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/sched.h>
44 #include <linux/file.h>
45 #include <linux/cdev.h>
46 #include <linux/anon_inodes.h>
47 #include <linux/slab.h>
48 
49 #include <linux/uaccess.h>
50 
51 #include <rdma/ib.h>
52 #include <rdma/uverbs_std_types.h>
53 
54 #include "uverbs.h"
55 #include "core_priv.h"
56 #include "rdma_core.h"
57 
58 MODULE_AUTHOR("Roland Dreier");
59 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
60 MODULE_LICENSE("Dual BSD/GPL");
61 
62 enum {
63 	IB_UVERBS_MAJOR       = 231,
64 	IB_UVERBS_BASE_MINOR  = 192,
65 	IB_UVERBS_MAX_DEVICES = RDMA_MAX_PORTS,
66 	IB_UVERBS_NUM_FIXED_MINOR = 32,
67 	IB_UVERBS_NUM_DYNAMIC_MINOR = IB_UVERBS_MAX_DEVICES - IB_UVERBS_NUM_FIXED_MINOR,
68 };
69 
70 #define IB_UVERBS_BASE_DEV	MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
71 
72 static dev_t dynamic_uverbs_dev;
73 static struct class *uverbs_class;
74 
75 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
76 
77 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
78 				     struct ib_device *ib_dev,
79 				     const char __user *buf, int in_len,
80 				     int out_len) = {
81 	[IB_USER_VERBS_CMD_GET_CONTEXT]		= ib_uverbs_get_context,
82 	[IB_USER_VERBS_CMD_QUERY_DEVICE]	= ib_uverbs_query_device,
83 	[IB_USER_VERBS_CMD_QUERY_PORT]		= ib_uverbs_query_port,
84 	[IB_USER_VERBS_CMD_ALLOC_PD]		= ib_uverbs_alloc_pd,
85 	[IB_USER_VERBS_CMD_DEALLOC_PD]		= ib_uverbs_dealloc_pd,
86 	[IB_USER_VERBS_CMD_REG_MR]		= ib_uverbs_reg_mr,
87 	[IB_USER_VERBS_CMD_REREG_MR]		= ib_uverbs_rereg_mr,
88 	[IB_USER_VERBS_CMD_DEREG_MR]		= ib_uverbs_dereg_mr,
89 	[IB_USER_VERBS_CMD_ALLOC_MW]		= ib_uverbs_alloc_mw,
90 	[IB_USER_VERBS_CMD_DEALLOC_MW]		= ib_uverbs_dealloc_mw,
91 	[IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
92 	[IB_USER_VERBS_CMD_CREATE_CQ]		= ib_uverbs_create_cq,
93 	[IB_USER_VERBS_CMD_RESIZE_CQ]		= ib_uverbs_resize_cq,
94 	[IB_USER_VERBS_CMD_POLL_CQ]		= ib_uverbs_poll_cq,
95 	[IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]	= ib_uverbs_req_notify_cq,
96 	[IB_USER_VERBS_CMD_DESTROY_CQ]		= ib_uverbs_destroy_cq,
97 	[IB_USER_VERBS_CMD_CREATE_QP]		= ib_uverbs_create_qp,
98 	[IB_USER_VERBS_CMD_QUERY_QP]		= ib_uverbs_query_qp,
99 	[IB_USER_VERBS_CMD_MODIFY_QP]		= ib_uverbs_modify_qp,
100 	[IB_USER_VERBS_CMD_DESTROY_QP]		= ib_uverbs_destroy_qp,
101 	[IB_USER_VERBS_CMD_POST_SEND]		= ib_uverbs_post_send,
102 	[IB_USER_VERBS_CMD_POST_RECV]		= ib_uverbs_post_recv,
103 	[IB_USER_VERBS_CMD_POST_SRQ_RECV]	= ib_uverbs_post_srq_recv,
104 	[IB_USER_VERBS_CMD_CREATE_AH]		= ib_uverbs_create_ah,
105 	[IB_USER_VERBS_CMD_DESTROY_AH]		= ib_uverbs_destroy_ah,
106 	[IB_USER_VERBS_CMD_ATTACH_MCAST]	= ib_uverbs_attach_mcast,
107 	[IB_USER_VERBS_CMD_DETACH_MCAST]	= ib_uverbs_detach_mcast,
108 	[IB_USER_VERBS_CMD_CREATE_SRQ]		= ib_uverbs_create_srq,
109 	[IB_USER_VERBS_CMD_MODIFY_SRQ]		= ib_uverbs_modify_srq,
110 	[IB_USER_VERBS_CMD_QUERY_SRQ]		= ib_uverbs_query_srq,
111 	[IB_USER_VERBS_CMD_DESTROY_SRQ]		= ib_uverbs_destroy_srq,
112 	[IB_USER_VERBS_CMD_OPEN_XRCD]		= ib_uverbs_open_xrcd,
113 	[IB_USER_VERBS_CMD_CLOSE_XRCD]		= ib_uverbs_close_xrcd,
114 	[IB_USER_VERBS_CMD_CREATE_XSRQ]		= ib_uverbs_create_xsrq,
115 	[IB_USER_VERBS_CMD_OPEN_QP]		= ib_uverbs_open_qp,
116 };
117 
118 static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file,
119 				    struct ib_device *ib_dev,
120 				    struct ib_udata *ucore,
121 				    struct ib_udata *uhw) = {
122 	[IB_USER_VERBS_EX_CMD_CREATE_FLOW]	= ib_uverbs_ex_create_flow,
123 	[IB_USER_VERBS_EX_CMD_DESTROY_FLOW]	= ib_uverbs_ex_destroy_flow,
124 	[IB_USER_VERBS_EX_CMD_QUERY_DEVICE]	= ib_uverbs_ex_query_device,
125 	[IB_USER_VERBS_EX_CMD_CREATE_CQ]	= ib_uverbs_ex_create_cq,
126 	[IB_USER_VERBS_EX_CMD_CREATE_QP]        = ib_uverbs_ex_create_qp,
127 	[IB_USER_VERBS_EX_CMD_CREATE_WQ]        = ib_uverbs_ex_create_wq,
128 	[IB_USER_VERBS_EX_CMD_MODIFY_WQ]        = ib_uverbs_ex_modify_wq,
129 	[IB_USER_VERBS_EX_CMD_DESTROY_WQ]       = ib_uverbs_ex_destroy_wq,
130 	[IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL] = ib_uverbs_ex_create_rwq_ind_table,
131 	[IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL] = ib_uverbs_ex_destroy_rwq_ind_table,
132 	[IB_USER_VERBS_EX_CMD_MODIFY_QP]        = ib_uverbs_ex_modify_qp,
133 	[IB_USER_VERBS_EX_CMD_MODIFY_CQ]        = ib_uverbs_ex_modify_cq,
134 };
135 
136 static void ib_uverbs_add_one(struct ib_device *device);
137 static void ib_uverbs_remove_one(struct ib_device *device, void *client_data);
138 
139 int uverbs_dealloc_mw(struct ib_mw *mw)
140 {
141 	struct ib_pd *pd = mw->pd;
142 	int ret;
143 
144 	ret = mw->device->dealloc_mw(mw);
145 	if (!ret)
146 		atomic_dec(&pd->usecnt);
147 	return ret;
148 }
149 
150 static void ib_uverbs_release_dev(struct kobject *kobj)
151 {
152 	struct ib_uverbs_device *dev =
153 		container_of(kobj, struct ib_uverbs_device, kobj);
154 
155 	cleanup_srcu_struct(&dev->disassociate_srcu);
156 	kfree(dev);
157 }
158 
159 static struct kobj_type ib_uverbs_dev_ktype = {
160 	.release = ib_uverbs_release_dev,
161 };
162 
