blob: a564c1a39ec5f833b3a5dd9f369269615a047940 [file] [log] [blame]
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -04001/*
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/init.h>
52#include <linux/seq_file.h>
53
54#include "xprt_rdma.h"
55
56#ifdef RPC_DEBUG
57# define RPCDBG_FACILITY RPCDBG_TRANS
58#endif
59
60MODULE_LICENSE("Dual BSD/GPL");
61
62MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
63MODULE_AUTHOR("Network Appliance, Inc.");
64
65/*
66 * tunables
67 */
68
69static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
70static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
71static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
72static unsigned int xprt_rdma_inline_write_padding;
73#if !RPCRDMA_PERSISTENT_REGISTRATION
74static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_REGISTER; /* FMR? */
75#else
76static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_ALLPHYSICAL;
77#endif
78
79#ifdef RPC_DEBUG
80
81static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
82static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
83static unsigned int zero;
84static unsigned int max_padding = PAGE_SIZE;
85static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
86static unsigned int max_memreg = RPCRDMA_LAST - 1;
87
88static struct ctl_table_header *sunrpc_table_header;
89
90static ctl_table xr_tunables_table[] = {
91 {
James Lentinicfcb43f2007-11-26 12:42:44 -050092 .ctl_name = CTL_UNNUMBERED,
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040093 .procname = "rdma_slot_table_entries",
94 .data = &xprt_rdma_slot_table_entries,
95 .maxlen = sizeof(unsigned int),
96 .mode = 0644,
97 .proc_handler = &proc_dointvec_minmax,
98 .strategy = &sysctl_intvec,
99 .extra1 = &min_slot_table_size,
100 .extra2 = &max_slot_table_size
101 },
102 {
James Lentinicfcb43f2007-11-26 12:42:44 -0500103 .ctl_name = CTL_UNNUMBERED,
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400104 .procname = "rdma_max_inline_read",
105 .data = &xprt_rdma_max_inline_read,
106 .maxlen = sizeof(unsigned int),
107 .mode = 0644,
108 .proc_handler = &proc_dointvec,
109 .strategy = &sysctl_intvec,
110 },
111 {
James Lentinicfcb43f2007-11-26 12:42:44 -0500112 .ctl_name = CTL_UNNUMBERED,
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400113 .procname = "rdma_max_inline_write",
114 .data = &xprt_rdma_max_inline_write,
115 .maxlen = sizeof(unsigned int),
116 .mode = 0644,
117 .proc_handler = &proc_dointvec,
118 .strategy = &sysctl_intvec,
119 },
120 {
James Lentinicfcb43f2007-11-26 12:42:44 -0500121 .ctl_name = CTL_UNNUMBERED,
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400122 .procname = "rdma_inline_write_padding",
123 .data = &xprt_rdma_inline_write_padding,
124 .maxlen = sizeof(unsigned int),
125 .mode = 0644,
126 .proc_handler = &proc_dointvec_minmax,
127 .strategy = &sysctl_intvec,
128 .extra1 = &zero,
129 .extra2 = &max_padding,
130 },
131 {
James Lentinicfcb43f2007-11-26 12:42:44 -0500132 .ctl_name = CTL_UNNUMBERED,
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400133 .procname = "rdma_memreg_strategy",
134 .data = &xprt_rdma_memreg_strategy,
135 .maxlen = sizeof(unsigned int),
136 .mode = 0644,
137 .proc_handler = &proc_dointvec_minmax,
138 .strategy = &sysctl_intvec,
139 .extra1 = &min_memreg,
140 .extra2 = &max_memreg,
141 },
142 {
143 .ctl_name = 0,
144 },
145};
146
147static ctl_table sunrpc_table[] = {
148 {
149 .ctl_name = CTL_SUNRPC,
150 .procname = "sunrpc",
151 .mode = 0555,
152 .child = xr_tunables_table
153 },
154 {
155 .ctl_name = 0,
156 },
157};
158
159#endif
160
161static struct rpc_xprt_ops xprt_rdma_procs; /* forward reference */
162
163static void
164xprt_rdma_format_addresses(struct rpc_xprt *xprt)
165{
166 struct sockaddr_in *addr = (struct sockaddr_in *)
167 &rpcx_to_rdmad(xprt).addr;
