xref: /linux/net/sunrpc/xprtrdma/rpc_rdma.c (revision 4fd18fc38757217c746aa063ba9e4729814dc737)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (c) 2014-2020, Oracle and/or its affiliates.
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 	return size;
82 }
83 
84 /* Returns size of largest RPC-over-RDMA header in a Reply message
85  *
86  * There is only one Write list or one Reply chunk per Reply
87  * message.  The larger list is the Write list.
88  */
89 static unsigned int rpcrdma_max_reply_header_size(unsigned int maxsegs)
90 {
91 	unsigned int size;
92 
93 	/* Fixed header fields and list discriminators */
94 	size = RPCRDMA_HDRLEN_MIN;
95 
96 	/* Maximum Write list size */
97 	size += sizeof(__be32);		/* segment count */
98 	size += maxsegs * rpcrdma_segment_maxsz * sizeof(__be32);
99 	size += sizeof(__be32);	/* list discriminator */
100 
101 	return size;
102 }
103 
104 /**
105  * rpcrdma_set_max_header_sizes - Initialize inline payload sizes
106  * @ep: endpoint to initialize
107  *
108  * The max_inline fields contain the maximum size of an RPC message
109  * so the marshaling code doesn't have to repeat this calculation
110  * for every RPC.
111  */
112 void rpcrdma_set_max_header_sizes(struct rpcrdma_ep *ep)
113 {
114 	unsigned int maxsegs = ep->re_max_rdma_segs;
115 
116 	ep->re_max_inline_send =
117 		ep->re_inline_send - rpcrdma_max_call_header_size(maxsegs);
118 	ep->re_max_inline_recv =
119 		ep->re_inline_recv - rpcrdma_max_reply_header_size(maxsegs);
120 }
121 
122 /* The client can send a request inline as long as the RPCRDMA header
123  * plus the RPC call fit under the transport's inline limit. If the
124  * combined call message size exceeds that limit, the client must use
125  * a Read chunk for this operation.
126  *
127  * A Read chunk is also required if sending the RPC call inline would
128  * exceed this device's max_sge limit.
129  */
130 static bool rpcrdma_args_inline(struct rpcrdma_xprt *r_xprt,
131 				struct rpc_rqst *rqst)
132 {
133 	struct xdr_buf *xdr = &rqst->rq_snd_buf;
134 	struct rpcrdma_ep *ep = r_xprt->rx_ep;
135 	unsigned int count, remaining, offset;
136 
137 	if (xdr->len > ep->re_max_inline_send)
138 		return false;
139 
140 	if (xdr->page_len) {
141 		remaining = xdr->page_len;
142 		offset = offset_in_page(xdr->page_base);
143 		count = RPCRDMA_MIN_SEND_SGES;
144 		while (remaining) {
145 			remaining -= min_t(unsigned int,
146 					   PAGE_SIZE - offset, remaining);
147 			offset = 0;
148 			if (++count > ep->re_attr.cap.max_send_sge)
149 				return false;
150 		}
151 	}
152 
153 	return true;
154 }
155 
156 /* The client can't know how large the actual reply will be. Thus it
157  * plans for the largest possible reply for that particular ULP
158  * operation. If the maximum combined reply message size exceeds that
159  * limit, the client must provide a write list or a reply chunk for
160  * this request.
161  */
162 static bool rpcrdma_results_inline(struct rpcrdma_xprt *r_xprt,
163 				   struct rpc_rqst *rqst)
164 {
165 	return rqst->rq_rcv_buf.buflen <= r_xprt->rx_ep->re_max_inline_recv;
166 }
167 
168 /* The client is required to provide a Reply chunk if the maximum
169  * size of the non-payload part of the RPC Reply is larger than
170  * the inline threshold.
171  */
172 static bool
173 rpcrdma_nonpayload_inline(const struct rpcrdma_xprt *r_xprt,
174 			  const struct rpc_rqst *rqst)
175 {
176 	const struct xdr_buf *buf = &rqst->rq_rcv_buf;
177 
178 	return (buf->head[0].iov_len + buf->tail[0].iov_len) <
179 		r_xprt->rx_ep->re_max_inline_recv;
180 }
181 
182 /* ACL likes to be lazy in allocating pages. For TCP, these
183  * pages can be allocated during receive processing. Not true
184  * for RDMA, which must always provision receive buffers
185  * up front.
186  */
187 static noinline int
188 rpcrdma_alloc_sparse_pages(struct xdr_buf *buf)
189 {
190 	struct page **ppages;
191 	int len;
192 
193 	len = buf->page_len;
194 	ppages = buf->pages + (buf->page_base >> PAGE_SHIFT);
195 	while (len > 0) {
196 		if (!*ppages)
197 			*ppages = alloc_page(GFP_NOWAIT | __GFP_NOWARN);
198 		if (!*ppages)
199 			return -ENOBUFS;
200 		ppages++;
201 		len -= PAGE_SIZE;
202 	}
203 
204 	return 0;
205 }
206 
207 /* Split @vec on page boundaries into SGEs. FMR registers pages, not
208  * a byte range. Other modes coalesce these SGEs into a single MR
209  * when they can.
210  *
211  * Returns pointer to next available SGE, and bumps the total number
212  * of SGEs consumed.
213  */
214 static struct rpcrdma_mr_seg *
215 rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg,
216 		     unsigned int *n)
217 {
218 	u32 remaining, page_offset;
219 	char *base;
220 
221 	base = vec->iov_base;
222 	page_offset = offset_in_page(base);
223 	remaining = vec->iov_len;
224 	while (remaining) {
225 		seg->mr_page = NULL;
226 		seg->mr_offset = base;
227 		seg->mr_len = min_t(u32, PAGE_SIZE - page_offset, remaining);
228 		remaining -= seg->mr_len;
229 		base += seg->mr_len;
230 		++seg;
231 		++(*n);
232 		page_offset = 0;
233 	}
234 	return seg;
235 }
236 
237 /* Convert @xdrbuf into SGEs no larger than a page each. As they
238  * are registered, these SGEs are then coalesced into RDMA segments
239  * when the selected memreg mode supports it.
240  *
241  * Returns positive number of SGEs consumed, or a negative errno.
