Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/9p/trans_rdma.c
4 *
5 * RDMA transport layer based on the trans_fd.c implementation.
6 *
7 * Copyright (C) 2008 by Tom Tucker <tom@opengridcomputing.com>
8 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
9 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
10 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
11 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16#include <linux/in.h>
17#include <linux/module.h>
18#include <linux/net.h>
19#include <linux/ipv6.h>
20#include <linux/kthread.h>
21#include <linux/errno.h>
22#include <linux/kernel.h>
23#include <linux/un.h>
24#include <linux/uaccess.h>
25#include <linux/inet.h>
26#include <linux/idr.h>
27#include <linux/file.h>
28#include <linux/parser.h>
29#include <linux/semaphore.h>
30#include <linux/slab.h>
31#include <linux/seq_file.h>
32#include <net/9p/9p.h>
33#include <net/9p/client.h>
34#include <net/9p/transport.h>
35#include <rdma/ib_verbs.h>
36#include <rdma/rdma_cm.h>
37
38#define P9_PORT 5640
39#define P9_RDMA_SQ_DEPTH 32
40#define P9_RDMA_RQ_DEPTH 32
41#define P9_RDMA_SEND_SGE 4
42#define P9_RDMA_RECV_SGE 4
43#define P9_RDMA_IRD 0
44#define P9_RDMA_ORD 0
45#define P9_RDMA_TIMEOUT 30000 /* 30 seconds */
46#define P9_RDMA_MAXSIZE (1024*1024) /* 1MB */
47
48/**
49 * struct p9_trans_rdma - RDMA transport instance
50 *
51 * @state: tracks the transport state machine for connection setup and tear down
52 * @cm_id: The RDMA CM ID
53 * @pd: Protection Domain pointer
54 * @qp: Queue Pair pointer
55 * @cq: Completion Queue pointer
56 * @timeout: Number of uSecs to wait for connection management events
57 * @privport: Whether a privileged port may be used
58 * @port: The port to use
59 * @sq_depth: The depth of the Send Queue
60 * @sq_sem: Semaphore for the SQ
61 * @rq_depth: The depth of the Receive Queue.
62 * @rq_sem: Semaphore for the RQ
63 * @excess_rc : Amount of posted Receive Contexts without a pending request.
64 * See rdma_request()
65 * @addr: The remote peer's address
66 * @req_lock: Protects the active request list
67 * @cm_done: Completion event for connection management tracking
68 */
69struct p9_trans_rdma {
70 enum {
71 P9_RDMA_INIT,
72 P9_RDMA_ADDR_RESOLVED,
73 P9_RDMA_ROUTE_RESOLVED,
74 P9_RDMA_CONNECTED,
75 P9_RDMA_FLUSHING,
76 P9_RDMA_CLOSING,
77 P9_RDMA_CLOSED,
78 } state;
79 struct rdma_cm_id *cm_id;
80 struct ib_pd *pd;
81 struct ib_qp *qp;
82 struct ib_cq *cq;
83 long timeout;
84 bool privport;
85 u16 port;
86 int sq_depth;
87 struct semaphore sq_sem;
88 int rq_depth;
89 struct semaphore rq_sem;
90 atomic_t excess_rc;
91 struct sockaddr_in addr;
92 spinlock_t req_lock;
93
94 struct completion cm_done;
95};
96
97struct p9_rdma_req;
98
99/**
100 * struct p9_rdma_context - Keeps track of in-process WR
101 *
102 * @cqe: completion queue entry
103 * @busa: Bus address to unmap when the WR completes
104 * @req: Keeps track of requests (send)
105 * @rc: Keepts track of replies (receive)
106 */
107struct p9_rdma_context {
108 struct ib_cqe cqe;
109 dma_addr_t busa;
110 union {
111 struct p9_req_t *req;
112 struct p9_fcall rc;
113 };
114};
115
116/**
117 * struct p9_rdma_opts - Collection of mount options
118 * @port: port of connection
119 * @privport: Whether a privileged port may be used
120 * @sq_depth: The requested depth of the SQ. This really doesn't need
121 * to be any deeper than the number of threads used in the client
122 * @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
123 * @timeout: Time to wait in msecs for CM events
124 */
125struct p9_rdma_opts {
126 short port;
127 bool privport;
128 int sq_depth;
129 int rq_depth;
130 long timeout;
131};
132
133/*
134 * Option Parsing (code inspired by NFS code)
135 */
136enum {
137 /* Options that take integer arguments */
138 Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout,
139 /* Options that take no argument */
140 Opt_privport,
141 Opt_err,
142};
143
144static match_table_t tokens = {
145 {Opt_port, "port=%u"},
146 {Opt_sq_depth, "sq=%u"},
147 {Opt_rq_depth, "rq=%u"},
148 {Opt_timeout, "timeout=%u"},
149 {Opt_privport, "privport"},
150 {Opt_err, NULL},
151};
152
153static int p9_rdma_show_options(struct seq_file *m, struct p9_client *clnt)
154{
155 struct p9_trans_rdma *rdma = clnt->trans;
