Loading...
1/*
2 * Copyright (c) 2003-2007 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
40/*
41 * transport.c
42 *
43 * This file contains the top-level implementation of an RPC RDMA
44 * transport.
45 *
46 * Naming convention: functions beginning with xprt_ are part of the
47 * transport switch. All others are RPC RDMA internal.
48 */
49
50#include <linux/module.h>
51#include <linux/slab.h>
52#include <linux/seq_file.h>
53#include <linux/sunrpc/addr.h>
54
55#include "xprt_rdma.h"
56
57#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
58# define RPCDBG_FACILITY RPCDBG_TRANS
59#endif
60
61/*
62 * tunables
63 */
64
65static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
66unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
67static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
68static unsigned int xprt_rdma_inline_write_padding;
69static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR;
70 int xprt_rdma_pad_optimize = 1;
71
72#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
73
74static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
75static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
76static unsigned int zero;
77static unsigned int max_padding = PAGE_SIZE;
78static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
79static unsigned int max_memreg = RPCRDMA_LAST - 1;
80
81static struct ctl_table_header *sunrpc_table_header;
82
83static struct ctl_table xr_tunables_table[] = {
84 {
85 .procname = "rdma_slot_table_entries",
86 .data = &xprt_rdma_slot_table_entries,
87 .maxlen = sizeof(unsigned int),
88 .mode = 0644,
89 .proc_handler = proc_dointvec_minmax,
90 .extra1 = &min_slot_table_size,
91 .extra2 = &max_slot_table_size
92 },
93 {
94 .procname = "rdma_max_inline_read",
95 .data = &xprt_rdma_max_inline_read,
96 .maxlen = sizeof(unsigned int),
97 .mode = 0644,
98 .proc_handler = proc_dointvec,
99 },
100 {
101 .procname = "rdma_max_inline_write",
102 .data = &xprt_rdma_max_inline_write,
103 .maxlen = sizeof(unsigned int),
104 .mode = 0644,
105 .proc_handler = proc_dointvec,
106 },
107 {
108 .procname = "rdma_inline_write_padding",
109 .data = &xprt_rdma_inline_write_padding,
110 .maxlen = sizeof(unsigned int),
111 .mode = 0644,
112 .proc_handler = proc_dointvec_minmax,
113 .extra1 = &zero,
114 .extra2 = &max_padding,
115 },
116 {
117 .procname = "rdma_memreg_strategy",
118 .data = &xprt_rdma_memreg_strategy,
119 .maxlen = sizeof(unsigned int),
120 .mode = 0644,
121 .proc_handler = proc_dointvec_minmax,
122 .extra1 = &min_memreg,
123 .extra2 = &max_memreg,
124 },
125 {
126 .procname = "rdma_pad_optimize",
127 .data = &xprt_rdma_pad_optimize,
128 .maxlen = sizeof(unsigned int),
129 .mode = 0644,
130 .proc_handler = proc_dointvec,
131 },
132 { },
133};
134
135static struct ctl_table sunrpc_table[] = {
136 {
137 .procname = "sunrpc",
138 .mode = 0555,
139 .child = xr_tunables_table
140 },
141 { },
142};
143
144#endif
145
146static struct rpc_xprt_ops xprt_rdma_procs; /*forward reference */
147
148static void
149xprt_rdma_format_addresses4(struct rpc_xprt *xprt, struct sockaddr *sap)
150{
151 struct sockaddr_in *sin = (struct sockaddr_in *)sap;
152 char buf[20];
153
154 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
155 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
156
157 xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA;
158}
159
160static void
161xprt_rdma_format_addresses6(struct rpc_xprt *xprt, struct sockaddr *sap)
162{
163 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
164 char buf[40];
165
166 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
167 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
168
169 xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA6;
170}
171
172void
173xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap)
174{
175 char buf[128];
176
177 switch (sap->sa_family) {
178 case AF_INET:
179 xprt_rdma_format_addresses4(xprt, sap);
180 break;
181 case AF_INET6:
182 xprt_rdma_format_addresses6(xprt, sap);
183 break;
184 default:
185 pr_err("rpcrdma: Unrecognized address family\n");
186 return;
187 }
188
189 (void)rpc_ntop(sap, buf, sizeof(buf));
190 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
191
192 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
193 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
194
195 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
196 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
197
198 xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
199}
200
201void
202xprt_rdma_free_addresses(struct rpc_xprt *xprt)
203{
204 unsigned int i;
205
206 for (i = 0; i < RPC_DISPLAY_MAX; i++)
207 switch (i) {
208 case RPC_DISPLAY_PROTO:
209 case RPC_DISPLAY_NETID:
210 continue;
211 default:
212 kfree(xprt->address_strings[i]);
213 }
214}
215
216static void
217xprt_rdma_connect_worker(struct work_struct *work)
218{
219 struct rpcrdma_xprt *r_xprt = container_of(work, struct rpcrdma_xprt,
220 rx_connect_worker.work);
221 struct rpc_xprt *xprt = &r_xprt->rx_xprt;
222 int rc = 0;
223
224 xprt_clear_connected(xprt);
225
226 dprintk("RPC: %s: %sconnect\n", __func__,
227 r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
228 rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
229 if (rc)
230 xprt_wake_pending_tasks(xprt, rc);
231
232 dprintk("RPC: %s: exit\n", __func__);
233 xprt_clear_connecting(xprt);
234}
235
236static void
237xprt_rdma_inject_disconnect(struct rpc_xprt *xprt)
238{
239 struct rpcrdma_xprt *r_xprt = container_of(xprt, struct rpcrdma_xprt,
240 rx_xprt);
241
242 pr_info("rpcrdma: injecting transport disconnect on xprt=%p\n", xprt);
243 rdma_disconnect(r_xprt->rx_ia.ri_id);
244}
245
246/*
247 * xprt_rdma_destroy
248 *
249 * Destroy the xprt.
250 * Free all memory associated with the object, including its own.
251 * NOTE: none of the *destroy methods free memory for their top-level
252 * objects, even though they may have allocated it (they do free
253 * private memory). It's up to the caller to handle it. In this
254 * case (RDMA transport), all structure memory is inlined with the
255 * struct rpcrdma_xprt.