242  */
243 
244 static int
245 rpcrdma_convert_iovs(struct rpcrdma_xprt *r_xprt, struct xdr_buf *xdrbuf,
246 		     unsigned int pos, enum rpcrdma_chunktype type,
247 		     struct rpcrdma_mr_seg *seg)
248 {
249 	unsigned long page_base;
250 	unsigned int len, n;
251 	struct page **ppages;
252 
253 	n = 0;
254 	if (pos == 0)
255 		seg = rpcrdma_convert_kvec(&xdrbuf->head[0], seg, &n);
256 
257 	len = xdrbuf->page_len;
258 	ppages = xdrbuf->pages + (xdrbuf->page_base >> PAGE_SHIFT);
259 	page_base = offset_in_page(xdrbuf->page_base);
260 	while (len) {
261 		seg->mr_page = *ppages;
262 		seg->mr_offset = (char *)page_base;
263 		seg->mr_len = min_t(u32, PAGE_SIZE - page_base, len);
264 		len -= seg->mr_len;
265 		++ppages;
266 		++seg;
267 		++n;
268 		page_base = 0;
269 	}
270 
271 	/* When encoding a Read chunk, the tail iovec contains an
272 	 * XDR pad and may be omitted.
273 	 */
274 	if (type == rpcrdma_readch && r_xprt->rx_ep->re_implicit_roundup)
275 		goto out;
276 
277 	/* When encoding a Write chunk, some servers need to see an
278 	 * extra segment for non-XDR-aligned Write chunks. The upper
279 	 * layer provides space in the tail iovec that may be used
280 	 * for this purpose.
281 	 */
282 	if (type == rpcrdma_writech && r_xprt->rx_ep->re_implicit_roundup)
283 		goto out;
284 
285 	if (xdrbuf->tail[0].iov_len)
286 		seg = rpcrdma_convert_kvec(&xdrbuf->tail[0], seg, &n);
287 
288 out:
289 	if (unlikely(n > RPCRDMA_MAX_SEGS))
290 		return -EIO;
291 	return n;
292 }
293 
294 static int
295 encode_rdma_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr)
296 {
297 	__be32 *p;
298 
299 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
300 	if (unlikely(!p))
301 		return -EMSGSIZE;
302 
303 	xdr_encode_rdma_segment(p, mr->mr_handle, mr->mr_length, mr->mr_offset);
304 	return 0;
305 }
306 
307 static int
308 encode_read_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr,
309 		    u32 position)
310 {
311 	__be32 *p;
312 
313 	p = xdr_reserve_space(xdr, 6 * sizeof(*p));
314 	if (unlikely(!p))
315 		return -EMSGSIZE;
316 
317 	*p++ = xdr_one;			/* Item present */
318 	xdr_encode_read_segment(p, position, mr->mr_handle, mr->mr_length,
319 				mr->mr_offset);
320 	return 0;
321 }
322 
323 static struct rpcrdma_mr_seg *rpcrdma_mr_prepare(struct rpcrdma_xprt *r_xprt,
324 						 struct rpcrdma_req *req,
325 						 struct rpcrdma_mr_seg *seg,
326 						 int nsegs, bool writing,
327 						 struct rpcrdma_mr **mr)
328 {
329 	*mr = rpcrdma_mr_pop(&req->rl_free_mrs);
330 	if (!*mr) {
331 		*mr = rpcrdma_mr_get(r_xprt);
332 		if (!*mr)
333 			goto out_getmr_err;
334 		(*mr)->mr_req = req;
335 	}
336 
337 	rpcrdma_mr_push(*mr, &req->rl_registered);
338 	return frwr_map(r_xprt, seg, nsegs, writing, req->rl_slot.rq_xid, *mr);
339 
340 out_getmr_err:
341 	trace_xprtrdma_nomrs_err(r_xprt, req);
342 	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
343 	rpcrdma_mrs_refresh(r_xprt);
344 	return ERR_PTR(-EAGAIN);
345 }
346 
347 /* Register and XDR encode the Read list. Supports encoding a list of read
348  * segments that belong to a single read chunk.
349  *
350  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
351  *
352  *  Read chunklist (a linked list):
353  *   N elements, position P (same P for all chunks of same arg!):
354  *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
355  *
356  * Returns zero on success, or a negative errno if a failure occurred.
357  * @xdr is advanced to the next position in the stream.
358  *
359  * Only a single @pos value is currently supported.
360  */
361 static int rpcrdma_encode_read_list(struct rpcrdma_xprt *r_xprt,
362 				    struct rpcrdma_req *req,
363 				    struct rpc_rqst *rqst,
364 				    enum rpcrdma_chunktype rtype)
365 {
366 	struct xdr_stream *xdr = &req->rl_stream;
367 	struct rpcrdma_mr_seg *seg;
368 	struct rpcrdma_mr *mr;
369 	unsigned int pos;
370 	int nsegs;
371 
372 	if (rtype == rpcrdma_noch_pullup || rtype == rpcrdma_noch_mapped)
373 		goto done;
374 
375 	pos = rqst->rq_snd_buf.head[0].iov_len;
376 	if (rtype == rpcrdma_areadch)
377 		pos = 0;
378 	seg = req->rl_segments;
379 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_snd_buf, pos,
380 				     rtype, seg);
381 	if (nsegs < 0)
382 		return nsegs;
383 
384 	do {
385 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, false, &mr);
386 		if (IS_ERR(seg))
387 			return PTR_ERR(seg);
388 
389 		if (encode_read_segment(xdr, mr, pos) < 0)
390 			return -EMSGSIZE;
391 
392 		trace_xprtrdma_chunk_read(rqst->rq_task, pos, mr, nsegs);
393 		r_xprt->rx_stats.read_chunk_count++;
394 		nsegs -= mr->mr_nents;
395 	} while (nsegs);
396 
397 done:
398 	if (xdr_stream_encode_item_absent(xdr) < 0)
399 		return -EMSGSIZE;
400 	return 0;
401 }
402 
403 /* Register and XDR encode the Write list. Supports encoding a list
404  * containing one array of plain segments that belong to a single
405  * write chunk.
406  *
407  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
408  *
409  *  Write chunklist (a list of (one) counted array):
410  *   N elements:
411  *    1 - N - HLOO - HLOO - ... - HLOO - 0
412  *
413  * Returns zero on success, or a negative errno if a failure occurred.
414  * @xdr is advanced to the next position in the stream.
415  *
416  * Only a single Write chunk is currently supported.
