xref: /openbmc/linux/net/sunrpc/xprtrdma/rpc_rdma.c (revision 63705da3)
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 /* Convert @vec to a single SGL element.
208  *
209  * Returns pointer to next available SGE, and bumps the total number
210  * of SGEs consumed.
211  */
212 static struct rpcrdma_mr_seg *
213 rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg,
214 		     unsigned int *n)
215 {
216 	seg->mr_page = virt_to_page(vec->iov_base);
217 	seg->mr_offset = offset_in_page(vec->iov_base);
218 	seg->mr_len = vec->iov_len;
219 	++seg;
220 	++(*n);
221 	return seg;
222 }
223 
224 /* Convert @xdrbuf into SGEs no larger than a page each. As they
225  * are registered, these SGEs are then coalesced into RDMA segments
226  * when the selected memreg mode supports it.
227  *
228  * Returns positive number of SGEs consumed, or a negative errno.
229  */
230 
231 static int
232 rpcrdma_convert_iovs(struct rpcrdma_xprt *r_xprt, struct xdr_buf *xdrbuf,
233 		     unsigned int pos, enum rpcrdma_chunktype type,
234 		     struct rpcrdma_mr_seg *seg)
235 {
236 	unsigned long page_base;
237 	unsigned int len, n;
238 	struct page **ppages;
239 
240 	n = 0;
241 	if (pos == 0)
242 		seg = rpcrdma_convert_kvec(&xdrbuf->head[0], seg, &n);
243 
244 	len = xdrbuf->page_len;
245 	ppages = xdrbuf->pages + (xdrbuf->page_base >> PAGE_SHIFT);
246 	page_base = offset_in_page(xdrbuf->page_base);
247 	while (len) {
248 		seg->mr_page = *ppages;
249 		seg->mr_offset = page_base;
250 		seg->mr_len = min_t(u32, PAGE_SIZE - page_base, len);
251 		len -= seg->mr_len;
252 		++ppages;
253 		++seg;
254 		++n;
255 		page_base = 0;
256 	}
257 
258 	if (type == rpcrdma_readch || type == rpcrdma_writech)
259 		goto out;
260 
261 	if (xdrbuf->tail[0].iov_len)
262 		rpcrdma_convert_kvec(&xdrbuf->tail[0], seg, &n);
263 
264 out:
265 	if (unlikely(n > RPCRDMA_MAX_SEGS))
266 		return -EIO;
267 	return n;
268 }
269 
270 static int
271 encode_rdma_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr)
272 {
273 	__be32 *p;
274 
275 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
276 	if (unlikely(!p))
277 		return -EMSGSIZE;
278 
279 	xdr_encode_rdma_segment(p, mr->mr_handle, mr->mr_length, mr->mr_offset);
280 	return 0;
281 }
282 
283 static int
284 encode_read_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr,
285 		    u32 position)
286 {
287 	__be32 *p;
288 
289 	p = xdr_reserve_space(xdr, 6 * sizeof(*p));
290 	if (unlikely(!p))
291 		return -EMSGSIZE;
292 
293 	*p++ = xdr_one;			/* Item present */
294 	xdr_encode_read_segment(p, position, mr->mr_handle, mr->mr_length,
295 				mr->mr_offset);
296 	return 0;
297 }
298 
299 static struct rpcrdma_mr_seg *rpcrdma_mr_prepare(struct rpcrdma_xprt *r_xprt,
300 						 struct rpcrdma_req *req,
301 						 struct rpcrdma_mr_seg *seg,
302 						 int nsegs, bool writing,
303 						 struct rpcrdma_mr **mr)
304 {
305 	*mr = rpcrdma_mr_pop(&req->rl_free_mrs);
306 	if (!*mr) {
307 		*mr = rpcrdma_mr_get(r_xprt);
308 		if (!*mr)
309 			goto out_getmr_err;
310 		(*mr)->mr_req = req;
311 	}
312 
313 	rpcrdma_mr_push(*mr, &req->rl_registered);
314 	return frwr_map(r_xprt, seg, nsegs, writing, req->rl_slot.rq_xid, *mr);
315 
316 out_getmr_err:
317 	trace_xprtrdma_nomrs_err(r_xprt, req);
318 	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
319 	rpcrdma_mrs_refresh(r_xprt);
320 	return ERR_PTR(-EAGAIN);
321 }
322 
323 /* Register and XDR encode the Read list. Supports encoding a list of read
324  * segments that belong to a single read chunk.
325  *
326  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
327  *
328  *  Read chunklist (a linked list):
329  *   N elements, position P (same P for all chunks of same arg!):
330  *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
331  *
332  * Returns zero on success, or a negative errno if a failure occurred.
333  * @xdr is advanced to the next position in the stream.
334  *
335  * Only a single @pos value is currently supported.
336  */
337 static int rpcrdma_encode_read_list(struct rpcrdma_xprt *r_xprt,
338 				    struct rpcrdma_req *req,
339 				    struct rpc_rqst *rqst,
340 				    enum rpcrdma_chunktype rtype)
341 {
342 	struct xdr_stream *xdr = &req->rl_stream;
343 	struct rpcrdma_mr_seg *seg;
344 	struct rpcrdma_mr *mr;
345 	unsigned int pos;
346 	int nsegs;
347 
348 	if (rtype == rpcrdma_noch_pullup || rtype == rpcrdma_noch_mapped)
349 		goto done;
350 
351 	pos = rqst->rq_snd_buf.head[0].iov_len;
352 	if (rtype == rpcrdma_areadch)
353 		pos = 0;
354 	seg = req->rl_segments;
355 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_snd_buf, pos,
356 				     rtype, seg);
357 	if (nsegs < 0)
358 		return nsegs;
359 
360 	do {
361 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, false, &mr);
362 		if (IS_ERR(seg))
363 			return PTR_ERR(seg);
364 
365 		if (encode_read_segment(xdr, mr, pos) < 0)
366 			return -EMSGSIZE;
367 
368 		trace_xprtrdma_chunk_read(rqst->rq_task, pos, mr, nsegs);
369 		r_xprt->rx_stats.read_chunk_count++;
370 		nsegs -= mr->mr_nents;
371 	} while (nsegs);
372 
373 done:
374 	if (xdr_stream_encode_item_absent(xdr) < 0)
375 		return -EMSGSIZE;
376 	return 0;
377 }
378 
379 /* Register and XDR encode the Write list. Supports encoding a list
380  * containing one array of plain segments that belong to a single
381  * write chunk.
