Merge branch 'master' of /home/trondmy/kernel/linux-2.6/
[linux-2.6] / drivers / infiniband / core / uverbs_main.c
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_SPINLOCK(ib_uverbs_idr_lock);
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         list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
187                 struct ib_ah *ah = uobj->object;
188
189                 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
190                 ib_destroy_ah(ah);
191                 list_del(&uobj->list);
192                 kfree(uobj);
193         }
194
195         list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
196                 struct ib_qp *qp = uobj->object;
197                 struct ib_uqp_object *uqp =
198                         container_of(uobj, struct ib_uqp_object, uevent.uobject);
199
200                 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
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 = uobj->object;
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
214                 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
215                 ib_destroy_cq(cq);
216                 list_del(&uobj->list);
217                 ib_uverbs_release_ucq(file, ev_file, ucq);
218                 kfree(ucq);
219         }
220
221         list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
222                 struct ib_srq *srq = uobj->object;
223                 struct ib_uevent_object *uevent =
224                         container_of(uobj, struct ib_uevent_object, uobject);
225
226                 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
227                 ib_destroy_srq(srq);
228                 list_del(&uobj->list);
229                 ib_uverbs_release_uevent(file, uevent);
230                 kfree(uevent);
231         }
232
233         /* XXX Free MWs */
234
235         list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
236                 struct ib_mr *mr = uobj->object;
237                 struct ib_device *mrdev = mr->device;
238                 struct ib_umem_object *memobj;
239
240                 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
241                 ib_dereg_mr(mr);
242
243                 memobj = container_of(uobj, struct ib_umem_object, uobject);
244                 ib_umem_release_on_close(mrdev, &memobj->umem);
245
246                 list_del(&uobj->list);
247                 kfree(memobj);
248         }
249
250         list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
251                 struct ib_pd *pd = uobj->object;
252
253                 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
254                 ib_dealloc_pd(pd);
255                 list_del(&uobj->list);
256                 kfree(uobj);
257         }
258
259         return context->device->dealloc_ucontext(context);
260 }
261
262 static void ib_uverbs_release_file(struct kref *ref)
263 {
264         struct ib_uverbs_file *file =
265                 container_of(ref, struct ib_uverbs_file, ref);
266
267         module_put(file->device->ib_dev->owner);
268         kref_put(&file->device->ref, ib_uverbs_release_dev);
269
270         kfree(file);
271 }
272
273 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
274                                     size_t count, loff_t *pos)
275 {
276         struct ib_uverbs_event_file *file = filp->private_data;
277         struct ib_uverbs_event *event;
278         int eventsz;
279         int ret = 0;
280
281         spin_lock_irq(&file->lock);
282
283         while (list_empty(&file->event_list)) {
284                 spin_unlock_irq(&file->lock);
285
286                 if (filp->f_flags & O_NONBLOCK)
287                         return -EAGAIN;
288
289                 if (wait_event_interruptible(file->poll_wait,
290                                              !list_empty(&file->event_list)))
291                         return -ERESTARTSYS;
292
293                 spin_lock_irq(&file->lock);
294         }
295
296         event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
297
298         if (file->is_async)
299                 eventsz = sizeof (struct ib_uverbs_async_event_desc);
300         else
301                 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
302
303         if (eventsz > count) {
304                 ret   = -EINVAL;
305                 event = NULL;
306         } else {
307                 list_del(file->event_list.next);
308                 if (event->counter) {
309                         ++(*event->counter);
310                         list_del(&event->obj_list);
311                 }
312         }
313
314         spin_unlock_irq(&file->lock);
315
316         if (event) {
317                 if (copy_to_user(buf, event, eventsz))
318                         ret = -EFAULT;
319                 else
320                         ret = eventsz;
321         }
322
323         kfree(event);
324
325         return ret;
326 }
327
