xref: /linux/drivers/infiniband/core/uverbs_main.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
1 /*
2  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 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_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_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_DESTROY_SRQ]   	= ib_uverbs_destroy_srq,
110 };
111 
112 static struct vfsmount *uverbs_event_mnt;
113 
114 static void ib_uverbs_add_one(struct ib_device *device);
115 static void ib_uverbs_remove_one(struct ib_device *device);
116 
117 static void ib_uverbs_release_dev(struct kref *ref)
118 {
119 	struct ib_uverbs_device *dev =
120 		container_of(ref, struct ib_uverbs_device, ref);
121 
122 	kfree(dev);
123 }
124 
125 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
126 			  struct ib_uverbs_event_file *ev_file,
127 			  struct ib_ucq_object *uobj)
128 {
129 	struct ib_uverbs_event *evt, *tmp;
130 
131 	if (ev_file) {
132 		spin_lock_irq(&ev_file->lock);
133 		list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
134 			list_del(&evt->list);
135 			kfree(evt);
136 		}
137 		spin_unlock_irq(&ev_file->lock);
138 
139 		kref_put(&ev_file->ref, ib_uverbs_release_event_file);
140 	}
141 
142 	spin_lock_irq(&file->async_file->lock);
143 	list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
144 		list_del(&evt->list);
145 		kfree(evt);
146 	}
147 	spin_unlock_irq(&file->async_file->lock);
148 }
149 
150 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
151 			      struct ib_uevent_object *uobj)
152 {
153 	struct ib_uverbs_event *evt, *tmp;
154 
155 	spin_lock_irq(&file->async_file->lock);
156 	list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
157 		list_del(&evt->list);
158 		kfree(evt);
159 	}
160 	spin_unlock_irq(&file->async_file->lock);
161 }
162 
163 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
164 				    struct ib_uqp_object *uobj)
165 {
166 	struct ib_uverbs_mcast_entry *mcast, *tmp;
167 
168 	list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
169 		ib_detach_mcast(qp, &mcast->gid, mcast->lid);
170 		list_del(&mcast->list);
171 		kfree(mcast);
172 	}
173 }
174 
175 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
176 				      struct ib_ucontext *context)
177 {
178 	struct ib_uobject *uobj, *tmp;
179 
180 	if (!context)
181 		return 0;
182 
183 	mutex_lock(&ib_uverbs_idr_mutex);
184 
185 	list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
186 		struct ib_ah *ah = idr_find(&ib_uverbs_ah_idr, uobj->id);
187 		idr_remove(&ib_uverbs_ah_idr, uobj->id);
188 		ib_destroy_ah(ah);
189 		list_del(&uobj->list);
190 		kfree(uobj);
191 	}
192 
193 	list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
194 		struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
195 		struct ib_uqp_object *uqp =
196 			container_of(uobj, struct ib_uqp_object, uevent.uobject);
197 		idr_remove(&ib_uverbs_qp_idr, uobj->id);
198 		ib_uverbs_detach_umcast(qp, uqp);
199 		ib_destroy_qp(qp);
200 		list_del(&uobj->list);
201 		ib_uverbs_release_uevent(file, &uqp->uevent);
202 		kfree(uqp);
203 	}
204 
205 	list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
206 		struct ib_cq *cq = idr_find(&ib_uverbs_cq_idr, uobj->id);
207 		struct ib_uverbs_event_file *ev_file = cq->cq_context;
208 		struct ib_ucq_object *ucq =
209 			container_of(uobj, struct ib_ucq_object, uobject);
210 		idr_remove(&ib_uverbs_cq_idr, uobj->id);
211 		ib_destroy_cq(cq);
212 		list_del(&uobj->list);
213 		ib_uverbs_release_ucq(file, ev_file, ucq);
214 		kfree(ucq);
215 	}
216 
217 	list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
218 		struct ib_srq *srq = idr_find(&ib_uverbs_srq_idr, uobj->id);
219 		struct ib_uevent_object *uevent =
