RDMA/cm: Add RDMA CM support for IBoE devices
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / core / ucma.c
1 /*
2 * Copyright (c) 2005-2006 Intel Corporation. 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
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/completion.h>
34 #include <linux/file.h>
35 #include <linux/mutex.h>
36 #include <linux/poll.h>
37 #include <linux/sched.h>
38 #include <linux/idr.h>
39 #include <linux/in.h>
40 #include <linux/in6.h>
41 #include <linux/miscdevice.h>
42 #include <linux/slab.h>
43
44 #include <rdma/rdma_user_cm.h>
45 #include <rdma/ib_marshall.h>
46 #include <rdma/rdma_cm.h>
47 #include <rdma/rdma_cm_ib.h>
48
49 MODULE_AUTHOR("Sean Hefty");
50 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
51 MODULE_LICENSE("Dual BSD/GPL");
52
53 enum {
54 UCMA_MAX_BACKLOG = 128
55 };
56
57 struct ucma_file {
58 struct mutex mut;
59 struct file *filp;
60 struct list_head ctx_list;
61 struct list_head event_list;
62 wait_queue_head_t poll_wait;
63 };
64
65 struct ucma_context {
66 int id;
67 struct completion comp;
68 atomic_t ref;
69 int events_reported;
70 int backlog;
71
72 struct ucma_file *file;
73 struct rdma_cm_id *cm_id;
74 u64 uid;
75
76 struct list_head list;
77 struct list_head mc_list;
78 };
79
80 struct ucma_multicast {
81 struct ucma_context *ctx;
82 int id;
83 int events_reported;
84
85 u64 uid;
86 struct list_head list;
87 struct sockaddr_storage addr;
88 };
89
90 struct ucma_event {
91 struct ucma_context *ctx;
92 struct ucma_multicast *mc;
93 struct list_head list;
94 struct rdma_cm_id *cm_id;
95 struct rdma_ucm_event_resp resp;
96 };
97
98 static DEFINE_MUTEX(mut);
99 static DEFINE_IDR(ctx_idr);
100 static DEFINE_IDR(multicast_idr);
101
102 static inline struct ucma_context *_ucma_find_context(int id,
103 struct ucma_file *file)
104 {
105 struct ucma_context *ctx;
106
107 ctx = idr_find(&ctx_idr, id);
108 if (!ctx)
109 ctx = ERR_PTR(-ENOENT);
110 else if (ctx->file != file)
111 ctx = ERR_PTR(-EINVAL);
112 return ctx;
113 }
114
115 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
116 {
117 struct ucma_context *ctx;
118
119 mutex_lock(&mut);
120 ctx = _ucma_find_context(id, file);
121 if (!IS_ERR(ctx))
122 atomic_inc(&ctx->ref);
123 mutex_unlock(&mut);
124 return ctx;
125 }
126
127 static void ucma_put_ctx(struct ucma_context *ctx)
128 {
129 if (atomic_dec_and_test(&ctx->ref))
130 complete(&ctx->comp);
131 }
132
133 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
134 {
135 struct ucma_context *ctx;
136 int ret;
137
138 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
139 if (!ctx)
140 return NULL;
141
142 atomic_set(&ctx->ref, 1);
143 init_completion(&ctx->comp);
144 INIT_LIST_HEAD(&ctx->mc_list);
145 ctx->file = file;
146
147 do {
148 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
149 if (!ret)
150 goto error;
151
152 mutex_lock(&mut);
153 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
154 mutex_unlock(&mut);
155 } while (ret == -EAGAIN);
156
157 if (ret)
158 goto error;
159
160 list_add_tail(&ctx->list, &file->ctx_list);
161 return ctx;
162
163 error:
164 kfree(ctx);
165 return NULL;
166 }
167
168 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
169 {
170 struct ucma_multicast *mc;
171 int ret;
172
173 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
174 if (!mc)
175 return NULL;
176
177 do {
178 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
179 if (!ret)
180 goto error;
181
182 mutex_lock(&mut);
183 ret = idr_get_new(&multicast_idr, mc, &mc->id);
184 mutex_unlock(&mut);
185 } while (ret == -EAGAIN);
186
187 if (ret)
188 goto error;
189
190 mc->ctx = ctx;
191 list_add_tail(&mc->list, &ctx->mc_list);
192 return mc;
193
194 error:
195 kfree(mc);
196 return NULL;
197 }
198
199 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
200 struct rdma_conn_param *src)
201 {
202 if (src->private_data_len)
203 memcpy(dst->private_data, src->private_data,
204 src->private_data_len);
