1 /*
2  * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
3  * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the BSD-type
9  * license below:
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  *      Redistributions of source code must retain the above copyright
16  *      notice, this list of conditions and the following disclaimer.
17  *
18  *      Redistributions in binary form must reproduce the above
19  *      copyright notice, this list of conditions and the following
20  *      disclaimer in the documentation and/or other materials provided
21  *      with the distribution.
22  *
23  *      Neither the name of the Network Appliance, Inc. nor the names of
24  *      its contributors may be used to endorse or promote products
25  *      derived from this software without specific prior written
26  *      permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39  *
40  * Author: Tom Tucker <tom@opengridcomputing.com>
41  */
42 
43 #include <linux/sunrpc/debug.h>
44 #include <linux/sunrpc/rpc_rdma.h>
45 #include <linux/spinlock.h>
46 #include <asm/unaligned.h>
47 #include <rdma/ib_verbs.h>
48 #include <rdma/rdma_cm.h>
49 #include <linux/sunrpc/svc_rdma.h>
50 
51 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
52 
53 /*
54  * Replace the pages in the rq_argpages array with the pages from the SGE in
55  * the RDMA_RECV completion. The SGL should contain full pages up until the
56  * last one.
57  */
58 static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
59 			       struct svc_rdma_op_ctxt *ctxt,
60 			       u32 byte_count)
61 {
62 	struct rpcrdma_msg *rmsgp;
63 	struct page *page;
64 	u32 bc;
65 	int sge_no;
66 
67 	/* Swap the page in the SGE with the page in argpages */
68 	page = ctxt->pages[0];
69 	put_page(rqstp->rq_pages[0]);
70 	rqstp->rq_pages[0] = page;
71 
72 	/* Set up the XDR head */
73 	rqstp->rq_arg.head[0].iov_base = page_address(page);
74 	rqstp->rq_arg.head[0].iov_len =
75 		min_t(size_t, byte_count, ctxt->sge[0].length);
76 	rqstp->rq_arg.len = byte_count;
77 	rqstp->rq_arg.buflen = byte_count;
78 
79 	/* Compute bytes past head in the SGL */
80 	bc = byte_count - rqstp->rq_arg.head[0].iov_len;
81 
82 	/* If data remains, store it in the pagelist */
83 	rqstp->rq_arg.page_len = bc;
84 	rqstp->rq_arg.page_base = 0;
85 
86 	/* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
87 	rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
88 	if (be32_to_cpu(rmsgp->rm_type) == RDMA_NOMSG)
89 		rqstp->rq_arg.pages = &rqstp->rq_pages[0];
90 	else
91 		rqstp->rq_arg.pages = &rqstp->rq_pages[1];
92 
93 	sge_no = 1;
94 	while (bc && sge_no < ctxt->count) {
95 		page = ctxt->pages[sge_no];
96 		put_page(rqstp->rq_pages[sge_no]);
97 		rqstp->rq_pages[sge_no] = page;
98 		bc -= min_t(u32, bc, ctxt->sge[sge_no].length);
99 		rqstp->rq_arg.buflen += ctxt->sge[sge_no].length;
100 		sge_no++;
101 	}
102 	rqstp->rq_respages = &rqstp->rq_pages[sge_no];
103 	rqstp->rq_next_page = rqstp->rq_respages + 1;
104 
105 	/* If not all pages were used from the SGL, free the remaining ones */
106 	bc = sge_no;
107 	while (sge_no < ctxt->count) {
108 		page = ctxt->pages[sge_no++];
109 		put_page(page);
110 	}
111 	ctxt->count = bc;
112 
113 	/* Set up tail */
114 	rqstp->rq_arg.tail[0].iov_base = NULL;
115 	rqstp->rq_arg.tail[0].iov_len = 0;
116 }
117 
118 static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count)
119 {
120 	if (rdma_node_get_transport(xprt->sc_cm_id->device->node_type) ==
121 	     RDMA_TRANSPORT_IWARP)
122 		return 1;
123 	else
124 		return min_t(int, sge_count, xprt->sc_max_sge);
125 }
126 
127 /* Issue an RDMA_READ using the local lkey to map the data sink */
128 int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
129 			struct svc_rqst *rqstp,
130 			struct svc_rdma_op_ctxt *head,
131 			int *page_no,
132 			u32 *page_offset,
