Loading...
1/*
2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
6 * Copyright (c) 2005 PathScale, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
17 *
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
21 *
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 */
36
37#include <linux/module.h>
38#include <linux/init.h>
39#include <linux/device.h>
40#include <linux/err.h>
41#include <linux/fs.h>
42#include <linux/poll.h>
43#include <linux/sched.h>
44#include <linux/file.h>
45#include <linux/cdev.h>
46#include <linux/anon_inodes.h>
47#include <linux/slab.h>
48
49#include <asm/uaccess.h>
50
51#include "uverbs.h"
52
53MODULE_AUTHOR("Roland Dreier");
54MODULE_DESCRIPTION("InfiniBand userspace verbs access");
55MODULE_LICENSE("Dual BSD/GPL");
56
57enum {
58 IB_UVERBS_MAJOR = 231,
59 IB_UVERBS_BASE_MINOR = 192,
60 IB_UVERBS_MAX_DEVICES = 32
61};
62
63#define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
64
65static struct class *uverbs_class;
66
67DEFINE_SPINLOCK(ib_uverbs_idr_lock);
68DEFINE_IDR(ib_uverbs_pd_idr);
69DEFINE_IDR(ib_uverbs_mr_idr);
70DEFINE_IDR(ib_uverbs_mw_idr);
71DEFINE_IDR(ib_uverbs_ah_idr);
72DEFINE_IDR(ib_uverbs_cq_idr);
73DEFINE_IDR(ib_uverbs_qp_idr);
74DEFINE_IDR(ib_uverbs_srq_idr);
75
76static DEFINE_SPINLOCK(map_lock);
77static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
78
79static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
80 const char __user *buf, int in_len,
81 int out_len) = {
82 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
83 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
84 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
85 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
86 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
87 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
88 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
89 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
90 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
91 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq,
92 [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq,
93 [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq,
94 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
95 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
96 [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp,
97 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
98 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
99 [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send,
100 [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv,
101 [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv,
102 [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah,
103 [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah,
104 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
105 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
106 [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
107 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
108 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq,
109 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
110};
111
112static void ib_uverbs_add_one(struct ib_device *device);
113static void ib_uverbs_remove_one(struct ib_device *device);
114
115static void ib_uverbs_release_dev(struct kref *ref)
116{
117 struct ib_uverbs_device *dev =
118 container_of(ref, struct ib_uverbs_device, ref);
119
120 complete(&dev->comp);
121}
122
123static void ib_uverbs_release_event_file(struct kref *ref)
124{
125 struct ib_uverbs_event_file *file =
126 container_of(ref, struct ib_uverbs_event_file, ref);
127
128 kfree(file);
129}
130
131void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
132 struct ib_uverbs_event_file *ev_file,
133 struct ib_ucq_object *uobj)
134{
135 struct ib_uverbs_event *evt, *tmp;
136
137 if (ev_file) {
138 spin_lock_irq(&ev_file->lock);
139 list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
140 list_del(&evt->list);
141 kfree(evt);
142 }
143 spin_unlock_irq(&ev_file->lock);
144
145 kref_put(&ev_file->ref, ib_uverbs_release_event_file);
146 }
147
148 spin_lock_irq(&file->async_file->lock);
149 list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
150 list_del(&evt->list);
151 kfree(evt);
152 }
153 spin_unlock_irq(&file->async_file->lock);
154}
155
156void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
157 struct ib_uevent_object *uobj)
158{
159 struct ib_uverbs_event *evt, *tmp;
160
161 spin_lock_irq(&file->async_file->lock);
162 list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
163 list_del(&evt->list);
164 kfree(evt);
165 }
166 spin_unlock_irq(&file->async_file->lock);
167}
168
169static void ib_uverbs_detach_umcast(struct ib_qp *qp,
170 struct ib_uqp_object *uobj)
171{
172 struct ib_uverbs_mcast_entry *mcast, *tmp;
173
174 list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
175 ib_detach_mcast(qp, &mcast->gid, mcast->lid);
176 list_del(&mcast->list);
177 kfree(mcast);
178 }
179}
180
181static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
182 struct ib_ucontext *context)
183{
184 struct ib_uobject *uobj, *tmp;
185
186 if (!context)
187 return 0;
188
189 context->closing = 1;
190
191 list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
192 struct ib_ah *ah = uobj->object;
193
194 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
195 ib_destroy_ah(ah);
196 kfree(uobj);
197 }
198
199 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
200 struct ib_qp *qp = uobj->object;
201 struct ib_uqp_object *uqp =
202 container_of(uobj, struct ib_uqp_object, uevent.uobject);
203
204 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
205 ib_uverbs_detach_umcast(qp, uqp);
206 ib_destroy_qp(qp);
207 ib_uverbs_release_uevent(file, &uqp->uevent);
208 kfree(uqp);
209 }
210
211 list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
212 struct ib_cq *cq = uobj->object;
213 struct ib_uverbs_event_file *ev_file = cq->cq_context;
214 struct ib_ucq_object *ucq =
215 container_of(uobj, struct ib_ucq_object, uobject);
216
217 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
218 ib_destroy_cq(cq);
219 ib_uverbs_release_ucq(file, ev_file, ucq);
220 kfree(ucq);
221 }
222
223 list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
224 struct ib_srq *srq = uobj->object;
225 struct ib_uevent_object *uevent =
226 container_of(uobj, struct ib_uevent_object, uobject);
227
228 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
229 ib_destroy_srq(srq);
230 ib_uverbs_release_uevent(file, uevent);
231 kfree(uevent);
232 }
233
234 /* XXX Free MWs */
235
236 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
237 struct ib_mr *mr = uobj->object;
238
239 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
240 ib_dereg_mr(mr);
241 kfree(uobj);
242 }
243
244 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
245 struct ib_pd *pd = uobj->object;
246
247 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
248 ib_dealloc_pd(pd);
249 kfree(uobj);
250 }
251
252 return context->device->dealloc_ucontext(context);
253}
254
255static void ib_uverbs_release_file(struct kref *ref)
256{
257 struct ib_uverbs_file *file =
258 container_of(ref, struct ib_uverbs_file, ref);
259
260 module_put(file->device->ib_dev->owner);
261 kref_put(&file->device->ref, ib_uverbs_release_dev);
262
263 kfree(file);
264}
265
266static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
267 size_t count, loff_t *pos)
268{
269 struct ib_uverbs_event_file *file = filp->private_data;
270 struct ib_uverbs_event *event;
271 int eventsz;
272 int ret = 0;
273
274 spin_lock_irq(&file->lock);
275
276 while (list_empty(&file->event_list)) {
277 spin_unlock_irq(&file->lock);
278
279 if (filp->f_flags & O_NONBLOCK)
280 return -EAGAIN;
281
282 if (wait_event_interruptible(file->poll_wait,
283 !list_empty(&file->event_list)))
284 return -ERESTARTSYS;
285
286 spin_lock_irq(&file->lock);
287 }
288
289 event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
290
291 if (file->is_async)
292 eventsz = sizeof (struct ib_uverbs_async_event_desc);
293 else
294 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
