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