xref: /linux/drivers/infiniband/core/uverbs_main.c (revision de2fe5e07d58424bc286fff3fd3c1b0bf933cd58)
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  * $Id: uverbs_main.c 2733 2005-06-28 19:14:34Z roland $
37  */
38 
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/device.h>
42 #include <linux/err.h>
43 #include <linux/fs.h>
44 #include <linux/poll.h>
45 #include <linux/file.h>
46 #include <linux/mount.h>
47 #include <linux/cdev.h>
48 
49 #include <asm/uaccess.h>
50 
51 #include "uverbs.h"
52 
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
55 MODULE_LICENSE("Dual BSD/GPL");
56 
57 #define INFINIBANDEVENTFS_MAGIC	0x49426576	/* "IBev" */
58 
59 enum {
60 	IB_UVERBS_MAJOR       = 231,
61 	IB_UVERBS_BASE_MINOR  = 192,
62 	IB_UVERBS_MAX_DEVICES = 32
63 };
64 
65 #define IB_UVERBS_BASE_DEV	MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
66 
67 static struct class *uverbs_class;
68 
69 DEFINE_MUTEX(ib_uverbs_idr_mutex);
70 DEFINE_IDR(ib_uverbs_pd_idr);
71 DEFINE_IDR(ib_uverbs_mr_idr);
72 DEFINE_IDR(ib_uverbs_mw_idr);
73 DEFINE_IDR(ib_uverbs_ah_idr);
74 DEFINE_IDR(ib_uverbs_cq_idr);
75 DEFINE_IDR(ib_uverbs_qp_idr);
76 DEFINE_IDR(ib_uverbs_srq_idr);
77 
78 static spinlock_t map_lock;
79 static struct ib_uverbs_device *dev_table[IB_UVERBS_MAX_DEVICES];
80 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
81 
82 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
83 				     const char __user *buf, int in_len,
84 				     int out_len) = {
85 	[IB_USER_VERBS_CMD_GET_CONTEXT]   	= ib_uverbs_get_context,
86 	[IB_USER_VERBS_CMD_QUERY_DEVICE]  	= ib_uverbs_query_device,
87 	[IB_USER_VERBS_CMD_QUERY_PORT]    	= ib_uverbs_query_port,
88 	[IB_USER_VERBS_CMD_ALLOC_PD]      	= ib_uverbs_alloc_pd,
89 	[IB_USER_VERBS_CMD_DEALLOC_PD]    	= ib_uverbs_dealloc_pd,
90 	[IB_USER_VERBS_CMD_REG_MR]        	= ib_uverbs_reg_mr,
91 	[IB_USER_VERBS_CMD_DEREG_MR]      	= ib_uverbs_dereg_mr,
92 	[IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
93 	[IB_USER_VERBS_CMD_CREATE_CQ]     	= ib_uverbs_create_cq,
94 	[IB_USER_VERBS_CMD_RESIZE_CQ]     	= ib_uverbs_resize_cq,
95 	[IB_USER_VERBS_CMD_POLL_CQ]     	= ib_uverbs_poll_cq,
96 	[IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]     	= ib_uverbs_req_notify_cq,
97 	[IB_USER_VERBS_CMD_DESTROY_CQ]    	= ib_uverbs_destroy_cq,
98 	[IB_USER_VERBS_CMD_CREATE_QP]     	= ib_uverbs_create_qp,
99 	[IB_USER_VERBS_CMD_QUERY_QP]     	= ib_uverbs_query_qp,
100 	[IB_USER_VERBS_CMD_MODIFY_QP]     	= ib_uverbs_modify_qp,
101 	[IB_USER_VERBS_CMD_DESTROY_QP]    	= ib_uverbs_destroy_qp,
102 	[IB_USER_VERBS_CMD_POST_SEND]    	= ib_uverbs_post_send,
103 	[IB_USER_VERBS_CMD_POST_RECV]    	= ib_uverbs_post_recv,
104 	[IB_USER_VERBS_CMD_POST_SRQ_RECV]    	= ib_uverbs_post_srq_recv,
105 	[IB_USER_VERBS_CMD_CREATE_AH]    	= ib_uverbs_create_ah,
106 	[IB_USER_VERBS_CMD_DESTROY_AH]    	= ib_uverbs_destroy_ah,
107 	[IB_USER_VERBS_CMD_ATTACH_MCAST]  	= ib_uverbs_attach_mcast,
108 	[IB_USER_VERBS_CMD_DETACH_MCAST]  	= ib_uverbs_detach_mcast,
109 	[IB_USER_VERBS_CMD_CREATE_SRQ]    	= ib_uverbs_create_srq,
110 	[IB_USER_VERBS_CMD_MODIFY_SRQ]    	= ib_uverbs_modify_srq,
111 	[IB_USER_VERBS_CMD_QUERY_SRQ]     	= ib_uverbs_query_srq,
112 	[IB_USER_VERBS_CMD_DESTROY_SRQ]   	= ib_uverbs_destroy_srq,
113 };
114 
115 static struct vfsmount *uverbs_event_mnt;
116 
117 static void ib_uverbs_add_one(struct ib_device *device);
118 static void ib_uverbs_remove_one(struct ib_device *device);
119 
120 static void ib_uverbs_release_dev(struct kref *ref)
121 {
122 	struct ib_uverbs_device *dev =
123 		container_of(ref, struct ib_uverbs_device, ref);
124 
125 	kfree(dev);
126 }
127 
128 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