256 */
257static void
258xprt_rdma_destroy(struct rpc_xprt *xprt)
259{
260 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
261
262 dprintk("RPC: %s: called\n", __func__);
263
264 cancel_delayed_work_sync(&r_xprt->rx_connect_worker);
265
266 xprt_clear_connected(xprt);
267
268 rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
269 rpcrdma_buffer_destroy(&r_xprt->rx_buf);
270 rpcrdma_ia_close(&r_xprt->rx_ia);
271
272 xprt_rdma_free_addresses(xprt);
273
274 xprt_free(xprt);
275
276 dprintk("RPC: %s: returning\n", __func__);
277
278 module_put(THIS_MODULE);
279}
280
281static const struct rpc_timeout xprt_rdma_default_timeout = {
282 .to_initval = 60 * HZ,
283 .to_maxval = 60 * HZ,
284};
285
286/**
287 * xprt_setup_rdma - Set up transport to use RDMA
288 *
289 * @args: rpc transport arguments
290 */
291static struct rpc_xprt *
292xprt_setup_rdma(struct xprt_create *args)
293{
294 struct rpcrdma_create_data_internal cdata;
295 struct rpc_xprt *xprt;
296 struct rpcrdma_xprt *new_xprt;
297 struct rpcrdma_ep *new_ep;
298 struct sockaddr *sap;
299 int rc;
300
301 if (args->addrlen > sizeof(xprt->addr)) {
302 dprintk("RPC: %s: address too large\n", __func__);
303 return ERR_PTR(-EBADF);
304 }
305
306 xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt),
307 xprt_rdma_slot_table_entries,
308 xprt_rdma_slot_table_entries);
309 if (xprt == NULL) {
310 dprintk("RPC: %s: couldn't allocate rpcrdma_xprt\n",
311 __func__);
312 return ERR_PTR(-ENOMEM);
313 }
314
315 /* 60 second timeout, no retries */
316 xprt->timeout = &xprt_rdma_default_timeout;
317 xprt->bind_timeout = RPCRDMA_BIND_TO;
318 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
319 xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO;
320
321 xprt->resvport = 0; /* privileged port not needed */
322 xprt->tsh_size = 0; /* RPC-RDMA handles framing */
323 xprt->ops = &xprt_rdma_procs;
324
325 /*
326 * Set up RDMA-specific connect data.
327 */
328
329 sap = (struct sockaddr *)&cdata.addr;
330 memcpy(sap, args->dstaddr, args->addrlen);
331
332 /* Ensure xprt->addr holds valid server TCP (not RDMA)
333 * address, for any side protocols which peek at it */
334 xprt->prot = IPPROTO_TCP;
335 xprt->addrlen = args->addrlen;
336 memcpy(&xprt->addr, sap, xprt->addrlen);
337
338 if (rpc_get_port(sap))
339 xprt_set_bound(xprt);
340
341 cdata.max_requests = xprt->max_reqs;
342
343 cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
344 cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
345
346 cdata.inline_wsize = xprt_rdma_max_inline_write;
347 if (cdata.inline_wsize > cdata.wsize)
348 cdata.inline_wsize = cdata.wsize;
349
350 cdata.inline_rsize = xprt_rdma_max_inline_read;
351 if (cdata.inline_rsize > cdata.rsize)
352 cdata.inline_rsize = cdata.rsize;
353
354 cdata.padding = xprt_rdma_inline_write_padding;
355
356 /*
357 * Create new transport instance, which includes initialized
358 * o ia
359 * o endpoint
360 * o buffers
361 */
362
363 new_xprt = rpcx_to_rdmax(xprt);
364
365 rc = rpcrdma_ia_open(new_xprt, sap, xprt_rdma_memreg_strategy);
366 if (rc)
367 goto out1;
368
369 /*
370 * initialize and create ep
371 */
372 new_xprt->rx_data = cdata;
373 new_ep = &new_xprt->rx_ep;
374 new_ep->rep_remote_addr = cdata.addr;
375
376 rc = rpcrdma_ep_create(&new_xprt->rx_ep,
377 &new_xprt->rx_ia, &new_xprt->rx_data);
378 if (rc)
379 goto out2;
380
381 /*
382 * Allocate pre-registered send and receive buffers for headers and
383 * any inline data. Also specify any padding which will be provided
384 * from a preregistered zero buffer.
385 */
386 rc = rpcrdma_buffer_create(new_xprt);
387 if (rc)
388 goto out3;
389
390 /*
391 * Register a callback for connection events. This is necessary because
392 * connection loss notification is async. We also catch connection loss
393 * when reaping receives.
394 */
395 INIT_DELAYED_WORK(&new_xprt->rx_connect_worker,
396 xprt_rdma_connect_worker);
397
398 xprt_rdma_format_addresses(xprt, sap);
399 xprt->max_payload = new_xprt->rx_ia.ri_ops->ro_maxpages(new_xprt);
400 if (xprt->max_payload == 0)
401 goto out4;
402 xprt->max_payload <<= PAGE_SHIFT;
403 dprintk("RPC: %s: transport data payload maximum: %zu bytes\n",
404 __func__, xprt->max_payload);
405
406 if (!try_module_get(THIS_MODULE))
407 goto out4;
408
409 dprintk("RPC: %s: %s:%s\n", __func__,
410 xprt->address_strings[RPC_DISPLAY_ADDR],
411 xprt->address_strings[RPC_DISPLAY_PORT]);
412 return xprt;
413
414out4:
415 xprt_rdma_free_addresses(xprt);
416 rc = -EINVAL;
417out3:
418 rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
419out2:
420 rpcrdma_ia_close(&new_xprt->rx_ia);
421out1:
422 xprt_free(xprt);
423 return ERR_PTR(rc);
424}
425
426/*
427 * Close a connection, during shutdown or timeout/reconnect
428 */
429static void
430xprt_rdma_close(struct rpc_xprt *xprt)
431{
432 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
433
434 dprintk("RPC: %s: closing\n", __func__);
435 if (r_xprt->rx_ep.rep_connected > 0)
436 xprt->reestablish_timeout = 0;
437 xprt_disconnect_done(xprt);
438 rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
439}
440
441static void
442xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
443{
444 struct sockaddr_in *sap;
445
446 sap = (struct sockaddr_in *)&xprt->addr;
447 sap->sin_port = htons(port);
448 sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
449 sap->sin_port = htons(port);
450 dprintk("RPC: %s: %u\n", __func__, port);
451}
452
453static void
454xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task)
455{
456 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
457
458 if (r_xprt->rx_ep.rep_connected != 0) {
459 /* Reconnect */
460 schedule_delayed_work(&r_xprt->rx_connect_worker,
461 xprt->reestablish_timeout);
462 xprt->reestablish_timeout <<= 1;
463 if (xprt->reestablish_timeout > RPCRDMA_MAX_REEST_TO)
464 xprt->reestablish_timeout = RPCRDMA_MAX_REEST_TO;
465 else if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO)
466 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
467 } else {
468 schedule_delayed_work(&r_xprt->rx_connect_worker, 0);
469 if (!RPC_IS_ASYNC(task))
470 flush_delayed_work(&r_xprt->rx_connect_worker);
471 }
472}
473
474/*
475 * The RDMA allocate/free functions need the task structure as a place
476 * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
477 * sequence.
