xref: /linux/net/sunrpc/xprtrdma/svc_rdma_rw.c (revision 3a07362fab1653d3aca31a9155c8cc776138fd02)
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 llist_node	rw_node;
39 	struct list_head	rw_list;
40 	struct rdma_rw_ctx	rw_ctx;
41 	unsigned int		rw_nents;
42 	unsigned int		rw_first_sgl_nents;
43 	struct sg_table		rw_sg_table;
44 	struct scatterlist	rw_first_sgl[];
45 };
46 
47 static inline struct svc_rdma_rw_ctxt *
svc_rdma_next_ctxt(struct list_head * list)48 svc_rdma_next_ctxt(struct list_head *list)
49 {
50 	return list_first_entry_or_null(list, struct svc_rdma_rw_ctxt,
51 					rw_list);
52 }
53 
54 static struct svc_rdma_rw_ctxt *
svc_rdma_get_rw_ctxt(struct svcxprt_rdma * rdma,unsigned int sges)55 svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges)
56 {
57 	struct ib_device *dev = rdma->sc_cm_id->device;
58 	unsigned int first_sgl_nents = dev->attrs.max_send_sge;
59 	struct svc_rdma_rw_ctxt *ctxt;
60 	struct llist_node *node;
61 
62 	spin_lock(&rdma->sc_rw_ctxt_lock);
63 	node = llist_del_first(&rdma->sc_rw_ctxts);
64 	spin_unlock(&rdma->sc_rw_ctxt_lock);
65 	if (node) {
66 		ctxt = llist_entry(node, struct svc_rdma_rw_ctxt, rw_node);
67 	} else {
68 		ctxt = kmalloc_node(struct_size(ctxt, rw_first_sgl, first_sgl_nents),
69 				    GFP_KERNEL, ibdev_to_node(dev));
70 		if (!ctxt)
71 			goto out_noctx;
72 
73 		INIT_LIST_HEAD(&ctxt->rw_list);
74 		ctxt->rw_first_sgl_nents = first_sgl_nents;
75 	}
76 
77 	ctxt->rw_sg_table.sgl = ctxt->rw_first_sgl;
78 	if (sg_alloc_table_chained(&ctxt->rw_sg_table, sges,
79 				   ctxt->rw_sg_table.sgl,
80 				   first_sgl_nents))
81 		goto out_free;
82 	return ctxt;
83 
84 out_free:
85 	kfree(ctxt);
86 out_noctx:
87 	trace_svcrdma_rwctx_empty(rdma, sges);
88 	return NULL;
89 }
90 
__svc_rdma_put_rw_ctxt(struct svc_rdma_rw_ctxt * ctxt,struct llist_head * list)91 static void __svc_rdma_put_rw_ctxt(struct svc_rdma_rw_ctxt *ctxt,
92 				   struct llist_head *list)
93 {
94 	sg_free_table_chained(&ctxt->rw_sg_table, ctxt->rw_first_sgl_nents);
95 	llist_add(&ctxt->rw_node, list);
96 }
97 
svc_rdma_put_rw_ctxt(struct svcxprt_rdma * rdma,struct svc_rdma_rw_ctxt * ctxt)98 static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
99 				 struct svc_rdma_rw_ctxt *ctxt)
100 {
101 	__svc_rdma_put_rw_ctxt(ctxt, &rdma->sc_rw_ctxts);
102 }
103 
104 /**
105  * svc_rdma_destroy_rw_ctxts - Free accumulated R/W contexts
106  * @rdma: transport about to be destroyed
107  *
108  */
svc_rdma_destroy_rw_ctxts(struct svcxprt_rdma * rdma)109 void svc_rdma_destroy_rw_ctxts(struct svcxprt_rdma *rdma)
110 {
111 	struct svc_rdma_rw_ctxt *ctxt;
112 	struct llist_node *node;
113 
114 	while ((node = llist_del_first(&rdma->sc_rw_ctxts)) != NULL) {
115 		ctxt = llist_entry(node, struct svc_rdma_rw_ctxt, rw_node);
116 		kfree(ctxt);
117 	}
118 }
119 
120 /**
121  * svc_rdma_rw_ctx_init - Prepare a R/W context for I/O
122  * @rdma: controlling transport instance
123  * @ctxt: R/W context to prepare
124  * @offset: RDMA offset
125  * @handle: RDMA tag/handle
126  * @direction: I/O direction
127  *
128  * Returns on success, the number of WQEs that will be needed
129  * on the workqueue, or a negative errno.
