xref: /openbmc/linux/net/sunrpc/xprtrdma/rpc_rdma.c (revision 29c37341)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (c) 2014-2017 Oracle.  All rights reserved.
4  * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the BSD-type
10  * license below:
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  *
16  *      Redistributions of source code must retain the above copyright
17  *      notice, this list of conditions and the following disclaimer.
18  *
19  *      Redistributions in binary form must reproduce the above
20  *      copyright notice, this list of conditions and the following
21  *      disclaimer in the documentation and/or other materials provided
22  *      with the distribution.
23  *
24  *      Neither the name of the Network Appliance, Inc. nor the names of
25  *      its contributors may be used to endorse or promote products
26  *      derived from this software without specific prior written
27  *      permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*
43  * rpc_rdma.c
44  *
45  * This file contains the guts of the RPC RDMA protocol, and
46  * does marshaling/unmarshaling, etc. It is also where interfacing
47  * to the Linux RPC framework lives.
48  */
49 
50 #include <linux/highmem.h>
51 
52 #include <linux/sunrpc/svc_rdma.h>
53 
54 #include "xprt_rdma.h"
55 #include <trace/events/rpcrdma.h>
56 
57 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
58 # define RPCDBG_FACILITY	RPCDBG_TRANS
59 #endif
60 
61 /* Returns size of largest RPC-over-RDMA header in a Call message
62  *
63  * The largest Call header contains a full-size Read list and a
64  * minimal Reply chunk.
65  */
66 static unsigned int rpcrdma_max_call_header_size(unsigned int maxsegs)
67 {
68 	unsigned int size;
69 
70 	/* Fixed header fields and list discriminators */
71 	size = RPCRDMA_HDRLEN_MIN;
72 
73 	/* Maximum Read list size */
74 	size += maxsegs * rpcrdma_readchunk_maxsz * sizeof(__be32);
75 
76 	/* Minimal Read chunk size */
77 	size += sizeof(__be32);	/* segment count */
78 	size += rpcrdma_segment_maxsz * sizeof(__be32);
79 	size += sizeof(__be32);	/* list discriminator */
80 
81 	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 /* Split @vec on page boundaries into SGEs. FMR registers pages, not
183  * a byte range. Other modes coalesce these SGEs into a single MR
184  * when they can.
185  *
186  * Returns pointer to next available SGE, and bumps the total number
187  * of SGEs consumed.
188  */
189 static struct rpcrdma_mr_seg *
190 rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg,
191 		     unsigned int *n)
192 {
193 	u32 remaining, page_offset;
194 	char *base;
195 
196 	base = vec->iov_base;
197 	page_offset = offset_in_page(base);
198 	remaining = vec->iov_len;
199 	while (remaining) {
200 		seg->mr_page = NULL;
201 		seg->mr_offset = base;
202 		seg->mr_len = min_t(u32, PAGE_SIZE - page_offset, remaining);
203 		remaining -= seg->mr_len;
204 		base += seg->mr_len;
205 		++seg;
206 		++(*n);
207 		page_offset = 0;
208 	}
209 	return seg;
210 }
211 
212 /* Convert @xdrbuf into SGEs no larger than a page each. As they
213  * are registered, these SGEs are then coalesced into RDMA segments
214  * when the selected memreg mode supports it.
215  *
216  * Returns positive number of SGEs consumed, or a negative errno.
217  */
218 
219 static int
220 rpcrdma_convert_iovs(struct rpcrdma_xprt *r_xprt, struct xdr_buf *xdrbuf,
221 		     unsigned int pos, enum rpcrdma_chunktype type,
222 		     struct rpcrdma_mr_seg *seg)
223 {
224 	unsigned long page_base;
225 	unsigned int len, n;
226 	struct page **ppages;
227 
228 	n = 0;
229 	if (pos == 0)
230 		seg = rpcrdma_convert_kvec(&xdrbuf->head[0], seg, &n);
231 
232 	len = xdrbuf->page_len;
233 	ppages = xdrbuf->pages + (xdrbuf->page_base >> PAGE_SHIFT);
234 	page_base = offset_in_page(xdrbuf->page_base);
235 	while (len) {
236 		/* ACL likes to be lazy in allocating pages - ACLs
237 		 * are small by default but can get huge.
238 		 */
239 		if (unlikely(xdrbuf->flags & XDRBUF_SPARSE_PAGES)) {
240 			if (!*ppages)
241 				*ppages = alloc_page(GFP_NOWAIT | __GFP_NOWARN);
242 			if (!*ppages)
243 				return -ENOBUFS;
244 		}
245 		seg->mr_page = *ppages;
246 		seg->mr_offset = (char *)page_base;
247 		seg->mr_len = min_t(u32, PAGE_SIZE - page_base, len);
248 		len -= seg->mr_len;
249 		++ppages;
250 		++seg;
251 		++n;
252 		page_base = 0;
253 	}
254 
255 	/* When encoding a Read chunk, the tail iovec contains an
256 	 * XDR pad and may be omitted.
257 	 */
258 	if (type == rpcrdma_readch && r_xprt->rx_ep->re_implicit_roundup)
259 		goto out;
260 
261 	/* When encoding a Write chunk, some servers need to see an
262 	 * extra segment for non-XDR-aligned Write chunks. The upper
263 	 * layer provides space in the tail iovec that may be used
264 	 * for this purpose.