205 dst->private_data_len = src->private_data_len;
206 dst->responder_resources =src->responder_resources;
207 dst->initiator_depth = src->initiator_depth;
208 dst->flow_control = src->flow_control;
209 dst->retry_count = src->retry_count;
210 dst->rnr_retry_count = src->rnr_retry_count;
211 dst->srq = src->srq;
212 dst->qp_num = src->qp_num;
213 }
214
215 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
216 struct rdma_ud_param *src)
217 {
218 if (src->private_data_len)
219 memcpy(dst->private_data, src->private_data,
220 src->private_data_len);
221 dst->private_data_len = src->private_data_len;
222 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
223 dst->qp_num = src->qp_num;
224 dst->qkey = src->qkey;
225 }
226
227 static void ucma_set_event_context(struct ucma_context *ctx,
228 struct rdma_cm_event *event,
229 struct ucma_event *uevent)
230 {
231 uevent->ctx = ctx;
232 switch (event->event) {
233 case RDMA_CM_EVENT_MULTICAST_JOIN:
234 case RDMA_CM_EVENT_MULTICAST_ERROR:
235 uevent->mc = (struct ucma_multicast *)
236 event->param.ud.private_data;
237 uevent->resp.uid = uevent->mc->uid;
238 uevent->resp.id = uevent->mc->id;
239 break;
240 default:
241 uevent->resp.uid = ctx->uid;
242 uevent->resp.id = ctx->id;
243 break;
244 }
245 }
246
247 static int ucma_event_handler(struct rdma_cm_id *cm_id,
248 struct rdma_cm_event *event)
249 {
250 struct ucma_event *uevent;
251 struct ucma_context *ctx = cm_id->context;
252 int ret = 0;
253
254 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
255 if (!uevent)
256 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
257
258 uevent->cm_id = cm_id;
259 ucma_set_event_context(ctx, event, uevent);
260 uevent->resp.event = event->event;
261 uevent->resp.status = event->status;
262 if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB)
263 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
264 else
265 ucma_copy_conn_event(&uevent->resp.param.conn,
266 &event->param.conn);
267
268 mutex_lock(&ctx->file->mut);
269 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
270 if (!ctx->backlog) {
271 ret = -ENOMEM;
272 kfree(uevent);
273 goto out;
274 }
275 ctx->backlog--;
276 } else if (!ctx->uid) {
277 /*
278 * We ignore events for new connections until userspace has set
279 * their context. This can only happen if an error occurs on a
280 * new connection before the user accepts it. This is okay,
281 * since the accept will just fail later.
282 */
283 kfree(uevent);
284 goto out;
285 }
286
287 list_add_tail(&uevent->list, &ctx->file->event_list);
288 wake_up_interruptible(&ctx->file->poll_wait);
289 out:
290 mutex_unlock(&ctx->file->mut);
291 return ret;
292 }
293
294 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
295 int in_len, int out_len)
296 {
297 struct ucma_context *ctx;
298 struct rdma_ucm_get_event cmd;
299 struct ucma_event *uevent;
300 int ret = 0;
301 DEFINE_WAIT(wait);
302
303 if (out_len < sizeof uevent->resp)
304 return -ENOSPC;
305
306 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
307 return -EFAULT;
308
309 mutex_lock(&file->mut);
310 while (list_empty(&file->event_list)) {
311 mutex_unlock(&file->mut);
312
313 if (file->filp->f_flags & O_NONBLOCK)
314 return -EAGAIN;
315
316 if (wait_event_interruptible(file->poll_wait,
317 !list_empty(&file->event_list)))
318 return -ERESTARTSYS;
319
320 mutex_lock(&file->mut);
321 }
322
323 uevent = list_entry(file->event_list.next, struct ucma_event, list);
324
325 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
326 ctx = ucma_alloc_ctx(file);
327 if (!ctx) {
328 ret = -ENOMEM;
329 goto done;
330 }
331 uevent->ctx->backlog++;
332 ctx->cm_id = uevent->cm_id;
333 ctx->cm_id->context = ctx;
334 uevent->resp.id = ctx->id;
335 }
336
337 if (copy_to_user((void __user *)(unsigned long)cmd.response,
338 &uevent->resp, sizeof uevent->resp)) {
339 ret = -EFAULT;
340 goto done;
341 }
342
343 list_del(&uevent->list);
344 uevent->ctx->events_reported++;
