xref: /linux/net/sunrpc/xprtrdma/rpc_rdma.c (revision 52338415)
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 static struct rpcrdma_mr_seg *rpcrdma_mr_prepare(struct rpcrdma_xprt *r_xprt,
346 						 struct rpcrdma_req *req,
347 						 struct rpcrdma_mr_seg *seg,
348 						 int nsegs, bool writing,
349 						 struct rpcrdma_mr **mr)
350 {
351 	*mr = rpcrdma_mr_pop(&req->rl_free_mrs);
352 	if (!*mr) {
353 		*mr = rpcrdma_mr_get(r_xprt);
354 		if (!*mr)
355 			goto out_getmr_err;
356 		trace_xprtrdma_mr_get(req);
357 		(*mr)->mr_req = req;
358 	}
359 
360 	rpcrdma_mr_push(*mr, &req->rl_registered);
361 	return frwr_map(r_xprt, seg, nsegs, writing, req->rl_slot.rq_xid, *mr);
362 
363 out_getmr_err:
364 	trace_xprtrdma_nomrs(req);
365 	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
366 	if (r_xprt->rx_ep.rep_connected != -ENODEV)
367 		schedule_work(&r_xprt->rx_buf.rb_refresh_worker);
368 	return ERR_PTR(-EAGAIN);
369 }
370 
371 /* Register and XDR encode the Read list. Supports encoding a list of read
372  * segments that belong to a single read chunk.
373  *
374  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
375  *
376  *  Read chunklist (a linked list):
377  *   N elements, position P (same P for all chunks of same arg!):
378  *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
379  *
380  * Returns zero on success, or a negative errno if a failure occurred.
381  * @xdr is advanced to the next position in the stream.
382  *
383  * Only a single @pos value is currently supported.
384  */
385 static int rpcrdma_encode_read_list(struct rpcrdma_xprt *r_xprt,
386 				    struct rpcrdma_req *req,
387 				    struct rpc_rqst *rqst,
388 				    enum rpcrdma_chunktype rtype)
389 {
390 	struct xdr_stream *xdr = &req->rl_stream;
391 	struct rpcrdma_mr_seg *seg;
392 	struct rpcrdma_mr *mr;
393 	unsigned int pos;
394 	int nsegs;
395 
396 	if (rtype == rpcrdma_noch)
397 		goto done;
398 
399 	pos = rqst->rq_snd_buf.head[0].iov_len;
400 	if (rtype == rpcrdma_areadch)
401 		pos = 0;
402 	seg = req->rl_segments;
403 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_snd_buf, pos,
404 				     rtype, seg);
405 	if (nsegs < 0)
406 		return nsegs;
407 
408 	do {
409 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, false, &mr);
410 		if (IS_ERR(seg))
411 			return PTR_ERR(seg);
412 
413 		if (encode_read_segment(xdr, mr, pos) < 0)
414 			return -EMSGSIZE;
415 
416 		trace_xprtrdma_chunk_read(rqst->rq_task, pos, mr, nsegs);
417 		r_xprt->rx_stats.read_chunk_count++;
418 		nsegs -= mr->mr_nents;
419 	} while (nsegs);
420 
421 done:
422 	return encode_item_not_present(xdr);
423 }
424 
425 /* Register and XDR encode the Write list. Supports encoding a list
426  * containing one array of plain segments that belong to a single
427  * write chunk.
428  *
429  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
430  *
431  *  Write chunklist (a list of (one) counted array):
432  *   N elements:
433  *    1 - N - HLOO - HLOO - ... - HLOO - 0
434  *
435  * Returns zero on success, or a negative errno if a failure occurred.
436  * @xdr is advanced to the next position in the stream.
437  *
438  * Only a single Write chunk is currently supported.
439  */
440 static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt,
441 				     struct rpcrdma_req *req,
442 				     struct rpc_rqst *rqst,
443 				     enum rpcrdma_chunktype wtype)
444 {
445 	struct xdr_stream *xdr = &req->rl_stream;
446 	struct rpcrdma_mr_seg *seg;
447 	struct rpcrdma_mr *mr;
448 	int nsegs, nchunks;
449 	__be32 *segcount;
450 
451 	if (wtype != rpcrdma_writech)
452 		goto done;
453 
454 	seg = req->rl_segments;
455 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf,
456 				     rqst->rq_rcv_buf.head[0].iov_len,
457 				     wtype, seg);
458 	if (nsegs < 0)
459 		return nsegs;
460 
461 	if (encode_item_present(xdr) < 0)
462 		return -EMSGSIZE;
463 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
464 	if (unlikely(!segcount))
465 		return -EMSGSIZE;
466 	/* Actual value encoded below */
467 
468 	nchunks = 0;
469 	do {
470 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
471 		if (IS_ERR(seg))
472 			return PTR_ERR(seg);
473 
474 		if (encode_rdma_segment(xdr, mr) < 0)
475 			return -EMSGSIZE;
476 
477 		trace_xprtrdma_chunk_write(rqst->rq_task, mr, nsegs);
478 		r_xprt->rx_stats.write_chunk_count++;
479 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
480 		nchunks++;
481 		nsegs -= mr->mr_nents;
482 	} while (nsegs);
483 
484 	/* Update count of segments in this Write chunk */
485 	*segcount = cpu_to_be32(nchunks);
486 
487 done:
488 	return encode_item_not_present(xdr);
489 }
490 
491 /* Register and XDR encode the Reply chunk. Supports encoding an array
492  * of plain segments that belong to a single write (reply) chunk.