478 *
479 * The RPC layer allocates both send and receive buffers in the same call
480 * (rq_send_buf and rq_rcv_buf are both part of a single contiguous buffer).
481 * We may register rq_rcv_buf when using reply chunks.
482 */
483static void *
484xprt_rdma_allocate(struct rpc_task *task, size_t size)
485{
486 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
487 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
488 struct rpcrdma_regbuf *rb;
489 struct rpcrdma_req *req;
490 size_t min_size;
491 gfp_t flags;
492
493 req = rpcrdma_buffer_get(&r_xprt->rx_buf);
494 if (req == NULL)
495 return NULL;
496
497 flags = RPCRDMA_DEF_GFP;
498 if (RPC_IS_SWAPPER(task))
499 flags = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
500
501 if (req->rl_rdmabuf == NULL)
502 goto out_rdmabuf;
503 if (req->rl_sendbuf == NULL)
504 goto out_sendbuf;
505 if (size > req->rl_sendbuf->rg_size)
506 goto out_sendbuf;
507
508out:
509 dprintk("RPC: %s: size %zd, request 0x%p\n", __func__, size, req);
510 req->rl_connect_cookie = 0; /* our reserved value */
511 return req->rl_sendbuf->rg_base;
512
513out_rdmabuf:
514 min_size = RPCRDMA_INLINE_WRITE_THRESHOLD(task->tk_rqstp);
515 rb = rpcrdma_alloc_regbuf(&r_xprt->rx_ia, min_size, flags);
516 if (IS_ERR(rb))
517 goto out_fail;
518 req->rl_rdmabuf = rb;
519
520out_sendbuf:
521 /* XDR encoding and RPC/RDMA marshaling of this request has not
522 * yet occurred. Thus a lower bound is needed to prevent buffer
523 * overrun during marshaling.
524 *
525 * RPC/RDMA marshaling may choose to send payload bearing ops
526 * inline, if the result is smaller than the inline threshold.
527 * The value of the "size" argument accounts for header
528 * requirements but not for the payload in these cases.
529 *
530 * Likewise, allocate enough space to receive a reply up to the
531 * size of the inline threshold.
532 *
533 * It's unlikely that both the send header and the received
534 * reply will be large, but slush is provided here to allow
535 * flexibility when marshaling.
536 */
537 min_size = RPCRDMA_INLINE_READ_THRESHOLD(task->tk_rqstp);
538 min_size += RPCRDMA_INLINE_WRITE_THRESHOLD(task->tk_rqstp);
539 if (size < min_size)
540 size = min_size;
541
542 rb = rpcrdma_alloc_regbuf(&r_xprt->rx_ia, size, flags);
543 if (IS_ERR(rb))
544 goto out_fail;
545 rb->rg_owner = req;
546
547 r_xprt->rx_stats.hardway_register_count += size;
548 rpcrdma_free_regbuf(&r_xprt->rx_ia, req->rl_sendbuf);
549 req->rl_sendbuf = rb;
550 goto out;
551
552out_fail:
553 rpcrdma_buffer_put(req);
554 r_xprt->rx_stats.failed_marshal_count++;
555 return NULL;
556}
557
558/*
559 * This function returns all RDMA resources to the pool.
560 */
561static void
562xprt_rdma_free(void *buffer)
563{
564 struct rpcrdma_req *req;
565 struct rpcrdma_xprt *r_xprt;
566 struct rpcrdma_regbuf *rb;
567 int i;
568
569 if (buffer == NULL)
570 return;
571
572 rb = container_of(buffer, struct rpcrdma_regbuf, rg_base[0]);
573 req = rb->rg_owner;
574 if (req->rl_backchannel)
575 return;
576
577 r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
578
579 dprintk("RPC: %s: called on 0x%p\n", __func__, req->rl_reply);
580
581 for (i = 0; req->rl_nchunks;) {
582 --req->rl_nchunks;
583 i += r_xprt->rx_ia.ri_ops->ro_unmap(r_xprt,
584 &req->rl_segments[i]);
585 }
586
587 rpcrdma_buffer_put(req);
588}
589
590/*
591 * send_request invokes the meat of RPC RDMA. It must do the following:
592 * 1. Marshal the RPC request into an RPC RDMA request, which means
593 * putting a header in front of data, and creating IOVs for RDMA
594 * from those in the request.
595 * 2. In marshaling, detect opportunities for RDMA, and use them.
596 * 3. Post a recv message to set up asynch completion, then send
597 * the request (rpcrdma_ep_post).
598 * 4. No partial sends are possible in the RPC-RDMA protocol (as in UDP).