133 			u32 rs_handle,
134 			u32 rs_length,
135 			u64 rs_offset,
136 			bool last)
137 {
138 	struct ib_send_wr read_wr;
139 	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
140 	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
141 	int ret, read, pno;
142 	u32 pg_off = *page_offset;
143 	u32 pg_no = *page_no;
144 
145 	ctxt->direction = DMA_FROM_DEVICE;
146 	ctxt->read_hdr = head;
147 	pages_needed =
148 		min_t(int, pages_needed, rdma_read_max_sge(xprt, pages_needed));
149 	read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
150 
151 	for (pno = 0; pno < pages_needed; pno++) {
152 		int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
153 
154 		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
155 		head->arg.page_len += len;
156 		head->arg.len += len;
157 		if (!pg_off)
158 			head->count++;
159 		rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
160 		rqstp->rq_next_page = rqstp->rq_respages + 1;
161 		ctxt->sge[pno].addr =
162 			ib_dma_map_page(xprt->sc_cm_id->device,
163 					head->arg.pages[pg_no], pg_off,
164 					PAGE_SIZE - pg_off,
165 					DMA_FROM_DEVICE);
166 		ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
167 					   ctxt->sge[pno].addr);
168 		if (ret)
169 			goto err;
170 		atomic_inc(&xprt->sc_dma_used);
171 
172 		/* The lkey here is either a local dma lkey or a dma_mr lkey */
173 		ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
174 		ctxt->sge[pno].length = len;
175 		ctxt->count++;
176 
177 		/* adjust offset and wrap to next page if needed */
178 		pg_off += len;
179 		if (pg_off == PAGE_SIZE) {
180 			pg_off = 0;
181 			pg_no++;
182 		}
183 		rs_length -= len;
184 	}
185 
186 	if (last && rs_length == 0)
187 		set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
188 	else
189 		clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
190 
191 	memset(&read_wr, 0, sizeof(read_wr));
192 	read_wr.wr_id = (unsigned long)ctxt;
193 	read_wr.opcode = IB_WR_RDMA_READ;
194 	ctxt->wr_op = read_wr.opcode;
195 	read_wr.send_flags = IB_SEND_SIGNALED;
196 	read_wr.wr.rdma.rkey = rs_handle;
197 	read_wr.wr.rdma.remote_addr = rs_offset;
198 	read_wr.sg_list = ctxt->sge;
199 	read_wr.num_sge = pages_needed;
200 
201 	ret = svc_rdma_send(xprt, &read_wr);
202 	if (ret) {
203 		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
204 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
205 		goto err;
206 	}
207 
208 	/* return current location in page array */
209 	*page_no = pg_no;
210 	*page_offset = pg_off;
211 	ret = read;
212 	atomic_inc(&rdma_stat_read);
213 	return ret;
214  err:
215 	svc_rdma_unmap_dma(ctxt);
216 	svc_rdma_put_context(ctxt, 0);
217 	return ret;
218 }
219 
220 /* Issue an RDMA_READ using an FRMR to map the data sink */
221 int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
222 			 struct svc_rqst *rqstp,
223 			 struct svc_rdma_op_ctxt *head,
224 			 int *page_no,
225 			 u32 *page_offset,
226 			 u32 rs_handle,
227 			 u32 rs_length,
228 			 u64 rs_offset,
229 			 bool last)
230 {
231 	struct ib_send_wr read_wr;
232 	struct ib_send_wr inv_wr;
233 	struct ib_send_wr fastreg_wr;
234 	u8 key;
235 	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
236 	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
237 	struct svc_rdma_fastreg_mr *frmr = svc_rdma_get_frmr(xprt);
238 	int ret, read, pno;
239 	u32 pg_off = *page_offset;
240 	u32 pg_no = *page_no;
241 
242 	if (IS_ERR(frmr))
243 		return -ENOMEM;
244 
245 	ctxt->direction = DMA_FROM_DEVICE;
246 	ctxt->frmr = frmr;
247 	pages_needed = min_t(int, pages_needed, xprt->sc_frmr_pg_list_len);
248 	read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
249 
250 	frmr->kva = page_address(rqstp->rq_arg.pages[pg_no]);
251 	frmr->direction = DMA_FROM_DEVICE;
252 	frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE);
