summaryrefslogtreecommitdiff
path: root/drivers/infiniband/core/rdma_core.c
blob: 6c72773faf2911406191b67c76caef4da54a04d7 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
/*
 * Copyright (c) 2016, Mellanox Technologies inc.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/file.h>
#include <linux/anon_inodes.h>
#include <linux/sched/mm.h>
#include <rdma/ib_verbs.h>
#include <rdma/uverbs_types.h>
#include <linux/rcupdate.h>
#include <rdma/uverbs_ioctl.h>
#include <rdma/rdma_user_ioctl.h>
#include "uverbs.h"
#include "core_priv.h"
#include "rdma_core.h"

void uverbs_uobject_get(struct ib_uobject *uobject)
{
	kref_get(&uobject->ref);
}

static void uverbs_uobject_free(struct kref *ref)
{
	struct ib_uobject *uobj =
		container_of(ref, struct ib_uobject, ref);

	if (uobj->uapi_object->type_class->needs_kfree_rcu)
		kfree_rcu(uobj, rcu);
	else
		kfree(uobj);
}

void uverbs_uobject_put(struct ib_uobject *uobject)
{
	kref_put(&uobject->ref, uverbs_uobject_free);
}

static int uverbs_try_lock_object(struct ib_uobject *uobj,
				  enum rdma_lookup_mode mode)
{
	/*
	 * When a shared access is required, we use a positive counter. Each
	 * shared access request checks that the value != -1 and increment it.
	 * Exclusive access is required for operations like write or destroy.
	 * In exclusive access mode, we check that the counter is zero (nobody
	 * claimed this object) and we set it to -1. Releasing a shared access
	 * lock is done simply by decreasing the counter. As for exclusive
	 * access locks, since only a single one of them is is allowed
	 * concurrently, setting the counter to zero is enough for releasing
	 * this lock.
	 */
	switch (mode) {
	case UVERBS_LOOKUP_READ:
		return atomic_fetch_add_unless(&uobj->usecnt, 1, -1) == -1 ?
			-EBUSY : 0;
	case UVERBS_LOOKUP_WRITE:
		/* lock is exclusive */
		return atomic_cmpxchg(&uobj->usecnt, 0, -1) == 0 ? 0 : -EBUSY;
	case UVERBS_LOOKUP_DESTROY:
		return 0;
	}
	return 0;
}

static void assert_uverbs_usecnt(struct ib_uobject *uobj,
				 enum rdma_lookup_mode mode)
{
#ifdef CONFIG_LOCKDEP
	switch (mode) {
	case UVERBS_LOOKUP_READ:
		WARN_ON(atomic_read(&uobj->usecnt) <= 0);
		break;
	case UVERBS_LOOKUP_WRITE:
		WARN_ON(atomic_read(&uobj->usecnt) != -1);
		break;
	case UVERBS_LOOKUP_DESTROY:
		break;
	}
#endif
}

/*
 * This must be called with the hw_destroy_rwsem locked for read or write,
 * also the uobject itself must be locked for write.
 *
 * Upon return the HW object is guaranteed to be destroyed.
 *
 * For RDMA_REMOVE_ABORT, the hw_destroy_rwsem is not required to be held,
 * however the type's allocat_commit function cannot have been called and the
 * uobject cannot be on the uobjects_lists
 *
 * For RDMA_REMOVE_DESTROY the caller shold be holding a kref (eg via
 * rdma_lookup_get_uobject) and the object is left in a state where the caller
 * needs to call rdma_lookup_put_uobject.
 *
 * For all other destroy modes this function internally unlocks the uobject
 * and consumes the kref on the uobj.
 */
static int uverbs_destroy_uobject(struct ib_uobject *uobj,
				  enum rdma_remove_reason reason,
				  struct uverbs_attr_bundle *attrs)
{
	struct ib_uverbs_file *ufile = attrs->ufile;
	unsigned long flags;
	int ret;

	lockdep_assert_held(&ufile->hw_destroy_rwsem);
	assert_uverbs_usecnt(uobj, UVERBS_LOOKUP_WRITE);

	if (uobj->object) {
		ret = uobj->uapi_object->type_class->destroy_hw(uobj, reason,
								attrs);
		if (ret) {
			if (ib_is_destroy_retryable(ret, reason, uobj))
				return ret;