328 static unsigned int ib_uverbs_event_poll(struct file *filp,
329                                          struct poll_table_struct *wait)
330 {
331         unsigned int pollflags = 0;
332         struct ib_uverbs_event_file *file = filp->private_data;
333
334         poll_wait(filp, &file->poll_wait, wait);
335
336         spin_lock_irq(&file->lock);
337         if (!list_empty(&file->event_list))
338                 pollflags = POLLIN | POLLRDNORM;
339         spin_unlock_irq(&file->lock);
340
341         return pollflags;
342 }
343
344 void ib_uverbs_release_event_file(struct kref *ref)
345 {
346         struct ib_uverbs_event_file *file =
347                 container_of(ref, struct ib_uverbs_event_file, ref);
348
349         kfree(file);
350 }
351
352 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
353 {
354         struct ib_uverbs_event_file *file = filp->private_data;
355
356         return fasync_helper(fd, filp, on, &file->async_queue);
357 }
358
359 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
360 {
361         struct ib_uverbs_event_file *file = filp->private_data;
362         struct ib_uverbs_event *entry, *tmp;
363
364         spin_lock_irq(&file->lock);
365         file->file = NULL;
366         list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
367                 if (entry->counter)
368                         list_del(&entry->obj_list);
369                 kfree(entry);
370         }
371         spin_unlock_irq(&file->lock);
372
373         ib_uverbs_event_fasync(-1, filp, 0);
374
375         if (file->is_async) {
376                 ib_unregister_event_handler(&file->uverbs_file->event_handler);
377                 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
378         }
379         kref_put(&file->ref, ib_uverbs_release_event_file);
380
381         return 0;
382 }
383
384 static struct file_operations uverbs_event_fops = {
385         .owner   = THIS_MODULE,
386         .read    = ib_uverbs_event_read,
387         .poll    = ib_uverbs_event_poll,
388         .release = ib_uverbs_event_close,
389         .fasync  = ib_uverbs_event_fasync
390 };
391
392 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
393 {
394         struct ib_uverbs_event_file    *file = cq_context;
395         struct ib_ucq_object           *uobj;
396         struct ib_uverbs_event         *entry;
397         unsigned long                   flags;
398
399         if (!file)
400                 return;
401
402         spin_lock_irqsave(&file->lock, flags);
403         if (!file->file) {
404                 spin_unlock_irqrestore(&file->lock, flags);
405                 return;
406         }
407
408         entry = kmalloc(sizeof *entry, GFP_ATOMIC);
409         if (!entry) {
410                 spin_unlock_irqrestore(&file->lock, flags);
411                 return;
412         }
413
414         uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
415
416         entry->desc.comp.cq_handle = cq->uobject->user_handle;
417         entry->counter             = &uobj->comp_events_reported;
418
419         list_add_tail(&entry->list, &file->event_list);
420         list_add_tail(&entry->obj_list, &uobj->comp_list);
421         spin_unlock_irqrestore(&file->lock, flags);
422
423         wake_up_interruptible(&file->poll_wait);
424         kill_fasync(&file->async_queue, SIGIO, POLL_IN);
425 }
426
427 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
428                                     __u64 element, __u64 event,
429                                     struct list_head *obj_list,
430                                     u32 *counter)
431 {
432         struct ib_uverbs_event *entry;
433         unsigned long flags;
434
435         spin_lock_irqsave(&file->async_file->lock, flags);
436         if (!file->async_file->file) {
437                 spin_unlock_irqrestore(&file->async_file->lock, flags);
438                 return;
439         }
440
441         entry = kmalloc(sizeof *entry, GFP_ATOMIC);
442         if (!entry) {
443                 spin_unlock_irqrestore(&file->async_file->lock, flags);
444                 return;
445         }
446
447         entry->desc.async.element    = element;
448         entry->desc.async.event_type = event;
449         entry->counter               = counter;
450
451         list_add_tail(&entry->list, &file->async_file->event_list);
452         if (obj_list)
453                 list_add_tail(&entry->obj_list, obj_list);
454         spin_unlock_irqrestore(&file->async_file->lock, flags);
455
456         wake_up_interruptible(&file->async_file->poll_wait);
457         kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
458 }
459
460 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
461 {
462         struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
463                                                   struct ib_ucq_object, uobject);
464
465         ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