417  */
418 static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt,
419 				     struct rpcrdma_req *req,
420 				     struct rpc_rqst *rqst,
421 				     enum rpcrdma_chunktype wtype)
422 {
423 	struct xdr_stream *xdr = &req->rl_stream;
424 	struct rpcrdma_mr_seg *seg;
425 	struct rpcrdma_mr *mr;
426 	int nsegs, nchunks;
427 	__be32 *segcount;
428 
429 	if (wtype != rpcrdma_writech)
430 		goto done;
431 
432 	seg = req->rl_segments;
433 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf,
434 				     rqst->rq_rcv_buf.head[0].iov_len,
435 				     wtype, seg);
436 	if (nsegs < 0)
437 		return nsegs;
438 
439 	if (xdr_stream_encode_item_present(xdr) < 0)
440 		return -EMSGSIZE;
441 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
442 	if (unlikely(!segcount))
443 		return -EMSGSIZE;
444 	/* Actual value encoded below */
445 
446 	nchunks = 0;
447 	do {
448 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
449 		if (IS_ERR(seg))
450 			return PTR_ERR(seg);
451 
452 		if (encode_rdma_segment(xdr, mr) < 0)
453 			return -EMSGSIZE;
454 
455 		trace_xprtrdma_chunk_write(rqst->rq_task, mr, nsegs);
456 		r_xprt->rx_stats.write_chunk_count++;
457 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
458 		nchunks++;
459 		nsegs -= mr->mr_nents;
460 	} while (nsegs);
461 
462 	/* Update count of segments in this Write chunk */
463 	*segcount = cpu_to_be32(nchunks);
464 
465 done:
466 	if (xdr_stream_encode_item_absent(xdr) < 0)
467 		return -EMSGSIZE;
468 	return 0;
469 }
470 
471 /* Register and XDR encode the Reply chunk. Supports encoding an array
472  * of plain segments that belong to a single write (reply) chunk.
473  *
474  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
475  *
476  *  Reply chunk (a counted array):
477  *   N elements:
478  *    1 - N - HLOO - HLOO - ... - HLOO
479  *
480  * Returns zero on success, or a negative errno if a failure occurred.
481  * @xdr is advanced to the next position in the stream.
482  */
483 static int rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt,
484 				      struct rpcrdma_req *req,
485 				      struct rpc_rqst *rqst,
486 				      enum rpcrdma_chunktype wtype)
487 {
488 	struct xdr_stream *xdr = &req->rl_stream;
489 	struct rpcrdma_mr_seg *seg;
490 	struct rpcrdma_mr *mr;
491 	int nsegs, nchunks;
492 	__be32 *segcount;
493 
494 	if (wtype != rpcrdma_replych) {
495 		if (xdr_stream_encode_item_absent(xdr) < 0)
496 			return -EMSGSIZE;
497 		return 0;
498 	}
499 
500 	seg = req->rl_segments;
501 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf, 0, wtype, seg);
502 	if (nsegs < 0)
503 		return nsegs;
504 
505 	if (xdr_stream_encode_item_present(xdr) < 0)
506 		return -EMSGSIZE;
507 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
508 	if (unlikely(!segcount))
509 		return -EMSGSIZE;
510 	/* Actual value encoded below */
511 
512 	nchunks = 0;
513 	do {
514 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
515 		if (IS_ERR(seg))
516 			return PTR_ERR(seg);
517 
518 		if (encode_rdma_segment(xdr, mr) < 0)
519 			return -EMSGSIZE;
520 
521 		trace_xprtrdma_chunk_reply(rqst->rq_task, mr, nsegs);
522 		r_xprt->rx_stats.reply_chunk_count++;
523 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
524 		nchunks++;
525 		nsegs -= mr->mr_nents;
526 	} while (nsegs);
527 
528 	/* Update count of segments in the Reply chunk */
529 	*segcount = cpu_to_be32(nchunks);
530 
531 	return 0;
532 }
533 
534 static void rpcrdma_sendctx_done(struct kref *kref)
535 {
536 	struct rpcrdma_req *req =
537 		container_of(kref, struct rpcrdma_req, rl_kref);
538 	struct rpcrdma_rep *rep = req->rl_reply;
539 
540 	rpcrdma_complete_rqst(rep);
541 	rep->rr_rxprt->rx_stats.reply_waits_for_send++;
542 }
543 
544 /**
545  * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
546  * @sc: sendctx containing SGEs to unmap
547  *
548  */
549 void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
550 {
551 	struct rpcrdma_regbuf *rb = sc->sc_req->rl_sendbuf;
552 	struct ib_sge *sge;
553 
554 	if (!sc->sc_unmap_count)
555 		return;
556 
557 	/* The first two SGEs contain the transport header and
558 	 * the inline buffer. These are always left mapped so
559 	 * they can be cheaply re-used.
560 	 */
561 	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
562 	     ++sge, --sc->sc_unmap_count)
563 		ib_dma_unmap_page(rdmab_device(rb), sge->addr, sge->length,
564 				  DMA_TO_DEVICE);
565 
566 	kref_put(&sc->sc_req->rl_kref, rpcrdma_sendctx_done);
567 }
568 
569 /* Prepare an SGE for the RPC-over-RDMA transport header.
570  */
571 static void rpcrdma_prepare_hdr_sge(struct rpcrdma_xprt *r_xprt,
572 				    struct rpcrdma_req *req, u32 len)
573 {
574 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
575 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
576 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
577 
578 	sge->addr = rdmab_addr(rb);
579 	sge->length = len;
580 	sge->lkey = rdmab_lkey(rb);
581 
582 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
583 				      DMA_TO_DEVICE);
584 }
585 
586 /* The head iovec is straightforward, as it is usually already
587  * DMA-mapped. Sync the content that has changed.
588  */
589 static bool rpcrdma_prepare_head_iov(struct rpcrdma_xprt *r_xprt,
590 				     struct rpcrdma_req *req, unsigned int len)
591 {
592 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
593 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
594 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
595 
596 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
597 		return false;
598 
599 	sge->addr = rdmab_addr(rb);
600 	sge->length = len;
601 	sge->lkey = rdmab_lkey(rb);
602 
603 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
604 				      DMA_TO_DEVICE);
605 	return true;
606 }
607 
608 /* If there is a page list present, DMA map and prepare an
609  * SGE for each page to be sent.