382  *
383  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
384  *
385  *  Write chunklist (a list of (one) counted array):
386  *   N elements:
387  *    1 - N - HLOO - HLOO - ... - HLOO - 0
388  *
389  * Returns zero on success, or a negative errno if a failure occurred.
390  * @xdr is advanced to the next position in the stream.
391  *
392  * Only a single Write chunk is currently supported.
393  */
394 static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt,
395 				     struct rpcrdma_req *req,
396 				     struct rpc_rqst *rqst,
397 				     enum rpcrdma_chunktype wtype)
398 {
399 	struct xdr_stream *xdr = &req->rl_stream;
400 	struct rpcrdma_ep *ep = r_xprt->rx_ep;
401 	struct rpcrdma_mr_seg *seg;
402 	struct rpcrdma_mr *mr;
403 	int nsegs, nchunks;
404 	__be32 *segcount;
405 
406 	if (wtype != rpcrdma_writech)
407 		goto done;
408 
409 	seg = req->rl_segments;
410 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf,
411 				     rqst->rq_rcv_buf.head[0].iov_len,
412 				     wtype, seg);
413 	if (nsegs < 0)
414 		return nsegs;
415 
416 	if (xdr_stream_encode_item_present(xdr) < 0)
417 		return -EMSGSIZE;
418 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
419 	if (unlikely(!segcount))
420 		return -EMSGSIZE;
421 	/* Actual value encoded below */
422 
423 	nchunks = 0;
424 	do {
425 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
426 		if (IS_ERR(seg))
427 			return PTR_ERR(seg);
428 
429 		if (encode_rdma_segment(xdr, mr) < 0)
430 			return -EMSGSIZE;
431 
432 		trace_xprtrdma_chunk_write(rqst->rq_task, mr, nsegs);
433 		r_xprt->rx_stats.write_chunk_count++;
434 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
435 		nchunks++;
436 		nsegs -= mr->mr_nents;
437 	} while (nsegs);
438 
439 	if (xdr_pad_size(rqst->rq_rcv_buf.page_len)) {
440 		if (encode_rdma_segment(xdr, ep->re_write_pad_mr) < 0)
441 			return -EMSGSIZE;
442 
443 		trace_xprtrdma_chunk_wp(rqst->rq_task, ep->re_write_pad_mr,
444 					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 	}
450 
451 	/* Update count of segments in this Write chunk */
452 	*segcount = cpu_to_be32(nchunks);
453 
454 done:
455 	if (xdr_stream_encode_item_absent(xdr) < 0)
456 		return -EMSGSIZE;
457 	return 0;
458 }
459 
460 /* Register and XDR encode the Reply chunk. Supports encoding an array
461  * of plain segments that belong to a single write (reply) chunk.
462  *
463  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
464  *
465  *  Reply chunk (a counted array):
466  *   N elements:
467  *    1 - N - HLOO - HLOO - ... - HLOO
468  *
469  * Returns zero on success, or a negative errno if a failure occurred.
470  * @xdr is advanced to the next position in the stream.
471  */
472 static int rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt,
473 				      struct rpcrdma_req *req,
474 				      struct rpc_rqst *rqst,
475 				      enum rpcrdma_chunktype wtype)
476 {
477 	struct xdr_stream *xdr = &req->rl_stream;
478 	struct rpcrdma_mr_seg *seg;
479 	struct rpcrdma_mr *mr;
480 	int nsegs, nchunks;
481 	__be32 *segcount;
482 
483 	if (wtype != rpcrdma_replych) {
484 		if (xdr_stream_encode_item_absent(xdr) < 0)
485 			return -EMSGSIZE;
486 		return 0;
487 	}
488 
489 	seg = req->rl_segments;
490 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf, 0, wtype, seg);
491 	if (nsegs < 0)
492 		return nsegs;
493 
494 	if (xdr_stream_encode_item_present(xdr) < 0)
495 		return -EMSGSIZE;
496 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
497 	if (unlikely(!segcount))
498 		return -EMSGSIZE;
499 	/* Actual value encoded below */
500 
501 	nchunks = 0;
502 	do {
503 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
504 		if (IS_ERR(seg))
505 			return PTR_ERR(seg);
506 
507 		if (encode_rdma_segment(xdr, mr) < 0)
508 			return -EMSGSIZE;
509 
510 		trace_xprtrdma_chunk_reply(rqst->rq_task, mr, nsegs);
511 		r_xprt->rx_stats.reply_chunk_count++;
512 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
513 		nchunks++;
514 		nsegs -= mr->mr_nents;
515 	} while (nsegs);
516 
517 	/* Update count of segments in the Reply chunk */
518 	*segcount = cpu_to_be32(nchunks);
519 
520 	return 0;
521 }
522 
523 static void rpcrdma_sendctx_done(struct kref *kref)
524 {
525 	struct rpcrdma_req *req =
526 		container_of(kref, struct rpcrdma_req, rl_kref);
527 	struct rpcrdma_rep *rep = req->rl_reply;
528 
529 	rpcrdma_complete_rqst(rep);
530 	rep->rr_rxprt->rx_stats.reply_waits_for_send++;
531 }
532 
533 /**
534  * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
535  * @sc: sendctx containing SGEs to unmap
536  *
537  */
538 void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
539 {
540 	struct rpcrdma_regbuf *rb = sc->sc_req->rl_sendbuf;
541 	struct ib_sge *sge;
542 
543 	if (!sc->sc_unmap_count)
544 		return;
545 
546 	/* The first two SGEs contain the transport header and
547 	 * the inline buffer. These are always left mapped so
548 	 * they can be cheaply re-used.
