xref: /linux/net/sunrpc/xprtrdma/svc_rdma_sendto.c (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
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_sendto. This is called by the
48  * RPC server when an RPC Reply is ready to be transmitted to a client.
49  *
50  * The passed-in svc_rqst contains a struct xdr_buf which holds an
51  * XDR-encoded RPC Reply message. sendto must construct the RPC-over-RDMA
52  * transport header, post all Write WRs needed for this Reply, then post
53  * a Send WR conveying the transport header and the RPC message itself to
54  * the client.
55  *
56  * svc_rdma_sendto must fully transmit the Reply before returning, as
57  * the svc_rqst will be recycled as soon as sendto returns. Remaining
58  * resources referred to by the svc_rqst are also recycled at that time.
59  * Therefore any resources that must remain longer must be detached
60  * from the svc_rqst and released later.
61  *
62  * Page Management
63  *
64  * The I/O that performs Reply transmission is asynchronous, and may
65  * complete well after sendto returns. Thus pages under I/O must be
66  * removed from the svc_rqst before sendto returns.
67  *
68  * The logic here depends on Send Queue and completion ordering. Since
69  * the Send WR is always posted last, it will always complete last. Thus
70  * when it completes, it is guaranteed that all previous Write WRs have
71  * also completed.
72  *
73  * Write WRs are constructed and posted. Each Write segment gets its own
74  * svc_rdma_rw_ctxt, allowing the Write completion handler to find and
75  * DMA-unmap the pages under I/O for that Write segment. The Write
76  * completion handler does not release any pages.
77  *
78  * When the Send WR is constructed, it also gets its own svc_rdma_send_ctxt.
79  * The ownership of all of the Reply's pages are transferred into that
80  * ctxt, the Send WR is posted, and sendto returns.
81  *
82  * The svc_rdma_send_ctxt is presented when the Send WR completes. The
83  * Send completion handler finally releases the Reply's pages.
84  *
85  * This mechanism also assumes that completions on the transport's Send
86  * Completion Queue do not run in parallel. Otherwise a Write completion
87  * and Send completion running at the same time could release pages that
88  * are still DMA-mapped.
89  *
90  * Error Handling
91  *
92  * - If the Send WR is posted successfully, it will either complete
93  *   successfully, or get flushed. Either way, the Send completion
94  *   handler releases the Reply's pages.
95  * - If the Send WR cannot be not posted, the forward path releases
96  *   the Reply's pages.
97  *
98  * This handles the case, without the use of page reference counting,
99  * where two different Write segments send portions of the same page.
100  */
101 
102 #include <linux/spinlock.h>
103 #include <asm/unaligned.h>
104 
105 #include <rdma/ib_verbs.h>
106 #include <rdma/rdma_cm.h>
107 
108 #include <linux/sunrpc/debug.h>
109 #include <linux/sunrpc/svc_rdma.h>
110 
111 #include "xprt_rdma.h"
112 #include <trace/events/rpcrdma.h>
113 
114 static void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc);
115 
116 static struct svc_rdma_send_ctxt *
117 svc_rdma_send_ctxt_alloc(struct svcxprt_rdma *rdma)
118 {
119 	int node = ibdev_to_node(rdma->sc_cm_id->device);
120 	struct svc_rdma_send_ctxt *ctxt;
121 	dma_addr_t addr;
122 	void *buffer;
123 	int i;
124 
125 	ctxt = kzalloc_node(struct_size(ctxt, sc_sges, rdma->sc_max_send_sges),
126 			    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_TO_DEVICE);
134 	if (ib_dma_mapping_error(rdma->sc_pd->device, addr))
135 		goto fail2;
136 
137 	svc_rdma_send_cid_init(rdma, &ctxt->sc_cid);
138 
139 	ctxt->sc_rdma = rdma;
140 	ctxt->sc_send_wr.next = NULL;
141 	ctxt->sc_send_wr.wr_cqe = &ctxt->sc_cqe;
142 	ctxt->sc_send_wr.sg_list = ctxt->sc_sges;
143 	ctxt->sc_send_wr.send_flags = IB_SEND_SIGNALED;
144 	ctxt->sc_cqe.done = svc_rdma_wc_send;
145 	ctxt->sc_xprt_buf = buffer;
146 	xdr_buf_init(&ctxt->sc_hdrbuf, ctxt->sc_xprt_buf,
147 		     rdma->sc_max_req_size);
148 	ctxt->sc_sges[0].addr = addr;
149 
150 	for (i = 0; i < rdma->sc_max_send_sges; i++)
151 		ctxt->sc_sges[i].lkey = rdma->sc_pd->local_dma_lkey;
152 	return ctxt;
153 
154 fail2:
155 	kfree(buffer);
156 fail1:
157 	kfree(ctxt);
158 fail0:
159 	return NULL;
160 }
161 
162 /**
163  * svc_rdma_send_ctxts_destroy - Release all send_ctxt's for an xprt
164  * @rdma: svcxprt_rdma being torn down
165  *
166  */
167 void svc_rdma_send_ctxts_destroy(struct svcxprt_rdma *rdma)
168 {
169 	struct svc_rdma_send_ctxt *ctxt;
170 	struct llist_node *node;
171 
172 	while ((node = llist_del_first(&rdma->sc_send_ctxts)) != NULL) {
173 		ctxt = llist_entry(node, struct svc_rdma_send_ctxt, sc_node);
174 		ib_dma_unmap_single(rdma->sc_pd->device,
175 				    ctxt->sc_sges[0].addr,
176 				    rdma->sc_max_req_size,
177 				    DMA_TO_DEVICE);
178 		kfree(ctxt->sc_xprt_buf);
179 		kfree(ctxt);
180 	}
181 }
182 
183 /**
184  * svc_rdma_send_ctxt_get - Get a free send_ctxt
185  * @rdma: controlling svcxprt_rdma
186  *
187  * Returns a ready-to-use send_ctxt, or NULL if none are
188  * available and a fresh one cannot be allocated.