220 			container_of(uobj, struct ib_uevent_object, uobject);
221 		idr_remove(&ib_uverbs_srq_idr, uobj->id);
222 		ib_destroy_srq(srq);
223 		list_del(&uobj->list);
224 		ib_uverbs_release_uevent(file, uevent);
225 		kfree(uevent);
226 	}
227 
228 	/* XXX Free MWs */
229 
230 	list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
231 		struct ib_mr *mr = idr_find(&ib_uverbs_mr_idr, uobj->id);
232 		struct ib_device *mrdev = mr->device;
233 		struct ib_umem_object *memobj;
234 
235 		idr_remove(&ib_uverbs_mr_idr, uobj->id);
236 		ib_dereg_mr(mr);
237 
238 		memobj = container_of(uobj, struct ib_umem_object, uobject);
239 		ib_umem_release_on_close(mrdev, &memobj->umem);
240 
241 		list_del(&uobj->list);
242 		kfree(memobj);
243 	}
244 
245 	list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
246 		struct ib_pd *pd = idr_find(&ib_uverbs_pd_idr, uobj->id);
247 		idr_remove(&ib_uverbs_pd_idr, uobj->id);
248 		ib_dealloc_pd(pd);
249 		list_del(&uobj->list);
250 		kfree(uobj);
251 	}
252 
253 	mutex_unlock(&ib_uverbs_idr_mutex);
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 void ib_uverbs_release_event_file(struct kref *ref)
341 {
342 	struct ib_uverbs_event_file *file =
343 		container_of(ref, struct ib_uverbs_event_file, ref);
344 
345 	kfree(file);
346 }
347 
348 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
349 {
350 	struct ib_uverbs_event_file *file = filp->private_data;
351 
352 	return fasync_helper(fd, filp, on, &file->async_queue);
353 }
354 
355 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
356 {
357 	struct ib_uverbs_event_file *file = filp->private_data;
358 	struct ib_uverbs_event *entry, *tmp;
359 
360 	spin_lock_irq(&file->lock);
361 	file->file = NULL;
362 	list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
363 		if (entry->counter)
364 			list_del(&entry->obj_list);
365 		kfree(entry);
366 	}
367 	spin_unlock_irq(&file->lock);
368 
369 	ib_uverbs_event_fasync(-1, filp, 0);
370 
371 	if (file->is_async) {
372 		ib_unregister_event_handler(&file->uverbs_file->event_handler);
373 		kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
374 	}
375 	kref_put(&file->ref, ib_uverbs_release_event_file);
376 
377 	return 0;
378 }
379 
380 static struct file_operations uverbs_event_fops = {
381 	.owner	 = THIS_MODULE,
382 	.read 	 = ib_uverbs_event_read,
383 	.poll    = ib_uverbs_event_poll,
384 	.release = ib_uverbs_event_close,
385 	.fasync  = ib_uverbs_event_fasync
386 };
387 
388 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
389 {
390 	struct ib_uverbs_event_file    *file = cq_context;
391 	struct ib_ucq_object	       *uobj;
392 	struct ib_uverbs_event	       *entry;
393 	unsigned long			flags;
394 
395 	if (!file)
396 		return;
397 
398 	spin_lock_irqsave(&file->lock, flags);
399 	if (!file->file) {
400 		spin_unlock_irqrestore(&file->lock, flags);
401 		return;
402 	}
403 
404 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
405 	if (!entry) {
406 		spin_unlock_irqrestore(&file->lock, flags);
407 		return;
408 	}
409 
410 	uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
411 
412 	entry->desc.comp.cq_handle = cq->uobject->user_handle;
413 	entry->counter		   = &uobj->comp_events_reported;
414 
415 	list_add_tail(&entry->list, &file->event_list);
416 	list_add_tail(&entry->obj_list, &uobj->comp_list);
417 	spin_unlock_irqrestore(&file->lock, flags);
418 
419 	wake_up_interruptible(&file->poll_wait);
420 	kill_fasync(&file->async_queue, SIGIO, POLL_IN);
421 }
422 
423 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
424 				    __u64 element, __u64 event,
425 				    struct list_head *obj_list,
426 				    u32 *counter)
427 {
428 	struct ib_uverbs_event *entry;
429 	unsigned long flags;
430 
431 	spin_lock_irqsave(&file->async_file->lock, flags);