163 static void ib_uverbs_release_async_event_file(struct kref *ref)
164 {
165 	struct ib_uverbs_async_event_file *file =
166 		container_of(ref, struct ib_uverbs_async_event_file, ref);
167 
168 	kfree(file);
169 }
170 
171 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
172 			  struct ib_uverbs_completion_event_file *ev_file,
173 			  struct ib_ucq_object *uobj)
174 {
175 	struct ib_uverbs_event *evt, *tmp;
176 
177 	if (ev_file) {
178 		spin_lock_irq(&ev_file->ev_queue.lock);
179 		list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
180 			list_del(&evt->list);
181 			kfree(evt);
182 		}
183 		spin_unlock_irq(&ev_file->ev_queue.lock);
184 
185 		uverbs_uobject_put(&ev_file->uobj_file.uobj);
186 	}
187 
188 	spin_lock_irq(&file->async_file->ev_queue.lock);
189 	list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
190 		list_del(&evt->list);
191 		kfree(evt);
192 	}
193 	spin_unlock_irq(&file->async_file->ev_queue.lock);
194 }
195 
196 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
197 			      struct ib_uevent_object *uobj)
198 {
199 	struct ib_uverbs_event *evt, *tmp;
200 
201 	spin_lock_irq(&file->async_file->ev_queue.lock);
202 	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
203 		list_del(&evt->list);
204 		kfree(evt);
205 	}
206 	spin_unlock_irq(&file->async_file->ev_queue.lock);
207 }
208 
209 void ib_uverbs_detach_umcast(struct ib_qp *qp,
210 			     struct ib_uqp_object *uobj)
211 {
212 	struct ib_uverbs_mcast_entry *mcast, *tmp;
213 
214 	list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
215 		ib_detach_mcast(qp, &mcast->gid, mcast->lid);
216 		list_del(&mcast->list);
217 		kfree(mcast);
218 	}
219 }
220 
221 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
222 				      struct ib_ucontext *context,
223 				      bool device_removed)
224 {
225 	context->closing = 1;
226 	uverbs_cleanup_ucontext(context, device_removed);
227 	put_pid(context->tgid);
228 
229 	ib_rdmacg_uncharge(&context->cg_obj, context->device,
230 			   RDMACG_RESOURCE_HCA_HANDLE);
231 
232 	return context->device->dealloc_ucontext(context);
233 }
234 
235 static void ib_uverbs_comp_dev(struct ib_uverbs_device *dev)
236 {
237 	complete(&dev->comp);
238 }
239 
240 void ib_uverbs_release_file(struct kref *ref)
241 {
242 	struct ib_uverbs_file *file =
243 		container_of(ref, struct ib_uverbs_file, ref);
244 	struct ib_device *ib_dev;
245 	int srcu_key;
246 
247 	srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
248 	ib_dev = srcu_dereference(file->device->ib_dev,
249 				  &file->device->disassociate_srcu);
250 	if (ib_dev && !ib_dev->disassociate_ucontext)
251 		module_put(ib_dev->owner);
252 	srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
253 
254 	if (atomic_dec_and_test(&file->device->refcount))
255 		ib_uverbs_comp_dev(file->device);
256 
257 	kobject_put(&file->device->kobj);
258 	kfree(file);
259 }
260 
261 static ssize_t ib_uverbs_event_read(struct ib_uverbs_event_queue *ev_queue,
262 				    struct ib_uverbs_file *uverbs_file,
263 				    struct file *filp, char __user *buf,
264 				    size_t count, loff_t *pos,
265 				    size_t eventsz)
266 {
267 	struct ib_uverbs_event *event;
268 	int ret = 0;
269 
270 	spin_lock_irq(&ev_queue->lock);
271 
272 	while (list_empty(&ev_queue->event_list)) {
273 		spin_unlock_irq(&ev_queue->lock);
274 
275 		if (filp->f_flags & O_NONBLOCK)
276 			return -EAGAIN;
277 
278 		if (wait_event_interruptible(ev_queue->poll_wait,
279 					     (!list_empty(&ev_queue->event_list) ||
280 			/* The barriers built into wait_event_interruptible()
281 			 * and wake_up() guarentee this will see the null set
282 			 * without using RCU
283 			 */
284 					     !uverbs_file->device->ib_dev)))
285 			return -ERESTARTSYS;
286 
287 		/* If device was disassociated and no event exists set an error */
288 		if (list_empty(&ev_queue->event_list) &&
289 		    !uverbs_file->device->ib_dev)
290 			return -EIO;
291 
292 		spin_lock_irq(&ev_queue->lock);
293 	}
294 
295 	event = list_entry(ev_queue->event_list.next, struct ib_uverbs_event, list);
296 
297 	if (eventsz > count) {
298 		ret   = -EINVAL;
299 		event = NULL;
300 	} else {
301 		list_del(ev_queue->event_list.next);
302 		if (event->counter) {
303 			++(*event->counter);
304 			list_del(&event->obj_list);
305 		}
306 	}
307 
308 	spin_unlock_irq(&ev_queue->lock);
309 
310 	if (event) {
311 		if (copy_to_user(buf, event, eventsz))
312 			ret = -EFAULT;
313 		else
314 			ret = eventsz;
315 	}
316 
317 	kfree(event);
318 
319 	return ret;
320 }
321 
322 static ssize_t ib_uverbs_async_event_read(struct file *filp, char __user *buf,
323 					  size_t count, loff_t *pos)
324 {
325 	struct ib_uverbs_async_event_file *file = filp->private_data;
326 
327 	return ib_uverbs_event_read(&file->ev_queue, file->uverbs_file, filp,
328 				    buf, count, pos,
329 				    sizeof(struct ib_uverbs_async_event_desc));
330 }
331 
332 static ssize_t ib_uverbs_comp_event_read(struct file *filp, char __user *buf,
333 					 size_t count, loff_t *pos)
334 {
335 	struct ib_uverbs_completion_event_file *comp_ev_file =
336 		filp->private_data;
