1 /*
2  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
4  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
6  * Copyright (c) 2005 PathScale, Inc. All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/device.h>
40 #include <linux/err.h>
41 #include <linux/fs.h>
42 #include <linux/poll.h>
43 #include <linux/file.h>
44 #include <linux/mount.h>
45 #include <linux/cdev.h>
46 
47 #include <asm/uaccess.h>
48 
49 #include "uverbs.h"
50 
51 MODULE_AUTHOR("Roland Dreier");
52 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
53 MODULE_LICENSE("Dual BSD/GPL");
54 
55 #define INFINIBANDEVENTFS_MAGIC	0x49426576	/* "IBev" */
56 
57 enum {
58 	IB_UVERBS_MAJOR       = 231,
59 	IB_UVERBS_BASE_MINOR  = 192,
60 	IB_UVERBS_MAX_DEVICES = 32
61 };
62 
63 #define IB_UVERBS_BASE_DEV	MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
64 
65 static struct class *uverbs_class;
66 
67 DEFINE_SPINLOCK(ib_uverbs_idr_lock);
68 DEFINE_IDR(ib_uverbs_pd_idr);
69 DEFINE_IDR(ib_uverbs_mr_idr);
70 DEFINE_IDR(ib_uverbs_mw_idr);
71 DEFINE_IDR(ib_uverbs_ah_idr);
72 DEFINE_IDR(ib_uverbs_cq_idr);
73 DEFINE_IDR(ib_uverbs_qp_idr);
74 DEFINE_IDR(ib_uverbs_srq_idr);
75 
76 static spinlock_t map_lock;
77 static struct ib_uverbs_device *dev_table[IB_UVERBS_MAX_DEVICES];
78 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
79 
80 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
81 				     const char __user *buf, int in_len,
82 				     int out_len) = {
83 	[IB_USER_VERBS_CMD_GET_CONTEXT]   	= ib_uverbs_get_context,
84 	[IB_USER_VERBS_CMD_QUERY_DEVICE]  	= ib_uverbs_query_device,
85 	[IB_USER_VERBS_CMD_QUERY_PORT]    	= ib_uverbs_query_port,
86 	[IB_USER_VERBS_CMD_ALLOC_PD]      	= ib_uverbs_alloc_pd,
87 	[IB_USER_VERBS_CMD_DEALLOC_PD]    	= ib_uverbs_dealloc_pd,
88 	[IB_USER_VERBS_CMD_REG_MR]        	= ib_uverbs_reg_mr,
89 	[IB_USER_VERBS_CMD_DEREG_MR]      	= ib_uverbs_dereg_mr,
90 	[IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
91 	[IB_USER_VERBS_CMD_CREATE_CQ]     	= ib_uverbs_create_cq,
92 	[IB_USER_VERBS_CMD_RESIZE_CQ]     	= ib_uverbs_resize_cq,
93 	[IB_USER_VERBS_CMD_POLL_CQ]     	= ib_uverbs_poll_cq,
94 	[IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]     	= ib_uverbs_req_notify_cq,
95 	[IB_USER_VERBS_CMD_DESTROY_CQ]    	= ib_uverbs_destroy_cq,
96 	[IB_USER_VERBS_CMD_CREATE_QP]     	= ib_uverbs_create_qp,
97 	[IB_USER_VERBS_CMD_QUERY_QP]     	= ib_uverbs_query_qp,
98 	[IB_USER_VERBS_CMD_MODIFY_QP]     	= ib_uverbs_modify_qp,
99 	[IB_USER_VERBS_CMD_DESTROY_QP]    	= ib_uverbs_destroy_qp,
100 	[IB_USER_VERBS_CMD_POST_SEND]    	= ib_uverbs_post_send,
101 	[IB_USER_VERBS_CMD_POST_RECV]    	= ib_uverbs_post_recv,
102 	[IB_USER_VERBS_CMD_POST_SRQ_RECV]    	= ib_uverbs_post_srq_recv,
103 	[IB_USER_VERBS_CMD_CREATE_AH]    	= ib_uverbs_create_ah,
104 	[IB_USER_VERBS_CMD_DESTROY_AH]    	= ib_uverbs_destroy_ah,
105 	[IB_USER_VERBS_CMD_ATTACH_MCAST]  	= ib_uverbs_attach_mcast,
106 	[IB_USER_VERBS_CMD_DETACH_MCAST]  	= ib_uverbs_detach_mcast,
107 	[IB_USER_VERBS_CMD_CREATE_SRQ]    	= ib_uverbs_create_srq,
108 	[IB_USER_VERBS_CMD_MODIFY_SRQ]    	= ib_uverbs_modify_srq,
109 	[IB_USER_VERBS_CMD_QUERY_SRQ]     	= ib_uverbs_query_srq,
110 	[IB_USER_VERBS_CMD_DESTROY_SRQ]   	= ib_uverbs_destroy_srq,
111 };
112 
113 static struct vfsmount *uverbs_event_mnt;
114 
115 static void ib_uverbs_add_one(struct ib_device *device);
116 static void ib_uverbs_remove_one(struct ib_device *device);
117 
118 static void ib_uverbs_release_dev(struct kref *ref)
119 {
120 	struct ib_uverbs_device *dev =
121 		container_of(ref, struct ib_uverbs_device, ref);
122 
123 	complete(&dev->comp);
124 }
125 
126 static void ib_uverbs_release_event_file(struct kref *ref)
127 {
128 	struct ib_uverbs_event_file *file =