295
296 if (eventsz > count) {
297 ret = -EINVAL;
298 event = NULL;
299 } else {
300 list_del(file->event_list.next);
301 if (event->counter) {
302 ++(*event->counter);
303 list_del(&event->obj_list);
304 }
305 }
306
307 spin_unlock_irq(&file->lock);
308
309 if (event) {
310 if (copy_to_user(buf, event, eventsz))
311 ret = -EFAULT;
312 else
313 ret = eventsz;
314 }
315
316 kfree(event);
317
318 return ret;
319}
320
321static unsigned int ib_uverbs_event_poll(struct file *filp,
322 struct poll_table_struct *wait)
323{
324 unsigned int pollflags = 0;
325 struct ib_uverbs_event_file *file = filp->private_data;
326
327 poll_wait(filp, &file->poll_wait, wait);
328
329 spin_lock_irq(&file->lock);
330 if (!list_empty(&file->event_list))
331 pollflags = POLLIN | POLLRDNORM;
332 spin_unlock_irq(&file->lock);
333
334 return pollflags;
335}
336
337static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
338{
339 struct ib_uverbs_event_file *file = filp->private_data;
340
341 return fasync_helper(fd, filp, on, &file->async_queue);
342}
343
344static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
345{
346 struct ib_uverbs_event_file *file = filp->private_data;
347 struct ib_uverbs_event *entry, *tmp;
348
349 spin_lock_irq(&file->lock);
350 file->is_closed = 1;
351 list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
352 if (entry->counter)
353 list_del(&entry->obj_list);
354 kfree(entry);
355 }
356 spin_unlock_irq(&file->lock);
357
358 if (file->is_async) {
359 ib_unregister_event_handler(&file->uverbs_file->event_handler);
360 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
361 }
362 kref_put(&file->ref, ib_uverbs_release_event_file);
363
364 return 0;
365}
366
367static const struct file_operations uverbs_event_fops = {
368 .owner = THIS_MODULE,
369 .read = ib_uverbs_event_read,
370 .poll = ib_uverbs_event_poll,
371 .release = ib_uverbs_event_close,
372 .fasync = ib_uverbs_event_fasync,
373 .llseek = no_llseek,
374};
375
376void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
377{
378 struct ib_uverbs_event_file *file = cq_context;
379 struct ib_ucq_object *uobj;
380 struct ib_uverbs_event *entry;
381 unsigned long flags;
382
383 if (!file)
384 return;
385
386 spin_lock_irqsave(&file->lock, flags);
387 if (file->is_closed) {
388 spin_unlock_irqrestore(&file->lock, flags);
389 return;
390 }
391
392 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
393 if (!entry) {
394 spin_unlock_irqrestore(&file->lock, flags);
395 return;
396 }
397
398 uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
399
400 entry->desc.comp.cq_handle = cq->uobject->user_handle;
401 entry->counter = &uobj->comp_events_reported;
402
403 list_add_tail(&entry->list, &file->event_list);
404 list_add_tail(&entry->obj_list, &uobj->comp_list);
405 spin_unlock_irqrestore(&file->lock, flags);
406
407 wake_up_interruptible(&file->poll_wait);
408 kill_fasync(&file->async_queue, SIGIO, POLL_IN);
409}
410
411static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
412 __u64 element, __u64 event,
413 struct list_head *obj_list,
414 u32 *counter)
415{
416 struct ib_uverbs_event *entry;
417 unsigned long flags;
418
419 spin_lock_irqsave(&file->async_file->lock, flags);
420 if (file->async_file->is_closed) {
421 spin_unlock_irqrestore(&file->async_file->lock, flags);
422 return;
423 }
424
425 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
426 if (!entry) {
427 spin_unlock_irqrestore(&file->async_file->lock, flags);
428 return;
429 }
430
431 entry->desc.async.element = element;
432 entry->desc.async.event_type = event;
433 entry->counter = counter;
434
435 list_add_tail(&entry->list, &file->async_file->event_list);
436 if (obj_list)
437 list_add_tail(&entry->obj_list, obj_list);
438 spin_unlock_irqrestore(&file->async_file->lock, flags);
439
440 wake_up_interruptible(&file->async_file->poll_wait);
441 kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
442}
443
444void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
445{
446 struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
447 struct ib_ucq_object, uobject);
448
449 ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
450 event->event, &uobj->async_list,
451 &uobj->async_events_reported);
452}
453
454void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
455{
456 struct ib_uevent_object *uobj;
457
458 uobj = container_of(event->element.qp->uobject,
459 struct ib_uevent_object, uobject);
460
461 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
462 event->event, &uobj->event_list,
463 &uobj->events_reported);
464}
465
466void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
467{
468 struct ib_uevent_object *uobj;
469
470 uobj = container_of(event->element.srq->uobject,
471 struct ib_uevent_object, uobject);
472
473 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
474 event->event, &uobj->event_list,
475 &uobj->events_reported);
476}
477
478void ib_uverbs_event_handler(struct ib_event_handler *handler,
479 struct ib_event *event)
480{
481 struct ib_uverbs_file *file =
482 container_of(handler, struct ib_uverbs_file, event_handler);
483
484 ib_uverbs_async_handler(file, event->element.port_num, event->event,
485 NULL, NULL);
486}
487
488struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
489 int is_async)
490{
491 struct ib_uverbs_event_file *ev_file;
492 struct file *filp;
493
494 ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
495 if (!ev_file)
496 return ERR_PTR(-ENOMEM);
497
498 kref_init(&ev_file->ref);
499 spin_lock_init(&ev_file->lock);
500 INIT_LIST_HEAD(&ev_file->event_list);
501 init_waitqueue_head(&ev_file->poll_wait);
502 ev_file->uverbs_file = uverbs_file;
503 ev_file->async_queue = NULL;
504 ev_file->is_async = is_async;
505 ev_file->is_closed = 0;
506
507 filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops,
508 ev_file, O_RDONLY);
509 if (IS_ERR(filp))
510 kfree(ev_file);
511
512 return filp;
513}
514
515/*
516 * Look up a completion event file by FD. If lookup is successful,
517 * takes a ref to the event file struct that it returns; if
518 * unsuccessful, returns NULL.
519 */
520struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
521{
522 struct ib_uverbs_event_file *ev_file = NULL;
523 struct file *filp;
524
525 filp = fget(fd);
526 if (!filp)
527 return NULL;
528
529 if (filp->f_op != &uverbs_event_fops)
530 goto out;
531
532 ev_file = filp->private_data;
533 if (ev_file->is_async) {
534 ev_file = NULL;
535 goto out;
536 }
537
538 kref_get(&ev_file->ref);
539
540out:
541 fput(filp);
542 return ev_file;
543}
544
545static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
546 size_t count, loff_t *pos)
547{
548 struct ib_uverbs_file *file = filp->private_data;
549 struct ib_uverbs_cmd_hdr hdr;
550
551 if (count < sizeof hdr)
552 return -EINVAL;
553
554 if (copy_from_user(&hdr, buf, sizeof hdr))
555 return -EFAULT;
556
557 if (hdr.in_words * 4 != count)
558 return -EINVAL;
559
560 if (hdr.command < 0 ||
561 hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
562 !uverbs_cmd_table[hdr.command])
563 return -EINVAL;
564
565 if (!file->ucontext &&
566 hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
567 return -EINVAL;
568
569 if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
570 return -ENOSYS;
571
572 return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
573 hdr.in_words * 4, hdr.out_words * 4);
574}
575
576static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
577{
578 struct ib_uverbs_file *file = filp->private_data;
579
580 if (!file->ucontext)
581 return -ENODEV;
582 else
583 return file->device->ib_dev->mmap(file->ucontext, vma);
584}
585
586/*
587 * ib_uverbs_open() does not need the BKL:
588 *
589 * - the ib_uverbs_device structures are properly reference counted and
590 * everything else is purely local to the file being created, so
591 * races against other open calls are not a problem;
592 * - there is no ioctl method to race against;
593 * - the open method will either immediately run -ENXIO, or all
594 * required initialization will be done.
