switch alloc_file() to passing struct path
[linux-2.6.git] / 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
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/device.h>
40 #include <linux/err.h>
41 #include <linux/fs.h>
42 #include <linux/poll.h>
43 #include <linux/sched.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/cdev.h>
47
48 #include <asm/uaccess.h>
49
50 #include "uverbs.h"
51
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
54 MODULE_LICENSE("Dual BSD/GPL");
55
56 #define INFINIBANDEVENTFS_MAGIC 0x49426576      /* "IBev" */
57
58 enum {
59         IB_UVERBS_MAJOR       = 231,
60         IB_UVERBS_BASE_MINOR  = 192,
61         IB_UVERBS_MAX_DEVICES = 32
62 };
63
64 #define IB_UVERBS_BASE_DEV      MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
65
66 static struct class *uverbs_class;
67
68 DEFINE_SPINLOCK(ib_uverbs_idr_lock);
69 DEFINE_IDR(ib_uverbs_pd_idr);
70 DEFINE_IDR(ib_uverbs_mr_idr);
71 DEFINE_IDR(ib_uverbs_mw_idr);
72 DEFINE_IDR(ib_uverbs_ah_idr);
73 DEFINE_IDR(ib_uverbs_cq_idr);
74 DEFINE_IDR(ib_uverbs_qp_idr);
75 DEFINE_IDR(ib_uverbs_srq_idr);
76
77 static DEFINE_SPINLOCK(map_lock);
78 static struct ib_uverbs_device *dev_table[IB_UVERBS_MAX_DEVICES];
79 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
80
81 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
82                                      const char __user *buf, int in_len,
83                                      int out_len) = {
84         [IB_USER_VERBS_CMD_GET_CONTEXT]         = ib_uverbs_get_context,
85         [IB_USER_VERBS_CMD_QUERY_DEVICE]        = ib_uverbs_query_device,
86         [IB_USER_VERBS_CMD_QUERY_PORT]          = ib_uverbs_query_port,
87         [IB_USER_VERBS_CMD_ALLOC_PD]            = ib_uverbs_alloc_pd,
88         [IB_USER_VERBS_CMD_DEALLOC_PD]          = ib_uverbs_dealloc_pd,
89         [IB_USER_VERBS_CMD_REG_MR]              = ib_uverbs_reg_mr,
90         [IB_USER_VERBS_CMD_DEREG_MR]            = ib_uverbs_dereg_mr,
91         [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
92         [IB_USER_VERBS_CMD_CREATE_CQ]           = ib_uverbs_create_cq,
93         [IB_USER_VERBS_CMD_RESIZE_CQ]           = ib_uverbs_resize_cq,
94         [IB_USER_VERBS_CMD_POLL_CQ]             = ib_uverbs_poll_cq,
95         [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]       = ib_uverbs_req_notify_cq,
96         [IB_USER_VERBS_CMD_DESTROY_CQ]          = ib_uverbs_destroy_cq,
97         [IB_USER_VERBS_CMD_CREATE_QP]           = ib_uverbs_create_qp,
98         [IB_USER_VERBS_CMD_QUERY_QP]            = ib_uverbs_query_qp,
99         [IB_USER_VERBS_CMD_MODIFY_QP]           = ib_uverbs_modify_qp,
100         [IB_USER_VERBS_CMD_DESTROY_QP]          = ib_uverbs_destroy_qp,
101         [IB_USER_VERBS_CMD_POST_SEND]           = ib_uverbs_post_send,
102         [IB_USER_VERBS_CMD_POST_RECV]           = ib_uverbs_post_recv,
103         [IB_USER_VERBS_CMD_POST_SRQ_RECV]       = ib_uverbs_post_srq_recv,
104         [IB_USER_VERBS_CMD_CREATE_AH]           = ib_uverbs_create_ah,
105         [IB_USER_VERBS_CMD_DESTROY_AH]          = ib_uverbs_destroy_ah,
106         [IB_USER_VERBS_CMD_ATTACH_MCAST]        = ib_uverbs_attach_mcast,
107         [IB_USER_VERBS_CMD_DETACH_MCAST]        = ib_uverbs_detach_mcast,
108         [IB_USER_VERBS_CMD_CREATE_SRQ]          = ib_uverbs_create_srq,
109         [IB_USER_VERBS_CMD_MODIFY_SRQ]          = ib_uverbs_modify_srq,
110         [IB_USER_VERBS_CMD_QUERY_SRQ]           = ib_uverbs_query_srq,
111         [IB_USER_VERBS_CMD_DESTROY_SRQ]         = ib_uverbs_destroy_srq,
112 };
113
114 static struct vfsmount *uverbs_event_mnt;
115
116 static void ib_uverbs_add_one(struct ib_device *device);
117 static void ib_uverbs_remove_one(struct ib_device *device);
118
119 static void ib_uverbs_release_dev(struct kref *ref)
120 {
121         struct ib_uverbs_device *dev =
122                 container_of(ref, struct ib_uverbs_device, ref);
123
124         complete(&dev->comp);
125 }
126
127 static void ib_uverbs_release_event_file(struct kref *ref)
128 {
129         struct ib_uverbs_event_file *file =
130                 container_of(ref, struct ib_uverbs_event_file, ref);
131
132         kfree(file);
133 }
134
135 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
136                           struct ib_uverbs_event_file *ev_file,
137                           struct ib_ucq_object *uobj)
138 {
139         struct ib_uverbs_event *evt, *tmp;
140
141         if (ev_file) {
142                 spin_lock_irq(&ev_file->lock);
143                 list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
144                         list_del(&evt->list);
145                         kfree(evt);
146                 }
