xref: /linux/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (c) 2016-2018 Oracle. All rights reserved.
4  * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
5  * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the BSD-type
11  * license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  *      Redistributions of source code must retain the above copyright
18  *      notice, this list of conditions and the following disclaimer.
19  *
20  *      Redistributions in binary form must reproduce the above
21  *      copyright notice, this list of conditions and the following
22  *      disclaimer in the documentation and/or other materials provided
23  *      with the distribution.
24  *
25  *      Neither the name of the Network Appliance, Inc. nor the names of
26  *      its contributors may be used to endorse or promote products
27  *      derived from this software without specific prior written
28  *      permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
31  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
32  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
33  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
34  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
35  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
36  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
37  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
38  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
39  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
40  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41  *
42  * Author: Tom Tucker <tom@opengridcomputing.com>
43  */
44 
45 /* Operation
46  *
47  * The main entry point is svc_rdma_recvfrom. This is called from
48  * svc_recv when the transport indicates there is incoming data to
49  * be read. "Data Ready" is signaled when an RDMA Receive completes,
50  * or when a set of RDMA Reads complete.
51  *
52  * An svc_rqst is passed in. This structure contains an array of
53  * free pages (rq_pages) that will contain the incoming RPC message.
54  *
55  * Short messages are moved directly into svc_rqst::rq_arg, and
56  * the RPC Call is ready to be processed by the Upper Layer.
57  * svc_rdma_recvfrom returns the length of the RPC Call message,
58  * completing the reception of the RPC Call.
59  *
60  * However, when an incoming message has Read chunks,
61  * svc_rdma_recvfrom must post RDMA Reads to pull the RPC Call's
62  * data payload from the client. svc_rdma_recvfrom sets up the
63  * RDMA Reads using pages in svc_rqst::rq_pages, which are
64  * transferred to an svc_rdma_recv_ctxt for the duration of the
65  * I/O. svc_rdma_recvfrom then returns zero, since the RPC message
66  * is still not yet ready.
67  *
68  * When the Read chunk payloads have become available on the
69  * server, "Data Ready" is raised again, and svc_recv calls
70  * svc_rdma_recvfrom again. This second call may use a different
71  * svc_rqst than the first one, thus any information that needs
72  * to be preserved across these two calls is kept in an
73  * svc_rdma_recv_ctxt.
74  *
75  * The second call to svc_rdma_recvfrom performs final assembly
76  * of the RPC Call message, using the RDMA Read sink pages kept in
77  * the svc_rdma_recv_ctxt. The xdr_buf is copied from the
78  * svc_rdma_recv_ctxt to the second svc_rqst. The second call returns
79  * the length of the completed RPC Call message.
80  *
81  * Page Management
82  *
83  * Pages under I/O must be transferred from the first svc_rqst to an
84  * svc_rdma_recv_ctxt before the first svc_rdma_recvfrom call returns.
85  *
86  * The first svc_rqst supplies pages for RDMA Reads. These are moved
87  * from rqstp::rq_pages into ctxt::pages. The consumed elements of
88  * the rq_pages array are set to NULL and refilled with the first
89  * svc_rdma_recvfrom call returns.
90  *
91  * During the second svc_rdma_recvfrom call, RDMA Read sink pages
92  * are transferred from the svc_rdma_recv_ctxt to the second svc_rqst.
93  */
94 
95 #include <linux/slab.h>
96 #include <linux/spinlock.h>
97 #include <linux/unaligned.h>
98 #include <rdma/ib_verbs.h>
99 #include <rdma/rdma_cm.h>
100 
101 #include <linux/sunrpc/xdr.h>
102 #include <linux/sunrpc/debug.h>
103 #include <linux/sunrpc/rpc_rdma.h>