130  */
svc_rdma_rw_ctx_init(struct svcxprt_rdma * rdma,struct svc_rdma_rw_ctxt * ctxt,u64 offset,u32 handle,enum dma_data_direction direction)131 static int svc_rdma_rw_ctx_init(struct svcxprt_rdma *rdma,
132 				struct svc_rdma_rw_ctxt *ctxt,
133 				u64 offset, u32 handle,
134 				enum dma_data_direction direction)
135 {
136 	int ret;
137 
138 	ret = rdma_rw_ctx_init(&ctxt->rw_ctx, rdma->sc_qp, rdma->sc_port_num,
139 			       ctxt->rw_sg_table.sgl, ctxt->rw_nents,
140 			       0, offset, handle, direction);
141 	if (unlikely(ret < 0)) {
142 		trace_svcrdma_dma_map_rw_err(rdma, offset, handle,
143 					     ctxt->rw_nents, ret);
144 		svc_rdma_put_rw_ctxt(rdma, ctxt);
145 	}
146 	return ret;
147 }
148 
149 /**
150  * svc_rdma_cc_init - Initialize an svc_rdma_chunk_ctxt
151  * @rdma: controlling transport instance
152  * @cc: svc_rdma_chunk_ctxt to be initialized
153  */
svc_rdma_cc_init(struct svcxprt_rdma * rdma,struct svc_rdma_chunk_ctxt * cc)154 void svc_rdma_cc_init(struct svcxprt_rdma *rdma,
155 		      struct svc_rdma_chunk_ctxt *cc)
156 {
157 	struct rpc_rdma_cid *cid = &cc->cc_cid;
158 
159 	if (unlikely(!cid->ci_completion_id))
160 		svc_rdma_send_cid_init(rdma, cid);
161 
162 	INIT_LIST_HEAD(&cc->cc_rwctxts);
163 	cc->cc_sqecount = 0;
164 }
165 
166 /**
167  * svc_rdma_cc_release - Release resources held by a svc_rdma_chunk_ctxt
168  * @rdma: controlling transport instance
169  * @cc: svc_rdma_chunk_ctxt to be released
170  * @dir: DMA direction
171  */
svc_rdma_cc_release(struct svcxprt_rdma * rdma,struct svc_rdma_chunk_ctxt * cc,enum dma_data_direction dir)172 void svc_rdma_cc_release(struct svcxprt_rdma *rdma,
173 			 struct svc_rdma_chunk_ctxt *cc,
174 			 enum dma_data_direction dir)
175 {
176 	struct llist_node *first, *last;
177 	struct svc_rdma_rw_ctxt *ctxt;
178 	LLIST_HEAD(free);
179 
180 	trace_svcrdma_cc_release(&cc->cc_cid, cc->cc_sqecount);
181 
182 	first = last = NULL;
183 	while ((ctxt = svc_rdma_next_ctxt(&cc->cc_rwctxts)) != NULL) {
184 		list_del(&ctxt->rw_list);
185 
186 		rdma_rw_ctx_destroy(&ctxt->rw_ctx, rdma->sc_qp,
187 				    rdma->sc_port_num, ctxt->rw_sg_table.sgl,
188 				    ctxt->rw_nents, dir);
189 		__svc_rdma_put_rw_ctxt(ctxt, &free);
190 
191 		ctxt->rw_node.next = first;
192 		first = &ctxt->rw_node;
193 		if (!last)
194 			last = first;
195 	}
196 	if (first)
197 		llist_add_batch(first, last, &rdma->sc_rw_ctxts);
198 }
199 
200 static struct svc_rdma_write_info *
svc_rdma_write_info_alloc(struct svcxprt_rdma * rdma,const struct svc_rdma_chunk * chunk)201 svc_rdma_write_info_alloc(struct svcxprt_rdma *rdma,
202 			  const struct svc_rdma_chunk *chunk)
203 {
204 	struct svc_rdma_write_info *info;
205 
206 	info = kzalloc_node(sizeof(*info), GFP_KERNEL,
207 			    ibdev_to_node(rdma->sc_cm_id->device));
208 	if (!info)
209 		return info;
210 
211 	info->wi_rdma = rdma;
212 	info->wi_chunk = chunk;
213 	svc_rdma_cc_init(rdma, &info->wi_cc);
214 	info->wi_cc.cc_cqe.done = svc_rdma_write_done;
215 	return info;
216 }
217 
svc_rdma_write_info_free_async(struct work_struct * work)218 static void svc_rdma_write_info_free_async(struct work_struct *work)
219 {
220 	struct svc_rdma_write_info *info;
221 
222 	info = container_of(work, struct svc_rdma_write_info, wi_work);
223 	svc_rdma_cc_release(info->wi_rdma, &info->wi_cc, DMA_TO_DEVICE);
224 	kfree(info);
225 }
226 
svc_rdma_write_info_free(struct svc_rdma_write_info * info)227 static void svc_rdma_write_info_free(struct svc_rdma_write_info *info)
228 {
229 	INIT_WORK(&info->wi_work, svc_rdma_write_info_free_async);
230 	queue_work(svcrdma_wq, &info->wi_work);
231 }
232 
233 /**
234  * svc_rdma_reply_chunk_release - Release Reply chunk I/O resources
235  * @rdma: controlling transport
236  * @ctxt: Send context that is being released
237  */
svc_rdma_reply_chunk_release(struct svcxprt_rdma * rdma,struct svc_rdma_send_ctxt * ctxt)238 void svc_rdma_reply_chunk_release(struct svcxprt_rdma *rdma,
239 				  struct svc_rdma_send_ctxt *ctxt)
240 {
241 	struct svc_rdma_chunk_ctxt *cc = &ctxt->sc_reply_info.wi_cc;
242 
243 	if (!cc->cc_sqecount)
244 		return;
245 	svc_rdma_cc_release(rdma, cc, DMA_TO_DEVICE);
246 }
247 
248 /**
249  * svc_rdma_reply_done - Reply chunk Write completion handler
250  * @cq: controlling Completion Queue
251  * @wc: Work Completion report
252  *
253  * Pages under I/O are released by a subsequent Send completion.
254  */
svc_rdma_reply_done(struct ib_cq * cq,struct ib_wc * wc)255 static void svc_rdma_reply_done(struct ib_cq *cq, struct ib_wc *wc)
256 {
257 	struct ib_cqe *cqe = wc->wr_cqe;
258 	struct svc_rdma_chunk_ctxt *cc =
259 			container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe);
260 	struct svcxprt_rdma *rdma = cq->cq_context;
261 
262 	switch (wc->status) {
263 	case IB_WC_SUCCESS:
264 		trace_svcrdma_wc_reply(&cc->cc_cid);
265 		return;
266 	case IB_WC_WR_FLUSH_ERR:
267 		trace_svcrdma_wc_reply_flush(wc, &cc->cc_cid);
268 		break;
269 	default:
270 		trace_svcrdma_wc_reply_err(wc, &cc->cc_cid);
271 	}
272 
273 	svc_xprt_deferred_close(&rdma->sc_xprt);
274 }
275 
276 /**
277  * svc_rdma_write_done - Write chunk completion
278  * @cq: controlling Completion Queue
279  * @wc: Work Completion
280  *
281  * Pages under I/O are freed by a subsequent Send completion.
