Fixed MTP to work with TWRP

net/sunrpc/xprtrdma/Makefile

@@ -0,0 +1,8 @@
+obj-$(CONFIG_SUNRPC_XPRT_RDMA_CLIENT) += xprtrdma.o
+
+xprtrdma-y := transport.o rpc_rdma.o verbs.o
+
+obj-$(CONFIG_SUNRPC_XPRT_RDMA_SERVER) += svcrdma.o
+
+svcrdma-y := svc_rdma.o svc_rdma_transport.o \
+	svc_rdma_marshal.o svc_rdma_sendto.o svc_rdma_recvfrom.o

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -0,0 +1,875 @@
+/*
+ * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * rpc_rdma.c
+ *
+ * This file contains the guts of the RPC RDMA protocol, and
+ * does marshaling/unmarshaling, etc. It is also where interfacing
+ * to the Linux RPC framework lives.
+ */
+
+#include "xprt_rdma.h"
+
+#include <linux/highmem.h>
+
+#ifdef RPC_DEBUG
+# define RPCDBG_FACILITY	RPCDBG_TRANS
+#endif
+
+#ifdef RPC_DEBUG
+static const char transfertypes[][12] = {
+	"pure inline",	/* no chunks */
+	" read chunk",	/* some argument via rdma read */
+	"*read chunk",	/* entire request via rdma read */
+	"write chunk",	/* some result via rdma write */
+	"reply chunk"	/* entire reply via rdma write */
+};
+#endif
+
+/*
+ * Chunk assembly from upper layer xdr_buf.
+ *
+ * Prepare the passed-in xdr_buf into representation as RPC/RDMA chunk
+ * elements. Segments are then coalesced when registered, if possible
+ * within the selected memreg mode.
+ *
+ * Returns positive number of segments converted, or a negative errno.
+ */
+
+static int
+rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos,
+	enum rpcrdma_chunktype type, struct rpcrdma_mr_seg *seg, int nsegs)
+{
+	int len, n = 0, p;
+	int page_base;
+	struct page **ppages;
+
+	if (pos == 0 && xdrbuf->head[0].iov_len) {
+		seg[n].mr_page = NULL;
+		seg[n].mr_offset = xdrbuf->head[0].iov_base;
+		seg[n].mr_len = xdrbuf->head[0].iov_len;
+		++n;
+	}
+
+	len = xdrbuf->page_len;
+	ppages = xdrbuf->pages + (xdrbuf->page_base >> PAGE_SHIFT);
+	page_base = xdrbuf->page_base & ~PAGE_MASK;
+	p = 0;
+	while (len && n < nsegs) {
+		if (!ppages[p]) {
+			/* alloc the pagelist for receiving buffer */
+			ppages[p] = alloc_page(GFP_ATOMIC);
+			if (!ppages[p])
+				return -ENOMEM;
+		}
+		seg[n].mr_page = ppages[p];
+		seg[n].mr_offset = (void *)(unsigned long) page_base;
+		seg[n].mr_len = min_t(u32, PAGE_SIZE - page_base, len);
+		if (seg[n].mr_len > PAGE_SIZE)
+			return -EIO;
+		len -= seg[n].mr_len;
+		++n;
+		++p;
+		page_base = 0;	/* page offset only applies to first page */
+	}
+
+	/* Message overflows the seg array */
+	if (len && n == nsegs)
+		return -EIO;
+
+	if (xdrbuf->tail[0].iov_len) {
+		/* the rpcrdma protocol allows us to omit any trailing
+		 * xdr pad bytes, saving the server an RDMA operation. */
+		if (xdrbuf->tail[0].iov_len < 4 && xprt_rdma_pad_optimize)
+			return n;
+		if (n == nsegs)
+			/* Tail remains, but we're out of segments */
+			return -EIO;
+		seg[n].mr_page = NULL;
+		seg[n].mr_offset = xdrbuf->tail[0].iov_base;
+		seg[n].mr_len = xdrbuf->tail[0].iov_len;
+		++n;
+	}
+
+	return n;
+}
+
+/*
+ * Create read/write chunk lists, and reply chunks, for RDMA
+ *
+ *   Assume check against THRESHOLD has been done, and chunks are required.
+ *   Assume only encoding one list entry for read|write chunks. The NFSv3
+ *     protocol is simple enough to allow this as it only has a single "bulk
+ *     result" in each procedure - complicated NFSv4 COMPOUNDs are not. (The
+ *     RDMA/Sessions NFSv4 proposal addresses this for future v4 revs.)
+ *
+ * When used for a single reply chunk (which is a special write
+ * chunk used for the entire reply, rather than just the data), it
+ * is used primarily for READDIR and READLINK which would otherwise
+ * be severely size-limited by a small rdma inline read max. The server
+ * response will come back as an RDMA Write, followed by a message
+ * of type RDMA_NOMSG carrying the xid and length. As a result, reply
+ * chunks do not provide data alignment, however they do not require
+ * "fixup" (moving the response to the upper layer buffer) either.
+ *
+ * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
+ *
+ *  Read chunklist (a linked list):
+ *   N elements, position P (same P for all chunks of same arg!):
+ *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
+ *
+ *  Write chunklist (a list of (one) counted array):
+ *   N elements:
+ *    1 - N - HLOO - HLOO - ... - HLOO - 0
+ *
+ *  Reply chunk (a counted array):
+ *   N elements:
+ *    1 - N - HLOO - HLOO - ... - HLOO
+ *
+ * Returns positive RPC/RDMA header size, or negative errno.
+ */
+
+static ssize_t
+rpcrdma_create_chunks(struct rpc_rqst *rqst, struct xdr_buf *target,
+		struct rpcrdma_msg *headerp, enum rpcrdma_chunktype type)
+{
+	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
+	int n, nsegs, nchunks = 0;
+	unsigned int pos;
+	struct rpcrdma_mr_seg *seg = req->rl_segments;
+	struct rpcrdma_read_chunk *cur_rchunk = NULL;
+	struct rpcrdma_write_array *warray = NULL;
+	struct rpcrdma_write_chunk *cur_wchunk = NULL;
+	__be32 *iptr = headerp->rm_body.rm_chunks;
+
+	if (type == rpcrdma_readch || type == rpcrdma_areadch) {
+		/* a read chunk - server will RDMA Read our memory */
+		cur_rchunk = (struct rpcrdma_read_chunk *) iptr;
+	} else {
+		/* a write or reply chunk - server will RDMA Write our memory */
+		*iptr++ = xdr_zero;	/* encode a NULL read chunk list */
+		if (type == rpcrdma_replych)
+			*iptr++ = xdr_zero;	/* a NULL write chunk list */
+		warray = (struct rpcrdma_write_array *) iptr;
+		cur_wchunk = (struct rpcrdma_write_chunk *) (warray + 1);
+	}
+
+	if (type == rpcrdma_replych || type == rpcrdma_areadch)
+		pos = 0;
+	else
+		pos = target->head[0].iov_len;
+
+	nsegs = rpcrdma_convert_iovs(target, pos, type, seg, RPCRDMA_MAX_SEGS);
+	if (nsegs < 0)
+		return nsegs;
+
+	do {
+		n = rpcrdma_register_external(seg, nsegs,
+						cur_wchunk != NULL, r_xprt);
+		if (n <= 0)
+			goto out;
+		if (cur_rchunk) {	/* read */
+			cur_rchunk->rc_discrim = xdr_one;
+			/* all read chunks have the same "position" */
+			cur_rchunk->rc_position = htonl(pos);
+			cur_rchunk->rc_target.rs_handle = htonl(seg->mr_rkey);
+			cur_rchunk->rc_target.rs_length = htonl(seg->mr_len);
+			xdr_encode_hyper(
+					(__be32 *)&cur_rchunk->rc_target.rs_offset,
+					seg->mr_base);
+			dprintk("RPC:       %s: read chunk "
+				"elem %d@0x%llx:0x%x pos %u (%s)\n", __func__,
+				seg->mr_len, (unsigned long long)seg->mr_base,
+				seg->mr_rkey, pos, n < nsegs ? "more" : "last");
+			cur_rchunk++;
+			r_xprt->rx_stats.read_chunk_count++;
+		} else {		/* write/reply */
+			cur_wchunk->wc_target.rs_handle = htonl(seg->mr_rkey);
+			cur_wchunk->wc_target.rs_length = htonl(seg->mr_len);
+			xdr_encode_hyper(
+					(__be32 *)&cur_wchunk->wc_target.rs_offset,
+					seg->mr_base);
+			dprintk("RPC:       %s: %s chunk "
+				"elem %d@0x%llx:0x%x (%s)\n", __func__,
+				(type == rpcrdma_replych) ? "reply" : "write",
+				seg->mr_len, (unsigned long long)seg->mr_base,
+				seg->mr_rkey, n < nsegs ? "more" : "last");
+			cur_wchunk++;
+			if (type == rpcrdma_replych)
+				r_xprt->rx_stats.reply_chunk_count++;
+			else
+				r_xprt->rx_stats.write_chunk_count++;
+			r_xprt->rx_stats.total_rdma_request += seg->mr_len;
+		}
+		nchunks++;
+		seg   += n;
+		nsegs -= n;
+	} while (nsegs);
+
+	/* success. all failures return above */
+	req->rl_nchunks = nchunks;
+
+	/*
+	 * finish off header. If write, marshal discrim and nchunks.
+	 */
+	if (cur_rchunk) {
+		iptr = (__be32 *) cur_rchunk;
+		*iptr++ = xdr_zero;	/* finish the read chunk list */
+		*iptr++ = xdr_zero;	/* encode a NULL write chunk list */
+		*iptr++ = xdr_zero;	/* encode a NULL reply chunk */
+	} else {
+		warray->wc_discrim = xdr_one;
+		warray->wc_nchunks = htonl(nchunks);
+		iptr = (__be32 *) cur_wchunk;
+		if (type == rpcrdma_writech) {
+			*iptr++ = xdr_zero; /* finish the write chunk list */
+			*iptr++ = xdr_zero; /* encode a NULL reply chunk */
+		}
+	}
+
+	/*
+	 * Return header size.
+	 */
+	return (unsigned char *)iptr - (unsigned char *)headerp;
+
+out:
+	if (r_xprt->rx_ia.ri_memreg_strategy != RPCRDMA_FRMR) {
+		for (pos = 0; nchunks--;)
+			pos += rpcrdma_deregister_external(
+					&req->rl_segments[pos], r_xprt);
+	}
+	return n;
+}
+
+/*
+ * Marshal chunks. This routine returns the header length
+ * consumed by marshaling.
+ *
+ * Returns positive RPC/RDMA header size, or negative errno.
+ */
+
+ssize_t
+rpcrdma_marshal_chunks(struct rpc_rqst *rqst, ssize_t result)
+{
+	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
+	struct rpcrdma_msg *headerp = (struct rpcrdma_msg *)req->rl_base;
+
+	if (req->rl_rtype != rpcrdma_noch)
+		result = rpcrdma_create_chunks(rqst, &rqst->rq_snd_buf,
+					       headerp, req->rl_rtype);
+	else if (req->rl_wtype != rpcrdma_noch)
+		result = rpcrdma_create_chunks(rqst, &rqst->rq_rcv_buf,
+					       headerp, req->rl_wtype);
+	return result;
+}
+
+/*
+ * Copy write data inline.
+ * This function is used for "small" requests. Data which is passed
+ * to RPC via iovecs (or page list) is copied directly into the
+ * pre-registered memory buffer for this request. For small amounts
+ * of data, this is efficient. The cutoff value is tunable.
+ */
+static int
+rpcrdma_inline_pullup(struct rpc_rqst *rqst, int pad)
+{
+	int i, npages, curlen;
+	int copy_len;
+	unsigned char *srcp, *destp;
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
+	int page_base;
+	struct page **ppages;
+
+	destp = rqst->rq_svec[0].iov_base;
+	curlen = rqst->rq_svec[0].iov_len;
+	destp += curlen;
+	/*
+	 * Do optional padding where it makes sense. Alignment of write
+	 * payload can help the server, if our setting is accurate.
+	 */
+	pad -= (curlen + 36/*sizeof(struct rpcrdma_msg_padded)*/);
+	if (pad < 0 || rqst->rq_slen - curlen < RPCRDMA_INLINE_PAD_THRESH)
+		pad = 0;	/* don't pad this request */
+
+	dprintk("RPC:       %s: pad %d destp 0x%p len %d hdrlen %d\n",
+		__func__, pad, destp, rqst->rq_slen, curlen);
+
+	copy_len = rqst->rq_snd_buf.page_len;
+
+	if (rqst->rq_snd_buf.tail[0].iov_len) {
+		curlen = rqst->rq_snd_buf.tail[0].iov_len;
+		if (destp + copy_len != rqst->rq_snd_buf.tail[0].iov_base) {
+			memmove(destp + copy_len,
+				rqst->rq_snd_buf.tail[0].iov_base, curlen);
+			r_xprt->rx_stats.pullup_copy_count += curlen;
+		}
+		dprintk("RPC:       %s: tail destp 0x%p len %d\n",
+			__func__, destp + copy_len, curlen);
+		rqst->rq_svec[0].iov_len += curlen;
+	}
+	r_xprt->rx_stats.pullup_copy_count += copy_len;
+
+	page_base = rqst->rq_snd_buf.page_base;
+	ppages = rqst->rq_snd_buf.pages + (page_base >> PAGE_SHIFT);
+	page_base &= ~PAGE_MASK;
+	npages = PAGE_ALIGN(page_base+copy_len) >> PAGE_SHIFT;
+	for (i = 0; copy_len && i < npages; i++) {
+		curlen = PAGE_SIZE - page_base;
+		if (curlen > copy_len)
+			curlen = copy_len;
+		dprintk("RPC:       %s: page %d destp 0x%p len %d curlen %d\n",
+			__func__, i, destp, copy_len, curlen);
+		srcp = kmap_atomic(ppages[i]);
+		memcpy(destp, srcp+page_base, curlen);
+		kunmap_atomic(srcp);
+		rqst->rq_svec[0].iov_len += curlen;
+		destp += curlen;
+		copy_len -= curlen;
+		page_base = 0;
+	}
+	/* header now contains entire send message */
+	return pad;
+}
+
+/*
+ * Marshal a request: the primary job of this routine is to choose
+ * the transfer modes. See comments below.
+ *
+ * Uses multiple RDMA IOVs for a request:
+ *  [0] -- RPC RDMA header, which uses memory from the *start* of the
+ *         preregistered buffer that already holds the RPC data in
+ *         its middle.
+ *  [1] -- the RPC header/data, marshaled by RPC and the NFS protocol.
+ *  [2] -- optional padding.
+ *  [3] -- if padded, header only in [1] and data here.
+ *
+ * Returns zero on success, otherwise a negative errno.
+ */
+
+int
+rpcrdma_marshal_req(struct rpc_rqst *rqst)
+{
+	struct rpc_xprt *xprt = rqst->rq_xprt;
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
+	char *base;
+	size_t rpclen, padlen;
+	ssize_t hdrlen;
+	struct rpcrdma_msg *headerp;
+
+	/*
+	 * rpclen gets amount of data in first buffer, which is the
+	 * pre-registered buffer.
+	 */
+	base = rqst->rq_svec[0].iov_base;
+	rpclen = rqst->rq_svec[0].iov_len;
+
+	/* build RDMA header in private area at front */
+	headerp = (struct rpcrdma_msg *) req->rl_base;
+	/* don't htonl XID, it's already done in request */
+	headerp->rm_xid = rqst->rq_xid;
+	headerp->rm_vers = xdr_one;
+	headerp->rm_credit = htonl(r_xprt->rx_buf.rb_max_requests);
+	headerp->rm_type = htonl(RDMA_MSG);
+
+	/*
+	 * Chunks needed for results?
+	 *
+	 * o If the expected result is under the inline threshold, all ops
+	 *   return as inline (but see later).
+	 * o Large non-read ops return as a single reply chunk.
+	 * o Large read ops return data as write chunk(s), header as inline.
+	 *
+	 * Note: the NFS code sending down multiple result segments implies
+	 * the op is one of read, readdir[plus], readlink or NFSv4 getacl.
+	 */
+
+	/*
+	 * This code can handle read chunks, write chunks OR reply
+	 * chunks -- only one type. If the request is too big to fit
+	 * inline, then we will choose read chunks. If the request is
+	 * a READ, then use write chunks to separate the file data
+	 * into pages; otherwise use reply chunks.
+	 */
+	if (rqst->rq_rcv_buf.buflen <= RPCRDMA_INLINE_READ_THRESHOLD(rqst))
+		req->rl_wtype = rpcrdma_noch;
+	else if (rqst->rq_rcv_buf.page_len == 0)
+		req->rl_wtype = rpcrdma_replych;
+	else if (rqst->rq_rcv_buf.flags & XDRBUF_READ)
+		req->rl_wtype = rpcrdma_writech;
+	else
+		req->rl_wtype = rpcrdma_replych;
+
+	/*
+	 * Chunks needed for arguments?
+	 *
+	 * o If the total request is under the inline threshold, all ops
+	 *   are sent as inline.