337 
338 	return ib_uverbs_event_read(&comp_ev_file->ev_queue,
339 				    comp_ev_file->uobj_file.ufile, filp,
340 				    buf, count, pos,
341 				    sizeof(struct ib_uverbs_comp_event_desc));
342 }
343 
344 static __poll_t ib_uverbs_event_poll(struct ib_uverbs_event_queue *ev_queue,
345 					 struct file *filp,
346 					 struct poll_table_struct *wait)
347 {
348 	__poll_t pollflags = 0;
349 
350 	poll_wait(filp, &ev_queue->poll_wait, wait);
351 
352 	spin_lock_irq(&ev_queue->lock);
353 	if (!list_empty(&ev_queue->event_list))
354 		pollflags = EPOLLIN | EPOLLRDNORM;
355 	spin_unlock_irq(&ev_queue->lock);
356 
357 	return pollflags;
358 }
359 
360 static __poll_t ib_uverbs_async_event_poll(struct file *filp,
361 					       struct poll_table_struct *wait)
362 {
363 	return ib_uverbs_event_poll(filp->private_data, filp, wait);
364 }
365 
366 static __poll_t ib_uverbs_comp_event_poll(struct file *filp,
367 					      struct poll_table_struct *wait)
368 {
369 	struct ib_uverbs_completion_event_file *comp_ev_file =
370 		filp->private_data;
371 
372 	return ib_uverbs_event_poll(&comp_ev_file->ev_queue, filp, wait);
373 }
374 
375 static int ib_uverbs_async_event_fasync(int fd, struct file *filp, int on)
376 {
377 	struct ib_uverbs_event_queue *ev_queue = filp->private_data;
378 
379 	return fasync_helper(fd, filp, on, &ev_queue->async_queue);
380 }
381 
382 static int ib_uverbs_comp_event_fasync(int fd, struct file *filp, int on)
383 {
384 	struct ib_uverbs_completion_event_file *comp_ev_file =
385 		filp->private_data;
386 
387 	return fasync_helper(fd, filp, on, &comp_ev_file->ev_queue.async_queue);
388 }
389 
390 static int ib_uverbs_async_event_close(struct inode *inode, struct file *filp)
391 {
392 	struct ib_uverbs_async_event_file *file = filp->private_data;
393 	struct ib_uverbs_file *uverbs_file = file->uverbs_file;
394 	struct ib_uverbs_event *entry, *tmp;
395 	int closed_already = 0;
396 
397 	mutex_lock(&uverbs_file->device->lists_mutex);
398 	spin_lock_irq(&file->ev_queue.lock);
399 	closed_already = file->ev_queue.is_closed;
400 	file->ev_queue.is_closed = 1;
401 	list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) {
402 		if (entry->counter)
403 			list_del(&entry->obj_list);
404 		kfree(entry);
405 	}
406 	spin_unlock_irq(&file->ev_queue.lock);
407 	if (!closed_already) {
408 		list_del(&file->list);
409 		ib_unregister_event_handler(&uverbs_file->event_handler);
410 	}
411 	mutex_unlock(&uverbs_file->device->lists_mutex);
412 
413 	kref_put(&uverbs_file->ref, ib_uverbs_release_file);
414 	kref_put(&file->ref, ib_uverbs_release_async_event_file);
415 
416 	return 0;
417 }
418 
419 static int ib_uverbs_comp_event_close(struct inode *inode, struct file *filp)
420 {
421 	struct ib_uverbs_completion_event_file *file = filp->private_data;
422 	struct ib_uverbs_event *entry, *tmp;
423 
424 	spin_lock_irq(&file->ev_queue.lock);
425 	list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) {
426 		if (entry->counter)
427 			list_del(&entry->obj_list);
428 		kfree(entry);
429 	}
430 	spin_unlock_irq(&file->ev_queue.lock);
431 
432 	uverbs_close_fd(filp);
433 
434 	return 0;
435 }
436 
437 const struct file_operations uverbs_event_fops = {
438 	.owner	 = THIS_MODULE,
439 	.read	 = ib_uverbs_comp_event_read,
440 	.poll    = ib_uverbs_comp_event_poll,
441 	.release = ib_uverbs_comp_event_close,
442 	.fasync  = ib_uverbs_comp_event_fasync,
443 	.llseek	 = no_llseek,
444 };
445 
446 static const struct file_operations uverbs_async_event_fops = {
447 	.owner	 = THIS_MODULE,
448 	.read	 = ib_uverbs_async_event_read,
449 	.poll    = ib_uverbs_async_event_poll,
450 	.release = ib_uverbs_async_event_close,
451 	.fasync  = ib_uverbs_async_event_fasync,
452 	.llseek	 = no_llseek,
453 };
454 
455 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
456 {
457 	struct ib_uverbs_event_queue   *ev_queue = cq_context;
458 	struct ib_ucq_object	       *uobj;
459 	struct ib_uverbs_event	       *entry;
460 	unsigned long			flags;
461 
462 	if (!ev_queue)
463 		return;
464 
465 	spin_lock_irqsave(&ev_queue->lock, flags);
466 	if (ev_queue->is_closed) {
467 		spin_unlock_irqrestore(&ev_queue->lock, flags);
468 		return;
469 	}
470 
471 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
472 	if (!entry) {
473 		spin_unlock_irqrestore(&ev_queue->lock, flags);
474 		return;
475 	}
476 
477 	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
478 
479 	entry->desc.comp.cq_handle = cq->uobject->user_handle;
480 	entry->counter		   = &uobj->comp_events_reported;
481 
482 	list_add_tail(&entry->list, &ev_queue->event_list);
483 	list_add_tail(&entry->obj_list, &uobj->comp_list);
484 	spin_unlock_irqrestore(&ev_queue->lock, flags);
485 
486 	wake_up_interruptible(&ev_queue->poll_wait);
487 	kill_fasync(&ev_queue->async_queue, SIGIO, POLL_IN);
488 }
489 
490 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
491 				    __u64 element, __u64 event,
492 				    struct list_head *obj_list,
493 				    u32 *counter)
494 {
495 	struct ib_uverbs_event *entry;
496 	unsigned long flags;
497 