282  */
svc_rdma_write_done(struct ib_cq * cq,struct ib_wc * wc)283 static void svc_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
284 {
285 	struct svcxprt_rdma *rdma = cq->cq_context;
286 	struct ib_cqe *cqe = wc->wr_cqe;
287 	struct svc_rdma_chunk_ctxt *cc =
288 			container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe);
289 	struct svc_rdma_write_info *info =
290 			container_of(cc, struct svc_rdma_write_info, wi_cc);
291 
292 	switch (wc->status) {
293 	case IB_WC_SUCCESS:
294 		trace_svcrdma_wc_write(&cc->cc_cid);
295 		break;
296 	case IB_WC_WR_FLUSH_ERR:
297 		trace_svcrdma_wc_write_flush(wc, &cc->cc_cid);
298 		break;
299 	default:
300 		trace_svcrdma_wc_write_err(wc, &cc->cc_cid);
301 	}
302 
303 	svc_rdma_wake_send_waiters(rdma, cc->cc_sqecount);
304 
305 	if (unlikely(wc->status != IB_WC_SUCCESS))
306 		svc_xprt_deferred_close(&rdma->sc_xprt);
307 
308 	svc_rdma_write_info_free(info);
309 }
310 
311 /**
312  * svc_rdma_wc_read_done - Handle completion of an RDMA Read ctx
313  * @cq: controlling Completion Queue
314  * @wc: Work Completion
315  *
316  */
svc_rdma_wc_read_done(struct ib_cq * cq,struct ib_wc * wc)317 static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc)
318 {
319 	struct svcxprt_rdma *rdma = cq->cq_context;
320 	struct ib_cqe *cqe = wc->wr_cqe;
321 	struct svc_rdma_chunk_ctxt *cc =
322 			container_of(cqe, struct svc_rdma_chunk_ctxt, cc_cqe);
323 	struct svc_rdma_recv_ctxt *ctxt;
324 
325 	svc_rdma_wake_send_waiters(rdma, cc->cc_sqecount);
326 
327 	ctxt = container_of(cc, struct svc_rdma_recv_ctxt, rc_cc);
328 	switch (wc->status) {
329 	case IB_WC_SUCCESS:
330 		trace_svcrdma_wc_read(wc, &cc->cc_cid, ctxt->rc_readbytes,
331 				      cc->cc_posttime);
332 
333 		spin_lock(&rdma->sc_rq_dto_lock);
334 		list_add_tail(&ctxt->rc_list, &rdma->sc_read_complete_q);
335 		/* the unlock pairs with the smp_rmb in svc_xprt_ready */
336 		set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags);
337 		spin_unlock(&rdma->sc_rq_dto_lock);
338 		svc_xprt_enqueue(&rdma->sc_xprt);
339 		return;
340 	case IB_WC_WR_FLUSH_ERR:
341 		trace_svcrdma_wc_read_flush(wc, &cc->cc_cid);
342 		break;
343 	default:
344 		trace_svcrdma_wc_read_err(wc, &cc->cc_cid);
345 	}
346 
347 	/* The RDMA Read has flushed, so the incoming RPC message
348 	 * cannot be constructed and must be dropped. Signal the
349 	 * loss to the client by closing the connection.
350 	 */
351 	svc_rdma_cc_release(rdma, cc, DMA_FROM_DEVICE);
352 	svc_rdma_recv_ctxt_put(rdma, ctxt);
353 	svc_xprt_deferred_close(&rdma->sc_xprt);
354 }
355 
356 /*
357  * Assumptions:
358  * - If ib_post_send() succeeds, only one completion is expected,
359  *   even if one or more WRs are flushed. This is true when posting
360  *   an rdma_rw_ctx or when posting a single signaled WR.
361  */
svc_rdma_post_chunk_ctxt(struct svcxprt_rdma * rdma,struct svc_rdma_chunk_ctxt * cc)362 static int svc_rdma_post_chunk_ctxt(struct svcxprt_rdma *rdma,
363 				    struct svc_rdma_chunk_ctxt *cc)
364 {
365 	struct ib_send_wr *first_wr;
366 	const struct ib_send_wr *bad_wr;
367 	struct list_head *tmp;
368 	struct ib_cqe *cqe;
369 	int ret;
370 
371 	might_sleep();
372 
373 	if (cc->cc_sqecount > rdma->sc_sq_depth)
374 		return -EINVAL;
375 
376 	first_wr = NULL;
377 	cqe = &cc->cc_cqe;
378 	list_for_each(tmp, &cc->cc_rwctxts) {
379 		struct svc_rdma_rw_ctxt *ctxt;
380 
381 		ctxt = list_entry(tmp, struct svc_rdma_rw_ctxt, rw_list);
382 		first_wr = rdma_rw_ctx_wrs(&ctxt->rw_ctx, rdma->sc_qp,
383 					   rdma->sc_port_num, cqe, first_wr);
384 		cqe = NULL;
385 	}
386 
387 	do {
388 		if (atomic_sub_return(cc->cc_sqecount,
389 				      &rdma->sc_sq_avail) > 0) {
390 			cc->cc_posttime = ktime_get();
391 			ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr);
392 			if (ret)
393 				break;
394 			return 0;
395 		}
396 
397 		percpu_counter_inc(&svcrdma_stat_sq_starve);
398 		trace_svcrdma_sq_full(rdma, &cc->cc_cid);
399 		atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
400 		wait_event(rdma->sc_send_wait,
401 			   atomic_read(&rdma->sc_sq_avail) > cc->cc_sqecount);
402 		trace_svcrdma_sq_retry(rdma, &cc->cc_cid);
403 	} while (1);
404 
405 	trace_svcrdma_sq_post_err(rdma, &cc->cc_cid, ret);
406 	svc_xprt_deferred_close(&rdma->sc_xprt);
407 
408 	/* If even one was posted, there will be a completion. */
409 	if (bad_wr != first_wr)
410 		return 0;
411 
412 	atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
413 	wake_up(&rdma->sc_send_wait);
414 	return -ENOTCONN;
415 }
416 
417 /* Build and DMA-map an SGL that covers one kvec in an xdr_buf
418  */
svc_rdma_vec_to_sg(struct svc_rdma_write_info * info,unsigned int len,struct svc_rdma_rw_ctxt * ctxt)419 static void svc_rdma_vec_to_sg(struct svc_rdma_write_info *info,
420 			       unsigned int len,
421 			       struct svc_rdma_rw_ctxt *ctxt)
422 {
423 	struct scatterlist *sg = ctxt->rw_sg_table.sgl;
424 
425 	sg_set_buf(&sg[0], info->wi_base, len);
426 	info->wi_base += len;
427 
428 	ctxt->rw_nents = 1;
429 }
430 
431 /* Build and DMA-map an SGL that covers part of an xdr_buf's pagelist.
