1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2016-2018 Oracle.  All rights reserved.
4  *
5  * Use the core R/W API to move RPC-over-RDMA Read and Write chunks.
6  */
7 
8 #include <rdma/rw.h>
9 
10 #include <linux/sunrpc/xdr.h>
11 #include <linux/sunrpc/rpc_rdma.h>
12 #include <linux/sunrpc/svc_rdma.h>
13 
14 #include "xprt_rdma.h"
15 #include <trace/events/rpcrdma.h>
16 
17 static void svc_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc);
18 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc);
19 
20 /* Each R/W context contains state for one chain of RDMA Read or
21  * Write Work Requests.
22  *
23  * Each WR chain handles a single contiguous server-side buffer,
24  * because scatterlist entries after the first have to start on
25  * page alignment. xdr_buf iovecs cannot guarantee alignment.
26  *
27  * Each WR chain handles only one R_key. Each RPC-over-RDMA segment
28  * from a client may contain a unique R_key, so each WR chain moves
29  * up to one segment at a time.
30  *
31  * The scatterlist makes this data structure over 4KB in size. To
32  * make it less likely to fail, and to handle the allocation for
33  * smaller I/O requests without disabling bottom-halves, these
34  * contexts are created on demand, but cached and reused until the
35  * controlling svcxprt_rdma is destroyed.
36  */
37 struct svc_rdma_rw_ctxt {
38 	struct list_head	rw_list;
39 	struct rdma_rw_ctx	rw_ctx;
40 	unsigned int		rw_nents;
41 	struct sg_table		rw_sg_table;
42 	struct scatterlist	rw_first_sgl[];
43 };
44 
45 static inline struct svc_rdma_rw_ctxt *
46 svc_rdma_next_ctxt(struct list_head *list)
47 {
48 	return list_first_entry_or_null(list, struct svc_rdma_rw_ctxt,
49 					rw_list);
50 }
51 
52 static struct svc_rdma_rw_ctxt *
53 svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges)
54 {
55 	struct svc_rdma_rw_ctxt *ctxt;
56 
57 	spin_lock(&rdma->sc_rw_ctxt_lock);
58 
59 	ctxt = svc_rdma_next_ctxt(&rdma->sc_rw_ctxts);
60 	if (ctxt) {
61 		list_del(&ctxt->rw_list);
62 		spin_unlock(&rdma->sc_rw_ctxt_lock);
63 	} else {
64 		spin_unlock(&rdma->sc_rw_ctxt_lock);
65 		ctxt = kmalloc(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE),
66 			       GFP_KERNEL);
67 		if (!ctxt)
68 			goto out_noctx;
69 		INIT_LIST_HEAD(&ctxt->rw_list);
70 	}
71 
72 	ctxt->rw_sg_table.sgl = ctxt->rw_first_sgl;
73 	if (sg_alloc_table_chained(&ctxt->rw_sg_table, sges,
74 				   ctxt->rw_sg_table.sgl,
75 				   SG_CHUNK_SIZE))
76 		goto out_free;
77 	return ctxt;
78 
79 out_free:
80 	kfree(ctxt);
81 out_noctx:
82 	trace_svcrdma_no_rwctx_err(rdma, sges);
83 	return NULL;
84 }
85 
86 static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
87 				 struct svc_rdma_rw_ctxt *ctxt)
88 {
89 	sg_free_table_chained(&ctxt->rw_sg_table, SG_CHUNK_SIZE);
90 
91 	spin_lock(&rdma->sc_rw_ctxt_lock);
92 	list_add(&ctxt->rw_list, &rdma->sc_rw_ctxts);
93 	spin_unlock(&rdma->sc_rw_ctxt_lock);
94 }
95 
96 /**
97  * svc_rdma_destroy_rw_ctxts - Free accumulated R/W contexts
98  * @rdma: transport about to be destroyed
99  *
100  */
101 void svc_rdma_destroy_rw_ctxts(struct svcxprt_rdma *rdma)
102 {
103 	struct svc_rdma_rw_ctxt *ctxt;
104 
105 	while ((ctxt = svc_rdma_next_ctxt(&rdma->sc_rw_ctxts)) != NULL) {
106 		list_del(&ctxt->rw_list);
107 		kfree(ctxt);
108 	}
109 }
110 
111 /**
112  * svc_rdma_rw_ctx_init - Prepare a R/W context for I/O
113  * @rdma: controlling transport instance
114  * @ctxt: R/W context to prepare
115  * @offset: RDMA offset
116  * @handle: RDMA tag/handle
117  * @direction: I/O direction
118  *
119  * Returns on success, the number of WQEs that will be needed
120  * on the workqueue, or a negative errno.
121  */
122 static int svc_rdma_rw_ctx_init(struct svcxprt_rdma *rdma,
123 				struct svc_rdma_rw_ctxt *ctxt,
124 				u64 offset, u32 handle,
125 				enum dma_data_direction direction)
126 {
127 	int ret;
128 
129 	ret = rdma_rw_ctx_init(&ctxt->rw_ctx, rdma->sc_qp, rdma->sc_port_num,
130 			       ctxt->rw_sg_table.sgl, ctxt->rw_nents,
131 			       0, offset, handle, direction);
132 	if (unlikely(ret < 0)) {
133 		svc_rdma_put_rw_ctxt(rdma, ctxt);
134 		trace_svcrdma_dma_map_rw_err(rdma, ctxt->rw_nents, ret);
135 	}
136 	return ret;
137 }
138 
139 /* A chunk context tracks all I/O for moving one Read or Write
140  * chunk. This is a set of rdma_rw's that handle data movement
141  * for all segments of one chunk.
142  *
143  * These are small, acquired with a single allocator call, and
144  * no more than one is needed per chunk. They are allocated on
145  * demand, and not cached.