104 #include <linux/sunrpc/svc_rdma.h>
105 
106 #include "xprt_rdma.h"
107 #include <trace/events/rpcrdma.h>
108 
109 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc);
110 
111 static inline struct svc_rdma_recv_ctxt *
112 svc_rdma_next_recv_ctxt(struct list_head *list)
113 {
114 	return list_first_entry_or_null(list, struct svc_rdma_recv_ctxt,
115 					rc_list);
116 }
117 
118 static struct svc_rdma_recv_ctxt *
119 svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma)
120 {
121 	int node = ibdev_to_node(rdma->sc_cm_id->device);
122 	struct svc_rdma_recv_ctxt *ctxt;
123 	dma_addr_t addr;
124 	void *buffer;
125 
126 	ctxt = kzalloc_node(sizeof(*ctxt), GFP_KERNEL, node);
127 	if (!ctxt)
128 		goto fail0;
129 	buffer = kmalloc_node(rdma->sc_max_req_size, GFP_KERNEL, node);
130 	if (!buffer)
131 		goto fail1;
132 	addr = ib_dma_map_single(rdma->sc_pd->device, buffer,
133 				 rdma->sc_max_req_size, DMA_FROM_DEVICE);
134 	if (ib_dma_mapping_error(rdma->sc_pd->device, addr))
135 		goto fail2;
136 
137 	svc_rdma_recv_cid_init(rdma, &ctxt->rc_cid);
138 	pcl_init(&ctxt->rc_call_pcl);
139 	pcl_init(&ctxt->rc_read_pcl);
140 	pcl_init(&ctxt->rc_write_pcl);
141 	pcl_init(&ctxt->rc_reply_pcl);
142 
143 	ctxt->rc_recv_wr.next = NULL;
144 	ctxt->rc_recv_wr.wr_cqe = &ctxt->rc_cqe;
145 	ctxt->rc_recv_wr.sg_list = &ctxt->rc_recv_sge;
146 	ctxt->rc_recv_wr.num_sge = 1;
147 	ctxt->rc_cqe.done = svc_rdma_wc_receive;
148 	ctxt->rc_recv_sge.addr = addr;
149 	ctxt->rc_recv_sge.length = rdma->sc_max_req_size;
150 	ctxt->rc_recv_sge.lkey = rdma->sc_pd->local_dma_lkey;
151 	ctxt->rc_recv_buf = buffer;
152 	svc_rdma_cc_init(rdma, &ctxt->rc_cc);
153 	return ctxt;
154 
155 fail2:
156 	kfree(buffer);
157 fail1:
158 	kfree(ctxt);
159 fail0:
160 	return NULL;
161 }
162 
163 static void svc_rdma_recv_ctxt_destroy(struct svcxprt_rdma *rdma,
164 				       struct svc_rdma_recv_ctxt *ctxt)
165 {
166 	ib_dma_unmap_single(rdma->sc_pd->device, ctxt->rc_recv_sge.addr,
167 			    ctxt->rc_recv_sge.length, DMA_FROM_DEVICE);
168 	kfree(ctxt->rc_recv_buf);
169 	kfree(ctxt);
170 }
171 
172 /**
173  * svc_rdma_recv_ctxts_destroy - Release all recv_ctxt's for an xprt
174  * @rdma: svcxprt_rdma being torn down
175  *
176  */
177 void svc_rdma_recv_ctxts_destroy(struct svcxprt_rdma *rdma)
178 {
179 	struct svc_rdma_recv_ctxt *ctxt;
180 	struct llist_node *node;
181 
182 	while ((node = llist_del_first(&rdma->sc_recv_ctxts))) {
183 		ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node);
184 		svc_rdma_recv_ctxt_destroy(rdma, ctxt);
185 	}
186 }
187 
188 /**
189  * svc_rdma_recv_ctxt_get - Allocate a recv_ctxt
190  * @rdma: controlling svcxprt_rdma
191  *
192  * Returns a recv_ctxt or (rarely) NULL if none are available.
193  */
194 struct svc_rdma_recv_ctxt *svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma)
195 {
196 	struct svc_rdma_recv_ctxt *ctxt;
197 	struct llist_node *node;
198 
199 	node = llist_del_first(&rdma->sc_recv_ctxts);
200 	if (!node)
201 		return NULL;
202 
203 	ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node);
204 	ctxt->rc_page_count = 0;
205 	return ctxt;
206 }
207 
208 /**
209  * svc_rdma_recv_ctxt_put - Return recv_ctxt to free list
210  * @rdma: controlling svcxprt_rdma
211  * @ctxt: object to return to the free list
212  *
213  */
214 void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma,
215 			    struct svc_rdma_recv_ctxt *ctxt)
216 {
217 	svc_rdma_cc_release(rdma, &ctxt->rc_cc, DMA_FROM_DEVICE);
218 
219 	/* @rc_page_count is normally zero here, but error flows
220 	 * can leave pages in @rc_pages.
221 	 */
222 	release_pages(ctxt->rc_pages, ctxt->rc_page_count);
223 
224 	pcl_free(&ctxt->rc_call_pcl);
225 	pcl_free(&ctxt->rc_read_pcl);
226 	pcl_free(&ctxt->rc_write_pcl);
227 	pcl_free(&ctxt->rc_reply_pcl);
228 
229 	llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts);
230 }
231 
232 /**
233  * svc_rdma_release_ctxt - Release transport-specific per-rqst resources
234  * @xprt: the transport which owned the context
235  * @vctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
236  *
237  * Ensure that the recv_ctxt is released whether or not a Reply
238  * was sent. For example, the client could close the connection,
239  * or svc_process could drop an RPC, before the Reply is sent.
