2 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
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
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
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.
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
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.
43 * This file contains the top-level implementation of an RPC RDMA
46 * Naming convention: functions beginning with xprt_ are part of the
47 * transport switch. All others are RPC RDMA internal.
50 #include <linux/module.h>
51 #include <linux/init.h>
52 #include <linux/slab.h>
53 #include <linux/seq_file.h>
55 #include "xprt_rdma.h"
58 # define RPCDBG_FACILITY RPCDBG_TRANS
61 MODULE_LICENSE("Dual BSD/GPL");
63 MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
64 MODULE_AUTHOR("Network Appliance, Inc.");
70 static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
71 static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
72 static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
73 static unsigned int xprt_rdma_inline_write_padding;
74 static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR;
75 int xprt_rdma_pad_optimize = 0;
79 static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
80 static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
81 static unsigned int zero;
82 static unsigned int max_padding = PAGE_SIZE;
83 static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
84 static unsigned int max_memreg = RPCRDMA_LAST - 1;
86 static struct ctl_table_header *sunrpc_table_header;
88 static ctl_table xr_tunables_table[] = {
90 .procname = "rdma_slot_table_entries",
91 .data = &xprt_rdma_slot_table_entries,
92 .maxlen = sizeof(unsigned int),
94 .proc_handler = proc_dointvec_minmax,
95 .extra1 = &min_slot_table_size,
96 .extra2 = &max_slot_table_size
99 .procname = "rdma_max_inline_read",
100 .data = &xprt_rdma_max_inline_read,
101 .maxlen = sizeof(unsigned int),
103 .proc_handler = proc_dointvec,
106 .procname = "rdma_max_inline_write",
107 .data = &xprt_rdma_max_inline_write,
108 .maxlen = sizeof(unsigned int),
110 .proc_handler = proc_dointvec,
113 .procname = "rdma_inline_write_padding",
114 .data = &xprt_rdma_inline_write_padding,
115 .maxlen = sizeof(unsigned int),
117 .proc_handler = proc_dointvec_minmax,
119 .extra2 = &max_padding,
122 .procname = "rdma_memreg_strategy",
123 .data = &xprt_rdma_memreg_strategy,
124 .maxlen = sizeof(unsigned int),
126 .proc_handler = proc_dointvec_minmax,
127 .extra1 = &min_memreg,
128 .extra2 = &max_memreg,
131 .procname = "rdma_pad_optimize",
132 .data = &xprt_rdma_pad_optimize,
133 .maxlen = sizeof(unsigned int),
135 .proc_handler = proc_dointvec,
140 static ctl_table sunrpc_table[] = {
142 .procname = "sunrpc",
144 .child = xr_tunables_table
151 static struct rpc_xprt_ops xprt_rdma_procs; /* forward reference */
154 xprt_rdma_format_addresses(struct rpc_xprt *xprt)
156 struct sockaddr *sap = (struct sockaddr *)
157 &rpcx_to_rdmad(xprt).addr;
158 struct sockaddr_in *sin = (struct sockaddr_in *)sap;
161 (void)rpc_ntop(sap, buf, sizeof(buf));
162 xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
164 snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
165 xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
167 xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
169 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
170 xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
172 snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
173 xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
176 xprt->address_strings[RPC_DISPLAY_NETID] = "rdma";
180 xprt_rdma_free_addresses(struct rpc_xprt *xprt)
184 for (i = 0; i < RPC_DISPLAY_MAX; i++)
186 case RPC_DISPLAY_PROTO:
187 case RPC_DISPLAY_NETID:
190 kfree(xprt->address_strings[i]);
195 xprt_rdma_connect_worker(struct work_struct *work)
197 struct rpcrdma_xprt *r_xprt =
198 container_of(work, struct rpcrdma_xprt, rdma_connect.work);
199 struct rpc_xprt *xprt = &r_xprt->xprt;
202 current->flags |= PF_FSTRANS;
203 xprt_clear_connected(xprt);
205 dprintk("RPC: %s: %sconnect\n", __func__,
206 r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
207 rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
209 xprt_wake_pending_tasks(xprt, rc);
211 dprintk("RPC: %s: exit\n", __func__);
212 xprt_clear_connecting(xprt);
213 current->flags &= ~PF_FSTRANS;
220 * Free all memory associated with the object, including its own.