498 	spin_lock_irqsave(&file->async_file->ev_queue.lock, flags);
499 	if (file->async_file->ev_queue.is_closed) {
500 		spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
501 		return;
502 	}
503 
504 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
505 	if (!entry) {
506 		spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
507 		return;
508 	}
509 
510 	entry->desc.async.element    = element;
511 	entry->desc.async.event_type = event;
512 	entry->desc.async.reserved   = 0;
513 	entry->counter               = counter;
514 
515 	list_add_tail(&entry->list, &file->async_file->ev_queue.event_list);
516 	if (obj_list)
517 		list_add_tail(&entry->obj_list, obj_list);
518 	spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
519 
520 	wake_up_interruptible(&file->async_file->ev_queue.poll_wait);
521 	kill_fasync(&file->async_file->ev_queue.async_queue, SIGIO, POLL_IN);
522 }
523 
524 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
525 {
526 	struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
527 						  struct ib_ucq_object, uobject);
528 
529 	ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
530 				event->event, &uobj->async_list,
531 				&uobj->async_events_reported);
532 }
533 
534 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
535 {
536 	struct ib_uevent_object *uobj;
537 
538 	/* for XRC target qp's, check that qp is live */
539 	if (!event->element.qp->uobject)
540 		return;
541 
542 	uobj = container_of(event->element.qp->uobject,
543 			    struct ib_uevent_object, uobject);
544 
545 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
546 				event->event, &uobj->event_list,
547 				&uobj->events_reported);
548 }
549 
550 void ib_uverbs_wq_event_handler(struct ib_event *event, void *context_ptr)
551 {
552 	struct ib_uevent_object *uobj = container_of(event->element.wq->uobject,
553 						  struct ib_uevent_object, uobject);
554 
555 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
556 				event->event, &uobj->event_list,
557 				&uobj->events_reported);
558 }
559 
560 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
561 {
562 	struct ib_uevent_object *uobj;
563 
564 	uobj = container_of(event->element.srq->uobject,
565 			    struct ib_uevent_object, uobject);
566 
567 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
568 				event->event, &uobj->event_list,
569 				&uobj->events_reported);
570 }
571 
572 void ib_uverbs_event_handler(struct ib_event_handler *handler,
573 			     struct ib_event *event)
574 {
575 	struct ib_uverbs_file *file =
576 		container_of(handler, struct ib_uverbs_file, event_handler);
577 
578 	ib_uverbs_async_handler(file, event->element.port_num, event->event,
579 				NULL, NULL);
580 }
581 
582 void ib_uverbs_free_async_event_file(struct ib_uverbs_file *file)
583 {
584 	kref_put(&file->async_file->ref, ib_uverbs_release_async_event_file);
585 	file->async_file = NULL;
586 }
587 
588 void ib_uverbs_init_event_queue(struct ib_uverbs_event_queue *ev_queue)
589 {
590 	spin_lock_init(&ev_queue->lock);
591 	INIT_LIST_HEAD(&ev_queue->event_list);
592 	init_waitqueue_head(&ev_queue->poll_wait);
593 	ev_queue->is_closed   = 0;
594 	ev_queue->async_queue = NULL;
595 }
596 
597 struct file *ib_uverbs_alloc_async_event_file(struct ib_uverbs_file *uverbs_file,
598 					      struct ib_device	*ib_dev)
599 {
600 	struct ib_uverbs_async_event_file *ev_file;
601 	struct file *filp;
602 
603 	ev_file = kzalloc(sizeof(*ev_file), GFP_KERNEL);
604 	if (!ev_file)
605 		return ERR_PTR(-ENOMEM);
606 
607 	ib_uverbs_init_event_queue(&ev_file->ev_queue);
608 	ev_file->uverbs_file = uverbs_file;
609 	kref_get(&ev_file->uverbs_file->ref);
610 	kref_init(&ev_file->ref);
611 	filp = anon_inode_getfile("[infinibandevent]", &uverbs_async_event_fops,
612 				  ev_file, O_RDONLY);
613 	if (IS_ERR(filp))
614 		goto err_put_refs;
615 
616 	mutex_lock(&uverbs_file->device->lists_mutex);
617 	list_add_tail(&ev_file->list,
618 		      &uverbs_file->device->uverbs_events_file_list);
619 	mutex_unlock(&uverbs_file->device->lists_mutex);
620 
621 	WARN_ON(uverbs_file->async_file);
622 	uverbs_file->async_file = ev_file;
623 	kref_get(&uverbs_file->async_file->ref);
624 	INIT_IB_EVENT_HANDLER(&uverbs_file->event_handler,
625 			      ib_dev,
626 			      ib_uverbs_event_handler);
627 	ib_register_event_handler(&uverbs_file->event_handler);
628 	/* At that point async file stuff was fully set */
629 
630 	return filp;
631 
632 err_put_refs:
633 	kref_put(&ev_file->uverbs_file->ref, ib_uverbs_release_file);
634 	kref_put(&ev_file->ref, ib_uverbs_release_async_event_file);
635 	return filp;
636 }
637 
638 static int verify_command_mask(struct ib_device *ib_dev, __u32 command)
639 {
640 	u64 mask;
641 
642 	if (command <= IB_USER_VERBS_CMD_OPEN_QP)
643 		mask = ib_dev->uverbs_cmd_mask;
644 	else
645 		mask = ib_dev->uverbs_ex_cmd_mask;
646 
647 	if (mask & ((u64)1 << command))
648 		return 0;
649 
650 	return -1;
651 }
652 
653 static bool verify_command_idx(u32 command, bool extended)
654 {
655 	if (extended)
656 		return command < ARRAY_SIZE(uverbs_ex_cmd_table);
657 