595 */
596static int ib_uverbs_open(struct inode *inode, struct file *filp)
597{
598 struct ib_uverbs_device *dev;
599 struct ib_uverbs_file *file;
600 int ret;
601
602 dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
603 if (dev)
604 kref_get(&dev->ref);
605 else
606 return -ENXIO;
607
608 if (!try_module_get(dev->ib_dev->owner)) {
609 ret = -ENODEV;
610 goto err;
611 }
612
613 file = kmalloc(sizeof *file, GFP_KERNEL);
614 if (!file) {
615 ret = -ENOMEM;
616 goto err_module;
617 }
618
619 file->device = dev;
620 file->ucontext = NULL;
621 file->async_file = NULL;
622 kref_init(&file->ref);
623 mutex_init(&file->mutex);
624
625 filp->private_data = file;
626
627 return nonseekable_open(inode, filp);
628
629err_module:
630 module_put(dev->ib_dev->owner);
631
632err:
633 kref_put(&dev->ref, ib_uverbs_release_dev);
634 return ret;
635}
636
637static int ib_uverbs_close(struct inode *inode, struct file *filp)
638{
639 struct ib_uverbs_file *file = filp->private_data;
640
641 ib_uverbs_cleanup_ucontext(file, file->ucontext);
642
643 if (file->async_file)
644 kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
645
646 kref_put(&file->ref, ib_uverbs_release_file);
647
648 return 0;
649}
650
651static const struct file_operations uverbs_fops = {
652 .owner = THIS_MODULE,
653 .write = ib_uverbs_write,
654 .open = ib_uverbs_open,
655 .release = ib_uverbs_close,
656 .llseek = no_llseek,
657};
658
659static const struct file_operations uverbs_mmap_fops = {
660 .owner = THIS_MODULE,
661 .write = ib_uverbs_write,
662 .mmap = ib_uverbs_mmap,
663 .open = ib_uverbs_open,
664 .release = ib_uverbs_close,
665 .llseek = no_llseek,
666};
667
668static struct ib_client uverbs_client = {
669 .name = "uverbs",
670 .add = ib_uverbs_add_one,
671 .remove = ib_uverbs_remove_one
672};
673
674static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
675 char *buf)
676{
677 struct ib_uverbs_device *dev = dev_get_drvdata(device);
678
679 if (!dev)
680 return -ENODEV;
681
682 return sprintf(buf, "%s\n", dev->ib_dev->name);
683}
684static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
685
686static ssize_t show_dev_abi_version(struct device *device,
687 struct device_attribute *attr, char *buf)
688{
689 struct ib_uverbs_device *dev = dev_get_drvdata(device);
690
691 if (!dev)
692 return -ENODEV;
693
694 return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
695}
696static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
697
698static CLASS_ATTR_STRING(abi_version, S_IRUGO,
699 __stringify(IB_USER_VERBS_ABI_VERSION));
700
701static dev_t overflow_maj;
702static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES);
703
704/*
705 * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by
706 * requesting a new major number and doubling the number of max devices we
707 * support. It's stupid, but simple.
708 */
709static int find_overflow_devnum(void)
710{
711 int ret;
712
713 if (!overflow_maj) {
714 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES,
715 "infiniband_verbs");
716 if (ret) {
717 printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n");
718 return ret;
719 }
720 }
721
722 ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES);
723 if (ret >= IB_UVERBS_MAX_DEVICES)
724 return -1;
725
726 return ret;
727}
728
729static void ib_uverbs_add_one(struct ib_device *device)
730{
731 int devnum;
732 dev_t base;
733 struct ib_uverbs_device *uverbs_dev;
734
735 if (!device->alloc_ucontext)
736 return;
737
738 uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
739 if (!uverbs_dev)
740 return;
741
742 kref_init(&uverbs_dev->ref);
743 init_completion(&uverbs_dev->comp);
744
745 spin_lock(&map_lock);
746 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
747 if (devnum >= IB_UVERBS_MAX_DEVICES) {
748 spin_unlock(&map_lock);
749 devnum = find_overflow_devnum();
750 if (devnum < 0)
751 goto err;
752
753 spin_lock(&map_lock);
754 uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES;
755 base = devnum + overflow_maj;
756 set_bit(devnum, overflow_map);
757 } else {
758 uverbs_dev->devnum = devnum;
759 base = devnum + IB_UVERBS_BASE_DEV;
760 set_bit(devnum, dev_map);
761 }
762 spin_unlock(&map_lock);
763
764 uverbs_dev->ib_dev = device;
765 uverbs_dev->num_comp_vectors = device->num_comp_vectors;
766
767 cdev_init(&uverbs_dev->cdev, NULL);
768 uverbs_dev->cdev.owner = THIS_MODULE;
769 uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
770 kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
771 if (cdev_add(&uverbs_dev->cdev, base, 1))
772 goto err_cdev;
773
774 uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
775 uverbs_dev->cdev.dev, uverbs_dev,
776 "uverbs%d", uverbs_dev->devnum);
777 if (IS_ERR(uverbs_dev->dev))
778 goto err_cdev;
779
780 if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
781 goto err_class;
782 if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
783 goto err_class;
784
785 ib_set_client_data(device, &uverbs_client, uverbs_dev);
786
787 return;
788
789err_class:
790 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
791
792err_cdev:
793 cdev_del(&uverbs_dev->cdev);
794 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
795 clear_bit(devnum, dev_map);
796 else
797 clear_bit(devnum, overflow_map);
798
799err:
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
806static 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 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
818 clear_bit(uverbs_dev->devnum, dev_map);
819 else
820 clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map);
821
822 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
823 wait_for_completion(&uverbs_dev->comp);
824 kfree(uverbs_dev);
825}
826
827static char *uverbs_devnode(struct device *dev, mode_t *mode)
828{
829 if (mode)
830 *mode = 0666;
831 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
832}
833
834static int __init ib_uverbs_init(void)
835{
836 int ret;
837
838 ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
839 "infiniband_verbs");
840 if (ret) {
841 printk(KERN_ERR "user_verbs: couldn't register device number\n");
842 goto out;
843 }
844
845 uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
846 if (IS_ERR(uverbs_class)) {
847 ret = PTR_ERR(uverbs_class);
848 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
849 goto out_chrdev;
850 }
851
852 uverbs_class->devnode = uverbs_devnode;
853
854 ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
855 if (ret) {
856 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
857 goto out_class;
858 }
859
860 ret = ib_register_client(&uverbs_client);
861 if (ret) {
862 printk(KERN_ERR "user_verbs: couldn't register client\n");
863 goto out_class;
864 }
865
866 return 0;
867
868out_class:
869 class_destroy(uverbs_class);
870
871out_chrdev:
872 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
873
874out:
875 return ret;
876}
877
878static void __exit ib_uverbs_cleanup(void)
879{
880 ib_unregister_client(&uverbs_client);
881 class_destroy(uverbs_class);
882 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
883 if (overflow_maj)
884 unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES);
885 idr_destroy(&ib_uverbs_pd_idr);
886 idr_destroy(&ib_uverbs_mr_idr);
887 idr_destroy(&ib_uverbs_mw_idr);
888 idr_destroy(&ib_uverbs_ah_idr);
889 idr_destroy(&ib_uverbs_cq_idr);
890 idr_destroy(&ib_uverbs_qp_idr);
891 idr_destroy(&ib_uverbs_srq_idr);
892}
893
894module_init(ib_uverbs_init);
895module_exit(ib_uverbs_cleanup);
1/*
2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
6 * Copyright (c) 2005 PathScale, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
17 *
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
21 *
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 */
36
37#include <linux/module.h>
38#include <linux/init.h>
39#include <linux/device.h>
40#include <linux/err.h>
41#include <linux/fs.h>
42#include <linux/poll.h>
43#include <linux/sched.h>