345 if (uevent->mc)
346 uevent->mc->events_reported++;
347 kfree(uevent);
348 done:
349 mutex_unlock(&file->mut);
350 return ret;
351 }
352
353 static ssize_t ucma_create_id(struct ucma_file *file,
354 const char __user *inbuf,
355 int in_len, int out_len)
356 {
357 struct rdma_ucm_create_id cmd;
358 struct rdma_ucm_create_id_resp resp;
359 struct ucma_context *ctx;
360 int ret;
361
362 if (out_len < sizeof(resp))
363 return -ENOSPC;
364
365 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
366 return -EFAULT;
367
368 mutex_lock(&file->mut);
369 ctx = ucma_alloc_ctx(file);
370 mutex_unlock(&file->mut);
371 if (!ctx)
372 return -ENOMEM;
373
374 ctx->uid = cmd.uid;
375 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps);
376 if (IS_ERR(ctx->cm_id)) {
377 ret = PTR_ERR(ctx->cm_id);
378 goto err1;
379 }
380
381 resp.id = ctx->id;
382 if (copy_to_user((void __user *)(unsigned long)cmd.response,
383 &resp, sizeof(resp))) {
384 ret = -EFAULT;
385 goto err2;
386 }
387 return 0;
388
389 err2:
390 rdma_destroy_id(ctx->cm_id);
391 err1:
392 mutex_lock(&mut);
393 idr_remove(&ctx_idr, ctx->id);
394 mutex_unlock(&mut);
395 kfree(ctx);
396 return ret;
397 }
398
399 static void ucma_cleanup_multicast(struct ucma_context *ctx)
400 {
401 struct ucma_multicast *mc, *tmp;
402
403 mutex_lock(&mut);
404 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
405 list_del(&mc->list);
406 idr_remove(&multicast_idr, mc->id);
407 kfree(mc);
408 }
409 mutex_unlock(&mut);
410 }
411
412 static void ucma_cleanup_events(struct ucma_context *ctx)
413 {
414 struct ucma_event *uevent, *tmp;
415
416 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
417 if (uevent->ctx != ctx)
418 continue;
419
420 list_del(&uevent->list);
421
422 /* clear incoming connections. */
423 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
424 rdma_destroy_id(uevent->cm_id);
425
426 kfree(uevent);
427 }
428 }
429
430 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
431 {
432 struct ucma_event *uevent, *tmp;
433
434 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
435 if (uevent->mc != mc)
436 continue;
437
438 list_del(&uevent->list);
439 kfree(uevent);
440 }
441 }
442
443 static int ucma_free_ctx(struct ucma_context *ctx)
444 {
445 int events_reported;
446
447 /* No new events will be generated after destroying the id. */
448 rdma_destroy_id(ctx->cm_id);
449
450 ucma_cleanup_multicast(ctx);
451
452 /* Cleanup events not yet reported to the user. */
453 mutex_lock(&ctx->file->mut);
454 ucma_cleanup_events(ctx);
455 list_del(&ctx->list);
456 mutex_unlock(&ctx->file->mut);
457
458 events_reported = ctx->events_reported;
459 kfree(ctx);
460 return events_reported;
461 }
462
463 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
464 int in_len, int out_len)
465 {
466 struct rdma_ucm_destroy_id cmd;
467 struct rdma_ucm_destroy_id_resp resp;
468 struct ucma_context *ctx;
469 int ret = 0;
470
471 if (out_len < sizeof(resp))
472 return -ENOSPC;
473
474 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
475 return -EFAULT;
476
477 mutex_lock(&mut);
478 ctx = _ucma_find_context(cmd.id, file);
479 if (!IS_ERR(ctx))
480 idr_remove(&ctx_idr, ctx->id);
481 mutex_unlock(&mut);
482
483 if (IS_ERR(ctx))
484 return PTR_ERR(ctx);
485
486 ucma_put_ctx(ctx);
487 wait_for_completion(&ctx->comp);
488 resp.events_reported = ucma_free_ctx(ctx);
489
490 if (copy_to_user((void __user *)(unsigned long)cmd.response,
491 &resp, sizeof(resp)))
492 ret = -EFAULT;
493
494 return ret;
495 }
496
497 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
498 int in_len, int out_len)
499 {
500 struct rdma_ucm_bind_addr cmd;
501 struct ucma_context *ctx;
502 int ret;
503
504 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
505 return -EFAULT;
506
507 ctx = ucma_get_ctx(file, cmd.id);
508 if (IS_ERR(ctx))
509 return PTR_ERR(ctx);
510
511 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
512 ucma_put_ctx(ctx);