610  */
611 static bool rpcrdma_prepare_pagelist(struct rpcrdma_req *req,
612 				     struct xdr_buf *xdr)
613 {
614 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
615 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
616 	unsigned int page_base, len, remaining;
617 	struct page **ppages;
618 	struct ib_sge *sge;
619 
620 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
621 	page_base = offset_in_page(xdr->page_base);
622 	remaining = xdr->page_len;
623 	while (remaining) {
624 		sge = &sc->sc_sges[req->rl_wr.num_sge++];
625 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
626 		sge->addr = ib_dma_map_page(rdmab_device(rb), *ppages,
627 					    page_base, len, DMA_TO_DEVICE);
628 		if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
629 			goto out_mapping_err;
630 
631 		sge->length = len;
632 		sge->lkey = rdmab_lkey(rb);
633 
634 		sc->sc_unmap_count++;
635 		ppages++;
636 		remaining -= len;
637 		page_base = 0;
638 	}
639 
640 	return true;
641 
642 out_mapping_err:
643 	trace_xprtrdma_dma_maperr(sge->addr);
644 	return false;
645 }
646 
647 /* The tail iovec may include an XDR pad for the page list,
648  * as well as additional content, and may not reside in the
649  * same page as the head iovec.
650  */
651 static bool rpcrdma_prepare_tail_iov(struct rpcrdma_req *req,
652 				     struct xdr_buf *xdr,
653 				     unsigned int page_base, unsigned int len)
654 {
655 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
656 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
657 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
658 	struct page *page = virt_to_page(xdr->tail[0].iov_base);
659 
660 	sge->addr = ib_dma_map_page(rdmab_device(rb), page, page_base, len,
661 				    DMA_TO_DEVICE);
662 	if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
663 		goto out_mapping_err;
664 
665 	sge->length = len;
666 	sge->lkey = rdmab_lkey(rb);
667 	++sc->sc_unmap_count;
668 	return true;
669 
670 out_mapping_err:
671 	trace_xprtrdma_dma_maperr(sge->addr);
672 	return false;
673 }
674 
675 /* Copy the tail to the end of the head buffer.
676  */
677 static void rpcrdma_pullup_tail_iov(struct rpcrdma_xprt *r_xprt,
678 				    struct rpcrdma_req *req,
679 				    struct xdr_buf *xdr)
680 {
681 	unsigned char *dst;
682 
683 	dst = (unsigned char *)xdr->head[0].iov_base;
684 	dst += xdr->head[0].iov_len + xdr->page_len;
685 	memmove(dst, xdr->tail[0].iov_base, xdr->tail[0].iov_len);
686 	r_xprt->rx_stats.pullup_copy_count += xdr->tail[0].iov_len;
687 }
688 
689 /* Copy pagelist content into the head buffer.
690  */
691 static void rpcrdma_pullup_pagelist(struct rpcrdma_xprt *r_xprt,
692 				    struct rpcrdma_req *req,
693 				    struct xdr_buf *xdr)
694 {
695 	unsigned int len, page_base, remaining;
696 	struct page **ppages;
697 	unsigned char *src, *dst;
698 
699 	dst = (unsigned char *)xdr->head[0].iov_base;
700 	dst += xdr->head[0].iov_len;
701 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
702 	page_base = offset_in_page(xdr->page_base);
703 	remaining = xdr->page_len;
704 	while (remaining) {
705 		src = page_address(*ppages);
706 		src += page_base;
707 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
708 		memcpy(dst, src, len);
709 		r_xprt->rx_stats.pullup_copy_count += len;
710 
711 		ppages++;
712 		dst += len;
713 		remaining -= len;
714 		page_base = 0;
715 	}
716 }
717 
718 /* Copy the contents of @xdr into @rl_sendbuf and DMA sync it.
719  * When the head, pagelist, and tail are small, a pull-up copy
720  * is considerably less costly than DMA mapping the components
721  * of @xdr.
722  *
723  * Assumptions:
724  *  - the caller has already verified that the total length
725  *    of the RPC Call body will fit into @rl_sendbuf.
726  */
727 static bool rpcrdma_prepare_noch_pullup(struct rpcrdma_xprt *r_xprt,
728 					struct rpcrdma_req *req,
729 					struct xdr_buf *xdr)
730 {
731 	if (unlikely(xdr->tail[0].iov_len))
732 		rpcrdma_pullup_tail_iov(r_xprt, req, xdr);
733 
734 	if (unlikely(xdr->page_len))
735 		rpcrdma_pullup_pagelist(r_xprt, req, xdr);
736 
737 	/* The whole RPC message resides in the head iovec now */
738 	return rpcrdma_prepare_head_iov(r_xprt, req, xdr->len);
739 }
740 
741 static bool rpcrdma_prepare_noch_mapped(struct rpcrdma_xprt *r_xprt,
742 					struct rpcrdma_req *req,
743 					struct xdr_buf *xdr)
744 {
745 	struct kvec *tail = &xdr->tail[0];
746 
747 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
748 		return false;
749 	if (xdr->page_len)
750 		if (!rpcrdma_prepare_pagelist(req, xdr))
751 			return false;
752 	if (tail->iov_len)
753 		if (!rpcrdma_prepare_tail_iov(req, xdr,
754 					      offset_in_page(tail->iov_base),
755 					      tail->iov_len))
756 			return false;
757 
758 	if (req->rl_sendctx->sc_unmap_count)
759 		kref_get(&req->rl_kref);
760 	return true;
761 }
762 
763 static bool rpcrdma_prepare_readch(struct rpcrdma_xprt *r_xprt,
764 				   struct rpcrdma_req *req,
765 				   struct xdr_buf *xdr)
766 {
767 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
768 		return false;
769 
770 	/* If there is a Read chunk, the page list is being handled
771 	 * via explicit RDMA, and thus is skipped here.
772 	 */
773 
774 	/* Do not include the tail if it is only an XDR pad */
775 	if (xdr->tail[0].iov_len > 3) {
776 		unsigned int page_base, len;
777 
778 		/* If the content in the page list is an odd length,
779 		 * xdr_write_pages() adds a pad at the beginning of
780 		 * the tail iovec. Force the tail's non-pad content to
781 		 * land at the next XDR position in the Send message.
782 		 */
783 		page_base = offset_in_page(xdr->tail[0].iov_base);
784 		len = xdr->tail[0].iov_len;
785 		page_base += len & 3;
786 		len -= len & 3;
787 		if (!rpcrdma_prepare_tail_iov(req, xdr, page_base, len))
788 			return false;
789 		kref_get(&req->rl_kref);
790 	}
791 
792 	return true;
793 }
794 
795 /**
796  * rpcrdma_prepare_send_sges - Construct SGEs for a Send WR
797  * @r_xprt: controlling transport
798  * @req: context of RPC Call being marshalled
799  * @hdrlen: size of transport header, in bytes
800  * @xdr: xdr_buf containing RPC Call
801  * @rtype: chunk type being encoded
802  *
803  * Returns 0 on success; otherwise a negative errno is returned.