189  */
190 struct svc_rdma_send_ctxt *svc_rdma_send_ctxt_get(struct svcxprt_rdma *rdma)
191 {
192 	struct svc_rdma_send_ctxt *ctxt;
193 	struct llist_node *node;
194 
195 	spin_lock(&rdma->sc_send_lock);
196 	node = llist_del_first(&rdma->sc_send_ctxts);
197 	spin_unlock(&rdma->sc_send_lock);
198 	if (!node)
199 		goto out_empty;
200 
201 	ctxt = llist_entry(node, struct svc_rdma_send_ctxt, sc_node);
202 
203 out:
204 	rpcrdma_set_xdrlen(&ctxt->sc_hdrbuf, 0);
205 	xdr_init_encode(&ctxt->sc_stream, &ctxt->sc_hdrbuf,
206 			ctxt->sc_xprt_buf, NULL);
207 
208 	svc_rdma_cc_init(rdma, &ctxt->sc_reply_info.wi_cc);
209 	ctxt->sc_send_wr.num_sge = 0;
210 	ctxt->sc_cur_sge_no = 0;
211 	ctxt->sc_page_count = 0;
212 	ctxt->sc_wr_chain = &ctxt->sc_send_wr;
213 	ctxt->sc_sqecount = 1;
214 
215 	return ctxt;
216 
217 out_empty:
218 	ctxt = svc_rdma_send_ctxt_alloc(rdma);
219 	if (!ctxt)
220 		return NULL;
221 	goto out;
222 }
223 
224 static void svc_rdma_send_ctxt_release(struct svcxprt_rdma *rdma,
225 				       struct svc_rdma_send_ctxt *ctxt)
226 {
227 	struct ib_device *device = rdma->sc_cm_id->device;
228 	unsigned int i;
229 
230 	svc_rdma_reply_chunk_release(rdma, ctxt);
231 
232 	if (ctxt->sc_page_count)
233 		release_pages(ctxt->sc_pages, ctxt->sc_page_count);
234 
235 	/* The first SGE contains the transport header, which
236 	 * remains mapped until @ctxt is destroyed.
237 	 */
238 	for (i = 1; i < ctxt->sc_send_wr.num_sge; i++) {
239 		trace_svcrdma_dma_unmap_page(&ctxt->sc_cid,
240 					     ctxt->sc_sges[i].addr,
241 					     ctxt->sc_sges[i].length);
242 		ib_dma_unmap_page(device,
243 				  ctxt->sc_sges[i].addr,
244 				  ctxt->sc_sges[i].length,
245 				  DMA_TO_DEVICE);
246 	}
247 
248 	llist_add(&ctxt->sc_node, &rdma->sc_send_ctxts);
249 }
250 
251 static void svc_rdma_send_ctxt_put_async(struct work_struct *work)
252 {
253 	struct svc_rdma_send_ctxt *ctxt;
254 
255 	ctxt = container_of(work, struct svc_rdma_send_ctxt, sc_work);
256 	svc_rdma_send_ctxt_release(ctxt->sc_rdma, ctxt);
257 }
258 
259 /**
260  * svc_rdma_send_ctxt_put - Return send_ctxt to free list
261  * @rdma: controlling svcxprt_rdma
262  * @ctxt: object to return to the free list
263  *
264  * Pages left in sc_pages are DMA unmapped and released.
265  */
266 void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
267 			    struct svc_rdma_send_ctxt *ctxt)
268 {
269 	INIT_WORK(&ctxt->sc_work, svc_rdma_send_ctxt_put_async);
270 	queue_work(svcrdma_wq, &ctxt->sc_work);
271 }
272 
273 /**
274  * svc_rdma_wake_send_waiters - manage Send Queue accounting
275  * @rdma: controlling transport
276  * @avail: Number of additional SQEs that are now available
277  *
278  */
279 void svc_rdma_wake_send_waiters(struct svcxprt_rdma *rdma, int avail)
280 {
281 	atomic_add(avail, &rdma->sc_sq_avail);
282 	smp_mb__after_atomic();
283 	if (unlikely(waitqueue_active(&rdma->sc_send_wait)))
284 		wake_up(&rdma->sc_send_wait);
285 }
286 
287 /**
288  * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC
289  * @cq: Completion Queue context
290  * @wc: Work Completion object
291  *
292  * NB: The svc_xprt/svcxprt_rdma is pinned whenever it's possible that
293  * the Send completion handler could be running.