156
157 if (rdma->port != P9_PORT)
158 seq_printf(m, ",port=%u", rdma->port);
159 if (rdma->sq_depth != P9_RDMA_SQ_DEPTH)
160 seq_printf(m, ",sq=%u", rdma->sq_depth);
161 if (rdma->rq_depth != P9_RDMA_RQ_DEPTH)
162 seq_printf(m, ",rq=%u", rdma->rq_depth);
163 if (rdma->timeout != P9_RDMA_TIMEOUT)
164 seq_printf(m, ",timeout=%lu", rdma->timeout);
165 if (rdma->privport)
166 seq_puts(m, ",privport");
167 return 0;
168}
169
170/**
171 * parse_opts - parse mount options into rdma options structure
172 * @params: options string passed from mount
173 * @opts: rdma transport-specific structure to parse options into
174 *
175 * Returns 0 upon success, -ERRNO upon failure
176 */
177static int parse_opts(char *params, struct p9_rdma_opts *opts)
178{
179 char *p;
180 substring_t args[MAX_OPT_ARGS];
181 int option;
182 char *options, *tmp_options;
183
184 opts->port = P9_PORT;
185 opts->sq_depth = P9_RDMA_SQ_DEPTH;
186 opts->rq_depth = P9_RDMA_RQ_DEPTH;
187 opts->timeout = P9_RDMA_TIMEOUT;
188 opts->privport = false;
189
190 if (!params)
191 return 0;
192
193 tmp_options = kstrdup(params, GFP_KERNEL);
194 if (!tmp_options) {
195 p9_debug(P9_DEBUG_ERROR,
196 "failed to allocate copy of option string\n");
197 return -ENOMEM;
198 }
199 options = tmp_options;
200
201 while ((p = strsep(&options, ",")) != NULL) {
202 int token;
203 int r;
204 if (!*p)
205 continue;
206 token = match_token(p, tokens, args);
207 if ((token != Opt_err) && (token != Opt_privport)) {
208 r = match_int(&args[0], &option);
209 if (r < 0) {
210 p9_debug(P9_DEBUG_ERROR,
211 "integer field, but no integer?\n");
212 continue;
213 }
214 }
215 switch (token) {
216 case Opt_port:
217 opts->port = option;
218 break;
219 case Opt_sq_depth:
220 opts->sq_depth = option;
221 break;
222 case Opt_rq_depth:
223 opts->rq_depth = option;
224 break;
225 case Opt_timeout:
226 opts->timeout = option;
227 break;
228 case Opt_privport:
229 opts->privport = true;
230 break;
231 default:
232 continue;
233 }
234 }
235 /* RQ must be at least as large as the SQ */
236 opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
237 kfree(tmp_options);
238 return 0;
239}
240
241static int
242p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
243{
244 struct p9_client *c = id->context;
245 struct p9_trans_rdma *rdma = c->trans;
246 switch (event->event) {
247 case RDMA_CM_EVENT_ADDR_RESOLVED:
248 BUG_ON(rdma->state != P9_RDMA_INIT);
249 rdma->state = P9_RDMA_ADDR_RESOLVED;
250 break;
251
252 case RDMA_CM_EVENT_ROUTE_RESOLVED:
253 BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
254 rdma->state = P9_RDMA_ROUTE_RESOLVED;
255 break;
256
257 case RDMA_CM_EVENT_ESTABLISHED:
258 BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
259 rdma->state = P9_RDMA_CONNECTED;
260 break;
261
262 case RDMA_CM_EVENT_DISCONNECTED:
263 if (rdma)
264 rdma->state = P9_RDMA_CLOSED;
265 c->status = Disconnected;
266 break;
267
268 case RDMA_CM_EVENT_TIMEWAIT_EXIT:
269 break;
270
271 case RDMA_CM_EVENT_ADDR_CHANGE:
272 case RDMA_CM_EVENT_ROUTE_ERROR:
273 case RDMA_CM_EVENT_DEVICE_REMOVAL:
274 case RDMA_CM_EVENT_MULTICAST_JOIN:
275 case RDMA_CM_EVENT_MULTICAST_ERROR:
276 case RDMA_CM_EVENT_REJECTED:
277 case RDMA_CM_EVENT_CONNECT_REQUEST:
278 case RDMA_CM_EVENT_CONNECT_RESPONSE:
279 case RDMA_CM_EVENT_CONNECT_ERROR:
280 case RDMA_CM_EVENT_ADDR_ERROR:
281 case RDMA_CM_EVENT_UNREACHABLE:
282 c->status = Disconnected;
283 rdma_disconnect(rdma->cm_id);
284 break;
285 default:
286 BUG();
287 }
288 complete(&rdma->cm_done);
289 return 0;
290}
291
292static void
293recv_done(struct ib_cq *cq, struct ib_wc *wc)
294{
295 struct p9_client *client = cq->cq_context;
296 struct p9_trans_rdma *rdma = client->trans;
297 struct p9_rdma_context *c =
298 container_of(wc->wr_cqe, struct p9_rdma_context, cqe);
299 struct p9_req_t *req;
300 int err = 0;
301 int16_t tag;
302
303 req = NULL;
304 ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
305 DMA_FROM_DEVICE);
306
307 if (wc->status != IB_WC_SUCCESS)
308 goto err_out;
309
310 c->rc.size = wc->byte_len;
311 err = p9_parse_header(&c->rc, NULL, NULL, &tag, 1);
312 if (err)
313 goto err_out;
314
315 req = p9_tag_lookup(client, tag);
316 if (!req)
317 goto err_out;
318
319 /* Check that we have not yet received a reply for this request.