599 */
600
601static int
602xprt_rdma_send_request(struct rpc_task *task)
603{
604 struct rpc_rqst *rqst = task->tk_rqstp;
605 struct rpc_xprt *xprt = rqst->rq_xprt;
606 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
607 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
608 int rc = 0;
609
610 rc = rpcrdma_marshal_req(rqst);
611 if (rc < 0)
612 goto failed_marshal;
613
614 if (req->rl_reply == NULL) /* e.g. reconnection */
615 rpcrdma_recv_buffer_get(req);
616
617 /* Must suppress retransmit to maintain credits */
618 if (req->rl_connect_cookie == xprt->connect_cookie)
619 goto drop_connection;
620 req->rl_connect_cookie = xprt->connect_cookie;
621
622 if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
623 goto drop_connection;
624
625 rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
626 rqst->rq_bytes_sent = 0;
627 return 0;
628
629failed_marshal:
630 r_xprt->rx_stats.failed_marshal_count++;
631 dprintk("RPC: %s: rpcrdma_marshal_req failed, status %i\n",
632 __func__, rc);
633 if (rc == -EIO)
634 return -EIO;
635drop_connection:
636 xprt_disconnect_done(xprt);
637 return -ENOTCONN; /* implies disconnect */
638}
639
640void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
641{
642 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
643 long idle_time = 0;
644
645 if (xprt_connected(xprt))
646 idle_time = (long)(jiffies - xprt->last_used) / HZ;
647
648 seq_puts(seq, "\txprt:\trdma ");
649 seq_printf(seq, "%u %lu %lu %lu %ld %lu %lu %lu %llu %llu ",
650 0, /* need a local port? */
651 xprt->stat.bind_count,
652 xprt->stat.connect_count,
653 xprt->stat.connect_time,
654 idle_time,
655 xprt->stat.sends,
656 xprt->stat.recvs,
657 xprt->stat.bad_xids,
658 xprt->stat.req_u,
659 xprt->stat.bklog_u);
660 seq_printf(seq, "%lu %lu %lu %llu %llu %llu %llu %lu %lu %lu %lu\n",
661 r_xprt->rx_stats.read_chunk_count,
662 r_xprt->rx_stats.write_chunk_count,
663 r_xprt->rx_stats.reply_chunk_count,
664 r_xprt->rx_stats.total_rdma_request,
665 r_xprt->rx_stats.total_rdma_reply,
666 r_xprt->rx_stats.pullup_copy_count,
667 r_xprt->rx_stats.fixup_copy_count,
668 r_xprt->rx_stats.hardway_register_count,
669 r_xprt->rx_stats.failed_marshal_count,
670 r_xprt->rx_stats.bad_reply_count,
671 r_xprt->rx_stats.nomsg_call_count);
672}
673
674static int
675xprt_rdma_enable_swap(struct rpc_xprt *xprt)
676{
677 return 0;
678}
679
680static void
681xprt_rdma_disable_swap(struct rpc_xprt *xprt)
682{
683}
684
685/*
686 * Plumbing for rpc transport switch and kernel module
687 */
688
689static struct rpc_xprt_ops xprt_rdma_procs = {
690 .reserve_xprt = xprt_reserve_xprt_cong,
691 .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */
692 .alloc_slot = xprt_alloc_slot,
693 .release_request = xprt_release_rqst_cong, /* ditto */
694 .set_retrans_timeout = xprt_set_retrans_timeout_def, /* ditto */
695 .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */
696 .set_port = xprt_rdma_set_port,
697 .connect = xprt_rdma_connect,
698 .buf_alloc = xprt_rdma_allocate,
699 .buf_free = xprt_rdma_free,
700 .send_request = xprt_rdma_send_request,
701 .close = xprt_rdma_close,
702 .destroy = xprt_rdma_destroy,
703 .print_stats = xprt_rdma_print_stats,
704 .enable_swap = xprt_rdma_enable_swap,
705 .disable_swap = xprt_rdma_disable_swap,
706 .inject_disconnect = xprt_rdma_inject_disconnect,
707#if defined(CONFIG_SUNRPC_BACKCHANNEL)
708 .bc_setup = xprt_rdma_bc_setup,
709 .bc_up = xprt_rdma_bc_up,
710 .bc_free_rqst = xprt_rdma_bc_free_rqst,
711 .bc_destroy = xprt_rdma_bc_destroy,
712#endif
713};
714
715static struct xprt_class xprt_rdma = {
716 .list = LIST_HEAD_INIT(xprt_rdma.list),
717 .name = "rdma",
718 .owner = THIS_MODULE,
719 .ident = XPRT_TRANSPORT_RDMA,
720 .setup = xprt_setup_rdma,
721};
722
723void xprt_rdma_cleanup(void)
724{
725 int rc;
726
727 dprintk("RPCRDMA Module Removed, deregister RPC RDMA transport\n");
728#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
729 if (sunrpc_table_header) {
730 unregister_sysctl_table(sunrpc_table_header);
731 sunrpc_table_header = NULL;
732 }
733#endif
734 rc = xprt_unregister_transport(&xprt_rdma);
735 if (rc)
736 dprintk("RPC: %s: xprt_unregister returned %i\n",
737 __func__, rc);
738
739 rpcrdma_destroy_wq();
740 frwr_destroy_recovery_wq();
741
742 rc = xprt_unregister_transport(&xprt_rdma_bc);
743 if (rc)
744 dprintk("RPC: %s: xprt_unregister(bc) returned %i\n",
745 __func__, rc);
746}
747
748int xprt_rdma_init(void)
749{
750 int rc;
751
752 rc = frwr_alloc_recovery_wq();
753 if (rc)
754 return rc;
755
756 rc = rpcrdma_alloc_wq();
757 if (rc) {
758 frwr_destroy_recovery_wq();
759 return rc;
760 }
761
762 rc = xprt_register_transport(&xprt_rdma);
763 if (rc) {
764 rpcrdma_destroy_wq();
765 frwr_destroy_recovery_wq();
766 return rc;
767 }
768
769 rc = xprt_register_transport(&xprt_rdma_bc);
770 if (rc) {
771 xprt_unregister_transport(&xprt_rdma);
772 rpcrdma_destroy_wq();
773 frwr_destroy_recovery_wq();
774 return rc;
775 }
776
777 dprintk("RPCRDMA Module Init, register RPC RDMA transport\n");
778
779 dprintk("Defaults:\n");
780 dprintk("\tSlots %d\n"
781 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
782 xprt_rdma_slot_table_entries,
783 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
784 dprintk("\tPadding %d\n\tMemreg %d\n",
785 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
786
787#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
788 if (!sunrpc_table_header)
789 sunrpc_table_header = register_sysctl_table(sunrpc_table);
790#endif
791 return 0;
792}
1// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2/*
3 * Copyright (c) 2014-2017 Oracle. All rights reserved.
4 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the BSD-type
10 * license below:
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 *
16 * Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 *
19 * Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials provided
22 * with the distribution.
23 *
24 * Neither the name of the Network Appliance, Inc. nor the names of
25 * its contributors may be used to endorse or promote products
26 * derived from this software without specific prior written
27 * permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 */
41
42/*
43 * transport.c
44 *
45 * This file contains the top-level implementation of an RPC RDMA
46 * transport.