+	 * o Large non-write ops are sent with the entire message as a
+	 *   single read chunk (protocol 0-position special case).
+	 * o Large write ops transmit data as read chunk(s), header as
+	 *   inline.
+	 *
+	 * Note: the NFS code sending down multiple argument segments
+	 * implies the op is a write.
+	 * TBD check NFSv4 setacl
+	 */
+	if (rqst->rq_snd_buf.len <= RPCRDMA_INLINE_WRITE_THRESHOLD(rqst))
+		req->rl_rtype = rpcrdma_noch;
+	else if (rqst->rq_snd_buf.page_len == 0)
+		req->rl_rtype = rpcrdma_areadch;
+	else
+		req->rl_rtype = rpcrdma_readch;
+
+	/* The following simplification is not true forever */
+	if (req->rl_rtype != rpcrdma_noch && req->rl_wtype == rpcrdma_replych)
+		req->rl_wtype = rpcrdma_noch;
+	if (req->rl_rtype != rpcrdma_noch && req->rl_wtype != rpcrdma_noch) {
+		dprintk("RPC:       %s: cannot marshal multiple chunk lists\n",
+			__func__);
+		return -EIO;
+	}
+
+	hdrlen = 28; /*sizeof *headerp;*/
+	padlen = 0;
+
+	/*
+	 * Pull up any extra send data into the preregistered buffer.
+	 * When padding is in use and applies to the transfer, insert
+	 * it and change the message type.
+	 */
+	if (req->rl_rtype == rpcrdma_noch) {
+
+		padlen = rpcrdma_inline_pullup(rqst,
+						RPCRDMA_INLINE_PAD_VALUE(rqst));
+
+		if (padlen) {
+			headerp->rm_type = htonl(RDMA_MSGP);
+			headerp->rm_body.rm_padded.rm_align =
+				htonl(RPCRDMA_INLINE_PAD_VALUE(rqst));
+			headerp->rm_body.rm_padded.rm_thresh =
+				htonl(RPCRDMA_INLINE_PAD_THRESH);
+			headerp->rm_body.rm_padded.rm_pempty[0] = xdr_zero;
+			headerp->rm_body.rm_padded.rm_pempty[1] = xdr_zero;
+			headerp->rm_body.rm_padded.rm_pempty[2] = xdr_zero;
+			hdrlen += 2 * sizeof(u32); /* extra words in padhdr */
+			if (req->rl_wtype != rpcrdma_noch) {
+				dprintk("RPC:       %s: invalid chunk list\n",
+					__func__);
+				return -EIO;
+			}
+		} else {
+			headerp->rm_body.rm_nochunks.rm_empty[0] = xdr_zero;
+			headerp->rm_body.rm_nochunks.rm_empty[1] = xdr_zero;
+			headerp->rm_body.rm_nochunks.rm_empty[2] = xdr_zero;
+			/* new length after pullup */
+			rpclen = rqst->rq_svec[0].iov_len;
+			/*
+			 * Currently we try to not actually use read inline.
+			 * Reply chunks have the desirable property that
+			 * they land, packed, directly in the target buffers
+			 * without headers, so they require no fixup. The
+			 * additional RDMA Write op sends the same amount
+			 * of data, streams on-the-wire and adds no overhead
+			 * on receive. Therefore, we request a reply chunk
+			 * for non-writes wherever feasible and efficient.
+			 */
+			if (req->rl_wtype == rpcrdma_noch)
+				req->rl_wtype = rpcrdma_replych;
+		}
+	}
+
+	hdrlen = rpcrdma_marshal_chunks(rqst, hdrlen);
+	if (hdrlen < 0)
+		return hdrlen;
+
+	dprintk("RPC:       %s: %s: hdrlen %zd rpclen %zd padlen %zd"
+		" headerp 0x%p base 0x%p lkey 0x%x\n",
+		__func__, transfertypes[req->rl_wtype], hdrlen, rpclen, padlen,
+		headerp, base, req->rl_iov.lkey);
+
+	/*
+	 * initialize send_iov's - normally only two: rdma chunk header and
+	 * single preregistered RPC header buffer, but if padding is present,
+	 * then use a preregistered (and zeroed) pad buffer between the RPC
+	 * header and any write data. In all non-rdma cases, any following
+	 * data has been copied into the RPC header buffer.
+	 */
+	req->rl_send_iov[0].addr = req->rl_iov.addr;
+	req->rl_send_iov[0].length = hdrlen;
+	req->rl_send_iov[0].lkey = req->rl_iov.lkey;
+
+	req->rl_send_iov[1].addr = req->rl_iov.addr + (base - req->rl_base);
+	req->rl_send_iov[1].length = rpclen;
+	req->rl_send_iov[1].lkey = req->rl_iov.lkey;
+
+	req->rl_niovs = 2;
+
+	if (padlen) {
+		struct rpcrdma_ep *ep = &r_xprt->rx_ep;
+
+		req->rl_send_iov[2].addr = ep->rep_pad.addr;
+		req->rl_send_iov[2].length = padlen;
+		req->rl_send_iov[2].lkey = ep->rep_pad.lkey;
+
+		req->rl_send_iov[3].addr = req->rl_send_iov[1].addr + rpclen;
+		req->rl_send_iov[3].length = rqst->rq_slen - rpclen;
+		req->rl_send_iov[3].lkey = req->rl_iov.lkey;
+
+		req->rl_niovs = 4;
+	}
+
+	return 0;
+}
+
+/*
+ * Chase down a received write or reply chunklist to get length
+ * RDMA'd by server. See map at rpcrdma_create_chunks()! :-)
+ */
+static int
+rpcrdma_count_chunks(struct rpcrdma_rep *rep, unsigned int max, int wrchunk, __be32 **iptrp)
+{
+	unsigned int i, total_len;
+	struct rpcrdma_write_chunk *cur_wchunk;
+
+	i = ntohl(**iptrp);	/* get array count */
+	if (i > max)
+		return -1;
+	cur_wchunk = (struct rpcrdma_write_chunk *) (*iptrp + 1);
+	total_len = 0;
+	while (i--) {
+		struct rpcrdma_segment *seg = &cur_wchunk->wc_target;
+		ifdebug(FACILITY) {
+			u64 off;
+			xdr_decode_hyper((__be32 *)&seg->rs_offset, &off);
+			dprintk("RPC:       %s: chunk %d@0x%llx:0x%x\n",
+				__func__,
+				ntohl(seg->rs_length),
+				(unsigned long long)off,
+				ntohl(seg->rs_handle));
+		}
+		total_len += ntohl(seg->rs_length);
+		++cur_wchunk;
+	}
+	/* check and adjust for properly terminated write chunk */
+	if (wrchunk) {
+		__be32 *w = (__be32 *) cur_wchunk;
+		if (*w++ != xdr_zero)
+			return -1;
+		cur_wchunk = (struct rpcrdma_write_chunk *) w;
+	}
+	if ((char *) cur_wchunk > rep->rr_base + rep->rr_len)
+		return -1;
+
+	*iptrp = (__be32 *) cur_wchunk;
+	return total_len;
+}
+
+/*
+ * Scatter inline received data back into provided iov's.
+ */
+static void
+rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
+{
+	int i, npages, curlen, olen;
+	char *destp;
+	struct page **ppages;
+	int page_base;
+
+	curlen = rqst->rq_rcv_buf.head[0].iov_len;
+	if (curlen > copy_len) {	/* write chunk header fixup */
+		curlen = copy_len;
+		rqst->rq_rcv_buf.head[0].iov_len = curlen;
+	}
+
+	dprintk("RPC:       %s: srcp 0x%p len %d hdrlen %d\n",
+		__func__, srcp, copy_len, curlen);
+
+	/* Shift pointer for first receive segment only */
+	rqst->rq_rcv_buf.head[0].iov_base = srcp;
+	srcp += curlen;
+	copy_len -= curlen;
+
+	olen = copy_len;
+	i = 0;
+	rpcx_to_rdmax(rqst->rq_xprt)->rx_stats.fixup_copy_count += olen;
+	page_base = rqst->rq_rcv_buf.page_base;
+	ppages = rqst->rq_rcv_buf.pages + (page_base >> PAGE_SHIFT);
+	page_base &= ~PAGE_MASK;
+
+	if (copy_len && rqst->rq_rcv_buf.page_len) {
+		npages = PAGE_ALIGN(page_base +
+			rqst->rq_rcv_buf.page_len) >> PAGE_SHIFT;
+		for (; i < npages; i++) {
+			curlen = PAGE_SIZE - page_base;
+			if (curlen > copy_len)
+				curlen = copy_len;
+			dprintk("RPC:       %s: page %d"
+				" srcp 0x%p len %d curlen %d\n",
+				__func__, i, srcp, copy_len, curlen);
+			destp = kmap_atomic(ppages[i]);
+			memcpy(destp + page_base, srcp, curlen);
+			flush_dcache_page(ppages[i]);
+			kunmap_atomic(destp);
+			srcp += curlen;
+			copy_len -= curlen;
+			if (copy_len == 0)
+				break;
+			page_base = 0;
+		}
+	}
+
+	if (copy_len && rqst->rq_rcv_buf.tail[0].iov_len) {
+		curlen = copy_len;
+		if (curlen > rqst->rq_rcv_buf.tail[0].iov_len)
+			curlen = rqst->rq_rcv_buf.tail[0].iov_len;
+		if (rqst->rq_rcv_buf.tail[0].iov_base != srcp)
+			memmove(rqst->rq_rcv_buf.tail[0].iov_base, srcp, curlen);
+		dprintk("RPC:       %s: tail srcp 0x%p len %d curlen %d\n",
+			__func__, srcp, copy_len, curlen);
+		rqst->rq_rcv_buf.tail[0].iov_len = curlen;
+		copy_len -= curlen; ++i;
+	} else
+		rqst->rq_rcv_buf.tail[0].iov_len = 0;
+
+	if (pad) {
+		/* implicit padding on terminal chunk */
+		unsigned char *p = rqst->rq_rcv_buf.tail[0].iov_base;
+		while (pad--)
+			p[rqst->rq_rcv_buf.tail[0].iov_len++] = 0;
+	}
+
+	if (copy_len)
+		dprintk("RPC:       %s: %d bytes in"
+			" %d extra segments (%d lost)\n",
+			__func__, olen, i, copy_len);
+
+	/* TBD avoid a warning from call_decode() */
+	rqst->rq_private_buf = rqst->rq_rcv_buf;
+}
+
+void
+rpcrdma_connect_worker(struct work_struct *work)
+{
+	struct rpcrdma_ep *ep =
+		container_of(work, struct rpcrdma_ep, rep_connect_worker.work);
+	struct rpc_xprt *xprt = ep->rep_xprt;
+
+	spin_lock_bh(&xprt->transport_lock);
+	if (++xprt->connect_cookie == 0)	/* maintain a reserved value */
+		++xprt->connect_cookie;
+	if (ep->rep_connected > 0) {
+		if (!xprt_test_and_set_connected(xprt))
+			xprt_wake_pending_tasks(xprt, 0);
+	} else {
+		if (xprt_test_and_clear_connected(xprt))
+			xprt_wake_pending_tasks(xprt, -ENOTCONN);
+	}
+	spin_unlock_bh(&xprt->transport_lock);
+}
+
+/*
+ * This function is called when an async event is posted to
+ * the connection which changes the connection state. All it
+ * does at this point is mark the connection up/down, the rpc
+ * timers do the rest.
+ */
+void
+rpcrdma_conn_func(struct rpcrdma_ep *ep)
+{
+	schedule_delayed_work(&ep->rep_connect_worker, 0);
+}
+
+/*
+ * Called as a tasklet to do req/reply match and complete a request
+ * Errors must result in the RPC task either being awakened, or
+ * allowed to timeout, to discover the errors at that time.
+ */
+void
+rpcrdma_reply_handler(struct rpcrdma_rep *rep)
+{
+	struct rpcrdma_msg *headerp;
+	struct rpcrdma_req *req;
+	struct rpc_rqst *rqst;
+	struct rpc_xprt *xprt = rep->rr_xprt;
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	__be32 *iptr;
+	int rdmalen, status;
+	unsigned long cwnd;
+
+	/* Check status. If bad, signal disconnect and return rep to pool */
+	if (rep->rr_len == ~0U) {
+		rpcrdma_recv_buffer_put(rep);
+		if (r_xprt->rx_ep.rep_connected == 1) {
+			r_xprt->rx_ep.rep_connected = -EIO;
+			rpcrdma_conn_func(&r_xprt->rx_ep);
+		}
+		return;
+	}
+	if (rep->rr_len < 28) {
+		dprintk("RPC:       %s: short/invalid reply\n", __func__);
+		goto repost;
+	}
+	headerp = (struct rpcrdma_msg *) rep->rr_base;
+	if (headerp->rm_vers != xdr_one) {
+		dprintk("RPC:       %s: invalid version %d\n",
+			__func__, ntohl(headerp->rm_vers));
+		goto repost;
+	}
+
+	/* Get XID and try for a match. */
+	spin_lock(&xprt->transport_lock);
+	rqst = xprt_lookup_rqst(xprt, headerp->rm_xid);
+	if (rqst == NULL) {
+		spin_unlock(&xprt->transport_lock);
+		dprintk("RPC:       %s: reply 0x%p failed "
+			"to match any request xid 0x%08x len %d\n",
+			__func__, rep, headerp->rm_xid, rep->rr_len);
+repost:
+		r_xprt->rx_stats.bad_reply_count++;
+		rep->rr_func = rpcrdma_reply_handler;
+		if (rpcrdma_ep_post_recv(&r_xprt->rx_ia, &r_xprt->rx_ep, rep))
+			rpcrdma_recv_buffer_put(rep);
+
+		return;
+	}
+
+	/* get request object */
+	req = rpcr_to_rdmar(rqst);
+	if (req->rl_reply) {
+		spin_unlock(&xprt->transport_lock);
+		dprintk("RPC:       %s: duplicate reply 0x%p to RPC "
+			"request 0x%p: xid 0x%08x\n", __func__, rep, req,
+			headerp->rm_xid);
+		goto repost;
+	}
+
+	dprintk("RPC:       %s: reply 0x%p completes request 0x%p\n"
+		"                   RPC request 0x%p xid 0x%08x\n",
+			__func__, rep, req, rqst, headerp->rm_xid);
+
+	/* from here on, the reply is no longer an orphan */
+	req->rl_reply = rep;
+	xprt->reestablish_timeout = 0;
+
+	/* check for expected message types */
+	/* The order of some of these tests is important. */
+	switch (headerp->rm_type) {
+	case htonl(RDMA_MSG):
+		/* never expect read chunks */
+		/* never expect reply chunks (two ways to check) */
+		/* never expect write chunks without having offered RDMA */
+		if (headerp->rm_body.rm_chunks[0] != xdr_zero ||
+		    (headerp->rm_body.rm_chunks[1] == xdr_zero &&
+		     headerp->rm_body.rm_chunks[2] != xdr_zero) ||
+		    (headerp->rm_body.rm_chunks[1] != xdr_zero &&
+		     req->rl_nchunks == 0))
+			goto badheader;
+		if (headerp->rm_body.rm_chunks[1] != xdr_zero) {
+			/* count any expected write chunks in read reply */
+			/* start at write chunk array count */
+			iptr = &headerp->rm_body.rm_chunks[2];
+			rdmalen = rpcrdma_count_chunks(rep,
+						req->rl_nchunks, 1, &iptr);
+			/* check for validity, and no reply chunk after */
+			if (rdmalen < 0 || *iptr++ != xdr_zero)
+				goto badheader;
+			rep->rr_len -=
+			    ((unsigned char *)iptr - (unsigned char *)headerp);
+			status = rep->rr_len + rdmalen;
+			r_xprt->rx_stats.total_rdma_reply += rdmalen;
+			/* special case - last chunk may omit padding */
+			if (rdmalen &= 3) {
+				rdmalen = 4 - rdmalen;
+				status += rdmalen;
+			}
+		} else {
+			/* else ordinary inline */
+			rdmalen = 0;
+			iptr = (__be32 *)((unsigned char *)headerp + 28);
+			rep->rr_len -= 28; /*sizeof *headerp;*/
+			status = rep->rr_len;
+		}
+		/* Fix up the rpc results for upper layer */
+		rpcrdma_inline_fixup(rqst, (char *)iptr, rep->rr_len, rdmalen);
+		break;
+
+	case htonl(RDMA_NOMSG):
+		/* never expect read or write chunks, always reply chunks */
+		if (headerp->rm_body.rm_chunks[0] != xdr_zero ||
+		    headerp->rm_body.rm_chunks[1] != xdr_zero ||
+		    headerp->rm_body.rm_chunks[2] != xdr_one ||
+		    req->rl_nchunks == 0)
+			goto badheader;
+		iptr = (__be32 *)((unsigned char *)headerp + 28);
+		rdmalen = rpcrdma_count_chunks(rep, req->rl_nchunks, 0, &iptr);
+		if (rdmalen < 0)
+			goto badheader;
+		r_xprt->rx_stats.total_rdma_reply += rdmalen;
+		/* Reply chunk buffer already is the reply vector - no fixup. */
+		status = rdmalen;
+		break;
+
+badheader:
+	default:
+		dprintk("%s: invalid rpcrdma reply header (type %d):"
+				" chunks[012] == %d %d %d"
+				" expected chunks <= %d\n",
+				__func__, ntohl(headerp->rm_type),
+				headerp->rm_body.rm_chunks[0],
+				headerp->rm_body.rm_chunks[1],
+				headerp->rm_body.rm_chunks[2],
+				req->rl_nchunks);
+		status = -EIO;
+		r_xprt->rx_stats.bad_reply_count++;
+		break;
+	}
+
+	cwnd = xprt->cwnd;
+	xprt->cwnd = atomic_read(&r_xprt->rx_buf.rb_credits) << RPC_CWNDSHIFT;
+	if (xprt->cwnd > cwnd)
+		xprt_release_rqst_cong(rqst->rq_task);
+
+	dprintk("RPC:       %s: xprt_complete_rqst(0x%p, 0x%p, %d)\n",
+			__func__, xprt, rqst, status);
+	xprt_complete_rqst(rqst->rq_task, status);
+	spin_unlock(&xprt->transport_lock);
+}

