1 /*
2  * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the BSD-type
8  * license below:
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  *      Redistributions of source code must retain the above copyright
15  *      notice, this list of conditions and the following disclaimer.
16  *
17  *      Redistributions in binary form must reproduce the above
18  *      copyright notice, this list of conditions and the following
19  *      disclaimer in the documentation and/or other materials provided
20  *      with the distribution.
21  *
22  *      Neither the name of the Network Appliance, Inc. nor the names of
23  *      its contributors may be used to endorse or promote products
24  *      derived from this software without specific prior written
25  *      permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
32  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
33  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
34  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
35  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38  *
39  * Author: Tom Tucker <tom@opengridcomputing.com>
40  */
41 
42 #include <linux/sunrpc/debug.h>
43 #include <linux/sunrpc/rpc_rdma.h>
44 #include <linux/spinlock.h>
45 #include <asm/unaligned.h>
46 #include <rdma/ib_verbs.h>
47 #include <rdma/rdma_cm.h>
48 #include <linux/sunrpc/svc_rdma.h>
49 
50 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
51 
52 /* Encode an XDR as an array of IB SGE
53  *
54  * Assumptions:
55  * - head[0] is physically contiguous.
56  * - tail[0] is physically contiguous.
57  * - pages[] is not physically or virtually contigous and consists of
58  *   PAGE_SIZE elements.
59  *
60  * Output:
61  * SGE[0]              reserved for RCPRDMA header
62  * SGE[1]              data from xdr->head[]
63  * SGE[2..sge_count-2] data from xdr->pages[]
64  * SGE[sge_count-1]    data from xdr->tail.
65  *
66  * The max SGE we need is the length of the XDR / pagesize + one for
67  * head + one for tail + one for RPCRDMA header. Since RPCSVC_MAXPAGES
68  * reserves a page for both the request and the reply header, and this
69  * array is only concerned with the reply we are assured that we have
70  * on extra page for the RPCRMDA header.
71  */
72 static void xdr_to_sge(struct svcxprt_rdma *xprt,
73 		       struct xdr_buf *xdr,
74 		       struct svc_rdma_req_map *vec)
75 {
76 	int sge_max = (xdr->len+PAGE_SIZE-1) / PAGE_SIZE + 3;
77 	int sge_no;
78 	u32 sge_bytes;
79 	u32 page_bytes;
80 	u32 page_off;
81 	int page_no;
82 
83 	BUG_ON(xdr->len !=
84 	       (xdr->head[0].iov_len + xdr->page_len + xdr->tail[0].iov_len));
85 
86 	/* Skip the first sge, this is for the RPCRDMA header */
87 	sge_no = 1;
88 
89 	/* Head SGE */
90 	vec->sge[sge_no].iov_base = xdr->head[0].iov_base;
91 	vec->sge[sge_no].iov_len = xdr->head[0].iov_len;
92 	sge_no++;
93 
94 	/* pages SGE */
95 	page_no = 0;
96 	page_bytes = xdr->page_len;
97 	page_off = xdr->page_base;
98 	while (page_bytes) {
99 		vec->sge[sge_no].iov_base =
100 			page_address(xdr->pages[page_no]) + page_off;
101 		sge_bytes = min_t(u32, page_bytes, (PAGE_SIZE - page_off));
102 		page_bytes -= sge_bytes;
103 		vec->sge[sge_no].iov_len = sge_bytes;
104 
105 		sge_no++;
106 		page_no++;
107 		page_off = 0; /* reset for next time through loop */
108 	}
109 
110 	/* Tail SGE */
111 	if (xdr->tail[0].iov_len) {
112 		vec->sge[sge_no].iov_base = xdr->tail[0].iov_base;
113 		vec->sge[sge_no].iov_len = xdr->tail[0].iov_len;
114 		sge_no++;
115 	}
116 
117 	BUG_ON(sge_no > sge_max);