129 			  struct ib_uverbs_event_file *ev_file,
130 			  struct ib_ucq_object *uobj)
131 {
132 	struct ib_uverbs_event *evt, *tmp;
133 
134 	if (ev_file) {
135 		spin_lock_irq(&ev_file->lock);
136 		list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
137 			list_del(&evt->list);
138 			kfree(evt);
139 		}
140 		spin_unlock_irq(&ev_file->lock);
141 
142 		kref_put(&ev_file->ref, ib_uverbs_release_event_file);
143 	}
144 
145 	spin_lock_irq(&file->async_file->lock);
146 	list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
147 		list_del(&evt->list);
148 		kfree(evt);
149 	}
150 	spin_unlock_irq(&file->async_file->lock);
151 }
152 
153 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
154 			      struct ib_uevent_object *uobj)
155 {
156 	struct ib_uverbs_event *evt, *tmp;
157 
158 	spin_lock_irq(&file->async_file->lock);
159 	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
160 		list_del(&evt->list);
161 		kfree(evt);
162 	}
163 	spin_unlock_irq(&file->async_file->lock);
164 }
165 
166 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
167 				    struct ib_uqp_object *uobj)
168 {
169 	struct ib_uverbs_mcast_entry *mcast, *tmp;
170 
171 	list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
172 		ib_detach_mcast(qp, &mcast->gid, mcast->lid);
173 		list_del(&mcast->list);
174 		kfree(mcast);
175 	}
176 }
177 
178 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
179 				      struct ib_ucontext *context)
180 {
181 	struct ib_uobject *uobj, *tmp;
182 
183 	if (!context)
184 		return 0;
185 
186 	mutex_lock(&ib_uverbs_idr_mutex);
187 
188 	list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
189 		struct ib_ah *ah = idr_find(&ib_uverbs_ah_idr, uobj->id);
190 		idr_remove(&ib_uverbs_ah_idr, uobj->id);
191 		ib_destroy_ah(ah);
192 		list_del(&uobj->list);
193 		kfree(uobj);
194 	}
195 
196 	list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
197 		struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
198 		struct ib_uqp_object *uqp =
199 			container_of(uobj, struct ib_uqp_object, uevent.uobject);
200 		idr_remove(&ib_uverbs_qp_idr, uobj->id);
201 		ib_uverbs_detach_umcast(qp, uqp);
202 		ib_destroy_qp(qp);
203 		list_del(&uobj->list);
204 		ib_uverbs_release_uevent(file, &uqp->uevent);
205 		kfree(uqp);
206 	}
207 
208 	list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
209 		struct ib_cq *cq = idr_find(&ib_uverbs_cq_idr, uobj->id);
210 		struct ib_uverbs_event_file *ev_file = cq->cq_context;
211 		struct ib_ucq_object *ucq =
212 			container_of(uobj, struct ib_ucq_object, uobject);
213 		idr_remove(&ib_uverbs_cq_idr, uobj->id);
214 		ib_destroy_cq(cq);
215 		list_del(&uobj->list);
216 		ib_uverbs_release_ucq(file, ev_file, ucq);
217 		kfree(ucq);
218 	}
219 
220 	list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
221 		struct ib_srq *srq = idr_find(&ib_uverbs_srq_idr, uobj->id);
222 		struct ib_uevent_object *uevent =
223 			container_of(uobj, struct ib_uevent_object, uobject);
224 		idr_remove(&ib_uverbs_srq_idr, uobj->id);
225 		ib_destroy_srq(srq);
226 		list_del(&uobj->list);
227 		ib_uverbs_release_uevent(file, uevent);
228 		kfree(uevent);
229 	}
230 
231 	/* XXX Free MWs */
232 
233 	list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
234 		struct ib_mr *mr = idr_find(&ib_uverbs_mr_idr, uobj->id);
235 		struct ib_device *mrdev = mr->device;
236 		struct ib_umem_object *memobj;
237 
238 		idr_remove(&ib_uverbs_mr_idr, uobj->id);
239 		ib_dereg_mr(mr);
240 
241 		memobj = container_of(uobj, struct ib_umem_object, uobject);
242 		ib_umem_release_on_close(mrdev, &memobj->umem);
243 
244 		list_del(&uobj->list);
245 		kfree(memobj);
246 	}