44#include <linux/file.h>
45#include <linux/cdev.h>
46#include <linux/anon_inodes.h>
47#include <linux/slab.h>
48
49#include <linux/uaccess.h>
50
51#include <rdma/ib.h>
52#include <rdma/uverbs_std_types.h>
53
54#include "uverbs.h"
55#include "core_priv.h"
56#include "rdma_core.h"
57
58MODULE_AUTHOR("Roland Dreier");
59MODULE_DESCRIPTION("InfiniBand userspace verbs access");
60MODULE_LICENSE("Dual BSD/GPL");
61
62enum {
63 IB_UVERBS_MAJOR = 231,
64 IB_UVERBS_BASE_MINOR = 192,
65 IB_UVERBS_MAX_DEVICES = RDMA_MAX_PORTS,
66 IB_UVERBS_NUM_FIXED_MINOR = 32,
67 IB_UVERBS_NUM_DYNAMIC_MINOR = IB_UVERBS_MAX_DEVICES - IB_UVERBS_NUM_FIXED_MINOR,
68};
69
70#define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
71
72static dev_t dynamic_uverbs_dev;
73static struct class *uverbs_class;
74
75static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
76
77static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
78 struct ib_device *ib_dev,
79 const char __user *buf, int in_len,
80 int out_len) = {
81 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
82 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
83 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
84 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
85 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
86 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
87 [IB_USER_VERBS_CMD_REREG_MR] = ib_uverbs_rereg_mr,
88 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
89 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw,
90 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw,
91 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
92 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
93 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq,
94 [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq,
95 [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq,
96 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
97 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
98 [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp,
99 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
100 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
101 [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send,
102 [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv,
103 [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv,
104 [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah,
105 [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah,
106 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
107 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
108 [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
109 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
110 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq,
111 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
112 [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd,
113 [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd,
114 [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq,
115 [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp,
116};
117
118static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file,
119 struct ib_device *ib_dev,
120 struct ib_udata *ucore,
121 struct ib_udata *uhw) = {
122 [IB_USER_VERBS_EX_CMD_CREATE_FLOW] = ib_uverbs_ex_create_flow,
123 [IB_USER_VERBS_EX_CMD_DESTROY_FLOW] = ib_uverbs_ex_destroy_flow,
124 [IB_USER_VERBS_EX_CMD_QUERY_DEVICE] = ib_uverbs_ex_query_device,
125 [IB_USER_VERBS_EX_CMD_CREATE_CQ] = ib_uverbs_ex_create_cq,
126 [IB_USER_VERBS_EX_CMD_CREATE_QP] = ib_uverbs_ex_create_qp,
127 [IB_USER_VERBS_EX_CMD_CREATE_WQ] = ib_uverbs_ex_create_wq,
128 [IB_USER_VERBS_EX_CMD_MODIFY_WQ] = ib_uverbs_ex_modify_wq,
129 [IB_USER_VERBS_EX_CMD_DESTROY_WQ] = ib_uverbs_ex_destroy_wq,
130 [IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL] = ib_uverbs_ex_create_rwq_ind_table,
131 [IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL] = ib_uverbs_ex_destroy_rwq_ind_table,
132 [IB_USER_VERBS_EX_CMD_MODIFY_QP] = ib_uverbs_ex_modify_qp,
133 [IB_USER_VERBS_EX_CMD_MODIFY_CQ] = ib_uverbs_ex_modify_cq,
134};
135
136static void ib_uverbs_add_one(struct ib_device *device);
137static void ib_uverbs_remove_one(struct ib_device *device, void *client_data);
138
139int uverbs_dealloc_mw(struct ib_mw *mw)
140{
141 struct ib_pd *pd = mw->pd;
142 int ret;
143
144 ret = mw->device->dealloc_mw(mw);
145 if (!ret)
146 atomic_dec(&pd->usecnt);
147 return ret;
148}
149
150static void ib_uverbs_release_dev(struct kobject *kobj)
151{
152 struct ib_uverbs_device *dev =
153 container_of(kobj, struct ib_uverbs_device, kobj);
154
155 cleanup_srcu_struct(&dev->disassociate_srcu);
156 kfree(dev);
157}
158
159static struct kobj_type ib_uverbs_dev_ktype = {
160 .release = ib_uverbs_release_dev,
161};
162
163static void ib_uverbs_release_async_event_file(struct kref *ref)
164{
165 struct ib_uverbs_async_event_file *file =
166 container_of(ref, struct ib_uverbs_async_event_file, ref);
167
168 kfree(file);
169}
170
171void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
172 struct ib_uverbs_completion_event_file *ev_file,
173 struct ib_ucq_object *uobj)
174{
175 struct ib_uverbs_event *evt, *tmp;
176
177 if (ev_file) {
178 spin_lock_irq(&ev_file->ev_queue.lock);
179 list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
180 list_del(&evt->list);
181 kfree(evt);
182 }
183 spin_unlock_irq(&ev_file->ev_queue.lock);
184
185 uverbs_uobject_put(&ev_file->uobj_file.uobj);
186 }
187
188 spin_lock_irq(&file->async_file->ev_queue.lock);
189 list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
190 list_del(&evt->list);
191 kfree(evt);
192 }
193 spin_unlock_irq(&file->async_file->ev_queue.lock);
194}
195
196void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
197 struct ib_uevent_object *uobj)
198{
199 struct ib_uverbs_event *evt, *tmp;
200
201 spin_lock_irq(&file->async_file->ev_queue.lock);
202 list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
203 list_del(&evt->list);
204 kfree(evt);
205 }
206 spin_unlock_irq(&file->async_file->ev_queue.lock);
207}
208
209void ib_uverbs_detach_umcast(struct ib_qp *qp,
210 struct ib_uqp_object *uobj)
211{
212 struct ib_uverbs_mcast_entry *mcast, *tmp;
213
214 list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
215 ib_detach_mcast(qp, &mcast->gid, mcast->lid);
216 list_del(&mcast->list);
217 kfree(mcast);
218 }
219}
220
221static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
222 struct ib_ucontext *context,
223 bool device_removed)
224{
225 context->closing = 1;
226 uverbs_cleanup_ucontext(context, device_removed);
227 put_pid(context->tgid);
228
229 ib_rdmacg_uncharge(&context->cg_obj, context->device,
230 RDMACG_RESOURCE_HCA_HANDLE);
231
232 return context->device->dealloc_ucontext(context);
233}
234
235static void ib_uverbs_comp_dev(struct ib_uverbs_device *dev)
236{
237 complete(&dev->comp);
238}
239
240void ib_uverbs_release_file(struct kref *ref)
241{
242 struct ib_uverbs_file *file =
243 container_of(ref, struct ib_uverbs_file, ref);
244 struct ib_device *ib_dev;
245 int srcu_key;
246
247 srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
248 ib_dev = srcu_dereference(file->device->ib_dev,
249 &file->device->disassociate_srcu);
250 if (ib_dev && !ib_dev->disassociate_ucontext)
251 module_put(ib_dev->owner);
252 srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
253
254 if (atomic_dec_and_test(&file->device->refcount))
255 ib_uverbs_comp_dev(file->device);
256
257 kobject_put(&file->device->kobj);
258 kfree(file);
259}
260
261static ssize_t ib_uverbs_event_read(struct ib_uverbs_event_queue *ev_queue,
262 struct ib_uverbs_file *uverbs_file,
263 struct file *filp, char __user *buf,
264 size_t count, loff_t *pos,
265 size_t eventsz)
266{
267 struct ib_uverbs_event *event;
268 int ret = 0;
269
270 spin_lock_irq(&ev_queue->lock);
271
272 while (list_empty(&ev_queue->event_list)) {
273 spin_unlock_irq(&ev_queue->lock);
274
275 if (filp->f_flags & O_NONBLOCK)