513 return ret;
514 }
515
516 static ssize_t ucma_resolve_addr(struct ucma_file *file,
517 const char __user *inbuf,
518 int in_len, int out_len)
519 {
520 struct rdma_ucm_resolve_addr cmd;
521 struct ucma_context *ctx;
522 int ret;
523
524 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
525 return -EFAULT;
526
527 ctx = ucma_get_ctx(file, cmd.id);
528 if (IS_ERR(ctx))
529 return PTR_ERR(ctx);
530
531 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
532 (struct sockaddr *) &cmd.dst_addr,
533 cmd.timeout_ms);
534 ucma_put_ctx(ctx);
535 return ret;
536 }
537
538 static ssize_t ucma_resolve_route(struct ucma_file *file,
539 const char __user *inbuf,
540 int in_len, int out_len)
541 {
542 struct rdma_ucm_resolve_route cmd;
543 struct ucma_context *ctx;
544 int ret;
545
546 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
547 return -EFAULT;
548
549 ctx = ucma_get_ctx(file, cmd.id);
550 if (IS_ERR(ctx))
551 return PTR_ERR(ctx);
552
553 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
554 ucma_put_ctx(ctx);
555 return ret;
556 }
557
558 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
559 struct rdma_route *route)
560 {
561 struct rdma_dev_addr *dev_addr;
562
563 resp->num_paths = route->num_paths;
564 switch (route->num_paths) {
565 case 0:
566 dev_addr = &route->addr.dev_addr;
567 rdma_addr_get_dgid(dev_addr,
568 (union ib_gid *) &resp->ib_route[0].dgid);
569 rdma_addr_get_sgid(dev_addr,
570 (union ib_gid *) &resp->ib_route[0].sgid);
571 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
572 break;
573 case 2:
574 ib_copy_path_rec_to_user(&resp->ib_route[1],
575 &route->path_rec[1]);
576 /* fall through */
577 case 1:
578 ib_copy_path_rec_to_user(&resp->ib_route[0],
579 &route->path_rec[0]);
580 break;
581 default:
582 break;
583 }
584 }
585
586 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
587 struct rdma_route *route)
588 {
589 struct rdma_dev_addr *dev_addr;
590
591 resp->num_paths = route->num_paths;
592 switch (route->num_paths) {
593 case 0:
594 dev_addr = &route->addr.dev_addr;
595 iboe_mac_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
596 dev_addr->dst_dev_addr);
597 iboe_addr_get_sgid(dev_addr,
598 (union ib_gid *) &resp->ib_route[0].sgid);
599 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
600 break;
601 case 2:
602 ib_copy_path_rec_to_user(&resp->ib_route[1],
603 &route->path_rec[1]);
604 /* fall through */
605 case 1:
606 ib_copy_path_rec_to_user(&resp->ib_route[0],
607 &route->path_rec[0]);
608 break;
609 default:
610 break;
611 }
612 }
613
614 static ssize_t ucma_query_route(struct ucma_file *file,
615 const char __user *inbuf,
616 int in_len, int out_len)
617 {
618 struct rdma_ucm_query_route cmd;
619 struct rdma_ucm_query_route_resp resp;
620 struct ucma_context *ctx;
621 struct sockaddr *addr;
622 int ret = 0;
623
624 if (out_len < sizeof(resp))
625 return -ENOSPC;
626
627 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
628 return -EFAULT;
629
630 ctx = ucma_get_ctx(file, cmd.id);
631 if (IS_ERR(ctx))
632 return PTR_ERR(ctx);
633
634 memset(&resp, 0, sizeof resp);
635 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
636 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
637 sizeof(struct sockaddr_in) :
638 sizeof(struct sockaddr_in6));
639 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
640 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
641 sizeof(struct sockaddr_in) :
642 sizeof(struct sockaddr_in6));
643 if (!ctx->cm_id->device)
644 goto out;
645
646 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
647 resp.port_num = ctx->cm_id->port_num;
648 if (rdma_node_get_transport(ctx->cm_id->device->node_type) == RDMA_TRANSPORT_IB) {
649 switch (rdma_port_get_link_layer(ctx->cm_id->device, ctx->cm_id->port_num)) {
650 case IB_LINK_LAYER_INFINIBAND:
651 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
652 break;
653 case IB_LINK_LAYER_ETHERNET:
654 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
655 break;
656 default:
657 break;
658 }
659 }
660
661 out:
662 if (copy_to_user((void __user *)(unsigned long)cmd.response,
663 &resp, sizeof(resp)))
664 ret = -EFAULT;
665
666 ucma_put_ctx(ctx);
667 return ret;
668 }
669
670 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
671 struct rdma_ucm_conn_param *src)
672 {
673 dst->private_data = src->private_data;
674 dst->private_data_len = src->private_data_len;
675 dst->responder_resources =src->responder_resources;
676 dst->initiator_depth = src->initiator_depth;
677 dst->flow_control = src->flow_control;
678 dst->retry_count = src->retry_count;
679 dst->rnr_retry_count = src->rnr_retry_count;
680 dst->srq = src->srq;
681 dst->qp_num = src->qp_num;
682 }
683
684 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
685 int in_len, int out_len)
686 {
687 struct rdma_ucm_connect cmd;
688 struct rdma_conn_param conn_param;
689 struct ucma_context *ctx;
690 int ret;
691
692 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
693 return -EFAULT;
694
695 if (!cmd.conn_param.valid)
696 return -EINVAL;
697
698 ctx = ucma_get_ctx(file, cmd.id);
699 if (IS_ERR(ctx))
700 return PTR_ERR(ctx);
701
702 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
703 ret = rdma_connect(ctx->cm_id, &conn_param);
704 ucma_put_ctx(ctx);
705 return ret;
706 }
707
708 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
709 int in_len, int out_len)
710 {
711 struct rdma_ucm_listen cmd;
712 struct ucma_context *ctx;
713 int ret;
714
715 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
716 return -EFAULT;
717
718 ctx = ucma_get_ctx(file, cmd.id);
719 if (IS_ERR(ctx))
720 return PTR_ERR(ctx);
721
722 ctx->backlog = cmd.backlog > 0 && cmd.backlog < UCMA_MAX_BACKLOG ?
723 cmd.backlog : UCMA_MAX_BACKLOG;
724 ret = rdma_listen(ctx->cm_id, ctx->backlog);
725 ucma_put_ctx(ctx);
726 return ret;
727 }
728
729 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
730 int in_len, int out_len)
731 {
732 struct rdma_ucm_accept cmd;
733 struct rdma_conn_param conn_param;
734 struct ucma_context *ctx;
735 int ret;
736
737 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
738 return -EFAULT;
739
740 ctx = ucma_get_ctx(file, cmd.id);
741 if (IS_ERR(ctx))
742 return PTR_ERR(ctx);
743
744 if (cmd.conn_param.valid) {
745 ctx->uid = cmd.uid;
746 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
747 ret = rdma_accept(ctx->cm_id, &conn_param);
748 } else
749 ret = rdma_accept(ctx->cm_id, NULL);
750
751 ucma_put_ctx(ctx);
752 return ret;
753 }
754
755 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
756 int in_len, int out_len)
757 {
758 struct rdma_ucm_reject cmd;
759 struct ucma_context *ctx;
760 int ret;
761
762 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
763 return -EFAULT;
764
765 ctx = ucma_get_ctx(file, cmd.id);
766 if (IS_ERR(ctx))
767 return PTR_ERR(ctx);
768
769 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
770 ucma_put_ctx(ctx);
771 return ret;
772 }
773
774 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
775 int in_len, int out_len)
776 {
777 struct rdma_ucm_disconnect cmd;
778 struct ucma_context *ctx;
779 int ret;
780
781 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
782 return -EFAULT;
783
784 ctx = ucma_get_ctx(file, cmd.id);
785 if (IS_ERR(ctx))
786 return PTR_ERR(ctx);
787
788 ret = rdma_disconnect(ctx->cm_id);
789 ucma_put_ctx(ctx);
790 return ret;
791 }
792
793 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
794 const char __user *inbuf,
795 int in_len, int out_len)
796 {
797 struct rdma_ucm_init_qp_attr cmd;
798 struct ib_uverbs_qp_attr resp;
799 struct ucma_context *ctx;
800 struct ib_qp_attr qp_attr;
801 int ret;
802
803 if (out_len < sizeof(resp))
804 return -ENOSPC;
805
806 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
807 return -EFAULT;
808
809 ctx = ucma_get_ctx(file, cmd.id);
810 if (IS_ERR(ctx))
811 return PTR_ERR(ctx);
812