168 char *buf;
169
170 buf = kzalloc(20, GFP_KERNEL);
171 if (buf)
172 snprintf(buf, 20, NIPQUAD_FMT, NIPQUAD(addr->sin_addr.s_addr));
173 xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
174
175 buf = kzalloc(8, GFP_KERNEL);
176 if (buf)
177 snprintf(buf, 8, "%u", ntohs(addr->sin_port));
178 xprt->address_strings[RPC_DISPLAY_PORT] = buf;
179
180 xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
181
182 buf = kzalloc(48, GFP_KERNEL);
183 if (buf)
184 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
185 NIPQUAD(addr->sin_addr.s_addr),
186 ntohs(addr->sin_port), "rdma");
187 xprt->address_strings[RPC_DISPLAY_ALL] = buf;
188
189 buf = kzalloc(10, GFP_KERNEL);
190 if (buf)
191 snprintf(buf, 10, "%02x%02x%02x%02x",
192 NIPQUAD(addr->sin_addr.s_addr));
193 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
194
195 buf = kzalloc(8, GFP_KERNEL);
196 if (buf)
197 snprintf(buf, 8, "%4hx", ntohs(addr->sin_port));
198 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
199
200 buf = kzalloc(30, GFP_KERNEL);
201 if (buf)
202 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
203 NIPQUAD(addr->sin_addr.s_addr),
204 ntohs(addr->sin_port) >> 8,
205 ntohs(addr->sin_port) & 0xff);
206 xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
207
208 /* netid */
209 xprt->address_strings[RPC_DISPLAY_NETID] = "rdma";
210}
211
212static void
213xprt_rdma_free_addresses(struct rpc_xprt *xprt)
214{
Chuck Lever33e01dc2008-01-14 12:32:20 -0500215 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 }
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400225}
226
227static void
228xprt_rdma_connect_worker(struct work_struct *work)
229{
230 struct rpcrdma_xprt *r_xprt =
231 container_of(work, struct rpcrdma_xprt, rdma_connect.work);
232 struct rpc_xprt *xprt = &r_xprt->xprt;
233 int rc = 0;
234
235 if (!xprt->shutdown) {
236 xprt_clear_connected(xprt);
237
238 dprintk("RPC: %s: %sconnect\n", __func__,
239 r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
240 rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
241 if (rc)
242 goto out;
243 }
244 goto out_clear;
245
246out:
247 xprt_wake_pending_tasks(xprt, rc);
248
249out_clear:
250 dprintk("RPC: %s: exit\n", __func__);
251 xprt_clear_connecting(xprt);
252}
253
254/*
255 * xprt_rdma_destroy
256 *
257 * Destroy the xprt.
258 * Free all memory associated with the object, including its own.
259 * NOTE: none of the *destroy methods free memory for their top-level
260 * objects, even though they may have allocated it (they do free
261 * private memory). It's up to the caller to handle it. In this
262 * case (RDMA transport), all structure memory is inlined with the
263 * struct rpcrdma_xprt.
264 */
265static void
266xprt_rdma_destroy(struct rpc_xprt *xprt)
267{
268 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
269 int rc;
270
271 dprintk("RPC: %s: called\n", __func__);
272
273 cancel_delayed_work(&r_xprt->rdma_connect);
274 flush_scheduled_work();
275
276 xprt_clear_connected(xprt);
277
278 rpcrdma_buffer_destroy(&r_xprt->rx_buf);
279 rc = rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
280 if (rc)
281 dprintk("RPC: %s: rpcrdma_ep_destroy returned %i\n",
282 __func__, rc);
283 rpcrdma_ia_close(&r_xprt->rx_ia);
284
285 xprt_rdma_free_addresses(xprt);
286
287 kfree(xprt->slot);
288 xprt->slot = NULL;
289 kfree(xprt);
290
291 dprintk("RPC: %s: returning\n", __func__);
292
293 module_put(THIS_MODULE);
294}
295
Trond Myklebust2881ae72007-12-20 16:03:54 -0500296static const struct rpc_timeout xprt_rdma_default_timeout = {
297 .to_initval = 60 * HZ,
298 .to_maxval = 60 * HZ,
299};
300
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400301/**
302 * xprt_setup_rdma - Set up transport to use RDMA
303 *
304 * @args: rpc transport arguments
305 */
306static struct rpc_xprt *