net/sunrpc/xprtrdma/svc_rdma.c

@@ -0,0 +1,302 @@
+/*
+ * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Author: Tom Tucker <tom@opengridcomputing.com>
+ */
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/fs.h>
+#include <linux/sysctl.h>
+#include <linux/workqueue.h>
+#include <linux/sunrpc/clnt.h>
+#include <linux/sunrpc/sched.h>
+#include <linux/sunrpc/svc_rdma.h>
+#include "xprt_rdma.h"
+
+#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
+
+/* RPC/RDMA parameters */
+unsigned int svcrdma_ord = RPCRDMA_ORD;
+static unsigned int min_ord = 1;
+static unsigned int max_ord = 4096;
+unsigned int svcrdma_max_requests = RPCRDMA_MAX_REQUESTS;
+static unsigned int min_max_requests = 4;
+static unsigned int max_max_requests = 16384;
+unsigned int svcrdma_max_req_size = RPCRDMA_MAX_REQ_SIZE;
+static unsigned int min_max_inline = 4096;
+static unsigned int max_max_inline = 65536;
+
+atomic_t rdma_stat_recv;
+atomic_t rdma_stat_read;
+atomic_t rdma_stat_write;
+atomic_t rdma_stat_sq_starve;
+atomic_t rdma_stat_rq_starve;
+atomic_t rdma_stat_rq_poll;
+atomic_t rdma_stat_rq_prod;
+atomic_t rdma_stat_sq_poll;
+atomic_t rdma_stat_sq_prod;
+
+/* Temporary NFS request map and context caches */
+struct kmem_cache *svc_rdma_map_cachep;
+struct kmem_cache *svc_rdma_ctxt_cachep;
+
+struct workqueue_struct *svc_rdma_wq;
+
+/*
+ * This function implements reading and resetting an atomic_t stat
+ * variable through read/write to a proc file. Any write to the file
+ * resets the associated statistic to zero. Any read returns it's
+ * current value.
+ */
+static int read_reset_stat(struct ctl_table *table, int write,
+			   void __user *buffer, size_t *lenp,
+			   loff_t *ppos)
+{
+	atomic_t *stat = (atomic_t *)table->data;
+
+	if (!stat)
+		return -EINVAL;
+
+	if (write)
+		atomic_set(stat, 0);
+	else {
+		char str_buf[32];
+		char *data;
+		int len = snprintf(str_buf, 32, "%d\n", atomic_read(stat));
+		if (len >= 32)
+			return -EFAULT;
+		len = strlen(str_buf);
+		if (*ppos > len) {
+			*lenp = 0;
+			return 0;
+		}
+		data = &str_buf[*ppos];
+		len -= *ppos;
+		if (len > *lenp)
+			len = *lenp;
+		if (len && copy_to_user(buffer, str_buf, len))
+			return -EFAULT;
+		*lenp = len;
+		*ppos += len;
+	}
+	return 0;
+}
+
+static struct ctl_table_header *svcrdma_table_header;
+static struct ctl_table svcrdma_parm_table[] = {
+	{
+		.procname	= "max_requests",
+		.data		= &svcrdma_max_requests,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_max_requests,
+		.extra2		= &max_max_requests
+	},
+	{
+		.procname	= "max_req_size",
+		.data		= &svcrdma_max_req_size,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_max_inline,
+		.extra2		= &max_max_inline
+	},
+	{
+		.procname	= "max_outbound_read_requests",
+		.data		= &svcrdma_ord,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_ord,
+		.extra2		= &max_ord,
+	},
+
+	{
+		.procname	= "rdma_stat_read",
+		.data		= &rdma_stat_read,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_recv",
+		.data		= &rdma_stat_recv,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_write",
+		.data		= &rdma_stat_write,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_sq_starve",
+		.data		= &rdma_stat_sq_starve,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_rq_starve",
+		.data		= &rdma_stat_rq_starve,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_rq_poll",
+		.data		= &rdma_stat_rq_poll,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_rq_prod",
+		.data		= &rdma_stat_rq_prod,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_sq_poll",
+		.data		= &rdma_stat_sq_poll,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{
+		.procname	= "rdma_stat_sq_prod",
+		.data		= &rdma_stat_sq_prod,
+		.maxlen		= sizeof(atomic_t),
+		.mode		= 0644,
+		.proc_handler	= read_reset_stat,
+	},
+	{ },
+};
+
+static struct ctl_table svcrdma_table[] = {
+	{
+		.procname	= "svc_rdma",
+		.mode		= 0555,
+		.child		= svcrdma_parm_table
+	},
+	{ },
+};
+
+static struct ctl_table svcrdma_root_table[] = {
+	{
+		.procname	= "sunrpc",
+		.mode		= 0555,
+		.child		= svcrdma_table
+	},
+	{ },
+};
+
+void svc_rdma_cleanup(void)
+{
+	dprintk("SVCRDMA Module Removed, deregister RPC RDMA transport\n");
+	destroy_workqueue(svc_rdma_wq);
+	if (svcrdma_table_header) {
+		unregister_sysctl_table(svcrdma_table_header);
+		svcrdma_table_header = NULL;
+	}
+	svc_unreg_xprt_class(&svc_rdma_class);
+	kmem_cache_destroy(svc_rdma_map_cachep);
+	kmem_cache_destroy(svc_rdma_ctxt_cachep);
+}
+
+int svc_rdma_init(void)
+{
+	dprintk("SVCRDMA Module Init, register RPC RDMA transport\n");
+	dprintk("\tsvcrdma_ord      : %d\n", svcrdma_ord);
+	dprintk("\tmax_requests     : %d\n", svcrdma_max_requests);
+	dprintk("\tsq_depth         : %d\n",
+		svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT);
+	dprintk("\tmax_inline       : %d\n", svcrdma_max_req_size);
+
+	svc_rdma_wq = alloc_workqueue("svc_rdma", 0, 0);
+	if (!svc_rdma_wq)
+		return -ENOMEM;
+
+	if (!svcrdma_table_header)
+		svcrdma_table_header =
+			register_sysctl_table(svcrdma_root_table);
+
+	/* Create the temporary map cache */
+	svc_rdma_map_cachep = kmem_cache_create("svc_rdma_map_cache",
+						sizeof(struct svc_rdma_req_map),
+						0,
+						SLAB_HWCACHE_ALIGN,
+						NULL);
+	if (!svc_rdma_map_cachep) {
+		printk(KERN_INFO "Could not allocate map cache.\n");
+		goto err0;
+	}
+
+	/* Create the temporary context cache */
+	svc_rdma_ctxt_cachep =
+		kmem_cache_create("svc_rdma_ctxt_cache",
+				  sizeof(struct svc_rdma_op_ctxt),
+				  0,
+				  SLAB_HWCACHE_ALIGN,
+				  NULL);
+	if (!svc_rdma_ctxt_cachep) {
+		printk(KERN_INFO "Could not allocate WR ctxt cache.\n");
+		goto err1;
+	}
+
+	/* Register RDMA with the SVC transport switch */
+	svc_reg_xprt_class(&svc_rdma_class);
+	return 0;
+ err1:
+	kmem_cache_destroy(svc_rdma_map_cachep);
+ err0:
+	unregister_sysctl_table(svcrdma_table_header);
+	destroy_workqueue(svc_rdma_wq);
+	return -ENOMEM;
+}
+MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
+MODULE_DESCRIPTION("SVC RDMA Transport");
+MODULE_LICENSE("Dual BSD/GPL");
+module_init(svc_rdma_init);
+module_exit(svc_rdma_cleanup);

net/sunrpc/xprtrdma/svc_rdma_marshal.c

@@ -0,0 +1,386 @@
+/*
+ * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Author: Tom Tucker <tom@opengridcomputing.com>
+ */
+
+#include <linux/sunrpc/xdr.h>
+#include <linux/sunrpc/debug.h>
+#include <asm/unaligned.h>
+#include <linux/sunrpc/rpc_rdma.h>
+#include <linux/sunrpc/svc_rdma.h>
+
+#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
+
+/*
+ * Decodes a read chunk list. The expected format is as follows:
+ *    descrim  : xdr_one
+ *    position : u32 offset into XDR stream
+ *    handle   : u32 RKEY
+ *    . . .
+ *  end-of-list: xdr_zero
+ */
+static u32 *decode_read_list(u32 *va, u32 *vaend)
+{
+	struct rpcrdma_read_chunk *ch = (struct rpcrdma_read_chunk *)va;
+
+	while (ch->rc_discrim != xdr_zero) {
+		if (((unsigned long)ch + sizeof(struct rpcrdma_read_chunk)) >
+		    (unsigned long)vaend) {
+			dprintk("svcrdma: vaend=%p, ch=%p\n", vaend, ch);
+			return NULL;
+		}
+		ch++;
+	}
+	return (u32 *)&ch->rc_position;
+}
+
+/*
+ * Determine number of chunks and total bytes in chunk list. The chunk
+ * list has already been verified to fit within the RPCRDMA header.
+ */
+void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *ch,
+			       int *ch_count, int *byte_count)
+{
+	/* compute the number of bytes represented by read chunks */
+	*byte_count = 0;
+	*ch_count = 0;
+	for (; ch->rc_discrim != 0; ch++) {
+		*byte_count = *byte_count + ntohl(ch->rc_target.rs_length);
+		*ch_count = *ch_count + 1;
+	}
+}
+
+/*
+ * Decodes a write chunk list. The expected format is as follows:
+ *    descrim  : xdr_one
+ *    nchunks  : <count>
+ *       handle   : u32 RKEY              ---+
+ *       length   : u32 <len of segment>     |
+ *       offset   : remove va                + <count>
+ *       . . .                               |
+ *                                        ---+
+ */
+static u32 *decode_write_list(u32 *va, u32 *vaend)
+{
+	unsigned long start, end;
+	int nchunks;
+
+	struct rpcrdma_write_array *ary =
+		(struct rpcrdma_write_array *)va;
+
+	/* Check for not write-array */
+	if (ary->wc_discrim == xdr_zero)
+		return (u32 *)&ary->wc_nchunks;
+
+	if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
+	    (unsigned long)vaend) {
+		dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
+		return NULL;
+	}
+	nchunks = ntohl(ary->wc_nchunks);
+
+	start = (unsigned long)&ary->wc_array[0];
+	end = (unsigned long)vaend;
+	if (nchunks < 0 ||
+	    nchunks > (SIZE_MAX - start) / sizeof(struct rpcrdma_write_chunk) ||
+	    (start + (sizeof(struct rpcrdma_write_chunk) * nchunks)) > end) {
+		dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n",
+			ary, nchunks, vaend);
+		return NULL;
+	}
+	/*
+	 * rs_length is the 2nd 4B field in wc_target and taking its
+	 * address skips the list terminator
+	 */
+	return (u32 *)&ary->wc_array[nchunks].wc_target.rs_length;
+}
+
+static u32 *decode_reply_array(u32 *va, u32 *vaend)
+{
+	unsigned long start, end;
+	int nchunks;
+	struct rpcrdma_write_array *ary =
+		(struct rpcrdma_write_array *)va;
+
+	/* Check for no reply-array */
+	if (ary->wc_discrim == xdr_zero)
+		return (u32 *)&ary->wc_nchunks;
+
+	if ((unsigned long)ary + sizeof(struct rpcrdma_write_array) >
+	    (unsigned long)vaend) {
+		dprintk("svcrdma: ary=%p, vaend=%p\n", ary, vaend);
+		return NULL;
+	}
+	nchunks = ntohl(ary->wc_nchunks);
+
+	start = (unsigned long)&ary->wc_array[0];
+	end = (unsigned long)vaend;
+	if (nchunks < 0 ||
+	    nchunks > (SIZE_MAX - start) / sizeof(struct rpcrdma_write_chunk) ||
+	    (start + (sizeof(struct rpcrdma_write_chunk) * nchunks)) > end) {
+		dprintk("svcrdma: ary=%p, wc_nchunks=%d, vaend=%p\n",
+			ary, nchunks, vaend);
+		return NULL;
+	}
+	return (u32 *)&ary->wc_array[nchunks];
+}
+
+int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req,
+			    struct svc_rqst *rqstp)
+{
+	struct rpcrdma_msg *rmsgp = NULL;
+	u32 *va;
+	u32 *vaend;
+	u32 hdr_len;
+
+	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
+
+	/* Verify that there's enough bytes for header + something */
+	if (rqstp->rq_arg.len <= RPCRDMA_HDRLEN_MIN) {
+		dprintk("svcrdma: header too short = %d\n",
+			rqstp->rq_arg.len);
+		return -EINVAL;
+	}
+
+	/* Decode the header */
+	rmsgp->rm_xid = ntohl(rmsgp->rm_xid);
+	rmsgp->rm_vers = ntohl(rmsgp->rm_vers);
+	rmsgp->rm_credit = ntohl(rmsgp->rm_credit);
+	rmsgp->rm_type = ntohl(rmsgp->rm_type);
+
+	if (rmsgp->rm_vers != RPCRDMA_VERSION)
+		return -ENOSYS;
+
+	/* Pull in the extra for the padded case and bump our pointer */
+	if (rmsgp->rm_type == RDMA_MSGP) {
+		int hdrlen;
+		rmsgp->rm_body.rm_padded.rm_align =
+			ntohl(rmsgp->rm_body.rm_padded.rm_align);
+		rmsgp->rm_body.rm_padded.rm_thresh =
+			ntohl(rmsgp->rm_body.rm_padded.rm_thresh);
+
+		va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
+		rqstp->rq_arg.head[0].iov_base = va;
+		hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp);
+		rqstp->rq_arg.head[0].iov_len -= hdrlen;
+		if (hdrlen > rqstp->rq_arg.len)
+			return -EINVAL;
+		return hdrlen;
+	}
+
+	/* The chunk list may contain either a read chunk list or a write
+	 * chunk list and a reply chunk list.
+	 */
+	va = &rmsgp->rm_body.rm_chunks[0];
+	vaend = (u32 *)((unsigned long)rmsgp + rqstp->rq_arg.len);
+	va = decode_read_list(va, vaend);
+	if (!va)
+		return -EINVAL;
+	va = decode_write_list(va, vaend);
+	if (!va)
+		return -EINVAL;
+	va = decode_reply_array(va, vaend);
+	if (!va)
+		return -EINVAL;
+
+	rqstp->rq_arg.head[0].iov_base = va;
+	hdr_len = (unsigned long)va - (unsigned long)rmsgp;
+	rqstp->rq_arg.head[0].iov_len -= hdr_len;
+
+	*rdma_req = rmsgp;
+	return hdr_len;
+}
+
+int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *rqstp)
+{
+	struct rpcrdma_msg *rmsgp = NULL;
+	struct rpcrdma_read_chunk *ch;
+	struct rpcrdma_write_array *ary;
+	u32 *va;
+	u32 hdrlen;
+
+	dprintk("svcrdma: processing deferred RDMA header on rqstp=%p\n",
+		rqstp);
+	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
+
+	/* Pull in the extra for the padded case and bump our pointer */
+	if (rmsgp->rm_type == RDMA_MSGP) {
+		va = &rmsgp->rm_body.rm_padded.rm_pempty[4];
+		rqstp->rq_arg.head[0].iov_base = va;
+		hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp);
+		rqstp->rq_arg.head[0].iov_len -= hdrlen;
+		return hdrlen;
+	}
+
+	/*
+	 * Skip all chunks to find RPC msg. These were previously processed
+	 */
+	va = &rmsgp->rm_body.rm_chunks[0];
+
+	/* Skip read-list */
+	for (ch = (struct rpcrdma_read_chunk *)va;
+	     ch->rc_discrim != xdr_zero; ch++);
+	va = (u32 *)&ch->rc_position;
+
+	/* Skip write-list */
+	ary = (struct rpcrdma_write_array *)va;
+	if (ary->wc_discrim == xdr_zero)
+		va = (u32 *)&ary->wc_nchunks;
+	else
+		/*
+		 * rs_length is the 2nd 4B field in wc_target and taking its
+		 * address skips the list terminator
+		 */
+		va = (u32 *)&ary->wc_array[ary->wc_nchunks].wc_target.rs_length;
+
+	/* Skip reply-array */
+	ary = (struct rpcrdma_write_array *)va;
+	if (ary->wc_discrim == xdr_zero)
+		va = (u32 *)&ary->wc_nchunks;
+	else
+		va = (u32 *)&ary->wc_array[ary->wc_nchunks];
+
+	rqstp->rq_arg.head[0].iov_base = va;
+	hdrlen = (unsigned long)va - (unsigned long)rmsgp;
+	rqstp->rq_arg.head[0].iov_len -= hdrlen;
+
+	return hdrlen;
+}
+
+int svc_rdma_xdr_encode_error(struct svcxprt_rdma *xprt,
+			      struct rpcrdma_msg *rmsgp,
+			      enum rpcrdma_errcode err, u32 *va)
+{
+	u32 *startp = va;
+
+	*va++ = htonl(rmsgp->rm_xid);
+	*va++ = htonl(rmsgp->rm_vers);
+	*va++ = htonl(xprt->sc_max_requests);
+	*va++ = htonl(RDMA_ERROR);
+	*va++ = htonl(err);
+	if (err == ERR_VERS) {
+		*va++ = htonl(RPCRDMA_VERSION);
+		*va++ = htonl(RPCRDMA_VERSION);
+	}
+
+	return (int)((unsigned long)va - (unsigned long)startp);
+}
+
+int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *rmsgp)
+{
+	struct rpcrdma_write_array *wr_ary;
+
+	/* There is no read-list in a reply */
+
+	/* skip write list */
+	wr_ary = (struct rpcrdma_write_array *)
+		&rmsgp->rm_body.rm_chunks[1];
+	if (wr_ary->wc_discrim)
+		wr_ary = (struct rpcrdma_write_array *)
+			&wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)].
+			wc_target.rs_length;
+	else
+		wr_ary = (struct rpcrdma_write_array *)
+			&wr_ary->wc_nchunks;
+
+	/* skip reply array */
+	if (wr_ary->wc_discrim)
+		wr_ary = (struct rpcrdma_write_array *)
+			&wr_ary->wc_array[ntohl(wr_ary->wc_nchunks)];
+	else
+		wr_ary = (struct rpcrdma_write_array *)
+			&wr_ary->wc_nchunks;
+
+	return (unsigned long) wr_ary - (unsigned long) rmsgp;
+}
+
+void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *rmsgp, int chunks)
+{
+	struct rpcrdma_write_array *ary;
+
+	/* no read-list */
+	rmsgp->rm_body.rm_chunks[0] = xdr_zero;
+
+	/* write-array discrim */
+	ary = (struct rpcrdma_write_array *)
+		&rmsgp->rm_body.rm_chunks[1];
+	ary->wc_discrim = xdr_one;
+	ary->wc_nchunks = htonl(chunks);
+
+	/* write-list terminator */
+	ary->wc_array[chunks].wc_target.rs_handle = xdr_zero;
+
+	/* reply-array discriminator */
+	ary->wc_array[chunks].wc_target.rs_length = xdr_zero;
+}
+
+void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *ary,
+				 int chunks)
+{
+	ary->wc_discrim = xdr_one;
+	ary->wc_nchunks = htonl(chunks);
+}
+
+void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *ary,
+				     int chunk_no,
+				     __be32 rs_handle,
+				     __be64 rs_offset,
+				     u32 write_len)
+{
+	struct rpcrdma_segment *seg = &ary->wc_array[chunk_no].wc_target;
+	seg->rs_handle = rs_handle;
+	seg->rs_offset = rs_offset;
+	seg->rs_length = htonl(write_len);
+}
+
+void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *xprt,
+				  struct rpcrdma_msg *rdma_argp,
+				  struct rpcrdma_msg *rdma_resp,
+				  enum rpcrdma_proc rdma_type)
+{
+	rdma_resp->rm_xid = htonl(rdma_argp->rm_xid);
+	rdma_resp->rm_vers = htonl(rdma_argp->rm_vers);
+	rdma_resp->rm_credit = htonl(xprt->sc_max_requests);
+	rdma_resp->rm_type = htonl(rdma_type);
+
+	/* Encode <nul> chunks lists */
+	rdma_resp->rm_body.rm_chunks[0] = xdr_zero;
+	rdma_resp->rm_body.rm_chunks[1] = xdr_zero;
+	rdma_resp->rm_body.rm_chunks[2] = xdr_zero;
+}

