xref: /openbmc/linux/net/sunrpc/xprtrdma/rpc_rdma.c (revision cd238eff)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (c) 2014-2017 Oracle.  All rights reserved.
4  * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the BSD-type
10  * license below:
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  *
16  *      Redistributions of source code must retain the above copyright
17  *      notice, this list of conditions and the following disclaimer.
18  *
19  *      Redistributions in binary form must reproduce the above
20  *      copyright notice, this list of conditions and the following
21  *      disclaimer in the documentation and/or other materials provided
22  *      with the distribution.
23  *
24  *      Neither the name of the Network Appliance, Inc. nor the names of
25  *      its contributors may be used to endorse or promote products
26  *      derived from this software without specific prior written
27  *      permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*
43  * rpc_rdma.c
44  *
45  * This file contains the guts of the RPC RDMA protocol, and
46  * does marshaling/unmarshaling, etc. It is also where interfacing
47  * to the Linux RPC framework lives.
48  */
49 
50 #include <linux/highmem.h>
51 
52 #include <linux/sunrpc/svc_rdma.h>
53 
54 #include "xprt_rdma.h"
55 #include <trace/events/rpcrdma.h>
56 
57 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
58 # define RPCDBG_FACILITY	RPCDBG_TRANS
59 #endif
60 
61 /* Returns size of largest RPC-over-RDMA header in a Call message
62  *
63  * The largest Call header contains a full-size Read list and a
64  * minimal Reply chunk.
65  */
66 static unsigned int rpcrdma_max_call_header_size(unsigned int maxsegs)
67 {
68 	unsigned int size;
69 
70 	/* Fixed header fields and list discriminators */
71 	size = RPCRDMA_HDRLEN_MIN;
72 
73 	/* Maximum Read list size */
74 	size = maxsegs * rpcrdma_readchunk_maxsz * sizeof(__be32);
75 
76 	/* Minimal Read chunk size */
77 	size += sizeof(__be32);	/* segment count */
78 	size += rpcrdma_segment_maxsz * sizeof(__be32);
79 	size += sizeof(__be32);	/* list discriminator */
80 
81 	dprintk("RPC:       %s: max call header size = %u\n",
82 		__func__, size);
83 	return size;
84 }
85 
86 /* Returns size of largest RPC-over-RDMA header in a Reply message
87  *
88  * There is only one Write list or one Reply chunk per Reply
89  * message.  The larger list is the Write list.
90  */
91 static unsigned int rpcrdma_max_reply_header_size(unsigned int maxsegs)
92 {
93 	unsigned int size;
94 
95 	/* Fixed header fields and list discriminators */
96 	size = RPCRDMA_HDRLEN_MIN;
97 
98 	/* Maximum Write list size */
99 	size = sizeof(__be32);		/* segment count */
100 	size += maxsegs * rpcrdma_segment_maxsz * sizeof(__be32);
101 	size += sizeof(__be32);	/* list discriminator */
102 
103 	dprintk("RPC:       %s: max reply header size = %u\n",
104 		__func__, size);
105 	return size;
106 }
107 
108 /**
109  * rpcrdma_set_max_header_sizes - Initialize inline payload sizes
110  * @r_xprt: transport instance to initialize
111  *
112  * The max_inline fields contain the maximum size of an RPC message
113  * so the marshaling code doesn't have to repeat this calculation
114  * for every RPC.
115  */
116 void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *r_xprt)
117 {
118 	unsigned int maxsegs = r_xprt->rx_ia.ri_max_segs;
119 	struct rpcrdma_ep *ep = &r_xprt->rx_ep;
120 
121 	ep->rep_max_inline_send =
122 		ep->rep_inline_send - rpcrdma_max_call_header_size(maxsegs);
123 	ep->rep_max_inline_recv =
124 		ep->rep_inline_recv - rpcrdma_max_reply_header_size(maxsegs);
125 }
126 
127 /* The client can send a request inline as long as the RPCRDMA header
128  * plus the RPC call fit under the transport's inline limit. If the
129  * combined call message size exceeds that limit, the client must use
130  * a Read chunk for this operation.
131  *
132  * A Read chunk is also required if sending the RPC call inline would
133  * exceed this device's max_sge limit.
134  */
135 static bool rpcrdma_args_inline(struct rpcrdma_xprt *r_xprt,
136 				struct rpc_rqst *rqst)
137 {
138 	struct xdr_buf *xdr = &rqst->rq_snd_buf;
139 	unsigned int count, remaining, offset;
140 
141 	if (xdr->len > r_xprt->rx_ep.rep_max_inline_send)
142 		return false;
143 
144 	if (xdr->page_len) {
145 		remaining = xdr->page_len;
146 		offset = offset_in_page(xdr->page_base);
147 		count = RPCRDMA_MIN_SEND_SGES;
148 		while (remaining) {
149 			remaining -= min_t(unsigned int,
150 					   PAGE_SIZE - offset, remaining);
151 			offset = 0;
152 			if (++count > r_xprt->rx_ia.ri_max_send_sges)
153 				return false;
154 		}
155 	}
156 
157 	return true;
158 }
159 
160 /* The client can't know how large the actual reply will be. Thus it
161  * plans for the largest possible reply for that particular ULP
162  * operation. If the maximum combined reply message size exceeds that
163  * limit, the client must provide a write list or a reply chunk for
164  * this request.
165  */
166 static bool rpcrdma_results_inline(struct rpcrdma_xprt *r_xprt,
167 				   struct rpc_rqst *rqst)
168 {
169 	return rqst->rq_rcv_buf.buflen <= r_xprt->rx_ep.rep_max_inline_recv;
170 }
171 
172 /* The client is required to provide a Reply chunk if the maximum
173  * size of the non-payload part of the RPC Reply is larger than
174  * the inline threshold.
175  */
176 static bool
177 rpcrdma_nonpayload_inline(const struct rpcrdma_xprt *r_xprt,
178 			  const struct rpc_rqst *rqst)
179 {
180 	const struct xdr_buf *buf = &rqst->rq_rcv_buf;
181 
182 	return (buf->head[0].iov_len + buf->tail[0].iov_len) <
183 		r_xprt->rx_ep.rep_max_inline_recv;
184 }
185 
186 /* Split @vec on page boundaries into SGEs. FMR registers pages, not
187  * a byte range. Other modes coalesce these SGEs into a single MR
188  * when they can.
