?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: fix put_user() for gcc-8
[linux/fpc-iii.git] / drivers / infiniband / core / ucma.c
blob795938edce3fce80e42623574db64937a506dc9b
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 #include <linux/sysctl.h>
44 #include <linux/module.h>
45 #include <linux/nsproxy.h>
46
47 #include <rdma/rdma_user_cm.h>
48 #include <rdma/ib_marshall.h>
49 #include <rdma/rdma_cm.h>
50 #include <rdma/rdma_cm_ib.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib.h>
53
54 MODULE_AUTHOR("Sean Hefty");
55 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
56 MODULE_LICENSE("Dual BSD/GPL");
57
58 static unsigned int max_backlog = 1024;
59
60 static struct ctl_table_header *ucma_ctl_table_hdr;
61 static struct ctl_table ucma_ctl_table[] = {
62 {
63 .procname = "max_backlog",
64 .data = &max_backlog,
65 .maxlen = sizeof max_backlog,
66 .mode = 0644,
67 .proc_handler = proc_dointvec,
68 },
69 { }
70 };
71
72 struct ucma_file {
73 struct mutex mut;
74 struct file *filp;
75 struct list_head ctx_list;
76 struct list_head event_list;
77 wait_queue_head_t poll_wait;
78 struct workqueue_struct *close_wq;
79 };
80
81 struct ucma_context {
82 int id;
83 struct completion comp;
84 atomic_t ref;
85 int events_reported;
86 int backlog;
87
88 struct ucma_file *file;
89 struct rdma_cm_id *cm_id;
90 u64 uid;
91
92 struct list_head list;
93 struct list_head mc_list;
94 /* mark that device is in process of destroying the internal HW
95 * resources, protected by the global mut
96 */
97 int closing;
98 /* sync between removal event and id destroy, protected by file mut */
99 int destroying;
100 struct work_struct close_work;
101 };
102
103 struct ucma_multicast {
104 struct ucma_context *ctx;
105 int id;
106 int events_reported;
107
108 u64 uid;
109 struct list_head list;
110 struct sockaddr_storage addr;
111 };
112
113 struct ucma_event {
114 struct ucma_context *ctx;
115 struct ucma_multicast *mc;
116 struct list_head list;
117 struct rdma_cm_id *cm_id;
118 struct rdma_ucm_event_resp resp;
119 struct work_struct close_work;
120 };
121
122 static DEFINE_MUTEX(mut);
123 static DEFINE_IDR(ctx_idr);
124 static DEFINE_IDR(multicast_idr);
125
126 static inline struct ucma_context *_ucma_find_context(int id,
127 struct ucma_file *file)
128 {
129 struct ucma_context *ctx;
130
131 ctx = idr_find(&ctx_idr, id);
132 if (!ctx)
133 ctx = ERR_PTR(-ENOENT);
134 else if (ctx->file != file || !ctx->cm_id)
135 ctx = ERR_PTR(-EINVAL);
136 return ctx;
137 }
138
139 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
140 {
141 struct ucma_context *ctx;
142
143 mutex_lock(&mut);
144 ctx = _ucma_find_context(id, file);
145 if (!IS_ERR(ctx)) {
146 if (ctx->closing)
147 ctx = ERR_PTR(-EIO);
148 else
149 atomic_inc(&ctx->ref);
150 }
151 mutex_unlock(&mut);
152 return ctx;
153 }
154
155 static void ucma_put_ctx(struct ucma_context *ctx)
156 {
157 if (atomic_dec_and_test(&ctx->ref))
158 complete(&ctx->comp);
159 }
160
161 static void ucma_close_event_id(struct work_struct *work)
162 {
163 struct ucma_event *uevent_close = container_of(work, struct ucma_event, close_work);
164
165 rdma_destroy_id(uevent_close->cm_id);
166 kfree(uevent_close);
167 }
168
169 static void ucma_close_id(struct work_struct *work)
170 {
171 struct ucma_context *ctx = container_of(work, struct ucma_context, close_work);
172
173 /* once all inflight tasks are finished, we close all underlying
174 * resources. The context is still alive till its explicit destryoing
175 * by its creator.
176 */
177 ucma_put_ctx(ctx);
178 wait_for_completion(&ctx->comp);
179 /* No new events will be generated after destroying the id. */
180 rdma_destroy_id(ctx->cm_id);
181 }
182
183 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
184 {
185 struct ucma_context *ctx;
186
187 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
188 if (!ctx)
189 return NULL;
190
191 INIT_WORK(&ctx->close_work, ucma_close_id);
192 atomic_set(&ctx->ref, 1);
193 init_completion(&ctx->comp);
194 INIT_LIST_HEAD(&ctx->mc_list);
195 ctx->file = file;
196
197 mutex_lock(&mut);
198 ctx->id = idr_alloc(&ctx_idr, ctx, 0, 0, GFP_KERNEL);
199 mutex_unlock(&mut);
200 if (ctx->id < 0)
201 goto error;
202
203 list_add_tail(&ctx->list, &file->ctx_list);
204 return ctx;
205
206 error:
207 kfree(ctx);
208 return NULL;
209 }
210
211 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
212 {
213 struct ucma_multicast *mc;
214
215 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
216 if (!mc)
217 return NULL;
218
219 mutex_lock(&mut);
220 mc->id = idr_alloc(&multicast_idr, mc, 0, 0, GFP_KERNEL);
221 mutex_unlock(&mut);
222 if (mc->id < 0)
223 goto error;
224
225 mc->ctx = ctx;
226 list_add_tail(&mc->list, &ctx->mc_list);
227 return mc;
228
229 error:
230 kfree(mc);
231 return NULL;
232 }
233
234 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
235 struct rdma_conn_param *src)
236 {
237 if (src->private_data_len)
238 memcpy(dst->private_data, src->private_data,
239 src->private_data_len);
240 dst->private_data_len = src->private_data_len;
241 dst->responder_resources =src->responder_resources;
242 dst->initiator_depth = src->initiator_depth;
243 dst->flow_control = src->flow_control;
244 dst->retry_count = src->retry_count;
245 dst->rnr_retry_count = src->rnr_retry_count;
246 dst->srq = src->srq;
247 dst->qp_num = src->qp_num;
248 }
249
250 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
251 struct rdma_ud_param *src)
252 {
253 if (src->private_data_len)
254 memcpy(dst->private_data, src->private_data,
255 src->private_data_len);
256 dst->private_data_len = src->private_data_len;
257 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
258 dst->qp_num = src->qp_num;
259 dst->qkey = src->qkey;
260 }
261
262 static void ucma_set_event_context(struct ucma_context *ctx,
263 struct rdma_cm_event *event,
264 struct ucma_event *uevent)
265 {
266 uevent->ctx = ctx;
267 switch (event->event) {
268 case RDMA_CM_EVENT_MULTICAST_JOIN:
269 case RDMA_CM_EVENT_MULTICAST_ERROR:
270 uevent->mc = (struct ucma_multicast *)
271 event->param.ud.private_data;
272 uevent->resp.uid = uevent->mc->uid;
273 uevent->resp.id = uevent->mc->id;
274 break;
275 default:
276 uevent->resp.uid = ctx->uid;
277 uevent->resp.id = ctx->id;
278 break;
279 }
280 }
281
282 /* Called with file->mut locked for the relevant context. */
283 static void ucma_removal_event_handler(struct rdma_cm_id *cm_id)
284 {
285 struct ucma_context *ctx = cm_id->context;
286 struct ucma_event *con_req_eve;
287 int event_found = 0;
288
289 if (ctx->destroying)
290 return;
291
292 /* only if context is pointing to cm_id that it owns it and can be
293 * queued to be closed, otherwise that cm_id is an inflight one that
294 * is part of that context event list pending to be detached and
295 * reattached to its new context as part of ucma_get_event,
296 * handled separately below.