247 
248 	list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
249 		struct ib_pd *pd = idr_find(&ib_uverbs_pd_idr, uobj->id);
250 		idr_remove(&ib_uverbs_pd_idr, uobj->id);
251 		ib_dealloc_pd(pd);
252 		list_del(&uobj->list);
253 		kfree(uobj);
254 	}
255 
256 	mutex_unlock(&ib_uverbs_idr_mutex);
257 
258 	return context->device->dealloc_ucontext(context);
259 }
260 
261 static void ib_uverbs_release_file(struct kref *ref)
262 {
263 	struct ib_uverbs_file *file =
264 		container_of(ref, struct ib_uverbs_file, ref);
265 
266 	module_put(file->device->ib_dev->owner);
267 	kref_put(&file->device->ref, ib_uverbs_release_dev);
268 
269 	kfree(file);
270 }
271 
272 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
273 				    size_t count, loff_t *pos)
274 {
275 	struct ib_uverbs_event_file *file = filp->private_data;
276 	struct ib_uverbs_event *event;
277 	int eventsz;
278 	int ret = 0;
279 
280 	spin_lock_irq(&file->lock);
281 
282 	while (list_empty(&file->event_list)) {
283 		spin_unlock_irq(&file->lock);
284 
285 		if (filp->f_flags & O_NONBLOCK)
286 			return -EAGAIN;
287 
288 		if (wait_event_interruptible(file->poll_wait,
289 					     !list_empty(&file->event_list)))
290 			return -ERESTARTSYS;
291 
292 		spin_lock_irq(&file->lock);
293 	}
294 
295 	event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
296 
297 	if (file->is_async)
298 		eventsz = sizeof (struct ib_uverbs_async_event_desc);
299 	else
300 		eventsz = sizeof (struct ib_uverbs_comp_event_desc);
301 
302 	if (eventsz > count) {
303 		ret   = -EINVAL;
304 		event = NULL;
305 	} else {
306 		list_del(file->event_list.next);
307 		if (event->counter) {
308 			++(*event->counter);
309 			list_del(&event->obj_list);
310 		}
311 	}
312 
313 	spin_unlock_irq(&file->lock);
314 
315 	if (event) {
316 		if (copy_to_user(buf, event, eventsz))
317 			ret = -EFAULT;
318 		else
319 			ret = eventsz;
320 	}
321 
322 	kfree(event);
323 
324 	return ret;
325 }
326 
327 static unsigned int ib_uverbs_event_poll(struct file *filp,
328 					 struct poll_table_struct *wait)
329 {
330 	unsigned int pollflags = 0;
331 	struct ib_uverbs_event_file *file = filp->private_data;
332 
333 	poll_wait(filp, &file->poll_wait, wait);
334 
335 	spin_lock_irq(&file->lock);
336 	if (!list_empty(&file->event_list))
337 		pollflags = POLLIN | POLLRDNORM;
338 	spin_unlock_irq(&file->lock);
339 
340 	return pollflags;
341 }
342 
343 void ib_uverbs_release_event_file(struct kref *ref)
344 {
345 	struct ib_uverbs_event_file *file =
346 		container_of(ref, struct ib_uverbs_event_file, ref);
347 
348 	kfree(file);
349 }
350 
351 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
352 {
353 	struct ib_uverbs_event_file *file = filp->private_data;
354 
355 	return fasync_helper(fd, filp, on, &file->async_queue);
356 }
357 
358 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
359 {
360 	struct ib_uverbs_event_file *file = filp->private_data;
361 	struct ib_uverbs_event *entry, *tmp;
362 
363 	spin_lock_irq(&file->lock);
364 	file->file = NULL;
365 	list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
366 		if (entry->counter)
367 			list_del(&entry->obj_list);
368 		kfree(entry);
369 	}
370 	spin_unlock_irq(&file->lock);
371 
372 	ib_uverbs_event_fasync(-1, filp, 0);
373 
374 	if (file->is_async) {
375 		ib_unregister_event_handler(&file->uverbs_file->event_handler);
376 		kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
377 	}
378 	kref_put(&file->ref, ib_uverbs_release_event_file);
379 
380 	return 0;
381 }
382 
383 static struct file_operations uverbs_event_fops = {
384 	.owner	 = THIS_MODULE,
385 	.read 	 = ib_uverbs_event_read,
386 	.poll    = ib_uverbs_event_poll,