189  *
190  * Returns pointer to next available SGE, and bumps the total number
191  * of SGEs consumed.
192  */
193 static struct rpcrdma_mr_seg *
194 rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg,
195 		     unsigned int *n)
196 {
197 	u32 remaining, page_offset;
198 	char *base;
199 
200 	base = vec->iov_base;
201 	page_offset = offset_in_page(base);
202 	remaining = vec->iov_len;
203 	while (remaining) {
204 		seg->mr_page = NULL;
205 		seg->mr_offset = base;
206 		seg->mr_len = min_t(u32, PAGE_SIZE - page_offset, remaining);
207 		remaining -= seg->mr_len;
208 		base += seg->mr_len;
209 		++seg;
210 		++(*n);
211 		page_offset = 0;
212 	}
213 	return seg;
214 }
215 
216 /* Convert @xdrbuf into SGEs no larger than a page each. As they
217  * are registered, these SGEs are then coalesced into RDMA segments
218  * when the selected memreg mode supports it.
219  *
220  * Returns positive number of SGEs consumed, or a negative errno.
221  */
222 
223 static int
224 rpcrdma_convert_iovs(struct rpcrdma_xprt *r_xprt, struct xdr_buf *xdrbuf,
225 		     unsigned int pos, enum rpcrdma_chunktype type,
226 		     struct rpcrdma_mr_seg *seg)
227 {
228 	unsigned long page_base;
229 	unsigned int len, n;
230 	struct page **ppages;
231 
232 	n = 0;
233 	if (pos == 0)
234 		seg = rpcrdma_convert_kvec(&xdrbuf->head[0], seg, &n);
235 
236 	len = xdrbuf->page_len;
237 	ppages = xdrbuf->pages + (xdrbuf->page_base >> PAGE_SHIFT);
238 	page_base = offset_in_page(xdrbuf->page_base);
239 	while (len) {
240 		/* ACL likes to be lazy in allocating pages - ACLs
241 		 * are small by default but can get huge.
242 		 */
243 		if (unlikely(xdrbuf->flags & XDRBUF_SPARSE_PAGES)) {
244 			if (!*ppages)
245 				*ppages = alloc_page(GFP_NOWAIT | __GFP_NOWARN);
246 			if (!*ppages)
247 				return -ENOBUFS;
248 		}
249 		seg->mr_page = *ppages;
250 		seg->mr_offset = (char *)page_base;
251 		seg->mr_len = min_t(u32, PAGE_SIZE - page_base, len);
252 		len -= seg->mr_len;
253 		++ppages;
254 		++seg;
255 		++n;
256 		page_base = 0;
257 	}
258 
259 	/* When encoding a Read chunk, the tail iovec contains an
260 	 * XDR pad and may be omitted.
261 	 */
262 	if (type == rpcrdma_readch && r_xprt->rx_ia.ri_implicit_roundup)
263 		goto out;
264 
265 	/* When encoding a Write chunk, some servers need to see an
266 	 * extra segment for non-XDR-aligned Write chunks. The upper
267 	 * layer provides space in the tail iovec that may be used
268 	 * for this purpose.
269 	 */
270 	if (type == rpcrdma_writech && r_xprt->rx_ia.ri_implicit_roundup)
271 		goto out;
272 
273 	if (xdrbuf->tail[0].iov_len)
274 		seg = rpcrdma_convert_kvec(&xdrbuf->tail[0], seg, &n);
275 
276 out:
277 	if (unlikely(n > RPCRDMA_MAX_SEGS))
278 		return -EIO;
279 	return n;
280 }
281 
282 static inline int
283 encode_item_present(struct xdr_stream *xdr)
284 {
285 	__be32 *p;
286 
287 	p = xdr_reserve_space(xdr, sizeof(*p));
288 	if (unlikely(!p))
289 		return -EMSGSIZE;
290 
291 	*p = xdr_one;
292 	return 0;
293 }
294 
295 static inline int
296 encode_item_not_present(struct xdr_stream *xdr)
297 {
298 	__be32 *p;
299 
300 	p = xdr_reserve_space(xdr, sizeof(*p));
301 	if (unlikely(!p))
302 		return -EMSGSIZE;
303 
304 	*p = xdr_zero;
305 	return 0;
306 }
307 
308 static void
309 xdr_encode_rdma_segment(__be32 *iptr, struct rpcrdma_mr *mr)
310 {
311 	*iptr++ = cpu_to_be32(mr->mr_handle);
312 	*iptr++ = cpu_to_be32(mr->mr_length);
313 	xdr_encode_hyper(iptr, mr->mr_offset);
314 }
315 
316 static int
317 encode_rdma_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr)
318 {
319 	__be32 *p;
320 
321 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
322 	if (unlikely(!p))
323 		return -EMSGSIZE;
324 
325 	xdr_encode_rdma_segment(p, mr);
326 	return 0;
327 }
328 
329 static int
330 encode_read_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr,
331 		    u32 position)
332 {
333 	__be32 *p;
334 
335 	p = xdr_reserve_space(xdr, 6 * sizeof(*p));
336 	if (unlikely(!p))
337 		return -EMSGSIZE;
338 
339 	*p++ = xdr_one;			/* Item present */
340 	*p++ = cpu_to_be32(position);
341 	xdr_encode_rdma_segment(p, mr);
342 	return 0;
343 }
344 
345 /* Register and XDR encode the Read list. Supports encoding a list of read
346  * segments that belong to a single read chunk.
347  *
348  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
349  *
350  *  Read chunklist (a linked list):
351  *   N elements, position P (same P for all chunks of same arg!):
352  *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
353  *
354  * Returns zero on success, or a negative errno if a failure occurred.
355  * @xdr is advanced to the next position in the stream.
356  *
357  * Only a single @pos value is currently supported.