549 	 */
550 	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
551 	     ++sge, --sc->sc_unmap_count)
552 		ib_dma_unmap_page(rdmab_device(rb), sge->addr, sge->length,
553 				  DMA_TO_DEVICE);
554 
555 	kref_put(&sc->sc_req->rl_kref, rpcrdma_sendctx_done);
556 }
557 
558 /* Prepare an SGE for the RPC-over-RDMA transport header.
559  */
560 static void rpcrdma_prepare_hdr_sge(struct rpcrdma_xprt *r_xprt,
561 				    struct rpcrdma_req *req, u32 len)
562 {
563 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
564 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
565 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
566 
567 	sge->addr = rdmab_addr(rb);
568 	sge->length = len;
569 	sge->lkey = rdmab_lkey(rb);
570 
571 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
572 				      DMA_TO_DEVICE);
573 }
574 
575 /* The head iovec is straightforward, as it is usually already
576  * DMA-mapped. Sync the content that has changed.
577  */
578 static bool rpcrdma_prepare_head_iov(struct rpcrdma_xprt *r_xprt,
579 				     struct rpcrdma_req *req, unsigned int len)
580 {
581 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
582 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
583 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
584 
585 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
586 		return false;
587 
588 	sge->addr = rdmab_addr(rb);
589 	sge->length = len;
590 	sge->lkey = rdmab_lkey(rb);
591 
592 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
593 				      DMA_TO_DEVICE);
594 	return true;
595 }
596 
597 /* If there is a page list present, DMA map and prepare an
598  * SGE for each page to be sent.
599  */
600 static bool rpcrdma_prepare_pagelist(struct rpcrdma_req *req,
601 				     struct xdr_buf *xdr)
602 {
603 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
604 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
605 	unsigned int page_base, len, remaining;
606 	struct page **ppages;
607 	struct ib_sge *sge;
608 
609 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
610 	page_base = offset_in_page(xdr->page_base);
611 	remaining = xdr->page_len;
612 	while (remaining) {
613 		sge = &sc->sc_sges[req->rl_wr.num_sge++];
614 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
615 		sge->addr = ib_dma_map_page(rdmab_device(rb), *ppages,
616 					    page_base, len, DMA_TO_DEVICE);
617 		if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
618 			goto out_mapping_err;
619 
620 		sge->length = len;
621 		sge->lkey = rdmab_lkey(rb);
622 
623 		sc->sc_unmap_count++;
624 		ppages++;
625 		remaining -= len;
626 		page_base = 0;
627 	}
628 
629 	return true;
630 
631 out_mapping_err:
632 	trace_xprtrdma_dma_maperr(sge->addr);
633 	return false;
634 }
635 
636 /* The tail iovec may include an XDR pad for the page list,
637  * as well as additional content, and may not reside in the
638  * same page as the head iovec.
639  */
640 static bool rpcrdma_prepare_tail_iov(struct rpcrdma_req *req,
641 				     struct xdr_buf *xdr,
642 				     unsigned int page_base, unsigned int len)
643 {
644 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
645 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
646 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
647 	struct page *page = virt_to_page(xdr->tail[0].iov_base);
648 
649 	sge->addr = ib_dma_map_page(rdmab_device(rb), page, page_base, len,
650 				    DMA_TO_DEVICE);
651 	if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
652 		goto out_mapping_err;
653 
654 	sge->length = len;
655 	sge->lkey = rdmab_lkey(rb);
656 	++sc->sc_unmap_count;
657 	return true;
658 
659 out_mapping_err:
660 	trace_xprtrdma_dma_maperr(sge->addr);
661 	return false;
662 }
663 
664 /* Copy the tail to the end of the head buffer.
665  */
666 static void rpcrdma_pullup_tail_iov(struct rpcrdma_xprt *r_xprt,
667 				    struct rpcrdma_req *req,
668 				    struct xdr_buf *xdr)
669 {
670 	unsigned char *dst;
671 
672 	dst = (unsigned char *)xdr->head[0].iov_base;
673 	dst += xdr->head[0].iov_len + xdr->page_len;
674 	memmove(dst, xdr->tail[0].iov_base, xdr->tail[0].iov_len);
675 	r_xprt->rx_stats.pullup_copy_count += xdr->tail[0].iov_len;
676 }
677 
678 /* Copy pagelist content into the head buffer.
679  */
680 static void rpcrdma_pullup_pagelist(struct rpcrdma_xprt *r_xprt,
681 				    struct rpcrdma_req *req,
682 				    struct xdr_buf *xdr)
683 {
684 	unsigned int len, page_base, remaining;
685 	struct page **ppages;
686 	unsigned char *src, *dst;
687 
688 	dst = (unsigned char *)xdr->head[0].iov_base;
689 	dst += xdr->head[0].iov_len;
690 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
691 	page_base = offset_in_page(xdr->page_base);
692 	remaining = xdr->page_len;
693 	while (remaining) {
694 		src = page_address(*ppages);
695 		src += page_base;
696 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
697 		memcpy(dst, src, len);
698 		r_xprt->rx_stats.pullup_copy_count += len;
699 
700 		ppages++;
701 		dst += len;
702 		remaining -= len;
703 		page_base = 0;
704 	}
705 }
706 
707 /* Copy the contents of @xdr into @rl_sendbuf and DMA sync it.
708  * When the head, pagelist, and tail are small, a pull-up copy
709  * is considerably less costly than DMA mapping the components
710  * of @xdr.
711  *
712  * Assumptions:
713  *  - the caller has already verified that the total length
714  *    of the RPC Call body will fit into @rl_sendbuf.