net/sunrpc/xprtrdma/svc_rdma_recvfrom.c

@@ -0,0 +1,614 @@
+/*
+ * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
+ * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Author: Tom Tucker <tom@opengridcomputing.com>
+ */
+
+#include <linux/sunrpc/debug.h>
+#include <linux/sunrpc/rpc_rdma.h>
+#include <linux/spinlock.h>
+#include <linux/highmem.h>
+#include <asm/unaligned.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_cm.h>
+#include <linux/sunrpc/svc_rdma.h>
+
+#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
+
+/*
+ * Replace the pages in the rq_argpages array with the pages from the SGE in
+ * the RDMA_RECV completion. The SGL should contain full pages up until the
+ * last one.
+ */
+static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
+			       struct svc_rdma_op_ctxt *ctxt,
+			       u32 byte_count)
+{
+	struct page *page;
+	u32 bc;
+	int sge_no;
+
+	/* Swap the page in the SGE with the page in argpages */
+	page = ctxt->pages[0];
+	put_page(rqstp->rq_pages[0]);
+	rqstp->rq_pages[0] = page;
+
+	/* Set up the XDR head */
+	rqstp->rq_arg.head[0].iov_base = page_address(page);
+	rqstp->rq_arg.head[0].iov_len =
+		min_t(size_t, byte_count, ctxt->sge[0].length);
+	rqstp->rq_arg.len = byte_count;
+	rqstp->rq_arg.buflen = byte_count;
+
+	/* Compute bytes past head in the SGL */
+	bc = byte_count - rqstp->rq_arg.head[0].iov_len;
+
+	/* If data remains, store it in the pagelist */
+	rqstp->rq_arg.page_len = bc;
+	rqstp->rq_arg.page_base = 0;
+	rqstp->rq_arg.pages = &rqstp->rq_pages[1];
+	sge_no = 1;
+	while (bc && sge_no < ctxt->count) {
+		page = ctxt->pages[sge_no];
+		put_page(rqstp->rq_pages[sge_no]);
+		rqstp->rq_pages[sge_no] = page;
+		bc -= min_t(u32, bc, ctxt->sge[sge_no].length);
+		rqstp->rq_arg.buflen += ctxt->sge[sge_no].length;
+		sge_no++;
+	}
+	rqstp->rq_respages = &rqstp->rq_pages[sge_no];
+	rqstp->rq_next_page = rqstp->rq_respages + 1;
+
+	/* We should never run out of SGE because the limit is defined to
+	 * support the max allowed RPC data length
+	 */
+	BUG_ON(bc && (sge_no == ctxt->count));
+	BUG_ON((rqstp->rq_arg.head[0].iov_len + rqstp->rq_arg.page_len)
+	       != byte_count);
+	BUG_ON(rqstp->rq_arg.len != byte_count);
+
+	/* If not all pages were used from the SGL, free the remaining ones */
+	bc = sge_no;
+	while (sge_no < ctxt->count) {
+		page = ctxt->pages[sge_no++];
+		put_page(page);
+	}
+	ctxt->count = bc;
+
+	/* Set up tail */
+	rqstp->rq_arg.tail[0].iov_base = NULL;
+	rqstp->rq_arg.tail[0].iov_len = 0;
+}
+
+static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count)
+{
+	if (rdma_node_get_transport(xprt->sc_cm_id->device->node_type) ==
+	     RDMA_TRANSPORT_IWARP)
+		return 1;
+	else
+		return min_t(int, sge_count, xprt->sc_max_sge);
+}
+
+typedef int (*rdma_reader_fn)(struct svcxprt_rdma *xprt,
+			      struct svc_rqst *rqstp,
+			      struct svc_rdma_op_ctxt *head,
+			      int *page_no,
+			      u32 *page_offset,
+			      u32 rs_handle,
+			      u32 rs_length,
+			      u64 rs_offset,
+			      int last);
+
+/* Issue an RDMA_READ using the local lkey to map the data sink */
+static int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
+			       struct svc_rqst *rqstp,
+			       struct svc_rdma_op_ctxt *head,
+			       int *page_no,
+			       u32 *page_offset,
+			       u32 rs_handle,
+			       u32 rs_length,
+			       u64 rs_offset,
+			       int last)
+{
+	struct ib_send_wr read_wr;
+	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
+	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
+	int ret, read, pno;
+	u32 pg_off = *page_offset;
+	u32 pg_no = *page_no;
+
+	ctxt->direction = DMA_FROM_DEVICE;
+	ctxt->read_hdr = head;
+	pages_needed =
+		min_t(int, pages_needed, rdma_read_max_sge(xprt, pages_needed));
+	read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
+
+	for (pno = 0; pno < pages_needed; pno++) {
+		int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
+
+		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
+		head->arg.page_len += len;
+		head->arg.len += len;
+		if (!pg_off)
+			head->count++;
+		rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
+		rqstp->rq_next_page = rqstp->rq_respages + 1;
+		ctxt->sge[pno].addr =
+			ib_dma_map_page(xprt->sc_cm_id->device,
+					head->arg.pages[pg_no], pg_off,
+					PAGE_SIZE - pg_off,
+					DMA_FROM_DEVICE);
+		ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
+					   ctxt->sge[pno].addr);
+		if (ret)
+			goto err;
+		atomic_inc(&xprt->sc_dma_used);
+
+		/* The lkey here is either a local dma lkey or a dma_mr lkey */
+		ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
+		ctxt->sge[pno].length = len;
+		ctxt->count++;
+
+		/* adjust offset and wrap to next page if needed */
+		pg_off += len;
+		if (pg_off == PAGE_SIZE) {
+			pg_off = 0;
+			pg_no++;
+		}
+		rs_length -= len;
+	}
+
+	if (last && rs_length == 0)
+		set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
+	else
+		clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
+
+	memset(&read_wr, 0, sizeof(read_wr));
+	read_wr.wr_id = (unsigned long)ctxt;
+	read_wr.opcode = IB_WR_RDMA_READ;
+	ctxt->wr_op = read_wr.opcode;
+	read_wr.send_flags = IB_SEND_SIGNALED;
+	read_wr.wr.rdma.rkey = rs_handle;
+	read_wr.wr.rdma.remote_addr = rs_offset;
+	read_wr.sg_list = ctxt->sge;
+	read_wr.num_sge = pages_needed;
+
+	ret = svc_rdma_send(xprt, &read_wr);
+	if (ret) {
+		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
+		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
+		goto err;
+	}
+
+	/* return current location in page array */
+	*page_no = pg_no;
+	*page_offset = pg_off;
+	ret = read;
+	atomic_inc(&rdma_stat_read);
+	return ret;
+ err:
+	svc_rdma_unmap_dma(ctxt);
+	svc_rdma_put_context(ctxt, 0);
+	return ret;
+}
+
+/* Issue an RDMA_READ using an FRMR to map the data sink */
+static int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
+				struct svc_rqst *rqstp,
+				struct svc_rdma_op_ctxt *head,
+				int *page_no,
+				u32 *page_offset,
+				u32 rs_handle,
+				u32 rs_length,
+				u64 rs_offset,
+				int last)
+{
+	struct ib_send_wr read_wr;
+	struct ib_send_wr inv_wr;
+	struct ib_send_wr fastreg_wr;
+	u8 key;
+	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
+	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
+	struct svc_rdma_fastreg_mr *frmr = svc_rdma_get_frmr(xprt);
+	int ret, read, pno;
+	u32 pg_off = *page_offset;
+	u32 pg_no = *page_no;
+
+	if (IS_ERR(frmr))
+		return -ENOMEM;
+
+	ctxt->direction = DMA_FROM_DEVICE;
+	ctxt->frmr = frmr;
+	pages_needed = min_t(int, pages_needed, xprt->sc_frmr_pg_list_len);
+	read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
+
+	frmr->kva = page_address(rqstp->rq_arg.pages[pg_no]);
+	frmr->direction = DMA_FROM_DEVICE;
+	frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE);
+	frmr->map_len = pages_needed << PAGE_SHIFT;
+	frmr->page_list_len = pages_needed;
+
+	for (pno = 0; pno < pages_needed; pno++) {
+		int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
+
+		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
+		head->arg.page_len += len;
+		head->arg.len += len;
+		if (!pg_off)
+			head->count++;
+		rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
+		rqstp->rq_next_page = rqstp->rq_respages + 1;
+		frmr->page_list->page_list[pno] =
+			ib_dma_map_page(xprt->sc_cm_id->device,
+					head->arg.pages[pg_no], 0,
+					PAGE_SIZE, DMA_FROM_DEVICE);
+		ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
+					   frmr->page_list->page_list[pno]);
+		if (ret)
+			goto err;
+		atomic_inc(&xprt->sc_dma_used);
+
+		/* adjust offset and wrap to next page if needed */
+		pg_off += len;
+		if (pg_off == PAGE_SIZE) {
+			pg_off = 0;
+			pg_no++;
+		}
+		rs_length -= len;
+	}
+
+	if (last && rs_length == 0)
+		set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
+	else
+		clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
+
+	/* Bump the key */
+	key = (u8)(frmr->mr->lkey & 0x000000FF);
+	ib_update_fast_reg_key(frmr->mr, ++key);
+
+	ctxt->sge[0].addr = (unsigned long)frmr->kva + *page_offset;
+	ctxt->sge[0].lkey = frmr->mr->lkey;
+	ctxt->sge[0].length = read;
+	ctxt->count = 1;
+	ctxt->read_hdr = head;
+
+	/* Prepare FASTREG WR */
+	memset(&fastreg_wr, 0, sizeof(fastreg_wr));
+	fastreg_wr.opcode = IB_WR_FAST_REG_MR;
+	fastreg_wr.send_flags = IB_SEND_SIGNALED;
+	fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
+	fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
+	fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
+	fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
+	fastreg_wr.wr.fast_reg.length = frmr->map_len;
+	fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
+	fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
+	fastreg_wr.next = &read_wr;
+
+	/* Prepare RDMA_READ */
+	memset(&read_wr, 0, sizeof(read_wr));
+	read_wr.send_flags = IB_SEND_SIGNALED;
+	read_wr.wr.rdma.rkey = rs_handle;
+	read_wr.wr.rdma.remote_addr = rs_offset;
+	read_wr.sg_list = ctxt->sge;
+	read_wr.num_sge = 1;
+	if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) {
+		read_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
+		read_wr.wr_id = (unsigned long)ctxt;
+		read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
+	} else {
+		read_wr.opcode = IB_WR_RDMA_READ;
+		read_wr.next = &inv_wr;
+		/* Prepare invalidate */
+		memset(&inv_wr, 0, sizeof(inv_wr));
+		inv_wr.wr_id = (unsigned long)ctxt;
+		inv_wr.opcode = IB_WR_LOCAL_INV;
+		inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE;
+		inv_wr.ex.invalidate_rkey = frmr->mr->lkey;
+	}
+	ctxt->wr_op = read_wr.opcode;
+
+	/* Post the chain */
+	ret = svc_rdma_send(xprt, &fastreg_wr);
+	if (ret) {
+		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
+		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
+		goto err;
+	}
+
+	/* return current location in page array */
+	*page_no = pg_no;
+	*page_offset = pg_off;
+	ret = read;
+	atomic_inc(&rdma_stat_read);
+	return ret;
+ err:
+	svc_rdma_unmap_dma(ctxt);
+	svc_rdma_put_context(ctxt, 0);
+	svc_rdma_put_frmr(xprt, frmr);
+	return ret;
+}
+
+static int rdma_read_chunks(struct svcxprt_rdma *xprt,
+			    struct rpcrdma_msg *rmsgp,
+			    struct svc_rqst *rqstp,
+			    struct svc_rdma_op_ctxt *head)
+{
+	int page_no, ch_count, ret;
+	struct rpcrdma_read_chunk *ch;
+	u32 page_offset, byte_count;
+	u64 rs_offset;
+	rdma_reader_fn reader;
+
+	/* If no read list is present, return 0 */
+	ch = svc_rdma_get_read_chunk(rmsgp);
+	if (!ch)
+		return 0;
+
+	svc_rdma_rcl_chunk_counts(ch, &ch_count, &byte_count);
+	if (ch_count > RPCSVC_MAXPAGES)
+		return -EINVAL;
+
+	/* The request is completed when the RDMA_READs complete. The
+	 * head context keeps all the pages that comprise the
+	 * request.
+	 */
+	head->arg.head[0] = rqstp->rq_arg.head[0];
+	head->arg.tail[0] = rqstp->rq_arg.tail[0];
+	head->arg.pages = &head->pages[head->count];
+	head->hdr_count = head->count;
+	head->arg.page_base = 0;
+	head->arg.page_len = 0;
+	head->arg.len = rqstp->rq_arg.len;
+	head->arg.buflen = rqstp->rq_arg.buflen;
+
+	/* Use FRMR if supported */
+	if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG)
+		reader = rdma_read_chunk_frmr;
+	else
+		reader = rdma_read_chunk_lcl;
+
+	page_no = 0; page_offset = 0;
+	for (ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
+	     ch->rc_discrim != 0; ch++) {
+
+		xdr_decode_hyper((__be32 *)&ch->rc_target.rs_offset,
+				 &rs_offset);
+		byte_count = ntohl(ch->rc_target.rs_length);
+
+		while (byte_count > 0) {
+			ret = reader(xprt, rqstp, head,
+				     &page_no, &page_offset,
+				     ntohl(ch->rc_target.rs_handle),
+				     byte_count, rs_offset,
+				     ((ch+1)->rc_discrim == 0) /* last */
+				     );
+			if (ret < 0)
+				goto err;
+			byte_count -= ret;
+			rs_offset += ret;
+			head->arg.buflen += ret;
+		}
+	}
+	ret = 1;
+ err:
+	/* Detach arg pages. svc_recv will replenish them */
+	for (page_no = 0;
+	     &rqstp->rq_pages[page_no] < rqstp->rq_respages; page_no++)
+		rqstp->rq_pages[page_no] = NULL;
+
+	return ret;
+}
+
+/*
+ * To avoid a separate RDMA READ just for a handful of zero bytes,
+ * RFC 5666 section 3.7 allows the client to omit the XDR zero pad
+ * in chunk lists.
+ */
+static void
+rdma_fix_xdr_pad(struct xdr_buf *buf)
+{
+	unsigned int page_len = buf->page_len;
+	unsigned int size = (XDR_QUADLEN(page_len) << 2) - page_len;
+	unsigned int offset, pg_no;
+	char *p;
+
+	if (size == 0)
+		return;
+
+	pg_no = page_len >> PAGE_SHIFT;
+	offset = page_len & ~PAGE_MASK;
+	p = page_address(buf->pages[pg_no]);
+	memset(p + offset, 0, size);
+
+	buf->page_len += size;
+	buf->buflen += size;
+	buf->len += size;
+}
+
+static int rdma_read_complete(struct svc_rqst *rqstp,
+			      struct svc_rdma_op_ctxt *head)
+{
+	int page_no;
+	int ret;
+
+	BUG_ON(!head);
+
+	/* Copy RPC pages */
+	for (page_no = 0; page_no < head->count; page_no++) {
+		put_page(rqstp->rq_pages[page_no]);
+		rqstp->rq_pages[page_no] = head->pages[page_no];
+	}
+	/* Point rq_arg.pages past header */
+	rdma_fix_xdr_pad(&head->arg);
+	rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count];
+	rqstp->rq_arg.page_len = head->arg.page_len;
+	rqstp->rq_arg.page_base = head->arg.page_base;
+
+	/* rq_respages starts after the last arg page */
+	rqstp->rq_respages = &rqstp->rq_arg.pages[page_no];
+	rqstp->rq_next_page = rqstp->rq_respages + 1;
+
+	/* Rebuild rq_arg head and tail. */
+	rqstp->rq_arg.head[0] = head->arg.head[0];
+	rqstp->rq_arg.tail[0] = head->arg.tail[0];
+	rqstp->rq_arg.len = head->arg.len;
+	rqstp->rq_arg.buflen = head->arg.buflen;
+
+	/* Free the context */
+	svc_rdma_put_context(head, 0);
+
+	/* XXX: What should this be? */
+	rqstp->rq_prot = IPPROTO_MAX;
+	svc_xprt_copy_addrs(rqstp, rqstp->rq_xprt);
+
+	ret = rqstp->rq_arg.head[0].iov_len
+		+ rqstp->rq_arg.page_len
+		+ rqstp->rq_arg.tail[0].iov_len;
+	dprintk("svcrdma: deferred read ret=%d, rq_arg.len =%d, "
+		"rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len = %zd\n",
+		ret, rqstp->rq_arg.len,	rqstp->rq_arg.head[0].iov_base,
+		rqstp->rq_arg.head[0].iov_len);
+
+	return ret;
+}
+
+/*
+ * Set up the rqstp thread context to point to the RQ buffer. If
+ * necessary, pull additional data from the client with an RDMA_READ
+ * request.
+ */
+int svc_rdma_recvfrom(struct svc_rqst *rqstp)
+{
+	struct svc_xprt *xprt = rqstp->rq_xprt;
+	struct svcxprt_rdma *rdma_xprt =
+		container_of(xprt, struct svcxprt_rdma, sc_xprt);
+	struct svc_rdma_op_ctxt *ctxt = NULL;
+	struct rpcrdma_msg *rmsgp;
+	int ret = 0;
+	int len;
+
+	dprintk("svcrdma: rqstp=%p\n", rqstp);
+
+	spin_lock_bh(&rdma_xprt->sc_rq_dto_lock);
+	if (!list_empty(&rdma_xprt->sc_read_complete_q)) {
+		ctxt = list_entry(rdma_xprt->sc_read_complete_q.next,
+				  struct svc_rdma_op_ctxt,
+				  dto_q);
+		list_del_init(&ctxt->dto_q);
+		spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
+		return rdma_read_complete(rqstp, ctxt);
+	} else if (!list_empty(&rdma_xprt->sc_rq_dto_q)) {
+		ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next,
+				  struct svc_rdma_op_ctxt,
+				  dto_q);
+		list_del_init(&ctxt->dto_q);
+	} else {
+		atomic_inc(&rdma_stat_rq_starve);
+		clear_bit(XPT_DATA, &xprt->xpt_flags);
+		ctxt = NULL;
+	}
+	spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
+	if (!ctxt) {
+		/* This is the EAGAIN path. The svc_recv routine will
+		 * return -EAGAIN, the nfsd thread will go to call into
+		 * svc_recv again and we shouldn't be on the active
+		 * transport list
+		 */
+		if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
+			goto close_out;
+
+		goto out;
+	}
+	dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n",
+		ctxt, rdma_xprt, rqstp, ctxt->wc_status);
+	BUG_ON(ctxt->wc_status != IB_WC_SUCCESS);
+	atomic_inc(&rdma_stat_recv);
+
+	/* Build up the XDR from the receive buffers. */
+	rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len);
+
+	/* Decode the RDMA header. */
+	len = svc_rdma_xdr_decode_req(&rmsgp, rqstp);
+	rqstp->rq_xprt_hlen = len;
+
+	/* If the request is invalid, reply with an error */
+	if (len < 0) {
+		if (len == -ENOSYS)
+			svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS);
+		goto close_out;
+	}
+
+	/* Read read-list data. */
+	ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt);
+	if (ret > 0) {
+		/* read-list posted, defer until data received from client. */
+		goto defer;
+	} else if (ret < 0) {
+		/* Post of read-list failed, free context. */
+		svc_rdma_put_context(ctxt, 1);
+		return 0;
+	}
+
+	ret = rqstp->rq_arg.head[0].iov_len
+		+ rqstp->rq_arg.page_len
+		+ rqstp->rq_arg.tail[0].iov_len;
+	svc_rdma_put_context(ctxt, 0);
+ out:
+	dprintk("svcrdma: ret = %d, rq_arg.len =%d, "
+		"rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len = %zd\n",
+		ret, rqstp->rq_arg.len,
+		rqstp->rq_arg.head[0].iov_base,
+		rqstp->rq_arg.head[0].iov_len);
+	rqstp->rq_prot = IPPROTO_MAX;
+	svc_xprt_copy_addrs(rqstp, xprt);
+	return ret;
+
+ close_out:
+	if (ctxt)
+		svc_rdma_put_context(ctxt, 1);
+	dprintk("svcrdma: transport %p is closing\n", xprt);
+	/*
+	 * Set the close bit and enqueue it. svc_recv will see the
+	 * close bit and call svc_xprt_delete
+	 */
+	set_bit(XPT_CLOSE, &xprt->xpt_flags);
+defer:
+	return 0;
+}