358  */
359 static noinline int
360 rpcrdma_encode_read_list(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req,
361 			 struct rpc_rqst *rqst, enum rpcrdma_chunktype rtype)
362 {
363 	struct xdr_stream *xdr = &req->rl_stream;
364 	struct rpcrdma_mr_seg *seg;
365 	struct rpcrdma_mr *mr;
366 	unsigned int pos;
367 	int nsegs;
368 
369 	pos = rqst->rq_snd_buf.head[0].iov_len;
370 	if (rtype == rpcrdma_areadch)
371 		pos = 0;
372 	seg = req->rl_segments;
373 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_snd_buf, pos,
374 				     rtype, seg);
375 	if (nsegs < 0)
376 		return nsegs;
377 
378 	do {
379 		seg = frwr_map(r_xprt, seg, nsegs, false, rqst->rq_xid, &mr);
380 		if (IS_ERR(seg))
381 			return PTR_ERR(seg);
382 		rpcrdma_mr_push(mr, &req->rl_registered);
383 
384 		if (encode_read_segment(xdr, mr, pos) < 0)
385 			return -EMSGSIZE;
386 
387 		trace_xprtrdma_chunk_read(rqst->rq_task, pos, mr, nsegs);
388 		r_xprt->rx_stats.read_chunk_count++;
389 		nsegs -= mr->mr_nents;
390 	} while (nsegs);
391 
392 	return 0;
393 }
394 
395 /* Register and XDR encode the Write list. Supports encoding a list
396  * containing one array of plain segments that belong to a single
397  * write chunk.
398  *
399  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
400  *
401  *  Write chunklist (a list of (one) counted array):
402  *   N elements:
403  *    1 - N - HLOO - HLOO - ... - HLOO - 0
404  *
405  * Returns zero on success, or a negative errno if a failure occurred.
406  * @xdr is advanced to the next position in the stream.
407  *
408  * Only a single Write chunk is currently supported.
409  */
410 static noinline int
411 rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req,
412 			  struct rpc_rqst *rqst, enum rpcrdma_chunktype wtype)
413 {
414 	struct xdr_stream *xdr = &req->rl_stream;
415 	struct rpcrdma_mr_seg *seg;
416 	struct rpcrdma_mr *mr;
417 	int nsegs, nchunks;
418 	__be32 *segcount;
419 
420 	seg = req->rl_segments;
421 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf,
422 				     rqst->rq_rcv_buf.head[0].iov_len,
423 				     wtype, seg);
424 	if (nsegs < 0)
425 		return nsegs;
426 
427 	if (encode_item_present(xdr) < 0)
428 		return -EMSGSIZE;
429 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
430 	if (unlikely(!segcount))
431 		return -EMSGSIZE;
432 	/* Actual value encoded below */
433 
434 	nchunks = 0;
435 	do {
436 		seg = frwr_map(r_xprt, seg, nsegs, true, rqst->rq_xid, &mr);
437 		if (IS_ERR(seg))
438 			return PTR_ERR(seg);
439 		rpcrdma_mr_push(mr, &req->rl_registered);
440 
441 		if (encode_rdma_segment(xdr, mr) < 0)
442 			return -EMSGSIZE;
443 
444 		trace_xprtrdma_chunk_write(rqst->rq_task, mr, nsegs);
445 		r_xprt->rx_stats.write_chunk_count++;
446 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
447 		nchunks++;
448 		nsegs -= mr->mr_nents;
449 	} while (nsegs);
450 
451 	/* Update count of segments in this Write chunk */
452 	*segcount = cpu_to_be32(nchunks);
453 
454 	return 0;
455 }
456 
457 /* Register and XDR encode the Reply chunk. Supports encoding an array
458  * of plain segments that belong to a single write (reply) chunk.
459  *
460  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
461  *
462  *  Reply chunk (a counted array):
463  *   N elements:
464  *    1 - N - HLOO - HLOO - ... - HLOO
465  *
466  * Returns zero on success, or a negative errno if a failure occurred.
467  * @xdr is advanced to the next position in the stream.
468  */
469 static noinline int
470 rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req,
471 			   struct rpc_rqst *rqst, enum rpcrdma_chunktype wtype)
472 {
473 	struct xdr_stream *xdr = &req->rl_stream;
474 	struct rpcrdma_mr_seg *seg;
475 	struct rpcrdma_mr *mr;
476 	int nsegs, nchunks;
477 	__be32 *segcount;
478 
479 	seg = req->rl_segments;
480 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf, 0, wtype, seg);
481 	if (nsegs < 0)
482 		return nsegs;
483 
484 	if (encode_item_present(xdr) < 0)
485 		return -EMSGSIZE;
486 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
487 	if (unlikely(!segcount))
488 		return -EMSGSIZE;
489 	/* Actual value encoded below */
490 
491 	nchunks = 0;
492 	do {
493 		seg = frwr_map(r_xprt, seg, nsegs, true, rqst->rq_xid, &mr);
494 		if (IS_ERR(seg))
495 			return PTR_ERR(seg);
496 		rpcrdma_mr_push(mr, &req->rl_registered);
497 
498 		if (encode_rdma_segment(xdr, mr) < 0)
499 			return -EMSGSIZE;
500 
501 		trace_xprtrdma_chunk_reply(rqst->rq_task, mr, nsegs);
502 		r_xprt->rx_stats.reply_chunk_count++;
503 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
504 		nchunks++;
505 		nsegs -= mr->mr_nents;
506 	} while (nsegs);
507 
508 	/* Update count of segments in the Reply chunk */
509 	*segcount = cpu_to_be32(nchunks);
510 
511 	return 0;
512 }
513 
514 /**
515  * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
516  * @sc: sendctx containing SGEs to unmap
517  *
518  */
519 void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
520 {
521 	struct ib_sge *sge;
522 
523 	/* The first two SGEs contain the transport header and
524 	 * the inline buffer. These are always left mapped so
525 	 * they can be cheaply re-used.
526 	 */
527 	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
528 	     ++sge, --sc->sc_unmap_count)
529 		ib_dma_unmap_page(sc->sc_device, sge->addr, sge->length,
530 				  DMA_TO_DEVICE);
531 
532 	if (test_and_clear_bit(RPCRDMA_REQ_F_TX_RESOURCES,
533 			       &sc->sc_req->rl_flags))
534 		wake_up_bit(&sc->sc_req->rl_flags, RPCRDMA_REQ_F_TX_RESOURCES);
535 }
536 
537 /* Prepare an SGE for the RPC-over-RDMA transport header.
538  */
539 static bool rpcrdma_prepare_hdr_sge(struct rpcrdma_xprt *r_xprt,
540 				    struct rpcrdma_req *req, u32 len)
541 {
542 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
543 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
544 	struct ib_sge *sge = sc->sc_sges;
545 
546 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
547 		goto out_regbuf;
548 	sge->addr = rdmab_addr(rb);
549 	sge->length = len;
550 	sge->lkey = rdmab_lkey(rb);
551 
552 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
553 				      DMA_TO_DEVICE);
554 	sc->sc_wr.num_sge++;
555 	return true;
556 
557 out_regbuf:
558 	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
559 	return false;
560 }
561 
562 /* Prepare the Send SGEs. The head and tail iovec, and each entry
563  * in the page list, gets its own SGE.