715  */
716 static bool rpcrdma_prepare_noch_pullup(struct rpcrdma_xprt *r_xprt,
717 					struct rpcrdma_req *req,
718 					struct xdr_buf *xdr)
719 {
720 	if (unlikely(xdr->tail[0].iov_len))
721 		rpcrdma_pullup_tail_iov(r_xprt, req, xdr);
722 
723 	if (unlikely(xdr->page_len))
724 		rpcrdma_pullup_pagelist(r_xprt, req, xdr);
725 
726 	/* The whole RPC message resides in the head iovec now */
727 	return rpcrdma_prepare_head_iov(r_xprt, req, xdr->len);
728 }
729 
730 static bool rpcrdma_prepare_noch_mapped(struct rpcrdma_xprt *r_xprt,
731 					struct rpcrdma_req *req,
732 					struct xdr_buf *xdr)
733 {
734 	struct kvec *tail = &xdr->tail[0];
735 
736 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
737 		return false;
738 	if (xdr->page_len)
739 		if (!rpcrdma_prepare_pagelist(req, xdr))
740 			return false;
741 	if (tail->iov_len)
742 		if (!rpcrdma_prepare_tail_iov(req, xdr,
743 					      offset_in_page(tail->iov_base),
744 					      tail->iov_len))
745 			return false;
746 
747 	if (req->rl_sendctx->sc_unmap_count)
748 		kref_get(&req->rl_kref);
749 	return true;
750 }
751 
752 static bool rpcrdma_prepare_readch(struct rpcrdma_xprt *r_xprt,
753 				   struct rpcrdma_req *req,
754 				   struct xdr_buf *xdr)
755 {
756 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
757 		return false;
758 
759 	/* If there is a Read chunk, the page list is being handled
760 	 * via explicit RDMA, and thus is skipped here.
761 	 */
762 
763 	/* Do not include the tail if it is only an XDR pad */
764 	if (xdr->tail[0].iov_len > 3) {
765 		unsigned int page_base, len;
766 
767 		/* If the content in the page list is an odd length,
768 		 * xdr_write_pages() adds a pad at the beginning of
769 		 * the tail iovec. Force the tail's non-pad content to
770 		 * land at the next XDR position in the Send message.
771 		 */
772 		page_base = offset_in_page(xdr->tail[0].iov_base);
773 		len = xdr->tail[0].iov_len;
774 		page_base += len & 3;
775 		len -= len & 3;
776 		if (!rpcrdma_prepare_tail_iov(req, xdr, page_base, len))
777 			return false;
778 		kref_get(&req->rl_kref);
779 	}
780 
781 	return true;
782 }
783 
784 /**
785  * rpcrdma_prepare_send_sges - Construct SGEs for a Send WR
786  * @r_xprt: controlling transport
787  * @req: context of RPC Call being marshalled
788  * @hdrlen: size of transport header, in bytes
789  * @xdr: xdr_buf containing RPC Call
790  * @rtype: chunk type being encoded
791  *
792  * Returns 0 on success; otherwise a negative errno is returned.
793  */
794 inline int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
795 				     struct rpcrdma_req *req, u32 hdrlen,
796 				     struct xdr_buf *xdr,
797 				     enum rpcrdma_chunktype rtype)
798 {
799 	int ret;
800 
801 	ret = -EAGAIN;
802 	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
803 	if (!req->rl_sendctx)
804 		goto out_nosc;
805 	req->rl_sendctx->sc_unmap_count = 0;
806 	req->rl_sendctx->sc_req = req;
807 	kref_init(&req->rl_kref);
808 	req->rl_wr.wr_cqe = &req->rl_sendctx->sc_cqe;
809 	req->rl_wr.sg_list = req->rl_sendctx->sc_sges;
810 	req->rl_wr.num_sge = 0;
811 	req->rl_wr.opcode = IB_WR_SEND;
812 
813 	rpcrdma_prepare_hdr_sge(r_xprt, req, hdrlen);
814 
815 	ret = -EIO;
816 	switch (rtype) {
817 	case rpcrdma_noch_pullup:
818 		if (!rpcrdma_prepare_noch_pullup(r_xprt, req, xdr))
819 			goto out_unmap;
820 		break;
821 	case rpcrdma_noch_mapped:
822 		if (!rpcrdma_prepare_noch_mapped(r_xprt, req, xdr))
823 			goto out_unmap;
824 		break;
825 	case rpcrdma_readch:
826 		if (!rpcrdma_prepare_readch(r_xprt, req, xdr))
827 			goto out_unmap;
828 		break;
829 	case rpcrdma_areadch:
830 		break;
831 	default:
832 		goto out_unmap;
833 	}
834 
835 	return 0;
836 
837 out_unmap:
838 	rpcrdma_sendctx_unmap(req->rl_sendctx);
839 out_nosc:
840 	trace_xprtrdma_prepsend_failed(&req->rl_slot, ret);
841 	return ret;
842 }
843 
844 /**
845  * rpcrdma_marshal_req - Marshal and send one RPC request
846  * @r_xprt: controlling transport
847  * @rqst: RPC request to be marshaled
848  *
849  * For the RPC in "rqst", this function:
850  *  - Chooses the transfer mode (eg., RDMA_MSG or RDMA_NOMSG)
851  *  - Registers Read, Write, and Reply chunks
852  *  - Constructs the transport header
853  *  - Posts a Send WR to send the transport header and request
854  *
855  * Returns:
856  *	%0 if the RPC was sent successfully,
857  *	%-ENOTCONN if the connection was lost,
858  *	%-EAGAIN if the caller should call again with the same arguments,
859  *	%-ENOBUFS if the caller should call again after a delay,
860  *	%-EMSGSIZE if the transport header is too small,
861  *	%-EIO if a permanent problem occurred while marshaling.