466                                 event->event, &uobj->async_list,
467                                 &uobj->async_events_reported);
468 }
469
470 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
471 {
472         struct ib_uevent_object *uobj;
473
474         uobj = container_of(event->element.qp->uobject,
475                             struct ib_uevent_object, uobject);
476
477         ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
478                                 event->event, &uobj->event_list,
479                                 &uobj->events_reported);
480 }
481
482 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
483 {
484         struct ib_uevent_object *uobj;
485
486         uobj = container_of(event->element.srq->uobject,
487                             struct ib_uevent_object, uobject);
488
489         ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
490                                 event->event, &uobj->event_list,
491                                 &uobj->events_reported);
492 }
493
494 void ib_uverbs_event_handler(struct ib_event_handler *handler,
495                              struct ib_event *event)
496 {
497         struct ib_uverbs_file *file =
498                 container_of(handler, struct ib_uverbs_file, event_handler);
499
500         ib_uverbs_async_handler(file, event->element.port_num, event->event,
501                                 NULL, NULL);
502 }
503
504 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
505                                         int is_async, int *fd)
506 {
507         struct ib_uverbs_event_file *ev_file;
508         struct file *filp;
509         int ret;
510
511         ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
512         if (!ev_file)
513                 return ERR_PTR(-ENOMEM);
514
515         kref_init(&ev_file->ref);
516         spin_lock_init(&ev_file->lock);
517         INIT_LIST_HEAD(&ev_file->event_list);
518         init_waitqueue_head(&ev_file->poll_wait);
519         ev_file->uverbs_file = uverbs_file;
520         ev_file->async_queue = NULL;
521         ev_file->is_async    = is_async;
522
523         *fd = get_unused_fd();
524         if (*fd < 0) {
525                 ret = *fd;
526                 goto err;
527         }
528
529         filp = get_empty_filp();
530         if (!filp) {
531                 ret = -ENFILE;
532                 goto err_fd;
533         }
534
535         ev_file->file      = filp;
536
537         /*
538          * fops_get() can't fail here, because we're coming from a
539          * system call on a uverbs file, which will already have a
540          * module reference.
541          */
542         filp->f_op         = fops_get(&uverbs_event_fops);
543         filp->f_vfsmnt     = mntget(uverbs_event_mnt);
544         filp->f_dentry     = dget(uverbs_event_mnt->mnt_root);
545         filp->f_mapping    = filp->f_dentry->d_inode->i_mapping;
546         filp->f_flags      = O_RDONLY;
547         filp->f_mode       = FMODE_READ;
548         filp->private_data = ev_file;
549
550         return filp;
551
552 err_fd:
553         put_unused_fd(*fd);
554
555 err:
556         kfree(ev_file);
557         return ERR_PTR(ret);
558 }
559
560 /*
561  * Look up a completion event file by FD.  If lookup is successful,
562  * takes a ref to the event file struct that it returns; if
563  * unsuccessful, returns NULL.
564  */
565 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
566 {
567         struct ib_uverbs_event_file *ev_file = NULL;
568         struct file *filp;
569
570         filp = fget(fd);
571         if (!filp)
572                 return NULL;
573
574         if (filp->f_op != &uverbs_event_fops)
575                 goto out;
576
577         ev_file = filp->private_data;
578         if (ev_file->is_async) {
579                 ev_file = NULL;
580                 goto out;
581         }
582
583         kref_get(&ev_file->ref);
584
585 out:
586         fput(filp);
587         return ev_file;
588 }
589
590 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
591                              size_t count, loff_t *pos)
592 {
593         struct ib_uverbs_file *file = filp->private_data;
594         struct ib_uverbs_cmd_hdr hdr;
595
596         if (count < sizeof hdr)
597                 return -EINVAL;
598
599         if (copy_from_user(&hdr, buf, sizeof hdr))
600                 return -EFAULT;
601
602         if (hdr.in_words * 4 != count)
603                 return -EINVAL;
604
605         if (hdr.command < 0                             ||
606             hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
607             !uverbs_cmd_table[hdr.command]              ||
608             !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
609                 return -EINVAL;
610
611         if (!file->ucontext &&
612             hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
613                 return -EINVAL;
614
615         return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