265 	 */
266 	if (type == rpcrdma_writech && r_xprt->rx_ep->re_implicit_roundup)
267 		goto out;
268 
269 	if (xdrbuf->tail[0].iov_len)
270 		seg = rpcrdma_convert_kvec(&xdrbuf->tail[0], seg, &n);
271 
272 out:
273 	if (unlikely(n > RPCRDMA_MAX_SEGS))
274 		return -EIO;
275 	return n;
276 }
277 
278 static int
279 encode_rdma_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr)
280 {
281 	__be32 *p;
282 
283 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
284 	if (unlikely(!p))
285 		return -EMSGSIZE;
286 
287 	xdr_encode_rdma_segment(p, mr->mr_handle, mr->mr_length, mr->mr_offset);
288 	return 0;
289 }
290 
291 static int
292 encode_read_segment(struct xdr_stream *xdr, struct rpcrdma_mr *mr,
293 		    u32 position)
294 {
295 	__be32 *p;
296 
297 	p = xdr_reserve_space(xdr, 6 * sizeof(*p));
298 	if (unlikely(!p))
299 		return -EMSGSIZE;
300 
301 	*p++ = xdr_one;			/* Item present */
302 	xdr_encode_read_segment(p, position, mr->mr_handle, mr->mr_length,
303 				mr->mr_offset);
304 	return 0;
305 }
306 
307 static struct rpcrdma_mr_seg *rpcrdma_mr_prepare(struct rpcrdma_xprt *r_xprt,
308 						 struct rpcrdma_req *req,
309 						 struct rpcrdma_mr_seg *seg,
310 						 int nsegs, bool writing,
311 						 struct rpcrdma_mr **mr)
312 {
313 	*mr = rpcrdma_mr_pop(&req->rl_free_mrs);
314 	if (!*mr) {
315 		*mr = rpcrdma_mr_get(r_xprt);
316 		if (!*mr)
317 			goto out_getmr_err;
318 		trace_xprtrdma_mr_get(req);
319 		(*mr)->mr_req = req;
320 	}
321 
322 	rpcrdma_mr_push(*mr, &req->rl_registered);
323 	return frwr_map(r_xprt, seg, nsegs, writing, req->rl_slot.rq_xid, *mr);
324 
325 out_getmr_err:
326 	trace_xprtrdma_nomrs(req);
327 	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
328 	rpcrdma_mrs_refresh(r_xprt);
329 	return ERR_PTR(-EAGAIN);
330 }
331 
332 /* Register and XDR encode the Read list. Supports encoding a list of read
333  * segments that belong to a single read chunk.
334  *
335  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
336  *
337  *  Read chunklist (a linked list):
338  *   N elements, position P (same P for all chunks of same arg!):
339  *    1 - PHLOO - 1 - PHLOO - ... - 1 - PHLOO - 0
340  *
341  * Returns zero on success, or a negative errno if a failure occurred.
342  * @xdr is advanced to the next position in the stream.
343  *
344  * Only a single @pos value is currently supported.
345  */
346 static int rpcrdma_encode_read_list(struct rpcrdma_xprt *r_xprt,
347 				    struct rpcrdma_req *req,
348 				    struct rpc_rqst *rqst,
349 				    enum rpcrdma_chunktype rtype)
350 {
351 	struct xdr_stream *xdr = &req->rl_stream;
352 	struct rpcrdma_mr_seg *seg;
353 	struct rpcrdma_mr *mr;
354 	unsigned int pos;
355 	int nsegs;
356 
357 	if (rtype == rpcrdma_noch_pullup || rtype == rpcrdma_noch_mapped)
358 		goto done;
359 
360 	pos = rqst->rq_snd_buf.head[0].iov_len;
361 	if (rtype == rpcrdma_areadch)
362 		pos = 0;
363 	seg = req->rl_segments;
364 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_snd_buf, pos,
365 				     rtype, seg);
366 	if (nsegs < 0)
367 		return nsegs;
368 
369 	do {
370 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, false, &mr);
371 		if (IS_ERR(seg))
372 			return PTR_ERR(seg);
373 
374 		if (encode_read_segment(xdr, mr, pos) < 0)
375 			return -EMSGSIZE;
376 
377 		trace_xprtrdma_chunk_read(rqst->rq_task, pos, mr, nsegs);
378 		r_xprt->rx_stats.read_chunk_count++;
379 		nsegs -= mr->mr_nents;
380 	} while (nsegs);
381 
382 done:
383 	if (xdr_stream_encode_item_absent(xdr) < 0)
384 		return -EMSGSIZE;
385 	return 0;
386 }
387 
388 /* Register and XDR encode the Write list. Supports encoding a list
389  * containing one array of plain segments that belong to a single
390  * write chunk.
391  *
392  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
393  *
394  *  Write chunklist (a list of (one) counted array):
395  *   N elements:
396  *    1 - N - HLOO - HLOO - ... - HLOO - 0
397  *
398  * Returns zero on success, or a negative errno if a failure occurred.
399  * @xdr is advanced to the next position in the stream.
400  *
401  * Only a single Write chunk is currently supported.
402  */
403 static int rpcrdma_encode_write_list(struct rpcrdma_xprt *r_xprt,
404 				     struct rpcrdma_req *req,
405 				     struct rpc_rqst *rqst,
406 				     enum rpcrdma_chunktype wtype)
407 {
408 	struct xdr_stream *xdr = &req->rl_stream;
409 	struct rpcrdma_mr_seg *seg;
410 	struct rpcrdma_mr *mr;
411 	int nsegs, nchunks;
412 	__be32 *segcount;
413 
414 	if (wtype != rpcrdma_writech)
415 		goto done;
416 
417 	seg = req->rl_segments;
418 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf,
419 				     rqst->rq_rcv_buf.head[0].iov_len,
420 				     wtype, seg);
421 	if (nsegs < 0)
422 		return nsegs;
423 
424 	if (xdr_stream_encode_item_present(xdr) < 0)
425 		return -EMSGSIZE;
426 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
427 	if (unlikely(!segcount))
428 		return -EMSGSIZE;
429 	/* Actual value encoded below */
430 
431 	nchunks = 0;
432 	do {
433 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
434 		if (IS_ERR(seg))
435 			return PTR_ERR(seg);
436 
437 		if (encode_rdma_segment(xdr, mr) < 0)
438 			return -EMSGSIZE;
439 
440 		trace_xprtrdma_chunk_write(rqst->rq_task, mr, nsegs);
441 		r_xprt->rx_stats.write_chunk_count++;
442 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
443 		nchunks++;
444 		nsegs -= mr->mr_nents;
445 	} while (nsegs);
446 
447 	/* Update count of segments in this Write chunk */
448 	*segcount = cpu_to_be32(nchunks);
449 
450 done:
451 	if (xdr_stream_encode_item_absent(xdr) < 0)
452 		return -EMSGSIZE;
453 	return 0;
454 }
455 
456 /* Register and XDR encode the Reply chunk. Supports encoding an array
457  * of plain segments that belong to a single write (reply) chunk.
458  *
459  * Encoding key for single-list chunks (HLOO = Handle32 Length32 Offset64):
460  *
461  *  Reply chunk (a counted array):
462  *   N elements:
463  *    1 - N - HLOO - HLOO - ... - HLOO
464  *
465  * Returns zero on success, or a negative errno if a failure occurred.
466  * @xdr is advanced to the next position in the stream.