147                 spin_unlock_irq(&ev_file->lock);
148
149                 kref_put(&ev_file->ref, ib_uverbs_release_event_file);
150         }
151
152         spin_lock_irq(&file->async_file->lock);
153         list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
154                 list_del(&evt->list);
155                 kfree(evt);
156         }
157         spin_unlock_irq(&file->async_file->lock);
158 }
159
160 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
161                               struct ib_uevent_object *uobj)
162 {
163         struct ib_uverbs_event *evt, *tmp;
164
165         spin_lock_irq(&file->async_file->lock);
166         list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
167                 list_del(&evt->list);
168                 kfree(evt);
169         }
170         spin_unlock_irq(&file->async_file->lock);
171 }
172
173 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
174                                     struct ib_uqp_object *uobj)
175 {
176         struct ib_uverbs_mcast_entry *mcast, *tmp;
177
178         list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
179                 ib_detach_mcast(qp, &mcast->gid, mcast->lid);
180                 list_del(&mcast->list);
181                 kfree(mcast);
182         }
183 }
184
185 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
186                                       struct ib_ucontext *context)
187 {
188         struct ib_uobject *uobj, *tmp;
189
190         if (!context)
191                 return 0;
192
193         context->closing = 1;
194
195         list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
196                 struct ib_ah *ah = uobj->object;
197
198                 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
199                 ib_destroy_ah(ah);
200                 kfree(uobj);
201         }
202
203         list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
204                 struct ib_qp *qp = uobj->object;
205                 struct ib_uqp_object *uqp =
206                         container_of(uobj, struct ib_uqp_object, uevent.uobject);
207
208                 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
209                 ib_uverbs_detach_umcast(qp, uqp);
210                 ib_destroy_qp(qp);
211                 ib_uverbs_release_uevent(file, &uqp->uevent);
212                 kfree(uqp);
213         }
214
215         list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
216                 struct ib_cq *cq = uobj->object;
217                 struct ib_uverbs_event_file *ev_file = cq->cq_context;
218                 struct ib_ucq_object *ucq =
219                         container_of(uobj, struct ib_ucq_object, uobject);
220
221                 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
222                 ib_destroy_cq(cq);
223                 ib_uverbs_release_ucq(file, ev_file, ucq);
224                 kfree(ucq);
225         }
226
227         list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
228                 struct ib_srq *srq = uobj->object;
229                 struct ib_uevent_object *uevent =
230                         container_of(uobj, struct ib_uevent_object, uobject);
231
232                 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
233                 ib_destroy_srq(srq);
234                 ib_uverbs_release_uevent(file, uevent);
235                 kfree(uevent);
236         }
237
238         /* XXX Free MWs */
239
240         list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
241                 struct ib_mr *mr = uobj->object;
242
243                 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
244                 ib_dereg_mr(mr);
245                 kfree(uobj);
246         }
247
248         list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
249                 struct ib_pd *pd = uobj->object;
250
251                 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
252                 ib_dealloc_pd(pd);
253                 kfree(uobj);
254         }
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 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
342 {
343         struct ib_uverbs_event_file *file = filp->private_data;
344
345         return fasync_helper(fd, filp, on, &file->async_queue);
346 }
347
348 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
349 {
350         struct ib_uverbs_event_file *file = filp->private_data;
351         struct ib_uverbs_event *entry, *tmp;
352
353         spin_lock_irq(&file->lock);
354         file->is_closed = 1;
355         list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
356                 if (entry->counter)
357                         list_del(&entry->obj_list);