253 	frmr->map_len = pages_needed << PAGE_SHIFT;
254 	frmr->page_list_len = pages_needed;
255 
256 	for (pno = 0; pno < pages_needed; pno++) {
257 		int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
258 
259 		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
260 		head->arg.page_len += len;
261 		head->arg.len += len;
262 		if (!pg_off)
263 			head->count++;
264 		rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
265 		rqstp->rq_next_page = rqstp->rq_respages + 1;
266 		frmr->page_list->page_list[pno] =
267 			ib_dma_map_page(xprt->sc_cm_id->device,
268 					head->arg.pages[pg_no], 0,
269 					PAGE_SIZE, DMA_FROM_DEVICE);
270 		ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
271 					   frmr->page_list->page_list[pno]);
272 		if (ret)
273 			goto err;
274 		atomic_inc(&xprt->sc_dma_used);
275 
276 		/* adjust offset and wrap to next page if needed */
277 		pg_off += len;
278 		if (pg_off == PAGE_SIZE) {
279 			pg_off = 0;
280 			pg_no++;
281 		}
282 		rs_length -= len;
283 	}
284 
285 	if (last && rs_length == 0)
286 		set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
287 	else
288 		clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
289 
290 	/* Bump the key */
291 	key = (u8)(frmr->mr->lkey & 0x000000FF);
292 	ib_update_fast_reg_key(frmr->mr, ++key);
293 
294 	ctxt->sge[0].addr = (unsigned long)frmr->kva + *page_offset;
295 	ctxt->sge[0].lkey = frmr->mr->lkey;
296 	ctxt->sge[0].length = read;
297 	ctxt->count = 1;
298 	ctxt->read_hdr = head;
299 
300 	/* Prepare FASTREG WR */
301 	memset(&fastreg_wr, 0, sizeof(fastreg_wr));
302 	fastreg_wr.opcode = IB_WR_FAST_REG_MR;
303 	fastreg_wr.send_flags = IB_SEND_SIGNALED;
304 	fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
305 	fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
306 	fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
307 	fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
308 	fastreg_wr.wr.fast_reg.length = frmr->map_len;
309 	fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
310 	fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
311 	fastreg_wr.next = &read_wr;
312 
313 	/* Prepare RDMA_READ */
314 	memset(&read_wr, 0, sizeof(read_wr));
315 	read_wr.send_flags = IB_SEND_SIGNALED;
316 	read_wr.wr.rdma.rkey = rs_handle;
317 	read_wr.wr.rdma.remote_addr = rs_offset;
318 	read_wr.sg_list = ctxt->sge;
319 	read_wr.num_sge = 1;
320 	if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) {
321 		read_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
322 		read_wr.wr_id = (unsigned long)ctxt;
323 		read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
324 	} else {
325 		read_wr.opcode = IB_WR_RDMA_READ;
326 		read_wr.next = &inv_wr;
327 		/* Prepare invalidate */
328 		memset(&inv_wr, 0, sizeof(inv_wr));
329 		inv_wr.wr_id = (unsigned long)ctxt;
330 		inv_wr.opcode = IB_WR_LOCAL_INV;
331 		inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE;
332 		inv_wr.ex.invalidate_rkey = frmr->mr->lkey;
333 	}
334 	ctxt->wr_op = read_wr.opcode;
335 
336 	/* Post the chain */
337 	ret = svc_rdma_send(xprt, &fastreg_wr);
338 	if (ret) {
339 		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
340 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
341 		goto err;
342 	}
343 
344 	/* return current location in page array */
345 	*page_no = pg_no;
346 	*page_offset = pg_off;
347 	ret = read;
348 	atomic_inc(&rdma_stat_read);
349 	return ret;
350  err:
351 	svc_rdma_unmap_dma(ctxt);
352 	svc_rdma_put_context(ctxt, 0);
353 	svc_rdma_put_frmr(xprt, frmr);
354 	return ret;
355 }
356 
357 static unsigned int
358 rdma_rcl_chunk_count(struct rpcrdma_read_chunk *ch)
359 {
360 	unsigned int count;
361 
362 	for (count = 0; ch->rc_discrim != xdr_zero; ch++)
363 		count++;
364 	return count;
365 }
366 
367 /* If there was additional inline content, append it to the end of arg.pages.