129 		container_of(ref, struct ib_uverbs_event_file, ref);
130 
131 	kfree(file);
132 }
133 
134 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
135 			  struct ib_uverbs_event_file *ev_file,
136 			  struct ib_ucq_object *uobj)
137 {
138 	struct ib_uverbs_event *evt, *tmp;
139 
140 	if (ev_file) {
141 		spin_lock_irq(&ev_file->lock);
142 		list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
143 			list_del(&evt->list);
144 			kfree(evt);
145 		}
146 		spin_unlock_irq(&ev_file->lock);
147 
148 		kref_put(&ev_file->ref, ib_uverbs_release_event_file);
149 	}
150 
151 	spin_lock_irq(&file->async_file->lock);
152 	list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
153 		list_del(&evt->list);
154 		kfree(evt);
155 	}
156 	spin_unlock_irq(&file->async_file->lock);
157 }
158 
159 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
160 			      struct ib_uevent_object *uobj)
161 {
162 	struct ib_uverbs_event *evt, *tmp;
163 
164 	spin_lock_irq(&file->async_file->lock);
165 	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
166 		list_del(&evt->list);
167 		kfree(evt);
168 	}
169 	spin_unlock_irq(&file->async_file->lock);
170 }
171 
172 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
173 				    struct ib_uqp_object *uobj)
174 {
175 	struct ib_uverbs_mcast_entry *mcast, *tmp;
176 
177 	list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
178 		ib_detach_mcast(qp, &mcast->gid, mcast->lid);
179 		list_del(&mcast->list);
180 		kfree(mcast);
181 	}
182 }
183 
184 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
185 				      struct ib_ucontext *context)
186 {
187 	struct ib_uobject *uobj, *tmp;
188 
189 	if (!context)
190 		return 0;
191 
192 	context->closing = 1;
193 
194 	list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
195 		struct ib_ah *ah = uobj->object;
196 
197 		idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
198 		ib_destroy_ah(ah);
199 		kfree(uobj);
200 	}
201 
202 	list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
203 		struct ib_qp *qp = uobj->object;
204 		struct ib_uqp_object *uqp =
205 			container_of(uobj, struct ib_uqp_object, uevent.uobject);
206 
207 		idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
208 		ib_uverbs_detach_umcast(qp, uqp);
209 		ib_destroy_qp(qp);
210 		ib_uverbs_release_uevent(file, &uqp->uevent);
211 		kfree(uqp);
212 	}
213 
214 	list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
215 		struct ib_cq *cq = uobj->object;
216 		struct ib_uverbs_event_file *ev_file = cq->cq_context;
217 		struct ib_ucq_object *ucq =
218 			container_of(uobj, struct ib_ucq_object, uobject);
219 
220 		idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
221 		ib_destroy_cq(cq);
222 		ib_uverbs_release_ucq(file, ev_file, ucq);
223 		kfree(ucq);
224 	}
225 
226 	list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
227 		struct ib_srq *srq = uobj->object;
228 		struct ib_uevent_object *uevent =
229 			container_of(uobj, struct ib_uevent_object, uobject);
230 
231 		idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
232 		ib_destroy_srq(srq);
233 		ib_uverbs_release_uevent(file, uevent);
234 		kfree(uevent);
235 	}
236 
237 	/* XXX Free MWs */
238 
239 	list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
240 		struct ib_mr *mr = uobj->object;
241 
242 		idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
243 		ib_dereg_mr(mr);
244 		kfree(uobj);
245 	}
246 
247 	list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
248 		struct ib_pd *pd = uobj->object;
249 
250 		idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
251 		ib_dealloc_pd(pd);
252 		kfree(uobj);
253 	}
254 
255 	return context->device->dealloc_ucontext(context);
256 }
257 