294  */
295 static void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
296 {
297 	struct svcxprt_rdma *rdma = cq->cq_context;
298 	struct ib_cqe *cqe = wc->wr_cqe;
299 	struct svc_rdma_send_ctxt *ctxt =
300 		container_of(cqe, struct svc_rdma_send_ctxt, sc_cqe);
301 
302 	svc_rdma_wake_send_waiters(rdma, ctxt->sc_sqecount);
303 
304 	if (unlikely(wc->status != IB_WC_SUCCESS))
305 		goto flushed;
306 
307 	trace_svcrdma_wc_send(&ctxt->sc_cid);
308 	svc_rdma_send_ctxt_put(rdma, ctxt);
309 	return;
310 
311 flushed:
312 	if (wc->status != IB_WC_WR_FLUSH_ERR)
313 		trace_svcrdma_wc_send_err(wc, &ctxt->sc_cid);
314 	else
315 		trace_svcrdma_wc_send_flush(wc, &ctxt->sc_cid);
316 	svc_rdma_send_ctxt_put(rdma, ctxt);
317 	svc_xprt_deferred_close(&rdma->sc_xprt);
318 }
319 
320 /**
321  * svc_rdma_post_send - Post a WR chain to the Send Queue
322  * @rdma: transport context
323  * @ctxt: WR chain to post
324  *
325  * Copy fields in @ctxt to stack variables in order to guarantee
326  * that these values remain available after the ib_post_send() call.
327  * In some error flow cases, svc_rdma_wc_send() releases @ctxt.
328  *
329  * Note there is potential for starvation when the Send Queue is
330  * full because there is no order to when waiting threads are
331  * awoken. The transport is typically provisioned with a deep
332  * enough Send Queue that SQ exhaustion should be a rare event.
333  *
334  * Return values:
335  *   %0: @ctxt's WR chain was posted successfully
336  *   %-ENOTCONN: The connection was lost
337  */
338 int svc_rdma_post_send(struct svcxprt_rdma *rdma,
339 		       struct svc_rdma_send_ctxt *ctxt)
340 {
341 	struct ib_send_wr *first_wr = ctxt->sc_wr_chain;
342 	struct ib_send_wr *send_wr = &ctxt->sc_send_wr;
343 	const struct ib_send_wr *bad_wr = first_wr;
344 	struct rpc_rdma_cid cid = ctxt->sc_cid;
345 	int ret, sqecount = ctxt->sc_sqecount;
346 
347 	might_sleep();
348 
349 	/* Sync the transport header buffer */
350 	ib_dma_sync_single_for_device(rdma->sc_pd->device,
351 				      send_wr->sg_list[0].addr,
352 				      send_wr->sg_list[0].length,
353 				      DMA_TO_DEVICE);
354 
355 	/* If the SQ is full, wait until an SQ entry is available */
356 	while (!test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags)) {
357 		if (atomic_sub_return(sqecount, &rdma->sc_sq_avail) < 0) {
358 			svc_rdma_wake_send_waiters(rdma, sqecount);
359 
360 			/* When the transport is torn down, assume
361 			 * ib_drain_sq() will trigger enough Send
362 			 * completions to wake us. The XPT_CLOSE test
363 			 * above should then cause the while loop to
364 			 * exit.
365 			 */
366 			percpu_counter_inc(&svcrdma_stat_sq_starve);
367 			trace_svcrdma_sq_full(rdma, &cid);
368 			wait_event(rdma->sc_send_wait,
369 				   atomic_read(&rdma->sc_sq_avail) > 0);
370 			trace_svcrdma_sq_retry(rdma, &cid);
371 			continue;
372 		}
373 
374 		trace_svcrdma_post_send(ctxt);
375 		ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr);
376 		if (ret) {
377 			trace_svcrdma_sq_post_err(rdma, &cid, ret);
378 			svc_xprt_deferred_close(&rdma->sc_xprt);
379 
380 			/* If even one WR was posted, there will be a
381 			 * Send completion that bumps sc_sq_avail.
382 			 */
383 			if (bad_wr == first_wr) {
384 				svc_rdma_wake_send_waiters(rdma, sqecount);
385 				break;
386 			}
387 		}
388 		return 0;
389 	}
390 	return -ENOTCONN;
391 }
392 
393 /**
394  * svc_rdma_encode_read_list - Encode RPC Reply's Read chunk list
395  * @sctxt: Send context for the RPC Reply
396  *
397  * Return values:
398  *   On success, returns length in bytes of the Reply XDR buffer
399  *   that was consumed by the Reply Read list
400  *   %-EMSGSIZE on XDR buffer overflow
401  */
402 static ssize_t svc_rdma_encode_read_list(struct svc_rdma_send_ctxt *sctxt)
403 {
404 	/* RPC-over-RDMA version 1 replies never have a Read list. */
405 	return xdr_stream_encode_item_absent(&sctxt->sc_stream);
406 }
407 
408 /**
409  * svc_rdma_encode_write_segment - Encode one Write segment
410  * @sctxt: Send context for the RPC Reply
411  * @chunk: Write chunk to push
412  * @remaining: remaining bytes of the payload left in the Write chunk
413  * @segno: which segment in the chunk
414  *
415  * Return values:
416  *   On success, returns length in bytes of the Reply XDR buffer
417  *   that was consumed by the Write segment, and updates @remaining
418  *   %-EMSGSIZE on XDR buffer overflow
419  */
420 static ssize_t svc_rdma_encode_write_segment(struct svc_rdma_send_ctxt *sctxt,
421 					     const struct svc_rdma_chunk *chunk,
422 					     u32 *remaining, unsigned int segno)
423 {
424 	const struct svc_rdma_segment *segment = &chunk->ch_segments[segno];
425 	const size_t len = rpcrdma_segment_maxsz * sizeof(__be32);
426 	u32 length;
427 	__be32 *p;
428 
429 	p = xdr_reserve_space(&sctxt->sc_stream, len);
430 	if (!p)
431 		return -EMSGSIZE;
432 
433 	length = min_t(u32, *remaining, segment->rs_length);
434 	*remaining -= length;
435 	xdr_encode_rdma_segment(p, segment->rs_handle, length,
436 				segment->rs_offset);
437 	trace_svcrdma_encode_wseg(sctxt, segno, segment->rs_handle, length,
438 				  segment->rs_offset);
439 	return len;
440 }
441 
442 /**
443  * svc_rdma_encode_write_chunk - Encode one Write chunk
444  * @sctxt: Send context for the RPC Reply
445  * @chunk: Write chunk to push
446  *
447  * Copy a Write chunk from the Call transport header to the
448  * Reply transport header. Update each segment's length field
449  * to reflect the number of bytes written in that segment.