			/* Nothing to be done, dangle the memory and move on */
			WARN(true,
			     "ib_uverbs: failed to remove uobject id %d, driver err=%d",
			     uobj->id, ret);
		}

		uobj->object = NULL;
	}

	if (reason == RDMA_REMOVE_ABORT) {
		WARN_ON(!list_empty(&uobj->list));
		WARN_ON(!uobj->context);
		uobj->uapi_object->type_class->alloc_abort(uobj);
	}

	uobj->context = NULL;

	/*
	 * For DESTROY the usecnt is held write locked, the caller is expected
	 * to put it unlock and put the object when done with it. Only DESTROY
	 * can remove the IDR handle.
	 */
	if (reason != RDMA_REMOVE_DESTROY)
		atomic_set(&uobj->usecnt, 0);
	else
		uobj->uapi_object->type_class->remove_handle(uobj);

	if (!list_empty(&uobj->list)) {
		spin_lock_irqsave(&ufile->uobjects_lock, flags);
		list_del_init(&uobj->list);
		spin_unlock_irqrestore(&ufile->uobjects_lock, flags);

		/*
		 * Pairs with the get in rdma_alloc_commit_uobject(), could
		 * destroy uobj.
		 */
		uverbs_uobject_put(uobj);
	}

	/*
	 * When aborting the stack kref remains owned by the core code, and is
	 * not transferred into the type. Pairs with the get in alloc_uobj
	 */
	if (reason == RDMA_REMOVE_ABORT)
		uverbs_uobject_put(uobj);

	return 0;
}

/*
 * This calls uverbs_destroy_uobject() using the RDMA_REMOVE_DESTROY
 * sequence. It should only be used from command callbacks. On success the
 * caller must pair this with rdma_lookup_put_uobject(LOOKUP_WRITE). This
 * version requires the caller to have already obtained an
 * LOOKUP_DESTROY uobject kref.
 */
int uobj_destroy(struct ib_uobject *uobj, struct uverbs_attr_bundle *attrs)
{
	struct ib_uverbs_file *ufile = attrs->ufile;
	int ret;

	down_read(&ufile->hw_destroy_rwsem);

	ret = uverbs_try_lock_object(uobj, UVERBS_LOOKUP_WRITE);
	if (ret)
		goto out_unlock;

	ret = uverbs_destroy_uobject(uobj, RDMA_REMOVE_DESTROY, attrs);
	if (ret) {
		atomic_set(&uobj->usecnt, 0);
		goto out_unlock;
	}

out_unlock:
	up_read(&ufile->hw_destroy_rwsem);
	return ret;
}

/*
 * uobj_get_destroy destroys the HW object and returns a handle to the uobj
 * with a NULL object pointer. The caller must pair this with
 * uverbs_put_destroy.
 */
struct ib_uobject *__uobj_get_destroy(const struct uverbs_api_object *obj,
				      u32 id, struct uverbs_attr_bundle *attrs)
{
	struct ib_uobject *uobj;
	int ret;

	uobj = rdma_lookup_get_uobject(obj, attrs->ufile, id,
				       UVERBS_LOOKUP_DESTROY, attrs);
	if (IS_ERR(uobj))
		return uobj;

	ret = uobj_destroy(uobj, attrs);
	if (ret) {
		rdma_lookup_put_uobject(uobj, UVERBS_LOOKUP_DESTROY);
		return ERR_PTR(ret);
	}

	return uobj;
}

/*
 * Does both uobj_get_destroy() and uobj_put_destroy().  Returns 0 on success
 * (negative errno on failure). For use by callers that do not need the uobj.
 */
int __uobj_perform_destroy(const struct uverbs_api_object *obj, u32 id,
			   struct uverbs_attr_bundle *attrs)
{
	struct ib_uobject *uobj;

	uobj = __uobj_get_destroy(obj, id, attrs);
	if (IS_ERR(uobj))
		return PTR_ERR(uobj);