358                 kfree(entry);
359         }
360         spin_unlock_irq(&file->lock);
361
362         if (file->is_async) {
363                 ib_unregister_event_handler(&file->uverbs_file->event_handler);
364                 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
365         }
366         kref_put(&file->ref, ib_uverbs_release_event_file);
367
368         return 0;
369 }
370
371 static const struct file_operations uverbs_event_fops = {
372         .owner   = THIS_MODULE,
373         .read    = ib_uverbs_event_read,
374         .poll    = ib_uverbs_event_poll,
375         .release = ib_uverbs_event_close,
376         .fasync  = ib_uverbs_event_fasync
377 };
378
379 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
380 {
381         struct ib_uverbs_event_file    *file = cq_context;
382         struct ib_ucq_object           *uobj;
383         struct ib_uverbs_event         *entry;
384         unsigned long                   flags;
385
386         if (!file)
387                 return;
388
389         spin_lock_irqsave(&file->lock, flags);
390         if (file->is_closed) {
391                 spin_unlock_irqrestore(&file->lock, flags);
392                 return;
393         }
394
395         entry = kmalloc(sizeof *entry, GFP_ATOMIC);
396         if (!entry) {
397                 spin_unlock_irqrestore(&file->lock, flags);
398                 return;
399         }
400
401         uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
402
403         entry->desc.comp.cq_handle = cq->uobject->user_handle;
404         entry->counter             = &uobj->comp_events_reported;
405
406         list_add_tail(&entry->list, &file->event_list);
407         list_add_tail(&entry->obj_list, &uobj->comp_list);
408         spin_unlock_irqrestore(&file->lock, flags);
409
410         wake_up_interruptible(&file->poll_wait);
411         kill_fasync(&file->async_queue, SIGIO, POLL_IN);
412 }
413
414 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
415                                     __u64 element, __u64 event,
416                                     struct list_head *obj_list,
417                                     u32 *counter)
418 {
419         struct ib_uverbs_event *entry;
420         unsigned long flags;
421
422         spin_lock_irqsave(&file->async_file->lock, flags);
423         if (file->async_file->is_closed) {
424                 spin_unlock_irqrestore(&file->async_file->lock, flags);
425                 return;
426         }
427
428         entry = kmalloc(sizeof *entry, GFP_ATOMIC);
429         if (!entry) {
430                 spin_unlock_irqrestore(&file->async_file->lock, flags);
431                 return;
432         }
433
434         entry->desc.async.element    = element;
435         entry->desc.async.event_type = event;
436         entry->counter               = counter;
437
438         list_add_tail(&entry->list, &file->async_file->event_list);
439         if (obj_list)
440                 list_add_tail(&entry->obj_list, obj_list);
441         spin_unlock_irqrestore(&file->async_file->lock, flags);
442
443         wake_up_interruptible(&file->async_file->poll_wait);
444         kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
445 }
446
447 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
448 {
449         struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
450                                                   struct ib_ucq_object, uobject);
451
452         ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
453                                 event->event, &uobj->async_list,
454                                 &uobj->async_events_reported);
455 }
456
457 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
458 {
459         struct ib_uevent_object *uobj;
460
461         uobj = container_of(event->element.qp->uobject,
462                             struct ib_uevent_object, uobject);
463
464         ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
465                                 event->event, &uobj->event_list,
466                                 &uobj->events_reported);
467 }
468
469 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
470 {
471         struct ib_uevent_object *uobj;
472
473         uobj = container_of(event->element.srq->uobject,
474                             struct ib_uevent_object, uobject);
475
476         ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
477                                 event->event, &uobj->event_list,
478                                 &uobj->events_reported);
479 }
480
481 void ib_uverbs_event_handler(struct ib_event_handler *handler,
482                              struct ib_event *event)
483 {
484         struct ib_uverbs_file *file =