387 	.release = ib_uverbs_event_close,
388 	.fasync  = ib_uverbs_event_fasync
389 };
390 
391 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
392 {
393 	struct ib_uverbs_event_file    *file = cq_context;
394 	struct ib_ucq_object	       *uobj;
395 	struct ib_uverbs_event	       *entry;
396 	unsigned long			flags;
397 
398 	if (!file)
399 		return;
400 
401 	spin_lock_irqsave(&file->lock, flags);
402 	if (!file->file) {
403 		spin_unlock_irqrestore(&file->lock, flags);
404 		return;
405 	}
406 
407 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
408 	if (!entry) {
409 		spin_unlock_irqrestore(&file->lock, flags);
410 		return;
411 	}
412 
413 	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
414 
415 	entry->desc.comp.cq_handle = cq->uobject->user_handle;
416 	entry->counter		   = &uobj->comp_events_reported;
417 
418 	list_add_tail(&entry->list, &file->event_list);
419 	list_add_tail(&entry->obj_list, &uobj->comp_list);
420 	spin_unlock_irqrestore(&file->lock, flags);
421 
422 	wake_up_interruptible(&file->poll_wait);
423 	kill_fasync(&file->async_queue, SIGIO, POLL_IN);
424 }
425 
426 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
427 				    __u64 element, __u64 event,
428 				    struct list_head *obj_list,
429 				    u32 *counter)
430 {
431 	struct ib_uverbs_event *entry;
432 	unsigned long flags;
433 
434 	spin_lock_irqsave(&file->async_file->lock, flags);
435 	if (!file->async_file->file) {
436 		spin_unlock_irqrestore(&file->async_file->lock, flags);
437 		return;
438 	}
439 
440 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
441 	if (!entry) {
442 		spin_unlock_irqrestore(&file->async_file->lock, flags);
443 		return;
444 	}
445 
446 	entry->desc.async.element    = element;
447 	entry->desc.async.event_type = event;
448 	entry->counter               = counter;
449 
450 	list_add_tail(&entry->list, &file->async_file->event_list);
451 	if (obj_list)
452 		list_add_tail(&entry->obj_list, obj_list);
453 	spin_unlock_irqrestore(&file->async_file->lock, flags);
454 
455 	wake_up_interruptible(&file->async_file->poll_wait);
456 	kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
457 }
458 
459 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
460 {
461 	struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
462 						  struct ib_ucq_object, uobject);
463 
464 	ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
465 				event->event, &uobj->async_list,
466 				&uobj->async_events_reported);
467 }
468 
469 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
470 {
471 	struct ib_uevent_object *uobj;
472 
473 	uobj = container_of(event->element.qp->uobject,
474 			    struct ib_uevent_object, uobject);
475 
476 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
477 				event->event, &uobj->event_list,
478 				&uobj->events_reported);
479 }
480 
481 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
482 {
483 	struct ib_uevent_object *uobj;
484 
485 	uobj = container_of(event->element.srq->uobject,
486 			    struct ib_uevent_object, uobject);
487 
488 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
489 				event->event, &uobj->event_list,
490 				&uobj->events_reported);
491 }
492 
493 void ib_uverbs_event_handler(struct ib_event_handler *handler,
494 			     struct ib_event *event)
495 {
496 	struct ib_uverbs_file *file =
497 		container_of(handler, struct ib_uverbs_file, event_handler);
498 
499 	ib_uverbs_async_handler(file, event->element.port_num, event->event,
500 				NULL, NULL);
501 }
502 
503 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
504 					int is_async, int *fd)
505 {
506 	struct ib_uverbs_event_file *ev_file;
507 	struct file *filp;
508 	int ret;