297 */
298 if (ctx->cm_id == cm_id) {
299 mutex_lock(&mut);
300 ctx->closing = 1;
301 mutex_unlock(&mut);
302 queue_work(ctx->file->close_wq, &ctx->close_work);
303 return;
304 }
305
306 list_for_each_entry(con_req_eve, &ctx->file->event_list, list) {
307 if (con_req_eve->cm_id == cm_id &&
308 con_req_eve->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
309 list_del(&con_req_eve->list);
310 INIT_WORK(&con_req_eve->close_work, ucma_close_event_id);
311 queue_work(ctx->file->close_wq, &con_req_eve->close_work);
312 event_found = 1;
313 break;
314 }
315 }
316 if (!event_found)
317 printk(KERN_ERR "ucma_removal_event_handler: warning: connect request event wasn't found\n");
318 }
319
320 static int ucma_event_handler(struct rdma_cm_id *cm_id,
321 struct rdma_cm_event *event)
322 {
323 struct ucma_event *uevent;
324 struct ucma_context *ctx = cm_id->context;
325 int ret = 0;
326
327 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
328 if (!uevent)
329 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
330
331 mutex_lock(&ctx->file->mut);
332 uevent->cm_id = cm_id;
333 ucma_set_event_context(ctx, event, uevent);
334 uevent->resp.event = event->event;
335 uevent->resp.status = event->status;
336 if (cm_id->qp_type == IB_QPT_UD)
337 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
338 else
339 ucma_copy_conn_event(&uevent->resp.param.conn,
340 &event->param.conn);
341
342 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
343 if (!ctx->backlog) {
344 ret = -ENOMEM;
345 kfree(uevent);
346 goto out;
347 }
348 ctx->backlog--;
349 } else if (!ctx->uid || ctx->cm_id != cm_id) {
350 /*
351 * We ignore events for new connections until userspace has set
352 * their context. This can only happen if an error occurs on a
353 * new connection before the user accepts it. This is okay,
354 * since the accept will just fail later. However, we do need
355 * to release the underlying HW resources in case of a device
356 * removal event.
357 */
358 if (event->event == RDMA_CM_EVENT_DEVICE_REMOVAL)
359 ucma_removal_event_handler(cm_id);
360
361 kfree(uevent);
362 goto out;
363 }
364
365 list_add_tail(&uevent->list, &ctx->file->event_list);
366 wake_up_interruptible(&ctx->file->poll_wait);
367 if (event->event == RDMA_CM_EVENT_DEVICE_REMOVAL)
368 ucma_removal_event_handler(cm_id);
369 out:
370 mutex_unlock(&ctx->file->mut);
371 return ret;
372 }
373
374 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
375 int in_len, int out_len)
376 {
377 struct ucma_context *ctx;
378 struct rdma_ucm_get_event cmd;
379 struct ucma_event *uevent;
380 int ret = 0;
381
382 if (out_len < sizeof uevent->resp)
383 return -ENOSPC;
384
385 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
386 return -EFAULT;
387
388 mutex_lock(&file->mut);
389 while (list_empty(&file->event_list)) {
390 mutex_unlock(&file->mut);
391
392 if (file->filp->f_flags & O_NONBLOCK)
393 return -EAGAIN;
394
395 if (wait_event_interruptible(file->poll_wait,
396 !list_empty(&file->event_list)))
397 return -ERESTARTSYS;
398
399 mutex_lock(&file->mut);
400 }
401
402 uevent = list_entry(file->event_list.next, struct ucma_event, list);
403
404 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
405 ctx = ucma_alloc_ctx(file);
406 if (!ctx) {
407 ret = -ENOMEM;
408 goto done;
409 }
410 uevent->ctx->backlog++;
411 ctx->cm_id = uevent->cm_id;
412 ctx->cm_id->context = ctx;
413 uevent->resp.id = ctx->id;
414 }
415
416 if (copy_to_user((void __user *)(unsigned long)cmd.response,
417 &uevent->resp, sizeof uevent->resp)) {
418 ret = -EFAULT;
419 goto done;
420 }
421
422 list_del(&uevent->list);
423 uevent->ctx->events_reported++;
424 if (uevent->mc)
425 uevent->mc->events_reported++;
426 kfree(uevent);
427 done:
428 mutex_unlock(&file->mut);
429 return ret;
430 }
431
432 static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_type)
433 {
434 switch (cmd->ps) {
435 case RDMA_PS_TCP:
436 *qp_type = IB_QPT_RC;
437 return 0;
438 case RDMA_PS_UDP:
439 case RDMA_PS_IPOIB:
440 *qp_type = IB_QPT_UD;
441 return 0;
442 case RDMA_PS_IB:
443 *qp_type = cmd->qp_type;
444 return 0;
445 default:
446 return -EINVAL;
447 }
448 }
449
450 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,
451 int in_len, int out_len)
452 {
453 struct rdma_ucm_create_id cmd;
454 struct rdma_ucm_create_id_resp resp;
455 struct ucma_context *ctx;
456 struct rdma_cm_id *cm_id;
457 enum ib_qp_type qp_type;
458 int ret;
459
460 if (out_len < sizeof(resp))
461 return -ENOSPC;
462
463 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
464 return -EFAULT;
465
466 ret = ucma_get_qp_type(&cmd, &qp_type);
467 if (ret)
468 return ret;
469
470 mutex_lock(&file->mut);
471 ctx = ucma_alloc_ctx(file);
472 mutex_unlock(&file->mut);
473 if (!ctx)
474 return -ENOMEM;
475
476 ctx->uid = cmd.uid;
477 cm_id = rdma_create_id(current->nsproxy->net_ns,
478 ucma_event_handler, ctx, cmd.ps, qp_type);
479 if (IS_ERR(cm_id)) {
480 ret = PTR_ERR(cm_id);
481 goto err1;
482 }
483
484 resp.id = ctx->id;
485 if (copy_to_user((void __user *)(unsigned long)cmd.response,
486 &resp, sizeof(resp))) {
487 ret = -EFAULT;
488 goto err2;
489 }
490
491 ctx->cm_id = cm_id;
492 return 0;
493
494 err2:
495 rdma_destroy_id(cm_id);
496 err1:
497 mutex_lock(&mut);
498 idr_remove(&ctx_idr, ctx->id);
499 mutex_unlock(&mut);
500 mutex_lock(&file->mut);
501 list_del(&ctx->list);
502 mutex_unlock(&file->mut);
503 kfree(ctx);
504 return ret;
505 }
506
507 static void ucma_cleanup_multicast(struct ucma_context *ctx)
508 {
509 struct ucma_multicast *mc, *tmp;
510
511 mutex_lock(&mut);
512 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
513 list_del(&mc->list);
514 idr_remove(&multicast_idr, mc->id);
515 kfree(mc);
516 }
517 mutex_unlock(&mut);
518 }
519
520 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
521 {
522 struct ucma_event *uevent, *tmp;
523
524 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
525 if (uevent->mc != mc)
526 continue;
527
528 list_del(&uevent->list);
529 kfree(uevent);
530 }
531 }
532
533 /*
534 * ucma_free_ctx is called after the underlying rdma CM-ID is destroyed. At
535 * this point, no new events will be reported from the hardware. However, we
536 * still need to cleanup the UCMA context for this ID. Specifically, there
537 * might be events that have not yet been consumed by the user space software.