813 resp.qp_attr_mask = 0;
814 memset(&qp_attr, 0, sizeof qp_attr);
815 qp_attr.qp_state = cmd.qp_state;
816 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
817 if (ret)
818 goto out;
819
820 ib_copy_qp_attr_to_user(&resp, &qp_attr);
821 if (copy_to_user((void __user *)(unsigned long)cmd.response,
822 &resp, sizeof(resp)))
823 ret = -EFAULT;
824
825 out:
826 ucma_put_ctx(ctx);
827 return ret;
828 }
829
830 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
831 void *optval, size_t optlen)
832 {
833 int ret = 0;
834
835 switch (optname) {
836 case RDMA_OPTION_ID_TOS:
837 if (optlen != sizeof(u8)) {
838 ret = -EINVAL;
839 break;
840 }
841 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
842 break;
843 default:
844 ret = -ENOSYS;
845 }
846
847 return ret;
848 }
849
850 static int ucma_set_ib_path(struct ucma_context *ctx,
851 struct ib_path_rec_data *path_data, size_t optlen)
852 {
853 struct ib_sa_path_rec sa_path;
854 struct rdma_cm_event event;
855 int ret;
856
857 if (optlen % sizeof(*path_data))
858 return -EINVAL;
859
860 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
861 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
862 IB_PATH_BIDIRECTIONAL))
863 break;
864 }
865
866 if (!optlen)
867 return -EINVAL;
868
869 ib_sa_unpack_path(path_data->path_rec, &sa_path);
870 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
871 if (ret)
872 return ret;
873
874 memset(&event, 0, sizeof event);
875 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
876 return ucma_event_handler(ctx->cm_id, &event);
877 }
878
879 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
880 void *optval, size_t optlen)
881 {
882 int ret;
883
884 switch (optname) {
885 case RDMA_OPTION_IB_PATH:
886 ret = ucma_set_ib_path(ctx, optval, optlen);
887 break;
888 default:
889 ret = -ENOSYS;
890 }
891
892 return ret;
893 }
894
895 static int ucma_set_option_level(struct ucma_context *ctx, int level,
896 int optname, void *optval, size_t optlen)
897 {
898 int ret;
899
900 switch (level) {
901 case RDMA_OPTION_ID:
902 ret = ucma_set_option_id(ctx, optname, optval, optlen);
903 break;
904 case RDMA_OPTION_IB:
905 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
906 break;
907 default:
908 ret = -ENOSYS;
909 }
910
911 return ret;
912 }
913
914 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
915 int in_len, int out_len)
916 {
917 struct rdma_ucm_set_option cmd;
918 struct ucma_context *ctx;
919 void *optval;
920 int ret;
921
922 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
923 return -EFAULT;
924
925 ctx = ucma_get_ctx(file, cmd.id);
926 if (IS_ERR(ctx))
927 return PTR_ERR(ctx);
928
929 optval = kmalloc(cmd.optlen, GFP_KERNEL);
930 if (!optval) {
931 ret = -ENOMEM;
932 goto out1;
933 }
934
935 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
936 cmd.optlen)) {
937 ret = -EFAULT;
938 goto out2;
939 }
940
941 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
942 cmd.optlen);
943 out2:
944 kfree(optval);
945 out1:
946 ucma_put_ctx(ctx);
947 return ret;
948 }
949
950 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
951 int in_len, int out_len)
952 {
953 struct rdma_ucm_notify cmd;
954 struct ucma_context *ctx;
955 int ret;
956
957 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
958 return -EFAULT;
959
960 ctx = ucma_get_ctx(file, cmd.id);
961 if (IS_ERR(ctx))
962 return PTR_ERR(ctx);
963
964 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
965 ucma_put_ctx(ctx);
966 return ret;
967 }
968
969 static ssize_t ucma_join_multicast(struct ucma_file *file,
970 const char __user *inbuf,
971 int in_len, int out_len)
972 {
973 struct rdma_ucm_join_mcast cmd;
974 struct rdma_ucm_create_id_resp resp;
975 struct ucma_context *ctx;
976 struct ucma_multicast *mc;
977 int ret;
978
979 if (out_len < sizeof(resp))
980 return -ENOSPC;
981
982 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
983 return -EFAULT;
984
985 ctx = ucma_get_ctx(file, cmd.id);
986 if (IS_ERR(ctx))