616                                              hdr.in_words * 4, hdr.out_words * 4);
617 }
618
619 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
620 {
621         struct ib_uverbs_file *file = filp->private_data;
622
623         if (!file->ucontext)
624                 return -ENODEV;
625         else
626                 return file->device->ib_dev->mmap(file->ucontext, vma);
627 }
628
629 static int ib_uverbs_open(struct inode *inode, struct file *filp)
630 {
631         struct ib_uverbs_device *dev;
632         struct ib_uverbs_file *file;
633         int ret;
634
635         spin_lock(&map_lock);
636         dev = dev_table[iminor(inode) - IB_UVERBS_BASE_MINOR];
637         if (dev)
638                 kref_get(&dev->ref);
639         spin_unlock(&map_lock);
640
641         if (!dev)
642                 return -ENXIO;
643
644         if (!try_module_get(dev->ib_dev->owner)) {
645                 ret = -ENODEV;
646                 goto err;
647         }
648
649         file = kmalloc(sizeof *file, GFP_KERNEL);
650         if (!file) {
651                 ret = -ENOMEM;
652                 goto err_module;
653         }
654
655         file->device     = dev;
656         file->ucontext   = NULL;
657         file->async_file = NULL;
658         kref_init(&file->ref);
659         mutex_init(&file->mutex);
660
661         filp->private_data = file;
662
663         return 0;
664
665 err_module:
666         module_put(dev->ib_dev->owner);
667
668 err:
669         kref_put(&dev->ref, ib_uverbs_release_dev);
670
671         return ret;
672 }
673
674 static int ib_uverbs_close(struct inode *inode, struct file *filp)
675 {
676         struct ib_uverbs_file *file = filp->private_data;
677
678         ib_uverbs_cleanup_ucontext(file, file->ucontext);
679
680         if (file->async_file)
681                 kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
682
683         kref_put(&file->ref, ib_uverbs_release_file);
684
685         return 0;
686 }
687
688 static struct file_operations uverbs_fops = {
689         .owner   = THIS_MODULE,
690         .write   = ib_uverbs_write,
691         .open    = ib_uverbs_open,
692         .release = ib_uverbs_close
693 };
694
695 static struct file_operations uverbs_mmap_fops = {
696         .owner   = THIS_MODULE,
697         .write   = ib_uverbs_write,
698         .mmap    = ib_uverbs_mmap,
699         .open    = ib_uverbs_open,
700         .release = ib_uverbs_close
701 };
702
703 static struct ib_client uverbs_client = {
704         .name   = "uverbs",
705         .add    = ib_uverbs_add_one,
706         .remove = ib_uverbs_remove_one
707 };
708
709 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
710 {
711         struct ib_uverbs_device *dev = class_get_devdata(class_dev);
712
713         if (!dev)
714                 return -ENODEV;
715
716         return sprintf(buf, "%s\n", dev->ib_dev->name);
717 }
718 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
719
720 static ssize_t show_dev_abi_version(struct class_device *class_dev, char *buf)
721 {
722         struct ib_uverbs_device *dev = class_get_devdata(class_dev);
723
724         if (!dev)
725                 return -ENODEV;
726
727         return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
728 }
729 static CLASS_DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
730
731 static ssize_t show_abi_version(struct class *class, char *buf)
732 {
733         return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
734 }
735 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
736
737 static void ib_uverbs_add_one(struct ib_device *device)
738 {
739         struct ib_uverbs_device *uverbs_dev;
740
741         if (!device->alloc_ucontext)
742                 return;
743
744         uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
745         if (!uverbs_dev)
746                 return;
747
748         kref_init(&uverbs_dev->ref);
749
750         spin_lock(&map_lock);
751         uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
752         if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
753                 spin_unlock(&map_lock);
754                 goto err;
755         }
756         set_bit(uverbs_dev->devnum, dev_map);
757         spin_unlock(&map_lock);
758
759         uverbs_dev->ib_dev           = device;
760         uverbs_dev->num_comp_vectors = 1;
761
762         uverbs_dev->dev = cdev_alloc();
763         if (!uverbs_dev->dev)
764                 goto err;
765         uverbs_dev->dev->owner = THIS_MODULE;
766         uverbs_dev->dev->ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
767         kobject_set_name(&uverbs_dev->dev->kobj, "uverbs%d", uverbs_dev->devnum);
768         if (cdev_add(uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
769                 goto err_cdev;
770
771         uverbs_dev->class_dev = class_device_create(uverbs_class, NULL,
772                                                     uverbs_dev->dev->dev,