320 */
321 if (unlikely(req->rc.sdata)) {
322 pr_err("Duplicate reply for request %d", tag);
323 goto err_out;
324 }
325
326 req->rc.size = c->rc.size;
327 req->rc.sdata = c->rc.sdata;
328 p9_client_cb(client, req, REQ_STATUS_RCVD);
329
330 out:
331 up(&rdma->rq_sem);
332 kfree(c);
333 return;
334
335 err_out:
336 p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n",
337 req, err, wc->status);
338 rdma->state = P9_RDMA_FLUSHING;
339 client->status = Disconnected;
340 goto out;
341}
342
343static void
344send_done(struct ib_cq *cq, struct ib_wc *wc)
345{
346 struct p9_client *client = cq->cq_context;
347 struct p9_trans_rdma *rdma = client->trans;
348 struct p9_rdma_context *c =
349 container_of(wc->wr_cqe, struct p9_rdma_context, cqe);
350
351 ib_dma_unmap_single(rdma->cm_id->device,
352 c->busa, c->req->tc.size,
353 DMA_TO_DEVICE);
354 up(&rdma->sq_sem);
355 p9_req_put(c->req);
356 kfree(c);
357}
358
359static void qp_event_handler(struct ib_event *event, void *context)
360{
361 p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
362 event->event, context);
363}
364
365static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
366{
367 if (!rdma)
368 return;
369
370 if (rdma->qp && !IS_ERR(rdma->qp))
371 ib_destroy_qp(rdma->qp);
372
373 if (rdma->pd && !IS_ERR(rdma->pd))
374 ib_dealloc_pd(rdma->pd);
375
376 if (rdma->cq && !IS_ERR(rdma->cq))
377 ib_free_cq(rdma->cq);
378
379 if (rdma->cm_id && !IS_ERR(rdma->cm_id))
380 rdma_destroy_id(rdma->cm_id);
381
382 kfree(rdma);
383}
384
385static int
386post_recv(struct p9_client *client, struct p9_rdma_context *c)
387{
388 struct p9_trans_rdma *rdma = client->trans;
389 struct ib_recv_wr wr;
390 struct ib_sge sge;
391
392 c->busa = ib_dma_map_single(rdma->cm_id->device,
393 c->rc.sdata, client->msize,
394 DMA_FROM_DEVICE);
395 if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
396 goto error;
397
398 c->cqe.done = recv_done;
399
400 sge.addr = c->busa;
401 sge.length = client->msize;
402 sge.lkey = rdma->pd->local_dma_lkey;
403
404 wr.next = NULL;
405 wr.wr_cqe = &c->cqe;
406 wr.sg_list = &sge;
407 wr.num_sge = 1;
408 return ib_post_recv(rdma->qp, &wr, NULL);
409
410 error:
411 p9_debug(P9_DEBUG_ERROR, "EIO\n");
412 return -EIO;
413}
414
415static int rdma_request(struct p9_client *client, struct p9_req_t *req)
416{
417 struct p9_trans_rdma *rdma = client->trans;
418 struct ib_send_wr wr;
419 struct ib_sge sge;
420 int err = 0;
421 unsigned long flags;
422 struct p9_rdma_context *c = NULL;
423 struct p9_rdma_context *rpl_context = NULL;
424
425 /* When an error occurs between posting the recv and the send,
426 * there will be a receive context posted without a pending request.
427 * Since there is no way to "un-post" it, we remember it and skip
428 * post_recv() for the next request.
429 * So here,
430 * see if we are this `next request' and need to absorb an excess rc.
431 * If yes, then drop and free our own, and do not recv_post().
432 **/
433 if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
434 if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
435 /* Got one! */
436 p9_fcall_fini(&req->rc);
437 req->rc.sdata = NULL;
438 goto dont_need_post_recv;
439 } else {
440 /* We raced and lost. */
441 atomic_inc(&rdma->excess_rc);
442 }
443 }
444
445 /* Allocate an fcall for the reply */
446 rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
447 if (!rpl_context) {
448 err = -ENOMEM;
449 goto recv_error;
450 }
451 rpl_context->rc.sdata = req->rc.sdata;
452
453 /*
454 * Post a receive buffer for this request. We need to ensure
455 * there is a reply buffer available for every outstanding
456 * request. A flushed request can result in no reply for an
457 * outstanding request, so we must keep a count to avoid
458 * overflowing the RQ.
459 */
460 if (down_interruptible(&rdma->rq_sem)) {
461 err = -EINTR;
462 goto recv_error;
463 }
464
465 err = post_recv(client, rpl_context);
466 if (err) {
467 p9_debug(P9_DEBUG_ERROR, "POST RECV failed: %d\n", err);
468 goto recv_error;
469 }
470 /* remove posted receive buffer from request structure */
471 req->rc.sdata = NULL;
472
473dont_need_post_recv:
474 /* Post the request */
475 c = kmalloc(sizeof *c, GFP_NOFS);
476 if (!c) {
477 err = -ENOMEM;
478 goto send_error;
479 }
480 c->req = req;
481
482 c->busa = ib_dma_map_single(rdma->cm_id->device,
483 c->req->tc.sdata, c->req->tc.size,
484 DMA_TO_DEVICE);
485 if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
486 err = -EIO;
487 goto send_error;
488 }
489
490 c->cqe.done = send_done;
491
492 sge.addr = c->busa;
493 sge.length = c->req->tc.size;
494 sge.lkey = rdma->pd->local_dma_lkey;
495
496 wr.next = NULL;
497 wr.wr_cqe = &c->cqe;
498 wr.opcode = IB_WR_SEND;
499 wr.send_flags = IB_SEND_SIGNALED;
500 wr.sg_list = &sge;
501 wr.num_sge = 1;
502
503 if (down_interruptible(&rdma->sq_sem)) {
504 err = -EINTR;
505 goto send_error;
506 }
507
508 /* Mark request as `sent' *before* we actually send it,
509 * because doing if after could erase the REQ_STATUS_RCVD
510 * status in case of a very fast reply.
511 */
512 req->status = REQ_STATUS_SENT;
513 err = ib_post_send(rdma->qp, &wr, NULL);
514 if (err)
515 goto send_error;
516
517 /* Success */
518 return 0;
519
520 /* Handle errors that happened during or while preparing the send: */
521 send_error:
522 req->status = REQ_STATUS_ERROR;
523 kfree(c);
524 p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);
525
526 /* Ach.
527 * We did recv_post(), but not send. We have one recv_post in excess.
528 */
529 atomic_inc(&rdma->excess_rc);
530 return err;
531
532 /* Handle errors that happened during or while preparing post_recv(): */
533 recv_error:
534 kfree(rpl_context);
535 spin_lock_irqsave(&rdma->req_lock, flags);
536 if (err != -EINTR && rdma->state < P9_RDMA_CLOSING) {
537 rdma->state = P9_RDMA_CLOSING;
538 spin_unlock_irqrestore(&rdma->req_lock, flags);
539 rdma_disconnect(rdma->cm_id);
540 } else
541 spin_unlock_irqrestore(&rdma->req_lock, flags);
542 return err;
543}
544
545static void rdma_close(struct p9_client *client)
546{
547 struct p9_trans_rdma *rdma;
548
549 if (!client)
550 return;
551
552 rdma = client->trans;
553 if (!rdma)
554 return;
555
556 client->status = Disconnected;
557 rdma_disconnect(rdma->cm_id);
558 rdma_destroy_trans(rdma);
559}
560
561/**
562 * alloc_rdma - Allocate and initialize the rdma transport structure
563 * @opts: Mount options structure
564 */
565static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
566{
567 struct p9_trans_rdma *rdma;
568
569 rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
570 if (!rdma)
571 return NULL;
572
573 rdma->port = opts->port;
574 rdma->privport = opts->privport;
575 rdma->sq_depth = opts->sq_depth;
576 rdma->rq_depth = opts->rq_depth;
577 rdma->timeout = opts->timeout;
578 spin_lock_init(&rdma->req_lock);
579 init_completion(&rdma->cm_done);
580 sema_init(&rdma->sq_sem, rdma->sq_depth);
581 sema_init(&rdma->rq_sem, rdma->rq_depth);
582 atomic_set(&rdma->excess_rc, 0);
583
584 return rdma;
585}
586
587static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
588{
589 /* Nothing to do here.