862  */
863 int
864 rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
865 {
866 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
867 	struct xdr_stream *xdr = &req->rl_stream;
868 	enum rpcrdma_chunktype rtype, wtype;
869 	struct xdr_buf *buf = &rqst->rq_snd_buf;
870 	bool ddp_allowed;
871 	__be32 *p;
872 	int ret;
873 
874 	if (unlikely(rqst->rq_rcv_buf.flags & XDRBUF_SPARSE_PAGES)) {
875 		ret = rpcrdma_alloc_sparse_pages(&rqst->rq_rcv_buf);
876 		if (ret)
877 			return ret;
878 	}
879 
880 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
881 	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
882 			rqst);
883 
884 	/* Fixed header fields */
885 	ret = -EMSGSIZE;
886 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
887 	if (!p)
888 		goto out_err;
889 	*p++ = rqst->rq_xid;
890 	*p++ = rpcrdma_version;
891 	*p++ = r_xprt->rx_buf.rb_max_requests;
892 
893 	/* When the ULP employs a GSS flavor that guarantees integrity
894 	 * or privacy, direct data placement of individual data items
895 	 * is not allowed.
896 	 */
897 	ddp_allowed = !test_bit(RPCAUTH_AUTH_DATATOUCH,
898 				&rqst->rq_cred->cr_auth->au_flags);
899 
900 	/*
901 	 * Chunks needed for results?
902 	 *
903 	 * o If the expected result is under the inline threshold, all ops
904 	 *   return as inline.
905 	 * o Large read ops return data as write chunk(s), header as
906 	 *   inline.
907 	 * o Large non-read ops return as a single reply chunk.
908 	 */
909 	if (rpcrdma_results_inline(r_xprt, rqst))
910 		wtype = rpcrdma_noch;
911 	else if ((ddp_allowed && rqst->rq_rcv_buf.flags & XDRBUF_READ) &&
912 		 rpcrdma_nonpayload_inline(r_xprt, rqst))
913 		wtype = rpcrdma_writech;
914 	else
915 		wtype = rpcrdma_replych;
916 
917 	/*
918 	 * Chunks needed for arguments?
919 	 *
920 	 * o If the total request is under the inline threshold, all ops
921 	 *   are sent as inline.
922 	 * o Large write ops transmit data as read chunk(s), header as
923 	 *   inline.
924 	 * o Large non-write ops are sent with the entire message as a
925 	 *   single read chunk (protocol 0-position special case).
926 	 *
927 	 * This assumes that the upper layer does not present a request
928 	 * that both has a data payload, and whose non-data arguments
929 	 * by themselves are larger than the inline threshold.
930 	 */
931 	if (rpcrdma_args_inline(r_xprt, rqst)) {
932 		*p++ = rdma_msg;
933 		rtype = buf->len < rdmab_length(req->rl_sendbuf) ?
934 			rpcrdma_noch_pullup : rpcrdma_noch_mapped;
935 	} else if (ddp_allowed && buf->flags & XDRBUF_WRITE) {
936 		*p++ = rdma_msg;
937 		rtype = rpcrdma_readch;
938 	} else {
939 		r_xprt->rx_stats.nomsg_call_count++;
940 		*p++ = rdma_nomsg;
941 		rtype = rpcrdma_areadch;
942 	}
943 
944 	/* This implementation supports the following combinations
945 	 * of chunk lists in one RPC-over-RDMA Call message:
946 	 *
947 	 *   - Read list
948 	 *   - Write list
949 	 *   - Reply chunk
950 	 *   - Read list + Reply chunk
951 	 *
952 	 * It might not yet support the following combinations:
953 	 *
954 	 *   - Read list + Write list
955 	 *
956 	 * It does not support the following combinations:
957 	 *
958 	 *   - Write list + Reply chunk
959 	 *   - Read list + Write list + Reply chunk
960 	 *
961 	 * This implementation supports only a single chunk in each
962 	 * Read or Write list. Thus for example the client cannot
963 	 * send a Call message with a Position Zero Read chunk and a
964 	 * regular Read chunk at the same time.
965 	 */
966 	ret = rpcrdma_encode_read_list(r_xprt, req, rqst, rtype);
967 	if (ret)
968 		goto out_err;
969 	ret = rpcrdma_encode_write_list(r_xprt, req, rqst, wtype);
970 	if (ret)
971 		goto out_err;
972 	ret = rpcrdma_encode_reply_chunk(r_xprt, req, rqst, wtype);
973 	if (ret)
974 		goto out_err;
975 
976 	ret = rpcrdma_prepare_send_sges(r_xprt, req, req->rl_hdrbuf.len,
977 					buf, rtype);
978 	if (ret)
979 		goto out_err;
980 
981 	trace_xprtrdma_marshal(req, rtype, wtype);
982 	return 0;
983 
984 out_err:
985 	trace_xprtrdma_marshal_failed(rqst, ret);
986 	r_xprt->rx_stats.failed_marshal_count++;
987 	frwr_reset(req);
988 	return ret;
989 }
990 
991 static void __rpcrdma_update_cwnd_locked(struct rpc_xprt *xprt,
992 					 struct rpcrdma_buffer *buf,
993 					 u32 grant)
994 {
995 	buf->rb_credits = grant;
996 	xprt->cwnd = grant << RPC_CWNDSHIFT;
997 }
998 
999 static void rpcrdma_update_cwnd(struct rpcrdma_xprt *r_xprt, u32 grant)
1000 {
1001 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1002 
1003 	spin_lock(&xprt->transport_lock);
1004 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, grant);
1005 	spin_unlock(&xprt->transport_lock);
1006 }
1007 
1008 /**
1009  * rpcrdma_reset_cwnd - Reset the xprt's congestion window
1010  * @r_xprt: controlling transport instance
1011  *
1012  * Prepare @r_xprt for the next connection by reinitializing
1013  * its credit grant to one (see RFC 8166, Section 3.3.3).