258 static void ib_uverbs_release_file(struct kref *ref)
259 {
260 	struct ib_uverbs_file *file =
261 		container_of(ref, struct ib_uverbs_file, ref);
262 
263 	module_put(file->device->ib_dev->owner);
264 	kref_put(&file->device->ref, ib_uverbs_release_dev);
265 
266 	kfree(file);
267 }
268 
269 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
270 				    size_t count, loff_t *pos)
271 {
272 	struct ib_uverbs_event_file *file = filp->private_data;
273 	struct ib_uverbs_event *event;
274 	int eventsz;
275 	int ret = 0;
276 
277 	spin_lock_irq(&file->lock);
278 
279 	while (list_empty(&file->event_list)) {
280 		spin_unlock_irq(&file->lock);
281 
282 		if (filp->f_flags & O_NONBLOCK)
283 			return -EAGAIN;
284 
285 		if (wait_event_interruptible(file->poll_wait,
286 					     !list_empty(&file->event_list)))
287 			return -ERESTARTSYS;
288 
289 		spin_lock_irq(&file->lock);
290 	}
291 
292 	event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
293 
294 	if (file->is_async)
295 		eventsz = sizeof (struct ib_uverbs_async_event_desc);
296 	else
297 		eventsz = sizeof (struct ib_uverbs_comp_event_desc);
298 
299 	if (eventsz > count) {
300 		ret   = -EINVAL;
301 		event = NULL;
302 	} else {
303 		list_del(file->event_list.next);
304 		if (event->counter) {
305 			++(*event->counter);
306 			list_del(&event->obj_list);
307 		}
308 	}
309 
310 	spin_unlock_irq(&file->lock);
311 
312 	if (event) {
313 		if (copy_to_user(buf, event, eventsz))
314 			ret = -EFAULT;
315 		else
316 			ret = eventsz;
317 	}
318 
319 	kfree(event);
320 
321 	return ret;
322 }
323 
324 static unsigned int ib_uverbs_event_poll(struct file *filp,
325 					 struct poll_table_struct *wait)
326 {
327 	unsigned int pollflags = 0;
328 	struct ib_uverbs_event_file *file = filp->private_data;
329 
330 	poll_wait(filp, &file->poll_wait, wait);
331 
332 	spin_lock_irq(&file->lock);
333 	if (!list_empty(&file->event_list))
334 		pollflags = POLLIN | POLLRDNORM;
335 	spin_unlock_irq(&file->lock);
336 
337 	return pollflags;
338 }
339 
340 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
341 {
342 	struct ib_uverbs_event_file *file = filp->private_data;
343 
344 	return fasync_helper(fd, filp, on, &file->async_queue);
345 }
346 
347 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
348 {
349 	struct ib_uverbs_event_file *file = filp->private_data;
350 	struct ib_uverbs_event *entry, *tmp;
351 
352 	spin_lock_irq(&file->lock);
353 	file->is_closed = 1;
354 	list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
355 		if (entry->counter)
356 			list_del(&entry->obj_list);
357 		kfree(entry);
358 	}
359 	spin_unlock_irq(&file->lock);
360 
361 	if (file->is_async) {
362 		ib_unregister_event_handler(&file->uverbs_file->event_handler);
363 		kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
364 	}
365 	kref_put(&file->ref, ib_uverbs_release_event_file);
366 
367 	return 0;
368 }
369 
370 static const struct file_operations uverbs_event_fops = {
371 	.owner	 = THIS_MODULE,
372 	.read 	 = ib_uverbs_event_read,
373 	.poll    = ib_uverbs_event_poll,
374 	.release = ib_uverbs_event_close,
375 	.fasync  = ib_uverbs_event_fasync
376 };
377 
378 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
379 {
380 	struct ib_uverbs_event_file    *file = cq_context;
381 	struct ib_ucq_object	       *uobj;
382 	struct ib_uverbs_event	       *entry;
383 	unsigned long			flags;
384 
385 	if (!file)
386 		return;
387 
388 	spin_lock_irqsave(&file->lock, flags);
389 	if (file->is_closed) {
390 		spin_unlock_irqrestore(&file->lock, flags);
391 		return;
392 	}
393 
394 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
395 	if (!entry) {
396 		spin_unlock_irqrestore(&file->lock, flags);
397 		return;