590 * We will take care of it (if we have to) in rdma_cancelled()
591 */
592 return 1;
593}
594
595/* A request has been fully flushed without a reply.
596 * That means we have posted one buffer in excess.
597 */
598static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
599{
600 struct p9_trans_rdma *rdma = client->trans;
601 atomic_inc(&rdma->excess_rc);
602 return 0;
603}
604
605static int p9_rdma_bind_privport(struct p9_trans_rdma *rdma)
606{
607 struct sockaddr_in cl = {
608 .sin_family = AF_INET,
609 .sin_addr.s_addr = htonl(INADDR_ANY),
610 };
611 int port, err = -EINVAL;
612
613 for (port = P9_DEF_MAX_RESVPORT; port >= P9_DEF_MIN_RESVPORT; port--) {
614 cl.sin_port = htons((ushort)port);
615 err = rdma_bind_addr(rdma->cm_id, (struct sockaddr *)&cl);
616 if (err != -EADDRINUSE)
617 break;
618 }
619 return err;
620}
621
622/**
623 * rdma_create_trans - Transport method for creating a transport instance
624 * @client: client instance
625 * @addr: IP address string
626 * @args: Mount options string
627 */
628static int
629rdma_create_trans(struct p9_client *client, const char *addr, char *args)
630{
631 int err;
632 struct p9_rdma_opts opts;
633 struct p9_trans_rdma *rdma;
634 struct rdma_conn_param conn_param;
635 struct ib_qp_init_attr qp_attr;
636
637 if (addr == NULL)
638 return -EINVAL;
639
640 /* Parse the transport specific mount options */
641 err = parse_opts(args, &opts);
642 if (err < 0)
643 return err;
644
645 /* Create and initialize the RDMA transport structure */
646 rdma = alloc_rdma(&opts);
647 if (!rdma)
648 return -ENOMEM;
649
650 /* Create the RDMA CM ID */
651 rdma->cm_id = rdma_create_id(&init_net, p9_cm_event_handler, client,
652 RDMA_PS_TCP, IB_QPT_RC);
653 if (IS_ERR(rdma->cm_id))
654 goto error;
655
656 /* Associate the client with the transport */
657 client->trans = rdma;
658
659 /* Bind to a privileged port if we need to */
660 if (opts.privport) {
661 err = p9_rdma_bind_privport(rdma);
662 if (err < 0) {
663 pr_err("%s (%d): problem binding to privport: %d\n",
664 __func__, task_pid_nr(current), -err);
665 goto error;
666 }
667 }
668
669 /* Resolve the server's address */
670 rdma->addr.sin_family = AF_INET;
671 rdma->addr.sin_addr.s_addr = in_aton(addr);
672 rdma->addr.sin_port = htons(opts.port);
673 err = rdma_resolve_addr(rdma->cm_id, NULL,
674 (struct sockaddr *)&rdma->addr,
675 rdma->timeout);
676 if (err)
677 goto error;
678 err = wait_for_completion_interruptible(&rdma->cm_done);
679 if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
680 goto error;
681
682 /* Resolve the route to the server */
683 err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
684 if (err)
685 goto error;
686 err = wait_for_completion_interruptible(&rdma->cm_done);
687 if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
688 goto error;
689
690 /* Create the Completion Queue */
691 rdma->cq = ib_alloc_cq_any(rdma->cm_id->device, client,
692 opts.sq_depth + opts.rq_depth + 1,
693 IB_POLL_SOFTIRQ);
694 if (IS_ERR(rdma->cq))
695 goto error;
696
697 /* Create the Protection Domain */
698 rdma->pd = ib_alloc_pd(rdma->cm_id->device, 0);
699 if (IS_ERR(rdma->pd))
700 goto error;
701
702 /* Create the Queue Pair */
703 memset(&qp_attr, 0, sizeof qp_attr);
704 qp_attr.event_handler = qp_event_handler;
705 qp_attr.qp_context = client;
706 qp_attr.cap.max_send_wr = opts.sq_depth;
707 qp_attr.cap.max_recv_wr = opts.rq_depth;
708 qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
709 qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
710 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
711 qp_attr.qp_type = IB_QPT_RC;
712 qp_attr.send_cq = rdma->cq;
713 qp_attr.recv_cq = rdma->cq;
714 err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
715 if (err)
716 goto error;
717 rdma->qp = rdma->cm_id->qp;
718
719 /* Request a connection */
720 memset(&conn_param, 0, sizeof(conn_param));
721 conn_param.private_data = NULL;
722 conn_param.private_data_len = 0;
723 conn_param.responder_resources = P9_RDMA_IRD;
724 conn_param.initiator_depth = P9_RDMA_ORD;
725 err = rdma_connect(rdma->cm_id, &conn_param);
726 if (err)
727 goto error;
728 err = wait_for_completion_interruptible(&rdma->cm_done);
729 if (err || (rdma->state != P9_RDMA_CONNECTED))
730 goto error;
731
732 client->status = Connected;
733
734 return 0;
735
736error:
737 rdma_destroy_trans(rdma);
738 return -ENOTCONN;
739}
740
741static struct p9_trans_module p9_rdma_trans = {
742 .name = "rdma",
743 .maxsize = P9_RDMA_MAXSIZE,
744 .def = 0,
745 .owner = THIS_MODULE,
746 .create = rdma_create_trans,
747 .close = rdma_close,
748 .request = rdma_request,
749 .cancel = rdma_cancel,
750 .cancelled = rdma_cancelled,
751 .show_options = p9_rdma_show_options,
752};
753
754/**
755 * p9_trans_rdma_init - Register the 9P RDMA transport driver
756 */
757static int __init p9_trans_rdma_init(void)
758{
759 v9fs_register_trans(&p9_rdma_trans);
760 return 0;
761}
762
763static void __exit p9_trans_rdma_exit(void)
764{
765 v9fs_unregister_trans(&p9_rdma_trans);
766}
767
768module_init(p9_trans_rdma_init);
769module_exit(p9_trans_rdma_exit);
770
771MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
772MODULE_DESCRIPTION("RDMA Transport for 9P");
773MODULE_LICENSE("Dual BSD/GPL");
1/*
2 * linux/fs/9p/trans_rdma.c
3 *
4 * RDMA transport layer based on the trans_fd.c implementation.