564  */
565 static bool rpcrdma_prepare_msg_sges(struct rpcrdma_xprt *r_xprt,
566 				     struct rpcrdma_req *req,
567 				     struct xdr_buf *xdr,
568 				     enum rpcrdma_chunktype rtype)
569 {
570 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
571 	unsigned int sge_no, page_base, len, remaining;
572 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
573 	struct ib_sge *sge = sc->sc_sges;
574 	struct page *page, **ppages;
575 
576 	/* The head iovec is straightforward, as it is already
577 	 * DMA-mapped. Sync the content that has changed.
578 	 */
579 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
580 		goto out_regbuf;
581 	sc->sc_device = rdmab_device(rb);
582 	sge_no = 1;
583 	sge[sge_no].addr = rdmab_addr(rb);
584 	sge[sge_no].length = xdr->head[0].iov_len;
585 	sge[sge_no].lkey = rdmab_lkey(rb);
586 	ib_dma_sync_single_for_device(rdmab_device(rb), sge[sge_no].addr,
587 				      sge[sge_no].length, DMA_TO_DEVICE);
588 
589 	/* If there is a Read chunk, the page list is being handled
590 	 * via explicit RDMA, and thus is skipped here. However, the
591 	 * tail iovec may include an XDR pad for the page list, as
592 	 * well as additional content, and may not reside in the
593 	 * same page as the head iovec.
594 	 */
595 	if (rtype == rpcrdma_readch) {
596 		len = xdr->tail[0].iov_len;
597 
598 		/* Do not include the tail if it is only an XDR pad */
599 		if (len < 4)
600 			goto out;
601 
602 		page = virt_to_page(xdr->tail[0].iov_base);
603 		page_base = offset_in_page(xdr->tail[0].iov_base);
604 
605 		/* If the content in the page list is an odd length,
606 		 * xdr_write_pages() has added a pad at the beginning
607 		 * of the tail iovec. Force the tail's non-pad content
608 		 * to land at the next XDR position in the Send message.
609 		 */
610 		page_base += len & 3;
611 		len -= len & 3;
612 		goto map_tail;
613 	}
614 
615 	/* If there is a page list present, temporarily DMA map
616 	 * and prepare an SGE for each page to be sent.
617 	 */
618 	if (xdr->page_len) {
619 		ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
620 		page_base = offset_in_page(xdr->page_base);
621 		remaining = xdr->page_len;
622 		while (remaining) {
623 			sge_no++;
624 			if (sge_no > RPCRDMA_MAX_SEND_SGES - 2)
625 				goto out_mapping_overflow;
626 
627 			len = min_t(u32, PAGE_SIZE - page_base, remaining);
628 			sge[sge_no].addr =
629 				ib_dma_map_page(rdmab_device(rb), *ppages,
630 						page_base, len, DMA_TO_DEVICE);
631 			if (ib_dma_mapping_error(rdmab_device(rb),
632 						 sge[sge_no].addr))
633 				goto out_mapping_err;
634 			sge[sge_no].length = len;
635 			sge[sge_no].lkey = rdmab_lkey(rb);
636 
637 			sc->sc_unmap_count++;
638 			ppages++;
639 			remaining -= len;
640 			page_base = 0;
641 		}
642 	}
643 
644 	/* The tail iovec is not always constructed in the same
645 	 * page where the head iovec resides (see, for example,
646 	 * gss_wrap_req_priv). To neatly accommodate that case,
647 	 * DMA map it separately.
648 	 */
649 	if (xdr->tail[0].iov_len) {
650 		page = virt_to_page(xdr->tail[0].iov_base);
651 		page_base = offset_in_page(xdr->tail[0].iov_base);
652 		len = xdr->tail[0].iov_len;
653 
654 map_tail:
655 		sge_no++;
656 		sge[sge_no].addr =
657 			ib_dma_map_page(rdmab_device(rb), page, page_base, len,
658 					DMA_TO_DEVICE);
659 		if (ib_dma_mapping_error(rdmab_device(rb), sge[sge_no].addr))
660 			goto out_mapping_err;
661 		sge[sge_no].length = len;
662 		sge[sge_no].lkey = rdmab_lkey(rb);
663 		sc->sc_unmap_count++;
664 	}
665 
666 out:
667 	sc->sc_wr.num_sge += sge_no;
668 	if (sc->sc_unmap_count)
669 		__set_bit(RPCRDMA_REQ_F_TX_RESOURCES, &req->rl_flags);
670 	return true;
671 
672 out_regbuf:
673 	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
674 	return false;
675 
676 out_mapping_overflow:
677 	rpcrdma_sendctx_unmap(sc);
678 	pr_err("rpcrdma: too many Send SGEs (%u)\n", sge_no);
679 	return false;
680 
681 out_mapping_err:
682 	rpcrdma_sendctx_unmap(sc);
683 	trace_xprtrdma_dma_maperr(sge[sge_no].addr);
684 	return false;
685 }
686 
687 /**
688  * rpcrdma_prepare_send_sges - Construct SGEs for a Send WR
689  * @r_xprt: controlling transport
690  * @req: context of RPC Call being marshalled
691  * @hdrlen: size of transport header, in bytes
692  * @xdr: xdr_buf containing RPC Call
693  * @rtype: chunk type being encoded
694  *
695  * Returns 0 on success; otherwise a negative errno is returned.