240  */
241 void svc_rdma_release_ctxt(struct svc_xprt *xprt, void *vctxt)
242 {
243 	struct svc_rdma_recv_ctxt *ctxt = vctxt;
244 	struct svcxprt_rdma *rdma =
245 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
246 
247 	if (ctxt)
248 		svc_rdma_recv_ctxt_put(rdma, ctxt);
249 }
250 
251 static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma,
252 				   unsigned int wanted)
253 {
254 	const struct ib_recv_wr *bad_wr = NULL;
255 	struct svc_rdma_recv_ctxt *ctxt;
256 	struct ib_recv_wr *recv_chain;
257 	int ret;
258 
259 	if (test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags))
260 		return false;
261 
262 	recv_chain = NULL;
263 	while (wanted--) {
264 		ctxt = svc_rdma_recv_ctxt_get(rdma);
265 		if (!ctxt)
266 			break;
267 
268 		trace_svcrdma_post_recv(&ctxt->rc_cid);
269 		ctxt->rc_recv_wr.next = recv_chain;
270 		recv_chain = &ctxt->rc_recv_wr;
271 		rdma->sc_pending_recvs++;
272 	}
273 	if (!recv_chain)
274 		return true;
275 
276 	ret = ib_post_recv(rdma->sc_qp, recv_chain, &bad_wr);
277 	if (ret)
278 		goto err_free;
279 	return true;
280 
281 err_free:
282 	trace_svcrdma_rq_post_err(rdma, ret);
283 	while (bad_wr) {
284 		ctxt = container_of(bad_wr, struct svc_rdma_recv_ctxt,
285 				    rc_recv_wr);
286 		bad_wr = bad_wr->next;
287 		svc_rdma_recv_ctxt_put(rdma, ctxt);
288 	}
289 	/* Since we're destroying the xprt, no need to reset
290 	 * sc_pending_recvs. */
291 	return false;
292 }
293 
294 /**
295  * svc_rdma_post_recvs - Post initial set of Recv WRs
296  * @rdma: fresh svcxprt_rdma
297  *
298  * Return values:
299  *   %true: Receive Queue initialization successful
300  *   %false: memory allocation or DMA error
301  */
302 bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma)
303 {
304 	unsigned int total;
305 
306 	/* For each credit, allocate enough recv_ctxts for one
307 	 * posted Receive and one RPC in process.
308 	 */
309 	total = (rdma->sc_max_requests * 2) + rdma->sc_recv_batch;
310 	while (total--) {
311 		struct svc_rdma_recv_ctxt *ctxt;
312 
313 		ctxt = svc_rdma_recv_ctxt_alloc(rdma);
314 		if (!ctxt)
315 			return false;
316 		llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts);
317 	}
318 
319 	return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests);
320 }
321 
322 /**
323  * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
324  * @cq: Completion Queue context
325  * @wc: Work Completion object
326  *
327  */
328 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
329 {
330 	struct svcxprt_rdma *rdma = cq->cq_context;
331 	struct ib_cqe *cqe = wc->wr_cqe;
332 	struct svc_rdma_recv_ctxt *ctxt;
333 
334 	rdma->sc_pending_recvs--;
335 
336 	/* WARNING: Only wc->wr_cqe and wc->status are reliable */
337 	ctxt = container_of(cqe, struct svc_rdma_recv_ctxt, rc_cqe);
338 
339 	if (wc->status != IB_WC_SUCCESS)
340 		goto flushed;
341 	trace_svcrdma_wc_recv(wc, &ctxt->rc_cid);
342 
343 	/* If receive posting fails, the connection is about to be
344 	 * lost anyway. The server will not be able to send a reply
345 	 * for this RPC, and the client will retransmit this RPC
346 	 * anyway when it reconnects.
347 	 *
348 	 * Therefore we drop the Receive, even if status was SUCCESS
349 	 * to reduce the likelihood of replayed requests once the
350 	 * client reconnects.
351 	 */
352 	if (rdma->sc_pending_recvs < rdma->sc_max_requests)
353 		if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch))
354 			goto dropped;
355 
356 	/* All wc fields are now known to be valid */
357 	ctxt->rc_byte_len = wc->byte_len;
358 
359 	spin_lock(&rdma->sc_rq_dto_lock);
360 	list_add_tail(&ctxt->rc_list, &rdma->sc_rq_dto_q);
361 	/* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */
362 	set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags);
363 	spin_unlock(&rdma->sc_rq_dto_lock);
364 	if (!test_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags))
365 		svc_xprt_enqueue(&rdma->sc_xprt);
366 	return;
367 
368 flushed:
369 	if (wc->status == IB_WC_WR_FLUSH_ERR)
370 		trace_svcrdma_wc_recv_flush(wc, &ctxt->rc_cid);
371 	else
372 		trace_svcrdma_wc_recv_err(wc, &ctxt->rc_cid);
373 dropped:
374 	svc_rdma_recv_ctxt_put(rdma, ctxt);
375 	svc_xprt_deferred_close(&rdma->sc_xprt);
376 }
377 
378 /**
379  * svc_rdma_flush_recv_queues - Drain pending Receive work
380  * @rdma: svcxprt_rdma being shut down
381  *
382  */
383 void svc_rdma_flush_recv_queues(struct svcxprt_rdma *rdma)
384 {
385 	struct svc_rdma_recv_ctxt *ctxt;
386 
387 	while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_read_complete_q))) {
388 		list_del(&ctxt->rc_list);
389 		svc_rdma_recv_ctxt_put(rdma, ctxt);
390 	}
391 	while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_rq_dto_q))) {
392 		list_del(&ctxt->rc_list);
393 		svc_rdma_recv_ctxt_put(rdma, ctxt);
394 	}
395 }
396 
397 static void svc_rdma_build_arg_xdr(struct svc_rqst *rqstp,
398 				   struct svc_rdma_recv_ctxt *ctxt)
399 {
400 	struct xdr_buf *arg = &rqstp->rq_arg;
401 
402 	arg->head[0].iov_base = ctxt->rc_recv_buf;
403 	arg->head[0].iov_len = ctxt->rc_byte_len;
404 	arg->tail[0].iov_base = NULL;
405 	arg->tail[0].iov_len = 0;
406 	arg->page_len = 0;
407 	arg->page_base = 0;
408 	arg->buflen = ctxt->rc_byte_len;
409 	arg->len = ctxt->rc_byte_len;
410 }
411 
412 /**
413  * xdr_count_read_segments - Count number of Read segments in Read list
414  * @rctxt: Ingress receive context
415  * @p: Start of an un-decoded Read list
416  *
417  * Before allocating anything, ensure the ingress Read list is safe
418  * to use.