5 *
6 * Copyright (C) 2008 by Tom Tucker <tom@opengridcomputing.com>
7 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
8 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
9 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
10 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2
14 * as published by the Free Software Foundation.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to:
23 * Free Software Foundation
24 * 51 Franklin Street, Fifth Floor
25 * Boston, MA 02111-1301 USA
26 *
27 */
28
29#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31#include <linux/in.h>
32#include <linux/module.h>
33#include <linux/net.h>
34#include <linux/ipv6.h>
35#include <linux/kthread.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/un.h>
39#include <linux/uaccess.h>
40#include <linux/inet.h>
41#include <linux/idr.h>
42#include <linux/file.h>
43#include <linux/parser.h>
44#include <linux/semaphore.h>
45#include <linux/slab.h>
46#include <net/9p/9p.h>
47#include <net/9p/client.h>
48#include <net/9p/transport.h>
49#include <rdma/ib_verbs.h>
50#include <rdma/rdma_cm.h>
51
52#define P9_PORT 5640
53#define P9_RDMA_SQ_DEPTH 32
54#define P9_RDMA_RQ_DEPTH 32
55#define P9_RDMA_SEND_SGE 4
56#define P9_RDMA_RECV_SGE 4
57#define P9_RDMA_IRD 0
58#define P9_RDMA_ORD 0
59#define P9_RDMA_TIMEOUT 30000 /* 30 seconds */
60#define P9_RDMA_MAXSIZE (1024*1024) /* 1MB */
61
62/**
63 * struct p9_trans_rdma - RDMA transport instance
64 *
65 * @state: tracks the transport state machine for connection setup and tear down
66 * @cm_id: The RDMA CM ID
67 * @pd: Protection Domain pointer
68 * @qp: Queue Pair pointer
69 * @cq: Completion Queue pointer
70 * @dm_mr: DMA Memory Region pointer
71 * @lkey: The local access only memory region key
72 * @timeout: Number of uSecs to wait for connection management events
73 * @sq_depth: The depth of the Send Queue
74 * @sq_sem: Semaphore for the SQ
75 * @rq_depth: The depth of the Receive Queue.
76 * @rq_sem: Semaphore for the RQ
77 * @excess_rc : Amount of posted Receive Contexts without a pending request.
78 * See rdma_request()
79 * @addr: The remote peer's address
80 * @req_lock: Protects the active request list
81 * @cm_done: Completion event for connection management tracking
82 */
83struct p9_trans_rdma {
84 enum {
85 P9_RDMA_INIT,
86 P9_RDMA_ADDR_RESOLVED,
87 P9_RDMA_ROUTE_RESOLVED,
88 P9_RDMA_CONNECTED,
89 P9_RDMA_FLUSHING,
90 P9_RDMA_CLOSING,
91 P9_RDMA_CLOSED,
92 } state;
93 struct rdma_cm_id *cm_id;
94 struct ib_pd *pd;
95 struct ib_qp *qp;
96 struct ib_cq *cq;
97 struct ib_mr *dma_mr;
98 u32 lkey;
99 long timeout;
100 int sq_depth;
101 struct semaphore sq_sem;
102 int rq_depth;
103 struct semaphore rq_sem;
104 atomic_t excess_rc;
105 struct sockaddr_in addr;
106 spinlock_t req_lock;
107
108 struct completion cm_done;
109};
110
111/**
112 * p9_rdma_context - Keeps track of in-process WR
113 *
114 * @wc_op: The original WR op for when the CQE completes in error.
115 * @busa: Bus address to unmap when the WR completes
116 * @req: Keeps track of requests (send)
117 * @rc: Keepts track of replies (receive)
118 */
119struct p9_rdma_req;
120struct p9_rdma_context {
121 enum ib_wc_opcode wc_op;
122 dma_addr_t busa;
123 union {
124 struct p9_req_t *req;
125 struct p9_fcall *rc;
126 };
127};
128
129/**
130 * p9_rdma_opts - Collection of mount options
131 * @port: port of connection
132 * @sq_depth: The requested depth of the SQ. This really doesn't need
133 * to be any deeper than the number of threads used in the client
134 * @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
135 * @timeout: Time to wait in msecs for CM events
136 */
137struct p9_rdma_opts {
138 short port;
139 int sq_depth;
140 int rq_depth;
141 long timeout;
142};
143
144/*
145 * Option Parsing (code inspired by NFS code)
146 */
147enum {
148 /* Options that take integer arguments */
149 Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout, Opt_err,
150};
151
152static match_table_t tokens = {
153 {Opt_port, "port=%u"},
154 {Opt_sq_depth, "sq=%u"},
155 {Opt_rq_depth, "rq=%u"},
156 {Opt_timeout, "timeout=%u"},
157 {Opt_err, NULL},
158};
159
160/**
161 * parse_opts - parse mount options into rdma options structure
162 * @params: options string passed from mount
163 * @opts: rdma transport-specific structure to parse options into
164 *
165 * Returns 0 upon success, -ERRNO upon failure
166 */
167static int parse_opts(char *params, struct p9_rdma_opts *opts)
168{
169 char *p;
170 substring_t args[MAX_OPT_ARGS];
171 int option;
172 char *options, *tmp_options;
173
174 opts->port = P9_PORT;
175 opts->sq_depth = P9_RDMA_SQ_DEPTH;
176 opts->rq_depth = P9_RDMA_RQ_DEPTH;
177 opts->timeout = P9_RDMA_TIMEOUT;
178
179 if (!params)
180 return 0;
181
182 tmp_options = kstrdup(params, GFP_KERNEL);
183 if (!tmp_options) {
184 p9_debug(P9_DEBUG_ERROR,
185 "failed to allocate copy of option string\n");
186 return -ENOMEM;
187 }
188 options = tmp_options;
189
190 while ((p = strsep(&options, ",")) != NULL) {
191 int token;
192 int r;
193 if (!*p)
194 continue;
195 token = match_token(p, tokens, args);
196 if (token == Opt_err)
197 continue;
198 r = match_int(&args[0], &option);