221 * NOTE: none of the *destroy methods free memory for their top-level
222 * objects, even though they may have allocated it (they do free
223 * private memory). It's up to the caller to handle it. In this
224 * case (RDMA transport), all structure memory is inlined with the
225 * struct rpcrdma_xprt.
228 xprt_rdma_destroy(struct rpc_xprt *xprt)
230 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
233 dprintk("RPC: %s: called\n", __func__);
235 cancel_delayed_work_sync(&r_xprt->rdma_connect);
237 xprt_clear_connected(xprt);
239 rpcrdma_buffer_destroy(&r_xprt->rx_buf);
240 rc = rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
242 dprintk("RPC: %s: rpcrdma_ep_destroy returned %i\n",
244 rpcrdma_ia_close(&r_xprt->rx_ia);
246 xprt_rdma_free_addresses(xprt);
250 dprintk("RPC: %s: returning\n", __func__);
252 module_put(THIS_MODULE);
255 static const struct rpc_timeout xprt_rdma_default_timeout = {
256 .to_initval = 60 * HZ,
257 .to_maxval = 60 * HZ,
261 * xprt_setup_rdma - Set up transport to use RDMA
263 * @args: rpc transport arguments
265 static struct rpc_xprt *
266 xprt_setup_rdma(struct xprt_create *args)
268 struct rpcrdma_create_data_internal cdata;
269 struct rpc_xprt *xprt;
270 struct rpcrdma_xprt *new_xprt;
271 struct rpcrdma_ep *new_ep;
272 struct sockaddr_in *sin;
275 if (args->addrlen > sizeof(xprt->addr)) {
276 dprintk("RPC: %s: address too large\n", __func__);
277 return ERR_PTR(-EBADF);
280 xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt),
281 xprt_rdma_slot_table_entries,
282 xprt_rdma_slot_table_entries);
284 dprintk("RPC: %s: couldn't allocate rpcrdma_xprt\n",
286 return ERR_PTR(-ENOMEM);
289 /* 60 second timeout, no retries */
290 xprt->timeout = &xprt_rdma_default_timeout;
291 xprt->bind_timeout = (60U * HZ);
292 xprt->reestablish_timeout = (5U * HZ);
293 xprt->idle_timeout = (5U * 60 * HZ);
295 xprt->resvport = 0; /* privileged port not needed */
296 xprt->tsh_size = 0; /* RPC-RDMA handles framing */
297 xprt->max_payload = RPCRDMA_MAX_DATA_SEGS * PAGE_SIZE;
298 xprt->ops = &xprt_rdma_procs;
301 * Set up RDMA-specific connect data.
304 /* Put server RDMA address in local cdata */
305 memcpy(&cdata.addr, args->dstaddr, args->addrlen);
307 /* Ensure xprt->addr holds valid server TCP (not RDMA)
308 * address, for any side protocols which peek at it */
309 xprt->prot = IPPROTO_TCP;
310 xprt->addrlen = args->addrlen;
311 memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
313 sin = (struct sockaddr_in *)&cdata.addr;
314 if (ntohs(sin->sin_port) != 0)
315 xprt_set_bound(xprt);
317 dprintk("RPC: %s: %pI4:%u\n",
318 __func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port));
320 /* Set max requests */
321 cdata.max_requests = xprt->max_reqs;
323 /* Set some length limits */
324 cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
325 cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
327 cdata.inline_wsize = xprt_rdma_max_inline_write;
328 if (cdata.inline_wsize > cdata.wsize)
329 cdata.inline_wsize = cdata.wsize;
331 cdata.inline_rsize = xprt_rdma_max_inline_read;
332 if (cdata.inline_rsize > cdata.rsize)
333 cdata.inline_rsize = cdata.rsize;
335 cdata.padding = xprt_rdma_inline_write_padding;
338 * Create new transport instance, which includes initialized
344 new_xprt = rpcx_to_rdmax(xprt);
346 rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
347 xprt_rdma_memreg_strategy);
352 * initialize and create ep
354 new_xprt->rx_data = cdata;
355 new_ep = &new_xprt->rx_ep;
356 new_ep->rep_remote_addr = cdata.addr;
358 rc = rpcrdma_ep_create(&new_xprt->rx_ep,
359 &new_xprt->rx_ia, &new_xprt->rx_data);
364 * Allocate pre-registered send and receive buffers for headers and
365 * any inline data. Also specify any padding which will be provided
366 * from a preregistered zero buffer.