net/sunrpc/xprtrdma/svc_rdma_sendto.c

@@ -0,0 +1,554 @@
+/*
+ * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
+ * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Author: Tom Tucker <tom@opengridcomputing.com>
+ */
+
+#include <linux/sunrpc/debug.h>
+#include <linux/sunrpc/rpc_rdma.h>
+#include <linux/spinlock.h>
+#include <asm/unaligned.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_cm.h>
+#include <linux/sunrpc/svc_rdma.h>
+
+#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
+
+static int map_xdr(struct svcxprt_rdma *xprt,
+		   struct xdr_buf *xdr,
+		   struct svc_rdma_req_map *vec)
+{
+	int sge_no;
+	u32 sge_bytes;
+	u32 page_bytes;
+	u32 page_off;
+	int page_no;
+
+	BUG_ON(xdr->len !=
+	       (xdr->head[0].iov_len + xdr->page_len + xdr->tail[0].iov_len));
+
+	/* Skip the first sge, this is for the RPCRDMA header */
+	sge_no = 1;
+
+	/* Head SGE */
+	vec->sge[sge_no].iov_base = xdr->head[0].iov_base;
+	vec->sge[sge_no].iov_len = xdr->head[0].iov_len;
+	sge_no++;
+
+	/* pages SGE */
+	page_no = 0;
+	page_bytes = xdr->page_len;
+	page_off = xdr->page_base;
+	while (page_bytes) {
+		vec->sge[sge_no].iov_base =
+			page_address(xdr->pages[page_no]) + page_off;
+		sge_bytes = min_t(u32, page_bytes, (PAGE_SIZE - page_off));
+		page_bytes -= sge_bytes;
+		vec->sge[sge_no].iov_len = sge_bytes;
+
+		sge_no++;
+		page_no++;
+		page_off = 0; /* reset for next time through loop */
+	}
+
+	/* Tail SGE */
+	if (xdr->tail[0].iov_len) {
+		vec->sge[sge_no].iov_base = xdr->tail[0].iov_base;
+		vec->sge[sge_no].iov_len = xdr->tail[0].iov_len;
+		sge_no++;
+	}
+
+	dprintk("svcrdma: map_xdr: sge_no %d page_no %d "
+		"page_base %u page_len %u head_len %zu tail_len %zu\n",
+		sge_no, page_no, xdr->page_base, xdr->page_len,
+		xdr->head[0].iov_len, xdr->tail[0].iov_len);
+
+	vec->count = sge_no;
+	return 0;
+}
+
+static dma_addr_t dma_map_xdr(struct svcxprt_rdma *xprt,
+			      struct xdr_buf *xdr,
+			      u32 xdr_off, size_t len, int dir)
+{
+	struct page *page;
+	dma_addr_t dma_addr;
+	if (xdr_off < xdr->head[0].iov_len) {
+		/* This offset is in the head */
+		xdr_off += (unsigned long)xdr->head[0].iov_base & ~PAGE_MASK;
+		page = virt_to_page(xdr->head[0].iov_base);
+	} else {
+		xdr_off -= xdr->head[0].iov_len;
+		if (xdr_off < xdr->page_len) {
+			/* This offset is in the page list */
+			xdr_off += xdr->page_base;
+			page = xdr->pages[xdr_off >> PAGE_SHIFT];
+			xdr_off &= ~PAGE_MASK;
+		} else {
+			/* This offset is in the tail */
+			xdr_off -= xdr->page_len;
+			xdr_off += (unsigned long)
+				xdr->tail[0].iov_base & ~PAGE_MASK;
+			page = virt_to_page(xdr->tail[0].iov_base);
+		}
+	}
+	dma_addr = ib_dma_map_page(xprt->sc_cm_id->device, page, xdr_off,
+				   min_t(size_t, PAGE_SIZE, len), dir);
+	return dma_addr;
+}
+
+/* Assumptions:
+ * - The specified write_len can be represented in sc_max_sge * PAGE_SIZE
+ */
+static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
+		      u32 rmr, u64 to,
+		      u32 xdr_off, int write_len,
+		      struct svc_rdma_req_map *vec)
+{
+	struct ib_send_wr write_wr;
+	struct ib_sge *sge;
+	int xdr_sge_no;
+	int sge_no;
+	int sge_bytes;
+	int sge_off;
+	int bc;
+	struct svc_rdma_op_ctxt *ctxt;
+
+	BUG_ON(vec->count > RPCSVC_MAXPAGES);
+	dprintk("svcrdma: RDMA_WRITE rmr=%x, to=%llx, xdr_off=%d, "
+		"write_len=%d, vec->sge=%p, vec->count=%lu\n",
+		rmr, (unsigned long long)to, xdr_off,
+		write_len, vec->sge, vec->count);
+
+	ctxt = svc_rdma_get_context(xprt);
+	ctxt->direction = DMA_TO_DEVICE;
+	sge = ctxt->sge;
+
+	/* Find the SGE associated with xdr_off */
+	for (bc = xdr_off, xdr_sge_no = 1; bc && xdr_sge_no < vec->count;
+	     xdr_sge_no++) {
+		if (vec->sge[xdr_sge_no].iov_len > bc)
+			break;
+		bc -= vec->sge[xdr_sge_no].iov_len;
+	}
+
+	sge_off = bc;
+	bc = write_len;
+	sge_no = 0;
+
+	/* Copy the remaining SGE */
+	while (bc != 0) {
+		sge_bytes = min_t(size_t,
+			  bc, vec->sge[xdr_sge_no].iov_len-sge_off);
+		sge[sge_no].length = sge_bytes;
+		sge[sge_no].addr =
+			dma_map_xdr(xprt, &rqstp->rq_res, xdr_off,
+				    sge_bytes, DMA_TO_DEVICE);
+		xdr_off += sge_bytes;
+		if (ib_dma_mapping_error(xprt->sc_cm_id->device,
+					 sge[sge_no].addr))
+			goto err;
+		atomic_inc(&xprt->sc_dma_used);
+		sge[sge_no].lkey = xprt->sc_dma_lkey;
+		ctxt->count++;
+		sge_off = 0;
+		sge_no++;
+		xdr_sge_no++;
+		BUG_ON(xdr_sge_no > vec->count);
+		bc -= sge_bytes;
+		if (sge_no == xprt->sc_max_sge)
+			break;
+	}
+
+	/* Prepare WRITE WR */
+	memset(&write_wr, 0, sizeof write_wr);
+	ctxt->wr_op = IB_WR_RDMA_WRITE;
+	write_wr.wr_id = (unsigned long)ctxt;
+	write_wr.sg_list = &sge[0];
+	write_wr.num_sge = sge_no;
+	write_wr.opcode = IB_WR_RDMA_WRITE;
+	write_wr.send_flags = IB_SEND_SIGNALED;
+	write_wr.wr.rdma.rkey = rmr;
+	write_wr.wr.rdma.remote_addr = to;
+
+	/* Post It */
+	atomic_inc(&rdma_stat_write);
+	if (svc_rdma_send(xprt, &write_wr))
+		goto err;
+	return write_len - bc;
+ err:
+	svc_rdma_unmap_dma(ctxt);
+	svc_rdma_put_context(ctxt, 0);
+	/* Fatal error, close transport */
+	return -EIO;
+}
+
+static int send_write_chunks(struct svcxprt_rdma *xprt,
+			     struct rpcrdma_msg *rdma_argp,
+			     struct rpcrdma_msg *rdma_resp,
+			     struct svc_rqst *rqstp,
+			     struct svc_rdma_req_map *vec)
+{
+	u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
+	int write_len;
+	u32 xdr_off;
+	int chunk_off;
+	int chunk_no;
+	struct rpcrdma_write_array *arg_ary;
+	struct rpcrdma_write_array *res_ary;
+	int ret;
+
+	arg_ary = svc_rdma_get_write_array(rdma_argp);
+	if (!arg_ary)
+		return 0;
+	res_ary = (struct rpcrdma_write_array *)
+		&rdma_resp->rm_body.rm_chunks[1];
+
+	/* Write chunks start at the pagelist */
+	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
+	     xfer_len && chunk_no < arg_ary->wc_nchunks;
+	     chunk_no++) {
+		struct rpcrdma_segment *arg_ch;
+		u64 rs_offset;
+
+		arg_ch = &arg_ary->wc_array[chunk_no].wc_target;
+		write_len = min(xfer_len, ntohl(arg_ch->rs_length));
+
+		/* Prepare the response chunk given the length actually
+		 * written */
+		xdr_decode_hyper((__be32 *)&arg_ch->rs_offset, &rs_offset);
+		svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no,
+						arg_ch->rs_handle,
+						arg_ch->rs_offset,
+						write_len);
+		chunk_off = 0;
+		while (write_len) {
+			ret = send_write(xprt, rqstp,
+					 ntohl(arg_ch->rs_handle),
+					 rs_offset + chunk_off,
+					 xdr_off,
+					 write_len,
+					 vec);
+			if (ret <= 0) {
+				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
+					ret);
+				return -EIO;
+			}
+			chunk_off += ret;
+			xdr_off += ret;
+			xfer_len -= ret;
+			write_len -= ret;
+		}
+	}
+	/* Update the req with the number of chunks actually used */
+	svc_rdma_xdr_encode_write_list(rdma_resp, chunk_no);
+
+	return rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
+}
+
+static int send_reply_chunks(struct svcxprt_rdma *xprt,
+			     struct rpcrdma_msg *rdma_argp,
+			     struct rpcrdma_msg *rdma_resp,
+			     struct svc_rqst *rqstp,
+			     struct svc_rdma_req_map *vec)
+{
+	u32 xfer_len = rqstp->rq_res.len;
+	int write_len;
+	u32 xdr_off;
+	int chunk_no;
+	int chunk_off;
+	int nchunks;
+	struct rpcrdma_segment *ch;
+	struct rpcrdma_write_array *arg_ary;
+	struct rpcrdma_write_array *res_ary;
+	int ret;
+
+	arg_ary = svc_rdma_get_reply_array(rdma_argp);
+	if (!arg_ary)
+		return 0;
+	/* XXX: need to fix when reply lists occur with read-list and or
+	 * write-list */
+	res_ary = (struct rpcrdma_write_array *)
+		&rdma_resp->rm_body.rm_chunks[2];
+
+	/* xdr offset starts at RPC message */
+	nchunks = ntohl(arg_ary->wc_nchunks);
+	for (xdr_off = 0, chunk_no = 0;
+	     xfer_len && chunk_no < nchunks;
+	     chunk_no++) {
+		u64 rs_offset;
+		ch = &arg_ary->wc_array[chunk_no].wc_target;
+		write_len = min(xfer_len, htonl(ch->rs_length));
+
+		/* Prepare the reply chunk given the length actually
+		 * written */
+		xdr_decode_hyper((__be32 *)&ch->rs_offset, &rs_offset);
+		svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no,
+						ch->rs_handle, ch->rs_offset,
+						write_len);
+		chunk_off = 0;
+		while (write_len) {
+			ret = send_write(xprt, rqstp,
+					 ntohl(ch->rs_handle),
+					 rs_offset + chunk_off,
+					 xdr_off,
+					 write_len,
+					 vec);
+			if (ret <= 0) {
+				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
+					ret);
+				return -EIO;
+			}
+			chunk_off += ret;
+			xdr_off += ret;
+			xfer_len -= ret;
+			write_len -= ret;
+		}
+	}
+	/* Update the req with the number of chunks actually used */
+	svc_rdma_xdr_encode_reply_array(res_ary, chunk_no);
+
+	return rqstp->rq_res.len;
+}
+
+/* This function prepares the portion of the RPCRDMA message to be
+ * sent in the RDMA_SEND. This function is called after data sent via
+ * RDMA has already been transmitted. There are three cases:
+ * - The RPCRDMA header, RPC header, and payload are all sent in a
+ *   single RDMA_SEND. This is the "inline" case.
+ * - The RPCRDMA header and some portion of the RPC header and data
+ *   are sent via this RDMA_SEND and another portion of the data is
+ *   sent via RDMA.
+ * - The RPCRDMA header [NOMSG] is sent in this RDMA_SEND and the RPC
+ *   header and data are all transmitted via RDMA.
+ * In all three cases, this function prepares the RPCRDMA header in
+ * sge[0], the 'type' parameter indicates the type to place in the
+ * RPCRDMA header, and the 'byte_count' field indicates how much of
+ * the XDR to include in this RDMA_SEND. NB: The offset of the payload
+ * to send is zero in the XDR.
+ */
+static int send_reply(struct svcxprt_rdma *rdma,
+		      struct svc_rqst *rqstp,
+		      struct page *page,
+		      struct rpcrdma_msg *rdma_resp,
+		      struct svc_rdma_op_ctxt *ctxt,
+		      struct svc_rdma_req_map *vec,
+		      int byte_count)
+{
+	struct ib_send_wr send_wr;
+	int sge_no;
+	int sge_bytes;
+	int page_no;
+	int pages;
+	int ret;
+
+	/* Post a recv buffer to handle another request. */
+	ret = svc_rdma_post_recv(rdma);
+	if (ret) {
+		printk(KERN_INFO
+		       "svcrdma: could not post a receive buffer, err=%d."
+		       "Closing transport %p.\n", ret, rdma);
+		set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
+		svc_rdma_put_context(ctxt, 0);
+		return -ENOTCONN;
+	}
+
+	/* Prepare the context */
+	ctxt->pages[0] = page;
+	ctxt->count = 1;
+
+	/* Prepare the SGE for the RPCRDMA Header */
+	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
+	ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp);
+	ctxt->sge[0].addr =
+	    ib_dma_map_page(rdma->sc_cm_id->device, page, 0,
+			    ctxt->sge[0].length, DMA_TO_DEVICE);
+	if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr))
+		goto err;
+	atomic_inc(&rdma->sc_dma_used);
+
+	ctxt->direction = DMA_TO_DEVICE;
+
+	/* Map the payload indicated by 'byte_count' */
+	for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) {
+		int xdr_off = 0;
+		sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count);
+		byte_count -= sge_bytes;
+		ctxt->sge[sge_no].addr =
+			dma_map_xdr(rdma, &rqstp->rq_res, xdr_off,
+				    sge_bytes, DMA_TO_DEVICE);
+		xdr_off += sge_bytes;
+		if (ib_dma_mapping_error(rdma->sc_cm_id->device,
+					 ctxt->sge[sge_no].addr))
+			goto err;
+		atomic_inc(&rdma->sc_dma_used);
+		ctxt->sge[sge_no].lkey = rdma->sc_dma_lkey;
+		ctxt->sge[sge_no].length = sge_bytes;
+	}
+	BUG_ON(byte_count != 0);
+
+	/* Save all respages in the ctxt and remove them from the
+	 * respages array. They are our pages until the I/O
+	 * completes.
+	 */
+	pages = rqstp->rq_next_page - rqstp->rq_respages;
+	for (page_no = 0; page_no < pages; page_no++) {
+		ctxt->pages[page_no+1] = rqstp->rq_respages[page_no];
+		ctxt->count++;
+		rqstp->rq_respages[page_no] = NULL;
+		/*
+		 * If there are more pages than SGE, terminate SGE
+		 * list so that svc_rdma_unmap_dma doesn't attempt to
+		 * unmap garbage.
+		 */
+		if (page_no+1 >= sge_no)
+			ctxt->sge[page_no+1].length = 0;
+	}
+	rqstp->rq_next_page = rqstp->rq_respages + 1;
+
+	BUG_ON(sge_no > rdma->sc_max_sge);
+	memset(&send_wr, 0, sizeof send_wr);
+	ctxt->wr_op = IB_WR_SEND;
+	send_wr.wr_id = (unsigned long)ctxt;
+	send_wr.sg_list = ctxt->sge;
+	send_wr.num_sge = sge_no;
+	send_wr.opcode = IB_WR_SEND;
+	send_wr.send_flags =  IB_SEND_SIGNALED;
+
+	ret = svc_rdma_send(rdma, &send_wr);
+	if (ret)
+		goto err;
+
+	return 0;
+
+ err:
+	svc_rdma_unmap_dma(ctxt);
+	svc_rdma_put_context(ctxt, 1);
+	return -EIO;
+}
+
+void svc_rdma_prep_reply_hdr(struct svc_rqst *rqstp)
+{
+}
+
+/*
+ * Return the start of an xdr buffer.
+ */
+static void *xdr_start(struct xdr_buf *xdr)
+{
+	return xdr->head[0].iov_base -
+		(xdr->len -
+		 xdr->page_len -
+		 xdr->tail[0].iov_len -
+		 xdr->head[0].iov_len);
+}
+
+int svc_rdma_sendto(struct svc_rqst *rqstp)
+{
+	struct svc_xprt *xprt = rqstp->rq_xprt;
+	struct svcxprt_rdma *rdma =
+		container_of(xprt, struct svcxprt_rdma, sc_xprt);
+	struct rpcrdma_msg *rdma_argp;
+	struct rpcrdma_msg *rdma_resp;
+	struct rpcrdma_write_array *reply_ary;
+	enum rpcrdma_proc reply_type;
+	int ret;
+	int inline_bytes;
+	struct page *res_page;
+	struct svc_rdma_op_ctxt *ctxt;
+	struct svc_rdma_req_map *vec;
+
+	dprintk("svcrdma: sending response for rqstp=%p\n", rqstp);
+
+	/* Get the RDMA request header. */
+	rdma_argp = xdr_start(&rqstp->rq_arg);
+
+	/* Build an req vec for the XDR */
+	ctxt = svc_rdma_get_context(rdma);
+	ctxt->direction = DMA_TO_DEVICE;
+	vec = svc_rdma_get_req_map();
+	ret = map_xdr(rdma, &rqstp->rq_res, vec);
+	if (ret)
+		goto err0;
+	inline_bytes = rqstp->rq_res.len;
+
+	/* Create the RDMA response header */
+	res_page = svc_rdma_get_page();
+	rdma_resp = page_address(res_page);
+	reply_ary = svc_rdma_get_reply_array(rdma_argp);
+	if (reply_ary)
+		reply_type = RDMA_NOMSG;
+	else
+		reply_type = RDMA_MSG;
+	svc_rdma_xdr_encode_reply_header(rdma, rdma_argp,
+					 rdma_resp, reply_type);
+
+	/* Send any write-chunk data and build resp write-list */
+	ret = send_write_chunks(rdma, rdma_argp, rdma_resp,
+				rqstp, vec);
+	if (ret < 0) {
+		printk(KERN_ERR "svcrdma: failed to send write chunks, rc=%d\n",
+		       ret);
+		goto err1;
+	}
+	inline_bytes -= ret;
+
+	/* Send any reply-list data and update resp reply-list */
+	ret = send_reply_chunks(rdma, rdma_argp, rdma_resp,
+				rqstp, vec);
+	if (ret < 0) {
+		printk(KERN_ERR "svcrdma: failed to send reply chunks, rc=%d\n",
+		       ret);
+		goto err1;
+	}
+	inline_bytes -= ret;
+
+	ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec,
+			 inline_bytes);
+	svc_rdma_put_req_map(vec);
+	dprintk("svcrdma: send_reply returns %d\n", ret);
+	return ret;
+
+ err1:
+	put_page(res_page);
+ err0:
+	svc_rdma_put_req_map(vec);
+	svc_rdma_put_context(ctxt, 0);
+	return ret;
+}