199 if (r < 0) {
200 p9_debug(P9_DEBUG_ERROR,
201 "integer field, but no integer?\n");
202 continue;
203 }
204 switch (token) {
205 case Opt_port:
206 opts->port = option;
207 break;
208 case Opt_sq_depth:
209 opts->sq_depth = option;
210 break;
211 case Opt_rq_depth:
212 opts->rq_depth = option;
213 break;
214 case Opt_timeout:
215 opts->timeout = option;
216 break;
217 default:
218 continue;
219 }
220 }
221 /* RQ must be at least as large as the SQ */
222 opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
223 kfree(tmp_options);
224 return 0;
225}
226
227static int
228p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
229{
230 struct p9_client *c = id->context;
231 struct p9_trans_rdma *rdma = c->trans;
232 switch (event->event) {
233 case RDMA_CM_EVENT_ADDR_RESOLVED:
234 BUG_ON(rdma->state != P9_RDMA_INIT);
235 rdma->state = P9_RDMA_ADDR_RESOLVED;
236 break;
237
238 case RDMA_CM_EVENT_ROUTE_RESOLVED:
239 BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
240 rdma->state = P9_RDMA_ROUTE_RESOLVED;
241 break;
242
243 case RDMA_CM_EVENT_ESTABLISHED:
244 BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
245 rdma->state = P9_RDMA_CONNECTED;
246 break;
247
248 case RDMA_CM_EVENT_DISCONNECTED:
249 if (rdma)
250 rdma->state = P9_RDMA_CLOSED;
251 if (c)
252 c->status = Disconnected;
253 break;
254
255 case RDMA_CM_EVENT_TIMEWAIT_EXIT:
256 break;
257
258 case RDMA_CM_EVENT_ADDR_CHANGE:
259 case RDMA_CM_EVENT_ROUTE_ERROR:
260 case RDMA_CM_EVENT_DEVICE_REMOVAL:
261 case RDMA_CM_EVENT_MULTICAST_JOIN:
262 case RDMA_CM_EVENT_MULTICAST_ERROR:
263 case RDMA_CM_EVENT_REJECTED:
264 case RDMA_CM_EVENT_CONNECT_REQUEST:
265 case RDMA_CM_EVENT_CONNECT_RESPONSE:
266 case RDMA_CM_EVENT_CONNECT_ERROR:
267 case RDMA_CM_EVENT_ADDR_ERROR:
268 case RDMA_CM_EVENT_UNREACHABLE:
269 c->status = Disconnected;
270 rdma_disconnect(rdma->cm_id);
271 break;
272 default:
273 BUG();
274 }
275 complete(&rdma->cm_done);
276 return 0;
277}
278
279static void
280handle_recv(struct p9_client *client, struct p9_trans_rdma *rdma,
281 struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
282{
283 struct p9_req_t *req;
284 int err = 0;
285 int16_t tag;
286
287 req = NULL;
288 ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
289 DMA_FROM_DEVICE);
290
291 if (status != IB_WC_SUCCESS)
292 goto err_out;
293
294 err = p9_parse_header(c->rc, NULL, NULL, &tag, 1);
295 if (err)
296 goto err_out;
297
298 req = p9_tag_lookup(client, tag);
299 if (!req)
300 goto err_out;
301
302 /* Check that we have not yet received a reply for this request.
303 */
304 if (unlikely(req->rc)) {
305 pr_err("Duplicate reply for request %d", tag);
306 goto err_out;
307 }
308
309 req->rc = c->rc;
310 p9_client_cb(client, req, REQ_STATUS_RCVD);
311
312 return;
313
314 err_out:
315 p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n", req, err, status);
316 rdma->state = P9_RDMA_FLUSHING;
317 client->status = Disconnected;
318}
319
320static void
321handle_send(struct p9_client *client, struct p9_trans_rdma *rdma,
322 struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
323{
324 ib_dma_unmap_single(rdma->cm_id->device,
325 c->busa, c->req->tc->size,
326 DMA_TO_DEVICE);
327}
328
329static void qp_event_handler(struct ib_event *event, void *context)
330{
331 p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
332 event->event, context);
333}
334
335static void cq_comp_handler(struct ib_cq *cq, void *cq_context)
336{
337 struct p9_client *client = cq_context;
338 struct p9_trans_rdma *rdma = client->trans;
339 int ret;
340 struct ib_wc wc;
341
342 ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
343 while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
344 struct p9_rdma_context *c = (void *) (unsigned long) wc.wr_id;
345
346 switch (c->wc_op) {
347 case IB_WC_RECV:
348 handle_recv(client, rdma, c, wc.status, wc.byte_len);
349 up(&rdma->rq_sem);
350 break;
351
352 case IB_WC_SEND:
353 handle_send(client, rdma, c, wc.status, wc.byte_len);
354 up(&rdma->sq_sem);
355 break;
356
357 default:
358 pr_err("unexpected completion type, c->wc_op=%d, wc.opcode=%d, status=%d\n",
359 c->wc_op, wc.opcode, wc.status);
360 break;
361 }
362 kfree(c);
363 }
364}
365
366static void cq_event_handler(struct ib_event *e, void *v)
367{
368 p9_debug(P9_DEBUG_ERROR, "CQ event %d context %p\n", e->event, v);
369}
370
371static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
372{
373 if (!rdma)
374 return;
375
376 if (rdma->dma_mr && !IS_ERR(rdma->dma_mr))
377 ib_dereg_mr(rdma->dma_mr);
378
379 if (rdma->qp && !IS_ERR(rdma->qp))
380 ib_destroy_qp(rdma->qp);
381
382 if (rdma->pd && !IS_ERR(rdma->pd))
383 ib_dealloc_pd(rdma->pd);
384
385 if (rdma->cq && !IS_ERR(rdma->cq))
386 ib_destroy_cq(rdma->cq);
387
388 if (rdma->cm_id && !IS_ERR(rdma->cm_id))
389 rdma_destroy_id(rdma->cm_id);
390
391 kfree(rdma);
392}
393
394static int
395post_recv(struct p9_client *client, struct p9_rdma_context *c)
396{
397 struct p9_trans_rdma *rdma = client->trans;
398 struct ib_recv_wr wr, *bad_wr;
399 struct ib_sge sge;
400
401 c->busa = ib_dma_map_single(rdma->cm_id->device,
402 c->rc->sdata, client->msize,
403 DMA_FROM_DEVICE);