118 	vec->count = sge_no;
119 }
120 
121 /* Assumptions:
122  * - The specified write_len can be represented in sc_max_sge * PAGE_SIZE
123  */
124 static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
125 		      u32 rmr, u64 to,
126 		      u32 xdr_off, int write_len,
127 		      struct svc_rdma_req_map *vec)
128 {
129 	struct ib_send_wr write_wr;
130 	struct ib_sge *sge;
131 	int xdr_sge_no;
132 	int sge_no;
133 	int sge_bytes;
134 	int sge_off;
135 	int bc;
136 	struct svc_rdma_op_ctxt *ctxt;
137 
138 	BUG_ON(vec->count > RPCSVC_MAXPAGES);
139 	dprintk("svcrdma: RDMA_WRITE rmr=%x, to=%llx, xdr_off=%d, "
140 		"write_len=%d, vec->sge=%p, vec->count=%lu\n",
141 		rmr, (unsigned long long)to, xdr_off,
142 		write_len, vec->sge, vec->count);
143 
144 	ctxt = svc_rdma_get_context(xprt);
145 	ctxt->direction = DMA_TO_DEVICE;
146 	sge = ctxt->sge;
147 
148 	/* Find the SGE associated with xdr_off */
149 	for (bc = xdr_off, xdr_sge_no = 1; bc && xdr_sge_no < vec->count;
150 	     xdr_sge_no++) {
151 		if (vec->sge[xdr_sge_no].iov_len > bc)
152 			break;
153 		bc -= vec->sge[xdr_sge_no].iov_len;
154 	}
155 
156 	sge_off = bc;
157 	bc = write_len;
158 	sge_no = 0;
159 
160 	/* Copy the remaining SGE */
161 	while (bc != 0 && xdr_sge_no < vec->count) {
162 		sge[sge_no].lkey = xprt->sc_phys_mr->lkey;
163 		sge_bytes = min((size_t)bc,
164 				(size_t)(vec->sge[xdr_sge_no].iov_len-sge_off));
165 		sge[sge_no].length = sge_bytes;
166 		atomic_inc(&xprt->sc_dma_used);
167 		sge[sge_no].addr =
168 			ib_dma_map_single(xprt->sc_cm_id->device,
169 					  (void *)
170 					  vec->sge[xdr_sge_no].iov_base + sge_off,
171 					  sge_bytes, DMA_TO_DEVICE);
172 		if (dma_mapping_error(xprt->sc_cm_id->device->dma_device,
173 					sge[sge_no].addr))
174 			goto err;
175 		sge_off = 0;
176 		sge_no++;
177 		ctxt->count++;
178 		xdr_sge_no++;
179 		bc -= sge_bytes;
180 	}
181 
182 	BUG_ON(bc != 0);
183 	BUG_ON(xdr_sge_no > vec->count);
184 
185 	/* Prepare WRITE WR */
186 	memset(&write_wr, 0, sizeof write_wr);
187 	ctxt->wr_op = IB_WR_RDMA_WRITE;
188 	write_wr.wr_id = (unsigned long)ctxt;
189 	write_wr.sg_list = &sge[0];
190 	write_wr.num_sge = sge_no;
191 	write_wr.opcode = IB_WR_RDMA_WRITE;
192 	write_wr.send_flags = IB_SEND_SIGNALED;
193 	write_wr.wr.rdma.rkey = rmr;
194 	write_wr.wr.rdma.remote_addr = to;
195 
196 	/* Post It */
197 	atomic_inc(&rdma_stat_write);
198 	if (svc_rdma_send(xprt, &write_wr))
199 		goto err;
200 	return 0;
201  err:
202 	svc_rdma_put_context(ctxt, 0);
203 	/* Fatal error, close transport */
204 	return -EIO;
205 }
206 
207 static int send_write_chunks(struct svcxprt_rdma *xprt,
208 			     struct rpcrdma_msg *rdma_argp,
209 			     struct rpcrdma_msg *rdma_resp,
210 			     struct svc_rqst *rqstp,
211 			     struct svc_rdma_req_map *vec)
212 {
213 	u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
214 	int write_len;
215 	int max_write;
216 	u32 xdr_off;
217 	int chunk_off;
218 	int chunk_no;
219 	struct rpcrdma_write_array *arg_ary;
220 	struct rpcrdma_write_array *res_ary;
221 	int ret;
222 
223 	arg_ary = svc_rdma_get_write_array(rdma_argp);
224 	if (!arg_ary)
225 		return 0;
226 	res_ary = (struct rpcrdma_write_array *)
227 		&rdma_resp->rm_body.rm_chunks[1];
228 
229 	max_write = xprt->sc_max_sge * PAGE_SIZE;
230 
231 	/* Write chunks start at the pagelist */
232 	for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0;
233 	     xfer_len && chunk_no < arg_ary->wc_nchunks;
234 	     chunk_no++) {
235 		struct rpcrdma_segment *arg_ch;
236 		u64 rs_offset;
237 