47 *
48 * Naming convention: functions beginning with xprt_ are part of the
49 * transport switch. All others are RPC RDMA internal.
50 */
51
52#include <linux/module.h>
53#include <linux/slab.h>
54#include <linux/seq_file.h>
55#include <linux/smp.h>
56
57#include <linux/sunrpc/addr.h>
58#include <linux/sunrpc/svc_rdma.h>
59
60#include "xprt_rdma.h"
61#include <trace/events/rpcrdma.h>
62
63#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
64# define RPCDBG_FACILITY RPCDBG_TRANS
65#endif
66
67/*
68 * tunables
69 */
70
71static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
72unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
73unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
74unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRWR;
75int xprt_rdma_pad_optimize;
76
77#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
78
79static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
80static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
81static unsigned int min_inline_size = RPCRDMA_MIN_INLINE;
82static unsigned int max_inline_size = RPCRDMA_MAX_INLINE;
83static unsigned int max_padding = PAGE_SIZE;
84static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
85static unsigned int max_memreg = RPCRDMA_LAST - 1;
86static unsigned int dummy;
87
88static struct ctl_table_header *sunrpc_table_header;
89
90static struct ctl_table xr_tunables_table[] = {
91 {
92 .procname = "rdma_slot_table_entries",
93 .data = &xprt_rdma_slot_table_entries,
94 .maxlen = sizeof(unsigned int),
95 .mode = 0644,
96 .proc_handler = proc_dointvec_minmax,
97 .extra1 = &min_slot_table_size,
98 .extra2 = &max_slot_table_size
99 },
100 {
101 .procname = "rdma_max_inline_read",
102 .data = &xprt_rdma_max_inline_read,
103 .maxlen = sizeof(unsigned int),
104 .mode = 0644,
105 .proc_handler = proc_dointvec_minmax,
106 .extra1 = &min_inline_size,
107 .extra2 = &max_inline_size,
108 },
109 {
110 .procname = "rdma_max_inline_write",
111 .data = &xprt_rdma_max_inline_write,
112 .maxlen = sizeof(unsigned int),
113 .mode = 0644,
114 .proc_handler = proc_dointvec_minmax,
115 .extra1 = &min_inline_size,
116 .extra2 = &max_inline_size,
117 },
118 {
119 .procname = "rdma_inline_write_padding",
120 .data = &dummy,
121 .maxlen = sizeof(unsigned int),
122 .mode = 0644,
123 .proc_handler = proc_dointvec_minmax,
124 .extra1 = SYSCTL_ZERO,
125 .extra2 = &max_padding,
126 },
127 {
128 .procname = "rdma_memreg_strategy",
129 .data = &xprt_rdma_memreg_strategy,
130 .maxlen = sizeof(unsigned int),
131 .mode = 0644,
132 .proc_handler = proc_dointvec_minmax,
133 .extra1 = &min_memreg,
134 .extra2 = &max_memreg,
135 },
136 {
137 .procname = "rdma_pad_optimize",
138 .data = &xprt_rdma_pad_optimize,
139 .maxlen = sizeof(unsigned int),
140 .mode = 0644,
141 .proc_handler = proc_dointvec,
142 },
143 { },
144};
145
146static struct ctl_table sunrpc_table[] = {
147 {
148 .procname = "sunrpc",
149 .mode = 0555,
150 .child = xr_tunables_table
151 },
152 { },
153};
154
155#endif
156
157static const struct rpc_xprt_ops xprt_rdma_procs;
158
159static void
160xprt_rdma_format_addresses4(struct rpc_xprt *xprt, struct sockaddr *sap)
161{
162 struct sockaddr_in *sin = (struct sockaddr_in *)sap;
163 char buf[20];
164
165 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
166 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
167
168 xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA;
169}
170
171static void
172xprt_rdma_format_addresses6(struct rpc_xprt *xprt, struct sockaddr *sap)
173{
174 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
175 char buf[40];
176
177 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
178 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
179
180 xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA6;
181}
182
183void
184xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap)
185{
186 char buf[128];
187
188 switch (sap->sa_family) {
189 case AF_INET:
190 xprt_rdma_format_addresses4(xprt, sap);
191 break;
192 case AF_INET6:
193 xprt_rdma_format_addresses6(xprt, sap);
194 break;
195 default:
196 pr_err("rpcrdma: Unrecognized address family\n");
197 return;
198 }
199
200 (void)rpc_ntop(sap, buf, sizeof(buf));
201 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
202
203 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
204 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
205
206 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
207 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
208
209 xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
210}
211
212void
213xprt_rdma_free_addresses(struct rpc_xprt *xprt)
214{
215 unsigned int i;
216
217 for (i = 0; i < RPC_DISPLAY_MAX; i++)
218 switch (i) {
219 case RPC_DISPLAY_PROTO:
220 case RPC_DISPLAY_NETID:
221 continue;
222 default:
223 kfree(xprt->address_strings[i]);
224 }
225}
226
227/**
228 * xprt_rdma_connect_worker - establish connection in the background
229 * @work: worker thread context
230 *
231 * Requester holds the xprt's send lock to prevent activity on this
232 * transport while a fresh connection is being established. RPC tasks
233 * sleep on the xprt's pending queue waiting for connect to complete.
234 */
235static void
236xprt_rdma_connect_worker(struct work_struct *work)
237{
238 struct rpcrdma_xprt *r_xprt = container_of(work, struct rpcrdma_xprt,
239 rx_connect_worker.work);
240 struct rpc_xprt *xprt = &r_xprt->rx_xprt;
241 int rc;
242
243 rc = rpcrdma_xprt_connect(r_xprt);
244 xprt_clear_connecting(xprt);
245 if (!rc) {
246 xprt->connect_cookie++;
247 xprt->stat.connect_count++;
248 xprt->stat.connect_time += (long)jiffies -
249 xprt->stat.connect_start;
250 xprt_set_connected(xprt);
251 rc = -EAGAIN;
252 } else {
253 /* Force a call to xprt_rdma_close to clean up */
254 spin_lock(&xprt->transport_lock);
255 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
256 spin_unlock(&xprt->transport_lock);
257 }
258 xprt_wake_pending_tasks(xprt, rc);
259}
260
261/**
262 * xprt_rdma_inject_disconnect - inject a connection fault
263 * @xprt: transport context
264 *
265 * If @xprt is connected, disconnect it to simulate spurious connection
266 * loss.