419  *
420  * The segment count is limited to how many segments can fit in the
421  * transport header without overflowing the buffer. That's about 40
422  * Read segments for a 1KB inline threshold.
423  *
424  * Return values:
425  *   %true: Read list is valid. @rctxt's xdr_stream is updated to point
426  *	    to the first byte past the Read list. rc_read_pcl and
427  *	    rc_call_pcl cl_count fields are set to the number of
428  *	    Read segments in the list.
429  *  %false: Read list is corrupt. @rctxt's xdr_stream is left in an
430  *	    unknown state.
431  */
432 static bool xdr_count_read_segments(struct svc_rdma_recv_ctxt *rctxt, __be32 *p)
433 {
434 	rctxt->rc_call_pcl.cl_count = 0;
435 	rctxt->rc_read_pcl.cl_count = 0;
436 	while (xdr_item_is_present(p)) {
437 		u32 position, handle, length;
438 		u64 offset;
439 
440 		p = xdr_inline_decode(&rctxt->rc_stream,
441 				      rpcrdma_readseg_maxsz * sizeof(*p));
442 		if (!p)
443 			return false;
444 
445 		xdr_decode_read_segment(p, &position, &handle,
446 					    &length, &offset);
447 		if (position) {
448 			if (position & 3)
449 				return false;
450 			++rctxt->rc_read_pcl.cl_count;
451 		} else {
452 			++rctxt->rc_call_pcl.cl_count;
453 		}
454 
455 		p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p));
456 		if (!p)
457 			return false;
458 	}
459 	return true;
460 }
461 
462 /* Sanity check the Read list.
463  *
464  * Sanity checks:
465  * - Read list does not overflow Receive buffer.
466  * - Chunk size limited by largest NFS data payload.
467  *
468  * Return values:
469  *   %true: Read list is valid. @rctxt's xdr_stream is updated
470  *	    to point to the first byte past the Read list.
471  *  %false: Read list is corrupt. @rctxt's xdr_stream is left
472  *	    in an unknown state.
473  */
474 static bool xdr_check_read_list(struct svc_rdma_recv_ctxt *rctxt)
475 {
476 	__be32 *p;
477 
478 	p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p));
479 	if (!p)
480 		return false;
481 	if (!xdr_count_read_segments(rctxt, p))
482 		return false;
483 	if (!pcl_alloc_call(rctxt, p))
484 		return false;
485 	return pcl_alloc_read(rctxt, p);
486 }
487 
488 static bool xdr_check_write_chunk(struct svc_rdma_recv_ctxt *rctxt)
489 {
490 	u32 segcount;
491 	__be32 *p;
492 
493 	if (xdr_stream_decode_u32(&rctxt->rc_stream, &segcount))
494 		return false;
495 
496 	/* A bogus segcount causes this buffer overflow check to fail. */
497 	p = xdr_inline_decode(&rctxt->rc_stream,
498 			      segcount * rpcrdma_segment_maxsz * sizeof(*p));
499 	return p != NULL;
500 }
501 
502 /**
503  * xdr_count_write_chunks - Count number of Write chunks in Write list
504  * @rctxt: Received header and decoding state
505  * @p: start of an un-decoded Write list
506  *
507  * Before allocating anything, ensure the ingress Write list is
508  * safe to use.
509  *
510  * Return values:
511  *       %true: Write list is valid. @rctxt's xdr_stream is updated
512  *		to point to the first byte past the Write list, and
513  *		the number of Write chunks is in rc_write_pcl.cl_count.
514  *      %false: Write list is corrupt. @rctxt's xdr_stream is left
515  *		in an indeterminate state.