	rdma_lookup_put_uobject(uobj, UVERBS_LOOKUP_WRITE);
	return 0;
}

/* alloc_uobj must be undone by uverbs_destroy_uobject() */
static struct ib_uobject *alloc_uobj(struct ib_uverbs_file *ufile,
				     const struct uverbs_api_object *obj)
{
	struct ib_uobject *uobj;
	struct ib_ucontext *ucontext;

	ucontext = ib_uverbs_get_ucontext_file(ufile);
	if (IS_ERR(ucontext))
		return ERR_CAST(ucontext);

	uobj = kzalloc(obj->type_attrs->obj_size, GFP_KERNEL);
	if (!uobj)
		return ERR_PTR(-ENOMEM);
	/*
	 * user_handle should be filled by the handler,
	 * The object is added to the list in the commit stage.
	 */
	uobj->ufile = ufile;
	uobj->context = ucontext;
	INIT_LIST_HEAD(&uobj->list);
	uobj->uapi_object = obj;
	/*
	 * Allocated objects start out as write locked to deny any other
	 * syscalls from accessing them until they are committed. See
	 * rdma_alloc_commit_uobject
	 */
	atomic_set(&uobj->usecnt, -1);
	kref_init(&uobj->ref);

	return uobj;
}

static int idr_add_uobj(struct ib_uobject *uobj)
{
       /*
        * We start with allocating an idr pointing to NULL. This represents an
        * object which isn't initialized yet. We'll replace it later on with
        * the real object once we commit.
        */
	return xa_alloc(&uobj->ufile->idr, &uobj->id, NULL, xa_limit_32b,
			GFP_KERNEL);
}

/* Returns the ib_uobject or an error. The caller should check for IS_ERR. */
static struct ib_uobject *
lookup_get_idr_uobject(const struct uverbs_api_object *obj,
		       struct ib_uverbs_file *ufile, s64 id,
		       enum rdma_lookup_mode mode)
{
	struct ib_uobject *uobj;

	if (id < 0 || id > ULONG_MAX)
		return ERR_PTR(-EINVAL);

	rcu_read_lock();
	/*
	 * The idr_find is guaranteed to return a pointer to something that
	 * isn't freed yet, or NULL, as the free after idr_remove goes through
	 * kfree_rcu(). However the object may still have been released and
	 * kfree() could be called at any time.
	 */
	uobj = xa_load(&ufile->idr, id);
	if (!uobj || !kref_get_unless_zero(&uobj->ref))
		uobj = ERR_PTR(-ENOENT);
	rcu_read_unlock();
	return uobj;
}

static struct ib_uobject *
lookup_get_fd_uobject(const struct uverbs_api_object *obj,
		      struct ib_uverbs_file *ufile, s64 id,
		      enum rdma_lookup_mode mode)
{
	const struct uverbs_obj_fd_type *fd_type;
	struct file *f;
	struct ib_uobject *uobject;
	int fdno = id;

	if (fdno != id)
		return ERR_PTR(-EINVAL);

	if (mode != UVERBS_LOOKUP_READ)
		return ERR_PTR(-EOPNOTSUPP);

	if (!obj->type_attrs)
		return ERR_PTR(-EIO);
	fd_type =
		container_of(obj->type_attrs, struct uverbs_obj_fd_type, type);

	f = fget(fdno);
	if (!f)
		return ERR_PTR(-EBADF);

	uobject = f->private_data;
	/*
	 * fget(id) ensures we are not currently running uverbs_close_fd,
	 * and the caller is expected to ensure that uverbs_close_fd is never
	 * done while a call top lookup is possible.
	 */
	if (f->f_op != fd_type->fops) {
		fput(f);
		return ERR_PTR(-EBADF);
	}

	uverbs_uobject_get(uobject);
	return uobject;
}

struct ib_uobject *rdma_lookup_get_uobject(const struct uverbs_api_object *obj,
					   struct ib_uverbs_file *ufile, s64 id,
					   enum rdma_lookup_mode mode,
					   struct uverbs_attr_bundle *attrs)
{
	struct ib_uobject *uobj;
	int ret;

	if (obj == ERR_PTR(-ENOMSG)) {
		/* must be UVERBS_IDR_ANY_OBJECT, see uapi_get_object() */
		uobj = lookup_get_idr_uobject(NULL, ufile, id, mode);
		if (IS_ERR(uobj))
			return uobj;
	} else {
		if (IS_ERR(obj))
			return ERR_PTR(-EINVAL);