485                 container_of(handler, struct ib_uverbs_file, event_handler);
486
487         ib_uverbs_async_handler(file, event->element.port_num, event->event,
488                                 NULL, NULL);
489 }
490
491 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
492                                         int is_async, int *fd)
493 {
494         struct ib_uverbs_event_file *ev_file;
495         struct path path;
496         struct file *filp;
497         int ret;
498
499         ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
500         if (!ev_file)
501                 return ERR_PTR(-ENOMEM);
502
503         kref_init(&ev_file->ref);
504         spin_lock_init(&ev_file->lock);
505         INIT_LIST_HEAD(&ev_file->event_list);
506         init_waitqueue_head(&ev_file->poll_wait);
507         ev_file->uverbs_file = uverbs_file;
508         ev_file->async_queue = NULL;
509         ev_file->is_async    = is_async;
510         ev_file->is_closed   = 0;
511
512         *fd = get_unused_fd();
513         if (*fd < 0) {
514                 ret = *fd;
515                 goto err;
516         }
517
518         /*
519          * fops_get() can't fail here, because we're coming from a
520          * system call on a uverbs file, which will already have a
521          * module reference.
522          */
523         path.mnt = uverbs_event_mnt;
524         path.dentry = uverbs_event_mnt->mnt_root;
525         path_get(&path);
526         filp = alloc_file(&path, FMODE_READ, fops_get(&uverbs_event_fops));
527         if (!filp) {
528                 ret = -ENFILE;
529                 goto err_fd;
530         }
531
532         filp->private_data = ev_file;
533
534         return filp;
535
536 err_fd:
537         fops_put(&uverbs_event_fops);
538         path_put(&path);
539         put_unused_fd(*fd);
540
541 err:
542         kfree(ev_file);
543         return ERR_PTR(ret);
544 }
545
546 /*
547  * Look up a completion event file by FD.  If lookup is successful,
548  * takes a ref to the event file struct that it returns; if
549  * unsuccessful, returns NULL.
550  */
551 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
552 {
553         struct ib_uverbs_event_file *ev_file = NULL;
554         struct file *filp;
555
556         filp = fget(fd);
557         if (!filp)
558                 return NULL;
559
560         if (filp->f_op != &uverbs_event_fops)
561                 goto out;
562
563         ev_file = filp->private_data;
564         if (ev_file->is_async) {
565                 ev_file = NULL;
566                 goto out;
567         }
568
569         kref_get(&ev_file->ref);
570
571 out:
572         fput(filp);
573         return ev_file;
574 }
575
576 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
577                              size_t count, loff_t *pos)
578 {
579         struct ib_uverbs_file *file = filp->private_data;
580         struct ib_uverbs_cmd_hdr hdr;
581
582         if (count < sizeof hdr)
583                 return -EINVAL;
584
585         if (copy_from_user(&hdr, buf, sizeof hdr))
586                 return -EFAULT;
587
588         if (hdr.in_words * 4 != count)
589                 return -EINVAL;
590
591         if (hdr.command < 0                             ||
592             hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
593             !uverbs_cmd_table[hdr.command])
594                 return -EINVAL;
595
596         if (!file->ucontext &&
597             hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
598                 return -EINVAL;
599
600         if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
601                 return -ENOSYS;
602
603         return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
604                                              hdr.in_words * 4, hdr.out_words * 4);
605 }
606
607 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
608 {
609         struct ib_uverbs_file *file = filp->private_data;
610
611         if (!file->ucontext)
612                 return -ENODEV;
613         else
614                 return file->device->ib_dev->mmap(file->ucontext, vma);
615 }
616
617 /*
618  * ib_uverbs_open() does not need the BKL:
619  *
620  *  - dev_table[] accesses are protected by map_lock, the
621  *    ib_uverbs_device structures are properly reference counted, and
622  *    everything else is purely local to the file being created, so
623  *    races against other open calls are not a problem;
624  *  - there is no ioctl method to race against;
625  *  - the device is added to dev_table[] as the last part of module
626  *    initialization, the open method will either immediately run
627  *    -ENXIO, or all required initialization will be done.