432 	if (!file->async_file->file) {
433 		spin_unlock_irqrestore(&file->async_file->lock, flags);
434 		return;
435 	}
436 
437 	entry = kmalloc(sizeof *entry, GFP_ATOMIC);
438 	if (!entry) {
439 		spin_unlock_irqrestore(&file->async_file->lock, flags);
440 		return;
441 	}
442 
443 	entry->desc.async.element    = element;
444 	entry->desc.async.event_type = event;
445 	entry->counter               = counter;
446 
447 	list_add_tail(&entry->list, &file->async_file->event_list);
448 	if (obj_list)
449 		list_add_tail(&entry->obj_list, obj_list);
450 	spin_unlock_irqrestore(&file->async_file->lock, flags);
451 
452 	wake_up_interruptible(&file->async_file->poll_wait);
453 	kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
454 }
455 
456 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
457 {
458 	struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
459 						  struct ib_ucq_object, uobject);
460 
461 	ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
462 				event->event, &uobj->async_list,
463 				&uobj->async_events_reported);
464 
465 }
466 
467 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
468 {
469 	struct ib_uevent_object *uobj;
470 
471 	uobj = container_of(event->element.qp->uobject,
472 			    struct ib_uevent_object, uobject);
473 
474 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
475 				event->event, &uobj->event_list,
476 				&uobj->events_reported);
477 }
478 
479 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
480 {
481 	struct ib_uevent_object *uobj;
482 
483 	uobj = container_of(event->element.srq->uobject,
484 			    struct ib_uevent_object, uobject);
485 
486 	ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
487 				event->event, &uobj->event_list,
488 				&uobj->events_reported);
489 }
490 
491 void ib_uverbs_event_handler(struct ib_event_handler *handler,
492 			     struct ib_event *event)
493 {
494 	struct ib_uverbs_file *file =
495 		container_of(handler, struct ib_uverbs_file, event_handler);
496 
497 	ib_uverbs_async_handler(file, event->element.port_num, event->event,
498 				NULL, NULL);
499 }
500 
501 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
502 					int is_async, int *fd)
503 {
504 	struct ib_uverbs_event_file *ev_file;
505 	struct file *filp;
506 	int ret;
507 
508 	ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
509 	if (!ev_file)
510 		return ERR_PTR(-ENOMEM);
511 
512 	kref_init(&ev_file->ref);
513 	spin_lock_init(&ev_file->lock);
514 	INIT_LIST_HEAD(&ev_file->event_list);
515 	init_waitqueue_head(&ev_file->poll_wait);
516 	ev_file->uverbs_file = uverbs_file;
517 	ev_file->async_queue = NULL;
518 	ev_file->is_async    = is_async;
519 
520 	*fd = get_unused_fd();
521 	if (*fd < 0) {
522 		ret = *fd;
523 		goto err;
524 	}
525 
526 	filp = get_empty_filp();
527 	if (!filp) {
528 		ret = -ENFILE;
529 		goto err_fd;
530 	}
531 
532 	ev_file->file      = filp;
533 
534 	/*
535 	 * fops_get() can't fail here, because we're coming from a
536 	 * system call on a uverbs file, which will already have a
537 	 * module reference.
538 	 */
539 	filp->f_op 	   = fops_get(&uverbs_event_fops);
540 	filp->f_vfsmnt 	   = mntget(uverbs_event_mnt);
541 	filp->f_dentry 	   = dget(uverbs_event_mnt->mnt_root);
542 	filp->f_mapping    = filp->f_dentry->d_inode->i_mapping;
543 	filp->f_flags      = O_RDONLY;
544 	filp->f_mode       = FMODE_READ;
545 	filp->private_data = ev_file;
546 
547 	return filp;
548 
549 err_fd:
550 	put_unused_fd(*fd);
551 
552 err:
553 	kfree(ev_file);
554 	return ERR_PTR(ret);
555 }
556 
557 /*
558  * Look up a completion event file by FD.  If lookup is successful,
559  * takes a ref to the event file struct that it returns; if
560  * unsuccessful, returns NULL.