267 */
268static void
269xprt_rdma_inject_disconnect(struct rpc_xprt *xprt)
270{
271 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
272
273 trace_xprtrdma_op_inject_dsc(r_xprt);
274 rdma_disconnect(r_xprt->rx_ep->re_id);
275}
276
277/**
278 * xprt_rdma_destroy - Full tear down of transport
279 * @xprt: doomed transport context
280 *
281 * Caller guarantees there will be no more calls to us with
282 * this @xprt.
283 */
284static void
285xprt_rdma_destroy(struct rpc_xprt *xprt)
286{
287 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
288
289 cancel_delayed_work_sync(&r_xprt->rx_connect_worker);
290
291 rpcrdma_xprt_disconnect(r_xprt);
292 rpcrdma_buffer_destroy(&r_xprt->rx_buf);
293
294 xprt_rdma_free_addresses(xprt);
295 xprt_free(xprt);
296
297 module_put(THIS_MODULE);
298}
299
300/* 60 second timeout, no retries */
301static const struct rpc_timeout xprt_rdma_default_timeout = {
302 .to_initval = 60 * HZ,
303 .to_maxval = 60 * HZ,
304};
305
306/**
307 * xprt_setup_rdma - Set up transport to use RDMA
308 *
309 * @args: rpc transport arguments
310 */
311static struct rpc_xprt *
312xprt_setup_rdma(struct xprt_create *args)
313{
314 struct rpc_xprt *xprt;
315 struct rpcrdma_xprt *new_xprt;
316 struct sockaddr *sap;
317 int rc;
318
319 if (args->addrlen > sizeof(xprt->addr))
320 return ERR_PTR(-EBADF);
321
322 if (!try_module_get(THIS_MODULE))
323 return ERR_PTR(-EIO);
324
325 xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt), 0,
326 xprt_rdma_slot_table_entries);
327 if (!xprt) {
328 module_put(THIS_MODULE);
329 return ERR_PTR(-ENOMEM);
330 }
331
332 xprt->timeout = &xprt_rdma_default_timeout;
333 xprt->connect_timeout = xprt->timeout->to_initval;
334 xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
335 xprt->bind_timeout = RPCRDMA_BIND_TO;
336 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
337 xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO;
338
339 xprt->resvport = 0; /* privileged port not needed */
340 xprt->ops = &xprt_rdma_procs;
341
342 /*
343 * Set up RDMA-specific connect data.
344 */
345 sap = args->dstaddr;
346
347 /* Ensure xprt->addr holds valid server TCP (not RDMA)
348 * address, for any side protocols which peek at it */
349 xprt->prot = IPPROTO_TCP;
350 xprt->addrlen = args->addrlen;
351 memcpy(&xprt->addr, sap, xprt->addrlen);
352
353 if (rpc_get_port(sap))
354 xprt_set_bound(xprt);
355 xprt_rdma_format_addresses(xprt, sap);
356
357 new_xprt = rpcx_to_rdmax(xprt);
358 rc = rpcrdma_buffer_create(new_xprt);
359 if (rc) {
360 xprt_rdma_free_addresses(xprt);
361 xprt_free(xprt);
362 module_put(THIS_MODULE);
363 return ERR_PTR(rc);
364 }
365
366 INIT_DELAYED_WORK(&new_xprt->rx_connect_worker,
367 xprt_rdma_connect_worker);
368
369 xprt->max_payload = RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT;
370
371 return xprt;
372}
373
374/**
375 * xprt_rdma_close - close a transport connection
376 * @xprt: transport context
377 *
378 * Called during autoclose or device removal.
379 *
380 * Caller holds @xprt's send lock to prevent activity on this
381 * transport while the connection is torn down.
382 */
383void xprt_rdma_close(struct rpc_xprt *xprt)
384{
385 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
386
387 rpcrdma_xprt_disconnect(r_xprt);
388
389 xprt->reestablish_timeout = 0;
390 ++xprt->connect_cookie;
391 xprt_disconnect_done(xprt);
392}
393
394/**
395 * xprt_rdma_set_port - update server port with rpcbind result
396 * @xprt: controlling RPC transport
397 * @port: new port value
398 *
399 * Transport connect status is unchanged.
400 */
401static void
402xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
403{
404 struct sockaddr *sap = (struct sockaddr *)&xprt->addr;
405 char buf[8];
406
407 rpc_set_port(sap, port);
408
409 kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
410 snprintf(buf, sizeof(buf), "%u", port);
411 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
412
413 kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
414 snprintf(buf, sizeof(buf), "%4hx", port);
415 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
416
417 trace_xprtrdma_op_setport(container_of(xprt, struct rpcrdma_xprt,
418 rx_xprt));
419}
420
421/**
422 * xprt_rdma_timer - invoked when an RPC times out
423 * @xprt: controlling RPC transport
424 * @task: RPC task that timed out
425 *
426 * Invoked when the transport is still connected, but an RPC
427 * retransmit timeout occurs.
428 *
429 * Since RDMA connections don't have a keep-alive, forcibly
430 * disconnect and retry to connect. This drives full
431 * detection of the network path, and retransmissions of
432 * all pending RPCs.