493  *
494  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
495  *
496  *  Reply chunk (a counted array):
497  *   N elements:
498  *    1 - N - HLOO - HLOO - ... - HLOO
499  *
500  * Returns zero on success, or a negative errno if a failure occurred.
501  * @xdr is advanced to the next position in the stream.
502  */
503 static int rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt,
504 				      struct rpcrdma_req *req,
505 				      struct rpc_rqst *rqst,
506 				      enum rpcrdma_chunktype wtype)
507 {
508 	struct xdr_stream *xdr = &req->rl_stream;
509 	struct rpcrdma_mr_seg *seg;
510 	struct rpcrdma_mr *mr;
511 	int nsegs, nchunks;
512 	__be32 *segcount;
513 
514 	if (wtype != rpcrdma_replych)
515 		return encode_item_not_present(xdr);
516 
517 	seg = req->rl_segments;
518 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf, 0, wtype, seg);
519 	if (nsegs < 0)
520 		return nsegs;
521 
522 	if (encode_item_present(xdr) < 0)
523 		return -EMSGSIZE;
524 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
525 	if (unlikely(!segcount))
526 		return -EMSGSIZE;
527 	/* Actual value encoded below */
528 
529 	nchunks = 0;
530 	do {
531 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
532 		if (IS_ERR(seg))
533 			return PTR_ERR(seg);
534 
535 		if (encode_rdma_segment(xdr, mr) < 0)
536 			return -EMSGSIZE;
537 
538 		trace_xprtrdma_chunk_reply(rqst->rq_task, mr, nsegs);
539 		r_xprt->rx_stats.reply_chunk_count++;
540 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
541 		nchunks++;
542 		nsegs -= mr->mr_nents;
543 	} while (nsegs);
544 
545 	/* Update count of segments in the Reply chunk */
546 	*segcount = cpu_to_be32(nchunks);
547 
548 	return 0;
549 }
550 
551 static void rpcrdma_sendctx_done(struct kref *kref)
552 {
553 	struct rpcrdma_req *req =
554 		container_of(kref, struct rpcrdma_req, rl_kref);
555 	struct rpcrdma_rep *rep = req->rl_reply;
556 
557 	rpcrdma_complete_rqst(rep);
558 	rep->rr_rxprt->rx_stats.reply_waits_for_send++;
559 }
560 
561 /**
562  * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
563  * @sc: sendctx containing SGEs to unmap
564  *
565  */
566 void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
567 {
568 	struct ib_sge *sge;
569 
570 	if (!sc->sc_unmap_count)
571 		return;
572 
573 	/* The first two SGEs contain the transport header and
574 	 * the inline buffer. These are always left mapped so
575 	 * they can be cheaply re-used.
576 	 */
577 	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
578 	     ++sge, --sc->sc_unmap_count)
579 		ib_dma_unmap_page(sc->sc_device, sge->addr, sge->length,
580 				  DMA_TO_DEVICE);
581 
582 	kref_put(&sc->sc_req->rl_kref, rpcrdma_sendctx_done);
583 }
584 
585 /* Prepare an SGE for the RPC-over-RDMA transport header.
586  */
587 static bool rpcrdma_prepare_hdr_sge(struct rpcrdma_xprt *r_xprt,
588 				    struct rpcrdma_req *req, u32 len)
589 {
590 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
591 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
592 	struct ib_sge *sge = sc->sc_sges;
593 
594 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
595 		goto out_regbuf;
596 	sge->addr = rdmab_addr(rb);
597 	sge->length = len;
598 	sge->lkey = rdmab_lkey(rb);
599 
600 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
601 				      DMA_TO_DEVICE);
602 	sc->sc_wr.num_sge++;
603 	return true;
604 
605 out_regbuf:
606 	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
607 	return false;
608 }
609 
610 /* Prepare the Send SGEs. The head and tail iovec, and each entry
611  * in the page list, gets its own SGE.
612  */
613 static bool rpcrdma_prepare_msg_sges(struct rpcrdma_xprt *r_xprt,
614 				     struct rpcrdma_req *req,
615 				     struct xdr_buf *xdr,
616 				     enum rpcrdma_chunktype rtype)
617 {
618 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
619 	unsigned int sge_no, page_base, len, remaining;
620 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
621 	struct ib_sge *sge = sc->sc_sges;
622 	struct page *page, **ppages;
623 
624 	/* The head iovec is straightforward, as it is already
625 	 * DMA-mapped. Sync the content that has changed.