307xprt_setup_rdma(struct xprt_create *args)
308{
309 struct rpcrdma_create_data_internal cdata;
310 struct rpc_xprt *xprt;
311 struct rpcrdma_xprt *new_xprt;
312 struct rpcrdma_ep *new_ep;
313 struct sockaddr_in *sin;
314 int rc;
315
316 if (args->addrlen > sizeof(xprt->addr)) {
317 dprintk("RPC: %s: address too large\n", __func__);
318 return ERR_PTR(-EBADF);
319 }
320
321 xprt = kzalloc(sizeof(struct rpcrdma_xprt), GFP_KERNEL);
322 if (xprt == NULL) {
323 dprintk("RPC: %s: couldn't allocate rpcrdma_xprt\n",
324 __func__);
325 return ERR_PTR(-ENOMEM);
326 }
327
328 xprt->max_reqs = xprt_rdma_slot_table_entries;
329 xprt->slot = kcalloc(xprt->max_reqs,
330 sizeof(struct rpc_rqst), GFP_KERNEL);
331 if (xprt->slot == NULL) {
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400332 dprintk("RPC: %s: couldn't allocate %d slots\n",
333 __func__, xprt->max_reqs);
Adrian Bunkd5cd9782007-11-14 17:00:00 -0800334 kfree(xprt);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400335 return ERR_PTR(-ENOMEM);
336 }
337
338 /* 60 second timeout, no retries */
Trond Myklebustba7392b2007-12-20 16:03:55 -0500339 xprt->timeout = &xprt_rdma_default_timeout;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400340 xprt->bind_timeout = (60U * HZ);
341 xprt->connect_timeout = (60U * HZ);
342 xprt->reestablish_timeout = (5U * HZ);
343 xprt->idle_timeout = (5U * 60 * HZ);
344
345 xprt->resvport = 0; /* privileged port not needed */
346 xprt->tsh_size = 0; /* RPC-RDMA handles framing */
347 xprt->max_payload = RPCRDMA_MAX_DATA_SEGS * PAGE_SIZE;
348 xprt->ops = &xprt_rdma_procs;
349
350 /*
351 * Set up RDMA-specific connect data.
352 */
353
354 /* Put server RDMA address in local cdata */
355 memcpy(&cdata.addr, args->dstaddr, args->addrlen);
356
357 /* Ensure xprt->addr holds valid server TCP (not RDMA)
358 * address, for any side protocols which peek at it */
359 xprt->prot = IPPROTO_TCP;
360 xprt->addrlen = args->addrlen;
361 memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
362
363 sin = (struct sockaddr_in *)&cdata.addr;
364 if (ntohs(sin->sin_port) != 0)
365 xprt_set_bound(xprt);
366
367 dprintk("RPC: %s: %u.%u.%u.%u:%u\n", __func__,
368 NIPQUAD(sin->sin_addr.s_addr), ntohs(sin->sin_port));
369
370 /* Set max requests */
371 cdata.max_requests = xprt->max_reqs;
372
373 /* Set some length limits */
374 cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
375 cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
376
377 cdata.inline_wsize = xprt_rdma_max_inline_write;
378 if (cdata.inline_wsize > cdata.wsize)
379 cdata.inline_wsize = cdata.wsize;
380
381 cdata.inline_rsize = xprt_rdma_max_inline_read;
382 if (cdata.inline_rsize > cdata.rsize)
383 cdata.inline_rsize = cdata.rsize;
384
385 cdata.padding = xprt_rdma_inline_write_padding;
386
387 /*
388 * Create new transport instance, which includes initialized
389 * o ia
390 * o endpoint
391 * o buffers
392 */
393
394 new_xprt = rpcx_to_rdmax(xprt);
395
396 rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
397 xprt_rdma_memreg_strategy);
398 if (rc)
399 goto out1;
400
401 /*
402 * initialize and create ep
403 */
404 new_xprt->rx_data = cdata;
405 new_ep = &new_xprt->rx_ep;
406 new_ep->rep_remote_addr = cdata.addr;
407
408 rc = rpcrdma_ep_create(&new_xprt->rx_ep,
409 &new_xprt->rx_ia, &new_xprt->rx_data);
410 if (rc)
411 goto out2;
412
413 /*
414 * Allocate pre-registered send and receive buffers for headers and
415 * any inline data. Also specify any padding which will be provided
416 * from a preregistered zero buffer.
417 */
418 rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
419 &new_xprt->rx_data);
420 if (rc)
421 goto out3;
422
423 /*
424 * Register a callback for connection events. This is necessary because
425 * connection loss notification is async. We also catch connection loss
426 * when reaping receives.