276 return -EAGAIN;
277
278 if (wait_event_interruptible(ev_queue->poll_wait,
279 (!list_empty(&ev_queue->event_list) ||
280 /* The barriers built into wait_event_interruptible()
281 * and wake_up() guarentee this will see the null set
282 * without using RCU
283 */
284 !uverbs_file->device->ib_dev)))
285 return -ERESTARTSYS;
286
287 /* If device was disassociated and no event exists set an error */
288 if (list_empty(&ev_queue->event_list) &&
289 !uverbs_file->device->ib_dev)
290 return -EIO;
291
292 spin_lock_irq(&ev_queue->lock);
293 }
294
295 event = list_entry(ev_queue->event_list.next, struct ib_uverbs_event, list);
296
297 if (eventsz > count) {
298 ret = -EINVAL;
299 event = NULL;
300 } else {
301 list_del(ev_queue->event_list.next);
302 if (event->counter) {
303 ++(*event->counter);
304 list_del(&event->obj_list);
305 }
306 }
307
308 spin_unlock_irq(&ev_queue->lock);
309
310 if (event) {
311 if (copy_to_user(buf, event, eventsz))
312 ret = -EFAULT;
313 else
314 ret = eventsz;
315 }
316
317 kfree(event);
318
319 return ret;
320}
321
322static ssize_t ib_uverbs_async_event_read(struct file *filp, char __user *buf,
323 size_t count, loff_t *pos)
324{
325 struct ib_uverbs_async_event_file *file = filp->private_data;
326
327 return ib_uverbs_event_read(&file->ev_queue, file->uverbs_file, filp,
328 buf, count, pos,
329 sizeof(struct ib_uverbs_async_event_desc));
330}
331
332static ssize_t ib_uverbs_comp_event_read(struct file *filp, char __user *buf,
333 size_t count, loff_t *pos)
334{
335 struct ib_uverbs_completion_event_file *comp_ev_file =
336 filp->private_data;
337
338 return ib_uverbs_event_read(&comp_ev_file->ev_queue,
339 comp_ev_file->uobj_file.ufile, filp,
340 buf, count, pos,
341 sizeof(struct ib_uverbs_comp_event_desc));
342}
343
344static __poll_t ib_uverbs_event_poll(struct ib_uverbs_event_queue *ev_queue,
345 struct file *filp,
346 struct poll_table_struct *wait)
347{
348 __poll_t pollflags = 0;
349
350 poll_wait(filp, &ev_queue->poll_wait, wait);
351
352 spin_lock_irq(&ev_queue->lock);
353 if (!list_empty(&ev_queue->event_list))
354 pollflags = EPOLLIN | EPOLLRDNORM;
355 spin_unlock_irq(&ev_queue->lock);
356
357 return pollflags;
358}
359
360static __poll_t ib_uverbs_async_event_poll(struct file *filp,
361 struct poll_table_struct *wait)
362{
363 return ib_uverbs_event_poll(filp->private_data, filp, wait);
364}
365
366static __poll_t ib_uverbs_comp_event_poll(struct file *filp,
367 struct poll_table_struct *wait)
368{
369 struct ib_uverbs_completion_event_file *comp_ev_file =
370 filp->private_data;
371
372 return ib_uverbs_event_poll(&comp_ev_file->ev_queue, filp, wait);
373}
374
375static int ib_uverbs_async_event_fasync(int fd, struct file *filp, int on)
376{
377 struct ib_uverbs_event_queue *ev_queue = filp->private_data;
378
379 return fasync_helper(fd, filp, on, &ev_queue->async_queue);
380}
381
382static int ib_uverbs_comp_event_fasync(int fd, struct file *filp, int on)
383{
384 struct ib_uverbs_completion_event_file *comp_ev_file =
385 filp->private_data;
386
387 return fasync_helper(fd, filp, on, &comp_ev_file->ev_queue.async_queue);
388}
389
390static int ib_uverbs_async_event_close(struct inode *inode, struct file *filp)
391{
392 struct ib_uverbs_async_event_file *file = filp->private_data;
393 struct ib_uverbs_file *uverbs_file = file->uverbs_file;
394 struct ib_uverbs_event *entry, *tmp;
395 int closed_already = 0;
396
397 mutex_lock(&uverbs_file->device->lists_mutex);
398 spin_lock_irq(&file->ev_queue.lock);
399 closed_already = file->ev_queue.is_closed;
400 file->ev_queue.is_closed = 1;
401 list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) {
402 if (entry->counter)
403 list_del(&entry->obj_list);
404 kfree(entry);
405 }
406 spin_unlock_irq(&file->ev_queue.lock);
407 if (!closed_already) {
408 list_del(&file->list);
409 ib_unregister_event_handler(&uverbs_file->event_handler);
410 }
411 mutex_unlock(&uverbs_file->device->lists_mutex);
412
413 kref_put(&uverbs_file->ref, ib_uverbs_release_file);
414 kref_put(&file->ref, ib_uverbs_release_async_event_file);
415
416 return 0;
417}
418
419static int ib_uverbs_comp_event_close(struct inode *inode, struct file *filp)
420{
421 struct ib_uverbs_completion_event_file *file = filp->private_data;
422 struct ib_uverbs_event *entry, *tmp;
423
424 spin_lock_irq(&file->ev_queue.lock);
425 list_for_each_entry_safe(entry, tmp, &file->ev_queue.event_list, list) {
426 if (entry->counter)
427 list_del(&entry->obj_list);
428 kfree(entry);
429 }
430 spin_unlock_irq(&file->ev_queue.lock);
431
432 uverbs_close_fd(filp);
433
434 return 0;
435}
436
437const struct file_operations uverbs_event_fops = {
438 .owner = THIS_MODULE,
439 .read = ib_uverbs_comp_event_read,
440 .poll = ib_uverbs_comp_event_poll,
441 .release = ib_uverbs_comp_event_close,
442 .fasync = ib_uverbs_comp_event_fasync,
443 .llseek = no_llseek,
444};
445
446static const struct file_operations uverbs_async_event_fops = {
447 .owner = THIS_MODULE,
448 .read = ib_uverbs_async_event_read,
449 .poll = ib_uverbs_async_event_poll,
450 .release = ib_uverbs_async_event_close,
451 .fasync = ib_uverbs_async_event_fasync,
452 .llseek = no_llseek,
453};
454
455void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
456{
457 struct ib_uverbs_event_queue *ev_queue = cq_context;
458 struct ib_ucq_object *uobj;
459 struct ib_uverbs_event *entry;
460 unsigned long flags;
461
462 if (!ev_queue)
463 return;
464
465 spin_lock_irqsave(&ev_queue->lock, flags);
466 if (ev_queue->is_closed) {
467 spin_unlock_irqrestore(&ev_queue->lock, flags);
468 return;
469 }
470
471 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
472 if (!entry) {
473 spin_unlock_irqrestore(&ev_queue->lock, flags);
474 return;
475 }
476
477 uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
478
479 entry->desc.comp.cq_handle = cq->uobject->user_handle;
480 entry->counter = &uobj->comp_events_reported;
481
482 list_add_tail(&entry->list, &ev_queue->event_list);
483 list_add_tail(&entry->obj_list, &uobj->comp_list);
484 spin_unlock_irqrestore(&ev_queue->lock, flags);
485
486 wake_up_interruptible(&ev_queue->poll_wait);
487 kill_fasync(&ev_queue->async_queue, SIGIO, POLL_IN);
488}
489
490static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
491 __u64 element, __u64 event,
492 struct list_head *obj_list,
493 u32 *counter)
494{
495 struct ib_uverbs_event *entry;
496 unsigned long flags;
497
498 spin_lock_irqsave(&file->async_file->ev_queue.lock, flags);
499 if (file->async_file->ev_queue.is_closed) {
500 spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
501 return;
502 }
503
504 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
505 if (!entry) {
506 spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
507 return;
508 }
509
510 entry->desc.async.element = element;
511 entry->desc.async.event_type = event;
512 entry->desc.async.reserved = 0;
513 entry->counter = counter;
514
515 list_add_tail(&entry->list, &file->async_file->ev_queue.event_list);
516 if (obj_list)
517 list_add_tail(&entry->obj_list, obj_list);
518 spin_unlock_irqrestore(&file->async_file->ev_queue.lock, flags);
519
520 wake_up_interruptible(&file->async_file->ev_queue.poll_wait);
521 kill_fasync(&file->async_file->ev_queue.async_queue, SIGIO, POLL_IN);
522}
523
524void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
525{
526 struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
527 struct ib_ucq_object, uobject);
528
529 ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
530 event->event, &uobj->async_list,
531 &uobj->async_events_reported);
532}
533
534void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
535{
536 struct ib_uevent_object *uobj;
537
538 /* for XRC target qp's, check that qp is live */
539 if (!event->element.qp->uobject)
540 return;
541
542 uobj = container_of(event->element.qp->uobject,
543 struct ib_uevent_object, uobject);
544
545 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