987 return PTR_ERR(ctx);
988
989 mutex_lock(&file->mut);
990 mc = ucma_alloc_multicast(ctx);
991 if (!mc) {
992 ret = -ENOMEM;
993 goto err1;
994 }
995
996 mc->uid = cmd.uid;
997 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
998 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
999 if (ret)
1000 goto err2;
1001
1002 resp.id = mc->id;
1003 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1004 &resp, sizeof(resp))) {
1005 ret = -EFAULT;
1006 goto err3;
1007 }
1008
1009 mutex_unlock(&file->mut);
1010 ucma_put_ctx(ctx);
1011 return 0;
1012
1013 err3:
1014 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1015 ucma_cleanup_mc_events(mc);
1016 err2:
1017 mutex_lock(&mut);
1018 idr_remove(&multicast_idr, mc->id);
1019 mutex_unlock(&mut);
1020 list_del(&mc->list);
1021 kfree(mc);
1022 err1:
1023 mutex_unlock(&file->mut);
1024 ucma_put_ctx(ctx);
1025 return ret;
1026 }
1027
1028 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1029 const char __user *inbuf,
1030 int in_len, int out_len)
1031 {
1032 struct rdma_ucm_destroy_id cmd;
1033 struct rdma_ucm_destroy_id_resp resp;
1034 struct ucma_multicast *mc;
1035 int ret = 0;
1036
1037 if (out_len < sizeof(resp))
1038 return -ENOSPC;
1039
1040 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1041 return -EFAULT;
1042
1043 mutex_lock(&mut);
1044 mc = idr_find(&multicast_idr, cmd.id);
1045 if (!mc)
1046 mc = ERR_PTR(-ENOENT);
1047 else if (mc->ctx->file != file)
1048 mc = ERR_PTR(-EINVAL);
1049 else {
1050 idr_remove(&multicast_idr, mc->id);
1051 atomic_inc(&mc->ctx->ref);
1052 }
1053 mutex_unlock(&mut);
1054
1055 if (IS_ERR(mc)) {
1056 ret = PTR_ERR(mc);
1057 goto out;
1058 }
1059
1060 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1061 mutex_lock(&mc->ctx->file->mut);
1062 ucma_cleanup_mc_events(mc);
1063 list_del(&mc->list);
1064 mutex_unlock(&mc->ctx->file->mut);
1065
1066 ucma_put_ctx(mc->ctx);
1067 resp.events_reported = mc->events_reported;
1068 kfree(mc);
1069
1070 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1071 &resp, sizeof(resp)))
1072 ret = -EFAULT;
1073 out:
1074 return ret;
1075 }
1076
1077 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1078 {
1079 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1080 if (file1 < file2) {
1081 mutex_lock(&file1->mut);
1082 mutex_lock(&file2->mut);
1083 } else {
1084 mutex_lock(&file2->mut);
1085 mutex_lock(&file1->mut);
1086 }
1087 }
1088
1089 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1090 {
1091 if (file1 < file2) {
1092 mutex_unlock(&file2->mut);
1093 mutex_unlock(&file1->mut);
1094 } else {
1095 mutex_unlock(&file1->mut);
1096 mutex_unlock(&file2->mut);
1097 }
1098 }
1099
1100 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1101 {
1102 struct ucma_event *uevent, *tmp;
1103
1104 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1105 if (uevent->ctx == ctx)
1106 list_move_tail(&uevent->list, &file->event_list);
1107 }
1108
1109 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1110 const char __user *inbuf,
1111 int in_len, int out_len)
1112 {
1113 struct rdma_ucm_migrate_id cmd;
1114 struct rdma_ucm_migrate_resp resp;
1115 struct ucma_context *ctx;
1116 struct file *filp;
1117 struct ucma_file *cur_file;
1118 int ret = 0;
1119
1120 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1121 return -EFAULT;
1122
1123 /* Get current fd to protect against it being closed */
1124 filp = fget(cmd.fd);
1125 if (!filp)
1126 return -ENOENT;
1127
1128 /* Validate current fd and prevent destruction of id. */
1129 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1130 if (IS_ERR(ctx)) {
1131 ret = PTR_ERR(ctx);
1132 goto file_put;
1133 }
1134
1135 cur_file = ctx->file;
1136 if (cur_file == new_file) {
1137 resp.events_reported = ctx->events_reported;
1138 goto response;
1139 }
1140
1141 /*
1142 * Migrate events between fd's, maintaining order, and avoiding new
1143 * events being added before existing events.