398 	}
399 
400 	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
401 
402 	entry->desc.comp.cq_handle = cq->uobject->user_handle;
403 	entry->counter		   = &uobj->comp_events_reported;
404 
405 	list_add_tail(&entry->list, &file->event_list);
406 	list_add_tail(&entry->obj_list, &uobj->comp_list);
407 	spin_unlock_irqrestore(&file->lock, flags);
408 
409 	wake_up_interruptible(&file->poll_wait);
410 	kill_fasync(&file->async_queue, SIGIO, POLL_IN);
411 }
412 
413 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
414 				    __u64 element, __u64 event,
415 				    struct list_head *obj_list,
416 				    u32 *counter)
417 {
418 	struct ib_uverbs_event *entry;
419 	unsigned long flags;
420 
421 	spin_lock_irqsave(&file->async_file->lock, flags);
422 	if (file->async_file->is_closed) {
423 		spin_unlock_irqrestore(&file->async_file->lock, flags);
424 		return;
425 	}
426 
427 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
428 	if (!entry) {
429 		spin_unlock_irqrestore(&file->async_file->lock, flags);
430 		return;
431 	}
432 
433 	entry->desc.async.element    = element;
434 	entry->desc.async.event_type = event;
435 	entry->counter               = counter;
436 
437 	list_add_tail(&entry->list, &file->async_file->event_list);
438 	if (obj_list)
439 		list_add_tail(&entry->obj_list, obj_list);
440 	spin_unlock_irqrestore(&file->async_file->lock, flags);
441 
442 	wake_up_interruptible(&file->async_file->poll_wait);
443 	kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
444 }
445 
446 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
447 {
448 	struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
449 						  struct ib_ucq_object, uobject);
450 
451 	ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
452 				event->event, &uobj->async_list,
453 				&uobj->async_events_reported);
454 }
455 
456 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
457 {
458 	struct ib_uevent_object *uobj;
459 
460 	uobj = container_of(event->element.qp->uobject,
461 			    struct ib_uevent_object, uobject);
462 
463 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
464 				event->event, &uobj->event_list,
465 				&uobj->events_reported);
466 }
467 
468 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
469 {
470 	struct ib_uevent_object *uobj;
471 
472 	uobj = container_of(event->element.srq->uobject,
473 			    struct ib_uevent_object, uobject);
474 
475 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
476 				event->event, &uobj->event_list,
477 				&uobj->events_reported);
478 }
479 
480 void ib_uverbs_event_handler(struct ib_event_handler *handler,
481 			     struct ib_event *event)
482 {
483 	struct ib_uverbs_file *file =
484 		container_of(handler, struct ib_uverbs_file, event_handler);
485 
486 	ib_uverbs_async_handler(file, event->element.port_num, event->event,
487 				NULL, NULL);
488 }
489 
490 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
491 					int is_async, int *fd)
492 {
493 	struct ib_uverbs_event_file *ev_file;
494 	struct file *filp;
495 	int ret;
496 
497 	ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
498 	if (!ev_file)
499 		return ERR_PTR(-ENOMEM);
500 
501 	kref_init(&ev_file->ref);
502 	spin_lock_init(&ev_file->lock);
503 	INIT_LIST_HEAD(&ev_file->event_list);
504 	init_waitqueue_head(&ev_file->poll_wait);
505 	ev_file->uverbs_file = uverbs_file;
506 	ev_file->async_queue = NULL;
507 	ev_file->is_async    = is_async;
508 	ev_file->is_closed   = 0;
509 
510 	*fd = get_unused_fd();
511 	if (*fd < 0) {
512 		ret = *fd;
513 		goto err;
514 	}
515 
516 	/*
517 	 * fops_get() can't fail here, because we're coming from a
518 	 * system call on a uverbs file, which will already have a
519 	 * module reference.