626 	 */
627 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
628 		goto out_regbuf;
629 	sc->sc_device = rdmab_device(rb);
630 	sge_no = 1;
631 	sge[sge_no].addr = rdmab_addr(rb);
632 	sge[sge_no].length = xdr->head[0].iov_len;
633 	sge[sge_no].lkey = rdmab_lkey(rb);
634 	ib_dma_sync_single_for_device(rdmab_device(rb), sge[sge_no].addr,
635 				      sge[sge_no].length, DMA_TO_DEVICE);
636 
637 	/* If there is a Read chunk, the page list is being handled
638 	 * via explicit RDMA, and thus is skipped here. However, the
639 	 * tail iovec may include an XDR pad for the page list, as
640 	 * well as additional content, and may not reside in the
641 	 * same page as the head iovec.
642 	 */
643 	if (rtype == rpcrdma_readch) {
644 		len = xdr->tail[0].iov_len;
645 
646 		/* Do not include the tail if it is only an XDR pad */
647 		if (len < 4)
648 			goto out;
649 
650 		page = virt_to_page(xdr->tail[0].iov_base);
651 		page_base = offset_in_page(xdr->tail[0].iov_base);
652 
653 		/* If the content in the page list is an odd length,
654 		 * xdr_write_pages() has added a pad at the beginning
655 		 * of the tail iovec. Force the tail's non-pad content
656 		 * to land at the next XDR position in the Send message.
657 		 */
658 		page_base += len & 3;
659 		len -= len & 3;
660 		goto map_tail;
661 	}
662 
663 	/* If there is a page list present, temporarily DMA map
664 	 * and prepare an SGE for each page to be sent.
665 	 */
666 	if (xdr->page_len) {
667 		ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
668 		page_base = offset_in_page(xdr->page_base);
669 		remaining = xdr->page_len;
670 		while (remaining) {
671 			sge_no++;
672 			if (sge_no > RPCRDMA_MAX_SEND_SGES - 2)
673 				goto out_mapping_overflow;
674 
675 			len = min_t(u32, PAGE_SIZE - page_base, remaining);
676 			sge[sge_no].addr =
677 				ib_dma_map_page(rdmab_device(rb), *ppages,
678 						page_base, len, DMA_TO_DEVICE);
679 			if (ib_dma_mapping_error(rdmab_device(rb),
680 						 sge[sge_no].addr))
681 				goto out_mapping_err;
682 			sge[sge_no].length = len;
683 			sge[sge_no].lkey = rdmab_lkey(rb);
684 
685 			sc->sc_unmap_count++;
686 			ppages++;
687 			remaining -= len;
688 			page_base = 0;
689 		}
690 	}
691 
692 	/* The tail iovec is not always constructed in the same
693 	 * page where the head iovec resides (see, for example,
694 	 * gss_wrap_req_priv). To neatly accommodate that case,
695 	 * DMA map it separately.
696 	 */
697 	if (xdr->tail[0].iov_len) {
698 		page = virt_to_page(xdr->tail[0].iov_base);
699 		page_base = offset_in_page(xdr->tail[0].iov_base);
700 		len = xdr->tail[0].iov_len;
701 
702 map_tail:
703 		sge_no++;
704 		sge[sge_no].addr =
705 			ib_dma_map_page(rdmab_device(rb), page, page_base, len,
706 					DMA_TO_DEVICE);
707 		if (ib_dma_mapping_error(rdmab_device(rb), sge[sge_no].addr))
708 			goto out_mapping_err;
709 		sge[sge_no].length = len;
710 		sge[sge_no].lkey = rdmab_lkey(rb);
711 		sc->sc_unmap_count++;
712 	}
713 
714 out:
715 	sc->sc_wr.num_sge += sge_no;
716 	if (sc->sc_unmap_count)
717 		kref_get(&req->rl_kref);
718 	return true;
719 
720 out_regbuf:
721 	pr_err("rpcrdma: failed to DMA map a Send buffer\n");
722 	return false;
723 
724 out_mapping_overflow:
725 	rpcrdma_sendctx_unmap(sc);
726 	pr_err("rpcrdma: too many Send SGEs (%u)\n", sge_no);
727 	return false;
728 
729 out_mapping_err:
730 	rpcrdma_sendctx_unmap(sc);
731 	trace_xprtrdma_dma_maperr(sge[sge_no].addr);
732 	return false;
733 }
734 
735 /**
736  * rpcrdma_prepare_send_sges - Construct SGEs for a Send WR
737  * @r_xprt: controlling transport
738  * @req: context of RPC Call being marshalled
739  * @hdrlen: size of transport header, in bytes
740  * @xdr: xdr_buf containing RPC Call
741  * @rtype: chunk type being encoded
742  *
743  * Returns 0 on success; otherwise a negative errno is returned.