658 	return command < ARRAY_SIZE(uverbs_cmd_table);
659 }
660 
661 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
662 			     size_t count, loff_t *pos)
663 {
664 	struct ib_uverbs_file *file = filp->private_data;
665 	struct ib_device *ib_dev;
666 	struct ib_uverbs_cmd_hdr hdr;
667 	bool extended_command;
668 	__u32 command;
669 	__u32 flags;
670 	int srcu_key;
671 	ssize_t ret;
672 
673 	if (!ib_safe_file_access(filp)) {
674 		pr_err_once("uverbs_write: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
675 			    task_tgid_vnr(current), current->comm);
676 		return -EACCES;
677 	}
678 
679 	if (count < sizeof hdr)
680 		return -EINVAL;
681 
682 	if (copy_from_user(&hdr, buf, sizeof hdr))
683 		return -EFAULT;
684 
685 	srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
686 	ib_dev = srcu_dereference(file->device->ib_dev,
687 				  &file->device->disassociate_srcu);
688 	if (!ib_dev) {
689 		ret = -EIO;
690 		goto out;
691 	}
692 
693 	if (hdr.command & ~(__u32)(IB_USER_VERBS_CMD_FLAGS_MASK |
694 				   IB_USER_VERBS_CMD_COMMAND_MASK)) {
695 		ret = -EINVAL;
696 		goto out;
697 	}
698 
699 	command = hdr.command & IB_USER_VERBS_CMD_COMMAND_MASK;
700 	flags = (hdr.command &
701 		 IB_USER_VERBS_CMD_FLAGS_MASK) >> IB_USER_VERBS_CMD_FLAGS_SHIFT;
702 
703 	extended_command = flags & IB_USER_VERBS_CMD_FLAG_EXTENDED;
704 	if (!verify_command_idx(command, extended_command)) {
705 		ret = -EINVAL;
706 		goto out;
707 	}
708 
709 	if (verify_command_mask(ib_dev, command)) {
710 		ret = -EOPNOTSUPP;
711 		goto out;
712 	}
713 
714 	if (!file->ucontext &&
715 	    command != IB_USER_VERBS_CMD_GET_CONTEXT) {
716 		ret = -EINVAL;
717 		goto out;
718 	}
719 
720 	if (!flags) {
721 		if (!uverbs_cmd_table[command]) {
722 			ret = -EINVAL;
723 			goto out;
724 		}
725 
726 		if (hdr.in_words * 4 != count) {
727 			ret = -EINVAL;
728 			goto out;
729 		}
730 
731 		ret = uverbs_cmd_table[command](file, ib_dev,
732 						 buf + sizeof(hdr),
733 						 hdr.in_words * 4,
734 						 hdr.out_words * 4);
735 
736 	} else if (flags == IB_USER_VERBS_CMD_FLAG_EXTENDED) {
737 		struct ib_uverbs_ex_cmd_hdr ex_hdr;
738 		struct ib_udata ucore;
739 		struct ib_udata uhw;
740 		size_t written_count = count;
741 
742 		if (!uverbs_ex_cmd_table[command]) {
743 			ret = -ENOSYS;
744 			goto out;
745 		}
746 
747 		if (!file->ucontext) {
748 			ret = -EINVAL;
749 			goto out;
750 		}
751 
752 		if (count < (sizeof(hdr) + sizeof(ex_hdr))) {
753 			ret = -EINVAL;
754 			goto out;
755 		}
756 
757 		if (copy_from_user(&ex_hdr, buf + sizeof(hdr), sizeof(ex_hdr))) {
758 			ret = -EFAULT;
759 			goto out;
760 		}
761 
762 		count -= sizeof(hdr) + sizeof(ex_hdr);
763 		buf += sizeof(hdr) + sizeof(ex_hdr);
764 
765 		if ((hdr.in_words + ex_hdr.provider_in_words) * 8 != count) {
766 			ret = -EINVAL;
767 			goto out;
768 		}
769 
770 		if (ex_hdr.cmd_hdr_reserved) {
771 			ret = -EINVAL;
772 			goto out;
773 		}
774 
775 		if (ex_hdr.response) {
776 			if (!hdr.out_words && !ex_hdr.provider_out_words) {
777 				ret = -EINVAL;
778 				goto out;
779 			}
780 
781 			if (!access_ok(VERIFY_WRITE,
782 				       u64_to_user_ptr(ex_hdr.response),
783 				       (hdr.out_words + ex_hdr.provider_out_words) * 8)) {
784 				ret = -EFAULT;
785 				goto out;
786 			}
787 		} else {
788 			if (hdr.out_words || ex_hdr.provider_out_words) {
789 				ret = -EINVAL;
790 				goto out;
791 			}
792 		}
793 
794 		ib_uverbs_init_udata_buf_or_null(&ucore, buf,
795 					u64_to_user_ptr(ex_hdr.response),
796 					hdr.in_words * 8, hdr.out_words * 8);
797 
798 		ib_uverbs_init_udata_buf_or_null(&uhw,
799 					buf + ucore.inlen,
800 					u64_to_user_ptr(ex_hdr.response) + ucore.outlen,
801 					ex_hdr.provider_in_words * 8,
802 					ex_hdr.provider_out_words * 8);
803 
804 		ret = uverbs_ex_cmd_table[command](file, ib_dev, &ucore, &uhw);
805 		if (!ret)
806 			ret = written_count;
807 	} else {
808 		ret = -ENOSYS;
809 	}
810 
811 out:
812 	srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
813 	return ret;
814 }
815 
816 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
817 {
818 	struct ib_uverbs_file *file = filp->private_data;
819 	struct ib_device *ib_dev;
820 	int ret = 0;
821 	int srcu_key;
822 
823 	srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
824 	ib_dev = srcu_dereference(file->device->ib_dev,
825 				  &file->device->disassociate_srcu);
826 	if (!ib_dev) {
827 		ret = -EIO;
828 		goto out;
829 	}
830 
831 	if (!file->ucontext)
832 		ret = -ENODEV;
833 	else
834 		ret = ib_dev->mmap(file->ucontext, vma);
835 out:
836 	srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
837 	return ret;
838 }
839 
840 /*
841  * ib_uverbs_open() does not need the BKL:
842  *
843  *  - the ib_uverbs_device structures are properly reference counted and
844  *    everything else is purely local to the file being created, so
845  *    races against other open calls are not a problem;
846  *  - there is no ioctl method to race against;
847  *  - the open method will either immediately run -ENXIO, or all
848  *    required initialization will be done.