432  */
svc_rdma_pagelist_to_sg(struct svc_rdma_write_info * info,unsigned int remaining,struct svc_rdma_rw_ctxt * ctxt)433 static void svc_rdma_pagelist_to_sg(struct svc_rdma_write_info *info,
434 				    unsigned int remaining,
435 				    struct svc_rdma_rw_ctxt *ctxt)
436 {
437 	unsigned int sge_no, sge_bytes, page_off, page_no;
438 	const struct xdr_buf *xdr = info->wi_xdr;
439 	struct scatterlist *sg;
440 	struct page **page;
441 
442 	page_off = info->wi_next_off + xdr->page_base;
443 	page_no = page_off >> PAGE_SHIFT;
444 	page_off = offset_in_page(page_off);
445 	page = xdr->pages + page_no;
446 	info->wi_next_off += remaining;
447 	sg = ctxt->rw_sg_table.sgl;
448 	sge_no = 0;
449 	do {
450 		sge_bytes = min_t(unsigned int, remaining,
451 				  PAGE_SIZE - page_off);
452 		sg_set_page(sg, *page, sge_bytes, page_off);
453 
454 		remaining -= sge_bytes;
455 		sg = sg_next(sg);
456 		page_off = 0;
457 		sge_no++;
458 		page++;
459 	} while (remaining);
460 
461 	ctxt->rw_nents = sge_no;
462 }
463 
464 /* Construct RDMA Write WRs to send a portion of an xdr_buf containing
465  * an RPC Reply.
466  */
467 static int
svc_rdma_build_writes(struct svc_rdma_write_info * info,void (* constructor)(struct svc_rdma_write_info * info,unsigned int len,struct svc_rdma_rw_ctxt * ctxt),unsigned int remaining)468 svc_rdma_build_writes(struct svc_rdma_write_info *info,
469 		      void (*constructor)(struct svc_rdma_write_info *info,
470 					  unsigned int len,
471 					  struct svc_rdma_rw_ctxt *ctxt),
472 		      unsigned int remaining)
473 {
474 	struct svc_rdma_chunk_ctxt *cc = &info->wi_cc;
475 	struct svcxprt_rdma *rdma = info->wi_rdma;
476 	const struct svc_rdma_segment *seg;
477 	struct svc_rdma_rw_ctxt *ctxt;
478 	int ret;
479 
480 	do {
481 		unsigned int write_len;
482 		u64 offset;
483 
484 		if (info->wi_seg_no >= info->wi_chunk->ch_segcount)
485 			goto out_overflow;
486 
487 		seg = &info->wi_chunk->ch_segments[info->wi_seg_no];
488 		write_len = min(remaining, seg->rs_length - info->wi_seg_off);
489 		if (!write_len)
490 			goto out_overflow;
491 		ctxt = svc_rdma_get_rw_ctxt(rdma,
492 					    (write_len >> PAGE_SHIFT) + 2);
493 		if (!ctxt)
494 			return -ENOMEM;
495 
496 		constructor(info, write_len, ctxt);
497 		offset = seg->rs_offset + info->wi_seg_off;
498 		ret = svc_rdma_rw_ctx_init(rdma, ctxt, offset, seg->rs_handle,
499 					   DMA_TO_DEVICE);
500 		if (ret < 0)
501 			return -EIO;
502 		percpu_counter_inc(&svcrdma_stat_write);
503 
504 		list_add(&ctxt->rw_list, &cc->cc_rwctxts);
505 		cc->cc_sqecount += ret;
506 		if (write_len == seg->rs_length - info->wi_seg_off) {
507 			info->wi_seg_no++;
508 			info->wi_seg_off = 0;
509 		} else {
510 			info->wi_seg_off += write_len;
511 		}
512 		remaining -= write_len;
513 	} while (remaining);
514 
515 	return 0;
516 
517 out_overflow:
518 	trace_svcrdma_small_wrch_err(&cc->cc_cid, remaining, info->wi_seg_no,
519 				     info->wi_chunk->ch_segcount);
520 	return -E2BIG;
521 }
522 
523 /**
524  * svc_rdma_iov_write - Construct RDMA Writes from an iov
525  * @info: pointer to write arguments
526  * @iov: kvec to write
527  *
528  * Returns:
529  *   On success, returns zero
530  *   %-E2BIG if the client-provided Write chunk is too small
531  *   %-ENOMEM if a resource has been exhausted
532  *   %-EIO if an rdma-rw error occurred
533  */
svc_rdma_iov_write(struct svc_rdma_write_info * info,const struct kvec * iov)534 static int svc_rdma_iov_write(struct svc_rdma_write_info *info,
535 			      const struct kvec *iov)
536 {
537 	info->wi_base = iov->iov_base;
538 	return svc_rdma_build_writes(info, svc_rdma_vec_to_sg,
539 				     iov->iov_len);
540 }
541 
542 /**
543  * svc_rdma_pages_write - Construct RDMA Writes from pages
544  * @info: pointer to write arguments
545  * @xdr: xdr_buf with pages to write
546  * @offset: offset into the content of @xdr
547  * @length: number of bytes to write
548  *
549  * Returns:
550  *   On success, returns zero
551  *   %-E2BIG if the client-provided Write chunk is too small
552  *   %-ENOMEM if a resource has been exhausted
553  *   %-EIO if an rdma-rw error occurred
554  */
svc_rdma_pages_write(struct svc_rdma_write_info * info,const struct xdr_buf * xdr,unsigned int offset,unsigned long length)555 static int svc_rdma_pages_write(struct svc_rdma_write_info *info,
556 				const struct xdr_buf *xdr,
557 				unsigned int offset,
558 				unsigned long length)
559 {
560 	info->wi_xdr = xdr;
561 	info->wi_next_off = offset - xdr->head[0].iov_len;
562 	return svc_rdma_build_writes(info, svc_rdma_pagelist_to_sg,
563 				     length);
564 }
565 
566 /**
567  * svc_rdma_xb_write - Construct RDMA Writes to write an xdr_buf
568  * @xdr: xdr_buf to write
569  * @data: pointer to write arguments
570  *
571  * Returns:
572  *   On success, returns zero
573  *   %-E2BIG if the client-provided Write chunk is too small
574  *   %-ENOMEM if a resource has been exhausted
575  *   %-EIO if an rdma-rw error occurred
576  */
svc_rdma_xb_write(const struct xdr_buf * xdr,void * data)577 static int svc_rdma_xb_write(const struct xdr_buf *xdr, void *data)