744  */
745 int
746 rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
747 			  struct rpcrdma_req *req, u32 hdrlen,
748 			  struct xdr_buf *xdr, enum rpcrdma_chunktype rtype)
749 {
750 	int ret;
751 
752 	ret = -EAGAIN;
753 	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
754 	if (!req->rl_sendctx)
755 		goto err;
756 	req->rl_sendctx->sc_wr.num_sge = 0;
757 	req->rl_sendctx->sc_unmap_count = 0;
758 	req->rl_sendctx->sc_req = req;
759 	kref_init(&req->rl_kref);
760 
761 	ret = -EIO;
762 	if (!rpcrdma_prepare_hdr_sge(r_xprt, req, hdrlen))
763 		goto err;
764 	if (rtype != rpcrdma_areadch)
765 		if (!rpcrdma_prepare_msg_sges(r_xprt, req, xdr, rtype))
766 			goto err;
767 	return 0;
768 
769 err:
770 	trace_xprtrdma_prepsend_failed(&req->rl_slot, ret);
771 	return ret;
772 }
773 
774 /**
775  * rpcrdma_marshal_req - Marshal and send one RPC request
776  * @r_xprt: controlling transport
777  * @rqst: RPC request to be marshaled
778  *
779  * For the RPC in "rqst", this function:
780  *  - Chooses the transfer mode (eg., RDMA_MSG or RDMA_NOMSG)
781  *  - Registers Read, Write, and Reply chunks
782  *  - Constructs the transport header
783  *  - Posts a Send WR to send the transport header and request
784  *
785  * Returns:
786  *	%0 if the RPC was sent successfully,
787  *	%-ENOTCONN if the connection was lost,
788  *	%-EAGAIN if the caller should call again with the same arguments,
789  *	%-ENOBUFS if the caller should call again after a delay,
790  *	%-EMSGSIZE if the transport header is too small,
791  *	%-EIO if a permanent problem occurred while marshaling.
792  */
793 int
794 rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
795 {
796 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
797 	struct xdr_stream *xdr = &req->rl_stream;
798 	enum rpcrdma_chunktype rtype, wtype;
799 	bool ddp_allowed;
800 	__be32 *p;
801 	int ret;
802 
803 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
804 	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
805 			rqst);
806 
807 	/* Fixed header fields */
808 	ret = -EMSGSIZE;
809 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
810 	if (!p)
811 		goto out_err;
812 	*p++ = rqst->rq_xid;
813 	*p++ = rpcrdma_version;
814 	*p++ = cpu_to_be32(r_xprt->rx_buf.rb_max_requests);
815 
816 	/* When the ULP employs a GSS flavor that guarantees integrity
817 	 * or privacy, direct data placement of individual data items
818 	 * is not allowed.
819 	 */
820 	ddp_allowed = !(rqst->rq_cred->cr_auth->au_flags &
821 						RPCAUTH_AUTH_DATATOUCH);
822 
823 	/*
824 	 * Chunks needed for results?
825 	 *
826 	 * o If the expected result is under the inline threshold, all ops
827 	 *   return as inline.
828 	 * o Large read ops return data as write chunk(s), header as
829 	 *   inline.
830 	 * o Large non-read ops return as a single reply chunk.
831 	 */
832 	if (rpcrdma_results_inline(r_xprt, rqst))
833 		wtype = rpcrdma_noch;
834 	else if ((ddp_allowed && rqst->rq_rcv_buf.flags & XDRBUF_READ) &&
835 		 rpcrdma_nonpayload_inline(r_xprt, rqst))
836 		wtype = rpcrdma_writech;
837 	else
838 		wtype = rpcrdma_replych;
839 
840 	/*
841 	 * Chunks needed for arguments?
842 	 *
843 	 * o If the total request is under the inline threshold, all ops
844 	 *   are sent as inline.
845 	 * o Large write ops transmit data as read chunk(s), header as
846 	 *   inline.
847 	 * o Large non-write ops are sent with the entire message as a
848 	 *   single read chunk (protocol 0-position special case).
849 	 *
850 	 * This assumes that the upper layer does not present a request
851 	 * that both has a data payload, and whose non-data arguments
852 	 * by themselves are larger than the inline threshold.
853 	 */
854 	if (rpcrdma_args_inline(r_xprt, rqst)) {
855 		*p++ = rdma_msg;
856 		rtype = rpcrdma_noch;
857 	} else if (ddp_allowed && rqst->rq_snd_buf.flags & XDRBUF_WRITE) {
858 		*p++ = rdma_msg;
859 		rtype = rpcrdma_readch;
860 	} else {
861 		r_xprt->rx_stats.nomsg_call_count++;
862 		*p++ = rdma_nomsg;
863 		rtype = rpcrdma_areadch;
864 	}
865 
866 	/* If this is a retransmit, discard previously registered
867 	 * chunks. Very likely the connection has been replaced,
868 	 * so these registrations are invalid and unusable.