804  */
805 inline int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
806 				     struct rpcrdma_req *req, u32 hdrlen,
807 				     struct xdr_buf *xdr,
808 				     enum rpcrdma_chunktype rtype)
809 {
810 	int ret;
811 
812 	ret = -EAGAIN;
813 	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
814 	if (!req->rl_sendctx)
815 		goto out_nosc;
816 	req->rl_sendctx->sc_unmap_count = 0;
817 	req->rl_sendctx->sc_req = req;
818 	kref_init(&req->rl_kref);
819 	req->rl_wr.wr_cqe = &req->rl_sendctx->sc_cqe;
820 	req->rl_wr.sg_list = req->rl_sendctx->sc_sges;
821 	req->rl_wr.num_sge = 0;
822 	req->rl_wr.opcode = IB_WR_SEND;
823 
824 	rpcrdma_prepare_hdr_sge(r_xprt, req, hdrlen);
825 
826 	ret = -EIO;
827 	switch (rtype) {
828 	case rpcrdma_noch_pullup:
829 		if (!rpcrdma_prepare_noch_pullup(r_xprt, req, xdr))
830 			goto out_unmap;
831 		break;
832 	case rpcrdma_noch_mapped:
833 		if (!rpcrdma_prepare_noch_mapped(r_xprt, req, xdr))
834 			goto out_unmap;
835 		break;
836 	case rpcrdma_readch:
837 		if (!rpcrdma_prepare_readch(r_xprt, req, xdr))
838 			goto out_unmap;
839 		break;
840 	case rpcrdma_areadch:
841 		break;
842 	default:
843 		goto out_unmap;
844 	}
845 
846 	return 0;
847 
848 out_unmap:
849 	rpcrdma_sendctx_unmap(req->rl_sendctx);
850 out_nosc:
851 	trace_xprtrdma_prepsend_failed(&req->rl_slot, ret);
852 	return ret;
853 }
854 
855 /**
856  * rpcrdma_marshal_req - Marshal and send one RPC request
857  * @r_xprt: controlling transport
858  * @rqst: RPC request to be marshaled
859  *
860  * For the RPC in "rqst", this function:
861  *  - Chooses the transfer mode (eg., RDMA_MSG or RDMA_NOMSG)
862  *  - Registers Read, Write, and Reply chunks
863  *  - Constructs the transport header
864  *  - Posts a Send WR to send the transport header and request
865  *
866  * Returns:
867  *	%0 if the RPC was sent successfully,
868  *	%-ENOTCONN if the connection was lost,
869  *	%-EAGAIN if the caller should call again with the same arguments,
870  *	%-ENOBUFS if the caller should call again after a delay,
871  *	%-EMSGSIZE if the transport header is too small,
872  *	%-EIO if a permanent problem occurred while marshaling.
873  */
874 int
875 rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
876 {
877 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
878 	struct xdr_stream *xdr = &req->rl_stream;
879 	enum rpcrdma_chunktype rtype, wtype;
880 	struct xdr_buf *buf = &rqst->rq_snd_buf;
881 	bool ddp_allowed;
882 	__be32 *p;
883 	int ret;
884 
885 	if (unlikely(rqst->rq_rcv_buf.flags & XDRBUF_SPARSE_PAGES)) {
886 		ret = rpcrdma_alloc_sparse_pages(&rqst->rq_rcv_buf);
887 		if (ret)
888 			return ret;
889 	}
890 
891 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
892 	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
893 			rqst);
894 
895 	/* Fixed header fields */
896 	ret = -EMSGSIZE;
897 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
898 	if (!p)
899 		goto out_err;
900 	*p++ = rqst->rq_xid;
901 	*p++ = rpcrdma_version;
902 	*p++ = r_xprt->rx_buf.rb_max_requests;
903 
904 	/* When the ULP employs a GSS flavor that guarantees integrity
905 	 * or privacy, direct data placement of individual data items
906 	 * is not allowed.
907 	 */
908 	ddp_allowed = !test_bit(RPCAUTH_AUTH_DATATOUCH,
909 				&rqst->rq_cred->cr_auth->au_flags);
910 
911 	/*
912 	 * Chunks needed for results?
913 	 *
914 	 * o If the expected result is under the inline threshold, all ops
915 	 *   return as inline.
916 	 * o Large read ops return data as write chunk(s), header as
917 	 *   inline.
918 	 * o Large non-read ops return as a single reply chunk.
919 	 */
920 	if (rpcrdma_results_inline(r_xprt, rqst))
921 		wtype = rpcrdma_noch;
922 	else if ((ddp_allowed && rqst->rq_rcv_buf.flags & XDRBUF_READ) &&
923 		 rpcrdma_nonpayload_inline(r_xprt, rqst))
924 		wtype = rpcrdma_writech;
925 	else
926 		wtype = rpcrdma_replych;
927 
928 	/*
929 	 * Chunks needed for arguments?
930 	 *
931 	 * o If the total request is under the inline threshold, all ops
932 	 *   are sent as inline.
933 	 * o Large write ops transmit data as read chunk(s), header as
934 	 *   inline.
935 	 * o Large non-write ops are sent with the entire message as a
936 	 *   single read chunk (protocol 0-position special case).
937 	 *
938 	 * This assumes that the upper layer does not present a request
939 	 * that both has a data payload, and whose non-data arguments
940 	 * by themselves are larger than the inline threshold.
941 	 */
942 	if (rpcrdma_args_inline(r_xprt, rqst)) {
943 		*p++ = rdma_msg;
944 		rtype = buf->len < rdmab_length(req->rl_sendbuf) ?
945 			rpcrdma_noch_pullup : rpcrdma_noch_mapped;
946 	} else if (ddp_allowed && buf->flags & XDRBUF_WRITE) {
947 		*p++ = rdma_msg;
948 		rtype = rpcrdma_readch;
949 	} else {
950 		r_xprt->rx_stats.nomsg_call_count++;
951 		*p++ = rdma_nomsg;
952 		rtype = rpcrdma_areadch;
953 	}
954 
955 	/* This implementation supports the following combinations
956 	 * of chunk lists in one RPC-over-RDMA Call message:
957 	 *
958 	 *   - Read list
959 	 *   - Write list
960 	 *   - Reply chunk
961 	 *   - Read list + Reply chunk
962 	 *
963 	 * It might not yet support the following combinations:
964 	 *
965 	 *   - Read list + Write list
966 	 *
967 	 * It does not support the following combinations:
968 	 *
969 	 *   - Write list + Reply chunk
970 	 *   - Read list + Write list + Reply chunk
971 	 *
972 	 * This implementation supports only a single chunk in each
973 	 * Read or Write list. Thus for example the client cannot
974 	 * send a Call message with a Position Zero Read chunk and a
975 	 * regular Read chunk at the same time.