696  */
697 int
698 rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
699 			  struct rpcrdma_req *req, u32 hdrlen,
700 			  struct xdr_buf *xdr, enum rpcrdma_chunktype rtype)
701 {
702 	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
703 	if (!req->rl_sendctx)
704 		return -EAGAIN;
705 	req->rl_sendctx->sc_wr.num_sge = 0;
706 	req->rl_sendctx->sc_unmap_count = 0;
707 	req->rl_sendctx->sc_req = req;
708 	__clear_bit(RPCRDMA_REQ_F_TX_RESOURCES, &req->rl_flags);
709 
710 	if (!rpcrdma_prepare_hdr_sge(r_xprt, req, hdrlen))
711 		return -EIO;
712 
713 	if (rtype != rpcrdma_areadch)
714 		if (!rpcrdma_prepare_msg_sges(r_xprt, req, xdr, rtype))
715 			return -EIO;
716 
717 	return 0;
718 }
719 
720 /**
721  * rpcrdma_marshal_req - Marshal and send one RPC request
722  * @r_xprt: controlling transport
723  * @rqst: RPC request to be marshaled
724  *
725  * For the RPC in "rqst", this function:
726  *  - Chooses the transfer mode (eg., RDMA_MSG or RDMA_NOMSG)
727  *  - Registers Read, Write, and Reply chunks
728  *  - Constructs the transport header
729  *  - Posts a Send WR to send the transport header and request
730  *
731  * Returns:
732  *	%0 if the RPC was sent successfully,
733  *	%-ENOTCONN if the connection was lost,
734  *	%-EAGAIN if the caller should call again with the same arguments,
735  *	%-ENOBUFS if the caller should call again after a delay,
736  *	%-EMSGSIZE if the transport header is too small,
737  *	%-EIO if a permanent problem occurred while marshaling.
738  */
739 int
740 rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
741 {
742 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
743 	struct xdr_stream *xdr = &req->rl_stream;
744 	enum rpcrdma_chunktype rtype, wtype;
745 	bool ddp_allowed;
746 	__be32 *p;
747 	int ret;
748 
749 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
750 	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
751 			rqst);
752 
753 	/* Fixed header fields */
754 	ret = -EMSGSIZE;
755 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
756 	if (!p)
757 		goto out_err;
758 	*p++ = rqst->rq_xid;
759 	*p++ = rpcrdma_version;
760 	*p++ = cpu_to_be32(r_xprt->rx_buf.rb_max_requests);
761 
762 	/* When the ULP employs a GSS flavor that guarantees integrity
763 	 * or privacy, direct data placement of individual data items
764 	 * is not allowed.
765 	 */
766 	ddp_allowed = !(rqst->rq_cred->cr_auth->au_flags &
767 						RPCAUTH_AUTH_DATATOUCH);
768 
769 	/*
770 	 * Chunks needed for results?
771 	 *
772 	 * o If the expected result is under the inline threshold, all ops
773 	 *   return as inline.
774 	 * o Large read ops return data as write chunk(s), header as
775 	 *   inline.
776 	 * o Large non-read ops return as a single reply chunk.
777 	 */
778 	if (rpcrdma_results_inline(r_xprt, rqst))
779 		wtype = rpcrdma_noch;
780 	else if ((ddp_allowed && rqst->rq_rcv_buf.flags & XDRBUF_READ) &&
781 		 rpcrdma_nonpayload_inline(r_xprt, rqst))
782 		wtype = rpcrdma_writech;
783 	else
784 		wtype = rpcrdma_replych;
785 
786 	/*
787 	 * Chunks needed for arguments?
788 	 *
789 	 * o If the total request is under the inline threshold, all ops
790 	 *   are sent as inline.
791 	 * o Large write ops transmit data as read chunk(s), header as
792 	 *   inline.
793 	 * o Large non-write ops are sent with the entire message as a
794 	 *   single read chunk (protocol 0-position special case).
795 	 *
796 	 * This assumes that the upper layer does not present a request
797 	 * that both has a data payload, and whose non-data arguments
798 	 * by themselves are larger than the inline threshold.
799 	 */
800 	if (rpcrdma_args_inline(r_xprt, rqst)) {
801 		*p++ = rdma_msg;
802 		rtype = rpcrdma_noch;
803 	} else if (ddp_allowed && rqst->rq_snd_buf.flags & XDRBUF_WRITE) {
804 		*p++ = rdma_msg;
805 		rtype = rpcrdma_readch;
806 	} else {
807 		r_xprt->rx_stats.nomsg_call_count++;
808 		*p++ = rdma_nomsg;
809 		rtype = rpcrdma_areadch;
810 	}
811 
812 	/* If this is a retransmit, discard previously registered
813 	 * chunks. Very likely the connection has been replaced,
814 	 * so these registrations are invalid and unusable.
815 	 */
816 	while (unlikely(!list_empty(&req->rl_registered))) {
817 		struct rpcrdma_mr *mr;
818 
819 		mr = rpcrdma_mr_pop(&req->rl_registered);
820 		rpcrdma_mr_recycle(mr);
821 	}
822 
823 	/* This implementation supports the following combinations
824 	 * of chunk lists in one RPC-over-RDMA Call message:
825 	 *
826 	 *   - Read list
827 	 *   - Write list
828 	 *   - Reply chunk
829 	 *   - Read list + Reply chunk
830 	 *
831 	 * It might not yet support the following combinations:
832 	 *
833 	 *   - Read list + Write list
834 	 *
835 	 * It does not support the following combinations:
836 	 *
837 	 *   - Write list + Reply chunk
838 	 *   - Read list + Write list + Reply chunk
839 	 *
840 	 * This implementation supports only a single chunk in each
841 	 * Read or Write list. Thus for example the client cannot
842 	 * send a Call message with a Position Zero Read chunk and a
843 	 * regular Read chunk at the same time.
844 	 */
845 	if (rtype != rpcrdma_noch) {
846 		ret = rpcrdma_encode_read_list(r_xprt, req, rqst, rtype);
847 		if (ret)
848 			goto out_err;
849 	}
850 	ret = encode_item_not_present(xdr);
851 	if (ret)
852 		goto out_err;
853 
854 	if (wtype == rpcrdma_writech) {
855 		ret = rpcrdma_encode_write_list(r_xprt, req, rqst, wtype);
856 		if (ret)
857 			goto out_err;
858 	}
859 	ret = encode_item_not_present(xdr);
860 	if (ret)
861 		goto out_err;
862 
863 	if (wtype != rpcrdma_replych)
864 		ret = encode_item_not_present(xdr);
865 	else
866 		ret = rpcrdma_encode_reply_chunk(r_xprt, req, rqst, wtype);
867 	if (ret)
868 		goto out_err;
869 
870 	trace_xprtrdma_marshal(rqst, xdr_stream_pos(xdr), rtype, wtype);
871 
872 	ret = rpcrdma_prepare_send_sges(r_xprt, req, xdr_stream_pos(xdr),
873 					&rqst->rq_snd_buf, rtype);
874 	if (ret)
875 		goto out_err;
876 	return 0;
877 
878 out_err:
879 	trace_xprtrdma_marshal_failed(rqst, ret);
880 	switch (ret) {
881 	case -EAGAIN:
882 		xprt_wait_for_buffer_space(rqst->rq_xprt);
883 		break;
884 	case -ENOBUFS:
885 		break;
886 	default:
887 		r_xprt->rx_stats.failed_marshal_count++;
888 	}
889 	return ret;
890 }
891 
892 /**
893  * rpcrdma_inline_fixup - Scatter inline received data into rqst's iovecs
894  * @rqst: controlling RPC request
895  * @srcp: points to RPC message payload in receive buffer
896  * @copy_len: remaining length of receive buffer content
897  * @pad: Write chunk pad bytes needed (zero for pure inline)
898  *
899  * The upper layer has set the maximum number of bytes it can
900  * receive in each component of rq_rcv_buf. These values are set in
901  * the head.iov_len, page_len, tail.iov_len, and buflen fields.