546 event->event, &uobj->event_list,
547 &uobj->events_reported);
548}
549
550void ib_uverbs_wq_event_handler(struct ib_event *event, void *context_ptr)
551{
552 struct ib_uevent_object *uobj = container_of(event->element.wq->uobject,
553 struct ib_uevent_object, uobject);
554
555 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
556 event->event, &uobj->event_list,
557 &uobj->events_reported);
558}
559
560void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
561{
562 struct ib_uevent_object *uobj;
563
564 uobj = container_of(event->element.srq->uobject,
565 struct ib_uevent_object, uobject);
566
567 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
568 event->event, &uobj->event_list,
569 &uobj->events_reported);
570}
571
572void ib_uverbs_event_handler(struct ib_event_handler *handler,
573 struct ib_event *event)
574{
575 struct ib_uverbs_file *file =
576 container_of(handler, struct ib_uverbs_file, event_handler);
577
578 ib_uverbs_async_handler(file, event->element.port_num, event->event,
579 NULL, NULL);
580}
581
582void ib_uverbs_free_async_event_file(struct ib_uverbs_file *file)
583{
584 kref_put(&file->async_file->ref, ib_uverbs_release_async_event_file);
585 file->async_file = NULL;
586}
587
588void ib_uverbs_init_event_queue(struct ib_uverbs_event_queue *ev_queue)
589{
590 spin_lock_init(&ev_queue->lock);
591 INIT_LIST_HEAD(&ev_queue->event_list);
592 init_waitqueue_head(&ev_queue->poll_wait);
593 ev_queue->is_closed = 0;
594 ev_queue->async_queue = NULL;
595}
596
597struct file *ib_uverbs_alloc_async_event_file(struct ib_uverbs_file *uverbs_file,
598 struct ib_device *ib_dev)
599{
600 struct ib_uverbs_async_event_file *ev_file;
601 struct file *filp;
602
603 ev_file = kzalloc(sizeof(*ev_file), GFP_KERNEL);
604 if (!ev_file)
605 return ERR_PTR(-ENOMEM);
606
607 ib_uverbs_init_event_queue(&ev_file->ev_queue);
608 ev_file->uverbs_file = uverbs_file;
609 kref_get(&ev_file->uverbs_file->ref);
610 kref_init(&ev_file->ref);
611 filp = anon_inode_getfile("[infinibandevent]", &uverbs_async_event_fops,
612 ev_file, O_RDONLY);
613 if (IS_ERR(filp))
614 goto err_put_refs;
615
616 mutex_lock(&uverbs_file->device->lists_mutex);
617 list_add_tail(&ev_file->list,
618 &uverbs_file->device->uverbs_events_file_list);
619 mutex_unlock(&uverbs_file->device->lists_mutex);
620
621 WARN_ON(uverbs_file->async_file);
622 uverbs_file->async_file = ev_file;
623 kref_get(&uverbs_file->async_file->ref);
624 INIT_IB_EVENT_HANDLER(&uverbs_file->event_handler,
625 ib_dev,
626 ib_uverbs_event_handler);
627 ib_register_event_handler(&uverbs_file->event_handler);
628 /* At that point async file stuff was fully set */
629
630 return filp;
631
632err_put_refs:
633 kref_put(&ev_file->uverbs_file->ref, ib_uverbs_release_file);
634 kref_put(&ev_file->ref, ib_uverbs_release_async_event_file);
635 return filp;
636}
637
638static bool verify_command_mask(struct ib_device *ib_dev,
639 u32 command, bool extended)
640{
641 if (!extended)
642 return ib_dev->uverbs_cmd_mask & BIT_ULL(command);
643
644 return ib_dev->uverbs_ex_cmd_mask & BIT_ULL(command);
645}
646
647static bool verify_command_idx(u32 command, bool extended)
648{
649 if (extended)
650 return command < ARRAY_SIZE(uverbs_ex_cmd_table) &&
651 uverbs_ex_cmd_table[command];
652
653 return command < ARRAY_SIZE(uverbs_cmd_table) &&
654 uverbs_cmd_table[command];
655}
656
657static ssize_t process_hdr(struct ib_uverbs_cmd_hdr *hdr,
658 u32 *command, bool *extended)
659{
660 if (hdr->command & ~(u32)(IB_USER_VERBS_CMD_FLAG_EXTENDED |
661 IB_USER_VERBS_CMD_COMMAND_MASK))
662 return -EINVAL;
663
664 *command = hdr->command & IB_USER_VERBS_CMD_COMMAND_MASK;
665 *extended = hdr->command & IB_USER_VERBS_CMD_FLAG_EXTENDED;
666
667 if (!verify_command_idx(*command, *extended))
668 return -EOPNOTSUPP;
669
670 return 0;
671}
672
673static ssize_t verify_hdr(struct ib_uverbs_cmd_hdr *hdr,
674 struct ib_uverbs_ex_cmd_hdr *ex_hdr,
675 size_t count, bool extended)
676{
677 if (extended) {
678 count -= sizeof(*hdr) + sizeof(*ex_hdr);
679
680 if ((hdr->in_words + ex_hdr->provider_in_words) * 8 != count)
681 return -EINVAL;
682
683 if (ex_hdr->cmd_hdr_reserved)
684 return -EINVAL;
685
686 if (ex_hdr->response) {
687 if (!hdr->out_words && !ex_hdr->provider_out_words)
688 return -EINVAL;
689
690 if (!access_ok(VERIFY_WRITE,
691 u64_to_user_ptr(ex_hdr->response),
692 (hdr->out_words + ex_hdr->provider_out_words) * 8))
693 return -EFAULT;
694 } else {
695 if (hdr->out_words || ex_hdr->provider_out_words)
696 return -EINVAL;
697 }
698
699 return 0;
700 }
701
702 /* not extended command */
703 if (hdr->in_words * 4 != count)
704 return -EINVAL;
705
706 return 0;
707}
708
709static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
710 size_t count, loff_t *pos)
711{
712 struct ib_uverbs_file *file = filp->private_data;
713 struct ib_uverbs_ex_cmd_hdr ex_hdr;
714 struct ib_device *ib_dev;
715 struct ib_uverbs_cmd_hdr hdr;
716 bool extended;
717 int srcu_key;
718 u32 command;
719 ssize_t ret;
720
721 if (!ib_safe_file_access(filp)) {
722 pr_err_once("uverbs_write: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
723 task_tgid_vnr(current), current->comm);
724 return -EACCES;
725 }
726
727 if (count < sizeof(hdr))
728 return -EINVAL;
729
730 if (copy_from_user(&hdr, buf, sizeof(hdr)))
731 return -EFAULT;
732
733 ret = process_hdr(&hdr, &command, &extended);
734 if (ret)
735 return ret;
736
737 if (!file->ucontext &&
738 (command != IB_USER_VERBS_CMD_GET_CONTEXT || extended))
739 return -EINVAL;
740
741 if (extended) {
742 if (count < (sizeof(hdr) + sizeof(ex_hdr)))
743 return -EINVAL;
744 if (copy_from_user(&ex_hdr, buf + sizeof(hdr), sizeof(ex_hdr)))
745 return -EFAULT;
746 }
747
748 ret = verify_hdr(&hdr, &ex_hdr, count, extended);
749 if (ret)
750 return ret;
751
752 srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
753 ib_dev = srcu_dereference(file->device->ib_dev,
754 &file->device->disassociate_srcu);
755 if (!ib_dev) {
756 ret = -EIO;
757 goto out;
758 }
759
760 if (!verify_command_mask(ib_dev, command, extended)) {
761 ret = -EOPNOTSUPP;
762 goto out;
763 }
764
765 buf += sizeof(hdr);
766
767 if (!extended) {
768 ret = uverbs_cmd_table[command](file, ib_dev, buf,
769 hdr.in_words * 4,
770 hdr.out_words * 4);
771 } else {
772 struct ib_udata ucore;
773 struct ib_udata uhw;
774
775 buf += sizeof(ex_hdr);
776
777 ib_uverbs_init_udata_buf_or_null(&ucore, buf,
778 u64_to_user_ptr(ex_hdr.response),
779 hdr.in_words * 8, hdr.out_words * 8);
780
781 ib_uverbs_init_udata_buf_or_null(&uhw,
782 buf + ucore.inlen,
783 u64_to_user_ptr(ex_hdr.response) + ucore.outlen,
784 ex_hdr.provider_in_words * 8,
785 ex_hdr.provider_out_words * 8);
786
787 ret = uverbs_ex_cmd_table[command](file, ib_dev, &ucore, &uhw);
788 ret = (ret) ? : count;
789 }
790
791out:
792 srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
793 return ret;
794}
795
796static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
797{
798 struct ib_uverbs_file *file = filp->private_data;
799 struct ib_device *ib_dev;
800 int ret = 0;
801 int srcu_key;
802
803 srcu_key = srcu_read_lock(&file->device->disassociate_srcu);
804 ib_dev = srcu_dereference(file->device->ib_dev,
805 &file->device->disassociate_srcu);
806 if (!ib_dev) {
807 ret = -EIO;
808 goto out;
809 }
810
811 if (!file->ucontext)
812 ret = -ENODEV;
813 else
814 ret = ib_dev->mmap(file->ucontext, vma);
815out:
816 srcu_read_unlock(&file->device->disassociate_srcu, srcu_key);
817 return ret;
818}
819
820/*
821 * ib_uverbs_open() does not need the BKL:
822 *
823 * - the ib_uverbs_device structures are properly reference counted and
824 * everything else is purely local to the file being created, so
825 * races against other open calls are not a problem;
826 * - there is no ioctl method to race against;
827 * - the open method will either immediately run -ENXIO, or all
828 * required initialization will be done.