450  *
451  * Return values:
452  *   On success, returns length in bytes of the Reply XDR buffer
453  *   that was consumed by the Write chunk
454  *   %-EMSGSIZE on XDR buffer overflow
455  */
456 static ssize_t svc_rdma_encode_write_chunk(struct svc_rdma_send_ctxt *sctxt,
457 					   const struct svc_rdma_chunk *chunk)
458 {
459 	u32 remaining = chunk->ch_payload_length;
460 	unsigned int segno;
461 	ssize_t len, ret;
462 
463 	len = 0;
464 	ret = xdr_stream_encode_item_present(&sctxt->sc_stream);
465 	if (ret < 0)
466 		return ret;
467 	len += ret;
468 
469 	ret = xdr_stream_encode_u32(&sctxt->sc_stream, chunk->ch_segcount);
470 	if (ret < 0)
471 		return ret;
472 	len += ret;
473 
474 	for (segno = 0; segno < chunk->ch_segcount; segno++) {
475 		ret = svc_rdma_encode_write_segment(sctxt, chunk, &remaining, segno);
476 		if (ret < 0)
477 			return ret;
478 		len += ret;
479 	}
480 
481 	return len;
482 }
483 
484 /**
485  * svc_rdma_encode_write_list - Encode RPC Reply's Write chunk list
486  * @rctxt: Reply context with information about the RPC Call
487  * @sctxt: Send context for the RPC Reply
488  *
489  * Return values:
490  *   On success, returns length in bytes of the Reply XDR buffer
491  *   that was consumed by the Reply's Write list
492  *   %-EMSGSIZE on XDR buffer overflow
493  */
494 static ssize_t svc_rdma_encode_write_list(struct svc_rdma_recv_ctxt *rctxt,
495 					  struct svc_rdma_send_ctxt *sctxt)
496 {
497 	struct svc_rdma_chunk *chunk;
498 	ssize_t len, ret;
499 
500 	len = 0;
501 	pcl_for_each_chunk(chunk, &rctxt->rc_write_pcl) {
502 		ret = svc_rdma_encode_write_chunk(sctxt, chunk);
503 		if (ret < 0)
504 			return ret;
505 		len += ret;
506 	}
507 
508 	/* Terminate the Write list */
509 	ret = xdr_stream_encode_item_absent(&sctxt->sc_stream);
510 	if (ret < 0)
511 		return ret;
512 
513 	return len + ret;
514 }
515 
516 /**
517  * svc_rdma_encode_reply_chunk - Encode RPC Reply's Reply chunk
518  * @rctxt: Reply context with information about the RPC Call
519  * @sctxt: Send context for the RPC Reply
520  * @length: size in bytes of the payload in the Reply chunk
521  *
522  * Return values:
523  *   On success, returns length in bytes of the Reply XDR buffer
524  *   that was consumed by the Reply's Reply chunk
525  *   %-EMSGSIZE on XDR buffer overflow
526  *   %-E2BIG if the RPC message is larger than the Reply chunk
527  */
528 static ssize_t
529 svc_rdma_encode_reply_chunk(struct svc_rdma_recv_ctxt *rctxt,
530 			    struct svc_rdma_send_ctxt *sctxt,
531 			    unsigned int length)
532 {
533 	struct svc_rdma_chunk *chunk;
534 
535 	if (pcl_is_empty(&rctxt->rc_reply_pcl))
536 		return xdr_stream_encode_item_absent(&sctxt->sc_stream);
537 
538 	chunk = pcl_first_chunk(&rctxt->rc_reply_pcl);
539 	if (length > chunk->ch_length)
540 		return -E2BIG;
541 
542 	chunk->ch_payload_length = length;
543 	return svc_rdma_encode_write_chunk(sctxt, chunk);
544 }
545 
546 struct svc_rdma_map_data {
547 	struct svcxprt_rdma		*md_rdma;
548 	struct svc_rdma_send_ctxt	*md_ctxt;
549 };
550 
551 /**
552  * svc_rdma_page_dma_map - DMA map one page
553  * @data: pointer to arguments
554  * @page: struct page to DMA map
555  * @offset: offset into the page
556  * @len: number of bytes to map
557  *
558  * Returns:
559  *   %0 if DMA mapping was successful
560  *   %-EIO if the page cannot be DMA mapped
561  */
562 static int svc_rdma_page_dma_map(void *data, struct page *page,
563 				 unsigned long offset, unsigned int len)
564 {
565 	struct svc_rdma_map_data *args = data;
566 	struct svcxprt_rdma *rdma = args->md_rdma;
567 	struct svc_rdma_send_ctxt *ctxt = args->md_ctxt;
568 	struct ib_device *dev = rdma->sc_cm_id->device;
569 	dma_addr_t dma_addr;
570 
571 	++ctxt->sc_cur_sge_no;
572 
573 	dma_addr = ib_dma_map_page(dev, page, offset, len, DMA_TO_DEVICE);
574 	if (ib_dma_mapping_error(dev, dma_addr))
575 		goto out_maperr;
576 
577 	trace_svcrdma_dma_map_page(&ctxt->sc_cid, dma_addr, len);
578 	ctxt->sc_sges[ctxt->sc_cur_sge_no].addr = dma_addr;
579 	ctxt->sc_sges[ctxt->sc_cur_sge_no].length = len;
580 	ctxt->sc_send_wr.num_sge++;
581 	return 0;
582 
583 out_maperr:
584 	trace_svcrdma_dma_map_err(&ctxt->sc_cid, dma_addr, len);
585 	return -EIO;
586 }
587 
588 /**
589  * svc_rdma_iov_dma_map - DMA map an iovec
590  * @data: pointer to arguments
591  * @iov: kvec to DMA map
592  *
593  * ib_dma_map_page() is used here because svc_rdma_dma_unmap()
594  * handles DMA-unmap and it uses ib_dma_unmap_page() exclusively.