869 	 */
870 	frwr_recycle(req);
871 
872 	/* This implementation supports the following combinations
873 	 * of chunk lists in one RPC-over-RDMA Call message:
874 	 *
875 	 *   - Read list
876 	 *   - Write list
877 	 *   - Reply chunk
878 	 *   - Read list + Reply chunk
879 	 *
880 	 * It might not yet support the following combinations:
881 	 *
882 	 *   - Read list + Write list
883 	 *
884 	 * It does not support the following combinations:
885 	 *
886 	 *   - Write list + Reply chunk
887 	 *   - Read list + Write list + Reply chunk
888 	 *
889 	 * This implementation supports only a single chunk in each
890 	 * Read or Write list. Thus for example the client cannot
891 	 * send a Call message with a Position Zero Read chunk and a
892 	 * regular Read chunk at the same time.
893 	 */
894 	ret = rpcrdma_encode_read_list(r_xprt, req, rqst, rtype);
895 	if (ret)
896 		goto out_err;
897 	ret = rpcrdma_encode_write_list(r_xprt, req, rqst, wtype);
898 	if (ret)
899 		goto out_err;
900 	ret = rpcrdma_encode_reply_chunk(r_xprt, req, rqst, wtype);
901 	if (ret)
902 		goto out_err;
903 
904 	ret = rpcrdma_prepare_send_sges(r_xprt, req, req->rl_hdrbuf.len,
905 					&rqst->rq_snd_buf, rtype);
906 	if (ret)
907 		goto out_err;
908 
909 	trace_xprtrdma_marshal(req, rtype, wtype);
910 	return 0;
911 
912 out_err:
913 	trace_xprtrdma_marshal_failed(rqst, ret);
914 	r_xprt->rx_stats.failed_marshal_count++;
915 	frwr_reset(req);
916 	return ret;
917 }
918 
919 /**
920  * rpcrdma_inline_fixup - Scatter inline received data into rqst's iovecs
921  * @rqst: controlling RPC request
922  * @srcp: points to RPC message payload in receive buffer
923  * @copy_len: remaining length of receive buffer content
924  * @pad: Write chunk pad bytes needed (zero for pure inline)
925  *
926  * The upper layer has set the maximum number of bytes it can
927  * receive in each component of rq_rcv_buf. These values are set in
928  * the head.iov_len, page_len, tail.iov_len, and buflen fields.
929  *
930  * Unlike the TCP equivalent (xdr_partial_copy_from_skb), in
931  * many cases this function simply updates iov_base pointers in
932  * rq_rcv_buf to point directly to the received reply data, to
933  * avoid copying reply data.
934  *
935  * Returns the count of bytes which had to be memcopied.
936  */
937 static unsigned long
938 rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
939 {
940 	unsigned long fixup_copy_count;
941 	int i, npages, curlen;
942 	char *destp;
943 	struct page **ppages;
944 	int page_base;
945 
946 	/* The head iovec is redirected to the RPC reply message
947 	 * in the receive buffer, to avoid a memcopy.
948 	 */
949 	rqst->rq_rcv_buf.head[0].iov_base = srcp;
950 	rqst->rq_private_buf.head[0].iov_base = srcp;
951 
952 	/* The contents of the receive buffer that follow
953 	 * head.iov_len bytes are copied into the page list.
954 	 */
955 	curlen = rqst->rq_rcv_buf.head[0].iov_len;
956 	if (curlen > copy_len)
957 		curlen = copy_len;
958 	trace_xprtrdma_fixup(rqst, copy_len, curlen);
959 	srcp += curlen;
960 	copy_len -= curlen;
961 
962 	ppages = rqst->rq_rcv_buf.pages +
963 		(rqst->rq_rcv_buf.page_base >> PAGE_SHIFT);
964 	page_base = offset_in_page(rqst->rq_rcv_buf.page_base);
965 	fixup_copy_count = 0;
966 	if (copy_len && rqst->rq_rcv_buf.page_len) {
967 		int pagelist_len;
968 
969 		pagelist_len = rqst->rq_rcv_buf.page_len;
970 		if (pagelist_len > copy_len)
971 			pagelist_len = copy_len;
972 		npages = PAGE_ALIGN(page_base + pagelist_len) >> PAGE_SHIFT;
973 		for (i = 0; i < npages; i++) {
974 			curlen = PAGE_SIZE - page_base;
975 			if (curlen > pagelist_len)
976 				curlen = pagelist_len;
977 
978 			trace_xprtrdma_fixup_pg(rqst, i, srcp,
979 						copy_len, curlen);
980 			destp = kmap_atomic(ppages[i]);
981 			memcpy(destp + page_base, srcp, curlen);
982 			flush_dcache_page(ppages[i]);
983 			kunmap_atomic(destp);
984 			srcp += curlen;
985 			copy_len -= curlen;
986 			fixup_copy_count += curlen;
987 			pagelist_len -= curlen;
988 			if (!pagelist_len)
989 				break;
990 			page_base = 0;
991 		}
992 
993 		/* Implicit padding for the last segment in a Write
994 		 * chunk is inserted inline at the front of the tail
995 		 * iovec. The upper layer ignores the content of
996 		 * the pad. Simply ensure inline content in the tail
997 		 * that follows the Write chunk is properly aligned.