538 * These might include pending connect requests which we have not completed
539 * processing. We cannot call rdma_destroy_id while holding the lock of the
540 * context (file->mut), as it might cause a deadlock. We therefore extract all
541 * relevant events from the context pending events list while holding the
542 * mutex. After that we release them as needed.
543 */
544 static int ucma_free_ctx(struct ucma_context *ctx)
545 {
546 int events_reported;
547 struct ucma_event *uevent, *tmp;
548 LIST_HEAD(list);
549
550
551 ucma_cleanup_multicast(ctx);
552
553 /* Cleanup events not yet reported to the user. */
554 mutex_lock(&ctx->file->mut);
555 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
556 if (uevent->ctx == ctx)
557 list_move_tail(&uevent->list, &list);
558 }
559 list_del(&ctx->list);
560 mutex_unlock(&ctx->file->mut);
561
562 list_for_each_entry_safe(uevent, tmp, &list, list) {
563 list_del(&uevent->list);
564 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
565 rdma_destroy_id(uevent->cm_id);
566 kfree(uevent);
567 }
568
569 events_reported = ctx->events_reported;
570 kfree(ctx);
571 return events_reported;
572 }
573
574 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
575 int in_len, int out_len)
576 {
577 struct rdma_ucm_destroy_id cmd;
578 struct rdma_ucm_destroy_id_resp resp;
579 struct ucma_context *ctx;
580 int ret = 0;
581
582 if (out_len < sizeof(resp))
583 return -ENOSPC;
584
585 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
586 return -EFAULT;
587
588 mutex_lock(&mut);
589 ctx = _ucma_find_context(cmd.id, file);
590 if (!IS_ERR(ctx))
591 idr_remove(&ctx_idr, ctx->id);
592 mutex_unlock(&mut);
593
594 if (IS_ERR(ctx))
595 return PTR_ERR(ctx);
596
597 mutex_lock(&ctx->file->mut);
598 ctx->destroying = 1;
599 mutex_unlock(&ctx->file->mut);
600
601 flush_workqueue(ctx->file->close_wq);
602 /* At this point it's guaranteed that there is no inflight
603 * closing task */
604 mutex_lock(&mut);
605 if (!ctx->closing) {
606 mutex_unlock(&mut);
607 ucma_put_ctx(ctx);
608 wait_for_completion(&ctx->comp);
609 rdma_destroy_id(ctx->cm_id);
610 } else {
611 mutex_unlock(&mut);
612 }
613
614 resp.events_reported = ucma_free_ctx(ctx);
615 if (copy_to_user((void __user *)(unsigned long)cmd.response,
616 &resp, sizeof(resp)))
617 ret = -EFAULT;
618
619 return ret;
620 }
621
622 static ssize_t ucma_bind_ip(struct ucma_file *file, const char __user *inbuf,
623 int in_len, int out_len)
624 {
625 struct rdma_ucm_bind_ip cmd;
626 struct ucma_context *ctx;
627 int ret;
628
629 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
630 return -EFAULT;
631
632 if (!rdma_addr_size_in6(&cmd.addr))
633 return -EINVAL;
634
635 ctx = ucma_get_ctx(file, cmd.id);
636 if (IS_ERR(ctx))
637 return PTR_ERR(ctx);
638
639 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
640 ucma_put_ctx(ctx);
641 return ret;
642 }
643
644 static ssize_t ucma_bind(struct ucma_file *file, const char __user *inbuf,
645 int in_len, int out_len)
646 {
647 struct rdma_ucm_bind cmd;
648 struct ucma_context *ctx;
649 int ret;
650
651 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
652 return -EFAULT;
653
654 if (cmd.reserved || !cmd.addr_size ||
655 cmd.addr_size != rdma_addr_size_kss(&cmd.addr))
656 return -EINVAL;
657
658 ctx = ucma_get_ctx(file, cmd.id);
659 if (IS_ERR(ctx))
660 return PTR_ERR(ctx);
661
662 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
663 ucma_put_ctx(ctx);
664 return ret;
665 }
666
667 static ssize_t ucma_resolve_ip(struct ucma_file *file,
668 const char __user *inbuf,
669 int in_len, int out_len)
670 {
671 struct rdma_ucm_resolve_ip cmd;
672 struct ucma_context *ctx;
673 int ret;
674
675 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
676 return -EFAULT;
677
678 if ((cmd.src_addr.sin6_family && !rdma_addr_size_in6(&cmd.src_addr)) ||
679 !rdma_addr_size_in6(&cmd.dst_addr))
680 return -EINVAL;
681
682 ctx = ucma_get_ctx(file, cmd.id);
683 if (IS_ERR(ctx))
684 return PTR_ERR(ctx);
685
686 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
687 (struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms);
688 ucma_put_ctx(ctx);
689 return ret;
690 }
691
692 static ssize_t ucma_resolve_addr(struct ucma_file *file,
693 const char __user *inbuf,
694 int in_len, int out_len)
695 {
696 struct rdma_ucm_resolve_addr cmd;
697 struct ucma_context *ctx;
698 int ret;
699
700 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
701 return -EFAULT;
702
703 if (cmd.reserved ||
704 (cmd.src_size && (cmd.src_size != rdma_addr_size_kss(&cmd.src_addr))) ||
705 !cmd.dst_size || (cmd.dst_size != rdma_addr_size_kss(&cmd.dst_addr)))
706 return -EINVAL;
707
708 ctx = ucma_get_ctx(file, cmd.id);
709 if (IS_ERR(ctx))
710 return PTR_ERR(ctx);
711
712 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
713 (struct sockaddr *) &cmd.dst_addr, cmd.timeout_ms);
714 ucma_put_ctx(ctx);
715 return ret;
716 }
717
718 static ssize_t ucma_resolve_route(struct ucma_file *file,
719 const char __user *inbuf,
720 int in_len, int out_len)
721 {
722 struct rdma_ucm_resolve_route cmd;
723 struct ucma_context *ctx;
724 int ret;
725
726 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
727 return -EFAULT;
728
729 ctx = ucma_get_ctx(file, cmd.id);
730 if (IS_ERR(ctx))
731 return PTR_ERR(ctx);
732
733 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
734 ucma_put_ctx(ctx);
735 return ret;
736 }
737