368 rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
374 * Register a callback for connection events. This is necessary because
375 * connection loss notification is async. We also catch connection loss
376 * when reaping receives.
378 INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
379 new_ep->rep_func = rpcrdma_conn_func;
380 new_ep->rep_xprt = xprt;
382 xprt_rdma_format_addresses(xprt);
384 if (!try_module_get(THIS_MODULE))
390 xprt_rdma_free_addresses(xprt);
393 (void) rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
395 rpcrdma_ia_close(&new_xprt->rx_ia);
402 * Close a connection, during shutdown or timeout/reconnect
405 xprt_rdma_close(struct rpc_xprt *xprt)
407 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
409 dprintk("RPC: %s: closing\n", __func__);
410 if (r_xprt->rx_ep.rep_connected > 0)
411 xprt->reestablish_timeout = 0;
412 xprt_disconnect_done(xprt);
413 (void) rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
417 xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
419 struct sockaddr_in *sap;
421 sap = (struct sockaddr_in *)&xprt->addr;
422 sap->sin_port = htons(port);
423 sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
424 sap->sin_port = htons(port);
425 dprintk("RPC: %s: %u\n", __func__, port);
429 xprt_rdma_connect(struct rpc_task *task)
431 struct rpc_xprt *xprt = (struct rpc_xprt *)task->tk_xprt;
432 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
434 if (r_xprt->rx_ep.rep_connected != 0) {
436 schedule_delayed_work(&r_xprt->rdma_connect,
437 xprt->reestablish_timeout);
438 xprt->reestablish_timeout <<= 1;
439 if (xprt->reestablish_timeout > (30 * HZ))
440 xprt->reestablish_timeout = (30 * HZ);
441 else if (xprt->reestablish_timeout < (5 * HZ))
442 xprt->reestablish_timeout = (5 * HZ);
444 schedule_delayed_work(&r_xprt->rdma_connect, 0);
445 if (!RPC_IS_ASYNC(task))
446 flush_delayed_work(&r_xprt->rdma_connect);
451 xprt_rdma_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
453 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
454 int credits = atomic_read(&r_xprt->rx_buf.rb_credits);
456 /* == RPC_CWNDSCALE @ init, but *after* setup */
457 if (r_xprt->rx_buf.rb_cwndscale == 0UL) {
458 r_xprt->rx_buf.rb_cwndscale = xprt->cwnd;
459 dprintk("RPC: %s: cwndscale %lu\n", __func__,
460 r_xprt->rx_buf.rb_cwndscale);
461 BUG_ON(r_xprt->rx_buf.rb_cwndscale <= 0);
463 xprt->cwnd = credits * r_xprt->rx_buf.rb_cwndscale;
464 return xprt_reserve_xprt_cong(xprt, task);
468 * The RDMA allocate/free functions need the task structure as a place
469 * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
470 * sequence. For this reason, the recv buffers are attached to send
471 * buffers for portions of the RPC. Note that the RPC layer allocates
472 * both send and receive buffers in the same call. We may register
473 * the receive buffer portion when using reply chunks.
476 xprt_rdma_allocate(struct rpc_task *task, size_t size)
478 struct rpc_xprt *xprt = task->tk_xprt;
479 struct rpcrdma_req *req, *nreq;
481 req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
484 if (size > req->rl_size) {
485 dprintk("RPC: %s: size %zd too large for buffer[%zd]: "
486 "prog %d vers %d proc %d\n",
487 __func__, size, req->rl_size,
488 task->tk_client->cl_prog, task->tk_client->cl_vers,
489 task->tk_msg.rpc_proc->p_proc);
491 * Outgoing length shortage. Our inline write max must have
492 * been configured to perform direct i/o.