849  */
850 static int ib_uverbs_open(struct inode *inode, struct file *filp)
851 {
852 	struct ib_uverbs_device *dev;
853 	struct ib_uverbs_file *file;
854 	struct ib_device *ib_dev;
855 	int ret;
856 	int module_dependent;
857 	int srcu_key;
858 
859 	dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
860 	if (!atomic_inc_not_zero(&dev->refcount))
861 		return -ENXIO;
862 
863 	srcu_key = srcu_read_lock(&dev->disassociate_srcu);
864 	mutex_lock(&dev->lists_mutex);
865 	ib_dev = srcu_dereference(dev->ib_dev,
866 				  &dev->disassociate_srcu);
867 	if (!ib_dev) {
868 		ret = -EIO;
869 		goto err;
870 	}
871 
872 	/* In case IB device supports disassociate ucontext, there is no hard
873 	 * dependency between uverbs device and its low level device.
874 	 */
875 	module_dependent = !(ib_dev->disassociate_ucontext);
876 
877 	if (module_dependent) {
878 		if (!try_module_get(ib_dev->owner)) {
879 			ret = -ENODEV;
880 			goto err;
881 		}
882 	}
883 
884 	file = kzalloc(sizeof(*file), GFP_KERNEL);
885 	if (!file) {
886 		ret = -ENOMEM;
887 		if (module_dependent)
888 			goto err_module;
889 
890 		goto err;
891 	}
892 
893 	file->device	 = dev;
894 	spin_lock_init(&file->idr_lock);
895 	idr_init(&file->idr);
896 	file->ucontext	 = NULL;
897 	file->async_file = NULL;
898 	kref_init(&file->ref);
899 	mutex_init(&file->mutex);
900 	mutex_init(&file->cleanup_mutex);
901 
902 	filp->private_data = file;
903 	kobject_get(&dev->kobj);
904 	list_add_tail(&file->list, &dev->uverbs_file_list);
905 	mutex_unlock(&dev->lists_mutex);
906 	srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
907 
908 	return nonseekable_open(inode, filp);
909 
910 err_module:
911 	module_put(ib_dev->owner);
912 
913 err:
914 	mutex_unlock(&dev->lists_mutex);
915 	srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
916 	if (atomic_dec_and_test(&dev->refcount))
917 		ib_uverbs_comp_dev(dev);
918 
919 	return ret;
920 }
921 
922 static int ib_uverbs_close(struct inode *inode, struct file *filp)
923 {
924 	struct ib_uverbs_file *file = filp->private_data;
925 
926 	mutex_lock(&file->cleanup_mutex);
927 	if (file->ucontext) {
928 		ib_uverbs_cleanup_ucontext(file, file->ucontext, false);
929 		file->ucontext = NULL;
930 	}
931 	mutex_unlock(&file->cleanup_mutex);
932 	idr_destroy(&file->idr);
933 
934 	mutex_lock(&file->device->lists_mutex);
935 	if (!file->is_closed) {
936 		list_del(&file->list);
937 		file->is_closed = 1;
938 	}
939 	mutex_unlock(&file->device->lists_mutex);
940 
941 	if (file->async_file)
942 		kref_put(&file->async_file->ref,
943 			 ib_uverbs_release_async_event_file);
944 
945 	kref_put(&file->ref, ib_uverbs_release_file);
946 
947 	return 0;
948 }
949 
950 static const struct file_operations uverbs_fops = {
951 	.owner	 = THIS_MODULE,
952 	.write	 = ib_uverbs_write,
953 	.open	 = ib_uverbs_open,
954 	.release = ib_uverbs_close,
955 	.llseek	 = no_llseek,
956 #if IS_ENABLED(CONFIG_INFINIBAND_EXP_USER_ACCESS)
957 	.unlocked_ioctl = ib_uverbs_ioctl,
958 	.compat_ioctl = ib_uverbs_ioctl,
959 #endif
960 };
961 
962 static const struct file_operations uverbs_mmap_fops = {
963 	.owner	 = THIS_MODULE,
964 	.write	 = ib_uverbs_write,
965 	.mmap    = ib_uverbs_mmap,
966 	.open	 = ib_uverbs_open,
967 	.release = ib_uverbs_close,
968 	.llseek	 = no_llseek,
969 #if IS_ENABLED(CONFIG_INFINIBAND_EXP_USER_ACCESS)
970 	.unlocked_ioctl = ib_uverbs_ioctl,
971 	.compat_ioctl = ib_uverbs_ioctl,
972 #endif
973 };
974 
975 static struct ib_client uverbs_client = {
976 	.name   = "uverbs",
977 	.add    = ib_uverbs_add_one,
978 	.remove = ib_uverbs_remove_one
979 };
980 
981 static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
982 			  char *buf)
983 {
984 	int ret = -ENODEV;
985 	int srcu_key;
986 	struct ib_uverbs_device *dev = dev_get_drvdata(device);
987 	struct ib_device *ib_dev;
988 
989 	if (!dev)
990 		return -ENODEV;
991 
992 	srcu_key = srcu_read_lock(&dev->disassociate_srcu);
993 	ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu);
994 	if (ib_dev)
995 		ret = sprintf(buf, "%s\n", ib_dev->name);
996 	srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
997 
998 	return ret;
999 }
1000 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1001 
1002 static ssize_t show_dev_abi_version(struct device *device,
1003 				    struct device_attribute *attr, char *buf)
1004 {
1005 	struct ib_uverbs_device *dev = dev_get_drvdata(device);
1006 	int ret = -ENODEV;
1007 	int srcu_key;
1008 	struct ib_device *ib_dev;
1009 
1010 	if (!dev)
1011 		return -ENODEV;
1012 	srcu_key = srcu_read_lock(&dev->disassociate_srcu);
1013 	ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu);
1014 	if (ib_dev)