902  *
903  * Unlike the TCP equivalent (xdr_partial_copy_from_skb), in
904  * many cases this function simply updates iov_base pointers in
905  * rq_rcv_buf to point directly to the received reply data, to
906  * avoid copying reply data.
907  *
908  * Returns the count of bytes which had to be memcopied.
909  */
910 static unsigned long
911 rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
912 {
913 	unsigned long fixup_copy_count;
914 	int i, npages, curlen;
915 	char *destp;
916 	struct page **ppages;
917 	int page_base;
918 
919 	/* The head iovec is redirected to the RPC reply message
920 	 * in the receive buffer, to avoid a memcopy.
921 	 */
922 	rqst->rq_rcv_buf.head[0].iov_base = srcp;
923 	rqst->rq_private_buf.head[0].iov_base = srcp;
924 
925 	/* The contents of the receive buffer that follow
926 	 * head.iov_len bytes are copied into the page list.
927 	 */
928 	curlen = rqst->rq_rcv_buf.head[0].iov_len;
929 	if (curlen > copy_len)
930 		curlen = copy_len;
931 	trace_xprtrdma_fixup(rqst, copy_len, curlen);
932 	srcp += curlen;
933 	copy_len -= curlen;
934 
935 	ppages = rqst->rq_rcv_buf.pages +
936 		(rqst->rq_rcv_buf.page_base >> PAGE_SHIFT);
937 	page_base = offset_in_page(rqst->rq_rcv_buf.page_base);
938 	fixup_copy_count = 0;
939 	if (copy_len && rqst->rq_rcv_buf.page_len) {
940 		int pagelist_len;
941 
942 		pagelist_len = rqst->rq_rcv_buf.page_len;
943 		if (pagelist_len > copy_len)
944 			pagelist_len = copy_len;
945 		npages = PAGE_ALIGN(page_base + pagelist_len) >> PAGE_SHIFT;
946 		for (i = 0; i < npages; i++) {
947 			curlen = PAGE_SIZE - page_base;
948 			if (curlen > pagelist_len)
949 				curlen = pagelist_len;
950 
951 			trace_xprtrdma_fixup_pg(rqst, i, srcp,
952 						copy_len, curlen);
953 			destp = kmap_atomic(ppages[i]);
954 			memcpy(destp + page_base, srcp, curlen);
955 			flush_dcache_page(ppages[i]);
956 			kunmap_atomic(destp);
957 			srcp += curlen;
958 			copy_len -= curlen;
959 			fixup_copy_count += curlen;
960 			pagelist_len -= curlen;
961 			if (!pagelist_len)
962 				break;
963 			page_base = 0;
964 		}
965 
966 		/* Implicit padding for the last segment in a Write
967 		 * chunk is inserted inline at the front of the tail
968 		 * iovec. The upper layer ignores the content of
969 		 * the pad. Simply ensure inline content in the tail
970 		 * that follows the Write chunk is properly aligned.
971 		 */
972 		if (pad)
973 			srcp -= pad;
974 	}
975 
976 	/* The tail iovec is redirected to the remaining data
977 	 * in the receive buffer, to avoid a memcopy.
978 	 */
979 	if (copy_len || pad) {
980 		rqst->rq_rcv_buf.tail[0].iov_base = srcp;
981 		rqst->rq_private_buf.tail[0].iov_base = srcp;
982 	}
983 
984 	return fixup_copy_count;
985 }
986 
987 /* By convention, backchannel calls arrive via rdma_msg type
988  * messages, and never populate the chunk lists. This makes
989  * the RPC/RDMA header small and fixed in size, so it is
990  * straightforward to check the RPC header's direction field.
991  */
992 static bool
993 rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
994 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
995 {
996 	struct xdr_stream *xdr = &rep->rr_stream;
997 	__be32 *p;
998 
999 	if (rep->rr_proc != rdma_msg)
1000 		return false;
1001 
1002 	/* Peek at stream contents without advancing. */
1003 	p = xdr_inline_decode(xdr, 0);
1004 
1005 	/* Chunk lists */
1006 	if (*p++ != xdr_zero)
1007 		return false;
1008 	if (*p++ != xdr_zero)
1009 		return false;
1010 	if (*p++ != xdr_zero)
1011 		return false;
1012 
1013 	/* RPC header */
1014 	if (*p++ != rep->rr_xid)
1015 		return false;
1016 	if (*p != cpu_to_be32(RPC_CALL))
1017 		return false;
1018 
1019 	/* Now that we are sure this is a backchannel call,
1020 	 * advance to the RPC header.
1021 	 */
1022 	p = xdr_inline_decode(xdr, 3 * sizeof(*p));
1023 	if (unlikely(!p))
1024 		goto out_short;
1025 
1026 	rpcrdma_bc_receive_call(r_xprt, rep);
1027 	return true;
1028 
1029 out_short:
1030 	pr_warn("RPC/RDMA short backward direction call\n");
1031 	return true;
1032 }
1033 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
1034 {
1035 	return false;
1036 }
1037 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
1038 
1039 static int decode_rdma_segment(struct xdr_stream *xdr, u32 *length)
1040 {
1041 	u32 handle;
1042 	u64 offset;
1043 	__be32 *p;
1044 
1045 	p = xdr_inline_decode(xdr, 4 * sizeof(*p));
1046 	if (unlikely(!p))
1047 		return -EIO;
1048 
1049 	handle = be32_to_cpup(p++);
1050 	*length = be32_to_cpup(p++);
1051 	xdr_decode_hyper(p, &offset);
1052 
1053 	trace_xprtrdma_decode_seg(handle, *length, offset);
1054 	return 0;
1055 }
1056 
1057 static int decode_write_chunk(struct xdr_stream *xdr, u32 *length)
1058 {
1059 	u32 segcount, seglength;
1060 	__be32 *p;
1061 
1062 	p = xdr_inline_decode(xdr, sizeof(*p));
1063 	if (unlikely(!p))
1064 		return -EIO;
1065 
1066 	*length = 0;
1067 	segcount = be32_to_cpup(p);
1068 	while (segcount--) {
1069 		if (decode_rdma_segment(xdr, &seglength))
1070 			return -EIO;
1071 		*length += seglength;
1072 	}
1073 
1074 	return 0;
1075 }
1076 
1077 /* In RPC-over-RDMA Version One replies, a Read list is never
1078  * expected. This decoder is a stub that returns an error if
1079  * a Read list is present.