976 	 */
977 	ret = rpcrdma_encode_read_list(r_xprt, req, rqst, rtype);
978 	if (ret)
979 		goto out_err;
980 	ret = rpcrdma_encode_write_list(r_xprt, req, rqst, wtype);
981 	if (ret)
982 		goto out_err;
983 	ret = rpcrdma_encode_reply_chunk(r_xprt, req, rqst, wtype);
984 	if (ret)
985 		goto out_err;
986 
987 	ret = rpcrdma_prepare_send_sges(r_xprt, req, req->rl_hdrbuf.len,
988 					buf, rtype);
989 	if (ret)
990 		goto out_err;
991 
992 	trace_xprtrdma_marshal(req, rtype, wtype);
993 	return 0;
994 
995 out_err:
996 	trace_xprtrdma_marshal_failed(rqst, ret);
997 	r_xprt->rx_stats.failed_marshal_count++;
998 	frwr_reset(req);
999 	return ret;
1000 }
1001 
1002 static void __rpcrdma_update_cwnd_locked(struct rpc_xprt *xprt,
1003 					 struct rpcrdma_buffer *buf,
1004 					 u32 grant)
1005 {
1006 	buf->rb_credits = grant;
1007 	xprt->cwnd = grant << RPC_CWNDSHIFT;
1008 }
1009 
1010 static void rpcrdma_update_cwnd(struct rpcrdma_xprt *r_xprt, u32 grant)
1011 {
1012 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1013 
1014 	spin_lock(&xprt->transport_lock);
1015 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, grant);
1016 	spin_unlock(&xprt->transport_lock);
1017 }
1018 
1019 /**
1020  * rpcrdma_reset_cwnd - Reset the xprt's congestion window
1021  * @r_xprt: controlling transport instance
1022  *
1023  * Prepare @r_xprt for the next connection by reinitializing
1024  * its credit grant to one (see RFC 8166, Section 3.3.3).
1025  */
1026 void rpcrdma_reset_cwnd(struct rpcrdma_xprt *r_xprt)
1027 {
1028 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1029 
1030 	spin_lock(&xprt->transport_lock);
1031 	xprt->cong = 0;
1032 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, 1);
1033 	spin_unlock(&xprt->transport_lock);
1034 }
1035 
1036 /**
1037  * rpcrdma_inline_fixup - Scatter inline received data into rqst's iovecs
1038  * @rqst: controlling RPC request
1039  * @srcp: points to RPC message payload in receive buffer
1040  * @copy_len: remaining length of receive buffer content
1041  * @pad: Write chunk pad bytes needed (zero for pure inline)
1042  *
1043  * The upper layer has set the maximum number of bytes it can
1044  * receive in each component of rq_rcv_buf. These values are set in
1045  * the head.iov_len, page_len, tail.iov_len, and buflen fields.
1046  *
1047  * Unlike the TCP equivalent (xdr_partial_copy_from_skb), in
1048  * many cases this function simply updates iov_base pointers in
1049  * rq_rcv_buf to point directly to the received reply data, to
1050  * avoid copying reply data.
1051  *
1052  * Returns the count of bytes which had to be memcopied.
1053  */
1054 static unsigned long
1055 rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
1056 {
1057 	unsigned long fixup_copy_count;
1058 	int i, npages, curlen;
1059 	char *destp;
1060 	struct page **ppages;
1061 	int page_base;
1062 
1063 	/* The head iovec is redirected to the RPC reply message
1064 	 * in the receive buffer, to avoid a memcopy.
1065 	 */
1066 	rqst->rq_rcv_buf.head[0].iov_base = srcp;
1067 	rqst->rq_private_buf.head[0].iov_base = srcp;
1068 
1069 	/* The contents of the receive buffer that follow
1070 	 * head.iov_len bytes are copied into the page list.
1071 	 */
1072 	curlen = rqst->rq_rcv_buf.head[0].iov_len;
1073 	if (curlen > copy_len)
1074 		curlen = copy_len;
1075 	srcp += curlen;
1076 	copy_len -= curlen;
1077 
1078 	ppages = rqst->rq_rcv_buf.pages +
1079 		(rqst->rq_rcv_buf.page_base >> PAGE_SHIFT);
1080 	page_base = offset_in_page(rqst->rq_rcv_buf.page_base);
1081 	fixup_copy_count = 0;
1082 	if (copy_len && rqst->rq_rcv_buf.page_len) {
1083 		int pagelist_len;
1084 
1085 		pagelist_len = rqst->rq_rcv_buf.page_len;
1086 		if (pagelist_len > copy_len)
1087 			pagelist_len = copy_len;
1088 		npages = PAGE_ALIGN(page_base + pagelist_len) >> PAGE_SHIFT;
1089 		for (i = 0; i < npages; i++) {
1090 			curlen = PAGE_SIZE - page_base;
1091 			if (curlen > pagelist_len)
1092 				curlen = pagelist_len;
1093 
1094 			destp = kmap_atomic(ppages[i]);
1095 			memcpy(destp + page_base, srcp, curlen);
1096 			flush_dcache_page(ppages[i]);
1097 			kunmap_atomic(destp);
1098 			srcp += curlen;
1099 			copy_len -= curlen;
1100 			fixup_copy_count += curlen;
1101 			pagelist_len -= curlen;
1102 			if (!pagelist_len)
1103 				break;
1104 			page_base = 0;
1105 		}
1106 
1107 		/* Implicit padding for the last segment in a Write
1108 		 * chunk is inserted inline at the front of the tail
1109 		 * iovec. The upper layer ignores the content of
1110 		 * the pad. Simply ensure inline content in the tail
1111 		 * that follows the Write chunk is properly aligned.
1112 		 */
1113 		if (pad)
1114 			srcp -= pad;
1115 	}
1116 
1117 	/* The tail iovec is redirected to the remaining data
1118 	 * in the receive buffer, to avoid a memcopy.