467  */
468 static int rpcrdma_encode_reply_chunk(struct rpcrdma_xprt *r_xprt,
469 				      struct rpcrdma_req *req,
470 				      struct rpc_rqst *rqst,
471 				      enum rpcrdma_chunktype wtype)
472 {
473 	struct xdr_stream *xdr = &req->rl_stream;
474 	struct rpcrdma_mr_seg *seg;
475 	struct rpcrdma_mr *mr;
476 	int nsegs, nchunks;
477 	__be32 *segcount;
478 
479 	if (wtype != rpcrdma_replych) {
480 		if (xdr_stream_encode_item_absent(xdr) < 0)
481 			return -EMSGSIZE;
482 		return 0;
483 	}
484 
485 	seg = req->rl_segments;
486 	nsegs = rpcrdma_convert_iovs(r_xprt, &rqst->rq_rcv_buf, 0, wtype, seg);
487 	if (nsegs < 0)
488 		return nsegs;
489 
490 	if (xdr_stream_encode_item_present(xdr) < 0)
491 		return -EMSGSIZE;
492 	segcount = xdr_reserve_space(xdr, sizeof(*segcount));
493 	if (unlikely(!segcount))
494 		return -EMSGSIZE;
495 	/* Actual value encoded below */
496 
497 	nchunks = 0;
498 	do {
499 		seg = rpcrdma_mr_prepare(r_xprt, req, seg, nsegs, true, &mr);
500 		if (IS_ERR(seg))
501 			return PTR_ERR(seg);
502 
503 		if (encode_rdma_segment(xdr, mr) < 0)
504 			return -EMSGSIZE;
505 
506 		trace_xprtrdma_chunk_reply(rqst->rq_task, mr, nsegs);
507 		r_xprt->rx_stats.reply_chunk_count++;
508 		r_xprt->rx_stats.total_rdma_request += mr->mr_length;
509 		nchunks++;
510 		nsegs -= mr->mr_nents;
511 	} while (nsegs);
512 
513 	/* Update count of segments in the Reply chunk */
514 	*segcount = cpu_to_be32(nchunks);
515 
516 	return 0;
517 }
518 
519 static void rpcrdma_sendctx_done(struct kref *kref)
520 {
521 	struct rpcrdma_req *req =
522 		container_of(kref, struct rpcrdma_req, rl_kref);
523 	struct rpcrdma_rep *rep = req->rl_reply;
524 
525 	rpcrdma_complete_rqst(rep);
526 	rep->rr_rxprt->rx_stats.reply_waits_for_send++;
527 }
528 
529 /**
530  * rpcrdma_sendctx_unmap - DMA-unmap Send buffer
531  * @sc: sendctx containing SGEs to unmap
532  *
533  */
534 void rpcrdma_sendctx_unmap(struct rpcrdma_sendctx *sc)
535 {
536 	struct rpcrdma_regbuf *rb = sc->sc_req->rl_sendbuf;
537 	struct ib_sge *sge;
538 
539 	if (!sc->sc_unmap_count)
540 		return;
541 
542 	/* The first two SGEs contain the transport header and
543 	 * the inline buffer. These are always left mapped so
544 	 * they can be cheaply re-used.
545 	 */
546 	for (sge = &sc->sc_sges[2]; sc->sc_unmap_count;
547 	     ++sge, --sc->sc_unmap_count)
548 		ib_dma_unmap_page(rdmab_device(rb), sge->addr, sge->length,
549 				  DMA_TO_DEVICE);
550 
551 	kref_put(&sc->sc_req->rl_kref, rpcrdma_sendctx_done);
552 }
553 
554 /* Prepare an SGE for the RPC-over-RDMA transport header.
555  */
556 static void rpcrdma_prepare_hdr_sge(struct rpcrdma_xprt *r_xprt,
557 				    struct rpcrdma_req *req, u32 len)
558 {
559 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
560 	struct rpcrdma_regbuf *rb = req->rl_rdmabuf;
561 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
562 
563 	sge->addr = rdmab_addr(rb);
564 	sge->length = len;
565 	sge->lkey = rdmab_lkey(rb);
566 
567 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
568 				      DMA_TO_DEVICE);
569 }
570 
571 /* The head iovec is straightforward, as it is usually already
572  * DMA-mapped. Sync the content that has changed.
573  */
574 static bool rpcrdma_prepare_head_iov(struct rpcrdma_xprt *r_xprt,
575 				     struct rpcrdma_req *req, unsigned int len)
576 {
577 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
578 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
579 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
580 
581 	if (!rpcrdma_regbuf_dma_map(r_xprt, rb))
582 		return false;
583 
584 	sge->addr = rdmab_addr(rb);
585 	sge->length = len;
586 	sge->lkey = rdmab_lkey(rb);
587 
588 	ib_dma_sync_single_for_device(rdmab_device(rb), sge->addr, sge->length,
589 				      DMA_TO_DEVICE);
590 	return true;
591 }
592 
593 /* If there is a page list present, DMA map and prepare an
594  * SGE for each page to be sent.
595  */
596 static bool rpcrdma_prepare_pagelist(struct rpcrdma_req *req,
597 				     struct xdr_buf *xdr)
598 {
599 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
600 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
601 	unsigned int page_base, len, remaining;
602 	struct page **ppages;
603 	struct ib_sge *sge;
604 
605 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
606 	page_base = offset_in_page(xdr->page_base);
607 	remaining = xdr->page_len;
608 	while (remaining) {
609 		sge = &sc->sc_sges[req->rl_wr.num_sge++];
610 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
611 		sge->addr = ib_dma_map_page(rdmab_device(rb), *ppages,
612 					    page_base, len, DMA_TO_DEVICE);
613 		if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
614 			goto out_mapping_err;
615 
616 		sge->length = len;
617 		sge->lkey = rdmab_lkey(rb);
618 
619 		sc->sc_unmap_count++;
620 		ppages++;
621 		remaining -= len;
622 		page_base = 0;
623 	}
624 
625 	return true;
626 
627 out_mapping_err:
628 	trace_xprtrdma_dma_maperr(sge->addr);
629 	return false;
630 }
631 
632 /* The tail iovec may include an XDR pad for the page list,
633  * as well as additional content, and may not reside in the
634  * same page as the head iovec.
635  */
636 static bool rpcrdma_prepare_tail_iov(struct rpcrdma_req *req,
637 				     struct xdr_buf *xdr,
638 				     unsigned int page_base, unsigned int len)
639 {
640 	struct rpcrdma_sendctx *sc = req->rl_sendctx;
641 	struct ib_sge *sge = &sc->sc_sges[req->rl_wr.num_sge++];
642 	struct rpcrdma_regbuf *rb = req->rl_sendbuf;
643 	struct page *page = virt_to_page(xdr->tail[0].iov_base);
644 
645 	sge->addr = ib_dma_map_page(rdmab_device(rb), page, page_base, len,
646 				    DMA_TO_DEVICE);
647 	if (ib_dma_mapping_error(rdmab_device(rb), sge->addr))
648 		goto out_mapping_err;
649 
650 	sge->length = len;
651 	sge->lkey = rdmab_lkey(rb);
652 	++sc->sc_unmap_count;
653 	return true;
654 
655 out_mapping_err:
656 	trace_xprtrdma_dma_maperr(sge->addr);
657 	return false;
658 }
659 
660 /* Copy the tail to the end of the head buffer.