1015 		ret = sprintf(buf, "%d\n", ib_dev->uverbs_abi_ver);
1016 	srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
1017 
1018 	return ret;
1019 }
1020 static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
1021 
1022 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1023 			 __stringify(IB_USER_VERBS_ABI_VERSION));
1024 
1025 static void ib_uverbs_add_one(struct ib_device *device)
1026 {
1027 	int devnum;
1028 	dev_t base;
1029 	struct ib_uverbs_device *uverbs_dev;
1030 	int ret;
1031 
1032 	if (!device->alloc_ucontext)
1033 		return;
1034 
1035 	uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
1036 	if (!uverbs_dev)
1037 		return;
1038 
1039 	ret = init_srcu_struct(&uverbs_dev->disassociate_srcu);
1040 	if (ret) {
1041 		kfree(uverbs_dev);
1042 		return;
1043 	}
1044 
1045 	atomic_set(&uverbs_dev->refcount, 1);
1046 	init_completion(&uverbs_dev->comp);
1047 	uverbs_dev->xrcd_tree = RB_ROOT;
1048 	mutex_init(&uverbs_dev->xrcd_tree_mutex);
1049 	kobject_init(&uverbs_dev->kobj, &ib_uverbs_dev_ktype);
1050 	mutex_init(&uverbs_dev->lists_mutex);
1051 	INIT_LIST_HEAD(&uverbs_dev->uverbs_file_list);
1052 	INIT_LIST_HEAD(&uverbs_dev->uverbs_events_file_list);
1053 
1054 	devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
1055 	if (devnum >= IB_UVERBS_MAX_DEVICES)
1056 		goto err;
1057 	uverbs_dev->devnum = devnum;
1058 	set_bit(devnum, dev_map);
1059 	if (devnum >= IB_UVERBS_NUM_FIXED_MINOR)
1060 		base = dynamic_uverbs_dev + devnum - IB_UVERBS_NUM_FIXED_MINOR;
1061 	else
1062 		base = IB_UVERBS_BASE_DEV + devnum;
1063 
1064 	rcu_assign_pointer(uverbs_dev->ib_dev, device);
1065 	uverbs_dev->num_comp_vectors = device->num_comp_vectors;
1066 
1067 	cdev_init(&uverbs_dev->cdev, NULL);
1068 	uverbs_dev->cdev.owner = THIS_MODULE;
1069 	uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
1070 	cdev_set_parent(&uverbs_dev->cdev, &uverbs_dev->kobj);
1071 	kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
1072 	if (cdev_add(&uverbs_dev->cdev, base, 1))
1073 		goto err_cdev;
1074 
1075 	uverbs_dev->dev = device_create(uverbs_class, device->dev.parent,
1076 					uverbs_dev->cdev.dev, uverbs_dev,
1077 					"uverbs%d", uverbs_dev->devnum);
1078 	if (IS_ERR(uverbs_dev->dev))
1079 		goto err_cdev;
1080 
1081 	if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
1082 		goto err_class;
1083 	if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
1084 		goto err_class;
1085 
1086 	if (!device->specs_root) {
1087 		const struct uverbs_object_tree_def *default_root[] = {
1088 			uverbs_default_get_objects()};
1089 
1090 		uverbs_dev->specs_root = uverbs_alloc_spec_tree(1,
1091 								default_root);
1092 		if (IS_ERR(uverbs_dev->specs_root))
1093 			goto err_class;
1094 
1095 		device->specs_root = uverbs_dev->specs_root;
1096 	}
1097 
1098 	ib_set_client_data(device, &uverbs_client, uverbs_dev);
1099 
1100 	return;
1101 
1102 err_class:
1103 	device_destroy(uverbs_class, uverbs_dev->cdev.dev);
1104 
1105 err_cdev:
1106 	cdev_del(&uverbs_dev->cdev);
1107 	clear_bit(devnum, dev_map);
1108 
1109 err:
1110 	if (atomic_dec_and_test(&uverbs_dev->refcount))
1111 		ib_uverbs_comp_dev(uverbs_dev);
1112 	wait_for_completion(&uverbs_dev->comp);
1113 	kobject_put(&uverbs_dev->kobj);
1114 	return;
1115 }
1116 
1117 static void ib_uverbs_free_hw_resources(struct ib_uverbs_device *uverbs_dev,
1118 					struct ib_device *ib_dev)
1119 {
1120 	struct ib_uverbs_file *file;
1121 	struct ib_uverbs_async_event_file *event_file;
1122 	struct ib_event event;
1123 
1124 	/* Pending running commands to terminate */
1125 	synchronize_srcu(&uverbs_dev->disassociate_srcu);
1126 	event.event = IB_EVENT_DEVICE_FATAL;
1127 	event.element.port_num = 0;
1128 	event.device = ib_dev;
1129 
1130 	mutex_lock(&uverbs_dev->lists_mutex);
1131 	while (!list_empty(&uverbs_dev->uverbs_file_list)) {
1132 		struct ib_ucontext *ucontext;
1133 		file = list_first_entry(&uverbs_dev->uverbs_file_list,
1134 					struct ib_uverbs_file, list);
1135 		file->is_closed = 1;
1136 		list_del(&file->list);
1137 		kref_get(&file->ref);
1138 		mutex_unlock(&uverbs_dev->lists_mutex);
1139 
1140 
1141 		mutex_lock(&file->cleanup_mutex);
1142 		ucontext = file->ucontext;
1143 		file->ucontext = NULL;
1144 		mutex_unlock(&file->cleanup_mutex);
1145 
1146 		/* At this point ib_uverbs_close cannot be running
1147 		 * ib_uverbs_cleanup_ucontext
1148 		 */
1149 		if (ucontext) {
1150 			/* We must release the mutex before going ahead and
1151 			 * calling disassociate_ucontext. disassociate_ucontext
1152 			 * might end up indirectly calling uverbs_close,
1153 			 * for example due to freeing the resources
1154 			 * (e.g mmput).