1080  */
1081 static int decode_read_list(struct xdr_stream *xdr)
1082 {
1083 	__be32 *p;
1084 
1085 	p = xdr_inline_decode(xdr, sizeof(*p));
1086 	if (unlikely(!p))
1087 		return -EIO;
1088 	if (unlikely(*p != xdr_zero))
1089 		return -EIO;
1090 	return 0;
1091 }
1092 
1093 /* Supports only one Write chunk in the Write list
1094  */
1095 static int decode_write_list(struct xdr_stream *xdr, u32 *length)
1096 {
1097 	u32 chunklen;
1098 	bool first;
1099 	__be32 *p;
1100 
1101 	*length = 0;
1102 	first = true;
1103 	do {
1104 		p = xdr_inline_decode(xdr, sizeof(*p));
1105 		if (unlikely(!p))
1106 			return -EIO;
1107 		if (*p == xdr_zero)
1108 			break;
1109 		if (!first)
1110 			return -EIO;
1111 
1112 		if (decode_write_chunk(xdr, &chunklen))
1113 			return -EIO;
1114 		*length += chunklen;
1115 		first = false;
1116 	} while (true);
1117 	return 0;
1118 }
1119 
1120 static int decode_reply_chunk(struct xdr_stream *xdr, u32 *length)
1121 {
1122 	__be32 *p;
1123 
1124 	p = xdr_inline_decode(xdr, sizeof(*p));
1125 	if (unlikely(!p))
1126 		return -EIO;
1127 
1128 	*length = 0;
1129 	if (*p != xdr_zero)
1130 		if (decode_write_chunk(xdr, length))
1131 			return -EIO;
1132 	return 0;
1133 }
1134 
1135 static int
1136 rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1137 		   struct rpc_rqst *rqst)
1138 {
1139 	struct xdr_stream *xdr = &rep->rr_stream;
1140 	u32 writelist, replychunk, rpclen;
1141 	char *base;
1142 
1143 	/* Decode the chunk lists */
1144 	if (decode_read_list(xdr))
1145 		return -EIO;
1146 	if (decode_write_list(xdr, &writelist))
1147 		return -EIO;
1148 	if (decode_reply_chunk(xdr, &replychunk))
1149 		return -EIO;
1150 
1151 	/* RDMA_MSG sanity checks */
1152 	if (unlikely(replychunk))
1153 		return -EIO;
1154 
1155 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
1156 	base = (char *)xdr_inline_decode(xdr, 0);
1157 	rpclen = xdr_stream_remaining(xdr);
1158 	r_xprt->rx_stats.fixup_copy_count +=
1159 		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
1160 
1161 	r_xprt->rx_stats.total_rdma_reply += writelist;
1162 	return rpclen + xdr_align_size(writelist);
1163 }
1164 
1165 static noinline int
1166 rpcrdma_decode_nomsg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1167 {
1168 	struct xdr_stream *xdr = &rep->rr_stream;
1169 	u32 writelist, replychunk;
1170 
1171 	/* Decode the chunk lists */
1172 	if (decode_read_list(xdr))
1173 		return -EIO;
1174 	if (decode_write_list(xdr, &writelist))
1175 		return -EIO;
1176 	if (decode_reply_chunk(xdr, &replychunk))
1177 		return -EIO;
1178 
1179 	/* RDMA_NOMSG sanity checks */
1180 	if (unlikely(writelist))
1181 		return -EIO;
1182 	if (unlikely(!replychunk))
1183 		return -EIO;
1184 
1185 	/* Reply chunk buffer already is the reply vector */
1186 	r_xprt->rx_stats.total_rdma_reply += replychunk;
1187 	return replychunk;
1188 }
1189 
1190 static noinline int
1191 rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1192 		     struct rpc_rqst *rqst)
1193 {
1194 	struct xdr_stream *xdr = &rep->rr_stream;
1195 	__be32 *p;
1196 
1197 	p = xdr_inline_decode(xdr, sizeof(*p));
1198 	if (unlikely(!p))
1199 		return -EIO;
1200 
1201 	switch (*p) {
1202 	case err_vers:
1203 		p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1204 		if (!p)
1205 			break;
1206 		dprintk("RPC:       %s: server reports "
1207 			"version error (%u-%u), xid %08x\n", __func__,
1208 			be32_to_cpup(p), be32_to_cpu(*(p + 1)),
1209 			be32_to_cpu(rep->rr_xid));
1210 		break;
1211 	case err_chunk:
1212 		dprintk("RPC:       %s: server reports "
1213 			"header decoding error, xid %08x\n", __func__,
1214 			be32_to_cpu(rep->rr_xid));
1215 		break;
1216 	default:
1217 		dprintk("RPC:       %s: server reports "
1218 			"unrecognized error %d, xid %08x\n", __func__,
1219 			be32_to_cpup(p), be32_to_cpu(rep->rr_xid));
1220 	}
1221 
1222 	r_xprt->rx_stats.bad_reply_count++;
1223 	return -EREMOTEIO;
1224 }
1225 
1226 /* Perform XID lookup, reconstruction of the RPC reply, and
1227  * RPC completion while holding the transport lock to ensure
1228  * the rep, rqst, and rq_task pointers remain stable.