433 */
434static void
435xprt_rdma_timer(struct rpc_xprt *xprt, struct rpc_task *task)
436{
437 xprt_force_disconnect(xprt);
438}
439
440/**
441 * xprt_rdma_set_connect_timeout - set timeouts for establishing a connection
442 * @xprt: controlling transport instance
443 * @connect_timeout: reconnect timeout after client disconnects
444 * @reconnect_timeout: reconnect timeout after server disconnects
445 *
446 */
447static void xprt_rdma_set_connect_timeout(struct rpc_xprt *xprt,
448 unsigned long connect_timeout,
449 unsigned long reconnect_timeout)
450{
451 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
452
453 trace_xprtrdma_op_set_cto(r_xprt, connect_timeout, reconnect_timeout);
454
455 spin_lock(&xprt->transport_lock);
456
457 if (connect_timeout < xprt->connect_timeout) {
458 struct rpc_timeout to;
459 unsigned long initval;
460
461 to = *xprt->timeout;
462 initval = connect_timeout;
463 if (initval < RPCRDMA_INIT_REEST_TO << 1)
464 initval = RPCRDMA_INIT_REEST_TO << 1;
465 to.to_initval = initval;
466 to.to_maxval = initval;
467 r_xprt->rx_timeout = to;
468 xprt->timeout = &r_xprt->rx_timeout;
469 xprt->connect_timeout = connect_timeout;
470 }
471
472 if (reconnect_timeout < xprt->max_reconnect_timeout)
473 xprt->max_reconnect_timeout = reconnect_timeout;
474
475 spin_unlock(&xprt->transport_lock);
476}
477
478/**
479 * xprt_rdma_connect - schedule an attempt to reconnect
480 * @xprt: transport state
481 * @task: RPC scheduler context (unused)
482 *
483 */
484static void
485xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task)
486{
487 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
488 struct rpcrdma_ep *ep = r_xprt->rx_ep;
489 unsigned long delay;
490
491 delay = 0;
492 if (ep && ep->re_connect_status != 0) {
493 delay = xprt_reconnect_delay(xprt);
494 xprt_reconnect_backoff(xprt, RPCRDMA_INIT_REEST_TO);
495 }
496 trace_xprtrdma_op_connect(r_xprt, delay);
497 queue_delayed_work(xprtiod_workqueue, &r_xprt->rx_connect_worker,
498 delay);
499}
500
501/**
502 * xprt_rdma_alloc_slot - allocate an rpc_rqst
503 * @xprt: controlling RPC transport
504 * @task: RPC task requesting a fresh rpc_rqst
505 *
506 * tk_status values:
507 * %0 if task->tk_rqstp points to a fresh rpc_rqst
508 * %-EAGAIN if no rpc_rqst is available; queued on backlog
509 */
510static void
511xprt_rdma_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
512{
513 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
514 struct rpcrdma_req *req;
515
516 req = rpcrdma_buffer_get(&r_xprt->rx_buf);
517 if (!req)
518 goto out_sleep;
519 task->tk_rqstp = &req->rl_slot;
520 task->tk_status = 0;
521 return;
522
523out_sleep:
524 set_bit(XPRT_CONGESTED, &xprt->state);
525 rpc_sleep_on(&xprt->backlog, task, NULL);
526 task->tk_status = -EAGAIN;
527}
528
529/**
530 * xprt_rdma_free_slot - release an rpc_rqst
531 * @xprt: controlling RPC transport
532 * @rqst: rpc_rqst to release
533 *
534 */
535static void
536xprt_rdma_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *rqst)
537{
538 struct rpcrdma_xprt *r_xprt =
539 container_of(xprt, struct rpcrdma_xprt, rx_xprt);
540
541 memset(rqst, 0, sizeof(*rqst));
542 rpcrdma_buffer_put(&r_xprt->rx_buf, rpcr_to_rdmar(rqst));
543 if (unlikely(!rpc_wake_up_next(&xprt->backlog)))
544 clear_bit(XPRT_CONGESTED, &xprt->state);
545}
546
547static bool rpcrdma_check_regbuf(struct rpcrdma_xprt *r_xprt,
548 struct rpcrdma_regbuf *rb, size_t size,
549 gfp_t flags)
550{
551 if (unlikely(rdmab_length(rb) < size)) {
552 if (!rpcrdma_regbuf_realloc(rb, size, flags))
553 return false;
554 r_xprt->rx_stats.hardway_register_count += size;
555 }
556 return true;
557}
558
559/**
560 * xprt_rdma_allocate - allocate transport resources for an RPC
561 * @task: RPC task
562 *
563 * Return values:
564 * 0: Success; rq_buffer points to RPC buffer to use
565 * ENOMEM: Out of memory, call again later
566 * EIO: A permanent error occurred, do not retry
567 */
568static int
569xprt_rdma_allocate(struct rpc_task *task)
570{
571 struct rpc_rqst *rqst = task->tk_rqstp;
572 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
573 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
574 gfp_t flags;
575
576 flags = RPCRDMA_DEF_GFP;
577 if (RPC_IS_SWAPPER(task))
578 flags = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
579
580 if (!rpcrdma_check_regbuf(r_xprt, req->rl_sendbuf, rqst->rq_callsize,
581 flags))
582 goto out_fail;
583 if (!rpcrdma_check_regbuf(r_xprt, req->rl_recvbuf, rqst->rq_rcvsize,
584 flags))
585 goto out_fail;
586
587 rqst->rq_buffer = rdmab_data(req->rl_sendbuf);
588 rqst->rq_rbuffer = rdmab_data(req->rl_recvbuf);
589 trace_xprtrdma_op_allocate(task, req);
590 return 0;
591
592out_fail:
593 trace_xprtrdma_op_allocate(task, NULL);
594 return -ENOMEM;
595}
596
597/**
598 * xprt_rdma_free - release resources allocated by xprt_rdma_allocate
599 * @task: RPC task
600 *
601 * Caller guarantees rqst->rq_buffer is non-NULL.
602 */
603static void
604xprt_rdma_free(struct rpc_task *task)
605{
606 struct rpc_rqst *rqst = task->tk_rqstp;
607 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt);
608 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
609
610 trace_xprtrdma_op_free(task, req);
611
612 if (!list_empty(&req->rl_registered))
613 frwr_unmap_sync(r_xprt, req);
614
615 /* XXX: If the RPC is completing because of a signal and
616 * not because a reply was received, we ought to ensure
617 * that the Send completion has fired, so that memory
618 * involved with the Send is not still visible to the NIC.
619 */
620}
621
622/**
623 * xprt_rdma_send_request - marshal and send an RPC request
624 * @rqst: RPC message in rq_snd_buf
625 *
626 * Caller holds the transport's write lock.
627 *
628 * Returns:
629 * %0 if the RPC message has been sent
630 * %-ENOTCONN if the caller should reconnect and call again
631 * %-EAGAIN if the caller should call again
632 * %-ENOBUFS if the caller should call again after a delay
633 * %-EMSGSIZE if encoding ran out of buffer space. The request
634 * was not sent. Do not try to send this message again.
635 * %-EIO if an I/O error occurred. The request was not sent.
636 * Do not try to send this message again.