1144 */
1145 ucma_lock_files(cur_file, new_file);
1146 mutex_lock(&mut);
1147
1148 list_move_tail(&ctx->list, &new_file->ctx_list);
1149 ucma_move_events(ctx, new_file);
1150 ctx->file = new_file;
1151 resp.events_reported = ctx->events_reported;
1152
1153 mutex_unlock(&mut);
1154 ucma_unlock_files(cur_file, new_file);
1155
1156 response:
1157 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1158 &resp, sizeof(resp)))
1159 ret = -EFAULT;
1160
1161 ucma_put_ctx(ctx);
1162 file_put:
1163 fput(filp);
1164 return ret;
1165 }
1166
1167 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1168 const char __user *inbuf,
1169 int in_len, int out_len) = {
1170 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1171 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1172 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1173 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1174 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1175 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1176 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1177 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1178 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1179 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1180 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1181 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1182 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1183 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1184 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1185 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1186 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1187 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1188 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1189 };
1190
1191 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1192 size_t len, loff_t *pos)
1193 {
1194 struct ucma_file *file = filp->private_data;
1195 struct rdma_ucm_cmd_hdr hdr;
1196 ssize_t ret;
1197
1198 if (len < sizeof(hdr))
1199 return -EINVAL;
1200
1201 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1202 return -EFAULT;
1203
1204 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1205 return -EINVAL;
1206
1207 if (hdr.in + sizeof(hdr) > len)
1208 return -EINVAL;
1209
1210 if (!ucma_cmd_table[hdr.cmd])
1211 return -ENOSYS;
1212
1213 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1214 if (!ret)
1215 ret = len;
1216
1217 return ret;
1218 }
1219
1220 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1221 {
1222 struct ucma_file *file = filp->private_data;
1223 unsigned int mask = 0;
1224
1225 poll_wait(filp, &file->poll_wait, wait);
1226
1227 if (!list_empty(&file->event_list))
1228 mask = POLLIN | POLLRDNORM;
1229
1230 return mask;
1231 }
1232
1233 /*
1234 * ucma_open() does not need the BKL:
1235 *
1236 * - no global state is referred to;
1237 * - there is no ioctl method to race against;
1238 * - no further module initialization is required for open to work
1239 * after the device is registered.
1240 */
1241 static int ucma_open(struct inode *inode, struct file *filp)
1242 {
1243 struct ucma_file *file;
1244
1245 file = kmalloc(sizeof *file, GFP_KERNEL);
1246 if (!file)
1247 return -ENOMEM;
1248
1249 INIT_LIST_HEAD(&file->event_list);
1250 INIT_LIST_HEAD(&file->ctx_list);
1251 init_waitqueue_head(&file->poll_wait);
1252 mutex_init(&file->mut);
1253
1254 filp->private_data = file;
1255 file->filp = filp;
1256
1257 return nonseekable_open(inode, filp);
1258 }
1259
1260 static int ucma_close(struct inode *inode, struct file *filp)
1261 {
1262 struct ucma_file *file = filp->private_data;
1263 struct ucma_context *ctx, *tmp;
1264
1265 mutex_lock(&file->mut);
1266 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1267 mutex_unlock(&file->mut);
1268
1269 mutex_lock(&mut);
1270 idr_remove(&ctx_idr, ctx->id);
1271 mutex_unlock(&mut);
1272
1273 ucma_free_ctx(ctx);
1274 mutex_lock(&file->mut);
1275 }
1276 mutex_unlock(&file->mut);
1277 kfree(file);
1278 return 0;
1279 }
1280
1281 static const struct file_operations ucma_fops = {
1282 .owner = THIS_MODULE,
1283 .open = ucma_open,
1284 .release = ucma_close,
1285 .write = ucma_write,
1286 .poll = ucma_poll,
1287 .llseek = no_llseek,
1288 };
1289
1290 static struct miscdevice ucma_misc = {
1291 .minor = MISC_DYNAMIC_MINOR,
1292 .name = "rdma_cm",
1293 .fops = &ucma_fops,
1294 };
1295
1296 static ssize_t show_abi_version(struct device *dev,
1297 struct device_attribute *attr,
1298 char *buf)
1299 {
1300 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1301 }
1302 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1303
1304 static int __init ucma_init(void)
1305 {
1306 int ret;
1307
1308 ret = misc_register(&ucma_misc);
1309 if (ret)
1310 return ret;
1311
1312 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1313 if (ret) {
1314 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1315 goto err;
1316 }
1317 return 0;
1318 err:
1319 misc_deregister(&ucma_misc);
1320 return ret;
1321 }
1322
1323 static void __exit ucma_cleanup(void)
1324 {
1325 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1326 misc_deregister(&ucma_misc);
1327 idr_destroy(&ctx_idr);
1328 }
1329
1330 module_init(ucma_init);
1331 module_exit(ucma_cleanup);