738 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
739 struct rdma_route *route)
740 {
741 struct rdma_dev_addr *dev_addr;
742
743 resp->num_paths = route->num_paths;
744 switch (route->num_paths) {
745 case 0:
746 dev_addr = &route->addr.dev_addr;
747 rdma_addr_get_dgid(dev_addr,
748 (union ib_gid *) &resp->ib_route[0].dgid);
749 rdma_addr_get_sgid(dev_addr,
750 (union ib_gid *) &resp->ib_route[0].sgid);
751 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
752 break;
753 case 2:
754 ib_copy_path_rec_to_user(&resp->ib_route[1],
755 &route->path_rec[1]);
756 /* fall through */
757 case 1:
758 ib_copy_path_rec_to_user(&resp->ib_route[0],
759 &route->path_rec[0]);
760 break;
761 default:
762 break;
763 }
764 }
765
766 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
767 struct rdma_route *route)
768 {
769
770 resp->num_paths = route->num_paths;
771 switch (route->num_paths) {
772 case 0:
773 rdma_ip2gid((struct sockaddr *)&route->addr.dst_addr,
774 (union ib_gid *)&resp->ib_route[0].dgid);
775 rdma_ip2gid((struct sockaddr *)&route->addr.src_addr,
776 (union ib_gid *)&resp->ib_route[0].sgid);
777 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
778 break;
779 case 2:
780 ib_copy_path_rec_to_user(&resp->ib_route[1],
781 &route->path_rec[1]);
782 /* fall through */
783 case 1:
784 ib_copy_path_rec_to_user(&resp->ib_route[0],
785 &route->path_rec[0]);
786 break;
787 default:
788 break;
789 }
790 }
791
792 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
793 struct rdma_route *route)
794 {
795 struct rdma_dev_addr *dev_addr;
796
797 dev_addr = &route->addr.dev_addr;
798 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
799 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
800 }
801
802 static ssize_t ucma_query_route(struct ucma_file *file,
803 const char __user *inbuf,
804 int in_len, int out_len)
805 {
806 struct rdma_ucm_query cmd;
807 struct rdma_ucm_query_route_resp resp;
808 struct ucma_context *ctx;
809 struct sockaddr *addr;
810 int ret = 0;
811
812 if (out_len < sizeof(resp))
813 return -ENOSPC;
814
815 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
816 return -EFAULT;
817
818 ctx = ucma_get_ctx(file, cmd.id);
819 if (IS_ERR(ctx))
820 return PTR_ERR(ctx);
821
822 memset(&resp, 0, sizeof resp);
823 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
824 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
825 sizeof(struct sockaddr_in) :
826 sizeof(struct sockaddr_in6));
827 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
828 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
829 sizeof(struct sockaddr_in) :
830 sizeof(struct sockaddr_in6));
831 if (!ctx->cm_id->device)
832 goto out;
833
834 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
835 resp.port_num = ctx->cm_id->port_num;
836
837 if (rdma_cap_ib_sa(ctx->cm_id->device, ctx->cm_id->port_num))
838 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
839 else if (rdma_protocol_roce(ctx->cm_id->device, ctx->cm_id->port_num))
840 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
841 else if (rdma_protocol_iwarp(ctx->cm_id->device, ctx->cm_id->port_num))
842 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
843
844 out:
845 if (copy_to_user((void __user *)(unsigned long)cmd.response,
846 &resp, sizeof(resp)))
847 ret = -EFAULT;
848
849 ucma_put_ctx(ctx);
850 return ret;
851 }
852
853 static void ucma_query_device_addr(struct rdma_cm_id *cm_id,
854 struct rdma_ucm_query_addr_resp *resp)
855 {
856 if (!cm_id->device)
857 return;
858
859 resp->node_guid = (__force __u64) cm_id->device->node_guid;
860 resp->port_num = cm_id->port_num;
861 resp->pkey = (__force __u16) cpu_to_be16(
862 ib_addr_get_pkey(&cm_id->route.addr.dev_addr));
863 }
864
865 static ssize_t ucma_query_addr(struct ucma_context *ctx,
866 void __user *response, int out_len)
867 {
868 struct rdma_ucm_query_addr_resp resp;
869 struct sockaddr *addr;
870 int ret = 0;
871
872 if (out_len < sizeof(resp))
873 return -ENOSPC;
874
875 memset(&resp, 0, sizeof resp);
876
877 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
878 resp.src_size = rdma_addr_size(addr);
879 memcpy(&resp.src_addr, addr, resp.src_size);
880
881 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
882 resp.dst_size = rdma_addr_size(addr);
883 memcpy(&resp.dst_addr, addr, resp.dst_size);
884
885 ucma_query_device_addr(ctx->cm_id, &resp);
886
887 if (copy_to_user(response, &resp, sizeof(resp)))
888 ret = -EFAULT;
889
890 return ret;
891 }
892
893 static ssize_t ucma_query_path(struct ucma_context *ctx,
894 void __user *response, int out_len)
895 {
896 struct rdma_ucm_query_path_resp *resp;
897 int i, ret = 0;
898
899 if (out_len < sizeof(*resp))
900 return -ENOSPC;
901
902 resp = kzalloc(out_len, GFP_KERNEL);
903 if (!resp)
904 return -ENOMEM;
905
906 resp->num_paths = ctx->cm_id->route.num_paths;
907 for (i = 0, out_len -= sizeof(*resp);
908 i < resp->num_paths && out_len > sizeof(struct ib_path_rec_data);
909 i++, out_len -= sizeof(struct ib_path_rec_data)) {
910
911 resp->path_data[i].flags = IB_PATH_GMP | IB_PATH_PRIMARY |
912 IB_PATH_BIDIRECTIONAL;
913 ib_sa_pack_path(&ctx->cm_id->route.path_rec[i],
914 &resp->path_data[i].path_rec);
915 }
916
917 if (copy_to_user(response, resp,
918 sizeof(*resp) + (i * sizeof(struct ib_path_rec_data))))
919 ret = -EFAULT;
920
921 kfree(resp);
922 return ret;
923 }
924
925 static ssize_t ucma_query_gid(struct ucma_context *ctx,