146  */
147 struct svc_rdma_chunk_ctxt {
148 	struct rpc_rdma_cid	cc_cid;
149 	struct ib_cqe		cc_cqe;
150 	struct svcxprt_rdma	*cc_rdma;
151 	struct list_head	cc_rwctxts;
152 	int			cc_sqecount;
153 };
154 
155 static void svc_rdma_cc_cid_init(struct svcxprt_rdma *rdma,
156 				 struct rpc_rdma_cid *cid)
157 {
158 	cid->ci_queue_id = rdma->sc_sq_cq->res.id;
159 	cid->ci_completion_id = atomic_inc_return(&rdma->sc_completion_ids);
160 }
161 
162 static void svc_rdma_cc_init(struct svcxprt_rdma *rdma,
163 			     struct svc_rdma_chunk_ctxt *cc)
164 {
165 	svc_rdma_cc_cid_init(rdma, &cc->cc_cid);
166 	cc->cc_rdma = rdma;
167 
168 	INIT_LIST_HEAD(&cc->cc_rwctxts);
169 	cc->cc_sqecount = 0;
170 }
171 
172 static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc,
173 				enum dma_data_direction dir)
174 {
175 	struct svcxprt_rdma *rdma = cc->cc_rdma;
176 	struct svc_rdma_rw_ctxt *ctxt;
177 
178 	while ((ctxt = svc_rdma_next_ctxt(&cc->cc_rwctxts)) != NULL) {
179 		list_del(&ctxt->rw_list);
180 
181 		rdma_rw_ctx_destroy(&ctxt->rw_ctx, rdma->sc_qp,
182 				    rdma->sc_port_num, ctxt->rw_sg_table.sgl,
183 				    ctxt->rw_nents, dir);
184 		svc_rdma_put_rw_ctxt(rdma, ctxt);
185 	}
186 }
187 
188 /* State for sending a Write or Reply chunk.
189  *  - Tracks progress of writing one chunk over all its segments
190  *  - Stores arguments for the SGL constructor functions
191  */
192 struct svc_rdma_write_info {
193 	const struct svc_rdma_chunk	*wi_chunk;
194 
195 	/* write state of this chunk */
196 	unsigned int		wi_seg_off;
197 	unsigned int		wi_seg_no;
198 
199 	/* SGL constructor arguments */
200 	const struct xdr_buf	*wi_xdr;
201 	unsigned char		*wi_base;
202 	unsigned int		wi_next_off;
203 
204 	struct svc_rdma_chunk_ctxt	wi_cc;
205 };
206 
207 static struct svc_rdma_write_info *
208 svc_rdma_write_info_alloc(struct svcxprt_rdma *rdma,
209 			  const struct svc_rdma_chunk *chunk)
210 {
211 	struct svc_rdma_write_info *info;
212 
213 	info = kmalloc(sizeof(*info), GFP_KERNEL);
214 	if (!info)
215 		return info;
216 
217 	info->wi_chunk = chunk;
218 	info->wi_seg_off = 0;
219 	info->wi_seg_no = 0;
220 	svc_rdma_cc_init(rdma, &info->wi_cc);
221 	info->wi_cc.cc_cqe.done = svc_rdma_write_done;
222 	return info;
223 }
224 
225 static void svc_rdma_write_info_free(struct svc_rdma_write_info *info)
226 {
227 	svc_rdma_cc_release(&info->wi_cc, DMA_TO_DEVICE);
228 	kfree(info);
229 }
230 
231 /**
232  * svc_rdma_write_done - Write chunk completion
233  * @cq: controlling Completion Queue
234  * @wc: Work Completion
235  *
236  * Pages under I/O are freed by a subsequent Send completion.
237  */
238 static void svc_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
239 {
240 	struct ib_cqe *cqe = wc->wr_cqe;
241 	struct svc_rdma_chunk_ctxt *cc =
242 			container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe);
243 	struct svcxprt_rdma *rdma = cc->cc_rdma;
244 	struct svc_rdma_write_info *info =
245 			container_of(cc, struct svc_rdma_write_info, wi_cc);
246 
247 	trace_svcrdma_wc_write(wc, &cc->cc_cid);
248 
249 	atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
250 	wake_up(&rdma->sc_send_wait);
251 
252 	if (unlikely(wc->status != IB_WC_SUCCESS))
253 		set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
254 
255 	svc_rdma_write_info_free(info);
256 }
257 
258 /* State for pulling a Read chunk.
259  */
260 struct svc_rdma_read_info {
261 	struct svc_rqst			*ri_rqst;
262 	struct svc_rdma_recv_ctxt	*ri_readctxt;
263 	unsigned int			ri_pageno;
264 	unsigned int			ri_pageoff;
265 	unsigned int			ri_totalbytes;
266 
267 	struct svc_rdma_chunk_ctxt	ri_cc;
268 };
269 
270 static struct svc_rdma_read_info *
271 svc_rdma_read_info_alloc(struct svcxprt_rdma *rdma)
272 {
273 	struct svc_rdma_read_info *info;
274 
275 	info = kmalloc(sizeof(*info), GFP_KERNEL);
276 	if (!info)
277 		return info;
278 
279 	svc_rdma_cc_init(rdma, &info->ri_cc);
280 	info->ri_cc.cc_cqe.done = svc_rdma_wc_read_done;
281 	return info;
282 }
283 
284 static void svc_rdma_read_info_free(struct svc_rdma_read_info *info)
285 {
286 	svc_rdma_cc_release(&info->ri_cc, DMA_FROM_DEVICE);
287 	kfree(info);
288 }
289 
290 /**
291  * svc_rdma_wc_read_done - Handle completion of an RDMA Read ctx
292  * @cq: controlling Completion Queue
293  * @wc: Work Completion
294  *
295  */
296 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc)
297 {
298 	struct ib_cqe *cqe = wc->wr_cqe;
299 	struct svc_rdma_chunk_ctxt *cc =
300 			container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe);
301 	struct svcxprt_rdma *rdma = cc->cc_rdma;
302 	struct svc_rdma_read_info *info =
303 			container_of(cc, struct svc_rdma_read_info, ri_cc);
304 
305 	trace_svcrdma_wc_read(wc, &cc->cc_cid);
306 
307 	atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
308 	wake_up(&rdma->sc_send_wait);
309 
310 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
311 		set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
312 		svc_rdma_recv_ctxt_put(rdma, info->ri_readctxt);
313 	} else {
314 		spin_lock(&rdma->sc_rq_dto_lock);
315 		list_add_tail(&info->ri_readctxt->rc_list,
316 			      &rdma->sc_read_complete_q);
317 		/* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */
318 		set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags);
319 		spin_unlock(&rdma->sc_rq_dto_lock);
320 
321 		svc_xprt_enqueue(&rdma->sc_xprt);
322 	}
323 
324 	svc_rdma_read_info_free(info);
325 }
326 
327 /* This function sleeps when the transport's Send Queue is congested.
