f04e1..64a0a0a 100644 --- a/include/linux/sunrpc/xprtrdma.h +++ b/include/linux/sunrpc/xprtrdma.h @@ -62,8 +62,6 @@ #define RPCRDMA_INLINE_PAD_THRESH (512)/* payload threshold to pad (bytes) */ /* memory registration strategies */ -#define RPCRDMA_PERSISTENT_REGISTRATION (1) - enum rpcrdma_memreg { RPCRDMA_BOUNCEBUFFERS = 0, RPCRDMA_REGISTER, diff --git a/net/sunrpc/Kconfig b/net/sunrpc/Kconfig index 0754d0f..c9a736e 100644 --- a/net/sunrpc/Kconfig +++ b/net/sunrpc/Kconfig @@ -58,6 +58,20 @@ config SUNRPC_XPRT_RDMA_CLIENT If unsure, say N. +config SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL + bool "Enable ALLPHYSICAL memory registration mode" + depends on SUNRPC_XPRT_RDMA_CLIENT + default y + help + This option enables support for the ALLPHYSICAL memory + registration mode. + + This mode is very fast but not safe because it registers + and exposes all of local memory. This could allow an + NFS server bug to corrupt client memory. + + If unsure, say Y. + config SUNRPC_XPRT_RDMA_SERVER tristate "RPC over RDMA Server Support" depends on SUNRPC && INFINIBAND && INFINIBAND_ADDR_TRANS diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 35dcd10..9b71896 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -510,7 +510,7 @@ rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg) if (!ia->ri_id->device->alloc_fmr) { dprintk("RPC: %s: MTHCAFMR registration " "not supported by HCA\n", __func__); -#if RPCRDMA_PERSISTENT_REGISTRATION +#ifdef CONFIG_SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL memreg = RPCRDMA_ALLPHYSICAL; #else rc = -ENOMEM; @@ -530,7 +530,7 @@ rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg) switch (memreg) { case RPCRDMA_FRMR: break; -#if RPCRDMA_PERSISTENT_REGISTRATION +#ifdef CONFIG_SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL case RPCRDMA_ALLPHYSICAL: mem_priv = IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | @@ -541,7 +541,7 @@ rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg) if (ia->ri_have_dma_lkey) break; mem_priv = IB_ACCESS_LOCAL_WRITE; -#if RPCRDMA_PERSISTENT_REGISTRATION +#ifdef CONFIG_SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL register_setup: #endif ia->ri_bind_mem = ib_get_dma_mr(ia->ri_pd, mem_priv); @@ -1601,7 +1601,7 @@ rpcrdma_register_external(struct rpcrdma_mr_seg *seg, switch (ia->ri_memreg_strategy) { -#if RPCRDMA_PERSISTENT_REGISTRATION +#ifdef CONFIG_SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL case RPCRDMA_ALLPHYSICAL: rpcrdma_map_one(ia, seg, writing); seg->mr_rkey = ia->ri_bind_mem->rkey; @@ -1639,7 +1639,7 @@ rpcrdma_deregister_external(struct rpcrdma_mr_seg *seg, switch (ia->ri_memreg_strategy) { -#if RPCRDMA_PERSISTENT_REGISTRATION +#ifdef CONFIG_SUNRPC_XPRT_RDMA_CLIENT_ALLPHYSICAL case RPCRDMA_ALLPHYSICAL: BUG_ON(nsegs != 1); rpcrdma_unmap_one(ia, seg);
ALLPHYSICAL is not a safe memory registration mode because it permits NFS servers to write anywhere in a client's memory. NFS server bugs could result in client memory being overwritten. This can be useful for embedded systems which do not support more surgical RDMA memory registration and protection methods, or for bring-up of new HCA hardware. However, enterprise Linux distributions have expressed a desire to disable it in production environments. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> --- include/linux/sunrpc/xprtrdma.h | 2 -- net/sunrpc/Kconfig | 14 ++++++++++++++ net/sunrpc/xprtrdma/verbs.c | 10 +++++----- 3 files changed, 19 insertions(+), 7 deletions(-) -- 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