595  *
596  * Returns:
597  *   %0 if DMA mapping was successful
598  *   %-EIO if the iovec cannot be DMA mapped
599  */
600 static int svc_rdma_iov_dma_map(void *data, const struct kvec *iov)
601 {
602 	if (!iov->iov_len)
603 		return 0;
604 	return svc_rdma_page_dma_map(data, virt_to_page(iov->iov_base),
605 				     offset_in_page(iov->iov_base),
606 				     iov->iov_len);
607 }
608 
609 /**
610  * svc_rdma_xb_dma_map - DMA map all segments of an xdr_buf
611  * @xdr: xdr_buf containing portion of an RPC message to transmit
612  * @data: pointer to arguments
613  *
614  * Returns:
615  *   %0 if DMA mapping was successful
616  *   %-EIO if DMA mapping failed
617  *
618  * On failure, any DMA mappings that have been already done must be
619  * unmapped by the caller.
620  */
621 static int svc_rdma_xb_dma_map(const struct xdr_buf *xdr, void *data)
622 {
623 	unsigned int len, remaining;
624 	unsigned long pageoff;
625 	struct page **ppages;
626 	int ret;
627 
628 	ret = svc_rdma_iov_dma_map(data, &xdr->head[0]);
629 	if (ret < 0)
630 		return ret;
631 
632 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
633 	pageoff = offset_in_page(xdr->page_base);
634 	remaining = xdr->page_len;
635 	while (remaining) {
636 		len = min_t(u32, PAGE_SIZE - pageoff, remaining);
637 
638 		ret = svc_rdma_page_dma_map(data, *ppages++, pageoff, len);
639 		if (ret < 0)
640 			return ret;
641 
642 		remaining -= len;
643 		pageoff = 0;
644 	}
645 
646 	ret = svc_rdma_iov_dma_map(data, &xdr->tail[0]);
647 	if (ret < 0)
648 		return ret;
649 
650 	return xdr->len;
651 }
652 
653 struct svc_rdma_pullup_data {
654 	u8		*pd_dest;
655 	unsigned int	pd_length;
656 	unsigned int	pd_num_sges;
657 };
658 
659 /**
660  * svc_rdma_xb_count_sges - Count how many SGEs will be needed
661  * @xdr: xdr_buf containing portion of an RPC message to transmit
662  * @data: pointer to arguments
663  *
664  * Returns:
665  *   Number of SGEs needed to Send the contents of @xdr inline
666  */
667 static int svc_rdma_xb_count_sges(const struct xdr_buf *xdr,
668 				  void *data)
669 {
670 	struct svc_rdma_pullup_data *args = data;
671 	unsigned int remaining;
672 	unsigned long offset;
673 
674 	if (xdr->head[0].iov_len)
675 		++args->pd_num_sges;
676 
677 	offset = offset_in_page(xdr->page_base);
678 	remaining = xdr->page_len;
679 	while (remaining) {
680 		++args->pd_num_sges;
681 		remaining -= min_t(u32, PAGE_SIZE - offset, remaining);
682 		offset = 0;
683 	}
684 
685 	if (xdr->tail[0].iov_len)
686 		++args->pd_num_sges;
687 
688 	args->pd_length += xdr->len;
689 	return 0;
690 }
691 
692 /**
693  * svc_rdma_pull_up_needed - Determine whether to use pull-up
694  * @rdma: controlling transport
695  * @sctxt: send_ctxt for the Send WR
696  * @write_pcl: Write chunk list provided by client
697  * @xdr: xdr_buf containing RPC message to transmit
698  *
699  * Returns:
700  *   %true if pull-up must be used
701  *   %false otherwise
702  */
703 static bool svc_rdma_pull_up_needed(const struct svcxprt_rdma *rdma,
704 				    const struct svc_rdma_send_ctxt *sctxt,
705 				    const struct svc_rdma_pcl *write_pcl,
706 				    const struct xdr_buf *xdr)
707 {
708 	/* Resources needed for the transport header */
709 	struct svc_rdma_pullup_data args = {
710 		.pd_length	= sctxt->sc_hdrbuf.len,
711 		.pd_num_sges	= 1,
712 	};
713 	int ret;
714 
715 	ret = pcl_process_nonpayloads(write_pcl, xdr,
716 				      svc_rdma_xb_count_sges, &args);
717 	if (ret < 0)
718 		return false;
719 
720 	if (args.pd_length < RPCRDMA_PULLUP_THRESH)
721 		return true;
722 	return args.pd_num_sges >= rdma->sc_max_send_sges;
723 }
724 
725 /**
726  * svc_rdma_xb_linearize - Copy region of xdr_buf to flat buffer
727  * @xdr: xdr_buf containing portion of an RPC message to copy
728  * @data: pointer to arguments
729  *
730  * Returns:
731  *   Always zero.