578 {
579 	struct svc_rdma_write_info *info = data;
580 	int ret;
581 
582 	if (xdr->head[0].iov_len) {
583 		ret = svc_rdma_iov_write(info, &xdr->head[0]);
584 		if (ret < 0)
585 			return ret;
586 	}
587 
588 	if (xdr->page_len) {
589 		ret = svc_rdma_pages_write(info, xdr, xdr->head[0].iov_len,
590 					   xdr->page_len);
591 		if (ret < 0)
592 			return ret;
593 	}
594 
595 	if (xdr->tail[0].iov_len) {
596 		ret = svc_rdma_iov_write(info, &xdr->tail[0]);
597 		if (ret < 0)
598 			return ret;
599 	}
600 
601 	return xdr->len;
602 }
603 
svc_rdma_send_write_chunk(struct svcxprt_rdma * rdma,const struct svc_rdma_chunk * chunk,const struct xdr_buf * xdr)604 static int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma,
605 				     const struct svc_rdma_chunk *chunk,
606 				     const struct xdr_buf *xdr)
607 {
608 	struct svc_rdma_write_info *info;
609 	struct svc_rdma_chunk_ctxt *cc;
610 	struct xdr_buf payload;
611 	int ret;
612 
613 	if (xdr_buf_subsegment(xdr, &payload, chunk->ch_position,
614 			       chunk->ch_payload_length))
615 		return -EMSGSIZE;
616 
617 	info = svc_rdma_write_info_alloc(rdma, chunk);
618 	if (!info)
619 		return -ENOMEM;
620 	cc = &info->wi_cc;
621 
622 	ret = svc_rdma_xb_write(&payload, info);
623 	if (ret != payload.len)
624 		goto out_err;
625 
626 	trace_svcrdma_post_write_chunk(&cc->cc_cid, cc->cc_sqecount);
627 	ret = svc_rdma_post_chunk_ctxt(rdma, cc);
628 	if (ret < 0)
629 		goto out_err;
630 	return 0;
631 
632 out_err:
633 	svc_rdma_write_info_free(info);
634 	return ret;
635 }
636 
637 /**
638  * svc_rdma_send_write_list - Send all chunks on the Write list
639  * @rdma: controlling RDMA transport
640  * @rctxt: Write list provisioned by the client
641  * @xdr: xdr_buf containing an RPC Reply message
642  *
643  * Returns zero on success, or a negative errno if one or more
644  * Write chunks could not be sent.
645  */
svc_rdma_send_write_list(struct svcxprt_rdma * rdma,const struct svc_rdma_recv_ctxt * rctxt,const struct xdr_buf * xdr)646 int svc_rdma_send_write_list(struct svcxprt_rdma *rdma,
647 			     const struct svc_rdma_recv_ctxt *rctxt,
648 			     const struct xdr_buf *xdr)
649 {
650 	struct svc_rdma_chunk *chunk;
651 	int ret;
652 
653 	pcl_for_each_chunk(chunk, &rctxt->rc_write_pcl) {
654 		if (!chunk->ch_payload_length)
655 			break;
656 		ret = svc_rdma_send_write_chunk(rdma, chunk, xdr);
657 		if (ret < 0)
658 			return ret;
659 	}
660 	return 0;
661 }
662 
663 /**
664  * svc_rdma_prepare_reply_chunk - Construct WR chain for writing the Reply chunk
665  * @rdma: controlling RDMA transport
666  * @write_pcl: Write chunk list provided by client
667  * @reply_pcl: Reply chunk provided by client
668  * @sctxt: Send WR resources
669  * @xdr: xdr_buf containing an RPC Reply
670  *
671  * Returns a non-negative number of bytes the chunk consumed, or
672  *	%-E2BIG if the payload was larger than the Reply chunk,
673  *	%-EINVAL if client provided too many segments,
674  *	%-ENOMEM if rdma_rw context pool was exhausted,
675  *	%-ENOTCONN if posting failed (connection is lost),
676  *	%-EIO if rdma_rw initialization failed (DMA mapping, etc).
677  */
svc_rdma_prepare_reply_chunk(struct svcxprt_rdma * rdma,const struct svc_rdma_pcl * write_pcl,const struct svc_rdma_pcl * reply_pcl,struct svc_rdma_send_ctxt * sctxt,const struct xdr_buf * xdr)678 int svc_rdma_prepare_reply_chunk(struct svcxprt_rdma *rdma,
679 				 const struct svc_rdma_pcl *write_pcl,
680 				 const struct svc_rdma_pcl *reply_pcl,
681 				 struct svc_rdma_send_ctxt *sctxt,
682 				 const struct xdr_buf *xdr)
683 {
684 	struct svc_rdma_write_info *info = &sctxt->sc_reply_info;
685 	struct svc_rdma_chunk_ctxt *cc = &info->wi_cc;
686 	struct ib_send_wr *first_wr;
687 	struct list_head *pos;
688 	struct ib_cqe *cqe;
689 	int ret;
690 
691 	info->wi_rdma = rdma;
692 	info->wi_chunk = pcl_first_chunk(reply_pcl);
693 	info->wi_seg_off = 0;
694 	info->wi_seg_no = 0;
695 	info->wi_cc.cc_cqe.done = svc_rdma_reply_done;
696 
697 	ret = pcl_process_nonpayloads(write_pcl, xdr,
698 				      svc_rdma_xb_write, info);
699 	if (ret < 0)
700 		return ret;
701 
702 	first_wr = sctxt->sc_wr_chain;
703 	cqe = &cc->cc_cqe;
704 	list_for_each(pos, &cc->cc_rwctxts) {
705 		struct svc_rdma_rw_ctxt *rwc;
706 
707 		rwc = list_entry(pos, struct svc_rdma_rw_ctxt, rw_list);
708 		first_wr = rdma_rw_ctx_wrs(&rwc->rw_ctx, rdma->sc_qp,
709 					   rdma->sc_port_num, cqe, first_wr);
710 		cqe = NULL;
711 	}
712 	sctxt->sc_wr_chain = first_wr;
713 	sctxt->sc_sqecount += cc->cc_sqecount;
714 
715 	trace_svcrdma_post_reply_chunk(&cc->cc_cid, cc->cc_sqecount);
716 	return xdr->len;
717 }
718 
719 /**
720  * svc_rdma_build_read_segment - Build RDMA Read WQEs to pull one RDMA segment
721  * @rqstp: RPC transaction context
722  * @head: context for ongoing I/O
723  * @segment: co-ordinates of remote memory to be read
724  *
725  * Returns:
726  *   %0: the Read WR chain was constructed successfully
727  *   %-EINVAL: there were not enough rq_pages to finish