998 		 */
999 		if (pad)
1000 			srcp -= pad;
1001 	}
1002 
1003 	/* The tail iovec is redirected to the remaining data
1004 	 * in the receive buffer, to avoid a memcopy.
1005 	 */
1006 	if (copy_len || pad) {
1007 		rqst->rq_rcv_buf.tail[0].iov_base = srcp;
1008 		rqst->rq_private_buf.tail[0].iov_base = srcp;
1009 	}
1010 
1011 	return fixup_copy_count;
1012 }
1013 
1014 /* By convention, backchannel calls arrive via rdma_msg type
1015  * messages, and never populate the chunk lists. This makes
1016  * the RPC/RDMA header small and fixed in size, so it is
1017  * straightforward to check the RPC header's direction field.
1018  */
1019 static bool
1020 rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1021 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1022 {
1023 	struct xdr_stream *xdr = &rep->rr_stream;
1024 	__be32 *p;
1025 
1026 	if (rep->rr_proc != rdma_msg)
1027 		return false;
1028 
1029 	/* Peek at stream contents without advancing. */
1030 	p = xdr_inline_decode(xdr, 0);
1031 
1032 	/* Chunk lists */
1033 	if (*p++ != xdr_zero)
1034 		return false;
1035 	if (*p++ != xdr_zero)
1036 		return false;
1037 	if (*p++ != xdr_zero)
1038 		return false;
1039 
1040 	/* RPC header */
1041 	if (*p++ != rep->rr_xid)
1042 		return false;
1043 	if (*p != cpu_to_be32(RPC_CALL))
1044 		return false;
1045 
1046 	/* Now that we are sure this is a backchannel call,
1047 	 * advance to the RPC header.
1048 	 */
1049 	p = xdr_inline_decode(xdr, 3 * sizeof(*p));
1050 	if (unlikely(!p))
1051 		goto out_short;
1052 
1053 	rpcrdma_bc_receive_call(r_xprt, rep);
1054 	return true;
1055 
1056 out_short:
1057 	pr_warn("RPC/RDMA short backward direction call\n");
1058 	return true;
1059 }
1060 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
1061 {
1062 	return false;
1063 }
1064 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
1065 
1066 static int decode_rdma_segment(struct xdr_stream *xdr, u32 *length)
1067 {
1068 	u32 handle;
1069 	u64 offset;
1070 	__be32 *p;
1071 
1072 	p = xdr_inline_decode(xdr, 4 * sizeof(*p));
1073 	if (unlikely(!p))
1074 		return -EIO;
1075 
1076 	handle = be32_to_cpup(p++);
1077 	*length = be32_to_cpup(p++);
1078 	xdr_decode_hyper(p, &offset);
1079 
1080 	trace_xprtrdma_decode_seg(handle, *length, offset);
1081 	return 0;
1082 }
1083 
1084 static int decode_write_chunk(struct xdr_stream *xdr, u32 *length)
1085 {
1086 	u32 segcount, seglength;
1087 	__be32 *p;
1088 
1089 	p = xdr_inline_decode(xdr, sizeof(*p));
1090 	if (unlikely(!p))
1091 		return -EIO;
1092 
1093 	*length = 0;
1094 	segcount = be32_to_cpup(p);
1095 	while (segcount--) {
1096 		if (decode_rdma_segment(xdr, &seglength))
1097 			return -EIO;
1098 		*length += seglength;
1099 	}
1100 
1101 	return 0;
1102 }
1103 
1104 /* In RPC-over-RDMA Version One replies, a Read list is never
1105  * expected. This decoder is a stub that returns an error if
1106  * a Read list is present.