238 		arg_ch = &arg_ary->wc_array[chunk_no].wc_target;
239 		write_len = min(xfer_len, arg_ch->rs_length);
240 
241 		/* Prepare the response chunk given the length actually
242 		 * written */
243 		rs_offset = get_unaligned(&(arg_ch->rs_offset));
244 		svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no,
245 					    arg_ch->rs_handle,
246 					    rs_offset,
247 					    write_len);
248 		chunk_off = 0;
249 		while (write_len) {
250 			int this_write;
251 			this_write = min(write_len, max_write);
252 			ret = send_write(xprt, rqstp,
253 					 arg_ch->rs_handle,
254 					 rs_offset + chunk_off,
255 					 xdr_off,
256 					 this_write,
257 					 vec);
258 			if (ret) {
259 				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
260 					ret);
261 				return -EIO;
262 			}
263 			chunk_off += this_write;
264 			xdr_off += this_write;
265 			xfer_len -= this_write;
266 			write_len -= this_write;
267 		}
268 	}
269 	/* Update the req with the number of chunks actually used */
270 	svc_rdma_xdr_encode_write_list(rdma_resp, chunk_no);
271 
272 	return rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len;
273 }
274 
275 static int send_reply_chunks(struct svcxprt_rdma *xprt,
276 			     struct rpcrdma_msg *rdma_argp,
277 			     struct rpcrdma_msg *rdma_resp,
278 			     struct svc_rqst *rqstp,
279 			     struct svc_rdma_req_map *vec)
280 {
281 	u32 xfer_len = rqstp->rq_res.len;
282 	int write_len;
283 	int max_write;
284 	u32 xdr_off;
285 	int chunk_no;
286 	int chunk_off;
287 	struct rpcrdma_segment *ch;
288 	struct rpcrdma_write_array *arg_ary;
289 	struct rpcrdma_write_array *res_ary;
290 	int ret;
291 
292 	arg_ary = svc_rdma_get_reply_array(rdma_argp);
293 	if (!arg_ary)
294 		return 0;
295 	/* XXX: need to fix when reply lists occur with read-list and or
296 	 * write-list */
297 	res_ary = (struct rpcrdma_write_array *)
298 		&rdma_resp->rm_body.rm_chunks[2];
299 
300 	max_write = xprt->sc_max_sge * PAGE_SIZE;
301 
302 	/* xdr offset starts at RPC message */
303 	for (xdr_off = 0, chunk_no = 0;
304 	     xfer_len && chunk_no < arg_ary->wc_nchunks;
305 	     chunk_no++) {
306 		u64 rs_offset;
307 		ch = &arg_ary->wc_array[chunk_no].wc_target;
308 		write_len = min(xfer_len, ch->rs_length);
309 
310 
311 		/* Prepare the reply chunk given the length actually
312 		 * written */
313 		rs_offset = get_unaligned(&(ch->rs_offset));
314 		svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no,
315 					    ch->rs_handle, rs_offset,
316 					    write_len);
317 		chunk_off = 0;
318 		while (write_len) {
319 			int this_write;
320 
321 			this_write = min(write_len, max_write);
322 			ret = send_write(xprt, rqstp,
323 					 ch->rs_handle,
324 					 rs_offset + chunk_off,
325 					 xdr_off,
326 					 this_write,
327 					 vec);
328 			if (ret) {
329 				dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n",
330 					ret);
331 				return -EIO;
332 			}
333 			chunk_off += this_write;
334 			xdr_off += this_write;
335 			xfer_len -= this_write;
336 			write_len -= this_write;
337 		}
338 	}
339 	/* Update the req with the number of chunks actually used */
340 	svc_rdma_xdr_encode_reply_array(res_ary, chunk_no);
341 
342 	return rqstp->rq_res.len;
343 }
344 
345 /* This function prepares the portion of the RPCRDMA message to be
346  * sent in the RDMA_SEND. This function is called after data sent via
347  * RDMA has already been transmitted. There are three cases:
348  * - The RPCRDMA header, RPC header, and payload are all sent in a
349  *   single RDMA_SEND. This is the "inline" case.