328  *
329  * Assumptions:
330  * - If ib_post_send() succeeds, only one completion is expected,
331  *   even if one or more WRs are flushed. This is true when posting
332  *   an rdma_rw_ctx or when posting a single signaled WR.
333  */
334 static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc)
335 {
336 	struct svcxprt_rdma *rdma = cc->cc_rdma;
337 	struct svc_xprt *xprt = &rdma->sc_xprt;
338 	struct ib_send_wr *first_wr;
339 	const struct ib_send_wr *bad_wr;
340 	struct list_head *tmp;
341 	struct ib_cqe *cqe;
342 	int ret;
343 
344 	if (cc->cc_sqecount > rdma->sc_sq_depth)
345 		return -EINVAL;
346 
347 	first_wr = NULL;
348 	cqe = &cc->cc_cqe;
349 	list_for_each(tmp, &cc->cc_rwctxts) {
350 		struct svc_rdma_rw_ctxt *ctxt;
351 
352 		ctxt = list_entry(tmp, struct svc_rdma_rw_ctxt, rw_list);
353 		first_wr = rdma_rw_ctx_wrs(&ctxt->rw_ctx, rdma->sc_qp,
354 					   rdma->sc_port_num, cqe, first_wr);
355 		cqe = NULL;
356 	}
357 
358 	do {
359 		if (atomic_sub_return(cc->cc_sqecount,
360 				      &rdma->sc_sq_avail) > 0) {
361 			ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr);
362 			if (ret)
363 				break;
364 			return 0;
365 		}
366 
367 		trace_svcrdma_sq_full(rdma);
368 		atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
369 		wait_event(rdma->sc_send_wait,
370 			   atomic_read(&rdma->sc_sq_avail) > cc->cc_sqecount);
371 		trace_svcrdma_sq_retry(rdma);
372 	} while (1);
373 
374 	trace_svcrdma_sq_post_err(rdma, ret);
375 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
376 
377 	/* If even one was posted, there will be a completion. */
378 	if (bad_wr != first_wr)
379 		return 0;
380 
381 	atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
382 	wake_up(&rdma->sc_send_wait);
383 	return -ENOTCONN;
384 }
385 
386 /* Build and DMA-map an SGL that covers one kvec in an xdr_buf
387  */
388 static void svc_rdma_vec_to_sg(struct svc_rdma_write_info *info,
389 			       unsigned int len,
390 			       struct svc_rdma_rw_ctxt *ctxt)
391 {
392 	struct scatterlist *sg = ctxt->rw_sg_table.sgl;
393 
394 	sg_set_buf(&sg[0], info->wi_base, len);
395 	info->wi_base += len;
396 
397 	ctxt->rw_nents = 1;
398 }
399 
400 /* Build and DMA-map an SGL that covers part of an xdr_buf's pagelist.
401  */
402 static void svc_rdma_pagelist_to_sg(struct svc_rdma_write_info *info,
403 				    unsigned int remaining,
404 				    struct svc_rdma_rw_ctxt *ctxt)
405 {
406 	unsigned int sge_no, sge_bytes, page_off, page_no;
407 	const struct xdr_buf *xdr = info->wi_xdr;
408 	struct scatterlist *sg;
409 	struct page **page;
410 
411 	page_off = info->wi_next_off + xdr->page_base;
412 	page_no = page_off >> PAGE_SHIFT;
413 	page_off = offset_in_page(page_off);
414 	page = xdr->pages + page_no;
415 	info->wi_next_off += remaining;
416 	sg = ctxt->rw_sg_table.sgl;
417 	sge_no = 0;
418 	do {
419 		sge_bytes = min_t(unsigned int, remaining,
420 				  PAGE_SIZE - page_off);
421 		sg_set_page(sg, *page, sge_bytes, page_off);
422 
423 		remaining -= sge_bytes;
424 		sg = sg_next(sg);
425 		page_off = 0;
426 		sge_no++;
427 		page++;
428 	} while (remaining);
429 
430 	ctxt->rw_nents = sge_no;
431 }
432 
433 /* Construct RDMA Write WRs to send a portion of an xdr_buf containing
434  * an RPC Reply.
