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