		uobj = obj->type_class->lookup_get(obj, ufile, id, mode);
		if (IS_ERR(uobj))
			return uobj;

		if (uobj->uapi_object != obj) {
			ret = -EINVAL;
			goto free;
		}
	}

	/*
	 * If we have been disassociated block every command except for
	 * DESTROY based commands.
	 */
	if (mode != UVERBS_LOOKUP_DESTROY &&
	    !srcu_dereference(ufile->device->ib_dev,
			      &ufile->device->disassociate_srcu)) {
		ret = -EIO;
		goto free;
	}

	ret = uverbs_try_lock_object(uobj, mode);
	if (ret)
		goto free;
	if (attrs)
		attrs->context = uobj->context;

	return uobj;
free:
	uobj->uapi_object->type_class->lookup_put(uobj, mode);
	uverbs_uobject_put(uobj);
	return ERR_PTR(ret);
}

static struct ib_uobject *
alloc_begin_idr_uobject(const struct uverbs_api_object *obj,
			struct ib_uverbs_file *ufile)
{
	int ret;
	struct ib_uobject *uobj;

	uobj = alloc_uobj(ufile, obj);
	if (IS_ERR(uobj))
		return uobj;

	ret = idr_add_uobj(uobj);
	if (ret)
		goto uobj_put;

	ret = ib_rdmacg_try_charge(&uobj->cg_obj, uobj->context->device,
				   RDMACG_RESOURCE_HCA_OBJECT);
	if (ret)
		goto remove;

	return uobj;

remove:
	xa_erase(&ufile->idr, uobj->id);
uobj_put:
	uverbs_uobject_put(uobj);
	return ERR_PTR(ret);
}

static struct ib_uobject *
alloc_begin_fd_uobject(const struct uverbs_api_object *obj,
		       struct ib_uverbs_file *ufile)
{
	int new_fd;
	struct ib_uobject *uobj;

	new_fd = get_unused_fd_flags(O_CLOEXEC);
	if (new_fd < 0)
		return ERR_PTR(new_fd);

	uobj = alloc_uobj(ufile, obj);
	if (IS_ERR(uobj)) {
		put_unused_fd(new_fd);
		return uobj;
	}

	uobj->id = new_fd;
	uobj->ufile = ufile;

	return uobj;
}

struct ib_uobject *rdma_alloc_begin_uobject(const struct uverbs_api_object *obj,
					    struct ib_uverbs_file *ufile,
					    struct uverbs_attr_bundle *attrs)
{
	struct ib_uobject *ret;

	if (IS_ERR(obj))
		return ERR_PTR(-EINVAL);

	/*
	 * The hw_destroy_rwsem is held across the entire object creation and
	 * released during rdma_alloc_commit_uobject or
	 * rdma_alloc_abort_uobject
	 */
	if (!down_read_trylock(&ufile->hw_destroy_rwsem))
		return ERR_PTR(-EIO);

	ret = obj->type_class->alloc_begin(obj, ufile);
	if (IS_ERR(ret)) {
		up_read(&ufile->hw_destroy_rwsem);
		return ret;
	}
	if (attrs)
		attrs->context = ret->context;
	return ret;
}

static void alloc_abort_idr_uobject(struct ib_uobject *uobj)
{
	ib_rdmacg_uncharge(&uobj->cg_obj, uobj->context->device,
			   RDMACG_RESOURCE_HCA_OBJECT);

	xa_erase(&uobj->ufile->idr, uobj->id);
}

static int __must_check destroy_hw_idr_uobject(struct ib_uobject *uobj,
					       enum rdma_remove_reason why,
					       struct uverbs_attr_bundle *attrs)
{
	const struct uverbs_obj_idr_type *idr_type =
		container_of(uobj->uapi_object->type_attrs,
			     struct uverbs_obj_idr_type, type);
	int ret = idr_type->destroy_object(uobj, why, attrs);