435  */
436 static int
437 svc_rdma_build_writes(struct svc_rdma_write_info *info,
438 		      void (*constructor)(struct svc_rdma_write_info *info,
439 					  unsigned int len,
440 					  struct svc_rdma_rw_ctxt *ctxt),
441 		      unsigned int remaining)
442 {
443 	struct svc_rdma_chunk_ctxt *cc = &info->wi_cc;
444 	struct svcxprt_rdma *rdma = cc->cc_rdma;
445 	const struct svc_rdma_segment *seg;
446 	struct svc_rdma_rw_ctxt *ctxt;
447 	int ret;
448 
449 	do {
450 		unsigned int write_len;
451 		u64 offset;
452 
453 		seg = &info->wi_chunk->ch_segments[info->wi_seg_no];
454 		if (!seg)
455 			goto out_overflow;
456 
457 		write_len = min(remaining, seg->rs_length - info->wi_seg_off);
458 		if (!write_len)
459 			goto out_overflow;
460 		ctxt = svc_rdma_get_rw_ctxt(rdma,
461 					    (write_len >> PAGE_SHIFT) + 2);
462 		if (!ctxt)
463 			return -ENOMEM;
464 
465 		constructor(info, write_len, ctxt);
466 		offset = seg->rs_offset + info->wi_seg_off;
467 		ret = svc_rdma_rw_ctx_init(rdma, ctxt, offset, seg->rs_handle,
468 					   DMA_TO_DEVICE);
469 		if (ret < 0)
470 			return -EIO;
471 
472 		list_add(&ctxt->rw_list, &cc->cc_rwctxts);
473 		cc->cc_sqecount += ret;
474 		if (write_len == seg->rs_length - info->wi_seg_off) {
475 			info->wi_seg_no++;
476 			info->wi_seg_off = 0;
477 		} else {
478 			info->wi_seg_off += write_len;
479 		}
480 		remaining -= write_len;
481 	} while (remaining);
482 
483 	return 0;
484 
485 out_overflow:
486 	trace_svcrdma_small_wrch_err(rdma, remaining, info->wi_seg_no,
487 				     info->wi_chunk->ch_segcount);
488 	return -E2BIG;
489 }
490 
491 /**
492  * svc_rdma_iov_write - Construct RDMA Writes from an iov
493  * @info: pointer to write arguments
494  * @iov: kvec to write
495  *
496  * Returns:
497  *   On succes, returns zero
498  *   %-E2BIG if the client-provided Write chunk is too small
499  *   %-ENOMEM if a resource has been exhausted
500  *   %-EIO if an rdma-rw error occurred
501  */
502 static int svc_rdma_iov_write(struct svc_rdma_write_info *info,
503 			      const struct kvec *iov)
504 {
505 	info->wi_base = iov->iov_base;
506 	return svc_rdma_build_writes(info, svc_rdma_vec_to_sg,
507 				     iov->iov_len);
508 }
509 
510 /**
511  * svc_rdma_pages_write - Construct RDMA Writes from pages
512  * @info: pointer to write arguments
513  * @xdr: xdr_buf with pages to write
514  * @offset: offset into the content of @xdr
515  * @length: number of bytes to write
516  *
517  * Returns:
518  *   On succes, returns zero
519  *   %-E2BIG if the client-provided Write chunk is too small
520  *   %-ENOMEM if a resource has been exhausted
521  *   %-EIO if an rdma-rw error occurred
522  */
523 static int svc_rdma_pages_write(struct svc_rdma_write_info *info,
524 				const struct xdr_buf *xdr,
525 				unsigned int offset,
526 				unsigned long length)
527 {
528 	info->wi_xdr = xdr;
529 	info->wi_next_off = offset - xdr->head[0].iov_len;
530 	return svc_rdma_build_writes(info, svc_rdma_pagelist_to_sg,
531 				     length);
532 }
533 
534 /**
535  * svc_rdma_xb_write - Construct RDMA Writes to write an xdr_buf
536  * @xdr: xdr_buf to write
537  * @data: pointer to write arguments
538  *
539  * Returns:
540  *   On succes, returns zero
541  *   %-E2BIG if the client-provided Write chunk is too small
542  *   %-ENOMEM if a resource has been exhausted
543  *   %-EIO if an rdma-rw error occurred
544  */
545 static int svc_rdma_xb_write(const struct xdr_buf *xdr, void *data)
546 {
547 	struct svc_rdma_write_info *info = data;
548 	int ret;
549 
550 	if (xdr->head[0].iov_len) {
551 		ret = svc_rdma_iov_write(info, &xdr->head[0]);
552 		if (ret < 0)
553 			return ret;
554 	}
555 
556 	if (xdr->page_len) {
557 		ret = svc_rdma_pages_write(info, xdr, xdr->head[0].iov_len,
558 					   xdr->page_len);
559 		if (ret < 0)
560 			return ret;
561 	}
562 
563 	if (xdr->tail[0].iov_len) {
564 		ret = svc_rdma_iov_write(info, &xdr->tail[0]);
565 		if (ret < 0)
566 			return ret;
567 	}
568 
569 	return xdr->len;
570 }
571 
572 /**
573  * svc_rdma_send_write_chunk - Write all segments in a Write chunk
574  * @rdma: controlling RDMA transport
575  * @chunk: Write chunk provided by the client
576  * @xdr: xdr_buf containing the data payload
577  *
578  * Returns a non-negative number of bytes the chunk consumed, or
579  *	%-E2BIG if the payload was larger than the Write chunk,
580  *	%-EINVAL if client provided too many segments,
581  *	%-ENOMEM if rdma_rw context pool was exhausted,
582  *	%-ENOTCONN if posting failed (connection is lost),
583  *	%-EIO if rdma_rw initialization failed (DMA mapping, etc).