728  *   %-ENOMEM: allocating a local resources failed
729  *   %-EIO: a DMA mapping error occurred
730  */
svc_rdma_build_read_segment(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head,const struct svc_rdma_segment * segment)731 static int svc_rdma_build_read_segment(struct svc_rqst *rqstp,
732 				       struct svc_rdma_recv_ctxt *head,
733 				       const struct svc_rdma_segment *segment)
734 {
735 	struct svcxprt_rdma *rdma = svc_rdma_rqst_rdma(rqstp);
736 	struct svc_rdma_chunk_ctxt *cc = &head->rc_cc;
737 	unsigned int sge_no, seg_len, len;
738 	struct svc_rdma_rw_ctxt *ctxt;
739 	struct scatterlist *sg;
740 	int ret;
741 
742 	len = segment->rs_length;
743 	sge_no = PAGE_ALIGN(head->rc_pageoff + len) >> PAGE_SHIFT;
744 	ctxt = svc_rdma_get_rw_ctxt(rdma, sge_no);
745 	if (!ctxt)
746 		return -ENOMEM;
747 	ctxt->rw_nents = sge_no;
748 
749 	sg = ctxt->rw_sg_table.sgl;
750 	for (sge_no = 0; sge_no < ctxt->rw_nents; sge_no++) {
751 		seg_len = min_t(unsigned int, len,
752 				PAGE_SIZE - head->rc_pageoff);
753 
754 		if (!head->rc_pageoff)
755 			head->rc_page_count++;
756 
757 		sg_set_page(sg, rqstp->rq_pages[head->rc_curpage],
758 			    seg_len, head->rc_pageoff);
759 		sg = sg_next(sg);
760 
761 		head->rc_pageoff += seg_len;
762 		if (head->rc_pageoff == PAGE_SIZE) {
763 			head->rc_curpage++;
764 			head->rc_pageoff = 0;
765 		}
766 		len -= seg_len;
767 
768 		if (len && ((head->rc_curpage + 1) > ARRAY_SIZE(rqstp->rq_pages)))
769 			goto out_overrun;
770 	}
771 
772 	ret = svc_rdma_rw_ctx_init(rdma, ctxt, segment->rs_offset,
773 				   segment->rs_handle, DMA_FROM_DEVICE);
774 	if (ret < 0)
775 		return -EIO;
776 	percpu_counter_inc(&svcrdma_stat_read);
777 
778 	list_add(&ctxt->rw_list, &cc->cc_rwctxts);
779 	cc->cc_sqecount += ret;
780 	return 0;
781 
782 out_overrun:
783 	trace_svcrdma_page_overrun_err(&cc->cc_cid, head->rc_curpage);
784 	return -EINVAL;
785 }
786 
787 /**
788  * svc_rdma_build_read_chunk - Build RDMA Read WQEs to pull one RDMA chunk
789  * @rqstp: RPC transaction context
790  * @head: context for ongoing I/O
791  * @chunk: Read chunk to pull
792  *
793  * Return values:
794  *   %0: the Read WR chain was constructed successfully
795  *   %-EINVAL: there were not enough resources to finish
796  *   %-ENOMEM: allocating a local resources failed
797  *   %-EIO: a DMA mapping error occurred
798  */
svc_rdma_build_read_chunk(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head,const struct svc_rdma_chunk * chunk)799 static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp,
800 				     struct svc_rdma_recv_ctxt *head,
801 				     const struct svc_rdma_chunk *chunk)
802 {
803 	const struct svc_rdma_segment *segment;
804 	int ret;
805 
806 	ret = -EINVAL;
807 	pcl_for_each_segment(segment, chunk) {
808 		ret = svc_rdma_build_read_segment(rqstp, head, segment);
809 		if (ret < 0)
810 			break;
811 		head->rc_readbytes += segment->rs_length;
812 	}
813 	return ret;
814 }
815 
816 /**
817  * svc_rdma_copy_inline_range - Copy part of the inline content into pages
818  * @rqstp: RPC transaction context
819  * @head: context for ongoing I/O
820  * @offset: offset into the Receive buffer of region to copy
821  * @remaining: length of region to copy
822  *
823  * Take a page at a time from rqstp->rq_pages and copy the inline
824  * content from the Receive buffer into that page. Update
825  * head->rc_curpage and head->rc_pageoff so that the next RDMA Read
826  * result will land contiguously with the copied content.
827  *
828  * Return values:
829  *   %0: Inline content was successfully copied
830  *   %-EINVAL: offset or length was incorrect
831  */
svc_rdma_copy_inline_range(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head,unsigned int offset,unsigned int remaining)832 static int svc_rdma_copy_inline_range(struct svc_rqst *rqstp,
833 				      struct svc_rdma_recv_ctxt *head,
834 				      unsigned int offset,
835 				      unsigned int remaining)
836 {
837 	unsigned char *dst, *src = head->rc_recv_buf;
838 	unsigned int page_no, numpages;
839 
840 	numpages = PAGE_ALIGN(head->rc_pageoff + remaining) >> PAGE_SHIFT;
841 	for (page_no = 0; page_no < numpages; page_no++) {
842 		unsigned int page_len;
843 
844 		page_len = min_t(unsigned int, remaining,
845 				 PAGE_SIZE - head->rc_pageoff);
846 
847 		if (!head->rc_pageoff)
848 			head->rc_page_count++;
849 
850 		dst = page_address(rqstp->rq_pages[head->rc_curpage]);
851 		memcpy(dst + head->rc_curpage, src + offset, page_len);
852 
853 		head->rc_readbytes += page_len;
854 		head->rc_pageoff += page_len;
855 		if (head->rc_pageoff == PAGE_SIZE) {
856 			head->rc_curpage++;
857 			head->rc_pageoff = 0;
858 		}
859 		remaining -= page_len;
860 		offset += page_len;
861 	}
862 
863 	return -EINVAL;
864 }
865 
866 /**
867  * svc_rdma_read_multiple_chunks - Construct RDMA Reads to pull data item Read chunks
868  * @rqstp: RPC transaction context
869  * @head: context for ongoing I/O
870  *
871  * The chunk data lands in rqstp->rq_arg as a series of contiguous pages,
872  * like an incoming TCP call.