520 	 */
521 	filp = alloc_file(uverbs_event_mnt, dget(uverbs_event_mnt->mnt_root),
522 			  FMODE_READ, fops_get(&uverbs_event_fops));
523 	if (!filp) {
524 		ret = -ENFILE;
525 		goto err_fd;
526 	}
527 
528 	filp->private_data = ev_file;
529 
530 	return filp;
531 
532 err_fd:
533 	put_unused_fd(*fd);
534 
535 err:
536 	kfree(ev_file);
537 	return ERR_PTR(ret);
538 }
539 
540 /*
541  * Look up a completion event file by FD.  If lookup is successful,
542  * takes a ref to the event file struct that it returns; if
543  * unsuccessful, returns NULL.
544  */
545 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
546 {
547 	struct ib_uverbs_event_file *ev_file = NULL;
548 	struct file *filp;
549 
550 	filp = fget(fd);
551 	if (!filp)
552 		return NULL;
553 
554 	if (filp->f_op != &uverbs_event_fops)
555 		goto out;
556 
557 	ev_file = filp->private_data;
558 	if (ev_file->is_async) {
559 		ev_file = NULL;
560 		goto out;
561 	}
562 
563 	kref_get(&ev_file->ref);
564 
565 out:
566 	fput(filp);
567 	return ev_file;
568 }
569 
570 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
571 			     size_t count, loff_t *pos)
572 {
573 	struct ib_uverbs_file *file = filp->private_data;
574 	struct ib_uverbs_cmd_hdr hdr;
575 
576 	if (count < sizeof hdr)
577 		return -EINVAL;
578 
579 	if (copy_from_user(&hdr, buf, sizeof hdr))
580 		return -EFAULT;
581 
582 	if (hdr.in_words * 4 != count)
583 		return -EINVAL;
584 
585 	if (hdr.command < 0				||
586 	    hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
587 	    !uverbs_cmd_table[hdr.command]		||
588 	    !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
589 		return -EINVAL;
590 
591 	if (!file->ucontext &&
592 	    hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
593 		return -EINVAL;
594 
595 	return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
596 					     hdr.in_words * 4, hdr.out_words * 4);
597 }
598 
599 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
600 {
601 	struct ib_uverbs_file *file = filp->private_data;
602 
603 	if (!file->ucontext)
604 		return -ENODEV;
605 	else
606 		return file->device->ib_dev->mmap(file->ucontext, vma);
607 }
608 
609 /*
610  * ib_uverbs_open() does not need the BKL:
611  *
612  *  - dev_table[] accesses are protected by map_lock, the
613  *    ib_uverbs_device structures are properly reference counted, and
614  *    everything else is purely local to the file being created, so
615  *    races against other open calls are not a problem;
616  *  - there is no ioctl method to race against;
617  *  - the device is added to dev_table[] as the last part of module
618  *    initialization, the open method will either immediately run
619  *    -ENXIO, or all required initialization will be done.