732  */
733 static int svc_rdma_xb_linearize(const struct xdr_buf *xdr,
734 				 void *data)
735 {
736 	struct svc_rdma_pullup_data *args = data;
737 	unsigned int len, remaining;
738 	unsigned long pageoff;
739 	struct page **ppages;
740 
741 	if (xdr->head[0].iov_len) {
742 		memcpy(args->pd_dest, xdr->head[0].iov_base, xdr->head[0].iov_len);
743 		args->pd_dest += xdr->head[0].iov_len;
744 	}
745 
746 	ppages = xdr->pages + (xdr->page_base >> PAGE_SHIFT);
747 	pageoff = offset_in_page(xdr->page_base);
748 	remaining = xdr->page_len;
749 	while (remaining) {
750 		len = min_t(u32, PAGE_SIZE - pageoff, remaining);
751 		memcpy(args->pd_dest, page_address(*ppages) + pageoff, len);
752 		remaining -= len;
753 		args->pd_dest += len;
754 		pageoff = 0;
755 		ppages++;
756 	}
757 
758 	if (xdr->tail[0].iov_len) {
759 		memcpy(args->pd_dest, xdr->tail[0].iov_base, xdr->tail[0].iov_len);
760 		args->pd_dest += xdr->tail[0].iov_len;
761 	}
762 
763 	args->pd_length += xdr->len;
764 	return 0;
765 }
766 
767 /**
768  * svc_rdma_pull_up_reply_msg - Copy Reply into a single buffer
769  * @rdma: controlling transport
770  * @sctxt: send_ctxt for the Send WR; xprt hdr is already prepared
771  * @write_pcl: Write chunk list provided by client
772  * @xdr: prepared xdr_buf containing RPC message
773  *
774  * The device is not capable of sending the reply directly.
775  * Assemble the elements of @xdr into the transport header buffer.
776  *
777  * Assumptions:
778  *  pull_up_needed has determined that @xdr will fit in the buffer.
779  *
780  * Returns:
781  *   %0 if pull-up was successful
782  *   %-EMSGSIZE if a buffer manipulation problem occurred
783  */
784 static int svc_rdma_pull_up_reply_msg(const struct svcxprt_rdma *rdma,
785 				      struct svc_rdma_send_ctxt *sctxt,
786 				      const struct svc_rdma_pcl *write_pcl,
787 				      const struct xdr_buf *xdr)
788 {
789 	struct svc_rdma_pullup_data args = {
790 		.pd_dest	= sctxt->sc_xprt_buf + sctxt->sc_hdrbuf.len,
791 	};
792 	int ret;
793 
794 	ret = pcl_process_nonpayloads(write_pcl, xdr,
795 				      svc_rdma_xb_linearize, &args);
796 	if (ret < 0)
797 		return ret;
798 
799 	sctxt->sc_sges[0].length = sctxt->sc_hdrbuf.len + args.pd_length;
800 	trace_svcrdma_send_pullup(sctxt, args.pd_length);
801 	return 0;
802 }
803 
804 /* svc_rdma_map_reply_msg - DMA map the buffer holding RPC message
805  * @rdma: controlling transport
806  * @sctxt: send_ctxt for the Send WR
807  * @write_pcl: Write chunk list provided by client
808  * @reply_pcl: Reply chunk provided by client
809  * @xdr: prepared xdr_buf containing RPC message
810  *
811  * Returns:
812  *   %0 if DMA mapping was successful.
813  *   %-EMSGSIZE if a buffer manipulation problem occurred
814  *   %-EIO if DMA mapping failed
815  *
816  * The Send WR's num_sge field is set in all cases.
817  */
818 int svc_rdma_map_reply_msg(struct svcxprt_rdma *rdma,
819 			   struct svc_rdma_send_ctxt *sctxt,
820 			   const struct svc_rdma_pcl *write_pcl,
821 			   const struct svc_rdma_pcl *reply_pcl,
822 			   const struct xdr_buf *xdr)
823 {
824 	struct svc_rdma_map_data args = {
825 		.md_rdma	= rdma,
826 		.md_ctxt	= sctxt,
827 	};
828 
829 	/* Set up the (persistently-mapped) transport header SGE. */
830 	sctxt->sc_send_wr.num_sge = 1;
831 	sctxt->sc_sges[0].length = sctxt->sc_hdrbuf.len;
832 
833 	/* If there is a Reply chunk, nothing follows the transport
834 	 * header, so there is nothing to map.
835 	 */
836 	if (!pcl_is_empty(reply_pcl))
837 		return 0;
838 
839 	/* For pull-up, svc_rdma_send() will sync the transport header.
840 	 * No additional DMA mapping is necessary.
841 	 */
842 	if (svc_rdma_pull_up_needed(rdma, sctxt, write_pcl, xdr))
843 		return svc_rdma_pull_up_reply_msg(rdma, sctxt, write_pcl, xdr);
844 
845 	return pcl_process_nonpayloads(write_pcl, xdr,
846 				       svc_rdma_xb_dma_map, &args);
847 }
848 
849 /* The svc_rqst and all resources it owns are released as soon as
850  * svc_rdma_sendto returns. Transfer pages under I/O to the ctxt
851  * so they are released by the Send completion handler.