1107  */
1108 static int decode_read_list(struct xdr_stream *xdr)
1109 {
1110 	__be32 *p;
1111 
1112 	p = xdr_inline_decode(xdr, sizeof(*p));
1113 	if (unlikely(!p))
1114 		return -EIO;
1115 	if (unlikely(*p != xdr_zero))
1116 		return -EIO;
1117 	return 0;
1118 }
1119 
1120 /* Supports only one Write chunk in the Write list
1121  */
1122 static int decode_write_list(struct xdr_stream *xdr, u32 *length)
1123 {
1124 	u32 chunklen;
1125 	bool first;
1126 	__be32 *p;
1127 
1128 	*length = 0;
1129 	first = true;
1130 	do {
1131 		p = xdr_inline_decode(xdr, sizeof(*p));
1132 		if (unlikely(!p))
1133 			return -EIO;
1134 		if (*p == xdr_zero)
1135 			break;
1136 		if (!first)
1137 			return -EIO;
1138 
1139 		if (decode_write_chunk(xdr, &chunklen))
1140 			return -EIO;
1141 		*length += chunklen;
1142 		first = false;
1143 	} while (true);
1144 	return 0;
1145 }
1146 
1147 static int decode_reply_chunk(struct xdr_stream *xdr, u32 *length)
1148 {
1149 	__be32 *p;
1150 
1151 	p = xdr_inline_decode(xdr, sizeof(*p));
1152 	if (unlikely(!p))
1153 		return -EIO;
1154 
1155 	*length = 0;
1156 	if (*p != xdr_zero)
1157 		if (decode_write_chunk(xdr, length))
1158 			return -EIO;
1159 	return 0;
1160 }
1161 
1162 static int
1163 rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1164 		   struct rpc_rqst *rqst)
1165 {
1166 	struct xdr_stream *xdr = &rep->rr_stream;
1167 	u32 writelist, replychunk, rpclen;
1168 	char *base;
1169 
1170 	/* Decode the chunk lists */
1171 	if (decode_read_list(xdr))
1172 		return -EIO;
1173 	if (decode_write_list(xdr, &writelist))
1174 		return -EIO;
1175 	if (decode_reply_chunk(xdr, &replychunk))
1176 		return -EIO;
1177 
1178 	/* RDMA_MSG sanity checks */
1179 	if (unlikely(replychunk))
1180 		return -EIO;
1181 
1182 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
1183 	base = (char *)xdr_inline_decode(xdr, 0);
1184 	rpclen = xdr_stream_remaining(xdr);
1185 	r_xprt->rx_stats.fixup_copy_count +=
1186 		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
1187 
1188 	r_xprt->rx_stats.total_rdma_reply += writelist;
1189 	return rpclen + xdr_align_size(writelist);
1190 }
1191 
1192 static noinline int
1193 rpcrdma_decode_nomsg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1194 {
1195 	struct xdr_stream *xdr = &rep->rr_stream;
1196 	u32 writelist, replychunk;
1197 
1198 	/* Decode the chunk lists */
1199 	if (decode_read_list(xdr))
1200 		return -EIO;
1201 	if (decode_write_list(xdr, &writelist))
1202 		return -EIO;
1203 	if (decode_reply_chunk(xdr, &replychunk))
1204 		return -EIO;
1205 
1206 	/* RDMA_NOMSG sanity checks */
1207 	if (unlikely(writelist))
1208 		return -EIO;
1209 	if (unlikely(!replychunk))
1210 		return -EIO;
1211 
1212 	/* Reply chunk buffer already is the reply vector */
1213 	r_xprt->rx_stats.total_rdma_reply += replychunk;
1214 	return replychunk;
1215 }
1216 
1217 static noinline int
1218 rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1219 		     struct rpc_rqst *rqst)
1220 {
1221 	struct xdr_stream *xdr = &rep->rr_stream;
1222 	__be32 *p;
1223 
1224 	p = xdr_inline_decode(xdr, sizeof(*p));
1225 	if (unlikely(!p))
1226 		return -EIO;
1227 
1228 	switch (*p) {
1229 	case err_vers:
1230 		p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1231 		if (!p)
1232 			break;
1233 		dprintk("RPC:       %s: server reports "
1234 			"version error (%u-%u), xid %08x\n", __func__,
1235 			be32_to_cpup(p), be32_to_cpu(*(p + 1)),
1236 			be32_to_cpu(rep->rr_xid));
1237 		break;
1238 	case err_chunk:
1239 		dprintk("RPC:       %s: server reports "
1240 			"header decoding error, xid %08x\n", __func__,
1241 			be32_to_cpu(rep->rr_xid));
1242 		break;
1243 	default:
1244 		dprintk("RPC:       %s: server reports "
1245 			"unrecognized error %d, xid %08x\n", __func__,
1246 			be32_to_cpup(p), be32_to_cpu(rep->rr_xid));
1247 	}
1248 
1249 	r_xprt->rx_stats.bad_reply_count++;
1250 	return -EREMOTEIO;
1251 }
1252 
1253 /* Perform XID lookup, reconstruction of the RPC reply, and
1254  * RPC completion while holding the transport lock to ensure
1255  * the rep, rqst, and rq_task pointers remain stable.