1014  */
1015 void rpcrdma_reset_cwnd(struct rpcrdma_xprt *r_xprt)
1016 {
1017 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1018 
1019 	spin_lock(&xprt->transport_lock);
1020 	xprt->cong = 0;
1021 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, 1);
1022 	spin_unlock(&xprt->transport_lock);
1023 }
1024 
1025 /**
1026  * rpcrdma_inline_fixup - Scatter inline received data into rqst's iovecs
1027  * @rqst: controlling RPC request
1028  * @srcp: points to RPC message payload in receive buffer
1029  * @copy_len: remaining length of receive buffer content
1030  * @pad: Write chunk pad bytes needed (zero for pure inline)
1031  *
1032  * The upper layer has set the maximum number of bytes it can
1033  * receive in each component of rq_rcv_buf. These values are set in
1034  * the head.iov_len, page_len, tail.iov_len, and buflen fields.
1035  *
1036  * Unlike the TCP equivalent (xdr_partial_copy_from_skb), in
1037  * many cases this function simply updates iov_base pointers in
1038  * rq_rcv_buf to point directly to the received reply data, to
1039  * avoid copying reply data.
1040  *
1041  * Returns the count of bytes which had to be memcopied.
1042  */
1043 static unsigned long
1044 rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
1045 {
1046 	unsigned long fixup_copy_count;
1047 	int i, npages, curlen;
1048 	char *destp;
1049 	struct page **ppages;
1050 	int page_base;
1051 
1052 	/* The head iovec is redirected to the RPC reply message
1053 	 * in the receive buffer, to avoid a memcopy.
1054 	 */
1055 	rqst->rq_rcv_buf.head[0].iov_base = srcp;
1056 	rqst->rq_private_buf.head[0].iov_base = srcp;
1057 
1058 	/* The contents of the receive buffer that follow
1059 	 * head.iov_len bytes are copied into the page list.
1060 	 */
1061 	curlen = rqst->rq_rcv_buf.head[0].iov_len;
1062 	if (curlen > copy_len)
1063 		curlen = copy_len;
1064 	srcp += curlen;
1065 	copy_len -= curlen;
1066 
1067 	ppages = rqst->rq_rcv_buf.pages +
1068 		(rqst->rq_rcv_buf.page_base >> PAGE_SHIFT);
1069 	page_base = offset_in_page(rqst->rq_rcv_buf.page_base);
1070 	fixup_copy_count = 0;
1071 	if (copy_len && rqst->rq_rcv_buf.page_len) {
1072 		int pagelist_len;
1073 
1074 		pagelist_len = rqst->rq_rcv_buf.page_len;
1075 		if (pagelist_len > copy_len)
1076 			pagelist_len = copy_len;
1077 		npages = PAGE_ALIGN(page_base + pagelist_len) >> PAGE_SHIFT;
1078 		for (i = 0; i < npages; i++) {
1079 			curlen = PAGE_SIZE - page_base;
1080 			if (curlen > pagelist_len)
1081 				curlen = pagelist_len;
1082 
1083 			destp = kmap_atomic(ppages[i]);
1084 			memcpy(destp + page_base, srcp, curlen);
1085 			flush_dcache_page(ppages[i]);
1086 			kunmap_atomic(destp);
1087 			srcp += curlen;
1088 			copy_len -= curlen;
1089 			fixup_copy_count += curlen;
1090 			pagelist_len -= curlen;
1091 			if (!pagelist_len)
1092 				break;
1093 			page_base = 0;
1094 		}
1095 
1096 		/* Implicit padding for the last segment in a Write
1097 		 * chunk is inserted inline at the front of the tail
1098 		 * iovec. The upper layer ignores the content of
1099 		 * the pad. Simply ensure inline content in the tail
1100 		 * that follows the Write chunk is properly aligned.
1101 		 */
1102 		if (pad)
1103 			srcp -= pad;
1104 	}
1105 
1106 	/* The tail iovec is redirected to the remaining data
1107 	 * in the receive buffer, to avoid a memcopy.
1108 	 */
1109 	if (copy_len || pad) {
1110 		rqst->rq_rcv_buf.tail[0].iov_base = srcp;
1111 		rqst->rq_private_buf.tail[0].iov_base = srcp;
1112 	}
1113 
1114 	if (fixup_copy_count)
1115 		trace_xprtrdma_fixup(rqst, fixup_copy_count);
1116 	return fixup_copy_count;
1117 }
1118 
1119 /* By convention, backchannel calls arrive via rdma_msg type
1120  * messages, and never populate the chunk lists. This makes
1121  * the RPC/RDMA header small and fixed in size, so it is
1122  * straightforward to check the RPC header's direction field.
1123  */
1124 static bool
1125 rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1126 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1127 {
1128 	struct xdr_stream *xdr = &rep->rr_stream;
1129 	__be32 *p;
1130 
1131 	if (rep->rr_proc != rdma_msg)
1132 		return false;
1133 
1134 	/* Peek at stream contents without advancing. */
1135 	p = xdr_inline_decode(xdr, 0);
1136 
1137 	/* Chunk lists */
1138 	if (xdr_item_is_present(p++))
1139 		return false;
1140 	if (xdr_item_is_present(p++))
1141 		return false;
1142 	if (xdr_item_is_present(p++))
1143 		return false;
1144 
1145 	/* RPC header */
1146 	if (*p++ != rep->rr_xid)
1147 		return false;
1148 	if (*p != cpu_to_be32(RPC_CALL))
1149 		return false;
1150 
1151 	/* Now that we are sure this is a backchannel call,
1152 	 * advance to the RPC header.