516  */
517 static bool xdr_count_write_chunks(struct svc_rdma_recv_ctxt *rctxt, __be32 *p)
518 {
519 	rctxt->rc_write_pcl.cl_count = 0;
520 	while (xdr_item_is_present(p)) {
521 		if (!xdr_check_write_chunk(rctxt))
522 			return false;
523 		++rctxt->rc_write_pcl.cl_count;
524 		p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p));
525 		if (!p)
526 			return false;
527 	}
528 	return true;
529 }
530 
531 /* Sanity check the Write list.
532  *
533  * Implementation limits:
534  * - This implementation currently supports only one Write chunk.
535  *
536  * Sanity checks:
537  * - Write list does not overflow Receive buffer.
538  * - Chunk size limited by largest NFS data payload.
539  *
540  * Return values:
541  *       %true: Write list is valid. @rctxt's xdr_stream is updated
542  *		to point to the first byte past the Write list.
543  *      %false: Write list is corrupt. @rctxt's xdr_stream is left
544  *		in an unknown state.
545  */
546 static bool xdr_check_write_list(struct svc_rdma_recv_ctxt *rctxt)
547 {
548 	__be32 *p;
549 
550 	p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p));
551 	if (!p)
552 		return false;
553 	if (!xdr_count_write_chunks(rctxt, p))
554 		return false;
555 	if (!pcl_alloc_write(rctxt, &rctxt->rc_write_pcl, p))
556 		return false;
557 
558 	rctxt->rc_cur_result_payload = pcl_first_chunk(&rctxt->rc_write_pcl);
559 	return true;
560 }
561 
562 /* Sanity check the Reply chunk.
563  *
564  * Sanity checks:
565  * - Reply chunk does not overflow Receive buffer.
566  * - Chunk size limited by largest NFS data payload.
567  *
568  * Return values:
569  *       %true: Reply chunk is valid. @rctxt's xdr_stream is updated
570  *		to point to the first byte past the Reply chunk.
571  *      %false: Reply chunk is corrupt. @rctxt's xdr_stream is left
572  *		in an unknown state.
573  */
574 static bool xdr_check_reply_chunk(struct svc_rdma_recv_ctxt *rctxt)
575 {
576 	__be32 *p;
577 
578 	p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p));
579 	if (!p)
580 		return false;
581 
582 	if (!xdr_item_is_present(p))
583 		return true;
584 	if (!xdr_check_write_chunk(rctxt))
585 		return false;
586 
587 	rctxt->rc_reply_pcl.cl_count = 1;
588 	return pcl_alloc_write(rctxt, &rctxt->rc_reply_pcl, p);
589 }
590 
591 /* RPC-over-RDMA Version One private extension: Remote Invalidation.
592  * Responder's choice: requester signals it can handle Send With
593  * Invalidate, and responder chooses one R_key to invalidate.
594  *
595  * If there is exactly one distinct R_key in the received transport
596  * header, set rc_inv_rkey to that R_key. Otherwise, set it to zero.
597  */
598 static void svc_rdma_get_inv_rkey(struct svcxprt_rdma *rdma,
599 				  struct svc_rdma_recv_ctxt *ctxt)
600 {
601 	struct svc_rdma_segment *segment;
602 	struct svc_rdma_chunk *chunk;
603 	u32 inv_rkey;
604 
605 	ctxt->rc_inv_rkey = 0;
606 
607 	if (!rdma->sc_snd_w_inv)
608 		return;
609 
610 	inv_rkey = 0;
611 	pcl_for_each_chunk(chunk, &ctxt->rc_call_pcl) {
612 		pcl_for_each_segment(segment, chunk) {
613 			if (inv_rkey == 0)
614 				inv_rkey = segment->rs_handle;
615 			else if (inv_rkey != segment->rs_handle)
616 				return;
617 		}
618 	}
619 	pcl_for_each_chunk(chunk, &ctxt->rc_read_pcl) {
620 		pcl_for_each_segment(segment, chunk) {
621 			if (inv_rkey == 0)
622 				inv_rkey = segment->rs_handle;
623 			else if (inv_rkey != segment->rs_handle)
624 				return;
625 		}
626 	}
627 	pcl_for_each_chunk(chunk, &ctxt->rc_write_pcl) {
628 		pcl_for_each_segment(segment, chunk) {
629 			if (inv_rkey == 0)
630 				inv_rkey = segment->rs_handle;
631 			else if (inv_rkey != segment->rs_handle)
632 				return;
633 		}
634 	}
635 	pcl_for_each_chunk(chunk, &ctxt->rc_reply_pcl) {
636 		pcl_for_each_segment(segment, chunk) {
637 			if (inv_rkey == 0)
638 				inv_rkey = segment->rs_handle;
639 			else if (inv_rkey != segment->rs_handle)
640 				return;
641 		}
642 	}
643 	ctxt->rc_inv_rkey = inv_rkey;
644 }
645 
646 /**
647  * svc_rdma_xdr_decode_req - Decode the transport header
648  * @rq_arg: xdr_buf containing ingress RPC/RDMA message
649  * @rctxt: state of decoding
650  *
651  * On entry, xdr->head[0].iov_base points to first byte of the
652  * RPC-over-RDMA transport header.