	/*
	 * We can only fail gracefully if the user requested to destroy the
	 * object or when a retry may be called upon an error.
	 * In the rest of the cases, just remove whatever you can.
	 */
	if (ib_is_destroy_retryable(ret, why, uobj))
		return ret;

	if (why == RDMA_REMOVE_ABORT)
		return 0;

	ib_rdmacg_uncharge(&uobj->cg_obj, uobj->context->device,
			   RDMACG_RESOURCE_HCA_OBJECT);

	return 0;
}

static void remove_handle_idr_uobject(struct ib_uobject *uobj)
{
	xa_erase(&uobj->ufile->idr, uobj->id);
	/* Matches the kref in alloc_commit_idr_uobject */
	uverbs_uobject_put(uobj);
}

static void alloc_abort_fd_uobject(struct ib_uobject *uobj)
{
	put_unused_fd(uobj->id);
}

static int __must_check destroy_hw_fd_uobject(struct ib_uobject *uobj,
					      enum rdma_remove_reason why,
					      struct uverbs_attr_bundle *attrs)
{
	const struct uverbs_obj_fd_type *fd_type = container_of(
		uobj->uapi_object->type_attrs, struct uverbs_obj_fd_type, type);
	int ret = fd_type->context_closed(uobj, why);

	if (ib_is_destroy_retryable(ret, why, uobj))
		return ret;

	return 0;
}

static void remove_handle_fd_uobject(struct ib_uobject *uobj)
{
}

static int alloc_commit_idr_uobject(struct ib_uobject *uobj)
{
	struct ib_uverbs_file *ufile = uobj->ufile;
	void *old;

	/*
	 * We already allocated this IDR with a NULL object, so
	 * this shouldn't fail.
	 *
	 * NOTE: Storing the uobj transfers our kref on uobj to the XArray.
	 * It will be put by remove_commit_idr_uobject()
	 */
	old = xa_store(&ufile->idr, uobj->id, uobj, GFP_KERNEL);
	WARN_ON(old != NULL);

	return 0;
}

static int alloc_commit_fd_uobject(struct ib_uobject *uobj)
{
	const struct uverbs_obj_fd_type *fd_type = container_of(
		uobj->uapi_object->type_attrs, struct uverbs_obj_fd_type, type);
	int fd = uobj->id;
	struct file *filp;

	/*
	 * The kref for uobj is moved into filp->private data and put in
	 * uverbs_close_fd(). Once alloc_commit() succeeds uverbs_close_fd()
	 * must be guaranteed to be called from the provided fops release
	 * callback.
	 */
	filp = anon_inode_getfile(fd_type->name,
				  fd_type->fops,
				  uobj,
				  fd_type->flags);
	if (IS_ERR(filp))
		return PTR_ERR(filp);

	uobj->object = filp;

	/* Matching put will be done in uverbs_close_fd() */
	kref_get(&uobj->ufile->ref);

	/* This shouldn't be used anymore. Use the file object instead */
	uobj->id = 0;

	/*
	 * NOTE: Once we install the file we loose ownership of our kref on
	 * uobj. It will be put by uverbs_close_fd()
	 */
	fd_install(fd, filp);

	return 0;
}

/*
 * In all cases rdma_alloc_commit_uobject() consumes the kref to uobj and the
 * caller can no longer assume uobj is valid. If this function fails it
 * destroys the uboject, including the attached HW object.
 */
int __must_check rdma_alloc_commit_uobject(struct ib_uobject *uobj,
					   struct uverbs_attr_bundle *attrs)
{
	struct ib_uverbs_file *ufile = attrs->ufile;
	int ret;

	/* alloc_commit consumes the uobj kref */
	ret = uobj->uapi_object->type_class->alloc_commit(uobj);
	if (ret) {
		uverbs_destroy_uobject(uobj, RDMA_REMOVE_ABORT, attrs);
		up_read(&ufile->hw_destroy_rwsem);
		return ret;
	}

	/* kref is held so long as the uobj is on the uobj list. */
	uverbs_uobject_get(uobj);
	spin_lock_irq(&ufile->uobjects_lock);
	list_add(&uobj->list, &ufile->uobjects);
	spin_unlock_irq(&ufile->uobjects_lock);

	/* matches atomic_set(-1) in alloc_uobj */
	atomic_set(&uobj->usecnt, 0);