852  */
853 static void svc_rdma_save_io_pages(struct svc_rqst *rqstp,
854 				   struct svc_rdma_send_ctxt *ctxt)
855 {
856 	int i, pages = rqstp->rq_next_page - rqstp->rq_respages;
857 
858 	ctxt->sc_page_count += pages;
859 	for (i = 0; i < pages; i++) {
860 		ctxt->sc_pages[i] = rqstp->rq_respages[i];
861 		rqstp->rq_respages[i] = NULL;
862 	}
863 
864 	/* Prevent svc_xprt_release from releasing pages in rq_pages */
865 	rqstp->rq_next_page = rqstp->rq_respages;
866 }
867 
868 /* Prepare the portion of the RPC Reply that will be transmitted
869  * via RDMA Send. The RPC-over-RDMA transport header is prepared
870  * in sc_sges[0], and the RPC xdr_buf is prepared in following sges.
871  *
872  * Depending on whether a Write list or Reply chunk is present,
873  * the server may Send all, a portion of, or none of the xdr_buf.
874  * In the latter case, only the transport header (sc_sges[0]) is
875  * transmitted.
876  *
877  * Assumptions:
878  * - The Reply's transport header will never be larger than a page.
879  */
880 static int svc_rdma_send_reply_msg(struct svcxprt_rdma *rdma,
881 				   struct svc_rdma_send_ctxt *sctxt,
882 				   const struct svc_rdma_recv_ctxt *rctxt,
883 				   struct svc_rqst *rqstp)
884 {
885 	struct ib_send_wr *send_wr = &sctxt->sc_send_wr;
886 	int ret;
887 
888 	ret = svc_rdma_map_reply_msg(rdma, sctxt, &rctxt->rc_write_pcl,
889 				     &rctxt->rc_reply_pcl, &rqstp->rq_res);
890 	if (ret < 0)
891 		return ret;
892 
893 	/* Transfer pages involved in RDMA Writes to the sctxt's
894 	 * page array. Completion handling releases these pages.
895 	 */
896 	svc_rdma_save_io_pages(rqstp, sctxt);
897 
898 	if (rctxt->rc_inv_rkey) {
899 		send_wr->opcode = IB_WR_SEND_WITH_INV;
900 		send_wr->ex.invalidate_rkey = rctxt->rc_inv_rkey;
901 	} else {
902 		send_wr->opcode = IB_WR_SEND;
903 	}
904 
905 	return svc_rdma_post_send(rdma, sctxt);
906 }
907 
908 /**
909  * svc_rdma_send_error_msg - Send an RPC/RDMA v1 error response
910  * @rdma: controlling transport context
911  * @sctxt: Send context for the response
912  * @rctxt: Receive context for incoming bad message
913  * @status: negative errno indicating error that occurred
914  *
915  * Given the client-provided Read, Write, and Reply chunks, the
916  * server was not able to parse the Call or form a complete Reply.
917  * Return an RDMA_ERROR message so the client can retire the RPC
918  * transaction.
919  *
920  * The caller does not have to release @sctxt. It is released by
921  * Send completion, or by this function on error.
922  */
923 void svc_rdma_send_error_msg(struct svcxprt_rdma *rdma,
924 			     struct svc_rdma_send_ctxt *sctxt,
925 			     struct svc_rdma_recv_ctxt *rctxt,
926 			     int status)
927 {
928 	__be32 *rdma_argp = rctxt->rc_recv_buf;
929 	__be32 *p;
930 
931 	rpcrdma_set_xdrlen(&sctxt->sc_hdrbuf, 0);
932 	xdr_init_encode(&sctxt->sc_stream, &sctxt->sc_hdrbuf,
933 			sctxt->sc_xprt_buf, NULL);
934 
935 	p = xdr_reserve_space(&sctxt->sc_stream,
936 			      rpcrdma_fixed_maxsz * sizeof(*p));
937 	if (!p)
938 		goto put_ctxt;
939 
940 	*p++ = *rdma_argp;
941 	*p++ = *(rdma_argp + 1);
942 	*p++ = rdma->sc_fc_credits;
943 	*p = rdma_error;
944 
945 	switch (status) {
946 	case -EPROTONOSUPPORT:
947 		p = xdr_reserve_space(&sctxt->sc_stream, 3 * sizeof(*p));
948 		if (!p)
949 			goto put_ctxt;
950 
951 		*p++ = err_vers;
952 		*p++ = rpcrdma_version;
953 		*p = rpcrdma_version;
954 		trace_svcrdma_err_vers(*rdma_argp);
955 		break;
956 	default:
957 		p = xdr_reserve_space(&sctxt->sc_stream, sizeof(*p));
958 		if (!p)
959 			goto put_ctxt;
960 
961 		*p = err_chunk;
962 		trace_svcrdma_err_chunk(*rdma_argp);
963 	}
964 
965 	/* Remote Invalidation is skipped for simplicity. */
966 	sctxt->sc_send_wr.num_sge = 1;
967 	sctxt->sc_send_wr.opcode = IB_WR_SEND;
968 	sctxt->sc_sges[0].length = sctxt->sc_hdrbuf.len;
969 	if (svc_rdma_post_send(rdma, sctxt))
970 		goto put_ctxt;
971 	return;
972 
973 put_ctxt:
974 	svc_rdma_send_ctxt_put(rdma, sctxt);
975 }
976 
977 /**
978  * svc_rdma_sendto - Transmit an RPC reply
979  * @rqstp: processed RPC request, reply XDR already in ::rq_res
980  *
981  * Any resources still associated with @rqstp are released upon return.