653  *
654  * On successful exit, head[0] points to first byte past the
655  * RPC-over-RDMA header. For RDMA_MSG, this is the RPC message.
656  *
657  * The length of the RPC-over-RDMA header is returned.
658  *
659  * Assumptions:
660  * - The transport header is entirely contained in the head iovec.
661  */
662 static int svc_rdma_xdr_decode_req(struct xdr_buf *rq_arg,
663 				   struct svc_rdma_recv_ctxt *rctxt)
664 {
665 	__be32 *p, *rdma_argp;
666 	unsigned int hdr_len;
667 
668 	rdma_argp = rq_arg->head[0].iov_base;
669 	xdr_init_decode(&rctxt->rc_stream, rq_arg, rdma_argp, NULL);
670 
671 	p = xdr_inline_decode(&rctxt->rc_stream,
672 			      rpcrdma_fixed_maxsz * sizeof(*p));
673 	if (unlikely(!p))
674 		goto out_short;
675 	p++;
676 	if (*p != rpcrdma_version)
677 		goto out_version;
678 	p += 2;
679 	rctxt->rc_msgtype = *p;
680 	switch (rctxt->rc_msgtype) {
681 	case rdma_msg:
682 		break;
683 	case rdma_nomsg:
684 		break;
685 	case rdma_done:
686 		goto out_drop;
687 	case rdma_error:
688 		goto out_drop;
689 	default:
690 		goto out_proc;
691 	}
692 
693 	if (!xdr_check_read_list(rctxt))
694 		goto out_inval;
695 	if (!xdr_check_write_list(rctxt))
696 		goto out_inval;
697 	if (!xdr_check_reply_chunk(rctxt))
698 		goto out_inval;
699 
700 	rq_arg->head[0].iov_base = rctxt->rc_stream.p;
701 	hdr_len = xdr_stream_pos(&rctxt->rc_stream);
702 	rq_arg->head[0].iov_len -= hdr_len;
703 	rq_arg->len -= hdr_len;
704 	trace_svcrdma_decode_rqst(rctxt, rdma_argp, hdr_len);
705 	return hdr_len;
706 
707 out_short:
708 	trace_svcrdma_decode_short_err(rctxt, rq_arg->len);
709 	return -EINVAL;
710 
711 out_version:
712 	trace_svcrdma_decode_badvers_err(rctxt, rdma_argp);
713 	return -EPROTONOSUPPORT;
714 
715 out_drop:
716 	trace_svcrdma_decode_drop_err(rctxt, rdma_argp);
717 	return 0;
718 
719 out_proc:
720 	trace_svcrdma_decode_badproc_err(rctxt, rdma_argp);
721 	return -EINVAL;
722 
723 out_inval:
724 	trace_svcrdma_decode_parse_err(rctxt, rdma_argp);
725 	return -EINVAL;
726 }
727 
728 static void svc_rdma_send_error(struct svcxprt_rdma *rdma,
729 				struct svc_rdma_recv_ctxt *rctxt,
730 				int status)
731 {
732 	struct svc_rdma_send_ctxt *sctxt;
733 
734 	sctxt = svc_rdma_send_ctxt_get(rdma);
735 	if (!sctxt)
736 		return;
737 	svc_rdma_send_error_msg(rdma, sctxt, rctxt, status);
738 }
739 
740 /* By convention, backchannel calls arrive via rdma_msg type
741  * messages, and never populate the chunk lists. This makes
742  * the RPC/RDMA header small and fixed in size, so it is
743  * straightforward to check the RPC header's direction field.
744  */
745 static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt,
746 						struct svc_rdma_recv_ctxt *rctxt)
747 {
748 	__be32 *p = rctxt->rc_recv_buf;
749 
750 	if (!xprt->xpt_bc_xprt)
751 		return false;
752 
753 	if (rctxt->rc_msgtype != rdma_msg)
754 		return false;
755 
756 	if (!pcl_is_empty(&rctxt->rc_call_pcl))
757 		return false;
758 	if (!pcl_is_empty(&rctxt->rc_read_pcl))
759 		return false;
760 	if (!pcl_is_empty(&rctxt->rc_write_pcl))
761 		return false;
762 	if (!pcl_is_empty(&rctxt->rc_reply_pcl))
763 		return false;
764 
765 	/* RPC call direction */
766 	if (*(p + 8) == cpu_to_be32(RPC_CALL))
767 		return false;
768 
769 	return true;
770 }
771 
772 /* Finish constructing the RPC Call message in rqstp::rq_arg.
773  *
774  * The incoming RPC/RDMA message is an RDMA_MSG type message
775  * with a single Read chunk (only the upper layer data payload
776  * was conveyed via RDMA Read).