661  */
662 static void rpcrdma_pullup_tail_iov(struct rpcrdma_xprt *r_xprt,
663 				    struct rpcrdma_req *req,
664 				    struct xdr_buf *xdr)
665 {
666 	unsigned char *dst;
667 
668 	dst = (unsigned char *)xdr->head[0].iov_base;
669 	dst += xdr->head[0].iov_len + xdr->page_len;
670 	memmove(dst, xdr->tail[0].iov_base, xdr->tail[0].iov_len);
671 	r_xprt->rx_stats.pullup_copy_count += xdr->tail[0].iov_len;
672 }
673 
674 /* Copy pagelist content into the head buffer.
675  */
676 static void rpcrdma_pullup_pagelist(struct rpcrdma_xprt *r_xprt,
677 				    struct rpcrdma_req *req,
678 				    struct xdr_buf *xdr)
679 {
680 	unsigned int len, page_base, remaining;
681 	struct page **ppages;
682 	unsigned char *src, *dst;
683 
684 	dst = (unsigned char *)xdr->head[0].iov_base;
685 	dst += xdr->head[0].iov_len;
686 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
687 	page_base = offset_in_page(xdr->page_base);
688 	remaining = xdr->page_len;
689 	while (remaining) {
690 		src = page_address(*ppages);
691 		src += page_base;
692 		len = min_t(unsigned int, PAGE_SIZE - page_base, remaining);
693 		memcpy(dst, src, len);
694 		r_xprt->rx_stats.pullup_copy_count += len;
695 
696 		ppages++;
697 		dst += len;
698 		remaining -= len;
699 		page_base = 0;
700 	}
701 }
702 
703 /* Copy the contents of @xdr into @rl_sendbuf and DMA sync it.
704  * When the head, pagelist, and tail are small, a pull-up copy
705  * is considerably less costly than DMA mapping the components
706  * of @xdr.
707  *
708  * Assumptions:
709  *  - the caller has already verified that the total length
710  *    of the RPC Call body will fit into @rl_sendbuf.
711  */
712 static bool rpcrdma_prepare_noch_pullup(struct rpcrdma_xprt *r_xprt,
713 					struct rpcrdma_req *req,
714 					struct xdr_buf *xdr)
715 {
716 	if (unlikely(xdr->tail[0].iov_len))
717 		rpcrdma_pullup_tail_iov(r_xprt, req, xdr);
718 
719 	if (unlikely(xdr->page_len))
720 		rpcrdma_pullup_pagelist(r_xprt, req, xdr);
721 
722 	/* The whole RPC message resides in the head iovec now */
723 	return rpcrdma_prepare_head_iov(r_xprt, req, xdr->len);
724 }
725 
726 static bool rpcrdma_prepare_noch_mapped(struct rpcrdma_xprt *r_xprt,
727 					struct rpcrdma_req *req,
728 					struct xdr_buf *xdr)
729 {
730 	struct kvec *tail = &xdr->tail[0];
731 
732 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
733 		return false;
734 	if (xdr->page_len)
735 		if (!rpcrdma_prepare_pagelist(req, xdr))
736 			return false;
737 	if (tail->iov_len)
738 		if (!rpcrdma_prepare_tail_iov(req, xdr,
739 					      offset_in_page(tail->iov_base),
740 					      tail->iov_len))
741 			return false;
742 
743 	if (req->rl_sendctx->sc_unmap_count)
744 		kref_get(&req->rl_kref);
745 	return true;
746 }
747 
748 static bool rpcrdma_prepare_readch(struct rpcrdma_xprt *r_xprt,
749 				   struct rpcrdma_req *req,
750 				   struct xdr_buf *xdr)
751 {
752 	if (!rpcrdma_prepare_head_iov(r_xprt, req, xdr->head[0].iov_len))
753 		return false;
754 
755 	/* If there is a Read chunk, the page list is being handled
756 	 * via explicit RDMA, and thus is skipped here.
757 	 */
758 
759 	/* Do not include the tail if it is only an XDR pad */
760 	if (xdr->tail[0].iov_len > 3) {
761 		unsigned int page_base, len;
762 
763 		/* If the content in the page list is an odd length,
764 		 * xdr_write_pages() adds a pad at the beginning of
765 		 * the tail iovec. Force the tail's non-pad content to
766 		 * land at the next XDR position in the Send message.
767 		 */
768 		page_base = offset_in_page(xdr->tail[0].iov_base);
769 		len = xdr->tail[0].iov_len;
770 		page_base += len & 3;
771 		len -= len & 3;
772 		if (!rpcrdma_prepare_tail_iov(req, xdr, page_base, len))
773 			return false;
774 		kref_get(&req->rl_kref);
775 	}
776 
777 	return true;
778 }
779 
780 /**
781  * rpcrdma_prepare_send_sges - Construct SGEs for a Send WR
782  * @r_xprt: controlling transport
783  * @req: context of RPC Call being marshalled
784  * @hdrlen: size of transport header, in bytes
785  * @xdr: xdr_buf containing RPC Call
786  * @rtype: chunk type being encoded
787  *
788  * Returns 0 on success; otherwise a negative errno is returned.