427 */
428 INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
429 new_ep->rep_func = rpcrdma_conn_func;
430 new_ep->rep_xprt = xprt;
431
432 xprt_rdma_format_addresses(xprt);
433
434 if (!try_module_get(THIS_MODULE))
435 goto out4;
436
437 return xprt;
438
439out4:
440 xprt_rdma_free_addresses(xprt);
441 rc = -EINVAL;
442out3:
443 (void) rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
444out2:
445 rpcrdma_ia_close(&new_xprt->rx_ia);
446out1:
447 kfree(xprt->slot);
448 kfree(xprt);
449 return ERR_PTR(rc);
450}
451
452/*
453 * Close a connection, during shutdown or timeout/reconnect
454 */
455static void
456xprt_rdma_close(struct rpc_xprt *xprt)
457{
458 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
459
460 dprintk("RPC: %s: closing\n", __func__);
Trond Myklebust62da3b22007-11-06 18:44:20 -0500461 xprt_disconnect_done(xprt);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400462 (void) rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
463}
464
465static void
466xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
467{
468 struct sockaddr_in *sap;
469
470 sap = (struct sockaddr_in *)&xprt->addr;
471 sap->sin_port = htons(port);
472 sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
473 sap->sin_port = htons(port);
474 dprintk("RPC: %s: %u\n", __func__, port);
475}
476
477static void
478xprt_rdma_connect(struct rpc_task *task)
479{
480 struct rpc_xprt *xprt = (struct rpc_xprt *)task->tk_xprt;
481 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
482
483 if (!xprt_test_and_set_connecting(xprt)) {
484 if (r_xprt->rx_ep.rep_connected != 0) {
485 /* Reconnect */
486 schedule_delayed_work(&r_xprt->rdma_connect,
487 xprt->reestablish_timeout);
488 } else {
489 schedule_delayed_work(&r_xprt->rdma_connect, 0);
490 if (!RPC_IS_ASYNC(task))
491 flush_scheduled_work();
492 }
493 }
494}
495
496static int
497xprt_rdma_reserve_xprt(struct rpc_task *task)
498{
499 struct rpc_xprt *xprt = task->tk_xprt;
500 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
501 int credits = atomic_read(&r_xprt->rx_buf.rb_credits);
502
503 /* == RPC_CWNDSCALE @ init, but *after* setup */
504 if (r_xprt->rx_buf.rb_cwndscale == 0UL) {
505 r_xprt->rx_buf.rb_cwndscale = xprt->cwnd;
506 dprintk("RPC: %s: cwndscale %lu\n", __func__,
507 r_xprt->rx_buf.rb_cwndscale);
508 BUG_ON(r_xprt->rx_buf.rb_cwndscale <= 0);
509 }
510 xprt->cwnd = credits * r_xprt->rx_buf.rb_cwndscale;
511 return xprt_reserve_xprt_cong(task);
512}
513
514/*
515 * The RDMA allocate/free functions need the task structure as a place
516 * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
517 * sequence. For this reason, the recv buffers are attached to send
518 * buffers for portions of the RPC. Note that the RPC layer allocates
519 * both send and receive buffers in the same call. We may register
520 * the receive buffer portion when using reply chunks.
521 */
522static void *
523xprt_rdma_allocate(struct rpc_task *task, size_t size)
524{
525 struct rpc_xprt *xprt = task->tk_xprt;
526 struct rpcrdma_req *req, *nreq;
527
528 req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
529 BUG_ON(NULL == req);
530
531 if (size > req->rl_size) {
532 dprintk("RPC: %s: size %zd too large for buffer[%zd]: "
533 "prog %d vers %d proc %d\n",
534 __func__, size, req->rl_size,
535 task->tk_client->cl_prog, task->tk_client->cl_vers,
536 task->tk_msg.rpc_proc->p_proc);
537 /*
538 * Outgoing length shortage. Our inline write max must have
539 * been configured to perform direct i/o.
540 *
541 * This is therefore a large metadata operation, and the
542 * allocate call was made on the maximum possible message,
543 * e.g. containing long filename(s) or symlink data. In
544 * fact, while these metadata operations *might* carry
545 * large outgoing payloads, they rarely *do*. However, we
546 * have to commit to the request here, so reallocate and
547 * register it now. The data path will never require this
548 * reallocation.
549 *
550 * If the allocation or registration fails, the RPC framework
551 * will (doggedly) retry.