829 */
830static int ib_uverbs_open(struct inode *inode, struct file *filp)
831{
832 struct ib_uverbs_device *dev;
833 struct ib_uverbs_file *file;
834 struct ib_device *ib_dev;
835 int ret;
836 int module_dependent;
837 int srcu_key;
838
839 dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
840 if (!atomic_inc_not_zero(&dev->refcount))
841 return -ENXIO;
842
843 srcu_key = srcu_read_lock(&dev->disassociate_srcu);
844 mutex_lock(&dev->lists_mutex);
845 ib_dev = srcu_dereference(dev->ib_dev,
846 &dev->disassociate_srcu);
847 if (!ib_dev) {
848 ret = -EIO;
849 goto err;
850 }
851
852 /* In case IB device supports disassociate ucontext, there is no hard
853 * dependency between uverbs device and its low level device.
854 */
855 module_dependent = !(ib_dev->disassociate_ucontext);
856
857 if (module_dependent) {
858 if (!try_module_get(ib_dev->owner)) {
859 ret = -ENODEV;
860 goto err;
861 }
862 }
863
864 file = kzalloc(sizeof(*file), GFP_KERNEL);
865 if (!file) {
866 ret = -ENOMEM;
867 if (module_dependent)
868 goto err_module;
869
870 goto err;
871 }
872
873 file->device = dev;
874 spin_lock_init(&file->idr_lock);
875 idr_init(&file->idr);
876 file->ucontext = NULL;
877 file->async_file = NULL;
878 kref_init(&file->ref);
879 mutex_init(&file->mutex);
880 mutex_init(&file->cleanup_mutex);
881
882 filp->private_data = file;
883 kobject_get(&dev->kobj);
884 list_add_tail(&file->list, &dev->uverbs_file_list);
885 mutex_unlock(&dev->lists_mutex);
886 srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
887
888 return nonseekable_open(inode, filp);
889
890err_module:
891 module_put(ib_dev->owner);
892
893err:
894 mutex_unlock(&dev->lists_mutex);
895 srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
896 if (atomic_dec_and_test(&dev->refcount))
897 ib_uverbs_comp_dev(dev);
898
899 return ret;
900}
901
902static int ib_uverbs_close(struct inode *inode, struct file *filp)
903{
904 struct ib_uverbs_file *file = filp->private_data;
905
906 mutex_lock(&file->cleanup_mutex);
907 if (file->ucontext) {
908 ib_uverbs_cleanup_ucontext(file, file->ucontext, false);
909 file->ucontext = NULL;
910 }
911 mutex_unlock(&file->cleanup_mutex);
912 idr_destroy(&file->idr);
913
914 mutex_lock(&file->device->lists_mutex);
915 if (!file->is_closed) {
916 list_del(&file->list);
917 file->is_closed = 1;
918 }
919 mutex_unlock(&file->device->lists_mutex);
920
921 if (file->async_file)
922 kref_put(&file->async_file->ref,
923 ib_uverbs_release_async_event_file);
924
925 kref_put(&file->ref, ib_uverbs_release_file);
926
927 return 0;
928}
929
930static const struct file_operations uverbs_fops = {
931 .owner = THIS_MODULE,
932 .write = ib_uverbs_write,
933 .open = ib_uverbs_open,
934 .release = ib_uverbs_close,
935 .llseek = no_llseek,
936 .unlocked_ioctl = ib_uverbs_ioctl,
937 .compat_ioctl = ib_uverbs_ioctl,
938};
939
940static const struct file_operations uverbs_mmap_fops = {
941 .owner = THIS_MODULE,
942 .write = ib_uverbs_write,
943 .mmap = ib_uverbs_mmap,
944 .open = ib_uverbs_open,
945 .release = ib_uverbs_close,
946 .llseek = no_llseek,
947 .unlocked_ioctl = ib_uverbs_ioctl,
948 .compat_ioctl = ib_uverbs_ioctl,
949};
950
951static struct ib_client uverbs_client = {
952 .name = "uverbs",
953 .add = ib_uverbs_add_one,
954 .remove = ib_uverbs_remove_one
955};
956
957static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
958 char *buf)
959{
960 int ret = -ENODEV;
961 int srcu_key;
962 struct ib_uverbs_device *dev = dev_get_drvdata(device);
963 struct ib_device *ib_dev;
964
965 if (!dev)
966 return -ENODEV;
967
968 srcu_key = srcu_read_lock(&dev->disassociate_srcu);
969 ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu);
970 if (ib_dev)
971 ret = sprintf(buf, "%s\n", ib_dev->name);
972 srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
973
974 return ret;
975}
976static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
977
978static ssize_t show_dev_abi_version(struct device *device,
979 struct device_attribute *attr, char *buf)
980{
981 struct ib_uverbs_device *dev = dev_get_drvdata(device);
982 int ret = -ENODEV;
983 int srcu_key;
984 struct ib_device *ib_dev;
985
986 if (!dev)
987 return -ENODEV;
988 srcu_key = srcu_read_lock(&dev->disassociate_srcu);
989 ib_dev = srcu_dereference(dev->ib_dev, &dev->disassociate_srcu);
990 if (ib_dev)
991 ret = sprintf(buf, "%d\n", ib_dev->uverbs_abi_ver);
992 srcu_read_unlock(&dev->disassociate_srcu, srcu_key);
993
994 return ret;
995}
996static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
997
998static CLASS_ATTR_STRING(abi_version, S_IRUGO,
999 __stringify(IB_USER_VERBS_ABI_VERSION));
1000
1001static void ib_uverbs_add_one(struct ib_device *device)
1002{
1003 int devnum;
1004 dev_t base;
1005 struct ib_uverbs_device *uverbs_dev;
1006 int ret;
1007
1008 if (!device->alloc_ucontext)
1009 return;
1010
1011 uverbs_dev = kzalloc(sizeof(*uverbs_dev), GFP_KERNEL);
1012 if (!uverbs_dev)
1013 return;
1014
1015 ret = init_srcu_struct(&uverbs_dev->disassociate_srcu);
1016 if (ret) {
1017 kfree(uverbs_dev);
1018 return;
1019 }
1020
1021 atomic_set(&uverbs_dev->refcount, 1);
1022 init_completion(&uverbs_dev->comp);
1023 uverbs_dev->xrcd_tree = RB_ROOT;
1024 mutex_init(&uverbs_dev->xrcd_tree_mutex);
1025 kobject_init(&uverbs_dev->kobj, &ib_uverbs_dev_ktype);
1026 mutex_init(&uverbs_dev->lists_mutex);
1027 INIT_LIST_HEAD(&uverbs_dev->uverbs_file_list);
1028 INIT_LIST_HEAD(&uverbs_dev->uverbs_events_file_list);
1029
1030 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
1031 if (devnum >= IB_UVERBS_MAX_DEVICES)
1032 goto err;
1033 uverbs_dev->devnum = devnum;
1034 set_bit(devnum, dev_map);
1035 if (devnum >= IB_UVERBS_NUM_FIXED_MINOR)
1036 base = dynamic_uverbs_dev + devnum - IB_UVERBS_NUM_FIXED_MINOR;
1037 else
1038 base = IB_UVERBS_BASE_DEV + devnum;
1039
1040 rcu_assign_pointer(uverbs_dev->ib_dev, device);
1041 uverbs_dev->num_comp_vectors = device->num_comp_vectors;
1042
1043 cdev_init(&uverbs_dev->cdev, NULL);
1044 uverbs_dev->cdev.owner = THIS_MODULE;
1045 uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
1046 cdev_set_parent(&uverbs_dev->cdev, &uverbs_dev->kobj);
1047 kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
1048 if (cdev_add(&uverbs_dev->cdev, base, 1))
1049 goto err_cdev;
1050
1051 uverbs_dev->dev = device_create(uverbs_class, device->dev.parent,
1052 uverbs_dev->cdev.dev, uverbs_dev,
1053 "uverbs%d", uverbs_dev->devnum);
1054 if (IS_ERR(uverbs_dev->dev))
1055 goto err_cdev;
1056
1057 if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
1058 goto err_class;
1059 if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
1060 goto err_class;
1061
1062 if (!device->specs_root) {