789  */
790 inline int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
791 				     struct rpcrdma_req *req, u32 hdrlen,
792 				     struct xdr_buf *xdr,
793 				     enum rpcrdma_chunktype rtype)
794 {
795 	int ret;
796 
797 	ret = -EAGAIN;
798 	req->rl_sendctx = rpcrdma_sendctx_get_locked(r_xprt);
799 	if (!req->rl_sendctx)
800 		goto out_nosc;
801 	req->rl_sendctx->sc_unmap_count = 0;
802 	req->rl_sendctx->sc_req = req;
803 	kref_init(&req->rl_kref);
804 	req->rl_wr.wr_cqe = &req->rl_sendctx->sc_cqe;
805 	req->rl_wr.sg_list = req->rl_sendctx->sc_sges;
806 	req->rl_wr.num_sge = 0;
807 	req->rl_wr.opcode = IB_WR_SEND;
808 
809 	rpcrdma_prepare_hdr_sge(r_xprt, req, hdrlen);
810 
811 	ret = -EIO;
812 	switch (rtype) {
813 	case rpcrdma_noch_pullup:
814 		if (!rpcrdma_prepare_noch_pullup(r_xprt, req, xdr))
815 			goto out_unmap;
816 		break;
817 	case rpcrdma_noch_mapped:
818 		if (!rpcrdma_prepare_noch_mapped(r_xprt, req, xdr))
819 			goto out_unmap;
820 		break;
821 	case rpcrdma_readch:
822 		if (!rpcrdma_prepare_readch(r_xprt, req, xdr))
823 			goto out_unmap;
824 		break;
825 	case rpcrdma_areadch:
826 		break;
827 	default:
828 		goto out_unmap;
829 	}
830 
831 	return 0;
832 
833 out_unmap:
834 	rpcrdma_sendctx_unmap(req->rl_sendctx);
835 out_nosc:
836 	trace_xprtrdma_prepsend_failed(&req->rl_slot, ret);
837 	return ret;
838 }
839 
840 /**
841  * rpcrdma_marshal_req - Marshal and send one RPC request
842  * @r_xprt: controlling transport
843  * @rqst: RPC request to be marshaled
844  *
845  * For the RPC in "rqst", this function:
846  *  - Chooses the transfer mode (eg., RDMA_MSG or RDMA_NOMSG)
847  *  - Registers Read, Write, and Reply chunks
848  *  - Constructs the transport header
849  *  - Posts a Send WR to send the transport header and request
850  *
851  * Returns:
852  *	%0 if the RPC was sent successfully,
853  *	%-ENOTCONN if the connection was lost,
854  *	%-EAGAIN if the caller should call again with the same arguments,
855  *	%-ENOBUFS if the caller should call again after a delay,
856  *	%-EMSGSIZE if the transport header is too small,
857  *	%-EIO if a permanent problem occurred while marshaling.
858  */
859 int
860 rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
861 {
862 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
863 	struct xdr_stream *xdr = &req->rl_stream;
864 	enum rpcrdma_chunktype rtype, wtype;
865 	struct xdr_buf *buf = &rqst->rq_snd_buf;
866 	bool ddp_allowed;
867 	__be32 *p;
868 	int ret;
869 
870 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
871 	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
872 			rqst);
873 
874 	/* Fixed header fields */
875 	ret = -EMSGSIZE;
876 	p = xdr_reserve_space(xdr, 4 * sizeof(*p));
877 	if (!p)
878 		goto out_err;
879 	*p++ = rqst->rq_xid;
880 	*p++ = rpcrdma_version;
881 	*p++ = r_xprt->rx_buf.rb_max_requests;
882 
883 	/* When the ULP employs a GSS flavor that guarantees integrity
884 	 * or privacy, direct data placement of individual data items
885 	 * is not allowed.
886 	 */
887 	ddp_allowed = !test_bit(RPCAUTH_AUTH_DATATOUCH,
888 				&rqst->rq_cred->cr_auth->au_flags);
889 
890 	/*
891 	 * Chunks needed for results?
892 	 *
893 	 * o If the expected result is under the inline threshold, all ops
894 	 *   return as inline.
895 	 * o Large read ops return data as write chunk(s), header as
896 	 *   inline.
897 	 * o Large non-read ops return as a single reply chunk.
898 	 */
899 	if (rpcrdma_results_inline(r_xprt, rqst))
900 		wtype = rpcrdma_noch;
901 	else if ((ddp_allowed && rqst->rq_rcv_buf.flags & XDRBUF_READ) &&
902 		 rpcrdma_nonpayload_inline(r_xprt, rqst))
903 		wtype = rpcrdma_writech;
904 	else
905 		wtype = rpcrdma_replych;
906 
907 	/*
908 	 * Chunks needed for arguments?
909 	 *
910 	 * o If the total request is under the inline threshold, all ops
911 	 *   are sent as inline.
912 	 * o Large write ops transmit data as read chunk(s), header as
913 	 *   inline.
914 	 * o Large non-write ops are sent with the entire message as a
915 	 *   single read chunk (protocol 0-position special case).
916 	 *
917 	 * This assumes that the upper layer does not present a request
918 	 * that both has a data payload, and whose non-data arguments
919 	 * by themselves are larger than the inline threshold.
920 	 */
921 	if (rpcrdma_args_inline(r_xprt, rqst)) {
922 		*p++ = rdma_msg;
923 		rtype = buf->len < rdmab_length(req->rl_sendbuf) ?
924 			rpcrdma_noch_pullup : rpcrdma_noch_mapped;
925 	} else if (ddp_allowed && buf->flags & XDRBUF_WRITE) {
926 		*p++ = rdma_msg;
927 		rtype = rpcrdma_readch;
928 	} else {
929 		r_xprt->rx_stats.nomsg_call_count++;
930 		*p++ = rdma_nomsg;
931 		rtype = rpcrdma_areadch;
932 	}
933 
934 	/* This implementation supports the following combinations
935 	 * of chunk lists in one RPC-over-RDMA Call message:
936 	 *
937 	 *   - Read list
938 	 *   - Write list
939 	 *   - Reply chunk
940 	 *   - Read list + Reply chunk
941 	 *
942 	 * It might not yet support the following combinations:
943 	 *
944 	 *   - Read list + Write list
945 	 *
946 	 * It does not support the following combinations:
947 	 *
948 	 *   - Write list + Reply chunk
949 	 *   - Read list + Write list + Reply chunk
950 	 *
951 	 * This implementation supports only a single chunk in each
952 	 * Read or Write list. Thus for example the client cannot
953 	 * send a Call message with a Position Zero Read chunk and a
954 	 * regular Read chunk at the same time.