561  */
562 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
563 {
564 	struct ib_uverbs_event_file *ev_file = NULL;
565 	struct file *filp;
566 
567 	filp = fget(fd);
568 	if (!filp)
569 		return NULL;
570 
571 	if (filp->f_op != &uverbs_event_fops)
572 		goto out;
573 
574 	ev_file = filp->private_data;
575 	if (ev_file->is_async) {
576 		ev_file = NULL;
577 		goto out;
578 	}
579 
580 	kref_get(&ev_file->ref);
581 
582 out:
583 	fput(filp);
584 	return ev_file;
585 }
586 
587 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
588 			     size_t count, loff_t *pos)
589 {
590 	struct ib_uverbs_file *file = filp->private_data;
591 	struct ib_uverbs_cmd_hdr hdr;
592 
593 	if (count < sizeof hdr)
594 		return -EINVAL;
595 
596 	if (copy_from_user(&hdr, buf, sizeof hdr))
597 		return -EFAULT;
598 
599 	if (hdr.in_words * 4 != count)
600 		return -EINVAL;
601 
602 	if (hdr.command < 0				||
603 	    hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
604 	    !uverbs_cmd_table[hdr.command]		||
605 	    !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
606 		return -EINVAL;
607 
608 	if (!file->ucontext &&
609 	    hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
610 		return -EINVAL;
611 
612 	return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
613 					     hdr.in_words * 4, hdr.out_words * 4);
614 }
615 
616 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
617 {
618 	struct ib_uverbs_file *file = filp->private_data;
619 
620 	if (!file->ucontext)
621 		return -ENODEV;
622 	else
623 		return file->device->ib_dev->mmap(file->ucontext, vma);
624 }
625 
626 static int ib_uverbs_open(struct inode *inode, struct file *filp)
627 {
628 	struct ib_uverbs_device *dev;
629 	struct ib_uverbs_file *file;
630 	int ret;
631 
632 	spin_lock(&map_lock);
633 	dev = dev_table[iminor(inode) - IB_UVERBS_BASE_MINOR];
634 	if (dev)
635 		kref_get(&dev->ref);
636 	spin_unlock(&map_lock);
637 
638 	if (!dev)
639 		return -ENXIO;
640 
641 	if (!try_module_get(dev->ib_dev->owner)) {
642 		ret = -ENODEV;
643 		goto err;
644 	}
645 
646 	file = kmalloc(sizeof *file, GFP_KERNEL);
647 	if (!file) {
648 		ret = -ENOMEM;
649 		goto err_module;
650 	}
651 
652 	file->device	 = dev;
653 	file->ucontext	 = NULL;
654 	file->async_file = NULL;
655 	kref_init(&file->ref);
656 	mutex_init(&file->mutex);
657 
658 	filp->private_data = file;
659 
660 	return 0;
661 
662 err_module:
663 	module_put(dev->ib_dev->owner);
664 
665 err:
666 	kref_put(&dev->ref, ib_uverbs_release_dev);
667 
668 	return ret;
669 }
670 
671 static int ib_uverbs_close(struct inode *inode, struct file *filp)
672 {
673 	struct ib_uverbs_file *file = filp->private_data;
674 
675 	ib_uverbs_cleanup_ucontext(file, file->ucontext);
676 
677 	if (file->async_file)
678 		kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
679 
680 	kref_put(&file->ref, ib_uverbs_release_file);
681 
682 	return 0;
683 }
684 
685 static struct file_operations uverbs_fops = {
686 	.owner 	 = THIS_MODULE,
687 	.write 	 = ib_uverbs_write,
688 	.open 	 = ib_uverbs_open,
689 	.release = ib_uverbs_close
690 };
691 
692 static struct file_operations uverbs_mmap_fops = {
693 	.owner 	 = THIS_MODULE,
694 	.write 	 = ib_uverbs_write,
695 	.mmap    = ib_uverbs_mmap,
696 	.open 	 = ib_uverbs_open,
697 	.release = ib_uverbs_close
698 };
699 
700 static struct ib_client uverbs_client = {
701 	.name   = "uverbs",
702 	.add    = ib_uverbs_add_one,
703 	.remove = ib_uverbs_remove_one
704 };
705 
706 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
707 {
708 	struct ib_uverbs_device *dev = class_get_devdata(class_dev);
709 
710 	if (!dev)
711 		return -ENODEV;
712 
713 	return sprintf(buf, "%s\n", dev->ib_dev->name);
714 }
715 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
716 