552 */
553 if (rpcx_to_rdmax(xprt)->rx_ia.ri_memreg_strategy ==
554 RPCRDMA_BOUNCEBUFFERS) {
555 /* forced to "pure inline" */
556 dprintk("RPC: %s: too much data (%zd) for inline "
557 "(r/w max %d/%d)\n", __func__, size,
558 rpcx_to_rdmad(xprt).inline_rsize,
559 rpcx_to_rdmad(xprt).inline_wsize);
560 size = req->rl_size;
561 rpc_exit(task, -EIO); /* fail the operation */
562 rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
563 goto out;
564 }
565 if (task->tk_flags & RPC_TASK_SWAPPER)
566 nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
567 else
568 nreq = kmalloc(sizeof *req + size, GFP_NOFS);
569 if (nreq == NULL)
570 goto outfail;
571
572 if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
573 nreq->rl_base, size + sizeof(struct rpcrdma_req)
574 - offsetof(struct rpcrdma_req, rl_base),
575 &nreq->rl_handle, &nreq->rl_iov)) {
576 kfree(nreq);
577 goto outfail;
578 }
579 rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
580 nreq->rl_size = size;
581 nreq->rl_niovs = 0;
582 nreq->rl_nchunks = 0;
583 nreq->rl_buffer = (struct rpcrdma_buffer *)req;
584 nreq->rl_reply = req->rl_reply;
585 memcpy(nreq->rl_segments,
586 req->rl_segments, sizeof nreq->rl_segments);
587 /* flag the swap with an unused field */
588 nreq->rl_iov.length = 0;
589 req->rl_reply = NULL;
590 req = nreq;
591 }
592 dprintk("RPC: %s: size %zd, request 0x%p\n", __func__, size, req);
593out:
594 return req->rl_xdr_buf;
595
596outfail:
597 rpcrdma_buffer_put(req);
598 rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
599 return NULL;
600}
601
602/*
603 * This function returns all RDMA resources to the pool.
604 */
605static void
606xprt_rdma_free(void *buffer)
607{
608 struct rpcrdma_req *req;
609 struct rpcrdma_xprt *r_xprt;
610 struct rpcrdma_rep *rep;
611 int i;
612
613 if (buffer == NULL)
614 return;
615
616 req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
Tom Talpeyee1a2c52008-02-27 15:04:26 -0500617 if (req->rl_iov.length == 0) { /* see allocate above */
618 r_xprt = container_of(((struct rpcrdma_req *) req->rl_buffer)->rl_buffer,
619 struct rpcrdma_xprt, rx_buf);
620 } else
621 r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400622 rep = req->rl_reply;
623
624 dprintk("RPC: %s: called on 0x%p%s\n",
625 __func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
626
627 /*
628 * Finish the deregistration. When using mw bind, this was
629 * begun in rpcrdma_reply_handler(). In all other modes, we
630 * do it here, in thread context. The process is considered
631 * complete when the rr_func vector becomes NULL - this
632 * was put in place during rpcrdma_reply_handler() - the wait
633 * call below will not block if the dereg is "done". If
634 * interrupted, our framework will clean up.
635 */
636 for (i = 0; req->rl_nchunks;) {
637 --req->rl_nchunks;
638 i += rpcrdma_deregister_external(
639 &req->rl_segments[i], r_xprt, NULL);
640 }
641
642 if (rep && wait_event_interruptible(rep->rr_unbind, !rep->rr_func)) {
643 rep->rr_func = NULL; /* abandon the callback */
644 req->rl_reply = NULL;
645 }
646
647 if (req->rl_iov.length == 0) { /* see allocate above */
648 struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
649 oreq->rl_reply = req->rl_reply;
650 (void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
651 req->rl_handle,
652 &req->rl_iov);
653 kfree(req);
654 req = oreq;
655 }
656
657 /* Put back request+reply buffers */
658 rpcrdma_buffer_put(req);
659}
660
661/*
662 * send_request invokes the meat of RPC RDMA. It must do the following:
663 * 1. Marshal the RPC request into an RPC RDMA request, which means
664 * putting a header in front of data, and creating IOVs for RDMA
665 * from those in the request.
666 * 2. In marshaling, detect opportunities for RDMA, and use them.
667 * 3. Post a recv message to set up asynch completion, then send
668 * the request (rpcrdma_ep_post).
669 * 4. No partial sends are possible in the RPC-RDMA protocol (as in UDP).