	/* Matches the down_read in rdma_alloc_begin_uobject */
	up_read(&ufile->hw_destroy_rwsem);

	return 0;
}

/*
 * This consumes the kref for uobj. It is up to the caller to unwind the HW
 * object and anything else connected to uobj before calling this.
 */
void rdma_alloc_abort_uobject(struct ib_uobject *uobj,
			      struct uverbs_attr_bundle *attrs)
{
	struct ib_uverbs_file *ufile = uobj->ufile;

	uobj->object = NULL;
	uverbs_destroy_uobject(uobj, RDMA_REMOVE_ABORT, attrs);

	/* Matches the down_read in rdma_alloc_begin_uobject */
	up_read(&ufile->hw_destroy_rwsem);
}

static void lookup_put_idr_uobject(struct ib_uobject *uobj,
				   enum rdma_lookup_mode mode)
{
}

static void lookup_put_fd_uobject(struct ib_uobject *uobj,
				  enum rdma_lookup_mode mode)
{
	struct file *filp = uobj->object;

	WARN_ON(mode != UVERBS_LOOKUP_READ);
	/* This indirectly calls uverbs_close_fd and free the object */
	fput(filp);
}

void rdma_lookup_put_uobject(struct ib_uobject *uobj,
			     enum rdma_lookup_mode mode)
{
	assert_uverbs_usecnt(uobj, mode);
	uobj->uapi_object->type_class->lookup_put(uobj, mode);
	/*
	 * In order to unlock an object, either decrease its usecnt for
	 * read access or zero it in case of exclusive access. See
	 * uverbs_try_lock_object for locking schema information.
	 */
	switch (mode) {
	case UVERBS_LOOKUP_READ:
		atomic_dec(&uobj->usecnt);
		break;
	case UVERBS_LOOKUP_WRITE:
		atomic_set(&uobj->usecnt, 0);
		break;
	case UVERBS_LOOKUP_DESTROY:
		break;
	}

	/* Pairs with the kref obtained by type->lookup_get */
	uverbs_uobject_put(uobj);
}

void setup_ufile_idr_uobject(struct ib_uverbs_file *ufile)
{
	xa_init_flags(&ufile->idr, XA_FLAGS_ALLOC);
}

void release_ufile_idr_uobject(struct ib_uverbs_file *ufile)
{
	struct ib_uobject *entry;
	unsigned long id;

	/*
	 * At this point uverbs_cleanup_ufile() is guaranteed to have run, and
	 * there are no HW objects left, however the xarray is still populated
	 * with anything that has not been cleaned up by userspace. Since the
	 * kref on ufile is 0, nothing is allowed to call lookup_get.
	 *
	 * This is an optimized equivalent to remove_handle_idr_uobject
	 */
	xa_for_each(&ufile->idr, id, entry) {
		WARN_ON(entry->object);
		uverbs_uobject_put(entry);
	}

	xa_destroy(&ufile->idr);
}

const struct uverbs_obj_type_class uverbs_idr_class = {
	.alloc_begin = alloc_begin_idr_uobject,
	.lookup_get = lookup_get_idr_uobject,
	.alloc_commit = alloc_commit_idr_uobject,
	.alloc_abort = alloc_abort_idr_uobject,
	.lookup_put = lookup_put_idr_uobject,
	.destroy_hw = destroy_hw_idr_uobject,
	.remove_handle = remove_handle_idr_uobject,
	/*
	 * When we destroy an object, we first just lock it for WRITE and
	 * actually DESTROY it in the finalize stage. So, the problematic
	 * scenario is when we just started the finalize stage of the
	 * destruction (nothing was executed yet). Now, the other thread
	 * fetched the object for READ access, but it didn't lock it yet.
	 * The DESTROY thread continues and starts destroying the object.
	 * When the other thread continue - without the RCU, it would
	 * access freed memory. However, the rcu_read_lock delays the free
	 * until the rcu_read_lock of the READ operation quits. Since the
	 * exclusive lock of the object is still taken by the DESTROY flow, the
	 * READ operation will get -EBUSY and it'll just bail out.
	 */
	.needs_kfree_rcu = true,
};
EXPORT_SYMBOL(uverbs_idr_class);