404 if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
405 goto error;
406
407 sge.addr = c->busa;
408 sge.length = client->msize;
409 sge.lkey = rdma->lkey;
410
411 wr.next = NULL;
412 c->wc_op = IB_WC_RECV;
413 wr.wr_id = (unsigned long) c;
414 wr.sg_list = &sge;
415 wr.num_sge = 1;
416 return ib_post_recv(rdma->qp, &wr, &bad_wr);
417
418 error:
419 p9_debug(P9_DEBUG_ERROR, "EIO\n");
420 return -EIO;
421}
422
423static int rdma_request(struct p9_client *client, struct p9_req_t *req)
424{
425 struct p9_trans_rdma *rdma = client->trans;
426 struct ib_send_wr wr, *bad_wr;
427 struct ib_sge sge;
428 int err = 0;
429 unsigned long flags;
430 struct p9_rdma_context *c = NULL;
431 struct p9_rdma_context *rpl_context = NULL;
432
433 /* When an error occurs between posting the recv and the send,
434 * there will be a receive context posted without a pending request.
435 * Since there is no way to "un-post" it, we remember it and skip
436 * post_recv() for the next request.
437 * So here,
438 * see if we are this `next request' and need to absorb an excess rc.
439 * If yes, then drop and free our own, and do not recv_post().
440 **/
441 if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
442 if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
443 /* Got one ! */
444 kfree(req->rc);
445 req->rc = NULL;
446 goto dont_need_post_recv;
447 } else {
448 /* We raced and lost. */
449 atomic_inc(&rdma->excess_rc);
450 }
451 }
452
453 /* Allocate an fcall for the reply */
454 rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
455 if (!rpl_context) {
456 err = -ENOMEM;
457 goto recv_error;
458 }
459 rpl_context->rc = req->rc;
460
461 /*
462 * Post a receive buffer for this request. We need to ensure
463 * there is a reply buffer available for every outstanding
464 * request. A flushed request can result in no reply for an
465 * outstanding request, so we must keep a count to avoid
466 * overflowing the RQ.
467 */
468 if (down_interruptible(&rdma->rq_sem)) {
469 err = -EINTR;
470 goto recv_error;
471 }
472
473 err = post_recv(client, rpl_context);
474 if (err) {
475 p9_debug(P9_DEBUG_FCALL, "POST RECV failed\n");
476 goto recv_error;
477 }
478 /* remove posted receive buffer from request structure */
479 req->rc = NULL;
480
481dont_need_post_recv:
482 /* Post the request */
483 c = kmalloc(sizeof *c, GFP_NOFS);
484 if (!c) {
485 err = -ENOMEM;
486 goto send_error;
487 }
488 c->req = req;
489
490 c->busa = ib_dma_map_single(rdma->cm_id->device,
491 c->req->tc->sdata, c->req->tc->size,
492 DMA_TO_DEVICE);
493 if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
494 err = -EIO;
495 goto send_error;
496 }
497
498 sge.addr = c->busa;
499 sge.length = c->req->tc->size;
500 sge.lkey = rdma->lkey;
501
502 wr.next = NULL;
503 c->wc_op = IB_WC_SEND;
504 wr.wr_id = (unsigned long) c;
505 wr.opcode = IB_WR_SEND;
506 wr.send_flags = IB_SEND_SIGNALED;
507 wr.sg_list = &sge;
508 wr.num_sge = 1;
509
510 if (down_interruptible(&rdma->sq_sem)) {
511 err = -EINTR;
512 goto send_error;
513 }
514
515 /* Mark request as `sent' *before* we actually send it,
516 * because doing if after could erase the REQ_STATUS_RCVD
517 * status in case of a very fast reply.
518 */
519 req->status = REQ_STATUS_SENT;
520 err = ib_post_send(rdma->qp, &wr, &bad_wr);
521 if (err)
522 goto send_error;
523
524 /* Success */
525 return 0;
526
527 /* Handle errors that happened during or while preparing the send: */
528 send_error:
529 req->status = REQ_STATUS_ERROR;
530 kfree(c);
531 p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);
532
533 /* Ach.
534 * We did recv_post(), but not send. We have one recv_post in excess.
535 */
536 atomic_inc(&rdma->excess_rc);
537 return err;
538
539 /* Handle errors that happened during or while preparing post_recv(): */
540 recv_error:
541 kfree(rpl_context);
542 spin_lock_irqsave(&rdma->req_lock, flags);
543 if (rdma->state < P9_RDMA_CLOSING) {
544 rdma->state = P9_RDMA_CLOSING;
545 spin_unlock_irqrestore(&rdma->req_lock, flags);
546 rdma_disconnect(rdma->cm_id);
547 } else
548 spin_unlock_irqrestore(&rdma->req_lock, flags);
549 return err;
550}
551
552static void rdma_close(struct p9_client *client)
553{
554 struct p9_trans_rdma *rdma;
555
556 if (!client)
557 return;
558
559 rdma = client->trans;
560 if (!rdma)
561 return;
562
563 client->status = Disconnected;
564 rdma_disconnect(rdma->cm_id);
565 rdma_destroy_trans(rdma);
566}
567
568/**
569 * alloc_rdma - Allocate and initialize the rdma transport structure
570 * @opts: Mount options structure
571 */
572static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
573{
574 struct p9_trans_rdma *rdma;
575
576 rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
577 if (!rdma)
578 return NULL;
579
580 rdma->sq_depth = opts->sq_depth;
581 rdma->rq_depth = opts->rq_depth;
582 rdma->timeout = opts->timeout;
583 spin_lock_init(&rdma->req_lock);
584 init_completion(&rdma->cm_done);
585 sema_init(&rdma->sq_sem, rdma->sq_depth);
586 sema_init(&rdma->rq_sem, rdma->rq_depth);
587 atomic_set(&rdma->excess_rc, 0);
588
589 return rdma;
590}
591
592static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
593{
594 /* Nothing to do here.