509 
510 	ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
511 	if (!ev_file)
512 		return ERR_PTR(-ENOMEM);
513 
514 	kref_init(&ev_file->ref);
515 	spin_lock_init(&ev_file->lock);
516 	INIT_LIST_HEAD(&ev_file->event_list);
517 	init_waitqueue_head(&ev_file->poll_wait);
518 	ev_file->uverbs_file = uverbs_file;
519 	ev_file->async_queue = NULL;
520 	ev_file->is_async    = is_async;
521 
522 	*fd = get_unused_fd();
523 	if (*fd < 0) {
524 		ret = *fd;
525 		goto err;
526 	}
527 
528 	filp = get_empty_filp();
529 	if (!filp) {
530 		ret = -ENFILE;
531 		goto err_fd;
532 	}
533 
534 	ev_file->file      = filp;
535 
536 	/*
537 	 * fops_get() can't fail here, because we're coming from a
538 	 * system call on a uverbs file, which will already have a
539 	 * module reference.
540 	 */
541 	filp->f_op 	   = fops_get(&uverbs_event_fops);
542 	filp->f_vfsmnt 	   = mntget(uverbs_event_mnt);
543 	filp->f_dentry 	   = dget(uverbs_event_mnt->mnt_root);
544 	filp->f_mapping    = filp->f_dentry->d_inode->i_mapping;
545 	filp->f_flags      = O_RDONLY;
546 	filp->f_mode       = FMODE_READ;
547 	filp->private_data = ev_file;
548 
549 	return filp;
550 
551 err_fd:
552 	put_unused_fd(*fd);
553 
554 err:
555 	kfree(ev_file);
556 	return ERR_PTR(ret);
557 }
558 
559 /*
560  * Look up a completion event file by FD.  If lookup is successful,
561  * takes a ref to the event file struct that it returns; if
562  * unsuccessful, returns NULL.
563  */
564 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
565 {
566 	struct ib_uverbs_event_file *ev_file = NULL;
567 	struct file *filp;
568 
569 	filp = fget(fd);
570 	if (!filp)
571 		return NULL;
572 
573 	if (filp->f_op != &uverbs_event_fops)
574 		goto out;
575 
576 	ev_file = filp->private_data;
577 	if (ev_file->is_async) {
578 		ev_file = NULL;
579 		goto out;
580 	}
581 
582 	kref_get(&ev_file->ref);
583 
584 out:
585 	fput(filp);
586 	return ev_file;
587 }
588 
589 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
590 			     size_t count, loff_t *pos)
591 {
592 	struct ib_uverbs_file *file = filp->private_data;
593 	struct ib_uverbs_cmd_hdr hdr;
594 
595 	if (count < sizeof hdr)
596 		return -EINVAL;
597 
598 	if (copy_from_user(&hdr, buf, sizeof hdr))
599 		return -EFAULT;
600 
601 	if (hdr.in_words * 4 != count)
602 		return -EINVAL;
603 
604 	if (hdr.command < 0				||
605 	    hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
606 	    !uverbs_cmd_table[hdr.command]		||
607 	    !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
608 		return -EINVAL;
609 
610 	if (!file->ucontext &&
611 	    hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
612 		return -EINVAL;
613 
614 	return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
615 					     hdr.in_words * 4, hdr.out_words * 4);
616 }
617 
618 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
619 {
620 	struct ib_uverbs_file *file = filp->private_data;
621 
622 	if (!file->ucontext)
623 		return -ENODEV;
624 	else
625 		return file->device->ib_dev->mmap(file->ucontext, vma);
626 }
627 
628 static int ib_uverbs_open(struct inode *inode, struct file *filp)
629 {
630 	struct ib_uverbs_device *dev;
631 	struct ib_uverbs_file *file;
632 	int ret;
633 
634 	spin_lock(&map_lock);
635 	dev = dev_table[iminor(inode) - IB_UVERBS_BASE_MINOR];
636 	if (dev)
637 		kref_get(&dev->ref);
638 	spin_unlock(&map_lock);
639 
640 	if (!dev)
641 		return -ENXIO;
642 
643 	if (!try_module_get(dev->ib_dev->owner)) {
644 		ret = -ENODEV;
645 		goto err;
646 	}
647 
648 	file = kmalloc(sizeof *file, GFP_KERNEL);
649 	if (!file) {
650 		ret = -ENOMEM;
651 		goto err_module;
652 	}
653 
654 	file->device	 = dev;
655 	file->ucontext	 = NULL;