873  *
874  * Return values:
875  *   %0: RDMA Read WQEs were successfully built
876  *   %-EINVAL: client provided too many chunks or segments,
877  *   %-ENOMEM: rdma_rw context pool was exhausted,
878  *   %-ENOTCONN: posting failed (connection is lost),
879  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
880  */
881 static noinline int
svc_rdma_read_multiple_chunks(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)882 svc_rdma_read_multiple_chunks(struct svc_rqst *rqstp,
883 			      struct svc_rdma_recv_ctxt *head)
884 {
885 	const struct svc_rdma_pcl *pcl = &head->rc_read_pcl;
886 	struct svc_rdma_chunk *chunk, *next;
887 	unsigned int start, length;
888 	int ret;
889 
890 	start = 0;
891 	chunk = pcl_first_chunk(pcl);
892 	length = chunk->ch_position;
893 	ret = svc_rdma_copy_inline_range(rqstp, head, start, length);
894 	if (ret < 0)
895 		return ret;
896 
897 	pcl_for_each_chunk(chunk, pcl) {
898 		ret = svc_rdma_build_read_chunk(rqstp, head, chunk);
899 		if (ret < 0)
900 			return ret;
901 
902 		next = pcl_next_chunk(pcl, chunk);
903 		if (!next)
904 			break;
905 
906 		start += length;
907 		length = next->ch_position - head->rc_readbytes;
908 		ret = svc_rdma_copy_inline_range(rqstp, head, start, length);
909 		if (ret < 0)
910 			return ret;
911 	}
912 
913 	start += length;
914 	length = head->rc_byte_len - start;
915 	return svc_rdma_copy_inline_range(rqstp, head, start, length);
916 }
917 
918 /**
919  * svc_rdma_read_data_item - Construct RDMA Reads to pull data item Read chunks
920  * @rqstp: RPC transaction context
921  * @head: context for ongoing I/O
922  *
923  * The chunk data lands in the page list of rqstp->rq_arg.pages.
924  *
925  * Currently NFSD does not look at the rqstp->rq_arg.tail[0] kvec.
926  * Therefore, XDR round-up of the Read chunk and trailing
927  * inline content must both be added at the end of the pagelist.
928  *
929  * Return values:
930  *   %0: RDMA Read WQEs were successfully built
931  *   %-EINVAL: client provided too many chunks or segments,
932  *   %-ENOMEM: rdma_rw context pool was exhausted,
933  *   %-ENOTCONN: posting failed (connection is lost),
934  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
935  */
svc_rdma_read_data_item(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)936 static int svc_rdma_read_data_item(struct svc_rqst *rqstp,
937 				   struct svc_rdma_recv_ctxt *head)
938 {
939 	return svc_rdma_build_read_chunk(rqstp, head,
940 					 pcl_first_chunk(&head->rc_read_pcl));
941 }
942 
943 /**
944  * svc_rdma_read_chunk_range - Build RDMA Read WRs for portion of a chunk
945  * @rqstp: RPC transaction context
946  * @head: context for ongoing I/O
947  * @chunk: parsed Call chunk to pull
948  * @offset: offset of region to pull
949  * @length: length of region to pull
950  *
951  * Return values:
952  *   %0: RDMA Read WQEs were successfully built
953  *   %-EINVAL: there were not enough resources to finish
954  *   %-ENOMEM: rdma_rw context pool was exhausted,
955  *   %-ENOTCONN: posting failed (connection is lost),
956  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
957  */
svc_rdma_read_chunk_range(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head,const struct svc_rdma_chunk * chunk,unsigned int offset,unsigned int length)958 static int svc_rdma_read_chunk_range(struct svc_rqst *rqstp,
959 				     struct svc_rdma_recv_ctxt *head,
960 				     const struct svc_rdma_chunk *chunk,
961 				     unsigned int offset, unsigned int length)
962 {
963 	const struct svc_rdma_segment *segment;
964 	int ret;
965 
966 	ret = -EINVAL;
967 	pcl_for_each_segment(segment, chunk) {
968 		struct svc_rdma_segment dummy;
969 
970 		if (offset > segment->rs_length) {
971 			offset -= segment->rs_length;
972 			continue;
973 		}
974 
975 		dummy.rs_handle = segment->rs_handle;
976 		dummy.rs_length = min_t(u32, length, segment->rs_length) - offset;
977 		dummy.rs_offset = segment->rs_offset + offset;
978 
979 		ret = svc_rdma_build_read_segment(rqstp, head, &dummy);
980 		if (ret < 0)
981 			break;
982 
983 		head->rc_readbytes += dummy.rs_length;
984 		length -= dummy.rs_length;
985 		offset = 0;
986 	}
987 	return ret;
988 }
989 
990 /**
991  * svc_rdma_read_call_chunk - Build RDMA Read WQEs to pull a Long Message
992  * @rqstp: RPC transaction context
993  * @head: context for ongoing I/O
994  *
995  * Return values:
996  *   %0: RDMA Read WQEs were successfully built
997  *   %-EINVAL: there were not enough resources to finish
998  *   %-ENOMEM: rdma_rw context pool was exhausted,
999  *   %-ENOTCONN: posting failed (connection is lost),
1000  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
1001  */
svc_rdma_read_call_chunk(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)1002 static int svc_rdma_read_call_chunk(struct svc_rqst *rqstp,
1003 				    struct svc_rdma_recv_ctxt *head)
1004 {
1005 	const struct svc_rdma_chunk *call_chunk =
1006 			pcl_first_chunk(&head->rc_call_pcl);
1007 	const struct svc_rdma_pcl *pcl = &head->rc_read_pcl;
1008 	struct svc_rdma_chunk *chunk, *next;
1009 	unsigned int start, length;
1010 	int ret;
1011 
1012 	if (pcl_is_empty(pcl))
1013 		return svc_rdma_build_read_chunk(rqstp, head, call_chunk);
1014 
1015 	start = 0;
1016 	chunk = pcl_first_chunk(pcl);
1017 	length = chunk->ch_position;
1018 	ret = svc_rdma_read_chunk_range(rqstp, head, call_chunk,
1019 					start, length);
1020 	if (ret < 0)
1021 		return ret;
1022 
1023 	pcl_for_each_chunk(chunk, pcl) {
1024 		ret = svc_rdma_build_read_chunk(rqstp, head, chunk);
1025 		if (ret < 0)
1026 			return ret;
1027 
1028 		next = pcl_next_chunk(pcl, chunk);
1029 		if (!next)
1030 			break;
1031 
1032 		start += length;
1033 		length = next->ch_position - head->rc_readbytes;
1034 		ret = svc_rdma_read_chunk_range(rqstp, head, call_chunk,
1035 						start, length);
1036 		if (ret < 0)
1037 			return ret;
1038 	}
1039 
1040 	start += length;
1041 	length = call_chunk->ch_length - start;
1042 	return svc_rdma_read_chunk_range(rqstp, head, call_chunk,
1043 					 start, length);
1044 }
1045 
1046 /**
1047  * svc_rdma_read_special - Build RDMA Read WQEs to pull a Long Message
1048  * @rqstp: RPC transaction context
1049  * @head: context for ongoing I/O
1050  *
1051  * The start of the data lands in the first page just after the
1052  * Transport header, and the rest lands in rqstp->rq_arg.pages.