1153 	 */
1154 	p = xdr_inline_decode(xdr, 3 * sizeof(*p));
1155 	if (unlikely(!p))
1156 		return true;
1157 
1158 	rpcrdma_bc_receive_call(r_xprt, rep);
1159 	return true;
1160 }
1161 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
1162 {
1163 	return false;
1164 }
1165 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
1166 
1167 static int decode_rdma_segment(struct xdr_stream *xdr, u32 *length)
1168 {
1169 	u32 handle;
1170 	u64 offset;
1171 	__be32 *p;
1172 
1173 	p = xdr_inline_decode(xdr, 4 * sizeof(*p));
1174 	if (unlikely(!p))
1175 		return -EIO;
1176 
1177 	xdr_decode_rdma_segment(p, &handle, length, &offset);
1178 	trace_xprtrdma_decode_seg(handle, *length, offset);
1179 	return 0;
1180 }
1181 
1182 static int decode_write_chunk(struct xdr_stream *xdr, u32 *length)
1183 {
1184 	u32 segcount, seglength;
1185 	__be32 *p;
1186 
1187 	p = xdr_inline_decode(xdr, sizeof(*p));
1188 	if (unlikely(!p))
1189 		return -EIO;
1190 
1191 	*length = 0;
1192 	segcount = be32_to_cpup(p);
1193 	while (segcount--) {
1194 		if (decode_rdma_segment(xdr, &seglength))
1195 			return -EIO;
1196 		*length += seglength;
1197 	}
1198 
1199 	return 0;
1200 }
1201 
1202 /* In RPC-over-RDMA Version One replies, a Read list is never
1203  * expected. This decoder is a stub that returns an error if
1204  * a Read list is present.
1205  */
1206 static int decode_read_list(struct xdr_stream *xdr)
1207 {
1208 	__be32 *p;
1209 
1210 	p = xdr_inline_decode(xdr, sizeof(*p));
1211 	if (unlikely(!p))
1212 		return -EIO;
1213 	if (unlikely(xdr_item_is_present(p)))
1214 		return -EIO;
1215 	return 0;
1216 }
1217 
1218 /* Supports only one Write chunk in the Write list
1219  */
1220 static int decode_write_list(struct xdr_stream *xdr, u32 *length)
1221 {
1222 	u32 chunklen;
1223 	bool first;
1224 	__be32 *p;
1225 
1226 	*length = 0;
1227 	first = true;
1228 	do {
1229 		p = xdr_inline_decode(xdr, sizeof(*p));
1230 		if (unlikely(!p))
1231 			return -EIO;
1232 		if (xdr_item_is_absent(p))
1233 			break;
1234 		if (!first)
1235 			return -EIO;
1236 
1237 		if (decode_write_chunk(xdr, &chunklen))
1238 			return -EIO;
1239 		*length += chunklen;
1240 		first = false;
1241 	} while (true);
1242 	return 0;
1243 }
1244 
1245 static int decode_reply_chunk(struct xdr_stream *xdr, u32 *length)
1246 {
1247 	__be32 *p;
1248 
1249 	p = xdr_inline_decode(xdr, sizeof(*p));
1250 	if (unlikely(!p))
1251 		return -EIO;
1252 
1253 	*length = 0;
1254 	if (xdr_item_is_present(p))
1255 		if (decode_write_chunk(xdr, length))
1256 			return -EIO;
1257 	return 0;
1258 }
1259 
1260 static int
1261 rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1262 		   struct rpc_rqst *rqst)
1263 {
1264 	struct xdr_stream *xdr = &rep->rr_stream;
1265 	u32 writelist, replychunk, rpclen;
1266 	char *base;
1267 
1268 	/* Decode the chunk lists */
1269 	if (decode_read_list(xdr))
1270 		return -EIO;
1271 	if (decode_write_list(xdr, &writelist))
1272 		return -EIO;
1273 	if (decode_reply_chunk(xdr, &replychunk))
1274 		return -EIO;
1275 
1276 	/* RDMA_MSG sanity checks */
1277 	if (unlikely(replychunk))
1278 		return -EIO;
1279 
1280 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
1281 	base = (char *)xdr_inline_decode(xdr, 0);
1282 	rpclen = xdr_stream_remaining(xdr);
1283 	r_xprt->rx_stats.fixup_copy_count +=
1284 		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
1285 
1286 	r_xprt->rx_stats.total_rdma_reply += writelist;
1287 	return rpclen + xdr_align_size(writelist);
1288 }
1289 
1290 static noinline int
1291 rpcrdma_decode_nomsg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1292 {
1293 	struct xdr_stream *xdr = &rep->rr_stream;
1294 	u32 writelist, replychunk;
1295 
1296 	/* Decode the chunk lists */
1297 	if (decode_read_list(xdr))
1298 		return -EIO;
1299 	if (decode_write_list(xdr, &writelist))
1300 		return -EIO;
1301 	if (decode_reply_chunk(xdr, &replychunk))
1302 		return -EIO;
1303 
1304 	/* RDMA_NOMSG sanity checks */
1305 	if (unlikely(writelist))
1306 		return -EIO;
1307 	if (unlikely(!replychunk))
1308 		return -EIO;
1309 
1310 	/* Reply chunk buffer already is the reply vector */
1311 	r_xprt->rx_stats.total_rdma_reply += replychunk;
1312 	return replychunk;
1313 }
1314 
1315 static noinline int
1316 rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1317 		     struct rpc_rqst *rqst)
1318 {
1319 	struct xdr_stream *xdr = &rep->rr_stream;
1320 	__be32 *p;
1321 
1322 	p = xdr_inline_decode(xdr, sizeof(*p));
1323 	if (unlikely(!p))
1324 		return -EIO;
1325 
1326 	switch (*p) {
1327 	case err_vers:
1328 		p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1329 		if (!p)
1330 			break;
1331 		trace_xprtrdma_err_vers(rqst, p, p + 1);
1332 		break;
1333 	case err_chunk:
1334 		trace_xprtrdma_err_chunk(rqst);
1335 		break;
1336 	default:
1337 		trace_xprtrdma_err_unrecognized(rqst, p);
1338 	}
1339 
1340 	return -EIO;
1341 }
1342 
1343 /**
1344  * rpcrdma_unpin_rqst - Release rqst without completing it
1345  * @rep: RPC/RDMA Receive context
1346  *
1347  * This is done when a connection is lost so that a Reply
1348  * can be dropped and its matching Call can be subsequently
1349  * retransmitted on a new connection.