656 	file->async_file = NULL;
657 	kref_init(&file->ref);
658 	mutex_init(&file->mutex);
659 
660 	filp->private_data = file;
661 
662 	return 0;
663 
664 err_module:
665 	module_put(dev->ib_dev->owner);
666 
667 err:
668 	kref_put(&dev->ref, ib_uverbs_release_dev);
669 
670 	return ret;
671 }
672 
673 static int ib_uverbs_close(struct inode *inode, struct file *filp)
674 {
675 	struct ib_uverbs_file *file = filp->private_data;
676 
677 	ib_uverbs_cleanup_ucontext(file, file->ucontext);
678 
679 	if (file->async_file)
680 		kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
681 
682 	kref_put(&file->ref, ib_uverbs_release_file);
683 
684 	return 0;
685 }
686 
687 static struct file_operations uverbs_fops = {
688 	.owner 	 = THIS_MODULE,
689 	.write 	 = ib_uverbs_write,
690 	.open 	 = ib_uverbs_open,
691 	.release = ib_uverbs_close
692 };
693 
694 static struct file_operations uverbs_mmap_fops = {
695 	.owner 	 = THIS_MODULE,
696 	.write 	 = ib_uverbs_write,
697 	.mmap    = ib_uverbs_mmap,
698 	.open 	 = ib_uverbs_open,
699 	.release = ib_uverbs_close
700 };
701 
702 static struct ib_client uverbs_client = {
703 	.name   = "uverbs",
704 	.add    = ib_uverbs_add_one,
705 	.remove = ib_uverbs_remove_one
706 };
707 
708 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
709 {
710 	struct ib_uverbs_device *dev = class_get_devdata(class_dev);
711 
712 	if (!dev)
713 		return -ENODEV;
714 
715 	return sprintf(buf, "%s\n", dev->ib_dev->name);
716 }
717 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
718 
719 static ssize_t show_dev_abi_version(struct class_device *class_dev, char *buf)
720 {
721 	struct ib_uverbs_device *dev = class_get_devdata(class_dev);
722 
723 	if (!dev)
724 		return -ENODEV;
725 
726 	return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
727 }
728 static CLASS_DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
729 
730 static ssize_t show_abi_version(struct class *class, char *buf)
731 {
732 	return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
733 }
734 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
735 
736 static void ib_uverbs_add_one(struct ib_device *device)
737 {
738 	struct ib_uverbs_device *uverbs_dev;
739 
740 	if (!device->alloc_ucontext)
741 		return;
742 
743 	uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
744 	if (!uverbs_dev)
745 		return;
746 
747 	kref_init(&uverbs_dev->ref);
748 
749 	spin_lock(&map_lock);
750 	uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
751 	if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
752 		spin_unlock(&map_lock);
753 		goto err;
754 	}
755 	set_bit(uverbs_dev->devnum, dev_map);
756 	spin_unlock(&map_lock);
757 
758 	uverbs_dev->ib_dev           = device;
759 	uverbs_dev->num_comp_vectors = 1;
760 
761 	uverbs_dev->dev = cdev_alloc();
762 	if (!uverbs_dev->dev)
763 		goto err;
764 	uverbs_dev->dev->owner = THIS_MODULE;
765 	uverbs_dev->dev->ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
766 	kobject_set_name(&uverbs_dev->dev->kobj, "uverbs%d", uverbs_dev->devnum);
767 	if (cdev_add(uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
768 		goto err_cdev;
769 
770 	uverbs_dev->class_dev = class_device_create(uverbs_class, NULL,
771 						    uverbs_dev->dev->dev,
772 						    device->dma_device,
773 						    "uverbs%d", uverbs_dev->devnum);
774 	if (IS_ERR(uverbs_dev->class_dev))
775 		goto err_cdev;
776 
777 	class_set_devdata(uverbs_dev->class_dev, uverbs_dev);
778 
779 	if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_ibdev))
780 		goto err_class;
781 	if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_abi_version))
782 		goto err_class;
783 
784 	spin_lock(&map_lock);