1155 			 */
1156 			ib_uverbs_event_handler(&file->event_handler, &event);
1157 			ib_dev->disassociate_ucontext(ucontext);
1158 			mutex_lock(&file->cleanup_mutex);
1159 			ib_uverbs_cleanup_ucontext(file, ucontext, true);
1160 			mutex_unlock(&file->cleanup_mutex);
1161 		}
1162 
1163 		mutex_lock(&uverbs_dev->lists_mutex);
1164 		kref_put(&file->ref, ib_uverbs_release_file);
1165 	}
1166 
1167 	while (!list_empty(&uverbs_dev->uverbs_events_file_list)) {
1168 		event_file = list_first_entry(&uverbs_dev->
1169 					      uverbs_events_file_list,
1170 					      struct ib_uverbs_async_event_file,
1171 					      list);
1172 		spin_lock_irq(&event_file->ev_queue.lock);
1173 		event_file->ev_queue.is_closed = 1;
1174 		spin_unlock_irq(&event_file->ev_queue.lock);
1175 
1176 		list_del(&event_file->list);
1177 		ib_unregister_event_handler(
1178 			&event_file->uverbs_file->event_handler);
1179 		event_file->uverbs_file->event_handler.device =
1180 			NULL;
1181 
1182 		wake_up_interruptible(&event_file->ev_queue.poll_wait);
1183 		kill_fasync(&event_file->ev_queue.async_queue, SIGIO, POLL_IN);
1184 	}
1185 	mutex_unlock(&uverbs_dev->lists_mutex);
1186 }
1187 
1188 static void ib_uverbs_remove_one(struct ib_device *device, void *client_data)
1189 {
1190 	struct ib_uverbs_device *uverbs_dev = client_data;
1191 	int wait_clients = 1;
1192 
1193 	if (!uverbs_dev)
1194 		return;
1195 
1196 	dev_set_drvdata(uverbs_dev->dev, NULL);
1197 	device_destroy(uverbs_class, uverbs_dev->cdev.dev);
1198 	cdev_del(&uverbs_dev->cdev);
1199 	clear_bit(uverbs_dev->devnum, dev_map);
1200 
1201 	if (device->disassociate_ucontext) {
1202 		/* We disassociate HW resources and immediately return.
1203 		 * Userspace will see a EIO errno for all future access.
1204 		 * Upon returning, ib_device may be freed internally and is not
1205 		 * valid any more.
1206 		 * uverbs_device is still available until all clients close
1207 		 * their files, then the uverbs device ref count will be zero
1208 		 * and its resources will be freed.
1209 		 * Note: At this point no more files can be opened since the
1210 		 * cdev was deleted, however active clients can still issue
1211 		 * commands and close their open files.
1212 		 */
1213 		rcu_assign_pointer(uverbs_dev->ib_dev, NULL);
1214 		ib_uverbs_free_hw_resources(uverbs_dev, device);
1215 		wait_clients = 0;
1216 	}
1217 
1218 	if (atomic_dec_and_test(&uverbs_dev->refcount))
1219 		ib_uverbs_comp_dev(uverbs_dev);
1220 	if (wait_clients)
1221 		wait_for_completion(&uverbs_dev->comp);
1222 	if (uverbs_dev->specs_root) {
1223 		uverbs_free_spec_tree(uverbs_dev->specs_root);
1224 		device->specs_root = NULL;
1225 	}
1226 
1227 	kobject_put(&uverbs_dev->kobj);
1228 }
1229 
1230 static char *uverbs_devnode(struct device *dev, umode_t *mode)
1231 {
1232 	if (mode)
1233 		*mode = 0666;
1234 	return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1235 }
1236 
1237 static int __init ib_uverbs_init(void)
1238 {
1239 	int ret;
1240 
1241 	ret = register_chrdev_region(IB_UVERBS_BASE_DEV,
1242 				     IB_UVERBS_NUM_FIXED_MINOR,
1243 				     "infiniband_verbs");
1244 	if (ret) {
1245 		pr_err("user_verbs: couldn't register device number\n");
1246 		goto out;
1247 	}
1248 
1249 	ret = alloc_chrdev_region(&dynamic_uverbs_dev, 0,
1250 				  IB_UVERBS_NUM_DYNAMIC_MINOR,
1251 				  "infiniband_verbs");
1252 	if (ret) {
1253 		pr_err("couldn't register dynamic device number\n");
1254 		goto out_alloc;
1255 	}
1256 
1257 	uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
1258 	if (IS_ERR(uverbs_class)) {
1259 		ret = PTR_ERR(uverbs_class);
1260 		pr_err("user_verbs: couldn't create class infiniband_verbs\n");
1261 		goto out_chrdev;
1262 	}
1263 
1264 	uverbs_class->devnode = uverbs_devnode;
1265 
1266 	ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
1267 	if (ret) {
1268 		pr_err("user_verbs: couldn't create abi_version attribute\n");
1269 		goto out_class;
1270 	}
1271 
1272 	ret = ib_register_client(&uverbs_client);
1273 	if (ret) {
1274 		pr_err("user_verbs: couldn't register client\n");
1275 		goto out_class;
1276 	}
1277 
1278 	return 0;
1279 
1280 out_class:
1281 	class_destroy(uverbs_class);
1282 
1283 out_chrdev:
1284 	unregister_chrdev_region(dynamic_uverbs_dev,
1285 				 IB_UVERBS_NUM_DYNAMIC_MINOR);
1286 
1287 out_alloc:
1288 	unregister_chrdev_region(IB_UVERBS_BASE_DEV,
1289 				 IB_UVERBS_NUM_FIXED_MINOR);
1290 
1291 out:
1292 	return ret;
1293 }
1294 
1295 static void __exit ib_uverbs_cleanup(void)
1296 {
1297 	ib_unregister_client(&uverbs_client);
1298 	class_destroy(uverbs_class);
1299 	unregister_chrdev_region(IB_UVERBS_BASE_DEV,
1300 				 IB_UVERBS_NUM_FIXED_MINOR);
1301 	unregister_chrdev_region(dynamic_uverbs_dev,
1302 				 IB_UVERBS_NUM_DYNAMIC_MINOR);
1303 }
1304 
1305 module_init(ib_uverbs_init);
1306 module_exit(ib_uverbs_cleanup);
1307