1350  */
1351 void rpcrdma_unpin_rqst(struct rpcrdma_rep *rep)
1352 {
1353 	struct rpc_xprt *xprt = &rep->rr_rxprt->rx_xprt;
1354 	struct rpc_rqst *rqst = rep->rr_rqst;
1355 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
1356 
1357 	req->rl_reply = NULL;
1358 	rep->rr_rqst = NULL;
1359 
1360 	spin_lock(&xprt->queue_lock);
1361 	xprt_unpin_rqst(rqst);
1362 	spin_unlock(&xprt->queue_lock);
1363 }
1364 
1365 /**
1366  * rpcrdma_complete_rqst - Pass completed rqst back to RPC
1367  * @rep: RPC/RDMA Receive context
1368  *
1369  * Reconstruct the RPC reply and complete the transaction
1370  * while @rqst is still pinned to ensure the rep, rqst, and
1371  * rq_task pointers remain stable.
1372  */
1373 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
1374 {
1375 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1376 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1377 	struct rpc_rqst *rqst = rep->rr_rqst;
1378 	int status;
1379 
1380 	switch (rep->rr_proc) {
1381 	case rdma_msg:
1382 		status = rpcrdma_decode_msg(r_xprt, rep, rqst);
1383 		break;
1384 	case rdma_nomsg:
1385 		status = rpcrdma_decode_nomsg(r_xprt, rep);
1386 		break;
1387 	case rdma_error:
1388 		status = rpcrdma_decode_error(r_xprt, rep, rqst);
1389 		break;
1390 	default:
1391 		status = -EIO;
1392 	}
1393 	if (status < 0)
1394 		goto out_badheader;
1395 
1396 out:
1397 	spin_lock(&xprt->queue_lock);
1398 	xprt_complete_rqst(rqst->rq_task, status);
1399 	xprt_unpin_rqst(rqst);
1400 	spin_unlock(&xprt->queue_lock);
1401 	return;
1402 
1403 out_badheader:
1404 	trace_xprtrdma_reply_hdr_err(rep);
1405 	r_xprt->rx_stats.bad_reply_count++;
1406 	rqst->rq_task->tk_status = status;
1407 	status = 0;
1408 	goto out;
1409 }
1410 
1411 static void rpcrdma_reply_done(struct kref *kref)
1412 {
1413 	struct rpcrdma_req *req =
1414 		container_of(kref, struct rpcrdma_req, rl_kref);
1415 
1416 	rpcrdma_complete_rqst(req->rl_reply);
1417 }
1418 
1419 /**
1420  * rpcrdma_reply_handler - Process received RPC/RDMA messages
1421  * @rep: Incoming rpcrdma_rep object to process
1422  *
1423  * Errors must result in the RPC task either being awakened, or
1424  * allowed to timeout, to discover the errors at that time.
1425  */
1426 void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
1427 {
1428 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1429 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1430 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1431 	struct rpcrdma_req *req;
1432 	struct rpc_rqst *rqst;
1433 	u32 credits;
1434 	__be32 *p;
1435 
1436 	/* Any data means we had a useful conversation, so
1437 	 * then we don't need to delay the next reconnect.
1438 	 */
1439 	if (xprt->reestablish_timeout)
1440 		xprt->reestablish_timeout = 0;
1441 
1442 	/* Fixed transport header fields */
1443 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
1444 			rep->rr_hdrbuf.head[0].iov_base, NULL);
1445 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
1446 	if (unlikely(!p))
1447 		goto out_shortreply;
1448 	rep->rr_xid = *p++;
1449 	rep->rr_vers = *p++;
1450 	credits = be32_to_cpu(*p++);
1451 	rep->rr_proc = *p++;
1452 
1453 	if (rep->rr_vers != rpcrdma_version)
1454 		goto out_badversion;
1455 
1456 	if (rpcrdma_is_bcall(r_xprt, rep))
1457 		return;
1458 
1459 	/* Match incoming rpcrdma_rep to an rpcrdma_req to
1460 	 * get context for handling any incoming chunks.
1461 	 */
1462 	spin_lock(&xprt->queue_lock);
1463 	rqst = xprt_lookup_rqst(xprt, rep->rr_xid);
1464 	if (!rqst)
1465 		goto out_norqst;
1466 	xprt_pin_rqst(rqst);
1467 	spin_unlock(&xprt->queue_lock);
1468 
1469 	if (credits == 0)
1470 		credits = 1;	/* don't deadlock */
1471 	else if (credits > r_xprt->rx_ep->re_max_requests)
1472 		credits = r_xprt->rx_ep->re_max_requests;
1473 	rpcrdma_post_recvs(r_xprt, credits + (buf->rb_bc_srv_max_requests << 1),
1474 			   false);
1475 	if (buf->rb_credits != credits)
1476 		rpcrdma_update_cwnd(r_xprt, credits);
1477 
1478 	req = rpcr_to_rdmar(rqst);
1479 	if (unlikely(req->rl_reply))
1480 		rpcrdma_rep_put(buf, req->rl_reply);
1481 	req->rl_reply = rep;
1482 	rep->rr_rqst = rqst;
1483 
1484 	trace_xprtrdma_reply(rqst->rq_task, rep, credits);
1485 
1486 	if (rep->rr_wc_flags & IB_WC_WITH_INVALIDATE)
1487 		frwr_reminv(rep, &req->rl_registered);
1488 	if (!list_empty(&req->rl_registered))
1489 		frwr_unmap_async(r_xprt, req);
1490 		/* LocalInv completion will complete the RPC */
1491 	else
1492 		kref_put(&req->rl_kref, rpcrdma_reply_done);
1493 	return;
1494 
1495 out_badversion:
1496 	trace_xprtrdma_reply_vers_err(rep);
1497 	goto out;
1498 
1499 out_norqst:
1500 	spin_unlock(&xprt->queue_lock);
1501 	trace_xprtrdma_reply_rqst_err(rep);
1502 	goto out;
1503 
1504 out_shortreply:
1505 	trace_xprtrdma_reply_short_err(rep);
1506 
1507 out:
1508 	rpcrdma_rep_put(buf, rep);
1509 }
1510