350  * - The RPCRDMA header and some portion of the RPC header and data
351  *   are sent via this RDMA_SEND and another portion of the data is
352  *   sent via RDMA.
353  * - The RPCRDMA header [NOMSG] is sent in this RDMA_SEND and the RPC
354  *   header and data are all transmitted via RDMA.
355  * In all three cases, this function prepares the RPCRDMA header in
356  * sge[0], the 'type' parameter indicates the type to place in the
357  * RPCRDMA header, and the 'byte_count' field indicates how much of
358  * the XDR to include in this RDMA_SEND.
359  */
360 static int send_reply(struct svcxprt_rdma *rdma,
361 		      struct svc_rqst *rqstp,
362 		      struct page *page,
363 		      struct rpcrdma_msg *rdma_resp,
364 		      struct svc_rdma_op_ctxt *ctxt,
365 		      struct svc_rdma_req_map *vec,
366 		      int byte_count)
367 {
368 	struct ib_send_wr send_wr;
369 	int sge_no;
370 	int sge_bytes;
371 	int page_no;
372 	int ret;
373 
374 	/* Post a recv buffer to handle another request. */
375 	ret = svc_rdma_post_recv(rdma);
376 	if (ret) {
377 		printk(KERN_INFO
378 		       "svcrdma: could not post a receive buffer, err=%d."
379 		       "Closing transport %p.\n", ret, rdma);
380 		set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
381 		svc_rdma_put_context(ctxt, 0);
382 		return -ENOTCONN;
383 	}
384 
385 	/* Prepare the context */
386 	ctxt->pages[0] = page;
387 	ctxt->count = 1;
388 
389 	/* Prepare the SGE for the RPCRDMA Header */
390 	atomic_inc(&rdma->sc_dma_used);
391 	ctxt->sge[0].addr =
392 		ib_dma_map_page(rdma->sc_cm_id->device,
393 				page, 0, PAGE_SIZE, DMA_TO_DEVICE);
394 	ctxt->direction = DMA_TO_DEVICE;
395 	ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp);
396 	ctxt->sge[0].lkey = rdma->sc_phys_mr->lkey;
397 
398 	/* Determine how many of our SGE are to be transmitted */
399 	for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) {
400 		sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count);
401 		byte_count -= sge_bytes;
402 		atomic_inc(&rdma->sc_dma_used);
403 		ctxt->sge[sge_no].addr =
404 			ib_dma_map_single(rdma->sc_cm_id->device,
405 					  vec->sge[sge_no].iov_base,
406 					  sge_bytes, DMA_TO_DEVICE);
407 		ctxt->sge[sge_no].length = sge_bytes;
408 		ctxt->sge[sge_no].lkey = rdma->sc_phys_mr->lkey;
409 	}
410 	BUG_ON(byte_count != 0);
411 
412 	/* Save all respages in the ctxt and remove them from the
413 	 * respages array. They are our pages until the I/O
414 	 * completes.