584  */
585 int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma,
586 			      const struct svc_rdma_chunk *chunk,
587 			      const struct xdr_buf *xdr)
588 {
589 	struct svc_rdma_write_info *info;
590 	struct svc_rdma_chunk_ctxt *cc;
591 	int ret;
592 
593 	info = svc_rdma_write_info_alloc(rdma, chunk);
594 	if (!info)
595 		return -ENOMEM;
596 	cc = &info->wi_cc;
597 
598 	ret = svc_rdma_xb_write(xdr, info);
599 	if (ret != xdr->len)
600 		goto out_err;
601 
602 	trace_svcrdma_post_write_chunk(&cc->cc_cid, cc->cc_sqecount);
603 	ret = svc_rdma_post_chunk_ctxt(cc);
604 	if (ret < 0)
605 		goto out_err;
606 	return xdr->len;
607 
608 out_err:
609 	svc_rdma_write_info_free(info);
610 	return ret;
611 }
612 
613 /**
614  * svc_rdma_send_reply_chunk - Write all segments in the Reply chunk
615  * @rdma: controlling RDMA transport
616  * @rctxt: Write and Reply chunks from client
617  * @xdr: xdr_buf containing an RPC Reply
618  *
619  * Returns a non-negative number of bytes the chunk consumed, or
620  *	%-E2BIG if the payload was larger than the Reply chunk,
621  *	%-EINVAL if client provided too many segments,
622  *	%-ENOMEM if rdma_rw context pool was exhausted,
623  *	%-ENOTCONN if posting failed (connection is lost),
624  *	%-EIO if rdma_rw initialization failed (DMA mapping, etc).
625  */
626 int svc_rdma_send_reply_chunk(struct svcxprt_rdma *rdma,
627 			      const struct svc_rdma_recv_ctxt *rctxt,
628 			      const struct xdr_buf *xdr)
629 {
630 	struct svc_rdma_write_info *info;
631 	struct svc_rdma_chunk_ctxt *cc;
632 	struct svc_rdma_chunk *chunk;
633 	int ret;
634 
635 	if (pcl_is_empty(&rctxt->rc_reply_pcl))
636 		return 0;
637 
638 	chunk = pcl_first_chunk(&rctxt->rc_reply_pcl);
639 	info = svc_rdma_write_info_alloc(rdma, chunk);
640 	if (!info)
641 		return -ENOMEM;
642 	cc = &info->wi_cc;
643 
644 	ret = pcl_process_nonpayloads(&rctxt->rc_write_pcl, xdr,
645 				      svc_rdma_xb_write, info);
646 	if (ret < 0)
647 		goto out_err;
648 
649 	trace_svcrdma_post_reply_chunk(&cc->cc_cid, cc->cc_sqecount);
650 	ret = svc_rdma_post_chunk_ctxt(cc);
651 	if (ret < 0)
652 		goto out_err;
653 
654 	return xdr->len;
655 
656 out_err:
657 	svc_rdma_write_info_free(info);
658 	return ret;
659 }
660 
661 /**
662  * svc_rdma_build_read_segment - Build RDMA Read WQEs to pull one RDMA segment
663  * @info: context for ongoing I/O
664  * @segment: co-ordinates of remote memory to be read
665  *
666  * Returns:
667  *   %0: the Read WR chain was constructed successfully
668  *   %-EINVAL: there were not enough rq_pages to finish
669  *   %-ENOMEM: allocating a local resources failed
670  *   %-EIO: a DMA mapping error occurred
671  */
672 static int svc_rdma_build_read_segment(struct svc_rdma_read_info *info,
673 				       const struct svc_rdma_segment *segment)
674 {
675 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
676 	struct svc_rdma_chunk_ctxt *cc = &info->ri_cc;
677 	struct svc_rqst *rqstp = info->ri_rqst;
678 	struct svc_rdma_rw_ctxt *ctxt;
679 	unsigned int sge_no, seg_len, len;
680 	struct scatterlist *sg;
681 	int ret;
682 
683 	len = segment->rs_length;
684 	sge_no = PAGE_ALIGN(info->ri_pageoff + len) >> PAGE_SHIFT;
685 	ctxt = svc_rdma_get_rw_ctxt(cc->cc_rdma, sge_no);
686 	if (!ctxt)
687 		return -ENOMEM;
688 	ctxt->rw_nents = sge_no;
689 
690 	sg = ctxt->rw_sg_table.sgl;
691 	for (sge_no = 0; sge_no < ctxt->rw_nents; sge_no++) {
692 		seg_len = min_t(unsigned int, len,
693 				PAGE_SIZE - info->ri_pageoff);
694 
695 		head->rc_arg.pages[info->ri_pageno] =
696 			rqstp->rq_pages[info->ri_pageno];
697 		if (!info->ri_pageoff)
698 			head->rc_page_count++;
699 
700 		sg_set_page(sg, rqstp->rq_pages[info->ri_pageno],
701 			    seg_len, info->ri_pageoff);
702 		sg = sg_next(sg);
703 
704 		info->ri_pageoff += seg_len;
705 		if (info->ri_pageoff == PAGE_SIZE) {
706 			info->ri_pageno++;
707 			info->ri_pageoff = 0;
708 		}
709 		len -= seg_len;
710 
711 		/* Safety check */
712 		if (len &&
713 		    &rqstp->rq_pages[info->ri_pageno + 1] > rqstp->rq_page_end)
714 			goto out_overrun;
715 	}
716 
717 	ret = svc_rdma_rw_ctx_init(cc->cc_rdma, ctxt, segment->rs_offset,
718 				   segment->rs_handle, DMA_FROM_DEVICE);
719 	if (ret < 0)
720 		return -EIO;
721 
722 	list_add(&ctxt->rw_list, &cc->cc_rwctxts);
723 	cc->cc_sqecount += ret;
724 	return 0;
725 
726 out_overrun:
727 	trace_svcrdma_page_overrun_err(cc->cc_rdma, rqstp, info->ri_pageno);
728 	return -EINVAL;
729 }
730 
731 /**
732  * svc_rdma_build_read_chunk - Build RDMA Read WQEs to pull one RDMA chunk
733  * @info: context for ongoing I/O
734  * @chunk: Read chunk to pull
735  *
736  * Return values:
737  *   %0: the Read WR chain was constructed successfully
738  *   %-EINVAL: there were not enough resources to finish
739  *   %-ENOMEM: allocating a local resources failed
740  *   %-EIO: a DMA mapping error occurred
741  */
742 static int svc_rdma_build_read_chunk(struct svc_rdma_read_info *info,
743 				     const struct svc_rdma_chunk *chunk)
744 {
745 	const struct svc_rdma_segment *segment;
746 	int ret;
747 
748 	ret = -EINVAL;
749 	pcl_for_each_segment(segment, chunk) {
750 		ret = svc_rdma_build_read_segment(info, segment);
751 		if (ret < 0)
752 			break;
753 		info->ri_totalbytes += segment->rs_length;
754 	}
755 	return ret;
756 }
757 
758 /**
759  * svc_rdma_copy_inline_range - Copy part of the inline content into pages
760  * @info: context for RDMA Reads
761  * @offset: offset into the Receive buffer of region to copy
762  * @remaining: length of region to copy
763  *
764  * Take a page at a time from rqstp->rq_pages and copy the inline
765  * content from the Receive buffer into that page. Update
766  * info->ri_pageno and info->ri_pageoff so that the next RDMA Read
767  * result will land contiguously with the copied content.