void uverbs_close_fd(struct file *f)
{
	struct ib_uobject *uobj = f->private_data;
	struct ib_uverbs_file *ufile = uobj->ufile;
	struct uverbs_attr_bundle attrs = {
		.context = uobj->context,
		.ufile = ufile,
	};

	if (down_read_trylock(&ufile->hw_destroy_rwsem)) {
		/*
		 * lookup_get_fd_uobject holds the kref on the struct file any
		 * time a FD uobj is locked, which prevents this release
		 * method from being invoked. Meaning we can always get the
		 * write lock here, or we have a kernel bug.
		 */
		WARN_ON(uverbs_try_lock_object(uobj, UVERBS_LOOKUP_WRITE));
		uverbs_destroy_uobject(uobj, RDMA_REMOVE_CLOSE, &attrs);
		up_read(&ufile->hw_destroy_rwsem);
	}

	/* Matches the get in alloc_begin_fd_uobject */
	kref_put(&ufile->ref, ib_uverbs_release_file);

	/* Pairs with filp->private_data in alloc_begin_fd_uobject */
	uverbs_uobject_put(uobj);
}
EXPORT_SYMBOL(uverbs_close_fd);

/*
 * Drop the ucontext off the ufile and completely disconnect it from the
 * ib_device
 */
static void ufile_destroy_ucontext(struct ib_uverbs_file *ufile,
				   enum rdma_remove_reason reason)
{
	struct ib_ucontext *ucontext = ufile->ucontext;
	struct ib_device *ib_dev = ucontext->device;

	/*
	 * If we are closing the FD then the user mmap VMAs must have
	 * already been destroyed as they hold on to the filep, otherwise
	 * they need to be zap'd.
	 */
	if (reason == RDMA_REMOVE_DRIVER_REMOVE) {
		uverbs_user_mmap_disassociate(ufile);
		if (ib_dev->ops.disassociate_ucontext)
			ib_dev->ops.disassociate_ucontext(ucontext);
	}

	ib_rdmacg_uncharge(&ucontext->cg_obj, ib_dev,
			   RDMACG_RESOURCE_HCA_HANDLE);

	rdma_restrack_del(&ucontext->res);

	ib_dev->ops.dealloc_ucontext(ucontext);
	WARN_ON(!xa_empty(&ucontext->mmap_xa));
	kfree(ucontext);

	ufile->ucontext = NULL;
}

static int __uverbs_cleanup_ufile(struct ib_uverbs_file *ufile,
				  enum rdma_remove_reason reason)
{
	struct ib_uobject *obj, *next_obj;
	int ret = -EINVAL;
	struct uverbs_attr_bundle attrs = { .ufile = ufile };

	/*
	 * This shouldn't run while executing other commands on this
	 * context. Thus, the only thing we should take care of is
	 * releasing a FD while traversing this list. The FD could be
	 * closed and released from the _release fop of this FD.
	 * In order to mitigate this, we add a lock.
	 * We take and release the lock per traversal in order to let
	 * other threads (which might still use the FDs) chance to run.
	 */
	list_for_each_entry_safe(obj, next_obj, &ufile->uobjects, list) {
		attrs.context = obj->context;
		/*
		 * if we hit this WARN_ON, that means we are
		 * racing with a lookup_get.
		 */
		WARN_ON(uverbs_try_lock_object(obj, UVERBS_LOOKUP_WRITE));
		if (!uverbs_destroy_uobject(obj, reason, &attrs))
			ret = 0;
		else
			atomic_set(&obj->usecnt, 0);
	}
	return ret;
}

/*
 * Destroy the uncontext and every uobject associated with it. If called with
 * reason != RDMA_REMOVE_CLOSE this will not return until the destruction has
 * been completed and ufile->ucontext is NULL.
 *
 * This is internally locked and can be called in parallel from multiple
 * contexts.
 */
void uverbs_destroy_ufile_hw(struct ib_uverbs_file *ufile,
			     enum rdma_remove_reason reason)
{
	if (reason == RDMA_REMOVE_CLOSE) {
		/*
		 * During destruction we might trigger something that
		 * synchronously calls release on any file descriptor. For
		 * this reason all paths that come from file_operations
		 * release must use try_lock. They can progress knowing that
		 * there is an ongoing uverbs_destroy_ufile_hw that will clean
		 * up the driver resources.
		 */
		if (!mutex_trylock(&ufile->ucontext_lock))
			return;