926 void __user *response, int out_len)
927 {
928 struct rdma_ucm_query_addr_resp resp;
929 struct sockaddr_ib *addr;
930 int ret = 0;
931
932 if (out_len < sizeof(resp))
933 return -ENOSPC;
934
935 memset(&resp, 0, sizeof resp);
936
937 ucma_query_device_addr(ctx->cm_id, &resp);
938
939 addr = (struct sockaddr_ib *) &resp.src_addr;
940 resp.src_size = sizeof(*addr);
941 if (ctx->cm_id->route.addr.src_addr.ss_family == AF_IB) {
942 memcpy(addr, &ctx->cm_id->route.addr.src_addr, resp.src_size);
943 } else {
944 addr->sib_family = AF_IB;
945 addr->sib_pkey = (__force __be16) resp.pkey;
946 rdma_addr_get_sgid(&ctx->cm_id->route.addr.dev_addr,
947 (union ib_gid *) &addr->sib_addr);
948 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *)
949 &ctx->cm_id->route.addr.src_addr);
950 }
951
952 addr = (struct sockaddr_ib *) &resp.dst_addr;
953 resp.dst_size = sizeof(*addr);
954 if (ctx->cm_id->route.addr.dst_addr.ss_family == AF_IB) {
955 memcpy(addr, &ctx->cm_id->route.addr.dst_addr, resp.dst_size);
956 } else {
957 addr->sib_family = AF_IB;
958 addr->sib_pkey = (__force __be16) resp.pkey;
959 rdma_addr_get_dgid(&ctx->cm_id->route.addr.dev_addr,
960 (union ib_gid *) &addr->sib_addr);
961 addr->sib_sid = rdma_get_service_id(ctx->cm_id, (struct sockaddr *)
962 &ctx->cm_id->route.addr.dst_addr);
963 }
964
965 if (copy_to_user(response, &resp, sizeof(resp)))
966 ret = -EFAULT;
967
968 return ret;
969 }
970
971 static ssize_t ucma_query(struct ucma_file *file,
972 const char __user *inbuf,
973 int in_len, int out_len)
974 {
975 struct rdma_ucm_query cmd;
976 struct ucma_context *ctx;
977 void __user *response;
978 int ret;
979
980 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
981 return -EFAULT;
982
983 response = (void __user *)(unsigned long) cmd.response;
984 ctx = ucma_get_ctx(file, cmd.id);
985 if (IS_ERR(ctx))
986 return PTR_ERR(ctx);
987
988 switch (cmd.option) {
989 case RDMA_USER_CM_QUERY_ADDR:
990 ret = ucma_query_addr(ctx, response, out_len);
991 break;
992 case RDMA_USER_CM_QUERY_PATH:
993 ret = ucma_query_path(ctx, response, out_len);
994 break;
995 case RDMA_USER_CM_QUERY_GID:
996 ret = ucma_query_gid(ctx, response, out_len);
997 break;
998 default:
999 ret = -ENOSYS;
1000 break;
1001 }
1002
1003 ucma_put_ctx(ctx);
1004 return ret;
1005 }
1006
1007 static void ucma_copy_conn_param(struct rdma_cm_id *id,
1008 struct rdma_conn_param *dst,
1009 struct rdma_ucm_conn_param *src)
1010 {
1011 dst->private_data = src->private_data;
1012 dst->private_data_len = src->private_data_len;
1013 dst->responder_resources =src->responder_resources;
1014 dst->initiator_depth = src->initiator_depth;
1015 dst->flow_control = src->flow_control;
1016 dst->retry_count = src->retry_count;
1017 dst->rnr_retry_count = src->rnr_retry_count;
1018 dst->srq = src->srq;
1019 dst->qp_num = src->qp_num;
1020 dst->qkey = (id->route.addr.src_addr.ss_family == AF_IB) ? src->qkey : 0;
1021 }
1022
1023 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
1024 int in_len, int out_len)
1025 {
1026 struct rdma_ucm_connect cmd;
1027 struct rdma_conn_param conn_param;
1028 struct ucma_context *ctx;
1029 int ret;
1030
1031 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1032 return -EFAULT;
1033
1034 if (!cmd.conn_param.valid)
1035 return -EINVAL;
1036
1037 ctx = ucma_get_ctx(file, cmd.id);
1038 if (IS_ERR(ctx))
1039 return PTR_ERR(ctx);
1040
1041 ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
1042 ret = rdma_connect(ctx->cm_id, &conn_param);
1043 ucma_put_ctx(ctx);
1044 return ret;
1045 }
1046
1047 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
1048 int in_len, int out_len)
1049 {
1050 struct rdma_ucm_listen cmd;
1051 struct ucma_context *ctx;
1052 int ret;
1053
1054 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1055 return -EFAULT;
1056
1057 ctx = ucma_get_ctx(file, cmd.id);
1058 if (IS_ERR(ctx))
1059 return PTR_ERR(ctx);
1060
1061 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
1062 cmd.backlog : max_backlog;
1063 ret = rdma_listen(ctx->cm_id, ctx->backlog);
1064 ucma_put_ctx(ctx);
1065 return ret;
1066 }
1067
1068 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
1069 int in_len, int out_len)
1070 {
1071 struct rdma_ucm_accept cmd;
1072 struct rdma_conn_param conn_param;
1073 struct ucma_context *ctx;
1074 int ret;
1075
1076 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1077 return -EFAULT;
1078
1079 ctx = ucma_get_ctx(file, cmd.id);
1080 if (IS_ERR(ctx))
1081 return PTR_ERR(ctx);
1082
1083 if (cmd.conn_param.valid) {
1084 ucma_copy_conn_param(ctx->cm_id, &conn_param, &cmd.conn_param);
1085 mutex_lock(&file->mut);
1086 ret = rdma_accept(ctx->cm_id, &conn_param);
1087 if (!ret)
1088 ctx->uid = cmd.uid;
1089 mutex_unlock(&file->mut);
1090 } else
1091 ret = rdma_accept(ctx->cm_id, NULL);
1092
1093 ucma_put_ctx(ctx);
1094 return ret;
1095 }
1096
1097 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
1098 int in_len, int out_len)
1099 {
1100 struct rdma_ucm_reject cmd;
1101 struct ucma_context *ctx;
1102 int ret;
1103
1104 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1105 return -EFAULT;
1106
1107 ctx = ucma_get_ctx(file, cmd.id);
1108 if (IS_ERR(ctx))
1109 return PTR_ERR(ctx);
1110
1111 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
1112 ucma_put_ctx(ctx);
1113 return ret;
1114 }
1115
1116 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
1117 int in_len, int out_len)