net/sunrpc/xprtrdma/transport.c

@@ -0,0 +1,747 @@
+/*
+ * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/*
+ * transport.c
+ *
+ * This file contains the top-level implementation of an RPC RDMA
+ * transport.
+ *
+ * Naming convention: functions beginning with xprt_ are part of the
+ * transport switch. All others are RPC RDMA internal.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/seq_file.h>
+#include <linux/sunrpc/addr.h>
+
+#include "xprt_rdma.h"
+
+#ifdef RPC_DEBUG
+# define RPCDBG_FACILITY	RPCDBG_TRANS
+#endif
+
+MODULE_LICENSE("Dual BSD/GPL");
+
+MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
+MODULE_AUTHOR("Network Appliance, Inc.");
+
+/*
+ * tunables
+ */
+
+static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
+static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
+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 = 0;
+
+#ifdef RPC_DEBUG
+
+static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
+static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
+static unsigned int zero;
+static unsigned int max_padding = PAGE_SIZE;
+static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
+static unsigned int max_memreg = RPCRDMA_LAST - 1;
+
+static struct ctl_table_header *sunrpc_table_header;
+
+static struct ctl_table xr_tunables_table[] = {
+	{
+		.procname	= "rdma_slot_table_entries",
+		.data		= &xprt_rdma_slot_table_entries,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_slot_table_size,
+		.extra2		= &max_slot_table_size
+	},
+	{
+		.procname	= "rdma_max_inline_read",
+		.data		= &xprt_rdma_max_inline_read,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "rdma_max_inline_write",
+		.data		= &xprt_rdma_max_inline_write,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "rdma_inline_write_padding",
+		.data		= &xprt_rdma_inline_write_padding,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &max_padding,
+	},
+	{
+		.procname	= "rdma_memreg_strategy",
+		.data		= &xprt_rdma_memreg_strategy,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_memreg,
+		.extra2		= &max_memreg,
+	},
+	{
+		.procname	= "rdma_pad_optimize",
+		.data		= &xprt_rdma_pad_optimize,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{ },
+};
+
+static struct ctl_table sunrpc_table[] = {
+	{
+		.procname	= "sunrpc",
+		.mode		= 0555,
+		.child		= xr_tunables_table
+	},
+	{ },
+};
+
+#endif
+
+#define RPCRDMA_BIND_TO		(60U * HZ)
+#define RPCRDMA_INIT_REEST_TO	(5U * HZ)
+#define RPCRDMA_MAX_REEST_TO	(30U * HZ)
+#define RPCRDMA_IDLE_DISC_TO	(5U * 60 * HZ)
+
+static struct rpc_xprt_ops xprt_rdma_procs;	/* forward reference */
+
+static void
+xprt_rdma_format_addresses(struct rpc_xprt *xprt)
+{
+	struct sockaddr *sap = (struct sockaddr *)
+					&rpcx_to_rdmad(xprt).addr;
+	struct sockaddr_in *sin = (struct sockaddr_in *)sap;
+	char buf[64];
+
+	(void)rpc_ntop(sap, buf, sizeof(buf));
+	xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
+
+	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
+	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
+
+	xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
+
+	snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
+	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
+
+	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
+	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
+
+	/* netid */
+	xprt->address_strings[RPC_DISPLAY_NETID] = "rdma";
+}
+
+static void
+xprt_rdma_free_addresses(struct rpc_xprt *xprt)
+{
+	unsigned int i;
+
+	for (i = 0; i < RPC_DISPLAY_MAX; i++)
+		switch (i) {
+		case RPC_DISPLAY_PROTO:
+		case RPC_DISPLAY_NETID:
+			continue;
+		default:
+			kfree(xprt->address_strings[i]);
+		}
+}
+
+static void
+xprt_rdma_connect_worker(struct work_struct *work)
+{
+	struct rpcrdma_xprt *r_xprt =
+		container_of(work, struct rpcrdma_xprt, rdma_connect.work);
+	struct rpc_xprt *xprt = &r_xprt->xprt;
+	int rc = 0;
+
+	xprt_clear_connected(xprt);
+
+	dprintk("RPC:       %s: %sconnect\n", __func__,
+			r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
+	rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
+	if (rc)
+		xprt_wake_pending_tasks(xprt, rc);
+
+	dprintk("RPC:       %s: exit\n", __func__);
+	xprt_clear_connecting(xprt);
+}
+
+/*
+ * xprt_rdma_destroy
+ *
+ * Destroy the xprt.
+ * Free all memory associated with the object, including its own.
+ * NOTE: none of the *destroy methods free memory for their top-level
+ * objects, even though they may have allocated it (they do free
+ * private memory). It's up to the caller to handle it. In this
+ * case (RDMA transport), all structure memory is inlined with the
+ * struct rpcrdma_xprt.
+ */
+static void
+xprt_rdma_destroy(struct rpc_xprt *xprt)
+{
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+
+	dprintk("RPC:       %s: called\n", __func__);
+
+	cancel_delayed_work_sync(&r_xprt->rdma_connect);
+
+	xprt_clear_connected(xprt);
+
+	rpcrdma_buffer_destroy(&r_xprt->rx_buf);
+	rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
+	rpcrdma_ia_close(&r_xprt->rx_ia);
+
+	xprt_rdma_free_addresses(xprt);
+
+	xprt_free(xprt);
+
+	dprintk("RPC:       %s: returning\n", __func__);
+
+	module_put(THIS_MODULE);
+}
+
+static const struct rpc_timeout xprt_rdma_default_timeout = {
+	.to_initval = 60 * HZ,
+	.to_maxval = 60 * HZ,
+};
+
+/**
+ * xprt_setup_rdma - Set up transport to use RDMA
+ *
+ * @args: rpc transport arguments
+ */
+static struct rpc_xprt *
+xprt_setup_rdma(struct xprt_create *args)
+{
+	struct rpcrdma_create_data_internal cdata;
+	struct rpc_xprt *xprt;
+	struct rpcrdma_xprt *new_xprt;
+	struct rpcrdma_ep *new_ep;
+	struct sockaddr_in *sin;
+	int rc;
+
+	if (args->addrlen > sizeof(xprt->addr)) {
+		dprintk("RPC:       %s: address too large\n", __func__);
+		return ERR_PTR(-EBADF);
+	}
+
+	xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt),
+			xprt_rdma_slot_table_entries,
+			xprt_rdma_slot_table_entries);
+	if (xprt == NULL) {
+		dprintk("RPC:       %s: couldn't allocate rpcrdma_xprt\n",
+			__func__);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* 60 second timeout, no retries */
+	xprt->timeout = &xprt_rdma_default_timeout;
+	xprt->bind_timeout = RPCRDMA_BIND_TO;
+	xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
+	xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO;
+
+	xprt->resvport = 0;		/* privileged port not needed */
+	xprt->tsh_size = 0;		/* RPC-RDMA handles framing */
+	xprt->ops = &xprt_rdma_procs;
+
+	/*
+	 * Set up RDMA-specific connect data.
+	 */
+
+	/* Put server RDMA address in local cdata */
+	memcpy(&cdata.addr, args->dstaddr, args->addrlen);
+
+	/* Ensure xprt->addr holds valid server TCP (not RDMA)
+	 * address, for any side protocols which peek at it */
+	xprt->prot = IPPROTO_TCP;
+	xprt->addrlen = args->addrlen;
+	memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
+
+	sin = (struct sockaddr_in *)&cdata.addr;
+	if (ntohs(sin->sin_port) != 0)
+		xprt_set_bound(xprt);
+
+	dprintk("RPC:       %s: %pI4:%u\n",
+		__func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port));
+
+	/* Set max requests */
+	cdata.max_requests = xprt->max_reqs;
+
+	/* Set some length limits */
+	cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
+	cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
+
+	cdata.inline_wsize = xprt_rdma_max_inline_write;
+	if (cdata.inline_wsize > cdata.wsize)
+		cdata.inline_wsize = cdata.wsize;
+
+	cdata.inline_rsize = xprt_rdma_max_inline_read;
+	if (cdata.inline_rsize > cdata.rsize)
+		cdata.inline_rsize = cdata.rsize;
+
+	cdata.padding = xprt_rdma_inline_write_padding;
+
+	/*
+	 * Create new transport instance, which includes initialized
+	 *  o ia
+	 *  o endpoint
+	 *  o buffers
+	 */
+
+	new_xprt = rpcx_to_rdmax(xprt);
+
+	rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
+				xprt_rdma_memreg_strategy);
+	if (rc)
+		goto out1;
+
+	/*
+	 * initialize and create ep
+	 */
+	new_xprt->rx_data = cdata;
+	new_ep = &new_xprt->rx_ep;
+	new_ep->rep_remote_addr = cdata.addr;
+
+	rc = rpcrdma_ep_create(&new_xprt->rx_ep,
+				&new_xprt->rx_ia, &new_xprt->rx_data);
+	if (rc)
+		goto out2;
+
+	/*
+	 * Allocate pre-registered send and receive buffers for headers and
+	 * any inline data. Also specify any padding which will be provided
+	 * from a preregistered zero buffer.
+	 */
+	rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
+				&new_xprt->rx_data);
+	if (rc)
+		goto out3;
+
+	/*
+	 * Register a callback for connection events. This is necessary because
+	 * connection loss notification is async. We also catch connection loss
+	 * when reaping receives.
+	 */
+	INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
+	new_ep->rep_func = rpcrdma_conn_func;
+	new_ep->rep_xprt = xprt;
+
+	xprt_rdma_format_addresses(xprt);
+	xprt->max_payload = rpcrdma_max_payload(new_xprt);
+	dprintk("RPC:       %s: transport data payload maximum: %zu bytes\n",
+		__func__, xprt->max_payload);
+
+	if (!try_module_get(THIS_MODULE))
+		goto out4;
+
+	return xprt;
+
+out4:
+	xprt_rdma_free_addresses(xprt);
+	rc = -EINVAL;
+out3:
+	rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
+out2:
+	rpcrdma_ia_close(&new_xprt->rx_ia);
+out1:
+	xprt_free(xprt);
+	return ERR_PTR(rc);
+}
+
+/*
+ * Close a connection, during shutdown or timeout/reconnect
+ */
+static void
+xprt_rdma_close(struct rpc_xprt *xprt)
+{
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+
+	dprintk("RPC:       %s: closing\n", __func__);
+	if (r_xprt->rx_ep.rep_connected > 0)
+		xprt->reestablish_timeout = 0;
+	xprt_disconnect_done(xprt);
+	rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
+}
+
+static void
+xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
+{
+	struct sockaddr_in *sap;
+
+	sap = (struct sockaddr_in *)&xprt->addr;
+	sap->sin_port = htons(port);
+	sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
+	sap->sin_port = htons(port);