777  */
778 static void svc_rdma_read_complete_one(struct svc_rqst *rqstp,
779 				       struct svc_rdma_recv_ctxt *ctxt)
780 {
781 	struct svc_rdma_chunk *chunk = pcl_first_chunk(&ctxt->rc_read_pcl);
782 	struct xdr_buf *buf = &rqstp->rq_arg;
783 	unsigned int length;
784 
785 	/* Split the Receive buffer between the head and tail
786 	 * buffers at Read chunk's position. XDR roundup of the
787 	 * chunk is not included in either the pagelist or in
788 	 * the tail.
789 	 */
790 	buf->tail[0].iov_base = buf->head[0].iov_base + chunk->ch_position;
791 	buf->tail[0].iov_len = buf->head[0].iov_len - chunk->ch_position;
792 	buf->head[0].iov_len = chunk->ch_position;
793 
794 	/* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2).
795 	 *
796 	 * If the client already rounded up the chunk length, the
797 	 * length does not change. Otherwise, the length of the page
798 	 * list is increased to include XDR round-up.
799 	 *
800 	 * Currently these chunks always start at page offset 0,
801 	 * thus the rounded-up length never crosses a page boundary.
802 	 */
803 	buf->pages = &rqstp->rq_pages[0];
804 	length = xdr_align_size(chunk->ch_length);
805 	buf->page_len = length;
806 	buf->len += length;
807 	buf->buflen += length;
808 }
809 
810 /* Finish constructing the RPC Call message in rqstp::rq_arg.
811  *
812  * The incoming RPC/RDMA message is an RDMA_MSG type message
813  * with payload in multiple Read chunks and no PZRC.
814  */
815 static void svc_rdma_read_complete_multiple(struct svc_rqst *rqstp,
816 					    struct svc_rdma_recv_ctxt *ctxt)
817 {
818 	struct xdr_buf *buf = &rqstp->rq_arg;
819 
820 	buf->len += ctxt->rc_readbytes;
821 	buf->buflen += ctxt->rc_readbytes;
822 
823 	buf->head[0].iov_base = page_address(rqstp->rq_pages[0]);
824 	buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes);
825 	buf->pages = &rqstp->rq_pages[1];
826 	buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len;
827 }
828 
829 /* Finish constructing the RPC Call message in rqstp::rq_arg.
830  *
831  * The incoming RPC/RDMA message is an RDMA_NOMSG type message
832  * (the RPC message body was conveyed via RDMA Read).
833  */
834 static void svc_rdma_read_complete_pzrc(struct svc_rqst *rqstp,
835 					struct svc_rdma_recv_ctxt *ctxt)
836 {
837 	struct xdr_buf *buf = &rqstp->rq_arg;
838 
839 	buf->len += ctxt->rc_readbytes;
840 	buf->buflen += ctxt->rc_readbytes;
841 
842 	buf->head[0].iov_base = page_address(rqstp->rq_pages[0]);
843 	buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes);
844 	buf->pages = &rqstp->rq_pages[1];
845 	buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len;
846 }
847 
848 static noinline void svc_rdma_read_complete(struct svc_rqst *rqstp,
849 					    struct svc_rdma_recv_ctxt *ctxt)
850 {
851 	unsigned int i;
852 
853 	/* Transfer the Read chunk pages into @rqstp.rq_pages, replacing
854 	 * the rq_pages that were already allocated for this rqstp.
855 	 */
856 	release_pages(rqstp->rq_respages, ctxt->rc_page_count);
857 	for (i = 0; i < ctxt->rc_page_count; i++)
858 		rqstp->rq_pages[i] = ctxt->rc_pages[i];
859 
860 	/* Update @rqstp's result send buffer to start after the
861 	 * last page in the RDMA Read payload.
862 	 */
863 	rqstp->rq_respages = &rqstp->rq_pages[ctxt->rc_page_count];
864 	rqstp->rq_next_page = rqstp->rq_respages + 1;
865 
866 	/* Prevent svc_rdma_recv_ctxt_put() from releasing the
867 	 * pages in ctxt::rc_pages a second time.
868 	 */
869 	ctxt->rc_page_count = 0;
870 
871 	/* Finish constructing the RPC Call message. The exact
872 	 * procedure for that depends on what kind of RPC/RDMA
873 	 * chunks were provided by the client.
874 	 */
875 	rqstp->rq_arg = ctxt->rc_saved_arg;
876 	if (pcl_is_empty(&ctxt->rc_call_pcl)) {
877 		if (ctxt->rc_read_pcl.cl_count == 1)
878 			svc_rdma_read_complete_one(rqstp, ctxt);
879 		else
880 			svc_rdma_read_complete_multiple(rqstp, ctxt);
881 	} else {
882 		svc_rdma_read_complete_pzrc(rqstp, ctxt);
883 	}
884 
885 	trace_svcrdma_read_finished(&ctxt->rc_cid);
886 }
887 
888 /**
889  * svc_rdma_recvfrom - Receive an RPC call
890  * @rqstp: request structure into which to receive an RPC Call
891  *
892  * Returns:
893  *	The positive number of bytes in the RPC Call message,
894  *	%0 if there were no Calls ready to return,
895  *	%-EINVAL if the Read chunk data is too large,
896  *	%-ENOMEM if rdma_rw context pool was exhausted,
897  *	%-ENOTCONN if posting failed (connection is lost),
898  *	%-EIO if rdma_rw initialization failed (DMA mapping, etc).