670 */
671
672static int
673xprt_rdma_send_request(struct rpc_task *task)
674{
675 struct rpc_rqst *rqst = task->tk_rqstp;
676 struct rpc_xprt *xprt = task->tk_xprt;
677 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
678 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
679
680 /* marshal the send itself */
681 if (req->rl_niovs == 0 && rpcrdma_marshal_req(rqst) != 0) {
682 r_xprt->rx_stats.failed_marshal_count++;
683 dprintk("RPC: %s: rpcrdma_marshal_req failed\n",
684 __func__);
685 return -EIO;
686 }
687
688 if (req->rl_reply == NULL) /* e.g. reconnection */
689 rpcrdma_recv_buffer_get(req);
690
691 if (req->rl_reply) {
692 req->rl_reply->rr_func = rpcrdma_reply_handler;
693 /* this need only be done once, but... */
694 req->rl_reply->rr_xprt = xprt;
695 }
696
697 if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req)) {
Trond Myklebust62da3b22007-11-06 18:44:20 -0500698 xprt_disconnect_done(xprt);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400699 return -ENOTCONN; /* implies disconnect */
700 }
701
702 rqst->rq_bytes_sent = 0;
703 return 0;
704}
705
706static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
707{
708 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
709 long idle_time = 0;
710
711 if (xprt_connected(xprt))
712 idle_time = (long)(jiffies - xprt->last_used) / HZ;
713
714 seq_printf(seq,
715 "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
716 "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
717
718 0, /* need a local port? */
719 xprt->stat.bind_count,
720 xprt->stat.connect_count,
721 xprt->stat.connect_time,
722 idle_time,
723 xprt->stat.sends,
724 xprt->stat.recvs,
725 xprt->stat.bad_xids,
726 xprt->stat.req_u,
727 xprt->stat.bklog_u,
728
729 r_xprt->rx_stats.read_chunk_count,
730 r_xprt->rx_stats.write_chunk_count,
731 r_xprt->rx_stats.reply_chunk_count,
732 r_xprt->rx_stats.total_rdma_request,
733 r_xprt->rx_stats.total_rdma_reply,
734 r_xprt->rx_stats.pullup_copy_count,
735 r_xprt->rx_stats.fixup_copy_count,
736 r_xprt->rx_stats.hardway_register_count,
737 r_xprt->rx_stats.failed_marshal_count,
738 r_xprt->rx_stats.bad_reply_count);
739}
740
741/*
742 * Plumbing for rpc transport switch and kernel module
743 */
744
745static struct rpc_xprt_ops xprt_rdma_procs = {
746 .reserve_xprt = xprt_rdma_reserve_xprt,
747 .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */
748 .release_request = xprt_release_rqst_cong, /* ditto */
749 .set_retrans_timeout = xprt_set_retrans_timeout_def, /* ditto */
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 .print_stats = xprt_rdma_print_stats
759};
760
761static struct xprt_class xprt_rdma = {
762 .list = LIST_HEAD_INIT(xprt_rdma.list),
763 .name = "rdma",
764 .owner = THIS_MODULE,
765 .ident = XPRT_TRANSPORT_RDMA,
766 .setup = xprt_setup_rdma,
767};
768
769static void __exit xprt_rdma_cleanup(void)
770{
771 int rc;
772
773 dprintk("RPCRDMA Module Removed, deregister RPC RDMA transport\n");
774#ifdef RPC_DEBUG
775 if (sunrpc_table_header) {
776 unregister_sysctl_table(sunrpc_table_header);
777 sunrpc_table_header = NULL;
778 }
779#endif
780 rc = xprt_unregister_transport(&xprt_rdma);
781 if (rc)
782 dprintk("RPC: %s: xprt_unregister returned %i\n",
783 __func__, rc);
784}
785
786static int __init xprt_rdma_init(void)
787{
788 int rc;
789
790 rc = xprt_register_transport(&xprt_rdma);
791
792 if (rc)
793 return rc;
794
795 dprintk(KERN_INFO "RPCRDMA Module Init, register RPC RDMA transport\n");
796
797 dprintk(KERN_INFO "Defaults:\n");
798 dprintk(KERN_INFO "\tSlots %d\n"
799 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
800 xprt_rdma_slot_table_entries,
801 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
802 dprintk(KERN_INFO "\tPadding %d\n\tMemreg %d\n",
803 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
804
805#ifdef RPC_DEBUG
806 if (!sunrpc_table_header)
807 sunrpc_table_header = register_sysctl_table(sunrpc_table);
808#endif
809 return 0;
810}
811
812module_init(xprt_rdma_init);
813module_exit(xprt_rdma_cleanup);