982  * If no reply message was possible, the connection is closed.
983  *
984  * Returns:
985  *	%0 if an RPC reply has been successfully posted,
986  *	%-ENOMEM if a resource shortage occurred (connection is lost),
987  *	%-ENOTCONN if posting failed (connection is lost).
988  */
989 int svc_rdma_sendto(struct svc_rqst *rqstp)
990 {
991 	struct svc_xprt *xprt = rqstp->rq_xprt;
992 	struct svcxprt_rdma *rdma =
993 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
994 	struct svc_rdma_recv_ctxt *rctxt = rqstp->rq_xprt_ctxt;
995 	__be32 *rdma_argp = rctxt->rc_recv_buf;
996 	struct svc_rdma_send_ctxt *sctxt;
997 	unsigned int rc_size;
998 	__be32 *p;
999 	int ret;
1000 
1001 	ret = -ENOTCONN;
1002 	if (svc_xprt_is_dead(xprt))
1003 		goto drop_connection;
1004 
1005 	ret = -ENOMEM;
1006 	sctxt = svc_rdma_send_ctxt_get(rdma);
1007 	if (!sctxt)
1008 		goto drop_connection;
1009 
1010 	ret = -EMSGSIZE;
1011 	p = xdr_reserve_space(&sctxt->sc_stream,
1012 			      rpcrdma_fixed_maxsz * sizeof(*p));
1013 	if (!p)
1014 		goto put_ctxt;
1015 
1016 	ret = svc_rdma_send_write_list(rdma, rctxt, &rqstp->rq_res);
1017 	if (ret < 0)
1018 		goto put_ctxt;
1019 
1020 	rc_size = 0;
1021 	if (!pcl_is_empty(&rctxt->rc_reply_pcl)) {
1022 		ret = svc_rdma_prepare_reply_chunk(rdma, &rctxt->rc_write_pcl,
1023 						   &rctxt->rc_reply_pcl, sctxt,
1024 						   &rqstp->rq_res);
1025 		if (ret < 0)
1026 			goto reply_chunk;
1027 		rc_size = ret;
1028 	}
1029 
1030 	*p++ = *rdma_argp;
1031 	*p++ = *(rdma_argp + 1);
1032 	*p++ = rdma->sc_fc_credits;
1033 	*p = pcl_is_empty(&rctxt->rc_reply_pcl) ? rdma_msg : rdma_nomsg;
1034 
1035 	ret = svc_rdma_encode_read_list(sctxt);
1036 	if (ret < 0)
1037 		goto put_ctxt;
1038 	ret = svc_rdma_encode_write_list(rctxt, sctxt);
1039 	if (ret < 0)
1040 		goto put_ctxt;
1041 	ret = svc_rdma_encode_reply_chunk(rctxt, sctxt, rc_size);
1042 	if (ret < 0)
1043 		goto put_ctxt;
1044 
1045 	ret = svc_rdma_send_reply_msg(rdma, sctxt, rctxt, rqstp);
1046 	if (ret < 0)
1047 		goto put_ctxt;
1048 	return 0;
1049 
1050 reply_chunk:
1051 	if (ret != -E2BIG && ret != -EINVAL)
1052 		goto put_ctxt;
1053 
1054 	/* Send completion releases payload pages that were part
1055 	 * of previously posted RDMA Writes.
1056 	 */
1057 	svc_rdma_save_io_pages(rqstp, sctxt);
1058 	svc_rdma_send_error_msg(rdma, sctxt, rctxt, ret);
1059 	return 0;
1060 
1061 put_ctxt:
1062 	svc_rdma_send_ctxt_put(rdma, sctxt);
1063 drop_connection:
1064 	trace_svcrdma_send_err(rqstp, ret);
1065 	svc_xprt_deferred_close(&rdma->sc_xprt);
1066 	return -ENOTCONN;
1067 }
1068 
1069 /**
1070  * svc_rdma_result_payload - special processing for a result payload
1071  * @rqstp: RPC transaction context
1072  * @offset: payload's byte offset in @rqstp->rq_res
1073  * @length: size of payload, in bytes
1074  *
1075  * Assign the passed-in result payload to the current Write chunk,
1076  * and advance to cur_result_payload to the next Write chunk, if
1077  * there is one.
1078  *
1079  * Return values:
1080  *   %0 if successful or nothing needed to be done
1081  *   %-E2BIG if the payload was larger than the Write chunk
1082  */
1083 int svc_rdma_result_payload(struct svc_rqst *rqstp, unsigned int offset,
1084 			    unsigned int length)
1085 {
1086 	struct svc_rdma_recv_ctxt *rctxt = rqstp->rq_xprt_ctxt;
1087 	struct svc_rdma_chunk *chunk;
1088 
1089 	chunk = rctxt->rc_cur_result_payload;
1090 	if (!length || !chunk)
1091 		return 0;
1092 	rctxt->rc_cur_result_payload =
1093 		pcl_next_chunk(&rctxt->rc_write_pcl, chunk);
1094 
1095 	if (length > chunk->ch_length)
1096 		return -E2BIG;
1097 	chunk->ch_position = offset;
1098 	chunk->ch_payload_length = length;
1099 	return 0;
1100 }
1101