1053  *
1054  * Assumptions:
1055  *	- A PZRC is never sent in an RDMA_MSG message, though it's
1056  *	  allowed by spec.
1057  *
1058  * Return values:
1059  *   %0: RDMA Read WQEs were successfully built
1060  *   %-EINVAL: client provided too many chunks or segments,
1061  *   %-ENOMEM: rdma_rw context pool was exhausted,
1062  *   %-ENOTCONN: posting failed (connection is lost),
1063  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
1064  */
svc_rdma_read_special(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)1065 static noinline int svc_rdma_read_special(struct svc_rqst *rqstp,
1066 					  struct svc_rdma_recv_ctxt *head)
1067 {
1068 	return svc_rdma_read_call_chunk(rqstp, head);
1069 }
1070 
1071 /* Pages under I/O have been copied to head->rc_pages. Ensure that
1072  * svc_xprt_release() does not put them when svc_rdma_recvfrom()
1073  * returns. This has to be done after all Read WRs are constructed
1074  * to properly handle a page that happens to be part of I/O on behalf
1075  * of two different RDMA segments.
1076  *
1077  * Note: if the subsequent post_send fails, these pages have already
1078  * been moved to head->rc_pages and thus will be cleaned up by
1079  * svc_rdma_recv_ctxt_put().
1080  */
svc_rdma_clear_rqst_pages(struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)1081 static void svc_rdma_clear_rqst_pages(struct svc_rqst *rqstp,
1082 				      struct svc_rdma_recv_ctxt *head)
1083 {
1084 	unsigned int i;
1085 
1086 	for (i = 0; i < head->rc_page_count; i++) {
1087 		head->rc_pages[i] = rqstp->rq_pages[i];
1088 		rqstp->rq_pages[i] = NULL;
1089 	}
1090 }
1091 
1092 /**
1093  * svc_rdma_process_read_list - Pull list of Read chunks from the client
1094  * @rdma: controlling RDMA transport
1095  * @rqstp: set of pages to use as Read sink buffers
1096  * @head: pages under I/O collect here
1097  *
1098  * The RPC/RDMA protocol assumes that the upper layer's XDR decoders
1099  * pull each Read chunk as they decode an incoming RPC message.
1100  *
1101  * On Linux, however, the server needs to have a fully-constructed RPC
1102  * message in rqstp->rq_arg when there is a positive return code from
1103  * ->xpo_recvfrom. So the Read list is safety-checked immediately when
1104  * it is received, then here the whole Read list is pulled all at once.
1105  * The ingress RPC message is fully reconstructed once all associated
1106  * RDMA Reads have completed.
1107  *
1108  * Return values:
1109  *   %1: all needed RDMA Reads were posted successfully,
1110  *   %-EINVAL: client provided too many chunks or segments,
1111  *   %-ENOMEM: rdma_rw context pool was exhausted,
1112  *   %-ENOTCONN: posting failed (connection is lost),
1113  *   %-EIO: rdma_rw initialization failed (DMA mapping, etc).
1114  */
svc_rdma_process_read_list(struct svcxprt_rdma * rdma,struct svc_rqst * rqstp,struct svc_rdma_recv_ctxt * head)1115 int svc_rdma_process_read_list(struct svcxprt_rdma *rdma,
1116 			       struct svc_rqst *rqstp,
1117 			       struct svc_rdma_recv_ctxt *head)
1118 {
1119 	struct svc_rdma_chunk_ctxt *cc = &head->rc_cc;
1120 	int ret;
1121 
1122 	cc->cc_cqe.done = svc_rdma_wc_read_done;
1123 	cc->cc_sqecount = 0;
1124 	head->rc_pageoff = 0;
1125 	head->rc_curpage = 0;
1126 	head->rc_readbytes = 0;
1127 
1128 	if (pcl_is_empty(&head->rc_call_pcl)) {
1129 		if (head->rc_read_pcl.cl_count == 1)
1130 			ret = svc_rdma_read_data_item(rqstp, head);
1131 		else
1132 			ret = svc_rdma_read_multiple_chunks(rqstp, head);
1133 	} else
1134 		ret = svc_rdma_read_special(rqstp, head);
1135 	svc_rdma_clear_rqst_pages(rqstp, head);
1136 	if (ret < 0)
1137 		return ret;
1138 
1139 	trace_svcrdma_post_read_chunk(&cc->cc_cid, cc->cc_sqecount);
1140 	ret = svc_rdma_post_chunk_ctxt(rdma, cc);
1141 	return ret < 0 ? ret : 1;
1142 }
1143