494 * This is therefore a large metadata operation, and the
495 * allocate call was made on the maximum possible message,
496 * e.g. containing long filename(s) or symlink data. In
497 * fact, while these metadata operations *might* carry
498 * large outgoing payloads, they rarely *do*. However, we
499 * have to commit to the request here, so reallocate and
500 * register it now. The data path will never require this
503 * If the allocation or registration fails, the RPC framework
504 * will (doggedly) retry.
506 if (rpcx_to_rdmax(xprt)->rx_ia.ri_memreg_strategy ==
507 RPCRDMA_BOUNCEBUFFERS) {
508 /* forced to "pure inline" */
509 dprintk("RPC: %s: too much data (%zd) for inline "
510 "(r/w max %d/%d)\n", __func__, size,
511 rpcx_to_rdmad(xprt).inline_rsize,
512 rpcx_to_rdmad(xprt).inline_wsize);
514 rpc_exit(task, -EIO); /* fail the operation */
515 rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
518 if (task->tk_flags & RPC_TASK_SWAPPER)
519 nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
521 nreq = kmalloc(sizeof *req + size, GFP_NOFS);
525 if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
526 nreq->rl_base, size + sizeof(struct rpcrdma_req)
527 - offsetof(struct rpcrdma_req, rl_base),
528 &nreq->rl_handle, &nreq->rl_iov)) {
532 rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
533 nreq->rl_size = size;
535 nreq->rl_nchunks = 0;
536 nreq->rl_buffer = (struct rpcrdma_buffer *)req;
537 nreq->rl_reply = req->rl_reply;
538 memcpy(nreq->rl_segments,
539 req->rl_segments, sizeof nreq->rl_segments);
540 /* flag the swap with an unused field */
541 nreq->rl_iov.length = 0;
542 req->rl_reply = NULL;
545 dprintk("RPC: %s: size %zd, request 0x%p\n", __func__, size, req);
547 req->rl_connect_cookie = 0; /* our reserved value */
548 return req->rl_xdr_buf;
551 rpcrdma_buffer_put(req);
552 rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
557 * This function returns all RDMA resources to the pool.
560 xprt_rdma_free(void *buffer)
562 struct rpcrdma_req *req;
563 struct rpcrdma_xprt *r_xprt;
564 struct rpcrdma_rep *rep;
570 req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
571 if (req->rl_iov.length == 0) { /* see allocate above */
572 r_xprt = container_of(((struct rpcrdma_req *) req->rl_buffer)->rl_buffer,
573 struct rpcrdma_xprt, rx_buf);
575 r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
578 dprintk("RPC: %s: called on 0x%p%s\n",
579 __func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
582 * Finish the deregistration. When using mw bind, this was
583 * begun in rpcrdma_reply_handler(). In all other modes, we
584 * do it here, in thread context. The process is considered
585 * complete when the rr_func vector becomes NULL - this
586 * was put in place during rpcrdma_reply_handler() - the wait
587 * call below will not block if the dereg is "done". If
588 * interrupted, our framework will clean up.
590 for (i = 0; req->rl_nchunks;) {
592 i += rpcrdma_deregister_external(
593 &req->rl_segments[i], r_xprt, NULL);
596 if (rep && wait_event_interruptible(rep->rr_unbind, !rep->rr_func)) {
597 rep->rr_func = NULL; /* abandon the callback */
598 req->rl_reply = NULL;
601 if (req->rl_iov.length == 0) { /* see allocate above */
602 struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
603 oreq->rl_reply = req->rl_reply;
604 (void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
611 /* Put back request+reply buffers */
612 rpcrdma_buffer_put(req);
616 * send_request invokes the meat of RPC RDMA. It must do the following:
617 * 1. Marshal the RPC request into an RPC RDMA request, which means
618 * putting a header in front of data, and creating IOVs for RDMA
619 * from those in the request.
620 * 2. In marshaling, detect opportunities for RDMA, and use them.
621 * 3. Post a recv message to set up asynch completion, then send
622 * the request (rpcrdma_ep_post).
623 * 4. No partial sends are possible in the RPC-RDMA protocol (as in UDP).