955 	 */
956 	ret = rpcrdma_encode_read_list(r_xprt, req, rqst, rtype);
957 	if (ret)
958 		goto out_err;
959 	ret = rpcrdma_encode_write_list(r_xprt, req, rqst, wtype);
960 	if (ret)
961 		goto out_err;
962 	ret = rpcrdma_encode_reply_chunk(r_xprt, req, rqst, wtype);
963 	if (ret)
964 		goto out_err;
965 
966 	ret = rpcrdma_prepare_send_sges(r_xprt, req, req->rl_hdrbuf.len,
967 					buf, rtype);
968 	if (ret)
969 		goto out_err;
970 
971 	trace_xprtrdma_marshal(req, rtype, wtype);
972 	return 0;
973 
974 out_err:
975 	trace_xprtrdma_marshal_failed(rqst, ret);
976 	r_xprt->rx_stats.failed_marshal_count++;
977 	frwr_reset(req);
978 	return ret;
979 }
980 
981 static void __rpcrdma_update_cwnd_locked(struct rpc_xprt *xprt,
982 					 struct rpcrdma_buffer *buf,
983 					 u32 grant)
984 {
985 	buf->rb_credits = grant;
986 	xprt->cwnd = grant << RPC_CWNDSHIFT;
987 }
988 
989 static void rpcrdma_update_cwnd(struct rpcrdma_xprt *r_xprt, u32 grant)
990 {
991 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
992 
993 	spin_lock(&xprt->transport_lock);
994 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, grant);
995 	spin_unlock(&xprt->transport_lock);
996 }
997 
998 /**
999  * rpcrdma_reset_cwnd - Reset the xprt's congestion window
1000  * @r_xprt: controlling transport instance
1001  *
1002  * Prepare @r_xprt for the next connection by reinitializing
1003  * its credit grant to one (see RFC 8166, Section 3.3.3).
1004  */
1005 void rpcrdma_reset_cwnd(struct rpcrdma_xprt *r_xprt)
1006 {
1007 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1008 
1009 	spin_lock(&xprt->transport_lock);
1010 	xprt->cong = 0;
1011 	__rpcrdma_update_cwnd_locked(xprt, &r_xprt->rx_buf, 1);
1012 	spin_unlock(&xprt->transport_lock);
1013 }
1014 
1015 /**
1016  * rpcrdma_inline_fixup - Scatter inline received data into rqst's iovecs
1017  * @rqst: controlling RPC request
1018  * @srcp: points to RPC message payload in receive buffer
1019  * @copy_len: remaining length of receive buffer content
1020  * @pad: Write chunk pad bytes needed (zero for pure inline)
1021  *
1022  * The upper layer has set the maximum number of bytes it can
1023  * receive in each component of rq_rcv_buf. These values are set in
1024  * the head.iov_len, page_len, tail.iov_len, and buflen fields.
1025  *
1026  * Unlike the TCP equivalent (xdr_partial_copy_from_skb), in
1027  * many cases this function simply updates iov_base pointers in
1028  * rq_rcv_buf to point directly to the received reply data, to
1029  * avoid copying reply data.
1030  *
1031  * Returns the count of bytes which had to be memcopied.
1032  */
1033 static unsigned long
1034 rpcrdma_inline_fixup(struct rpc_rqst *rqst, char *srcp, int copy_len, int pad)
1035 {
1036 	unsigned long fixup_copy_count;
1037 	int i, npages, curlen;
1038 	char *destp;
1039 	struct page **ppages;
1040 	int page_base;
1041 
1042 	/* The head iovec is redirected to the RPC reply message
1043 	 * in the receive buffer, to avoid a memcopy.
1044 	 */
1045 	rqst->rq_rcv_buf.head[0].iov_base = srcp;
1046 	rqst->rq_private_buf.head[0].iov_base = srcp;
1047 
1048 	/* The contents of the receive buffer that follow
1049 	 * head.iov_len bytes are copied into the page list.
1050 	 */
1051 	curlen = rqst->rq_rcv_buf.head[0].iov_len;
1052 	if (curlen > copy_len)
1053 		curlen = copy_len;
1054 	srcp += curlen;
1055 	copy_len -= curlen;
1056 
1057 	ppages = rqst->rq_rcv_buf.pages +
1058 		(rqst->rq_rcv_buf.page_base >> PAGE_SHIFT);
1059 	page_base = offset_in_page(rqst->rq_rcv_buf.page_base);
1060 	fixup_copy_count = 0;
1061 	if (copy_len && rqst->rq_rcv_buf.page_len) {
1062 		int pagelist_len;
1063 
1064 		pagelist_len = rqst->rq_rcv_buf.page_len;
1065 		if (pagelist_len > copy_len)
1066 			pagelist_len = copy_len;
1067 		npages = PAGE_ALIGN(page_base + pagelist_len) >> PAGE_SHIFT;
1068 		for (i = 0; i < npages; i++) {
1069 			curlen = PAGE_SIZE - page_base;
1070 			if (curlen > pagelist_len)
1071 				curlen = pagelist_len;
1072 
1073 			destp = kmap_atomic(ppages[i]);
1074 			memcpy(destp + page_base, srcp, curlen);
1075 			flush_dcache_page(ppages[i]);
1076 			kunmap_atomic(destp);
1077 			srcp += curlen;
1078 			copy_len -= curlen;
1079 			fixup_copy_count += curlen;
1080 			pagelist_len -= curlen;
1081 			if (!pagelist_len)
1082 				break;
1083 			page_base = 0;
1084 		}
1085 
1086 		/* Implicit padding for the last segment in a Write
1087 		 * chunk is inserted inline at the front of the tail
1088 		 * iovec. The upper layer ignores the content of
1089 		 * the pad. Simply ensure inline content in the tail
1090 		 * that follows the Write chunk is properly aligned.
1091 		 */
1092 		if (pad)
1093 			srcp -= pad;
1094 	}
1095 
1096 	/* The tail iovec is redirected to the remaining data
1097 	 * in the receive buffer, to avoid a memcopy.
1098 	 */
1099 	if (copy_len || pad) {
1100 		rqst->rq_rcv_buf.tail[0].iov_base = srcp;
1101 		rqst->rq_private_buf.tail[0].iov_base = srcp;
1102 	}
1103 
1104 	if (fixup_copy_count)
1105 		trace_xprtrdma_fixup(rqst, fixup_copy_count);
1106 	return fixup_copy_count;
1107 }
1108 
1109 /* By convention, backchannel calls arrive via rdma_msg type
1110  * messages, and never populate the chunk lists. This makes
1111  * the RPC/RDMA header small and fixed in size, so it is
1112  * straightforward to check the RPC header's direction field.
1113  */
1114 static bool
1115 rpcrdma_is_bcall(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1116 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1117 {
1118 	struct xdr_stream *xdr = &rep->rr_stream;
1119 	__be32 *p;
1120 
1121 	if (rep->rr_proc != rdma_msg)
1122 		return false;
1123 
1124 	/* Peek at stream contents without advancing. */
1125 	p = xdr_inline_decode(xdr, 0);
1126 
1127 	/* Chunk lists */
1128 	if (xdr_item_is_present(p++))
1129 		return false;
1130 	if (xdr_item_is_present(p++))
1131 		return false;
1132 	if (xdr_item_is_present(p++))
1133 		return false;
1134 
1135 	/* RPC header */
1136 	if (*p++ != rep->rr_xid)
1137 		return false;
1138 	if (*p != cpu_to_be32(RPC_CALL))
1139 		return false;
1140 
1141 	/* Now that we are sure this is a backchannel call,
1142 	 * advance to the RPC header.