1118 {
1119 struct rdma_ucm_disconnect cmd;
1120 struct ucma_context *ctx;
1121 int ret;
1122
1123 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1124 return -EFAULT;
1125
1126 ctx = ucma_get_ctx(file, cmd.id);
1127 if (IS_ERR(ctx))
1128 return PTR_ERR(ctx);
1129
1130 ret = rdma_disconnect(ctx->cm_id);
1131 ucma_put_ctx(ctx);
1132 return ret;
1133 }
1134
1135 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
1136 const char __user *inbuf,
1137 int in_len, int out_len)
1138 {
1139 struct rdma_ucm_init_qp_attr cmd;
1140 struct ib_uverbs_qp_attr resp;
1141 struct ucma_context *ctx;
1142 struct ib_qp_attr qp_attr;
1143 int ret;
1144
1145 if (out_len < sizeof(resp))
1146 return -ENOSPC;
1147
1148 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1149 return -EFAULT;
1150
1151 if (cmd.qp_state > IB_QPS_ERR)
1152 return -EINVAL;
1153
1154 ctx = ucma_get_ctx(file, cmd.id);
1155 if (IS_ERR(ctx))
1156 return PTR_ERR(ctx);
1157
1158 if (!ctx->cm_id->device) {
1159 ret = -EINVAL;
1160 goto out;
1161 }
1162
1163 resp.qp_attr_mask = 0;
1164 memset(&qp_attr, 0, sizeof qp_attr);
1165 qp_attr.qp_state = cmd.qp_state;
1166 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
1167 if (ret)
1168 goto out;
1169
1170 ib_copy_qp_attr_to_user(&resp, &qp_attr);
1171 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1172 &resp, sizeof(resp)))
1173 ret = -EFAULT;
1174
1175 out:
1176 ucma_put_ctx(ctx);
1177 return ret;
1178 }
1179
1180 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
1181 void *optval, size_t optlen)
1182 {
1183 int ret = 0;
1184
1185 switch (optname) {
1186 case RDMA_OPTION_ID_TOS:
1187 if (optlen != sizeof(u8)) {
1188 ret = -EINVAL;
1189 break;
1190 }
1191 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
1192 break;
1193 case RDMA_OPTION_ID_REUSEADDR:
1194 if (optlen != sizeof(int)) {
1195 ret = -EINVAL;
1196 break;
1197 }
1198 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
1199 break;
1200 case RDMA_OPTION_ID_AFONLY:
1201 if (optlen != sizeof(int)) {
1202 ret = -EINVAL;
1203 break;
1204 }
1205 ret = rdma_set_afonly(ctx->cm_id, *((int *) optval) ? 1 : 0);
1206 break;
1207 default:
1208 ret = -ENOSYS;
1209 }
1210
1211 return ret;
1212 }
1213
1214 static int ucma_set_ib_path(struct ucma_context *ctx,
1215 struct ib_path_rec_data *path_data, size_t optlen)
1216 {
1217 struct ib_sa_path_rec sa_path;
1218 struct rdma_cm_event event;
1219 int ret;
1220
1221 if (optlen % sizeof(*path_data))
1222 return -EINVAL;
1223
1224 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
1225 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
1226 IB_PATH_BIDIRECTIONAL))
1227 break;
1228 }
1229
1230 if (!optlen)
1231 return -EINVAL;
1232
1233 if (!ctx->cm_id->device)
1234 return -EINVAL;
1235
1236 memset(&sa_path, 0, sizeof(sa_path));
1237
1238 ib_sa_unpack_path(path_data->path_rec, &sa_path);
1239 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
1240 if (ret)
1241 return ret;
1242
1243 memset(&event, 0, sizeof event);
1244 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1245 return ucma_event_handler(ctx->cm_id, &event);
1246 }
1247
1248 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
1249 void *optval, size_t optlen)
1250 {
1251 int ret;
1252
1253 switch (optname) {
1254 case RDMA_OPTION_IB_PATH:
1255 ret = ucma_set_ib_path(ctx, optval, optlen);
1256 break;
1257 default:
1258 ret = -ENOSYS;
1259 }
1260
1261 return ret;
1262 }
1263
1264 static int ucma_set_option_level(struct ucma_context *ctx, int level,
1265 int optname, void *optval, size_t optlen)
1266 {
1267 int ret;
1268
1269 switch (level) {
1270 case RDMA_OPTION_ID:
1271 ret = ucma_set_option_id(ctx, optname, optval, optlen);
1272 break;
1273 case RDMA_OPTION_IB:
1274 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
1275 break;
1276 default:
1277 ret = -ENOSYS;
1278 }
1279
1280 return ret;
1281 }
1282
1283 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
1284 int in_len, int out_len)
1285 {
1286 struct rdma_ucm_set_option cmd;
1287 struct ucma_context *ctx;
1288 void *optval;
1289 int ret;
1290
1291 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1292 return -EFAULT;
1293
1294 ctx = ucma_get_ctx(file, cmd.id);
1295 if (IS_ERR(ctx))
1296 return PTR_ERR(ctx);
1297
1298 if (unlikely(cmd.optlen > KMALLOC_MAX_SIZE))
1299 return -EINVAL;
1300
1301 optval = memdup_user((void __user *) (unsigned long) cmd.optval,
1302 cmd.optlen);
1303 if (IS_ERR(optval)) {
1304 ret = PTR_ERR(optval);
1305 goto out;
1306 }
1307
1308 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1309 cmd.optlen);
1310 kfree(optval);
1311
1312 out:
1313 ucma_put_ctx(ctx);
1314 return ret;
1315 }
1316
1317 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1318 int in_len, int out_len)
1319 {
1320 struct rdma_ucm_notify cmd;
1321 struct ucma_context *ctx;
1322 int ret = -EINVAL;
1323
1324 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1325 return -EFAULT;
1326
1327 ctx = ucma_get_ctx(file, cmd.id);
1328 if (IS_ERR(ctx))
1329 return PTR_ERR(ctx);
1330
1331 if (ctx->cm_id->device)
1332 ret = rdma_notify(ctx->cm_id, (enum ib_event_type)cmd.event);
1333
1334 ucma_put_ctx(ctx);
1335 return ret;
1336 }
1337
1338 static ssize_t ucma_process_join(struct ucma_file *file,
1339 struct rdma_ucm_join_mcast *cmd, int out_len)
1340 {
1341 struct rdma_ucm_create_id_resp resp;
1342 struct ucma_context *ctx;
1343 struct ucma_multicast *mc;