627 xprt_rdma_send_request(struct rpc_task *task)
629 struct rpc_rqst *rqst = task->tk_rqstp;
630 struct rpc_xprt *xprt = task->tk_xprt;
631 struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
632 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
634 /* marshal the send itself */
635 if (req->rl_niovs == 0 && rpcrdma_marshal_req(rqst) != 0) {
636 r_xprt->rx_stats.failed_marshal_count++;
637 dprintk("RPC: %s: rpcrdma_marshal_req failed\n",
642 if (req->rl_reply == NULL) /* e.g. reconnection */
643 rpcrdma_recv_buffer_get(req);
646 req->rl_reply->rr_func = rpcrdma_reply_handler;
647 /* this need only be done once, but... */
648 req->rl_reply->rr_xprt = xprt;
651 /* Must suppress retransmit to maintain credits */
652 if (req->rl_connect_cookie == xprt->connect_cookie)
653 goto drop_connection;
654 req->rl_connect_cookie = xprt->connect_cookie;
656 if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
657 goto drop_connection;
659 rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
660 rqst->rq_bytes_sent = 0;
664 xprt_disconnect_done(xprt);
665 return -ENOTCONN; /* implies disconnect */
668 static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
670 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
673 if (xprt_connected(xprt))
674 idle_time = (long)(jiffies - xprt->last_used) / HZ;
677 "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
678 "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
680 0, /* need a local port? */
681 xprt->stat.bind_count,
682 xprt->stat.connect_count,
683 xprt->stat.connect_time,
691 r_xprt->rx_stats.read_chunk_count,
692 r_xprt->rx_stats.write_chunk_count,
693 r_xprt->rx_stats.reply_chunk_count,
694 r_xprt->rx_stats.total_rdma_request,
695 r_xprt->rx_stats.total_rdma_reply,
696 r_xprt->rx_stats.pullup_copy_count,
697 r_xprt->rx_stats.fixup_copy_count,
698 r_xprt->rx_stats.hardway_register_count,
699 r_xprt->rx_stats.failed_marshal_count,
700 r_xprt->rx_stats.bad_reply_count);
704 * Plumbing for rpc transport switch and kernel module
707 static struct rpc_xprt_ops xprt_rdma_procs = {
708 .reserve_xprt = xprt_rdma_reserve_xprt,
709 .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */
710 .alloc_slot = xprt_alloc_slot,
711 .release_request = xprt_release_rqst_cong, /* ditto */
712 .set_retrans_timeout = xprt_set_retrans_timeout_def, /* ditto */
713 .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */
714 .set_port = xprt_rdma_set_port,
715 .connect = xprt_rdma_connect,
716 .buf_alloc = xprt_rdma_allocate,
717 .buf_free = xprt_rdma_free,
718 .send_request = xprt_rdma_send_request,
719 .close = xprt_rdma_close,
720 .destroy = xprt_rdma_destroy,
721 .print_stats = xprt_rdma_print_stats
724 static struct xprt_class xprt_rdma = {
725 .list = LIST_HEAD_INIT(xprt_rdma.list),
727 .owner = THIS_MODULE,
728 .ident = XPRT_TRANSPORT_RDMA,
729 .setup = xprt_setup_rdma,
732 static void __exit xprt_rdma_cleanup(void)
736 dprintk(KERN_INFO "RPCRDMA Module Removed, deregister RPC RDMA transport\n");
738 if (sunrpc_table_header) {
739 unregister_sysctl_table(sunrpc_table_header);
740 sunrpc_table_header = NULL;
743 rc = xprt_unregister_transport(&xprt_rdma);
745 dprintk("RPC: %s: xprt_unregister returned %i\n",
749 static int __init xprt_rdma_init(void)
753 rc = xprt_register_transport(&xprt_rdma);
758 dprintk(KERN_INFO "RPCRDMA Module Init, register RPC RDMA transport\n");
760 dprintk(KERN_INFO "Defaults:\n");
761 dprintk(KERN_INFO "\tSlots %d\n"
762 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
763 xprt_rdma_slot_table_entries,
764 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
765 dprintk(KERN_INFO "\tPadding %d\n\tMemreg %d\n",
766 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
769 if (!sunrpc_table_header)
770 sunrpc_table_header = register_sysctl_table(sunrpc_table);
775 module_init(xprt_rdma_init);
776 module_exit(xprt_rdma_cleanup);