1229  */
1230 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
1231 {
1232 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1233 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1234 	struct rpc_rqst *rqst = rep->rr_rqst;
1235 	int status;
1236 
1237 	xprt->reestablish_timeout = 0;
1238 
1239 	switch (rep->rr_proc) {
1240 	case rdma_msg:
1241 		status = rpcrdma_decode_msg(r_xprt, rep, rqst);
1242 		break;
1243 	case rdma_nomsg:
1244 		status = rpcrdma_decode_nomsg(r_xprt, rep);
1245 		break;
1246 	case rdma_error:
1247 		status = rpcrdma_decode_error(r_xprt, rep, rqst);
1248 		break;
1249 	default:
1250 		status = -EIO;
1251 	}
1252 	if (status < 0)
1253 		goto out_badheader;
1254 
1255 out:
1256 	spin_lock(&xprt->queue_lock);
1257 	xprt_complete_rqst(rqst->rq_task, status);
1258 	xprt_unpin_rqst(rqst);
1259 	spin_unlock(&xprt->queue_lock);
1260 	return;
1261 
1262 /* If the incoming reply terminated a pending RPC, the next
1263  * RPC call will post a replacement receive buffer as it is
1264  * being marshaled.
1265  */
1266 out_badheader:
1267 	trace_xprtrdma_reply_hdr(rep);
1268 	r_xprt->rx_stats.bad_reply_count++;
1269 	goto out;
1270 }
1271 
1272 void rpcrdma_release_rqst(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
1273 {
1274 	/* Invalidate and unmap the data payloads before waking
1275 	 * the waiting application. This guarantees the memory
1276 	 * regions are properly fenced from the server before the
1277 	 * application accesses the data. It also ensures proper
1278 	 * send flow control: waking the next RPC waits until this
1279 	 * RPC has relinquished all its Send Queue entries.
1280 	 */
1281 	if (!list_empty(&req->rl_registered))
1282 		frwr_unmap_sync(r_xprt, &req->rl_registered);
1283 
1284 	/* Ensure that any DMA mapped pages associated with
1285 	 * the Send of the RPC Call have been unmapped before
1286 	 * allowing the RPC to complete. This protects argument
1287 	 * memory not controlled by the RPC client from being
1288 	 * re-used before we're done with it.
1289 	 */
1290 	if (test_bit(RPCRDMA_REQ_F_TX_RESOURCES, &req->rl_flags)) {
1291 		r_xprt->rx_stats.reply_waits_for_send++;
1292 		out_of_line_wait_on_bit(&req->rl_flags,
1293 					RPCRDMA_REQ_F_TX_RESOURCES,
1294 					bit_wait,
1295 					TASK_UNINTERRUPTIBLE);
1296 	}
1297 }
1298 
1299 /* Reply handling runs in the poll worker thread. Anything that
1300  * might wait is deferred to a separate workqueue.
1301  */
1302 void rpcrdma_deferred_completion(struct work_struct *work)
1303 {
1304 	struct rpcrdma_rep *rep =
1305 			container_of(work, struct rpcrdma_rep, rr_work);
1306 	struct rpcrdma_req *req = rpcr_to_rdmar(rep->rr_rqst);
1307 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1308 
1309 	trace_xprtrdma_defer_cmp(rep);
1310 	if (rep->rr_wc_flags & IB_WC_WITH_INVALIDATE)
1311 		frwr_reminv(rep, &req->rl_registered);
1312 	rpcrdma_release_rqst(r_xprt, req);
1313 	rpcrdma_complete_rqst(rep);
1314 }
1315 
1316 /* Process received RPC/RDMA messages.
1317  *
1318  * Errors must result in the RPC task either being awakened, or
1319  * allowed to timeout, to discover the errors at that time.
1320  */
1321 void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
1322 {
1323 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1324 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1325 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1326 	struct rpcrdma_req *req;
1327 	struct rpc_rqst *rqst;
1328 	u32 credits;
1329 	__be32 *p;
1330 
1331 	/* Fixed transport header fields */
1332 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
1333 			rep->rr_hdrbuf.head[0].iov_base, NULL);
1334 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
1335 	if (unlikely(!p))
1336 		goto out_shortreply;
1337 	rep->rr_xid = *p++;
1338 	rep->rr_vers = *p++;
1339 	credits = be32_to_cpu(*p++);
1340 	rep->rr_proc = *p++;
1341 
1342 	if (rep->rr_vers != rpcrdma_version)
1343 		goto out_badversion;
1344 
1345 	if (rpcrdma_is_bcall(r_xprt, rep))
1346 		return;
1347 
1348 	/* Match incoming rpcrdma_rep to an rpcrdma_req to
1349 	 * get context for handling any incoming chunks.
1350 	 */
1351 	spin_lock(&xprt->queue_lock);
1352 	rqst = xprt_lookup_rqst(xprt, rep->rr_xid);
1353 	if (!rqst)
1354 		goto out_norqst;
1355 	xprt_pin_rqst(rqst);
1356 	spin_unlock(&xprt->queue_lock);
1357 
1358 	if (credits == 0)
1359 		credits = 1;	/* don't deadlock */
1360 	else if (credits > buf->rb_max_requests)
1361 		credits = buf->rb_max_requests;
1362 	if (buf->rb_credits != credits) {
1363 		spin_lock_bh(&xprt->transport_lock);
1364 		buf->rb_credits = credits;
1365 		xprt->cwnd = credits << RPC_CWNDSHIFT;
1366 		spin_unlock_bh(&xprt->transport_lock);
1367 	}
1368 
1369 	req = rpcr_to_rdmar(rqst);
1370 	if (req->rl_reply) {
1371 		trace_xprtrdma_leaked_rep(rqst, req->rl_reply);
1372 		rpcrdma_recv_buffer_put(req->rl_reply);
1373 	}
1374 	req->rl_reply = rep;
1375 	rep->rr_rqst = rqst;
1376 	clear_bit(RPCRDMA_REQ_F_PENDING, &req->rl_flags);
1377 
1378 	trace_xprtrdma_reply(rqst->rq_task, rep, req, credits);
1379 	queue_work(buf->rb_completion_wq, &rep->rr_work);
1380 	return;
1381 
1382 out_badversion:
1383 	trace_xprtrdma_reply_vers(rep);
1384 	goto out;
1385 
1386 out_norqst:
1387 	spin_unlock(&xprt->queue_lock);
1388 	trace_xprtrdma_reply_rqst(rep);
1389 	goto out;
1390 
1391 out_shortreply:
1392 	trace_xprtrdma_reply_short(rep);
1393 
1394 out:
1395 	rpcrdma_recv_buffer_put(rep);
1396 }
1397