+	dprintk("RPC:       %s: %u\n", __func__, port);
+}
+
+static void
+xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task)
+{
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+
+	if (r_xprt->rx_ep.rep_connected != 0) {
+		/* Reconnect */
+		schedule_delayed_work(&r_xprt->rdma_connect,
+			xprt->reestablish_timeout);
+		xprt->reestablish_timeout <<= 1;
+		if (xprt->reestablish_timeout > RPCRDMA_MAX_REEST_TO)
+			xprt->reestablish_timeout = RPCRDMA_MAX_REEST_TO;
+		else if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO)
+			xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
+	} else {
+		schedule_delayed_work(&r_xprt->rdma_connect, 0);
+		if (!RPC_IS_ASYNC(task))
+			flush_delayed_work(&r_xprt->rdma_connect);
+	}
+}
+
+/*
+ * The RDMA allocate/free functions need the task structure as a place
+ * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
+ * sequence. For this reason, the recv buffers are attached to send
+ * buffers for portions of the RPC. Note that the RPC layer allocates
+ * both send and receive buffers in the same call. We may register
+ * the receive buffer portion when using reply chunks.
+ */
+static void *
+xprt_rdma_allocate(struct rpc_task *task, size_t size)
+{
+	struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
+	struct rpcrdma_req *req, *nreq;
+
+	req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
+	if (req == NULL)
+		return NULL;
+
+	if (size > req->rl_size) {
+		dprintk("RPC:       %s: size %zd too large for buffer[%zd]: "
+			"prog %d vers %d proc %d\n",
+			__func__, size, req->rl_size,
+			task->tk_client->cl_prog, task->tk_client->cl_vers,
+			task->tk_msg.rpc_proc->p_proc);
+		/*
+		 * Outgoing length shortage. Our inline write max must have
+		 * been configured to perform direct i/o.
+		 *
+		 * This is therefore a large metadata operation, and the
+		 * allocate call was made on the maximum possible message,
+		 * e.g. containing long filename(s) or symlink data. In
+		 * fact, while these metadata operations *might* carry
+		 * large outgoing payloads, they rarely *do*. However, we
+		 * have to commit to the request here, so reallocate and
+		 * register it now. The data path will never require this
+		 * reallocation.
+		 *
+		 * If the allocation or registration fails, the RPC framework
+		 * will (doggedly) retry.
+		 */
+		if (task->tk_flags & RPC_TASK_SWAPPER)
+			nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
+		else
+			nreq = kmalloc(sizeof *req + size, GFP_NOFS);
+		if (nreq == NULL)
+			goto outfail;
+
+		if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
+				nreq->rl_base, size + sizeof(struct rpcrdma_req)
+				- offsetof(struct rpcrdma_req, rl_base),
+				&nreq->rl_handle, &nreq->rl_iov)) {
+			kfree(nreq);
+			goto outfail;
+		}
+		rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
+		nreq->rl_size = size;
+		nreq->rl_niovs = 0;
+		nreq->rl_nchunks = 0;
+		nreq->rl_buffer = (struct rpcrdma_buffer *)req;
+		nreq->rl_reply = req->rl_reply;
+		memcpy(nreq->rl_segments,
+			req->rl_segments, sizeof nreq->rl_segments);
+		/* flag the swap with an unused field */
+		nreq->rl_iov.length = 0;
+		req->rl_reply = NULL;
+		req = nreq;
+	}
+	dprintk("RPC:       %s: size %zd, request 0x%p\n", __func__, size, req);
+	req->rl_connect_cookie = 0;	/* our reserved value */
+	return req->rl_xdr_buf;
+
+outfail:
+	rpcrdma_buffer_put(req);
+	rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
+	return NULL;
+}
+
+/*
+ * This function returns all RDMA resources to the pool.
+ */
+static void
+xprt_rdma_free(void *buffer)
+{
+	struct rpcrdma_req *req;
+	struct rpcrdma_xprt *r_xprt;
+	struct rpcrdma_rep *rep;
+	int i;
+
+	if (buffer == NULL)
+		return;
+
+	req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
+	if (req->rl_iov.length == 0) {	/* see allocate above */
+		r_xprt = container_of(((struct rpcrdma_req *) req->rl_buffer)->rl_buffer,
+				      struct rpcrdma_xprt, rx_buf);
+	} else
+		r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
+	rep = req->rl_reply;
+
+	dprintk("RPC:       %s: called on 0x%p%s\n",
+		__func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
+
+	/*
+	 * Finish the deregistration.  The process is considered
+	 * complete when the rr_func vector becomes NULL - this
+	 * was put in place during rpcrdma_reply_handler() - the wait
+	 * call below will not block if the dereg is "done". If
+	 * interrupted, our framework will clean up.
+	 */
+	for (i = 0; req->rl_nchunks;) {
+		--req->rl_nchunks;
+		i += rpcrdma_deregister_external(
+			&req->rl_segments[i], r_xprt);
+	}
+
+	if (req->rl_iov.length == 0) {	/* see allocate above */
+		struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
+		oreq->rl_reply = req->rl_reply;
+		(void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
+						   req->rl_handle,
+						   &req->rl_iov);
+		kfree(req);
+		req = oreq;
+	}
+
+	/* Put back request+reply buffers */
+	rpcrdma_buffer_put(req);
+}
+
+/*
+ * send_request invokes the meat of RPC RDMA. It must do the following:
+ *  1.  Marshal the RPC request into an RPC RDMA request, which means
+ *	putting a header in front of data, and creating IOVs for RDMA
+ *	from those in the request.
+ *  2.  In marshaling, detect opportunities for RDMA, and use them.
+ *  3.  Post a recv message to set up asynch completion, then send
+ *	the request (rpcrdma_ep_post).
+ *  4.  No partial sends are possible in the RPC-RDMA protocol (as in UDP).
+ */
+
+static int
+xprt_rdma_send_request(struct rpc_task *task)
+{
+	struct rpc_rqst *rqst = task->tk_rqstp;
+	struct rpc_xprt *xprt = rqst->rq_xprt;
+	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	int rc = 0;
+
+	if (req->rl_niovs == 0)
+		rc = rpcrdma_marshal_req(rqst);
+	else if (r_xprt->rx_ia.ri_memreg_strategy == RPCRDMA_FRMR)
+		rc = rpcrdma_marshal_chunks(rqst, 0);
+	if (rc < 0)
+		goto failed_marshal;
+
+	if (req->rl_reply == NULL) 		/* e.g. reconnection */
+		rpcrdma_recv_buffer_get(req);
+
+	if (req->rl_reply) {
+		req->rl_reply->rr_func = rpcrdma_reply_handler;
+		/* this need only be done once, but... */
+		req->rl_reply->rr_xprt = xprt;
+	}
+
+	/* Must suppress retransmit to maintain credits */
+	if (req->rl_connect_cookie == xprt->connect_cookie)
+		goto drop_connection;
+	req->rl_connect_cookie = xprt->connect_cookie;
+
+	if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
+		goto drop_connection;
+
+	rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
+	rqst->rq_bytes_sent = 0;
+	return 0;
+
+failed_marshal:
+	r_xprt->rx_stats.failed_marshal_count++;
+	dprintk("RPC:       %s: rpcrdma_marshal_req failed, status %i\n",
+		__func__, rc);
+	if (rc == -EIO)
+		return -EIO;
+drop_connection:
+	xprt_disconnect_done(xprt);
+	return -ENOTCONN;	/* implies disconnect */
+}
+
+static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
+{
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	long idle_time = 0;
+
+	if (xprt_connected(xprt))
+		idle_time = (long)(jiffies - xprt->last_used) / HZ;
+
+	seq_printf(seq,
+	  "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
+	  "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
+
+	   0,	/* need a local port? */
+	   xprt->stat.bind_count,
+	   xprt->stat.connect_count,
+	   xprt->stat.connect_time,
+	   idle_time,
+	   xprt->stat.sends,
+	   xprt->stat.recvs,
+	   xprt->stat.bad_xids,
+	   xprt->stat.req_u,
+	   xprt->stat.bklog_u,
+
+	   r_xprt->rx_stats.read_chunk_count,
+	   r_xprt->rx_stats.write_chunk_count,
+	   r_xprt->rx_stats.reply_chunk_count,
+	   r_xprt->rx_stats.total_rdma_request,
+	   r_xprt->rx_stats.total_rdma_reply,
+	   r_xprt->rx_stats.pullup_copy_count,
+	   r_xprt->rx_stats.fixup_copy_count,
+	   r_xprt->rx_stats.hardway_register_count,
+	   r_xprt->rx_stats.failed_marshal_count,
+	   r_xprt->rx_stats.bad_reply_count);
+}
+
+/*
+ * Plumbing for rpc transport switch and kernel module
+ */
+
+static struct rpc_xprt_ops xprt_rdma_procs = {
+	.reserve_xprt		= xprt_reserve_xprt_cong,
+	.release_xprt		= xprt_release_xprt_cong, /* sunrpc/xprt.c */
+	.alloc_slot		= xprt_alloc_slot,
+	.release_request	= xprt_release_rqst_cong,       /* ditto */
+	.set_retrans_timeout	= xprt_set_retrans_timeout_def, /* ditto */
+	.rpcbind		= rpcb_getport_async,	/* sunrpc/rpcb_clnt.c */
+	.set_port		= xprt_rdma_set_port,
+	.connect		= xprt_rdma_connect,
+	.buf_alloc		= xprt_rdma_allocate,
+	.buf_free		= xprt_rdma_free,
+	.send_request		= xprt_rdma_send_request,
+	.close			= xprt_rdma_close,
+	.destroy		= xprt_rdma_destroy,
+	.print_stats		= xprt_rdma_print_stats
+};
+
+static struct xprt_class xprt_rdma = {
+	.list			= LIST_HEAD_INIT(xprt_rdma.list),
+	.name			= "rdma",
+	.owner			= THIS_MODULE,
+	.ident			= XPRT_TRANSPORT_RDMA,
+	.setup			= xprt_setup_rdma,
+};
+
+static void __exit xprt_rdma_cleanup(void)
+{
+	int rc;
+
+	dprintk("RPCRDMA Module Removed, deregister RPC RDMA transport\n");
+#ifdef RPC_DEBUG
+	if (sunrpc_table_header) {
+		unregister_sysctl_table(sunrpc_table_header);
+		sunrpc_table_header = NULL;
+	}
+#endif
+	rc = xprt_unregister_transport(&xprt_rdma);
+	if (rc)
+		dprintk("RPC:       %s: xprt_unregister returned %i\n",
+			__func__, rc);
+}
+
+static int __init xprt_rdma_init(void)
+{
+	int rc;
+
+	rc = xprt_register_transport(&xprt_rdma);
+
+	if (rc)
+		return rc;
+
+	dprintk("RPCRDMA Module Init, register RPC RDMA transport\n");
+
+	dprintk("Defaults:\n");
+	dprintk("\tSlots %d\n"
+		"\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
+		xprt_rdma_slot_table_entries,
+		xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
+	dprintk("\tPadding %d\n\tMemreg %d\n",
+		xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
+
+#ifdef RPC_DEBUG
+	if (!sunrpc_table_header)
+		sunrpc_table_header = register_sysctl_table(sunrpc_table);
+#endif
+	return 0;
+}
+
+module_init(xprt_rdma_init);
+module_exit(xprt_rdma_cleanup);