717 static ssize_t show_dev_abi_version(struct class_device *class_dev, char *buf)
718 {
719 	struct ib_uverbs_device *dev = class_get_devdata(class_dev);
720 
721 	if (!dev)
722 		return -ENODEV;
723 
724 	return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
725 }
726 static CLASS_DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
727 
728 static ssize_t show_abi_version(struct class *class, char *buf)
729 {
730 	return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
731 }
732 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
733 
734 static void ib_uverbs_add_one(struct ib_device *device)
735 {
736 	struct ib_uverbs_device *uverbs_dev;
737 
738 	if (!device->alloc_ucontext)
739 		return;
740 
741 	uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
742 	if (!uverbs_dev)
743 		return;
744 
745 	kref_init(&uverbs_dev->ref);
746 
747 	spin_lock(&map_lock);
748 	uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
749 	if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
750 		spin_unlock(&map_lock);
751 		goto err;
752 	}
753 	set_bit(uverbs_dev->devnum, dev_map);
754 	spin_unlock(&map_lock);
755 
756 	uverbs_dev->ib_dev           = device;
757 	uverbs_dev->num_comp_vectors = 1;
758 
759 	uverbs_dev->dev = cdev_alloc();
760 	if (!uverbs_dev->dev)
761 		goto err;
762 	uverbs_dev->dev->owner = THIS_MODULE;
763 	uverbs_dev->dev->ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
764 	kobject_set_name(&uverbs_dev->dev->kobj, "uverbs%d", uverbs_dev->devnum);
765 	if (cdev_add(uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
766 		goto err_cdev;
767 
768 	uverbs_dev->class_dev = class_device_create(uverbs_class, NULL,
769 						    uverbs_dev->dev->dev,
770 						    device->dma_device,
771 						    "uverbs%d", uverbs_dev->devnum);
772 	if (IS_ERR(uverbs_dev->class_dev))
773 		goto err_cdev;
774 
775 	class_set_devdata(uverbs_dev->class_dev, uverbs_dev);
776 
777 	if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_ibdev))
778 		goto err_class;
779 	if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_abi_version))
780 		goto err_class;
781 
782 	spin_lock(&map_lock);
783 	dev_table[uverbs_dev->devnum] = uverbs_dev;
784 	spin_unlock(&map_lock);
785 
786 	ib_set_client_data(device, &uverbs_client, uverbs_dev);
787 
788 	return;
789 
790 err_class:
791 	class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
792 
793 err_cdev:
794 	cdev_del(uverbs_dev->dev);
795 	clear_bit(uverbs_dev->devnum, dev_map);
796 
797 err:
798 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
799 	return;
800 }
801 
802 static void ib_uverbs_remove_one(struct ib_device *device)
803 {
804 	struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
805 
806 	if (!uverbs_dev)
807 		return;
808 
809 	class_set_devdata(uverbs_dev->class_dev, NULL);
810 	class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
811 	cdev_del(uverbs_dev->dev);
812 
813 	spin_lock(&map_lock);
814 	dev_table[uverbs_dev->devnum] = NULL;
815 	spin_unlock(&map_lock);
816 
817 	clear_bit(uverbs_dev->devnum, dev_map);
818 	kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
819 }
820 
821 static struct super_block *uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
822 					       const char *dev_name, void *data)
823 {
824 	return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
825 			     INFINIBANDEVENTFS_MAGIC);
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 	flush_scheduled_work();
906 	idr_destroy(&ib_uverbs_pd_idr);
907 	idr_destroy(&ib_uverbs_mr_idr);
908 	idr_destroy(&ib_uverbs_mw_idr);
909 	idr_destroy(&ib_uverbs_ah_idr);
910 	idr_destroy(&ib_uverbs_cq_idr);
911 	idr_destroy(&ib_uverbs_qp_idr);
912 	idr_destroy(&ib_uverbs_srq_idr);
913 }
914 
915 module_init(ib_uverbs_init);
916 module_exit(ib_uverbs_cleanup);
917