620  */
621 static int ib_uverbs_open(struct inode *inode, struct file *filp)
622 {
623 	struct ib_uverbs_device *dev;
624 	struct ib_uverbs_file *file;
625 	int ret;
626 
627 	spin_lock(&map_lock);
628 	dev = dev_table[iminor(inode) - IB_UVERBS_BASE_MINOR];
629 	if (dev)
630 		kref_get(&dev->ref);
631 	spin_unlock(&map_lock);
632 
633 	if (!dev)
634 		return -ENXIO;
635 
636 	if (!try_module_get(dev->ib_dev->owner)) {
637 		ret = -ENODEV;
638 		goto err;
639 	}
640 
641 	file = kmalloc(sizeof *file, GFP_KERNEL);
642 	if (!file) {
643 		ret = -ENOMEM;
644 		goto err_module;
645 	}
646 
647 	file->device	 = dev;
648 	file->ucontext	 = NULL;
649 	file->async_file = NULL;
650 	kref_init(&file->ref);
651 	mutex_init(&file->mutex);
652 
653 	filp->private_data = file;
654 
655 	return 0;
656 
657 err_module:
658 	module_put(dev->ib_dev->owner);
659 
660 err:
661 	kref_put(&dev->ref, ib_uverbs_release_dev);
662 	return ret;
663 }
664 
665 static int ib_uverbs_close(struct inode *inode, struct file *filp)
666 {
667 	struct ib_uverbs_file *file = filp->private_data;
668 
669 	ib_uverbs_cleanup_ucontext(file, file->ucontext);
670 
671 	if (file->async_file)
672 		kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
673 
674 	kref_put(&file->ref, ib_uverbs_release_file);
675 
676 	return 0;
677 }
678 
679 static const struct file_operations uverbs_fops = {
680 	.owner 	 = THIS_MODULE,
681 	.write 	 = ib_uverbs_write,
682 	.open 	 = ib_uverbs_open,
683 	.release = ib_uverbs_close
684 };
685 
686 static const struct file_operations uverbs_mmap_fops = {
687 	.owner 	 = THIS_MODULE,
688 	.write 	 = ib_uverbs_write,
689 	.mmap    = ib_uverbs_mmap,
690 	.open 	 = ib_uverbs_open,
691 	.release = ib_uverbs_close
692 };
693 
694 static struct ib_client uverbs_client = {
695 	.name   = "uverbs",
696 	.add    = ib_uverbs_add_one,
697 	.remove = ib_uverbs_remove_one
698 };
699 
700 static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
701 			  char *buf)
702 {
703 	struct ib_uverbs_device *dev = dev_get_drvdata(device);
704 
705 	if (!dev)
706 		return -ENODEV;
707 
708 	return sprintf(buf, "%s\n", dev->ib_dev->name);
709 }
710 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
711 
712 static ssize_t show_dev_abi_version(struct device *device,
713 				    struct device_attribute *attr, char *buf)
714 {
715 	struct ib_uverbs_device *dev = dev_get_drvdata(device);
716 
717 	if (!dev)
718 		return -ENODEV;
719 
720 	return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
721 }
722 static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
723 
724 static ssize_t show_abi_version(struct class *class, char *buf)
725 {
726 	return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
727 }
728 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
729 
730 static void ib_uverbs_add_one(struct ib_device *device)
731 {
732 	struct ib_uverbs_device *uverbs_dev;
733 
734 	if (!device->alloc_ucontext)
735 		return;
736 
737 	uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
738 	if (!uverbs_dev)
739 		return;
740 
741 	kref_init(&uverbs_dev->ref);
742 	init_completion(&uverbs_dev->comp);
743 
744 	spin_lock(&map_lock);
745 	uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
746 	if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
747 		spin_unlock(&map_lock);
748 		goto err;
749 	}
750 	set_bit(uverbs_dev->devnum, dev_map);
751 	spin_unlock(&map_lock);
752 
753 	uverbs_dev->ib_dev           = device;
754 	uverbs_dev->num_comp_vectors = device->num_comp_vectors;
755 
756 	uverbs_dev->cdev = cdev_alloc();
757 	if (!uverbs_dev->cdev)
758 		goto err;
759 	uverbs_dev->cdev->owner = THIS_MODULE;
760 	uverbs_dev->cdev->ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
761 	kobject_set_name(&uverbs_dev->cdev->kobj, "uverbs%d", uverbs_dev->devnum);
762 	if (cdev_add(uverbs_dev->cdev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
763 		goto err_cdev;
764 
765 	uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
766 					uverbs_dev->cdev->dev, uverbs_dev,
767 					"uverbs%d", uverbs_dev->devnum);