773                                                     device->dma_device,
774                                                     "uverbs%d", uverbs_dev->devnum);
775         if (IS_ERR(uverbs_dev->class_dev))
776                 goto err_cdev;
777
778         class_set_devdata(uverbs_dev->class_dev, uverbs_dev);
779
780         if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_ibdev))
781                 goto err_class;
782         if (class_device_create_file(uverbs_dev->class_dev, &class_device_attr_abi_version))
783                 goto err_class;
784
785         spin_lock(&map_lock);
786         dev_table[uverbs_dev->devnum] = uverbs_dev;
787         spin_unlock(&map_lock);
788
789         ib_set_client_data(device, &uverbs_client, uverbs_dev);
790
791         return;
792
793 err_class:
794         class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
795
796 err_cdev:
797         cdev_del(uverbs_dev->dev);
798         clear_bit(uverbs_dev->devnum, dev_map);
799
800 err:
801         kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
802         return;
803 }
804
805 static void ib_uverbs_remove_one(struct ib_device *device)
806 {
807         struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
808
809         if (!uverbs_dev)
810                 return;
811
812         class_set_devdata(uverbs_dev->class_dev, NULL);
813         class_device_destroy(uverbs_class, uverbs_dev->dev->dev);
814         cdev_del(uverbs_dev->dev);
815
816         spin_lock(&map_lock);
817         dev_table[uverbs_dev->devnum] = NULL;
818         spin_unlock(&map_lock);
819
820         clear_bit(uverbs_dev->devnum, dev_map);
821         kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
822 }
823
824 static int uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
825                                const char *dev_name, void *data,
826                                struct vfsmount *mnt)
827 {
828         return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
829                              INFINIBANDEVENTFS_MAGIC, mnt);
830 }
831
832 static struct file_system_type uverbs_event_fs = {
833         /* No owner field so module can be unloaded */
834         .name    = "infinibandeventfs",
835         .get_sb  = uverbs_event_get_sb,
836         .kill_sb = kill_litter_super
837 };
838
839 static int __init ib_uverbs_init(void)
840 {
841         int ret;
842
843         spin_lock_init(&map_lock);
844
845         ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
846                                      "infiniband_verbs");
847         if (ret) {
848                 printk(KERN_ERR "user_verbs: couldn't register device number\n");
849                 goto out;
850         }
851
852         uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
853         if (IS_ERR(uverbs_class)) {
854                 ret = PTR_ERR(uverbs_class);
855                 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
856                 goto out_chrdev;
857         }
858
859         ret = class_create_file(uverbs_class, &class_attr_abi_version);
860         if (ret) {
861                 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
862                 goto out_class;
863         }
864
865         ret = register_filesystem(&uverbs_event_fs);
866         if (ret) {
867                 printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
868                 goto out_class;
869         }
870
871         uverbs_event_mnt = kern_mount(&uverbs_event_fs);
872         if (IS_ERR(uverbs_event_mnt)) {
873                 ret = PTR_ERR(uverbs_event_mnt);
874                 printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
875                 goto out_fs;
876         }
877
878         ret = ib_register_client(&uverbs_client);
879         if (ret) {
880                 printk(KERN_ERR "user_verbs: couldn't register client\n");
881                 goto out_mnt;
882         }
883
884         return 0;
885
886 out_mnt:
887         mntput(uverbs_event_mnt);
888
889 out_fs:
890         unregister_filesystem(&uverbs_event_fs);
891
892 out_class:
893         class_destroy(uverbs_class);
894
895 out_chrdev:
896         unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
897
898 out:
899         return ret;
900 }
901
902 static void __exit ib_uverbs_cleanup(void)
903 {
904         ib_unregister_client(&uverbs_client);
905         mntput(uverbs_event_mnt);
906         unregister_filesystem(&uverbs_event_fs);
907         class_destroy(uverbs_class);
908         unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
909         flush_scheduled_work();
910         idr_destroy(&ib_uverbs_pd_idr);
911         idr_destroy(&ib_uverbs_mr_idr);
912         idr_destroy(&ib_uverbs_mw_idr);
913         idr_destroy(&ib_uverbs_ah_idr);
914         idr_destroy(&ib_uverbs_cq_idr);
915         idr_destroy(&ib_uverbs_qp_idr);
916         idr_destroy(&ib_uverbs_srq_idr);
917 }
918
919 module_init(ib_uverbs_init);
920 module_exit(ib_uverbs_cleanup);