368  * Tail copy has to be done after the reader function has determined how many
369  * pages are needed for RDMA READ.
370  */
371 static int
372 rdma_copy_tail(struct svc_rqst *rqstp, struct svc_rdma_op_ctxt *head,
373 	       u32 position, u32 byte_count, u32 page_offset, int page_no)
374 {
375 	char *srcp, *destp;
376 	int ret;
377 
378 	ret = 0;
379 	srcp = head->arg.head[0].iov_base + position;
380 	byte_count = head->arg.head[0].iov_len - position;
381 	if (byte_count > PAGE_SIZE) {
382 		dprintk("svcrdma: large tail unsupported\n");
383 		return 0;
384 	}
385 
386 	/* Fit as much of the tail on the current page as possible */
387 	if (page_offset != PAGE_SIZE) {
388 		destp = page_address(rqstp->rq_arg.pages[page_no]);
389 		destp += page_offset;
390 		while (byte_count--) {
391 			*destp++ = *srcp++;
392 			page_offset++;
393 			if (page_offset == PAGE_SIZE && byte_count)
394 				goto more;
395 		}
396 		goto done;
397 	}
398 
399 more:
400 	/* Fit the rest on the next page */
401 	page_no++;
402 	destp = page_address(rqstp->rq_arg.pages[page_no]);
403 	while (byte_count--)
404 		*destp++ = *srcp++;
405 
406 	rqstp->rq_respages = &rqstp->rq_arg.pages[page_no+1];
407 	rqstp->rq_next_page = rqstp->rq_respages + 1;
408 
409 done:
410 	byte_count = head->arg.head[0].iov_len - position;
411 	head->arg.page_len += byte_count;
412 	head->arg.len += byte_count;
413 	head->arg.buflen += byte_count;
414 	return 1;
415 }
416 
417 static int rdma_read_chunks(struct svcxprt_rdma *xprt,
418 			    struct rpcrdma_msg *rmsgp,
419 			    struct svc_rqst *rqstp,
420 			    struct svc_rdma_op_ctxt *head)
421 {
422 	int page_no, ret;
423 	struct rpcrdma_read_chunk *ch;
424 	u32 handle, page_offset, byte_count;
425 	u32 position;
426 	u64 rs_offset;
427 	bool last;
428 
429 	/* If no read list is present, return 0 */
430 	ch = svc_rdma_get_read_chunk(rmsgp);
431 	if (!ch)
432 		return 0;
433 
434 	if (rdma_rcl_chunk_count(ch) > RPCSVC_MAXPAGES)
435 		return -EINVAL;
436 
437 	/* The request is completed when the RDMA_READs complete. The
438 	 * head context keeps all the pages that comprise the
439 	 * request.