637 */
638static int
639xprt_rdma_send_request(struct rpc_rqst *rqst)
640{
641 struct rpc_xprt *xprt = rqst->rq_xprt;
642 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
643 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
644 int rc = 0;
645
646#if defined(CONFIG_SUNRPC_BACKCHANNEL)
647 if (unlikely(!rqst->rq_buffer))
648 return xprt_rdma_bc_send_reply(rqst);
649#endif /* CONFIG_SUNRPC_BACKCHANNEL */
650
651 if (!xprt_connected(xprt))
652 return -ENOTCONN;
653
654 if (!xprt_request_get_cong(xprt, rqst))
655 return -EBADSLT;
656
657 rc = rpcrdma_marshal_req(r_xprt, rqst);
658 if (rc < 0)
659 goto failed_marshal;
660
661 /* Must suppress retransmit to maintain credits */
662 if (rqst->rq_connect_cookie == xprt->connect_cookie)
663 goto drop_connection;
664 rqst->rq_xtime = ktime_get();
665
666 if (rpcrdma_post_sends(r_xprt, req))
667 goto drop_connection;
668
669 rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
670
671 /* An RPC with no reply will throw off credit accounting,
672 * so drop the connection to reset the credit grant.
673 */
674 if (!rpc_reply_expected(rqst->rq_task))
675 goto drop_connection;
676 return 0;
677
678failed_marshal:
679 if (rc != -ENOTCONN)
680 return rc;
681drop_connection:
682 xprt_rdma_close(xprt);
683 return -ENOTCONN;
684}
685
686void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
687{
688 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
689 long idle_time = 0;
690
691 if (xprt_connected(xprt))
692 idle_time = (long)(jiffies - xprt->last_used) / HZ;
693
694 seq_puts(seq, "\txprt:\trdma ");
695 seq_printf(seq, "%u %lu %lu %lu %ld %lu %lu %lu %llu %llu ",
696 0, /* need a local port? */
697 xprt->stat.bind_count,
698 xprt->stat.connect_count,
699 xprt->stat.connect_time / HZ,
700 idle_time,
701 xprt->stat.sends,
702 xprt->stat.recvs,
703 xprt->stat.bad_xids,
704 xprt->stat.req_u,
705 xprt->stat.bklog_u);
706 seq_printf(seq, "%lu %lu %lu %llu %llu %llu %llu %lu %lu %lu %lu ",
707 r_xprt->rx_stats.read_chunk_count,
708 r_xprt->rx_stats.write_chunk_count,
709 r_xprt->rx_stats.reply_chunk_count,
710 r_xprt->rx_stats.total_rdma_request,
711 r_xprt->rx_stats.total_rdma_reply,
712 r_xprt->rx_stats.pullup_copy_count,
713 r_xprt->rx_stats.fixup_copy_count,
714 r_xprt->rx_stats.hardway_register_count,
715 r_xprt->rx_stats.failed_marshal_count,
716 r_xprt->rx_stats.bad_reply_count,
717 r_xprt->rx_stats.nomsg_call_count);
718 seq_printf(seq, "%lu %lu %lu %lu %lu %lu\n",
719 r_xprt->rx_stats.mrs_recycled,
720 r_xprt->rx_stats.mrs_orphaned,
721 r_xprt->rx_stats.mrs_allocated,
722 r_xprt->rx_stats.local_inv_needed,
723 r_xprt->rx_stats.empty_sendctx_q,
724 r_xprt->rx_stats.reply_waits_for_send);
725}
726
727static int
728xprt_rdma_enable_swap(struct rpc_xprt *xprt)
729{
730 return 0;
731}
732
733static void
734xprt_rdma_disable_swap(struct rpc_xprt *xprt)
735{
736}
737
738/*
739 * Plumbing for rpc transport switch and kernel module
740 */
741
742static const struct rpc_xprt_ops xprt_rdma_procs = {
743 .reserve_xprt = xprt_reserve_xprt_cong,
744 .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */
745 .alloc_slot = xprt_rdma_alloc_slot,
746 .free_slot = xprt_rdma_free_slot,
747 .release_request = xprt_release_rqst_cong, /* ditto */
748 .wait_for_reply_request = xprt_wait_for_reply_request_def, /* ditto */
749 .timer = xprt_rdma_timer,
750 .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */
751 .set_port = xprt_rdma_set_port,
752 .connect = xprt_rdma_connect,
753 .buf_alloc = xprt_rdma_allocate,
754 .buf_free = xprt_rdma_free,
755 .send_request = xprt_rdma_send_request,
756 .close = xprt_rdma_close,
757 .destroy = xprt_rdma_destroy,
758 .set_connect_timeout = xprt_rdma_set_connect_timeout,
759 .print_stats = xprt_rdma_print_stats,
760 .enable_swap = xprt_rdma_enable_swap,
761 .disable_swap = xprt_rdma_disable_swap,
762 .inject_disconnect = xprt_rdma_inject_disconnect,
763#if defined(CONFIG_SUNRPC_BACKCHANNEL)
764 .bc_setup = xprt_rdma_bc_setup,
765 .bc_maxpayload = xprt_rdma_bc_maxpayload,
766 .bc_num_slots = xprt_rdma_bc_max_slots,
767 .bc_free_rqst = xprt_rdma_bc_free_rqst,
768 .bc_destroy = xprt_rdma_bc_destroy,
769#endif
770};
771
772static struct xprt_class xprt_rdma = {
773 .list = LIST_HEAD_INIT(xprt_rdma.list),
774 .name = "rdma",
775 .owner = THIS_MODULE,
776 .ident = XPRT_TRANSPORT_RDMA,
777 .setup = xprt_setup_rdma,
778};
779
780void xprt_rdma_cleanup(void)
781{
782#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
783 if (sunrpc_table_header) {
784 unregister_sysctl_table(sunrpc_table_header);
785 sunrpc_table_header = NULL;
786 }
787#endif
788
789 xprt_unregister_transport(&xprt_rdma);
790 xprt_unregister_transport(&xprt_rdma_bc);
791}
792
793int xprt_rdma_init(void)
794{
795 int rc;
796
797 rc = xprt_register_transport(&xprt_rdma);
798 if (rc)
799 return rc;
800
801 rc = xprt_register_transport(&xprt_rdma_bc);
802 if (rc) {
803 xprt_unregister_transport(&xprt_rdma);
804 return rc;
805 }
806
807#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
808 if (!sunrpc_table_header)
809 sunrpc_table_header = register_sysctl_table(sunrpc_table);
810#endif
811 return 0;
812}