1143 	 */
1144 	p = xdr_inline_decode(xdr, 3 * sizeof(*p));
1145 	if (unlikely(!p))
1146 		goto out_short;
1147 
1148 	rpcrdma_bc_receive_call(r_xprt, rep);
1149 	return true;
1150 
1151 out_short:
1152 	pr_warn("RPC/RDMA short backward direction call\n");
1153 	return true;
1154 }
1155 #else	/* CONFIG_SUNRPC_BACKCHANNEL */
1156 {
1157 	return false;
1158 }
1159 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
1160 
1161 static int decode_rdma_segment(struct xdr_stream *xdr, u32 *length)
1162 {
1163 	u32 handle;
1164 	u64 offset;
1165 	__be32 *p;
1166 
1167 	p = xdr_inline_decode(xdr, 4 * sizeof(*p));
1168 	if (unlikely(!p))
1169 		return -EIO;
1170 
1171 	xdr_decode_rdma_segment(p, &handle, length, &offset);
1172 	trace_xprtrdma_decode_seg(handle, *length, offset);
1173 	return 0;
1174 }
1175 
1176 static int decode_write_chunk(struct xdr_stream *xdr, u32 *length)
1177 {
1178 	u32 segcount, seglength;
1179 	__be32 *p;
1180 
1181 	p = xdr_inline_decode(xdr, sizeof(*p));
1182 	if (unlikely(!p))
1183 		return -EIO;
1184 
1185 	*length = 0;
1186 	segcount = be32_to_cpup(p);
1187 	while (segcount--) {
1188 		if (decode_rdma_segment(xdr, &seglength))
1189 			return -EIO;
1190 		*length += seglength;
1191 	}
1192 
1193 	return 0;
1194 }
1195 
1196 /* In RPC-over-RDMA Version One replies, a Read list is never
1197  * expected. This decoder is a stub that returns an error if
1198  * a Read list is present.
1199  */
1200 static int decode_read_list(struct xdr_stream *xdr)
1201 {
1202 	__be32 *p;
1203 
1204 	p = xdr_inline_decode(xdr, sizeof(*p));
1205 	if (unlikely(!p))
1206 		return -EIO;
1207 	if (unlikely(xdr_item_is_present(p)))
1208 		return -EIO;
1209 	return 0;
1210 }
1211 
1212 /* Supports only one Write chunk in the Write list
1213  */
1214 static int decode_write_list(struct xdr_stream *xdr, u32 *length)
1215 {
1216 	u32 chunklen;
1217 	bool first;
1218 	__be32 *p;
1219 
1220 	*length = 0;
1221 	first = true;
1222 	do {
1223 		p = xdr_inline_decode(xdr, sizeof(*p));
1224 		if (unlikely(!p))
1225 			return -EIO;
1226 		if (xdr_item_is_absent(p))
1227 			break;
1228 		if (!first)
1229 			return -EIO;
1230 
1231 		if (decode_write_chunk(xdr, &chunklen))
1232 			return -EIO;
1233 		*length += chunklen;
1234 		first = false;
1235 	} while (true);
1236 	return 0;
1237 }
1238 
1239 static int decode_reply_chunk(struct xdr_stream *xdr, u32 *length)
1240 {
1241 	__be32 *p;
1242 
1243 	p = xdr_inline_decode(xdr, sizeof(*p));
1244 	if (unlikely(!p))
1245 		return -EIO;
1246 
1247 	*length = 0;
1248 	if (xdr_item_is_present(p))
1249 		if (decode_write_chunk(xdr, length))
1250 			return -EIO;
1251 	return 0;
1252 }
1253 
1254 static int
1255 rpcrdma_decode_msg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1256 		   struct rpc_rqst *rqst)
1257 {
1258 	struct xdr_stream *xdr = &rep->rr_stream;
1259 	u32 writelist, replychunk, rpclen;
1260 	char *base;
1261 
1262 	/* Decode the chunk lists */
1263 	if (decode_read_list(xdr))
1264 		return -EIO;
1265 	if (decode_write_list(xdr, &writelist))
1266 		return -EIO;
1267 	if (decode_reply_chunk(xdr, &replychunk))
1268 		return -EIO;
1269 
1270 	/* RDMA_MSG sanity checks */
1271 	if (unlikely(replychunk))
1272 		return -EIO;
1273 
1274 	/* Build the RPC reply's Payload stream in rqst->rq_rcv_buf */
1275 	base = (char *)xdr_inline_decode(xdr, 0);
1276 	rpclen = xdr_stream_remaining(xdr);
1277 	r_xprt->rx_stats.fixup_copy_count +=
1278 		rpcrdma_inline_fixup(rqst, base, rpclen, writelist & 3);
1279 
1280 	r_xprt->rx_stats.total_rdma_reply += writelist;
1281 	return rpclen + xdr_align_size(writelist);
1282 }
1283 
1284 static noinline int
1285 rpcrdma_decode_nomsg(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep)
1286 {
1287 	struct xdr_stream *xdr = &rep->rr_stream;
1288 	u32 writelist, replychunk;
1289 
1290 	/* Decode the chunk lists */
1291 	if (decode_read_list(xdr))
1292 		return -EIO;
1293 	if (decode_write_list(xdr, &writelist))
1294 		return -EIO;
1295 	if (decode_reply_chunk(xdr, &replychunk))
1296 		return -EIO;
1297 
1298 	/* RDMA_NOMSG sanity checks */
1299 	if (unlikely(writelist))
1300 		return -EIO;
1301 	if (unlikely(!replychunk))
1302 		return -EIO;
1303 
1304 	/* Reply chunk buffer already is the reply vector */
1305 	r_xprt->rx_stats.total_rdma_reply += replychunk;
1306 	return replychunk;
1307 }
1308 
1309 static noinline int
1310 rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep,
1311 		     struct rpc_rqst *rqst)
1312 {
1313 	struct xdr_stream *xdr = &rep->rr_stream;
1314 	__be32 *p;
1315 
1316 	p = xdr_inline_decode(xdr, sizeof(*p));
1317 	if (unlikely(!p))
1318 		return -EIO;
1319 
1320 	switch (*p) {
1321 	case err_vers:
1322 		p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1323 		if (!p)
1324 			break;
1325 		dprintk("RPC:       %s: server reports "
1326 			"version error (%u-%u), xid %08x\n", __func__,
1327 			be32_to_cpup(p), be32_to_cpu(*(p + 1)),
1328 			be32_to_cpu(rep->rr_xid));
1329 		break;
1330 	case err_chunk:
1331 		dprintk("RPC:       %s: server reports "
1332 			"header decoding error, xid %08x\n", __func__,
1333 			be32_to_cpu(rep->rr_xid));
1334 		break;
1335 	default:
1336 		dprintk("RPC:       %s: server reports "
1337 			"unrecognized error %d, xid %08x\n", __func__,
1338 			be32_to_cpup(p), be32_to_cpu(rep->rr_xid));
1339 	}
1340 
1341 	return -EIO;
1342 }
1343 
1344 /* Perform XID lookup, reconstruction of the RPC reply, and
1345  * RPC completion while holding the transport lock to ensure
1346  * the rep, rqst, and rq_task pointers remain stable.