440 	 */
441 	head->arg.head[0] = rqstp->rq_arg.head[0];
442 	head->arg.tail[0] = rqstp->rq_arg.tail[0];
443 	head->hdr_count = head->count;
444 	head->arg.page_base = 0;
445 	head->arg.page_len = 0;
446 	head->arg.len = rqstp->rq_arg.len;
447 	head->arg.buflen = rqstp->rq_arg.buflen;
448 
449 	ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
450 	position = be32_to_cpu(ch->rc_position);
451 
452 	/* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
453 	if (position == 0) {
454 		head->arg.pages = &head->pages[0];
455 		page_offset = head->byte_len;
456 	} else {
457 		head->arg.pages = &head->pages[head->count];
458 		page_offset = 0;
459 	}
460 
461 	ret = 0;
462 	page_no = 0;
463 	for (; ch->rc_discrim != xdr_zero; ch++) {
464 		if (be32_to_cpu(ch->rc_position) != position)
465 			goto err;
466 
467 		handle = be32_to_cpu(ch->rc_target.rs_handle),
468 		byte_count = be32_to_cpu(ch->rc_target.rs_length);
469 		xdr_decode_hyper((__be32 *)&ch->rc_target.rs_offset,
470 				 &rs_offset);
471 
472 		while (byte_count > 0) {
473 			last = (ch + 1)->rc_discrim == xdr_zero;
474 			ret = xprt->sc_reader(xprt, rqstp, head,
475 					      &page_no, &page_offset,
476 					      handle, byte_count,
477 					      rs_offset, last);
478 			if (ret < 0)
479 				goto err;
480 			byte_count -= ret;
481 			rs_offset += ret;
482 			head->arg.buflen += ret;
483 		}
484 	}
485 
486 	/* Read list may need XDR round-up (see RFC 5666, s. 3.7) */
487 	if (page_offset & 3) {
488 		u32 pad = 4 - (page_offset & 3);
489 
490 		head->arg.page_len += pad;
491 		head->arg.len += pad;
492 		head->arg.buflen += pad;
493 		page_offset += pad;
494 	}
495 
496 	ret = 1;
497 	if (position && position < head->arg.head[0].iov_len)
498 		ret = rdma_copy_tail(rqstp, head, position,
499 				     byte_count, page_offset, page_no);
500 	head->arg.head[0].iov_len = position;
501 	head->position = position;
502 
503  err:
504 	/* Detach arg pages. svc_recv will replenish them */
505 	for (page_no = 0;
506 	     &rqstp->rq_pages[page_no] < rqstp->rq_respages; page_no++)
507 		rqstp->rq_pages[page_no] = NULL;
508 
509 	return ret;
510 }
511 
512 static int rdma_read_complete(struct svc_rqst *rqstp,
513 			      struct svc_rdma_op_ctxt *head)
514 {
515 	int page_no;
516 	int ret;
517 
518 	/* Copy RPC pages */
519 	for (page_no = 0; page_no < head->count; page_no++) {
520 		put_page(rqstp->rq_pages[page_no]);
521 		rqstp->rq_pages[page_no] = head->pages[page_no];
522 	}
523 
524 	/* Adjustments made for RDMA_NOMSG type requests */
525 	if (head->position == 0) {
526 		if (head->arg.len <= head->sge[0].length) {
527 			head->arg.head[0].iov_len = head->arg.len -
528 							head->byte_len;
529 			head->arg.page_len = 0;
530 		} else {
531 			head->arg.head[0].iov_len = head->sge[0].length -
532 								head->byte_len;
533 			head->arg.page_len = head->arg.len -
534 						head->sge[0].length;
535 		}
536 	}
537 
538 	/* Point rq_arg.pages past header */
539 	rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count];
540 	rqstp->rq_arg.page_len = head->arg.page_len;
541 	rqstp->rq_arg.page_base = head->arg.page_base;
542 
543 	/* rq_respages starts after the last arg page */
544 	rqstp->rq_respages = &rqstp->rq_arg.pages[page_no];
545 	rqstp->rq_next_page = rqstp->rq_respages + 1;
546 
547 	/* Rebuild rq_arg head and tail. */
548 	rqstp->rq_arg.head[0] = head->arg.head[0];
549 	rqstp->rq_arg.tail[0] = head->arg.tail[0];
550 	rqstp->rq_arg.len = head->arg.len;
551 	rqstp->rq_arg.buflen = head->arg.buflen;
552 
553 	/* Free the context */
554 	svc_rdma_put_context(head, 0);
555 
556 	/* XXX: What should this be? */
557 	rqstp->rq_prot = IPPROTO_MAX;
558 	svc_xprt_copy_addrs(rqstp, rqstp->rq_xprt);
559 
560 	ret = rqstp->rq_arg.head[0].iov_len
561 		+ rqstp->rq_arg.page_len
562 		+ rqstp->rq_arg.tail[0].iov_len;
563 	dprintk("svcrdma: deferred read ret=%d, rq_arg.len=%u, "
564 		"rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zu\n",
565 		ret, rqstp->rq_arg.len,	rqstp->rq_arg.head[0].iov_base,
566 		rqstp->rq_arg.head[0].iov_len);
567 
568 	return ret;
569 }
570 
571 /*
572  * Set up the rqstp thread context to point to the RQ buffer. If
573  * necessary, pull additional data from the client with an RDMA_READ
574  * request.