1119 	 */
1120 	if (copy_len || pad) {
1121 		rqst->rq_rcv_buf.tail[0].iov_base = srcp;
1122 		rqst->rq_private_buf.tail[0].iov_base = srcp;
1123 	}
1124 
1125 	if (fixup_copy_count)
1126 		trace_xprtrdma_fixup(rqst, fixup_copy_count);
1127 	return fixup_copy_count;
1128 }
1129 
1130 /* By convention, backchannel calls arrive via rdma_msg type
1131  * messages, and never populate the chunk lists. This makes
1132  * the RPC/RDMA header small and fixed in size, so it is
1133  * straightforward to check the RPC header's direction field.
1134  */
1135 static bool
1136 rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1137 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1138 {
1139 	struct xdr_stream *xdr = &rep->rr_stream;
1140 	__be32 *p;
1141 
1142 	if (rep->rr_proc != rdma_msg)
1143 		return false;
1144 
1145 	/* Peek at stream contents without advancing. */
1146 	p = xdr_inline_decode(xdr, 0);
1147 
1148 	/* Chunk lists */
1149 	if (xdr_item_is_present(p++))
1150 		return false;
1151 	if (xdr_item_is_present(p++))
1152 		return false;
1153 	if (xdr_item_is_present(p++))
1154 		return false;
1155 
1156 	/* RPC header */
1157 	if (*p++ != rep->rr_xid)
1158 		return false;
1159 	if (*p != cpu_to_be32(RPC_CALL))
1160 		return false;
1161 
1162 	/* Now that we are sure this is a backchannel call,
1163 	 * advance to the RPC header.
1164 	 */
1165 	p = xdr_inline_decode(xdr, 3 * sizeof(*p));
1166 	if (unlikely(!p))
1167 		goto out_short;
1168 
1169 	rpcrdma_bc_receive_call(r_xprt, rep);
1170 	return true;
1171 
1172 out_short:
1173 	pr_warn("RPC/RDMA short backward direction call\n");
1174 	return true;
1175 }
1176 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
1177 {
1178 	return false;
1179 }
1180 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
1181 
1182 static int decode_rdma_segment(struct xdr_stream *xdr, u32 *length)
1183 {
1184 	u32 handle;
1185 	u64 offset;
1186 	__be32 *p;
1187 
1188 	p = xdr_inline_decode(xdr, 4 * sizeof(*p));
1189 	if (unlikely(!p))
1190 		return -EIO;
1191 
1192 	xdr_decode_rdma_segment(p, &handle, length, &offset);
1193 	trace_xprtrdma_decode_seg(handle, *length, offset);
1194 	return 0;
1195 }
1196 
1197 static int decode_write_chunk(struct xdr_stream *xdr, u32 *length)
1198 {
1199 	u32 segcount, seglength;
1200 	__be32 *p;
1201 
1202 	p = xdr_inline_decode(xdr, sizeof(*p));
1203 	if (unlikely(!p))
1204 		return -EIO;
1205 
1206 	*length = 0;
1207 	segcount = be32_to_cpup(p);
1208 	while (segcount--) {
1209 		if (decode_rdma_segment(xdr, &seglength))
1210 			return -EIO;
1211 		*length += seglength;
1212 	}
1213 
1214 	return 0;
1215 }
1216 
1217 /* In RPC-over-RDMA Version One replies, a Read list is never
1218  * expected. This decoder is a stub that returns an error if
1219  * a Read list is present.
1220  */
1221 static int decode_read_list(struct xdr_stream *xdr)
1222 {
1223 	__be32 *p;
1224 
1225 	p = xdr_inline_decode(xdr, sizeof(*p));
1226 	if (unlikely(!p))
1227 		return -EIO;
1228 	if (unlikely(xdr_item_is_present(p)))
1229 		return -EIO;
1230 	return 0;
1231 }
1232 
1233 /* Supports only one Write chunk in the Write list
1234  */
1235 static int decode_write_list(struct xdr_stream *xdr, u32 *length)
1236 {
1237 	u32 chunklen;
1238 	bool first;
1239 	__be32 *p;
1240 
1241 	*length = 0;
1242 	first = true;
1243 	do {
1244 		p = xdr_inline_decode(xdr, sizeof(*p));
1245 		if (unlikely(!p))
1246 			return -EIO;
1247 		if (xdr_item_is_absent(p))
1248 			break;
1249 		if (!first)
1250 			return -EIO;
1251 
1252 		if (decode_write_chunk(xdr, &chunklen))
1253 			return -EIO;
1254 		*length += chunklen;
1255 		first = false;
1256 	} while (true);
1257 	return 0;
1258 }
1259 
1260 static int decode_reply_chunk(struct xdr_stream *xdr, u32 *length)
1261 {
1262 	__be32 *p;
1263 
1264 	p = xdr_inline_decode(xdr, sizeof(*p));
1265 	if (unlikely(!p))
1266 		return -EIO;
1267 
1268 	*length = 0;
1269 	if (xdr_item_is_present(p))
1270 		if (decode_write_chunk(xdr, length))
1271 			return -EIO;
1272 	return 0;
1273 }
1274 
1275 static int
1276 rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1277 		   struct rpc_rqst *rqst)
1278 {
1279 	struct xdr_stream *xdr = &rep->rr_stream;
1280 	u32 writelist, replychunk, rpclen;
1281 	char *base;
1282 
1283 	/* Decode the chunk lists */
1284 	if (decode_read_list(xdr))
1285 		return -EIO;
1286 	if (decode_write_list(xdr, &writelist))
1287 		return -EIO;
1288 	if (decode_reply_chunk(xdr, &replychunk))
1289 		return -EIO;
1290 
1291 	/* RDMA_MSG sanity checks */
1292 	if (unlikely(replychunk))
1293 		return -EIO;
1294 
1295 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
1296 	base = (char *)xdr_inline_decode(xdr, 0);
1297 	rpclen = xdr_stream_remaining(xdr);
1298 	r_xprt->rx_stats.fixup_copy_count +=
1299 		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
1300 
1301 	r_xprt->rx_stats.total_rdma_reply += writelist;
1302 	return rpclen + xdr_align_size(writelist);
1303 }
1304 
1305 static noinline int
1306 rpcrdma_decode_nomsg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1307 {
1308 	struct xdr_stream *xdr = &rep->rr_stream;
1309 	u32 writelist, replychunk;
1310 
1311 	/* Decode the chunk lists */
1312 	if (decode_read_list(xdr))
1313 		return -EIO;
1314 	if (decode_write_list(xdr, &writelist))
1315 		return -EIO;
1316 	if (decode_reply_chunk(xdr, &replychunk))
1317 		return -EIO;
1318 
1319 	/* RDMA_NOMSG sanity checks */
1320 	if (unlikely(writelist))
1321 		return -EIO;
1322 	if (unlikely(!replychunk))
1323 		return -EIO;
1324 
1325 	/* Reply chunk buffer already is the reply vector */
1326 	r_xprt->rx_stats.total_rdma_reply += replychunk;
1327 	return replychunk;
1328 }
1329 
1330 static noinline int
1331 rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1332 		     struct rpc_rqst *rqst)
1333 {
1334 	struct xdr_stream *xdr = &rep->rr_stream;
1335 	__be32 *p;
1336 
1337 	p = xdr_inline_decode(xdr, sizeof(*p));
1338 	if (unlikely(!p))
1339 		return -EIO;
1340 
1341 	switch (*p) {
1342 	case err_vers:
1343 		p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1344 		if (!p)
1345 			break;
1346 		trace_xprtrdma_err_vers(rqst, p, p + 1);
1347 		break;
1348 	case err_chunk:
1349 		trace_xprtrdma_err_chunk(rqst);
1350 		break;
1351 	default:
1352 		trace_xprtrdma_err_unrecognized(rqst, p);
1353 	}
1354 
1355 	return -EIO;
1356 }
1357 
1358 /* Perform XID lookup, reconstruction of the RPC reply, and
1359  * RPC completion while holding the transport lock to ensure
1360  * the rep, rqst, and rq_task pointers remain stable.