768  *
769  * Return values:
770  *   %0: Inline content was successfully copied
771  *   %-EINVAL: offset or length was incorrect
772  */
773 static int svc_rdma_copy_inline_range(struct svc_rdma_read_info *info,
774 				      unsigned int offset,
775 				      unsigned int remaining)
776 {
777 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
778 	unsigned char *dst, *src = head->rc_recv_buf;
779 	struct svc_rqst *rqstp = info->ri_rqst;
780 	unsigned int page_no, numpages;
781 
782 	numpages = PAGE_ALIGN(info->ri_pageoff + remaining) >> PAGE_SHIFT;
783 	for (page_no = 0; page_no < numpages; page_no++) {
784 		unsigned int page_len;
785 
786 		page_len = min_t(unsigned int, remaining,
787 				 PAGE_SIZE - info->ri_pageoff);
788 
789 		head->rc_arg.pages[info->ri_pageno] =
790 			rqstp->rq_pages[info->ri_pageno];
791 		if (!info->ri_pageoff)
792 			head->rc_page_count++;
793 
794 		dst = page_address(head->rc_arg.pages[info->ri_pageno]);
795 		memcpy(dst + info->ri_pageno, src + offset, page_len);
796 
797 		info->ri_totalbytes += page_len;
798 		info->ri_pageoff += page_len;
799 		if (info->ri_pageoff == PAGE_SIZE) {
800 			info->ri_pageno++;
801 			info->ri_pageoff = 0;
802 		}
803 		remaining -= page_len;
804 		offset += page_len;
805 	}
806 
807 	return -EINVAL;
808 }
809 
810 /**
811  * svc_rdma_read_multiple_chunks - Construct RDMA Reads to pull data item Read chunks
812  * @info: context for RDMA Reads
813  *
814  * The chunk data lands in head->rc_arg as a series of contiguous pages,
815  * like an incoming TCP call.
816  *
817  * Return values:
818  *   %0: RDMA Read WQEs were successfully built
819  *   %-EINVAL: client provided too many chunks or segments,
820  *   %-ENOMEM: rdma_rw context pool was exhausted,
821  *   %-ENOTCONN: posting failed (connection is lost),
822  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
823  */
824 static noinline int svc_rdma_read_multiple_chunks(struct svc_rdma_read_info *info)
825 {
826 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
827 	const struct svc_rdma_pcl *pcl = &head->rc_read_pcl;
828 	struct svc_rdma_chunk *chunk, *next;
829 	struct xdr_buf *buf = &head->rc_arg;
830 	unsigned int start, length;
831 	int ret;
832 
833 	start = 0;
834 	chunk = pcl_first_chunk(pcl);
835 	length = chunk->ch_position;
836 	ret = svc_rdma_copy_inline_range(info, start, length);
837 	if (ret < 0)
838 		return ret;
839 
840 	pcl_for_each_chunk(chunk, pcl) {
841 		ret = svc_rdma_build_read_chunk(info, chunk);
842 		if (ret < 0)
843 			return ret;
844 
845 		next = pcl_next_chunk(pcl, chunk);
846 		if (!next)
847 			break;
848 
849 		start += length;
850 		length = next->ch_position - info->ri_totalbytes;
851 		ret = svc_rdma_copy_inline_range(info, start, length);
852 		if (ret < 0)
853 			return ret;
854 	}
855 
856 	start += length;
857 	length = head->rc_byte_len - start;
858 	ret = svc_rdma_copy_inline_range(info, start, length);
859 	if (ret < 0)
860 		return ret;
861 
862 	buf->len += info->ri_totalbytes;
863 	buf->buflen += info->ri_totalbytes;
864 
865 	head->rc_hdr_count = 1;
866 	buf->head[0].iov_base = page_address(head->rc_pages[0]);
867 	buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, info->ri_totalbytes);
868 	buf->page_len = info->ri_totalbytes - buf->head[0].iov_len;
869 	return 0;
870 }
871 
872 /**
873  * svc_rdma_read_data_item - Construct RDMA Reads to pull data item Read chunks
874  * @info: context for RDMA Reads
875  *
876  * The chunk data lands in the page list of head->rc_arg.pages.
877  *
878  * Currently NFSD does not look at the head->rc_arg.tail[0] iovec.
879  * Therefore, XDR round-up of the Read chunk and trailing
880  * inline content must both be added at the end of the pagelist.