net/sunrpc/xprtrdma/xprt_rdma.h

@@ -0,0 +1,402 @@
+/*
+ * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the BSD-type
+ * license below:
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *      Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *
+ *      Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following
+ *      disclaimer in the documentation and/or other materials provided
+ *      with the distribution.
+ *
+ *      Neither the name of the Network Appliance, Inc. nor the names of
+ *      its contributors may be used to endorse or promote products
+ *      derived from this software without specific prior written
+ *      permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
+#define _LINUX_SUNRPC_XPRT_RDMA_H
+
+#include <linux/wait.h> 		/* wait_queue_head_t, etc */
+#include <linux/spinlock.h> 		/* spinlock_t, etc */
+#include <linux/atomic.h>			/* atomic_t, etc */
+#include <linux/workqueue.h>		/* struct work_struct */
+
+#include <rdma/rdma_cm.h>		/* RDMA connection api */
+#include <rdma/ib_verbs.h>		/* RDMA verbs api */
+
+#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 */
+
+/*
+ * Interface Adapter -- one per transport instance
+ */
+struct rpcrdma_ia {
+	rwlock_t		ri_qplock;
+	struct rdma_cm_id 	*ri_id;
+	struct ib_pd		*ri_pd;
+	struct ib_mr		*ri_bind_mem;
+	u32			ri_dma_lkey;
+	int			ri_have_dma_lkey;
+	struct completion	ri_done;
+	int			ri_async_rc;
+	enum rpcrdma_memreg	ri_memreg_strategy;
+	unsigned int		ri_max_frmr_depth;
+};
+
+/*
+ * RDMA Endpoint -- one per transport instance
+ */
+
+#define RPCRDMA_WC_BUDGET	(128)
+#define RPCRDMA_POLLSIZE	(16)
+
+struct rpcrdma_ep {
+	atomic_t		rep_cqcount;
+	int			rep_cqinit;
+	int			rep_connected;
+	struct rpcrdma_ia	*rep_ia;
+	struct ib_qp_init_attr	rep_attr;
+	wait_queue_head_t 	rep_connect_wait;
+	struct ib_sge		rep_pad;	/* holds zeroed pad */
+	struct ib_mr		*rep_pad_mr;	/* holds zeroed pad */
+	void			(*rep_func)(struct rpcrdma_ep *);
+	struct rpc_xprt		*rep_xprt;	/* for rep_func */
+	struct rdma_conn_param	rep_remote_cma;
+	struct sockaddr_storage	rep_remote_addr;
+	struct delayed_work	rep_connect_worker;
+	struct ib_wc		rep_send_wcs[RPCRDMA_POLLSIZE];
+	struct ib_wc		rep_recv_wcs[RPCRDMA_POLLSIZE];
+};
+
+#define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit)
+#define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount)
+
+enum rpcrdma_chunktype {
+	rpcrdma_noch = 0,
+	rpcrdma_readch,
+	rpcrdma_areadch,
+	rpcrdma_writech,
+	rpcrdma_replych
+};
+
+/*
+ * struct rpcrdma_rep -- this structure encapsulates state required to recv
+ * and complete a reply, asychronously. It needs several pieces of
+ * state:
+ *   o recv buffer (posted to provider)
+ *   o ib_sge (also donated to provider)
+ *   o status of reply (length, success or not)
+ *   o bookkeeping state to get run by tasklet (list, etc)
+ *
+ * These are allocated during initialization, per-transport instance;
+ * however, the tasklet execution list itself is global, as it should
+ * always be pretty short.
+ *
+ * N of these are associated with a transport instance, and stored in
+ * struct rpcrdma_buffer. N is the max number of outstanding requests.
+ */
+
+/* temporary static scatter/gather max */
+#define RPCRDMA_MAX_DATA_SEGS	(64)	/* max scatter/gather */
+#define RPCRDMA_MAX_SEGS 	(RPCRDMA_MAX_DATA_SEGS + 2) /* head+tail = 2 */
+#define MAX_RPCRDMAHDR	(\
+	/* max supported RPC/RDMA header */ \
+	sizeof(struct rpcrdma_msg) + (2 * sizeof(u32)) + \
+	(sizeof(struct rpcrdma_read_chunk) * RPCRDMA_MAX_SEGS) + sizeof(u32))
+
+struct rpcrdma_buffer;
+
+struct rpcrdma_rep {
+	unsigned int	rr_len;		/* actual received reply length */
+	struct rpcrdma_buffer *rr_buffer; /* home base for this structure */
+	struct rpc_xprt	*rr_xprt;	/* needed for request/reply matching */
+	void (*rr_func)(struct rpcrdma_rep *);/* called by tasklet in softint */
+	struct list_head rr_list;	/* tasklet list */
+	struct ib_sge	rr_iov;		/* for posting */
+	struct ib_mr	*rr_handle;	/* handle for mem in rr_iov */
+	char	rr_base[MAX_RPCRDMAHDR]; /* minimal inline receive buffer */
+};
+
+/*
+ * struct rpcrdma_mw - external memory region metadata
+ *
+ * An external memory region is any buffer or page that is registered
+ * on the fly (ie, not pre-registered).
+ *
+ * Each rpcrdma_buffer has a list of free MWs anchored in rb_mws. During
+ * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
+ * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
+ * track of registration metadata while each RPC is pending.
+ * rpcrdma_deregister_external() uses this metadata to unmap and
+ * release these resources when an RPC is complete.
+ */
+enum rpcrdma_frmr_state {
+	FRMR_IS_INVALID,	/* ready to be used */
+	FRMR_IS_VALID,		/* in use */
+	FRMR_IS_STALE,		/* failed completion */
+};
+
+struct rpcrdma_frmr {
+	struct ib_fast_reg_page_list	*fr_pgl;
+	struct ib_mr			*fr_mr;
+	enum rpcrdma_frmr_state		fr_state;
+};
+
+struct rpcrdma_mw {
+	union {
+		struct ib_fmr		*fmr;
+		struct rpcrdma_frmr	frmr;
+	} r;
+	struct list_head	mw_list;
+	struct list_head	mw_all;
+};
+
+/*
+ * struct rpcrdma_req -- structure central to the request/reply sequence.
+ *
+ * N of these are associated with a transport instance, and stored in
+ * struct rpcrdma_buffer. N is the max number of outstanding requests.
+ *
+ * It includes pre-registered buffer memory for send AND recv.
+ * The recv buffer, however, is not owned by this structure, and
+ * is "donated" to the hardware when a recv is posted. When a
+ * reply is handled, the recv buffer used is given back to the
+ * struct rpcrdma_req associated with the request.
+ *
+ * In addition to the basic memory, this structure includes an array
+ * of iovs for send operations. The reason is that the iovs passed to
+ * ib_post_{send,recv} must not be modified until the work request
+ * completes.
+ *
+ * NOTES:
+ *   o RPCRDMA_MAX_SEGS is the max number of addressible chunk elements we
+ *     marshal. The number needed varies depending on the iov lists that
+ *     are passed to us, the memory registration mode we are in, and if
+ *     physical addressing is used, the layout.
+ */
+
+struct rpcrdma_mr_seg {		/* chunk descriptors */
+	union {				/* chunk memory handles */
+		struct ib_mr	*rl_mr;		/* if registered directly */
+		struct rpcrdma_mw *rl_mw;	/* if registered from region */
+	} mr_chunk;
+	u64		mr_base;	/* registration result */
+	u32		mr_rkey;	/* registration result */
+	u32		mr_len;		/* length of chunk or segment */
+	int		mr_nsegs;	/* number of segments in chunk or 0 */
+	enum dma_data_direction	mr_dir;	/* segment mapping direction */
+	dma_addr_t	mr_dma;		/* segment mapping address */
+	size_t		mr_dmalen;	/* segment mapping length */
+	struct page	*mr_page;	/* owning page, if any */
+	char		*mr_offset;	/* kva if no page, else offset */
+};
+
+struct rpcrdma_req {
+	size_t 		rl_size;	/* actual length of buffer */
+	unsigned int	rl_niovs;	/* 0, 2 or 4 */
+	unsigned int	rl_nchunks;	/* non-zero if chunks */
+	unsigned int	rl_connect_cookie;	/* retry detection */
+	enum rpcrdma_chunktype	rl_rtype, rl_wtype;
+	struct rpcrdma_buffer *rl_buffer; /* home base for this structure */
+	struct rpcrdma_rep	*rl_reply;/* holder for reply buffer */
+	struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];/* chunk segments */
+	struct ib_sge	rl_send_iov[4];	/* for active requests */
+	struct ib_sge	rl_iov;		/* for posting */
+	struct ib_mr	*rl_handle;	/* handle for mem in rl_iov */
+	char		rl_base[MAX_RPCRDMAHDR]; /* start of actual buffer */
+	__u32 		rl_xdr_buf[0];	/* start of returned rpc rq_buffer */
+};
+#define rpcr_to_rdmar(r) \
+	container_of((r)->rq_buffer, struct rpcrdma_req, rl_xdr_buf[0])
+
+/*
+ * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
+ * inline requests/replies, and client/server credits.
+ *
+ * One of these is associated with a transport instance
+ */
+struct rpcrdma_buffer {
+	spinlock_t	rb_lock;	/* protects indexes */
+	atomic_t	rb_credits;	/* most recent server credits */
+	int		rb_max_requests;/* client max requests */
+	struct list_head rb_mws;	/* optional memory windows/fmrs/frmrs */
+	struct list_head rb_all;
+	int		rb_send_index;
+	struct rpcrdma_req	**rb_send_bufs;
+	int		rb_recv_index;
+	struct rpcrdma_rep	**rb_recv_bufs;
+	char		*rb_pool;
+};
+#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
+
+/*
+ * Internal structure for transport instance creation. This
+ * exists primarily for modularity.
+ *
+ * This data should be set with mount options
+ */
+struct rpcrdma_create_data_internal {
+	struct sockaddr_storage	addr;	/* RDMA server address */
+	unsigned int	max_requests;	/* max requests (slots) in flight */
+	unsigned int	rsize;		/* mount rsize - max read hdr+data */
+	unsigned int	wsize;		/* mount wsize - max write hdr+data */
+	unsigned int	inline_rsize;	/* max non-rdma read data payload */
+	unsigned int	inline_wsize;	/* max non-rdma write data payload */
+	unsigned int	padding;	/* non-rdma write header padding */
+};
+
+#define RPCRDMA_INLINE_READ_THRESHOLD(rq) \
+	(rpcx_to_rdmad(rq->rq_xprt).inline_rsize)
+
+#define RPCRDMA_INLINE_WRITE_THRESHOLD(rq)\
+	(rpcx_to_rdmad(rq->rq_xprt).inline_wsize)
+
+#define RPCRDMA_INLINE_PAD_VALUE(rq)\
+	rpcx_to_rdmad(rq->rq_xprt).padding
+
+/*
+ * Statistics for RPCRDMA
+ */
+struct rpcrdma_stats {
+	unsigned long		read_chunk_count;
+	unsigned long		write_chunk_count;
+	unsigned long		reply_chunk_count;
+
+	unsigned long long	total_rdma_request;
+	unsigned long long	total_rdma_reply;
+
+	unsigned long long	pullup_copy_count;
+	unsigned long long	fixup_copy_count;
+	unsigned long		hardway_register_count;
+	unsigned long		failed_marshal_count;
+	unsigned long		bad_reply_count;
+};
+
+/*
+ * RPCRDMA transport -- encapsulates the structures above for
+ * integration with RPC.
+ *
+ * The contained structures are embedded, not pointers,
+ * for convenience. This structure need not be visible externally.
+ *
+ * It is allocated and initialized during mount, and released
+ * during unmount.
+ */
+struct rpcrdma_xprt {
+	struct rpc_xprt		xprt;
+	struct rpcrdma_ia	rx_ia;
+	struct rpcrdma_ep	rx_ep;
+	struct rpcrdma_buffer	rx_buf;
+	struct rpcrdma_create_data_internal rx_data;
+	struct delayed_work	rdma_connect;
+	struct rpcrdma_stats	rx_stats;
+};
+
+#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, xprt)
+#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
+
+/* Setting this to 0 ensures interoperability with early servers.
+ * Setting this to 1 enhances certain unaligned read/write performance.
+ * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
+extern int xprt_rdma_pad_optimize;
+
+/*
+ * Interface Adapter calls - xprtrdma/verbs.c
+ */
+int rpcrdma_ia_open(struct rpcrdma_xprt *, struct sockaddr *, int);
+void rpcrdma_ia_close(struct rpcrdma_ia *);
+
+/*
+ * Endpoint calls - xprtrdma/verbs.c
+ */
+int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
+				struct rpcrdma_create_data_internal *);
+void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
+int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
+void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
+
+int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
+				struct rpcrdma_req *);
+int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_ep *,
+				struct rpcrdma_rep *);
+
+/*
+ * Buffer calls - xprtrdma/verbs.c
+ */
+int rpcrdma_buffer_create(struct rpcrdma_buffer *, struct rpcrdma_ep *,
+				struct rpcrdma_ia *,
+				struct rpcrdma_create_data_internal *);
+void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
+
+struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
+void rpcrdma_buffer_put(struct rpcrdma_req *);
+void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
+void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
+
+int rpcrdma_register_internal(struct rpcrdma_ia *, void *, int,
+				struct ib_mr **, struct ib_sge *);
+int rpcrdma_deregister_internal(struct rpcrdma_ia *,
+				struct ib_mr *, struct ib_sge *);
+
+int rpcrdma_register_external(struct rpcrdma_mr_seg *,
+				int, int, struct rpcrdma_xprt *);
+int rpcrdma_deregister_external(struct rpcrdma_mr_seg *,
+				struct rpcrdma_xprt *);
+
+/*
+ * RPC/RDMA connection management calls - xprtrdma/rpc_rdma.c
+ */
+void rpcrdma_connect_worker(struct work_struct *);
+void rpcrdma_conn_func(struct rpcrdma_ep *);
+void rpcrdma_reply_handler(struct rpcrdma_rep *);
+
+/*
+ * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
+ */
+ssize_t rpcrdma_marshal_chunks(struct rpc_rqst *, ssize_t);
+int rpcrdma_marshal_req(struct rpc_rqst *);
+size_t rpcrdma_max_payload(struct rpcrdma_xprt *);
+
+/* Temporary NFS request map cache. Created in svc_rdma.c  */
+extern struct kmem_cache *svc_rdma_map_cachep;
+/* WR context cache. Created in svc_rdma.c  */
+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 */