415 	 */
416 	for (page_no = 0; page_no < rqstp->rq_resused; page_no++) {
417 		ctxt->pages[page_no+1] = rqstp->rq_respages[page_no];
418 		ctxt->count++;
419 		rqstp->rq_respages[page_no] = NULL;
420 		/* If there are more pages than SGE, terminate SGE list */
421 		if (page_no+1 >= sge_no)
422 			ctxt->sge[page_no+1].length = 0;
423 	}
424 	BUG_ON(sge_no > rdma->sc_max_sge);
425 	memset(&send_wr, 0, sizeof send_wr);
426 	ctxt->wr_op = IB_WR_SEND;
427 	send_wr.wr_id = (unsigned long)ctxt;
428 	send_wr.sg_list = ctxt->sge;
429 	send_wr.num_sge = sge_no;
430 	send_wr.opcode = IB_WR_SEND;
431 	send_wr.send_flags =  IB_SEND_SIGNALED;
432 
433 	ret = svc_rdma_send(rdma, &send_wr);
434 	if (ret)
435 		svc_rdma_put_context(ctxt, 1);
436 
437 	return ret;
438 }
439 
440 void svc_rdma_prep_reply_hdr(struct svc_rqst *rqstp)
441 {
442 }
443 
444 /*
445  * Return the start of an xdr buffer.
446  */
447 static void *xdr_start(struct xdr_buf *xdr)
448 {
449 	return xdr->head[0].iov_base -
450 		(xdr->len -
451 		 xdr->page_len -
452 		 xdr->tail[0].iov_len -
453 		 xdr->head[0].iov_len);
454 }
455 
456 int svc_rdma_sendto(struct svc_rqst *rqstp)
457 {
458 	struct svc_xprt *xprt = rqstp->rq_xprt;
459 	struct svcxprt_rdma *rdma =
460 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
461 	struct rpcrdma_msg *rdma_argp;
462 	struct rpcrdma_msg *rdma_resp;
463 	struct rpcrdma_write_array *reply_ary;
464 	enum rpcrdma_proc reply_type;
465 	int ret;
466 	int inline_bytes;
467 	struct page *res_page;
468 	struct svc_rdma_op_ctxt *ctxt;
469 	struct svc_rdma_req_map *vec;
470 
471 	dprintk("svcrdma: sending response for rqstp=%p\n", rqstp);
472 
473 	/* Get the RDMA request header. */
474 	rdma_argp = xdr_start(&rqstp->rq_arg);
475 
476 	/* Build an req vec for the XDR */
477 	ctxt = svc_rdma_get_context(rdma);
478 	ctxt->direction = DMA_TO_DEVICE;
479 	vec = svc_rdma_get_req_map();
480 	xdr_to_sge(rdma, &rqstp->rq_res, vec);
481 
482 	inline_bytes = rqstp->rq_res.len;
483 
484 	/* Create the RDMA response header */
485 	res_page = svc_rdma_get_page();
486 	rdma_resp = page_address(res_page);
487 	reply_ary = svc_rdma_get_reply_array(rdma_argp);
488 	if (reply_ary)
489 		reply_type = RDMA_NOMSG;
490 	else
491 		reply_type = RDMA_MSG;
492 	svc_rdma_xdr_encode_reply_header(rdma, rdma_argp,
493 					 rdma_resp, reply_type);
494 
495 	/* Send any write-chunk data and build resp write-list */
496 	ret = send_write_chunks(rdma, rdma_argp, rdma_resp,
497 				rqstp, vec);
498 	if (ret < 0) {
499 		printk(KERN_ERR "svcrdma: failed to send write chunks, rc=%d\n",
500 		       ret);
501 		goto error;
502 	}
503 	inline_bytes -= ret;
504 
505 	/* Send any reply-list data and update resp reply-list */
506 	ret = send_reply_chunks(rdma, rdma_argp, rdma_resp,
507 				rqstp, vec);
508 	if (ret < 0) {
509 		printk(KERN_ERR "svcrdma: failed to send reply chunks, rc=%d\n",
510 		       ret);
511 		goto error;
512 	}
513 	inline_bytes -= ret;
514 
515 	ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec,
516 			 inline_bytes);
517 	svc_rdma_put_req_map(vec);
518 	dprintk("svcrdma: send_reply returns %d\n", ret);
519 	return ret;
520  error:
521 	svc_rdma_put_req_map(vec);
522 	svc_rdma_put_context(ctxt, 0);
523 	put_page(res_page);
524 	return ret;
525 }
526