595 * We will take care of it (if we have to) in rdma_cancelled()
596 */
597 return 1;
598}
599
600/* A request has been fully flushed without a reply.
601 * That means we have posted one buffer in excess.
602 */
603static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
604{
605 struct p9_trans_rdma *rdma = client->trans;
606 atomic_inc(&rdma->excess_rc);
607 return 0;
608}
609
610/**
611 * trans_create_rdma - Transport method for creating atransport instance
612 * @client: client instance
613 * @addr: IP address string
614 * @args: Mount options string
615 */
616static int
617rdma_create_trans(struct p9_client *client, const char *addr, char *args)
618{
619 int err;
620 struct p9_rdma_opts opts;
621 struct p9_trans_rdma *rdma;
622 struct rdma_conn_param conn_param;
623 struct ib_qp_init_attr qp_attr;
624 struct ib_device_attr devattr;
625
626 /* Parse the transport specific mount options */
627 err = parse_opts(args, &opts);
628 if (err < 0)
629 return err;
630
631 /* Create and initialize the RDMA transport structure */
632 rdma = alloc_rdma(&opts);
633 if (!rdma)
634 return -ENOMEM;
635
636 /* Create the RDMA CM ID */
637 rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP,
638 IB_QPT_RC);
639 if (IS_ERR(rdma->cm_id))
640 goto error;
641
642 /* Associate the client with the transport */
643 client->trans = rdma;
644
645 /* Resolve the server's address */
646 rdma->addr.sin_family = AF_INET;
647 rdma->addr.sin_addr.s_addr = in_aton(addr);
648 rdma->addr.sin_port = htons(opts.port);
649 err = rdma_resolve_addr(rdma->cm_id, NULL,
650 (struct sockaddr *)&rdma->addr,
651 rdma->timeout);
652 if (err)
653 goto error;
654 err = wait_for_completion_interruptible(&rdma->cm_done);
655 if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
656 goto error;
657
658 /* Resolve the route to the server */
659 err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
660 if (err)
661 goto error;
662 err = wait_for_completion_interruptible(&rdma->cm_done);
663 if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
664 goto error;
665
666 /* Query the device attributes */
667 err = ib_query_device(rdma->cm_id->device, &devattr);
668 if (err)
669 goto error;
670
671 /* Create the Completion Queue */
672 rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
673 cq_event_handler, client,
674 opts.sq_depth + opts.rq_depth + 1, 0);
675 if (IS_ERR(rdma->cq))
676 goto error;
677 ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
678
679 /* Create the Protection Domain */
680 rdma->pd = ib_alloc_pd(rdma->cm_id->device);
681 if (IS_ERR(rdma->pd))
682 goto error;
683
684 /* Cache the DMA lkey in the transport */
685 rdma->dma_mr = NULL;
686 if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
687 rdma->lkey = rdma->cm_id->device->local_dma_lkey;
688 else {
689 rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
690 if (IS_ERR(rdma->dma_mr))
691 goto error;
692 rdma->lkey = rdma->dma_mr->lkey;
693 }
694
695 /* Create the Queue Pair */
696 memset(&qp_attr, 0, sizeof qp_attr);
697 qp_attr.event_handler = qp_event_handler;
698 qp_attr.qp_context = client;
699 qp_attr.cap.max_send_wr = opts.sq_depth;
700 qp_attr.cap.max_recv_wr = opts.rq_depth;
701 qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
702 qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
703 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
704 qp_attr.qp_type = IB_QPT_RC;
705 qp_attr.send_cq = rdma->cq;
706 qp_attr.recv_cq = rdma->cq;
707 err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
708 if (err)
709 goto error;
710 rdma->qp = rdma->cm_id->qp;
711
712 /* Request a connection */
713 memset(&conn_param, 0, sizeof(conn_param));
714 conn_param.private_data = NULL;
715 conn_param.private_data_len = 0;
716 conn_param.responder_resources = P9_RDMA_IRD;
717 conn_param.initiator_depth = P9_RDMA_ORD;
718 err = rdma_connect(rdma->cm_id, &conn_param);
719 if (err)
720 goto error;
721 err = wait_for_completion_interruptible(&rdma->cm_done);
722 if (err || (rdma->state != P9_RDMA_CONNECTED))
723 goto error;
724
725 client->status = Connected;
726
727 return 0;
728
729error:
730 rdma_destroy_trans(rdma);
731 return -ENOTCONN;
732}
733
734static struct p9_trans_module p9_rdma_trans = {
735 .name = "rdma",
736 .maxsize = P9_RDMA_MAXSIZE,
737 .def = 0,
738 .owner = THIS_MODULE,
739 .create = rdma_create_trans,
740 .close = rdma_close,
741 .request = rdma_request,
742 .cancel = rdma_cancel,
743 .cancelled = rdma_cancelled,
744};
745
746/**
747 * p9_trans_rdma_init - Register the 9P RDMA transport driver
748 */
749static int __init p9_trans_rdma_init(void)
750{
751 v9fs_register_trans(&p9_rdma_trans);
752 return 0;
753}
754
755static void __exit p9_trans_rdma_exit(void)
756{
757 v9fs_unregister_trans(&p9_rdma_trans);
758}
759
760module_init(p9_trans_rdma_init);
761module_exit(p9_trans_rdma_exit);
762
763MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
764MODULE_DESCRIPTION("RDMA Transport for 9P");
765MODULE_LICENSE("Dual BSD/GPL");