575  */
576 int svc_rdma_recvfrom(struct svc_rqst *rqstp)
577 {
578 	struct svc_xprt *xprt = rqstp->rq_xprt;
579 	struct svcxprt_rdma *rdma_xprt =
580 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
581 	struct svc_rdma_op_ctxt *ctxt = NULL;
582 	struct rpcrdma_msg *rmsgp;
583 	int ret = 0;
584 	int len;
585 
586 	dprintk("svcrdma: rqstp=%p\n", rqstp);
587 
588 	spin_lock_bh(&rdma_xprt->sc_rq_dto_lock);
589 	if (!list_empty(&rdma_xprt->sc_read_complete_q)) {
590 		ctxt = list_entry(rdma_xprt->sc_read_complete_q.next,
591 				  struct svc_rdma_op_ctxt,
592 				  dto_q);
593 		list_del_init(&ctxt->dto_q);
594 		spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
595 		return rdma_read_complete(rqstp, ctxt);
596 	} else if (!list_empty(&rdma_xprt->sc_rq_dto_q)) {
597 		ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next,
598 				  struct svc_rdma_op_ctxt,
599 				  dto_q);
600 		list_del_init(&ctxt->dto_q);
601 	} else {
602 		atomic_inc(&rdma_stat_rq_starve);
603 		clear_bit(XPT_DATA, &xprt->xpt_flags);
604 		ctxt = NULL;
605 	}
606 	spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
607 	if (!ctxt) {
608 		/* This is the EAGAIN path. The svc_recv routine will
609 		 * return -EAGAIN, the nfsd thread will go to call into
610 		 * svc_recv again and we shouldn't be on the active
611 		 * transport list
612 		 */
613 		if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
614 			goto close_out;
615 
616 		goto out;
617 	}
618 	dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n",
619 		ctxt, rdma_xprt, rqstp, ctxt->wc_status);
620 	atomic_inc(&rdma_stat_recv);
621 
622 	/* Build up the XDR from the receive buffers. */
623 	rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len);
624 
625 	/* Decode the RDMA header. */
626 	len = svc_rdma_xdr_decode_req(&rmsgp, rqstp);
627 	rqstp->rq_xprt_hlen = len;
628 
629 	/* If the request is invalid, reply with an error */
630 	if (len < 0) {
631 		if (len == -ENOSYS)
632 			svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS);
633 		goto close_out;
634 	}
635 
636 	/* Read read-list data. */
637 	ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt);
638 	if (ret > 0) {
639 		/* read-list posted, defer until data received from client. */
640 		goto defer;
641 	} else if (ret < 0) {
642 		/* Post of read-list failed, free context. */
643 		svc_rdma_put_context(ctxt, 1);
644 		return 0;
645 	}
646 
647 	ret = rqstp->rq_arg.head[0].iov_len
648 		+ rqstp->rq_arg.page_len
649 		+ rqstp->rq_arg.tail[0].iov_len;
650 	svc_rdma_put_context(ctxt, 0);
651  out:
652 	dprintk("svcrdma: ret=%d, rq_arg.len=%u, "
653 		"rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zd\n",
654 		ret, rqstp->rq_arg.len,
655 		rqstp->rq_arg.head[0].iov_base,
656 		rqstp->rq_arg.head[0].iov_len);
657 	rqstp->rq_prot = IPPROTO_MAX;
658 	svc_xprt_copy_addrs(rqstp, xprt);
659 	return ret;
660 
661  close_out:
662 	if (ctxt)
663 		svc_rdma_put_context(ctxt, 1);
664 	dprintk("svcrdma: transport %p is closing\n", xprt);
665 	/*
666 	 * Set the close bit and enqueue it. svc_recv will see the
667 	 * close bit and call svc_xprt_delete
668 	 */
669 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
670 defer:
671 	return 0;
672 }
673