1344 struct sockaddr *addr;
1345 int ret;
1346
1347 if (out_len < sizeof(resp))
1348 return -ENOSPC;
1349
1350 addr = (struct sockaddr *) &cmd->addr;
1351 if (cmd->reserved || (cmd->addr_size != rdma_addr_size(addr)))
1352 return -EINVAL;
1353
1354 ctx = ucma_get_ctx(file, cmd->id);
1355 if (IS_ERR(ctx))
1356 return PTR_ERR(ctx);
1357
1358 mutex_lock(&file->mut);
1359 mc = ucma_alloc_multicast(ctx);
1360 if (!mc) {
1361 ret = -ENOMEM;
1362 goto err1;
1363 }
1364
1365 mc->uid = cmd->uid;
1366 memcpy(&mc->addr, addr, cmd->addr_size);
1367 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1368 if (ret)
1369 goto err2;
1370
1371 resp.id = mc->id;
1372 if (copy_to_user((void __user *)(unsigned long) cmd->response,
1373 &resp, sizeof(resp))) {
1374 ret = -EFAULT;
1375 goto err3;
1376 }
1377
1378 mutex_unlock(&file->mut);
1379 ucma_put_ctx(ctx);
1380 return 0;
1381
1382 err3:
1383 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1384 ucma_cleanup_mc_events(mc);
1385 err2:
1386 mutex_lock(&mut);
1387 idr_remove(&multicast_idr, mc->id);
1388 mutex_unlock(&mut);
1389 list_del(&mc->list);
1390 kfree(mc);
1391 err1:
1392 mutex_unlock(&file->mut);
1393 ucma_put_ctx(ctx);
1394 return ret;
1395 }
1396
1397 static ssize_t ucma_join_ip_multicast(struct ucma_file *file,
1398 const char __user *inbuf,
1399 int in_len, int out_len)
1400 {
1401 struct rdma_ucm_join_ip_mcast cmd;
1402 struct rdma_ucm_join_mcast join_cmd;
1403
1404 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1405 return -EFAULT;
1406
1407 join_cmd.response = cmd.response;
1408 join_cmd.uid = cmd.uid;
1409 join_cmd.id = cmd.id;
1410 join_cmd.addr_size = rdma_addr_size_in6(&cmd.addr);
1411 if (!join_cmd.addr_size)
1412 return -EINVAL;
1413
1414 join_cmd.reserved = 0;
1415 memcpy(&join_cmd.addr, &cmd.addr, join_cmd.addr_size);
1416
1417 return ucma_process_join(file, &join_cmd, out_len);
1418 }
1419
1420 static ssize_t ucma_join_multicast(struct ucma_file *file,
1421 const char __user *inbuf,
1422 int in_len, int out_len)
1423 {
1424 struct rdma_ucm_join_mcast cmd;
1425
1426 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1427 return -EFAULT;
1428
1429 if (!rdma_addr_size_kss(&cmd.addr))
1430 return -EINVAL;
1431
1432 return ucma_process_join(file, &cmd, out_len);
1433 }
1434
1435 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1436 const char __user *inbuf,
1437 int in_len, int out_len)
1438 {
1439 struct rdma_ucm_destroy_id cmd;
1440 struct rdma_ucm_destroy_id_resp resp;
1441 struct ucma_multicast *mc;
1442 int ret = 0;
1443
1444 if (out_len < sizeof(resp))
1445 return -ENOSPC;
1446
1447 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1448 return -EFAULT;
1449
1450 mutex_lock(&mut);
1451 mc = idr_find(&multicast_idr, cmd.id);
1452 if (!mc)
1453 mc = ERR_PTR(-ENOENT);
1454 else if (mc->ctx->file != file)
1455 mc = ERR_PTR(-EINVAL);
1456 else if (!atomic_inc_not_zero(&mc->ctx->ref))
1457 mc = ERR_PTR(-ENXIO);
1458 else
1459 idr_remove(&multicast_idr, mc->id);
1460 mutex_unlock(&mut);
1461
1462 if (IS_ERR(mc)) {
1463 ret = PTR_ERR(mc);
1464 goto out;
1465 }
1466
1467 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1468 mutex_lock(&mc->ctx->file->mut);
1469 ucma_cleanup_mc_events(mc);
1470 list_del(&mc->list);
1471 mutex_unlock(&mc->ctx->file->mut);
1472
1473 ucma_put_ctx(mc->ctx);
1474 resp.events_reported = mc->events_reported;
1475 kfree(mc);
1476
1477 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1478 &resp, sizeof(resp)))
1479 ret = -EFAULT;
1480 out:
1481 return ret;
1482 }
1483
1484 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1485 {
1486 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1487 if (file1 < file2) {
1488 mutex_lock(&file1->mut);
1489 mutex_lock_nested(&file2->mut, SINGLE_DEPTH_NESTING);
1490 } else {
1491 mutex_lock(&file2->mut);
1492 mutex_lock_nested(&file1->mut, SINGLE_DEPTH_NESTING);
1493 }
1494 }
1495
1496 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1497 {
1498 if (file1 < file2) {
1499 mutex_unlock(&file2->mut);
1500 mutex_unlock(&file1->mut);
1501 } else {
1502 mutex_unlock(&file1->mut);
1503 mutex_unlock(&file2->mut);
1504 }
1505 }
1506
1507 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1508 {
1509 struct ucma_event *uevent, *tmp;
1510
1511 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1512 if (uevent->ctx == ctx)
1513 list_move_tail(&uevent->list, &file->event_list);
1514 }
1515
1516 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1517 const char __user *inbuf,
1518 int in_len, int out_len)
1519 {
1520 struct rdma_ucm_migrate_id cmd;
1521 struct rdma_ucm_migrate_resp resp;
1522 struct ucma_context *ctx;
1523 struct fd f;
1524 struct ucma_file *cur_file;
1525 int ret = 0;
1526
1527 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1528 return -EFAULT;
1529
1530 /* Get current fd to protect against it being closed */
1531 f = fdget(cmd.fd);
1532 if (!f.file)
1533 return -ENOENT;
1534
1535 /* Validate current fd and prevent destruction of id. */
1536 ctx = ucma_get_ctx(f.file->private_data, cmd.id);
1537 if (IS_ERR(ctx)) {
1538 ret = PTR_ERR(ctx);
1539 goto file_put;
1540 }
1541
1542 cur_file = ctx->file;
1543 if (cur_file == new_file) {
1544 resp.events_reported = ctx->events_reported;
1545 goto response;
1546 }
1547
1548 /*
1549 * Migrate events between fd's, maintaining order, and avoiding new
1550 * events being added before existing events.