	} else {
		mutex_lock(&ufile->ucontext_lock);
	}

	down_write(&ufile->hw_destroy_rwsem);

	/*
	 * If a ucontext was never created then we can't have any uobjects to
	 * cleanup, nothing to do.
	 */
	if (!ufile->ucontext)
		goto done;

	ufile->ucontext->closing = true;
	ufile->ucontext->cleanup_retryable = true;
	while (!list_empty(&ufile->uobjects))
		if (__uverbs_cleanup_ufile(ufile, reason)) {
			/*
			 * No entry was cleaned-up successfully during this
			 * iteration
			 */
			break;
		}

	ufile->ucontext->cleanup_retryable = false;
	if (!list_empty(&ufile->uobjects))
		__uverbs_cleanup_ufile(ufile, reason);

	ufile_destroy_ucontext(ufile, reason);

done:
	up_write(&ufile->hw_destroy_rwsem);
	mutex_unlock(&ufile->ucontext_lock);
}

const struct uverbs_obj_type_class uverbs_fd_class = {
	.alloc_begin = alloc_begin_fd_uobject,
	.lookup_get = lookup_get_fd_uobject,
	.alloc_commit = alloc_commit_fd_uobject,
	.alloc_abort = alloc_abort_fd_uobject,
	.lookup_put = lookup_put_fd_uobject,
	.destroy_hw = destroy_hw_fd_uobject,
	.remove_handle = remove_handle_fd_uobject,
	.needs_kfree_rcu = false,
};
EXPORT_SYMBOL(uverbs_fd_class);

struct ib_uobject *
uverbs_get_uobject_from_file(u16 object_id, enum uverbs_obj_access access,
			     s64 id, struct uverbs_attr_bundle *attrs)
{
	const struct uverbs_api_object *obj =
		uapi_get_object(attrs->ufile->device->uapi, object_id);

	switch (access) {
	case UVERBS_ACCESS_READ:
		return rdma_lookup_get_uobject(obj, attrs->ufile, id,
					       UVERBS_LOOKUP_READ, attrs);
	case UVERBS_ACCESS_DESTROY:
		/* Actual destruction is done inside uverbs_handle_method */
		return rdma_lookup_get_uobject(obj, attrs->ufile, id,
					       UVERBS_LOOKUP_DESTROY, attrs);
	case UVERBS_ACCESS_WRITE:
		return rdma_lookup_get_uobject(obj, attrs->ufile, id,
					       UVERBS_LOOKUP_WRITE, attrs);
	case UVERBS_ACCESS_NEW:
		return rdma_alloc_begin_uobject(obj, attrs->ufile, attrs);
	default:
		WARN_ON(true);
		return ERR_PTR(-EOPNOTSUPP);
	}
}

int uverbs_finalize_object(struct ib_uobject *uobj,
			   enum uverbs_obj_access access, bool commit,
			   struct uverbs_attr_bundle *attrs)
{
	int ret = 0;

	/*
	 * refcounts should be handled at the object level and not at the
	 * uobject level. Refcounts of the objects themselves are done in
	 * handlers.
	 */

	switch (access) {
	case UVERBS_ACCESS_READ:
		rdma_lookup_put_uobject(uobj, UVERBS_LOOKUP_READ);
		break;
	case UVERBS_ACCESS_WRITE:
		rdma_lookup_put_uobject(uobj, UVERBS_LOOKUP_WRITE);
		break;
	case UVERBS_ACCESS_DESTROY:
		if (uobj)
			rdma_lookup_put_uobject(uobj, UVERBS_LOOKUP_DESTROY);
		break;
	case UVERBS_ACCESS_NEW:
		if (commit)
			ret = rdma_alloc_commit_uobject(uobj, attrs);
		else
			rdma_alloc_abort_uobject(uobj, attrs);
		break;
	default:
		WARN_ON(true);
		ret = -EOPNOTSUPP;
	}

	return ret;
}