1347  */
1348 void rpcrdma_complete_rqst(struct rpcrdma_rep *rep)
1349 {
1350 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1351 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1352 	struct rpc_rqst *rqst = rep->rr_rqst;
1353 	int status;
1354 
1355 	switch (rep->rr_proc) {
1356 	case rdma_msg:
1357 		status = rpcrdma_decode_msg(r_xprt, rep, rqst);
1358 		break;
1359 	case rdma_nomsg:
1360 		status = rpcrdma_decode_nomsg(r_xprt, rep);
1361 		break;
1362 	case rdma_error:
1363 		status = rpcrdma_decode_error(r_xprt, rep, rqst);
1364 		break;
1365 	default:
1366 		status = -EIO;
1367 	}
1368 	if (status < 0)
1369 		goto out_badheader;
1370 
1371 out:
1372 	spin_lock(&xprt->queue_lock);
1373 	xprt_complete_rqst(rqst->rq_task, status);
1374 	xprt_unpin_rqst(rqst);
1375 	spin_unlock(&xprt->queue_lock);
1376 	return;
1377 
1378 out_badheader:
1379 	trace_xprtrdma_reply_hdr(rep);
1380 	r_xprt->rx_stats.bad_reply_count++;
1381 	rqst->rq_task->tk_status = status;
1382 	status = 0;
1383 	goto out;
1384 }
1385 
1386 static void rpcrdma_reply_done(struct kref *kref)
1387 {
1388 	struct rpcrdma_req *req =
1389 		container_of(kref, struct rpcrdma_req, rl_kref);
1390 
1391 	rpcrdma_complete_rqst(req->rl_reply);
1392 }
1393 
1394 /**
1395  * rpcrdma_reply_handler - Process received RPC/RDMA messages
1396  * @rep: Incoming rpcrdma_rep object to process
1397  *
1398  * Errors must result in the RPC task either being awakened, or
1399  * allowed to timeout, to discover the errors at that time.
1400  */
1401 void rpcrdma_reply_handler(struct rpcrdma_rep *rep)
1402 {
1403 	struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
1404 	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
1405 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1406 	struct rpcrdma_req *req;
1407 	struct rpc_rqst *rqst;
1408 	u32 credits;
1409 	__be32 *p;
1410 
1411 	/* Any data means we had a useful conversation, so
1412 	 * then we don't need to delay the next reconnect.
1413 	 */
1414 	if (xprt->reestablish_timeout)
1415 		xprt->reestablish_timeout = 0;
1416 
1417 	/* Fixed transport header fields */
1418 	xdr_init_decode(&rep->rr_stream, &rep->rr_hdrbuf,
1419 			rep->rr_hdrbuf.head[0].iov_base, NULL);
1420 	p = xdr_inline_decode(&rep->rr_stream, 4 * sizeof(*p));
1421 	if (unlikely(!p))
1422 		goto out_shortreply;
1423 	rep->rr_xid = *p++;
1424 	rep->rr_vers = *p++;
1425 	credits = be32_to_cpu(*p++);
1426 	rep->rr_proc = *p++;
1427 
1428 	if (rep->rr_vers != rpcrdma_version)
1429 		goto out_badversion;
1430 
1431 	if (rpcrdma_is_bcall(r_xprt, rep))
1432 		return;
1433 
1434 	/* Match incoming rpcrdma_rep to an rpcrdma_req to
1435 	 * get context for handling any incoming chunks.
1436 	 */
1437 	spin_lock(&xprt->queue_lock);
1438 	rqst = xprt_lookup_rqst(xprt, rep->rr_xid);
1439 	if (!rqst)
1440 		goto out_norqst;
1441 	xprt_pin_rqst(rqst);
1442 	spin_unlock(&xprt->queue_lock);
1443 
1444 	if (credits == 0)
1445 		credits = 1;	/* don't deadlock */
1446 	else if (credits > r_xprt->rx_ep->re_max_requests)
1447 		credits = r_xprt->rx_ep->re_max_requests;
1448 	if (buf->rb_credits != credits)
1449 		rpcrdma_update_cwnd(r_xprt, credits);
1450 	rpcrdma_post_recvs(r_xprt, false);
1451 
1452 	req = rpcr_to_rdmar(rqst);
1453 	if (req->rl_reply) {
1454 		trace_xprtrdma_leaked_rep(rqst, req->rl_reply);
1455 		rpcrdma_recv_buffer_put(req->rl_reply);
1456 	}
1457 	req->rl_reply = rep;
1458 	rep->rr_rqst = rqst;
1459 
1460 	trace_xprtrdma_reply(rqst->rq_task, rep, req, credits);
1461 
1462 	if (rep->rr_wc_flags & IB_WC_WITH_INVALIDATE)
1463 		frwr_reminv(rep, &req->rl_registered);
1464 	if (!list_empty(&req->rl_registered))
1465 		frwr_unmap_async(r_xprt, req);
1466 		/* LocalInv completion will complete the RPC */
1467 	else
1468 		kref_put(&req->rl_kref, rpcrdma_reply_done);
1469 	return;
1470 
1471 out_badversion:
1472 	trace_xprtrdma_reply_vers(rep);
1473 	goto out;
1474 
1475 out_norqst:
1476 	spin_unlock(&xprt->queue_lock);
1477 	trace_xprtrdma_reply_rqst(rep);
1478 	goto out;
1479 
1480 out_shortreply:
1481 	trace_xprtrdma_reply_short(rep);
1482 
1483 out:
1484 	rpcrdma_recv_buffer_put(rep);
1485 }
1486