1063 const struct uverbs_object_tree_def *default_root[] = {
1064 uverbs_default_get_objects()};
1065
1066 uverbs_dev->specs_root = uverbs_alloc_spec_tree(1,
1067 default_root);
1068 if (IS_ERR(uverbs_dev->specs_root))
1069 goto err_class;
1070
1071 device->specs_root = uverbs_dev->specs_root;
1072 }
1073
1074 ib_set_client_data(device, &uverbs_client, uverbs_dev);
1075
1076 return;
1077
1078err_class:
1079 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
1080
1081err_cdev:
1082 cdev_del(&uverbs_dev->cdev);
1083 clear_bit(devnum, dev_map);
1084
1085err:
1086 if (atomic_dec_and_test(&uverbs_dev->refcount))
1087 ib_uverbs_comp_dev(uverbs_dev);
1088 wait_for_completion(&uverbs_dev->comp);
1089 kobject_put(&uverbs_dev->kobj);
1090 return;
1091}
1092
1093static void ib_uverbs_free_hw_resources(struct ib_uverbs_device *uverbs_dev,
1094 struct ib_device *ib_dev)
1095{
1096 struct ib_uverbs_file *file;
1097 struct ib_uverbs_async_event_file *event_file;
1098 struct ib_event event;
1099
1100 /* Pending running commands to terminate */
1101 synchronize_srcu(&uverbs_dev->disassociate_srcu);
1102 event.event = IB_EVENT_DEVICE_FATAL;
1103 event.element.port_num = 0;
1104 event.device = ib_dev;
1105
1106 mutex_lock(&uverbs_dev->lists_mutex);
1107 while (!list_empty(&uverbs_dev->uverbs_file_list)) {
1108 struct ib_ucontext *ucontext;
1109 file = list_first_entry(&uverbs_dev->uverbs_file_list,
1110 struct ib_uverbs_file, list);
1111 file->is_closed = 1;
1112 list_del(&file->list);
1113 kref_get(&file->ref);
1114 mutex_unlock(&uverbs_dev->lists_mutex);
1115
1116
1117 mutex_lock(&file->cleanup_mutex);
1118 ucontext = file->ucontext;
1119 file->ucontext = NULL;
1120 mutex_unlock(&file->cleanup_mutex);
1121
1122 /* At this point ib_uverbs_close cannot be running
1123 * ib_uverbs_cleanup_ucontext
1124 */
1125 if (ucontext) {
1126 /* We must release the mutex before going ahead and
1127 * calling disassociate_ucontext. disassociate_ucontext
1128 * might end up indirectly calling uverbs_close,
1129 * for example due to freeing the resources
1130 * (e.g mmput).
1131 */
1132 ib_uverbs_event_handler(&file->event_handler, &event);
1133 ib_dev->disassociate_ucontext(ucontext);
1134 mutex_lock(&file->cleanup_mutex);
1135 ib_uverbs_cleanup_ucontext(file, ucontext, true);
1136 mutex_unlock(&file->cleanup_mutex);
1137 }
1138
1139 mutex_lock(&uverbs_dev->lists_mutex);
1140 kref_put(&file->ref, ib_uverbs_release_file);
1141 }
1142
1143 while (!list_empty(&uverbs_dev->uverbs_events_file_list)) {
1144 event_file = list_first_entry(&uverbs_dev->
1145 uverbs_events_file_list,
1146 struct ib_uverbs_async_event_file,
1147 list);
1148 spin_lock_irq(&event_file->ev_queue.lock);
1149 event_file->ev_queue.is_closed = 1;
1150 spin_unlock_irq(&event_file->ev_queue.lock);
1151
1152 list_del(&event_file->list);
1153 ib_unregister_event_handler(
1154 &event_file->uverbs_file->event_handler);
1155 event_file->uverbs_file->event_handler.device =
1156 NULL;
1157
1158 wake_up_interruptible(&event_file->ev_queue.poll_wait);
1159 kill_fasync(&event_file->ev_queue.async_queue, SIGIO, POLL_IN);
1160 }
1161 mutex_unlock(&uverbs_dev->lists_mutex);
1162}
1163
1164static void ib_uverbs_remove_one(struct ib_device *device, void *client_data)
1165{
1166 struct ib_uverbs_device *uverbs_dev = client_data;
1167 int wait_clients = 1;
1168
1169 if (!uverbs_dev)
1170 return;
1171
1172 dev_set_drvdata(uverbs_dev->dev, NULL);
1173 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
1174 cdev_del(&uverbs_dev->cdev);
1175 clear_bit(uverbs_dev->devnum, dev_map);
1176
1177 if (device->disassociate_ucontext) {
1178 /* We disassociate HW resources and immediately return.
1179 * Userspace will see a EIO errno for all future access.
1180 * Upon returning, ib_device may be freed internally and is not
1181 * valid any more.
1182 * uverbs_device is still available until all clients close
1183 * their files, then the uverbs device ref count will be zero
1184 * and its resources will be freed.
1185 * Note: At this point no more files can be opened since the
1186 * cdev was deleted, however active clients can still issue
1187 * commands and close their open files.
1188 */
1189 rcu_assign_pointer(uverbs_dev->ib_dev, NULL);
1190 ib_uverbs_free_hw_resources(uverbs_dev, device);
1191 wait_clients = 0;
1192 }
1193
1194 if (atomic_dec_and_test(&uverbs_dev->refcount))
1195 ib_uverbs_comp_dev(uverbs_dev);
1196 if (wait_clients)
1197 wait_for_completion(&uverbs_dev->comp);
1198 if (uverbs_dev->specs_root) {
1199 uverbs_free_spec_tree(uverbs_dev->specs_root);
1200 device->specs_root = NULL;
1201 }
1202
1203 kobject_put(&uverbs_dev->kobj);
1204}
1205
1206static char *uverbs_devnode(struct device *dev, umode_t *mode)
1207{
1208 if (mode)
1209 *mode = 0666;
1210 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1211}
1212
1213static int __init ib_uverbs_init(void)
1214{
1215 int ret;
1216
1217 ret = register_chrdev_region(IB_UVERBS_BASE_DEV,
1218 IB_UVERBS_NUM_FIXED_MINOR,
1219 "infiniband_verbs");
1220 if (ret) {
1221 pr_err("user_verbs: couldn't register device number\n");
1222 goto out;
1223 }
1224
1225 ret = alloc_chrdev_region(&dynamic_uverbs_dev, 0,
1226 IB_UVERBS_NUM_DYNAMIC_MINOR,
1227 "infiniband_verbs");
1228 if (ret) {
1229 pr_err("couldn't register dynamic device number\n");
1230 goto out_alloc;
1231 }
1232
1233 uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
1234 if (IS_ERR(uverbs_class)) {
1235 ret = PTR_ERR(uverbs_class);
1236 pr_err("user_verbs: couldn't create class infiniband_verbs\n");
1237 goto out_chrdev;
1238 }
1239
1240 uverbs_class->devnode = uverbs_devnode;
1241
1242 ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
1243 if (ret) {
1244 pr_err("user_verbs: couldn't create abi_version attribute\n");
1245 goto out_class;
1246 }
1247
1248 ret = ib_register_client(&uverbs_client);
1249 if (ret) {
1250 pr_err("user_verbs: couldn't register client\n");
1251 goto out_class;
1252 }
1253
1254 return 0;
1255
1256out_class:
1257 class_destroy(uverbs_class);
1258
1259out_chrdev:
1260 unregister_chrdev_region(dynamic_uverbs_dev,
1261 IB_UVERBS_NUM_DYNAMIC_MINOR);
1262
1263out_alloc:
1264 unregister_chrdev_region(IB_UVERBS_BASE_DEV,
1265 IB_UVERBS_NUM_FIXED_MINOR);
1266
1267out:
1268 return ret;
1269}
1270
1271static void __exit ib_uverbs_cleanup(void)
1272{
1273 ib_unregister_client(&uverbs_client);
1274 class_destroy(uverbs_class);
1275 unregister_chrdev_region(IB_UVERBS_BASE_DEV,
1276 IB_UVERBS_NUM_FIXED_MINOR);
1277 unregister_chrdev_region(dynamic_uverbs_dev,
1278 IB_UVERBS_NUM_DYNAMIC_MINOR);
1279}
1280
1281module_init(ib_uverbs_init);
1282module_exit(ib_uverbs_cleanup);