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         return ret;
671 }
672
673 static int ib_uverbs_close(struct inode *inode, struct file *filp)
674 {
675         struct ib_uverbs_file *file = filp->private_data;
676
677         ib_uverbs_cleanup_ucontext(file, file->ucontext);
678
679         if (file->async_file)
680                 kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
681
682         kref_put(&file->ref, ib_uverbs_release_file);
683
684         return 0;
685 }
686
687 static const struct file_operations uverbs_fops = {
688         .owner   = THIS_MODULE,
689         .write   = ib_uverbs_write,
690         .open    = ib_uverbs_open,
691         .release = ib_uverbs_close
692 };
693
694 static const struct file_operations uverbs_mmap_fops = {
695         .owner   = THIS_MODULE,
696         .write   = ib_uverbs_write,
697         .mmap    = ib_uverbs_mmap,
698         .open    = ib_uverbs_open,
699         .release = ib_uverbs_close
700 };
701
702 static struct ib_client uverbs_client = {
703         .name   = "uverbs",
704         .add    = ib_uverbs_add_one,
705         .remove = ib_uverbs_remove_one
706 };
707
708 static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
709                           char *buf)
710 {
711         struct ib_uverbs_device *dev = dev_get_drvdata(device);
712
713         if (!dev)
714                 return -ENODEV;
715
716         return sprintf(buf, "%s\n", dev->ib_dev->name);
717 }
718 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
719
720 static ssize_t show_dev_abi_version(struct device *device,
721                                     struct device_attribute *attr, char *buf)
722 {
723         struct ib_uverbs_device *dev = dev_get_drvdata(device);
724
725         if (!dev)
726                 return -ENODEV;
727
728         return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
729 }
730 static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
731
732 static ssize_t show_abi_version(struct class *class, char *buf)
733 {
734         return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
735 }
736 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
737
738 static void ib_uverbs_add_one(struct ib_device *device)
739 {
740         struct ib_uverbs_device *uverbs_dev;
741
742         if (!device->alloc_ucontext)
743                 return;
744
745         uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
746         if (!uverbs_dev)
747                 return;
748
749         kref_init(&uverbs_dev->ref);
750         init_completion(&uverbs_dev->comp);
751
752         spin_lock(&map_lock);
753         uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
754         if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
755                 spin_unlock(&map_lock);
756                 goto err;
757         }
758         set_bit(uverbs_dev->devnum, dev_map);
759         spin_unlock(&map_lock);
760
761         uverbs_dev->ib_dev           = device;
762         uverbs_dev->num_comp_vectors = device->num_comp_vectors;
763
764         uverbs_dev->cdev = cdev_alloc();
765         if (!uverbs_dev->cdev)
766                 goto err;
767         uverbs_dev->cdev->owner = THIS_MODULE;
768         uverbs_dev->cdev->ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
769         kobject_set_name(&uverbs_dev->cdev->kobj, "uverbs%d", uverbs_dev->devnum);
770         if (cdev_add(uverbs_dev->cdev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
771                 goto err_cdev;
772
773         uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
774                                         uverbs_dev->cdev->dev, uverbs_dev,
775                                         "uverbs%d", uverbs_dev->devnum);
776         if (IS_ERR(uverbs_dev->dev))
777                 goto err_cdev;
778
779         if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
780                 goto err_class;
781         if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
782                 goto err_class;
783
784         spin_lock(&map_lock);
785         dev_table[uverbs_dev->devnum] = uverbs_dev;
786         spin_unlock(&map_lock);
787
788         ib_set_client_data(device, &uverbs_client, uverbs_dev);
789
790         return;
791
792 err_class:
793         device_destroy(uverbs_class, uverbs_dev->cdev->dev);
794
795 err_cdev:
796         cdev_del(uverbs_dev->cdev);
797         clear_bit(uverbs_dev->devnum, dev_map);
798
799 err:
800         kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
801         wait_for_completion(&uverbs_dev->comp);
802         kfree(uverbs_dev);
803         return;
804 }
805
806 static void ib_uverbs_remove_one(struct ib_device *device)
807 {
808         struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
809
810         if (!uverbs_dev)
811                 return;
812
813         dev_set_drvdata(uverbs_dev->dev, NULL);
814         device_destroy(uverbs_class, uverbs_dev->cdev->dev);
815         cdev_del(uverbs_dev->cdev);
816
817         spin_lock(&map_lock);
818         dev_table[uverbs_dev->devnum] = NULL;
819         spin_unlock(&map_lock);
820
821         clear_bit(uverbs_dev->devnum, dev_map);
822
823         kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
824         wait_for_completion(&uverbs_dev->comp);
825         kfree(uverbs_dev);
826 }
827
828 static int uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
829                                const char *dev_name, void *data,
830                                struct vfsmount *mnt)
831 {
832         return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
833                              INFINIBANDEVENTFS_MAGIC, mnt);
834 }
835
836 static struct file_system_type uverbs_event_fs = {
837         /* No owner field so module can be unloaded */
838         .name    = "infinibandeventfs",
839         .get_sb  = uverbs_event_get_sb,
840         .kill_sb = kill_litter_super
841 };
842
843 static int __init ib_uverbs_init(void)
844 {
845         int ret;
846
847         ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
848                                      "infiniband_verbs");
849         if (ret) {
850                 printk(KERN_ERR "user_verbs: couldn't register device number\n");
851                 goto out;
852         }
853
854         uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
855         if (IS_ERR(uverbs_class)) {
856                 ret = PTR_ERR(uverbs_class);
857                 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
858                 goto out_chrdev;
859         }
860
861         ret = class_create_file(uverbs_class, &class_attr_abi_version);
862         if (ret) {
863                 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
864                 goto out_class;
865         }
866
867         ret = register_filesystem(&uverbs_event_fs);
868         if (ret) {
869                 printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
870                 goto out_class;
871         }
872
873         uverbs_event_mnt = kern_mount(&uverbs_event_fs);
874         if (IS_ERR(uverbs_event_mnt)) {
875                 ret = PTR_ERR(uverbs_event_mnt);
876                 printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
877                 goto out_fs;
878         }
879
880         ret = ib_register_client(&uverbs_client);
881         if (ret) {
882                 printk(KERN_ERR "user_verbs: couldn't register client\n");
883                 goto out_mnt;
884         }
885
886         return 0;
887
888 out_mnt:
889         mntput(uverbs_event_mnt);
890
891 out_fs:
892         unregister_filesystem(&uverbs_event_fs);
893
894 out_class:
895         class_destroy(uverbs_class);
896
897 out_chrdev:
898         unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
899
900 out:
901         return ret;
902 }
903
904 static void __exit ib_uverbs_cleanup(void)
905 {
906         ib_unregister_client(&uverbs_client);
907         mntput(uverbs_event_mnt);
908         unregister_filesystem(&uverbs_event_fs);
909         class_destroy(uverbs_class);
910         unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
911         idr_destroy(&ib_uverbs_pd_idr);
912         idr_destroy(&ib_uverbs_mr_idr);
913         idr_destroy(&ib_uverbs_mw_idr);
914         idr_destroy(&ib_uverbs_ah_idr);
915         idr_destroy(&ib_uverbs_cq_idr);
916         idr_destroy(&ib_uverbs_qp_idr);
917         idr_destroy(&ib_uverbs_srq_idr);
918 }
919
920 module_init(ib_uverbs_init);
921 module_exit(ib_uverbs_cleanup);