881  *
882  * Return values:
883  *   %0: RDMA Read WQEs were successfully built
884  *   %-EINVAL: client provided too many chunks or segments,
885  *   %-ENOMEM: rdma_rw context pool was exhausted,
886  *   %-ENOTCONN: posting failed (connection is lost),
887  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
888  */
889 static int svc_rdma_read_data_item(struct svc_rdma_read_info *info)
890 {
891 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
892 	struct xdr_buf *buf = &head->rc_arg;
893 	struct svc_rdma_chunk *chunk;
894 	unsigned int length;
895 	int ret;
896 
897 	chunk = pcl_first_chunk(&head->rc_read_pcl);
898 	ret = svc_rdma_build_read_chunk(info, chunk);
899 	if (ret < 0)
900 		goto out;
901 
902 	head->rc_hdr_count = 0;
903 
904 	/* Split the Receive buffer between the head and tail
905 	 * buffers at Read chunk's position. XDR roundup of the
906 	 * chunk is not included in either the pagelist or in
907 	 * the tail.
908 	 */
909 	buf->tail[0].iov_base = buf->head[0].iov_base + chunk->ch_position;
910 	buf->tail[0].iov_len = buf->head[0].iov_len - chunk->ch_position;
911 	buf->head[0].iov_len = chunk->ch_position;
912 
913 	/* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2).
914 	 *
915 	 * If the client already rounded up the chunk length, the
916 	 * length does not change. Otherwise, the length of the page
917 	 * list is increased to include XDR round-up.
918 	 *
919 	 * Currently these chunks always start at page offset 0,
920 	 * thus the rounded-up length never crosses a page boundary.
921 	 */
922 	length = XDR_QUADLEN(info->ri_totalbytes) << 2;
923 	buf->page_len = length;
924 	buf->len += length;
925 	buf->buflen += length;
926 
927 out:
928 	return ret;
929 }
930 
931 /**
932  * svc_rdma_read_chunk_range - Build RDMA Read WQEs for portion of a chunk
933  * @info: context for RDMA Reads
934  * @chunk: parsed Call chunk to pull
935  * @offset: offset of region to pull
936  * @length: length of region to pull
937  *
938  * Return values:
939  *   %0: RDMA Read WQEs were successfully built
940  *   %-EINVAL: there were not enough resources to finish
941  *   %-ENOMEM: rdma_rw context pool was exhausted,
942  *   %-ENOTCONN: posting failed (connection is lost),
943  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
944  */
945 static int svc_rdma_read_chunk_range(struct svc_rdma_read_info *info,
946 				     const struct svc_rdma_chunk *chunk,
947 				     unsigned int offset, unsigned int length)
948 {
949 	const struct svc_rdma_segment *segment;
950 	int ret;
951 
952 	ret = -EINVAL;
953 	pcl_for_each_segment(segment, chunk) {
954 		struct svc_rdma_segment dummy;
955 
956 		if (offset > segment->rs_length) {
957 			offset -= segment->rs_length;
958 			continue;
959 		}
960 
961 		dummy.rs_handle = segment->rs_handle;
962 		dummy.rs_length = min_t(u32, length, segment->rs_length) - offset;
963 		dummy.rs_offset = segment->rs_offset + offset;
964 
965 		ret = svc_rdma_build_read_segment(info, &dummy);
966 		if (ret < 0)
967 			break;
968 
969 		info->ri_totalbytes += dummy.rs_length;
970 		length -= dummy.rs_length;
971 		offset = 0;
972 	}
973 	return ret;
974 }
975 
976 /**
977  * svc_rdma_read_call_chunk - Build RDMA Read WQEs to pull a Long Message
978  * @info: context for RDMA Reads
979  *
980  * Return values:
981  *   %0: RDMA Read WQEs were successfully built
982  *   %-EINVAL: there were not enough resources to finish
983  *   %-ENOMEM: rdma_rw context pool was exhausted,
984  *   %-ENOTCONN: posting failed (connection is lost),
985  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
986  */
987 static int svc_rdma_read_call_chunk(struct svc_rdma_read_info *info)
988 {
989 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
990 	const struct svc_rdma_chunk *call_chunk =
991 			pcl_first_chunk(&head->rc_call_pcl);
992 	const struct svc_rdma_pcl *pcl = &head->rc_read_pcl;
993 	struct svc_rdma_chunk *chunk, *next;
994 	unsigned int start, length;
995 	int ret;
996 
997 	if (pcl_is_empty(pcl))
998 		return svc_rdma_build_read_chunk(info, call_chunk);
999 
1000 	start = 0;
1001 	chunk = pcl_first_chunk(pcl);
1002 	length = chunk->ch_position;
1003 	ret = svc_rdma_read_chunk_range(info, call_chunk, start, length);
1004 	if (ret < 0)
1005 		return ret;
1006 
1007 	pcl_for_each_chunk(chunk, pcl) {
1008 		ret = svc_rdma_build_read_chunk(info, chunk);
1009 		if (ret < 0)
1010 			return ret;
1011 
1012 		next = pcl_next_chunk(pcl, chunk);
1013 		if (!next)
1014 			break;
1015 
1016 		start += length;
1017 		length = next->ch_position - info->ri_totalbytes;
1018 		ret = svc_rdma_read_chunk_range(info, call_chunk,
1019 						start, length);
1020 		if (ret < 0)
1021 			return ret;
1022 	}
1023 
1024 	start += length;
1025 	length = call_chunk->ch_length - start;
1026 	return svc_rdma_read_chunk_range(info, call_chunk, start, length);
1027 }
1028 
1029 /**
1030  * svc_rdma_read_special - Build RDMA Read WQEs to pull a Long Message
1031  * @info: context for RDMA Reads
1032  *
1033  * The start of the data lands in the first page just after the
1034  * Transport header, and the rest lands in the page list of
1035  * head->rc_arg.pages.