1551 */
1552 ucma_lock_files(cur_file, new_file);
1553 mutex_lock(&mut);
1554
1555 list_move_tail(&ctx->list, &new_file->ctx_list);
1556 ucma_move_events(ctx, new_file);
1557 ctx->file = new_file;
1558 resp.events_reported = ctx->events_reported;
1559
1560 mutex_unlock(&mut);
1561 ucma_unlock_files(cur_file, new_file);
1562
1563 response:
1564 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1565 &resp, sizeof(resp)))
1566 ret = -EFAULT;
1567
1568 ucma_put_ctx(ctx);
1569 file_put:
1570 fdput(f);
1571 return ret;
1572 }
1573
1574 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1575 const char __user *inbuf,
1576 int in_len, int out_len) = {
1577 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1578 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1579 [RDMA_USER_CM_CMD_BIND_IP] = ucma_bind_ip,
1580 [RDMA_USER_CM_CMD_RESOLVE_IP] = ucma_resolve_ip,
1581 [RDMA_USER_CM_CMD_RESOLVE_ROUTE] = ucma_resolve_route,
1582 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1583 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1584 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1585 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1586 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1587 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1588 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1589 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1590 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1591 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1592 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1593 [RDMA_USER_CM_CMD_JOIN_IP_MCAST] = ucma_join_ip_multicast,
1594 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1595 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id,
1596 [RDMA_USER_CM_CMD_QUERY] = ucma_query,
1597 [RDMA_USER_CM_CMD_BIND] = ucma_bind,
1598 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1599 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast
1600 };
1601
1602 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1603 size_t len, loff_t *pos)
1604 {
1605 struct ucma_file *file = filp->private_data;
1606 struct rdma_ucm_cmd_hdr hdr;
1607 ssize_t ret;
1608
1609 if (WARN_ON_ONCE(!ib_safe_file_access(filp)))
1610 return -EACCES;
1611
1612 if (len < sizeof(hdr))
1613 return -EINVAL;
1614
1615 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1616 return -EFAULT;
1617
1618 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1619 return -EINVAL;
1620
1621 if (hdr.in + sizeof(hdr) > len)
1622 return -EINVAL;
1623
1624 if (!ucma_cmd_table[hdr.cmd])
1625 return -ENOSYS;
1626
1627 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1628 if (!ret)
1629 ret = len;
1630
1631 return ret;
1632 }
1633
1634 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1635 {
1636 struct ucma_file *file = filp->private_data;
1637 unsigned int mask = 0;
1638
1639 poll_wait(filp, &file->poll_wait, wait);
1640
1641 if (!list_empty(&file->event_list))
1642 mask = POLLIN | POLLRDNORM;
1643
1644 return mask;
1645 }
1646
1647 /*
1648 * ucma_open() does not need the BKL:
1649 *
1650 * - no global state is referred to;
1651 * - there is no ioctl method to race against;
1652 * - no further module initialization is required for open to work
1653 * after the device is registered.
1654 */
1655 static int ucma_open(struct inode *inode, struct file *filp)
1656 {
1657 struct ucma_file *file;
1658
1659 file = kmalloc(sizeof *file, GFP_KERNEL);
1660 if (!file)
1661 return -ENOMEM;
1662
1663 file->close_wq = create_singlethread_workqueue("ucma_close_id");
1664 if (!file->close_wq) {
1665 kfree(file);
1666 return -ENOMEM;
1667 }
1668
1669 INIT_LIST_HEAD(&file->event_list);
1670 INIT_LIST_HEAD(&file->ctx_list);
1671 init_waitqueue_head(&file->poll_wait);
1672 mutex_init(&file->mut);
1673
1674 filp->private_data = file;
1675 file->filp = filp;
1676
1677 return nonseekable_open(inode, filp);
1678 }
1679
1680 static int ucma_close(struct inode *inode, struct file *filp)
1681 {
1682 struct ucma_file *file = filp->private_data;
1683 struct ucma_context *ctx, *tmp;
1684
1685 mutex_lock(&file->mut);
1686 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1687 ctx->destroying = 1;
1688 mutex_unlock(&file->mut);
1689
1690 mutex_lock(&mut);
1691 idr_remove(&ctx_idr, ctx->id);
1692 mutex_unlock(&mut);
1693
1694 flush_workqueue(file->close_wq);
1695 /* At that step once ctx was marked as destroying and workqueue
1696 * was flushed we are safe from any inflights handlers that
1697 * might put other closing task.
1698 */
1699 mutex_lock(&mut);
1700 if (!ctx->closing) {
1701 mutex_unlock(&mut);
1702 /* rdma_destroy_id ensures that no event handlers are
1703 * inflight for that id before releasing it.
1704 */
1705 rdma_destroy_id(ctx->cm_id);
1706 } else {
1707 mutex_unlock(&mut);
1708 }
1709
1710 ucma_free_ctx(ctx);
1711 mutex_lock(&file->mut);
1712 }
1713 mutex_unlock(&file->mut);
1714 destroy_workqueue(file->close_wq);
1715 kfree(file);
1716 return 0;
1717 }
1718
1719 static const struct file_operations ucma_fops = {
1720 .owner = THIS_MODULE,
1721 .open = ucma_open,
1722 .release = ucma_close,
1723 .write = ucma_write,
1724 .poll = ucma_poll,
1725 .llseek = no_llseek,
1726 };
1727
1728 static struct miscdevice ucma_misc = {
1729 .minor = MISC_DYNAMIC_MINOR,
1730 .name = "rdma_cm",
1731 .nodename = "infiniband/rdma_cm",
1732 .mode = 0666,
1733 .fops = &ucma_fops,
1734 };
1735
1736 static ssize_t show_abi_version(struct device *dev,
1737 struct device_attribute *attr,
1738 char *buf)
1739 {
1740 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1741 }
1742 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1743
1744 static int __init ucma_init(void)
1745 {
1746 int ret;
1747
1748 ret = misc_register(&ucma_misc);
1749 if (ret)
1750 return ret;
1751
1752 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1753 if (ret) {
1754 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1755 goto err1;
1756 }
1757
1758 ucma_ctl_table_hdr = register_net_sysctl(&init_net, "net/rdma_ucm", ucma_ctl_table);
1759 if (!ucma_ctl_table_hdr) {
1760 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1761 ret = -ENOMEM;
1762 goto err2;
1763 }
1764 return 0;
1765 err2:
1766 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1767 err1:
1768 misc_deregister(&ucma_misc);
1769 return ret;
1770 }
1771
1772 static void __exit ucma_cleanup(void)
1773 {
1774 unregister_net_sysctl_table(ucma_ctl_table_hdr);
1775 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1776 misc_deregister(&ucma_misc);
1777 idr_destroy(&ctx_idr);
1778 idr_destroy(&multicast_idr);
1779 }
1780
1781 module_init(ucma_init);
1782 module_exit(ucma_cleanup);
Linux with added FPC-III support