785 	dev_table[uverbs_dev->devnum] = uverbs_dev;
786 	spin_unlock(&map_lock);
787 
788 	ib_set_client_data(device, &uverbs_client, uverbs_dev);
789 
790 	return;
791 
792 err_class:
793 	class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
794 
795 err_cdev:
796 	cdev_del(uverbs_dev->dev);
797 	clear_bit(uverbs_dev->devnum, dev_map);
798 
799 err:
800 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
801 	return;
802 }
803 
804 static void ib_uverbs_remove_one(struct ib_device *device)
805 {
806 	struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
807 
808 	if (!uverbs_dev)
809 		return;
810 
811 	class_set_devdata(uverbs_dev->class_dev, NULL);
812 	class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
813 	cdev_del(uverbs_dev->dev);
814 
815 	spin_lock(&map_lock);
816 	dev_table[uverbs_dev->devnum] = NULL;
817 	spin_unlock(&map_lock);
818 
819 	clear_bit(uverbs_dev->devnum, dev_map);
820 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
821 }
822 
823 static struct super_block *uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
824 					       const char *dev_name, void *data)
825 {
826 	return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
827 			     INFINIBANDEVENTFS_MAGIC);
828 }
829 
830 static struct file_system_type uverbs_event_fs = {
831 	/* No owner field so module can be unloaded */
832 	.name    = "infinibandeventfs",
833 	.get_sb  = uverbs_event_get_sb,
834 	.kill_sb = kill_litter_super
835 };
836 
837 static int __init ib_uverbs_init(void)
838 {
839 	int ret;
840 
841 	spin_lock_init(&map_lock);
842 
843 	ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
844 				     "infiniband_verbs");
845 	if (ret) {
846 		printk(KERN_ERR "user_verbs: couldn't register device number\n");
847 		goto out;
848 	}
849 
850 	uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
851 	if (IS_ERR(uverbs_class)) {
852 		ret = PTR_ERR(uverbs_class);
853 		printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
854 		goto out_chrdev;
855 	}
856 
857 	ret = class_create_file(uverbs_class, &class_attr_abi_version);
858 	if (ret) {
859 		printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
860 		goto out_class;
861 	}
862 
863 	ret = register_filesystem(&uverbs_event_fs);
864 	if (ret) {
865 		printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
866 		goto out_class;
867 	}
868 
869 	uverbs_event_mnt = kern_mount(&uverbs_event_fs);
870 	if (IS_ERR(uverbs_event_mnt)) {
871 		ret = PTR_ERR(uverbs_event_mnt);
872 		printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
873 		goto out_fs;
874 	}
875 
876 	ret = ib_register_client(&uverbs_client);
877 	if (ret) {
878 		printk(KERN_ERR "user_verbs: couldn't register client\n");
879 		goto out_mnt;
880 	}
881 
882 	return 0;
883 
884 out_mnt:
885 	mntput(uverbs_event_mnt);
886 
887 out_fs:
888 	unregister_filesystem(&uverbs_event_fs);
889 
890 out_class:
891 	class_destroy(uverbs_class);
892 
893 out_chrdev:
894 	unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
895 
896 out:
897 	return ret;
898 }
899 
900 static void __exit ib_uverbs_cleanup(void)
901 {
902 	ib_unregister_client(&uverbs_client);
903 	mntput(uverbs_event_mnt);
904 	unregister_filesystem(&uverbs_event_fs);
905 	class_destroy(uverbs_class);
906 	unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
907 	flush_scheduled_work();
908 	idr_destroy(&ib_uverbs_pd_idr);
909 	idr_destroy(&ib_uverbs_mr_idr);
910 	idr_destroy(&ib_uverbs_mw_idr);
911 	idr_destroy(&ib_uverbs_ah_idr);
912 	idr_destroy(&ib_uverbs_cq_idr);
913 	idr_destroy(&ib_uverbs_qp_idr);
914 	idr_destroy(&ib_uverbs_srq_idr);
915 }
916 
917 module_init(ib_uverbs_init);
918 module_exit(ib_uverbs_cleanup);
919