1361  */
1362 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
1363 {
1364 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1365 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1366 	struct rpc_rqst *rqst = rep->rr_rqst;
1367 	int status;
1368 
1369 	switch (rep->rr_proc) {
1370 	case rdma_msg:
1371 		status = rpcrdma_decode_msg(r_xprt, rep, rqst);
1372 		break;
1373 	case rdma_nomsg:
1374 		status = rpcrdma_decode_nomsg(r_xprt, rep);
1375 		break;
1376 	case rdma_error:
1377 		status = rpcrdma_decode_error(r_xprt, rep, rqst);
1378 		break;
1379 	default:
1380 		status = -EIO;
1381 	}
1382 	if (status < 0)
1383 		goto out_badheader;
1384 
1385 out:
1386 	spin_lock(&xprt->queue_lock);
1387 	xprt_complete_rqst(rqst->rq_task, status);
1388 	xprt_unpin_rqst(rqst);
1389 	spin_unlock(&xprt->queue_lock);
1390 	return;
1391 
1392 out_badheader:
1393 	trace_xprtrdma_reply_hdr_err(rep);
1394 	r_xprt->rx_stats.bad_reply_count++;
1395 	rqst->rq_task->tk_status = status;
1396 	status = 0;
1397 	goto out;
1398 }
1399 
1400 static void rpcrdma_reply_done(struct kref *kref)
1401 {
1402 	struct rpcrdma_req *req =
1403 		container_of(kref, struct rpcrdma_req, rl_kref);
1404 
1405 	rpcrdma_complete_rqst(req->rl_reply);
1406 }
1407 
1408 /**
1409  * rpcrdma_reply_handler - Process received RPC/RDMA messages
1410  * @rep: Incoming rpcrdma_rep object to process
1411  *
1412  * Errors must result in the RPC task either being awakened, or
1413  * allowed to timeout, to discover the errors at that time.
1414  */
1415 void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
1416 {
1417 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1418 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1419 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1420 	struct rpcrdma_req *req;
1421 	struct rpc_rqst *rqst;
1422 	u32 credits;
1423 	__be32 *p;
1424 
1425 	/* Any data means we had a useful conversation, so
1426 	 * then we don't need to delay the next reconnect.
1427 	 */
1428 	if (xprt->reestablish_timeout)
1429 		xprt->reestablish_timeout = 0;
1430 
1431 	/* Fixed transport header fields */
1432 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
1433 			rep->rr_hdrbuf.head[0].iov_base, NULL);
1434 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
1435 	if (unlikely(!p))
1436 		goto out_shortreply;
1437 	rep->rr_xid = *p++;
1438 	rep->rr_vers = *p++;
1439 	credits = be32_to_cpu(*p++);
1440 	rep->rr_proc = *p++;
1441 
1442 	if (rep->rr_vers != rpcrdma_version)
1443 		goto out_badversion;
1444 
1445 	if (rpcrdma_is_bcall(r_xprt, rep))
1446 		return;
1447 
1448 	/* Match incoming rpcrdma_rep to an rpcrdma_req to
1449 	 * get context for handling any incoming chunks.
1450 	 */
1451 	spin_lock(&xprt->queue_lock);
1452 	rqst = xprt_lookup_rqst(xprt, rep->rr_xid);
1453 	if (!rqst)
1454 		goto out_norqst;
1455 	xprt_pin_rqst(rqst);
1456 	spin_unlock(&xprt->queue_lock);
1457 
1458 	if (credits == 0)
1459 		credits = 1;	/* don't deadlock */
1460 	else if (credits > r_xprt->rx_ep->re_max_requests)
1461 		credits = r_xprt->rx_ep->re_max_requests;
1462 	if (buf->rb_credits != credits)
1463 		rpcrdma_update_cwnd(r_xprt, credits);
1464 	rpcrdma_post_recvs(r_xprt, false);
1465 
1466 	req = rpcr_to_rdmar(rqst);
1467 	if (unlikely(req->rl_reply))
1468 		rpcrdma_recv_buffer_put(req->rl_reply);
1469 	req->rl_reply = rep;
1470 	rep->rr_rqst = rqst;
1471 
1472 	trace_xprtrdma_reply(rqst->rq_task, rep, credits);
1473 
1474 	if (rep->rr_wc_flags & IB_WC_WITH_INVALIDATE)
1475 		frwr_reminv(rep, &req->rl_registered);
1476 	if (!list_empty(&req->rl_registered))
1477 		frwr_unmap_async(r_xprt, req);
1478 		/* LocalInv completion will complete the RPC */
1479 	else
1480 		kref_put(&req->rl_kref, rpcrdma_reply_done);
1481 	return;
1482 
1483 out_badversion:
1484 	trace_xprtrdma_reply_vers_err(rep);
1485 	goto out;
1486 
1487 out_norqst:
1488 	spin_unlock(&xprt->queue_lock);
1489 	trace_xprtrdma_reply_rqst_err(rep);
1490 	goto out;
1491 
1492 out_shortreply:
1493 	trace_xprtrdma_reply_short_err(rep);
1494 
1495 out:
1496 	rpcrdma_recv_buffer_put(rep);
1497 }
1498