1036  *
1037  * Assumptions:
1038  *	- A PZRC is never sent in an RDMA_MSG message, though it's
1039  *	  allowed by spec.
1040  *
1041  * Return values:
1042  *   %0: RDMA Read WQEs were successfully built
1043  *   %-EINVAL: client provided too many chunks or segments,
1044  *   %-ENOMEM: rdma_rw context pool was exhausted,
1045  *   %-ENOTCONN: posting failed (connection is lost),
1046  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
1047  */
1048 static noinline int svc_rdma_read_special(struct svc_rdma_read_info *info)
1049 {
1050 	struct svc_rdma_recv_ctxt *head = info->ri_readctxt;
1051 	struct xdr_buf *buf = &head->rc_arg;
1052 	int ret;
1053 
1054 	ret = svc_rdma_read_call_chunk(info);
1055 	if (ret < 0)
1056 		goto out;
1057 
1058 	buf->len += info->ri_totalbytes;
1059 	buf->buflen += info->ri_totalbytes;
1060 
1061 	head->rc_hdr_count = 1;
1062 	buf->head[0].iov_base = page_address(head->rc_pages[0]);
1063 	buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, info->ri_totalbytes);
1064 	buf->page_len = info->ri_totalbytes - buf->head[0].iov_len;
1065 
1066 out:
1067 	return ret;
1068 }
1069 
1070 /* Pages under I/O have been copied to head->rc_pages. Ensure they
1071  * are not released by svc_xprt_release() until the I/O is complete.
1072  *
1073  * This has to be done after all Read WRs are constructed to properly
1074  * handle a page that is part of I/O on behalf of two different RDMA
1075  * segments.
1076  *
1077  * Do this only if I/O has been posted. Otherwise, we do indeed want
1078  * svc_xprt_release() to clean things up properly.
1079  */
1080 static void svc_rdma_save_io_pages(struct svc_rqst *rqstp,
1081 				   const unsigned int start,
1082 				   const unsigned int num_pages)
1083 {
1084 	unsigned int i;
1085 
1086 	for (i = start; i < num_pages + start; i++)
1087 		rqstp->rq_pages[i] = NULL;
1088 }
1089 
1090 /**
1091  * svc_rdma_process_read_list - Pull list of Read chunks from the client
1092  * @rdma: controlling RDMA transport
1093  * @rqstp: set of pages to use as Read sink buffers
1094  * @head: pages under I/O collect here
1095  *
1096  * The RPC/RDMA protocol assumes that the upper layer's XDR decoders
1097  * pull each Read chunk as they decode an incoming RPC message.
1098  *
1099  * On Linux, however, the server needs to have a fully-constructed RPC
1100  * message in rqstp->rq_arg when there is a positive return code from
1101  * ->xpo_recvfrom. So the Read list is safety-checked immediately when
1102  * it is received, then here the whole Read list is pulled all at once.
1103  * The ingress RPC message is fully reconstructed once all associated
1104  * RDMA Reads have completed.
1105  *
1106  * Return values:
1107  *   %1: all needed RDMA Reads were posted successfully,
1108  *   %-EINVAL: client provided too many chunks or segments,
1109  *   %-ENOMEM: rdma_rw context pool was exhausted,
1110  *   %-ENOTCONN: posting failed (connection is lost),
1111  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
1112  */
1113 int svc_rdma_process_read_list(struct svcxprt_rdma *rdma,
1114 			       struct svc_rqst *rqstp,
1115 			       struct svc_rdma_recv_ctxt *head)
1116 {
1117 	struct svc_rdma_read_info *info;
1118 	struct svc_rdma_chunk_ctxt *cc;
1119 	int ret;
1120 
1121 	/* The request (with page list) is constructed in
1122 	 * head->rc_arg. Pages involved with RDMA Read I/O are
1123 	 * transferred there.
1124 	 */
1125 	head->rc_arg.head[0] = rqstp->rq_arg.head[0];
1126 	head->rc_arg.tail[0] = rqstp->rq_arg.tail[0];
1127 	head->rc_arg.pages = head->rc_pages;
1128 	head->rc_arg.page_base = 0;
1129 	head->rc_arg.page_len = 0;
1130 	head->rc_arg.len = rqstp->rq_arg.len;
1131 	head->rc_arg.buflen = rqstp->rq_arg.buflen;
1132 
1133 	info = svc_rdma_read_info_alloc(rdma);
1134 	if (!info)
1135 		return -ENOMEM;
1136 	cc = &info->ri_cc;
1137 	info->ri_rqst = rqstp;
1138 	info->ri_readctxt = head;
1139 	info->ri_pageno = 0;
1140 	info->ri_pageoff = 0;
1141 	info->ri_totalbytes = 0;
1142 
1143 	if (pcl_is_empty(&head->rc_call_pcl)) {
1144 		if (head->rc_read_pcl.cl_count == 1)
1145 			ret = svc_rdma_read_data_item(info);
1146 		else
1147 			ret = svc_rdma_read_multiple_chunks(info);
1148 	} else
1149 		ret = svc_rdma_read_special(info);
1150 	if (ret < 0)
1151 		goto out_err;
1152 
1153 	trace_svcrdma_post_read_chunk(&cc->cc_cid, cc->cc_sqecount);
1154 	ret = svc_rdma_post_chunk_ctxt(cc);
1155 	if (ret < 0)
1156 		goto out_err;
1157 	svc_rdma_save_io_pages(rqstp, 0, head->rc_page_count);
1158 	return 1;
1159 
1160 out_err:
1161 	svc_rdma_read_info_free(info);
1162 	return ret;
1163 }
1164