1256  */
1257 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
1258 {
1259 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1260 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1261 	struct rpc_rqst *rqst = rep->rr_rqst;
1262 	int status;
1263 
1264 	switch (rep->rr_proc) {
1265 	case rdma_msg:
1266 		status = rpcrdma_decode_msg(r_xprt, rep, rqst);
1267 		break;
1268 	case rdma_nomsg:
1269 		status = rpcrdma_decode_nomsg(r_xprt, rep);
1270 		break;
1271 	case rdma_error:
1272 		status = rpcrdma_decode_error(r_xprt, rep, rqst);
1273 		break;
1274 	default:
1275 		status = -EIO;
1276 	}
1277 	if (status < 0)
1278 		goto out_badheader;
1279 
1280 out:
1281 	spin_lock(&xprt->queue_lock);
1282 	xprt_complete_rqst(rqst->rq_task, status);
1283 	xprt_unpin_rqst(rqst);
1284 	spin_unlock(&xprt->queue_lock);
1285 	return;
1286 
1287 /* If the incoming reply terminated a pending RPC, the next
1288  * RPC call will post a replacement receive buffer as it is
1289  * being marshaled.
1290  */
1291 out_badheader:
1292 	trace_xprtrdma_reply_hdr(rep);
1293 	r_xprt->rx_stats.bad_reply_count++;
1294 	goto out;
1295 }
1296 
1297 static void rpcrdma_reply_done(struct kref *kref)
1298 {
1299 	struct rpcrdma_req *req =
1300 		container_of(kref, struct rpcrdma_req, rl_kref);
1301 
1302 	rpcrdma_complete_rqst(req->rl_reply);
1303 }
1304 
1305 /**
1306  * rpcrdma_reply_handler - Process received RPC/RDMA messages
1307  * @rep: Incoming rpcrdma_rep object to process
1308  *
1309  * Errors must result in the RPC task either being awakened, or
1310  * allowed to timeout, to discover the errors at that time.
1311  */
1312 void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
1313 {
1314 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1315 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1316 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1317 	struct rpcrdma_req *req;
1318 	struct rpc_rqst *rqst;
1319 	u32 credits;
1320 	__be32 *p;
1321 
1322 	/* Any data means we had a useful conversation, so
1323 	 * then we don't need to delay the next reconnect.
1324 	 */
1325 	if (xprt->reestablish_timeout)
1326 		xprt->reestablish_timeout = 0;
1327 
1328 	/* Fixed transport header fields */
1329 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
1330 			rep->rr_hdrbuf.head[0].iov_base, NULL);
1331 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
1332 	if (unlikely(!p))
1333 		goto out_shortreply;
1334 	rep->rr_xid = *p++;
1335 	rep->rr_vers = *p++;
1336 	credits = be32_to_cpu(*p++);
1337 	rep->rr_proc = *p++;
1338 
1339 	if (rep->rr_vers != rpcrdma_version)
1340 		goto out_badversion;
1341 
1342 	if (rpcrdma_is_bcall(r_xprt, rep))
1343 		return;
1344 
1345 	/* Match incoming rpcrdma_rep to an rpcrdma_req to
1346 	 * get context for handling any incoming chunks.
1347 	 */
1348 	spin_lock(&xprt->queue_lock);
1349 	rqst = xprt_lookup_rqst(xprt, rep->rr_xid);
1350 	if (!rqst)
1351 		goto out_norqst;
1352 	xprt_pin_rqst(rqst);
1353 	spin_unlock(&xprt->queue_lock);
1354 
1355 	if (credits == 0)
1356 		credits = 1;	/* don't deadlock */
1357 	else if (credits > buf->rb_max_requests)
1358 		credits = buf->rb_max_requests;
1359 	if (buf->rb_credits != credits) {
1360 		spin_lock(&xprt->transport_lock);
1361 		buf->rb_credits = credits;
1362 		xprt->cwnd = credits << RPC_CWNDSHIFT;
1363 		spin_unlock(&xprt->transport_lock);
1364 	}
1365 
1366 	req = rpcr_to_rdmar(rqst);
1367 	if (req->rl_reply) {
1368 		trace_xprtrdma_leaked_rep(rqst, req->rl_reply);
1369 		rpcrdma_recv_buffer_put(req->rl_reply);
1370 	}
1371 	req->rl_reply = rep;
1372 	rep->rr_rqst = rqst;
1373 
1374 	trace_xprtrdma_reply(rqst->rq_task, rep, req, credits);
1375 
1376 	if (rep->rr_wc_flags & IB_WC_WITH_INVALIDATE)
1377 		frwr_reminv(rep, &req->rl_registered);
1378 	if (!list_empty(&req->rl_registered))
1379 		frwr_unmap_async(r_xprt, req);
1380 		/* LocalInv completion will complete the RPC */
1381 	else
1382 		kref_put(&req->rl_kref, rpcrdma_reply_done);
1383 	return;
1384 
1385 out_badversion:
1386 	trace_xprtrdma_reply_vers(rep);
1387 	goto out;
1388 
1389 out_norqst:
1390 	spin_unlock(&xprt->queue_lock);
1391 	trace_xprtrdma_reply_rqst(rep);
1392 	goto out;
1393 
1394 out_shortreply:
1395 	trace_xprtrdma_reply_short(rep);
1396 
1397 out:
1398 	rpcrdma_recv_buffer_put(rep);
1399 }
1400