899  *
900  * Called in a loop when XPT_DATA is set. XPT_DATA is cleared only
901  * when there are no remaining ctxt's to process.
902  *
903  * The next ctxt is removed from the "receive" lists.
904  *
905  * - If the ctxt completes a Receive, then construct the Call
906  *   message from the contents of the Receive buffer.
907  *
908  *   - If there are no Read chunks in this message, then finish
909  *     assembling the Call message and return the number of bytes
910  *     in the message.
911  *
912  *   - If there are Read chunks in this message, post Read WRs to
913  *     pull that payload. When the Read WRs complete, build the
914  *     full message and return the number of bytes in it.
915  */
916 int svc_rdma_recvfrom(struct svc_rqst *rqstp)
917 {
918 	struct svc_xprt *xprt = rqstp->rq_xprt;
919 	struct svcxprt_rdma *rdma_xprt =
920 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
921 	struct svc_rdma_recv_ctxt *ctxt;
922 	int ret;
923 
924 	/* Prevent svc_xprt_release() from releasing pages in rq_pages
925 	 * when returning 0 or an error.
926 	 */
927 	rqstp->rq_respages = rqstp->rq_pages;
928 	rqstp->rq_next_page = rqstp->rq_respages;
929 
930 	rqstp->rq_xprt_ctxt = NULL;
931 
932 	spin_lock(&rdma_xprt->sc_rq_dto_lock);
933 	ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_read_complete_q);
934 	if (ctxt) {
935 		list_del(&ctxt->rc_list);
936 		spin_unlock(&rdma_xprt->sc_rq_dto_lock);
937 		svc_xprt_received(xprt);
938 		svc_rdma_read_complete(rqstp, ctxt);
939 		goto complete;
940 	}
941 	ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_rq_dto_q);
942 	if (ctxt)
943 		list_del(&ctxt->rc_list);
944 	else
945 		/* No new incoming requests, terminate the loop */
946 		clear_bit(XPT_DATA, &xprt->xpt_flags);
947 	spin_unlock(&rdma_xprt->sc_rq_dto_lock);
948 
949 	/* Unblock the transport for the next receive */
950 	svc_xprt_received(xprt);
951 	if (!ctxt)
952 		return 0;
953 
954 	percpu_counter_inc(&svcrdma_stat_recv);
955 	ib_dma_sync_single_for_cpu(rdma_xprt->sc_pd->device,
956 				   ctxt->rc_recv_sge.addr, ctxt->rc_byte_len,
957 				   DMA_FROM_DEVICE);
958 	svc_rdma_build_arg_xdr(rqstp, ctxt);
959 
960 	ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg, ctxt);
961 	if (ret < 0)
962 		goto out_err;
963 	if (ret == 0)
964 		goto out_drop;
965 
966 	if (svc_rdma_is_reverse_direction_reply(xprt, ctxt))
967 		goto out_backchannel;
968 
969 	svc_rdma_get_inv_rkey(rdma_xprt, ctxt);
970 
971 	if (!pcl_is_empty(&ctxt->rc_read_pcl) ||
972 	    !pcl_is_empty(&ctxt->rc_call_pcl))
973 		goto out_readlist;
974 
975 complete:
976 	rqstp->rq_xprt_ctxt = ctxt;
977 	rqstp->rq_prot = IPPROTO_MAX;
978 	svc_xprt_copy_addrs(rqstp, xprt);
979 	set_bit(RQ_SECURE, &rqstp->rq_flags);
980 	return rqstp->rq_arg.len;
981 
982 out_err:
983 	svc_rdma_send_error(rdma_xprt, ctxt, ret);
984 	svc_rdma_recv_ctxt_put(rdma_xprt, ctxt);
985 	return 0;
986 
987 out_readlist:
988 	/* This @rqstp is about to be recycled. Save the work
989 	 * already done constructing the Call message in rq_arg
990 	 * so it can be restored when the RDMA Reads have
991 	 * completed.
992 	 */
993 	ctxt->rc_saved_arg = rqstp->rq_arg;
994 
995 	ret = svc_rdma_process_read_list(rdma_xprt, rqstp, ctxt);
996 	if (ret < 0) {
997 		if (ret == -EINVAL)
998 			svc_rdma_send_error(rdma_xprt, ctxt, ret);
999 		svc_rdma_recv_ctxt_put(rdma_xprt, ctxt);
1000 		svc_xprt_deferred_close(xprt);
1001 		return ret;
1002 	}
1003 	return 0;
1004 
1005 out_backchannel:
1006 	svc_rdma_handle_bc_reply(rqstp, ctxt);
1007 out_drop:
1008 	svc_rdma_recv_ctxt_put(rdma_xprt, ctxt);
1009 	return 0;
1010 }
1011