768 	if (IS_ERR(uverbs_dev->dev))
769 		goto err_cdev;
770 
771 	if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
772 		goto err_class;
773 	if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
774 		goto err_class;
775 
776 	spin_lock(&map_lock);
777 	dev_table[uverbs_dev->devnum] = uverbs_dev;
778 	spin_unlock(&map_lock);
779 
780 	ib_set_client_data(device, &uverbs_client, uverbs_dev);
781 
782 	return;
783 
784 err_class:
785 	device_destroy(uverbs_class, uverbs_dev->cdev->dev);
786 
787 err_cdev:
788 	cdev_del(uverbs_dev->cdev);
789 	clear_bit(uverbs_dev->devnum, dev_map);
790 
791 err:
792 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
793 	wait_for_completion(&uverbs_dev->comp);
794 	kfree(uverbs_dev);
795 	return;
796 }
797 
798 static void ib_uverbs_remove_one(struct ib_device *device)
799 {
800 	struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
801 
802 	if (!uverbs_dev)
803 		return;
804 
805 	dev_set_drvdata(uverbs_dev->dev, NULL);
806 	device_destroy(uverbs_class, uverbs_dev->cdev->dev);
807 	cdev_del(uverbs_dev->cdev);
808 
809 	spin_lock(&map_lock);
810 	dev_table[uverbs_dev->devnum] = NULL;
811 	spin_unlock(&map_lock);
812 
813 	clear_bit(uverbs_dev->devnum, dev_map);
814 
815 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
816 	wait_for_completion(&uverbs_dev->comp);
817 	kfree(uverbs_dev);
818 }
819 
820 static int uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
821 			       const char *dev_name, void *data,
822 			       struct vfsmount *mnt)
823 {
824 	return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
825 			     INFINIBANDEVENTFS_MAGIC, mnt);
826 }
827 
828 static struct file_system_type uverbs_event_fs = {
829 	/* No owner field so module can be unloaded */
830 	.name    = "infinibandeventfs",
831 	.get_sb  = uverbs_event_get_sb,
832 	.kill_sb = kill_litter_super
833 };
834 
835 static int __init ib_uverbs_init(void)
836 {
837 	int ret;
838 
839 	spin_lock_init(&map_lock);
840 
841 	ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
842 				     "infiniband_verbs");
843 	if (ret) {
844 		printk(KERN_ERR "user_verbs: couldn't register device number\n");
845 		goto out;
846 	}
847 
848 	uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
849 	if (IS_ERR(uverbs_class)) {
850 		ret = PTR_ERR(uverbs_class);
851 		printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
852 		goto out_chrdev;
853 	}
854 
855 	ret = class_create_file(uverbs_class, &class_attr_abi_version);
856 	if (ret) {
857 		printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
858 		goto out_class;
859 	}
860 
861 	ret = register_filesystem(&uverbs_event_fs);
862 	if (ret) {
863 		printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
864 		goto out_class;
865 	}
866 
867 	uverbs_event_mnt = kern_mount(&uverbs_event_fs);
868 	if (IS_ERR(uverbs_event_mnt)) {
869 		ret = PTR_ERR(uverbs_event_mnt);
870 		printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
871 		goto out_fs;
872 	}
873 
874 	ret = ib_register_client(&uverbs_client);
875 	if (ret) {
876 		printk(KERN_ERR "user_verbs: couldn't register client\n");
877 		goto out_mnt;
878 	}
879 
880 	return 0;
881 
882 out_mnt:
883 	mntput(uverbs_event_mnt);
884 
885 out_fs:
886 	unregister_filesystem(&uverbs_event_fs);
887 
888 out_class:
889 	class_destroy(uverbs_class);
890 
891 out_chrdev:
892 	unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
893 
894 out:
895 	return ret;
896 }
897 
898 static void __exit ib_uverbs_cleanup(void)
899 {
900 	ib_unregister_client(&uverbs_client);
901 	mntput(uverbs_event_mnt);
902 	unregister_filesystem(&uverbs_event_fs);
903 	class_destroy(uverbs_class);
904 	unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
905 	idr_destroy(&ib_uverbs_pd_idr);
906 	idr_destroy(&ib_uverbs_mr_idr);
907 	idr_destroy(&ib_uverbs_mw_idr);
908 	idr_destroy(&ib_uverbs_ah_idr);
909 	idr_destroy(&ib_uverbs_cq_idr);
910 	idr_destroy(&ib_uverbs_qp_idr);
911 	idr_destroy(&ib_uverbs_srq_idr);
912 }
913 
914 module_init(ib_uverbs_init);
915 module_exit(ib_uverbs_cleanup);
916