nvmm.c revision 1.21 1 /* $NetBSD: nvmm.c,v 1.21 2019/05/11 07:31:56 maxv Exp $ */
2
3 /*
4 * Copyright (c) 2018-2019 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Maxime Villard.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: nvmm.c,v 1.21 2019/05/11 07:31:56 maxv Exp $");
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38
39 #include <sys/cpu.h>
40 #include <sys/conf.h>
41 #include <sys/kmem.h>
42 #include <sys/module.h>
43 #include <sys/proc.h>
44 #include <sys/mman.h>
45 #include <sys/file.h>
46 #include <sys/filedesc.h>
47 #include <sys/kauth.h>
48
49 #include <uvm/uvm.h>
50 #include <uvm/uvm_page.h>
51
52 #include "ioconf.h"
53
54 #include <dev/nvmm/nvmm.h>
55 #include <dev/nvmm/nvmm_internal.h>
56 #include <dev/nvmm/nvmm_ioctl.h>
57
58 static struct nvmm_machine machines[NVMM_MAX_MACHINES];
59 static volatile unsigned int nmachines __cacheline_aligned;
60
61 static const struct nvmm_impl *nvmm_impl_list[] = {
62 &nvmm_x86_svm, /* x86 AMD SVM */
63 &nvmm_x86_vmx /* x86 Intel VMX */
64 };
65
66 static const struct nvmm_impl *nvmm_impl = NULL;
67
68 static struct nvmm_owner root_owner;
69
70 /* -------------------------------------------------------------------------- */
71
72 static int
73 nvmm_machine_alloc(struct nvmm_machine **ret)
74 {
75 struct nvmm_machine *mach;
76 size_t i;
77
78 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
79 mach = &machines[i];
80
81 rw_enter(&mach->lock, RW_WRITER);
82 if (mach->present) {
83 rw_exit(&mach->lock);
84 continue;
85 }
86
87 mach->present = true;
88 mach->time = time_second;
89 *ret = mach;
90 atomic_inc_uint(&nmachines);
91 return 0;
92 }
93
94 return ENOBUFS;
95 }
96
97 static void
98 nvmm_machine_free(struct nvmm_machine *mach)
99 {
100 KASSERT(rw_write_held(&mach->lock));
101 KASSERT(mach->present);
102 mach->present = false;
103 atomic_dec_uint(&nmachines);
104 }
105
106 static int
107 nvmm_machine_get(struct nvmm_owner *owner, nvmm_machid_t machid,
108 struct nvmm_machine **ret, bool writer)
109 {
110 struct nvmm_machine *mach;
111 krw_t op = writer ? RW_WRITER : RW_READER;
112
113 if (machid >= NVMM_MAX_MACHINES) {
114 return EINVAL;
115 }
116 mach = &machines[machid];
117
118 rw_enter(&mach->lock, op);
119 if (!mach->present) {
120 rw_exit(&mach->lock);
121 return ENOENT;
122 }
123 if (owner != &root_owner && mach->owner != owner) {
124 rw_exit(&mach->lock);
125 return EPERM;
126 }
127 *ret = mach;
128
129 return 0;
130 }
131
132 static void
133 nvmm_machine_put(struct nvmm_machine *mach)
134 {
135 rw_exit(&mach->lock);
136 }
137
138 /* -------------------------------------------------------------------------- */
139
140 static int
141 nvmm_vcpu_alloc(struct nvmm_machine *mach, nvmm_cpuid_t cpuid,
142 struct nvmm_cpu **ret)
143 {
144 struct nvmm_cpu *vcpu;
145
146 if (cpuid >= NVMM_MAX_VCPUS) {
147 return EINVAL;
148 }
149 vcpu = &mach->cpus[cpuid];
150
151 mutex_enter(&vcpu->lock);
152 if (vcpu->present) {
153 mutex_exit(&vcpu->lock);
154 return EBUSY;
155 }
156
157 vcpu->present = true;
158 vcpu->comm = NULL;
159 vcpu->hcpu_last = -1;
160 *ret = vcpu;
161 return 0;
162 }
163
164 static void
165 nvmm_vcpu_free(struct nvmm_machine *mach, struct nvmm_cpu *vcpu)
166 {
167 KASSERT(mutex_owned(&vcpu->lock));
168 vcpu->present = false;
169 if (vcpu->comm != NULL) {
170 uvm_deallocate(kernel_map, (vaddr_t)vcpu->comm, PAGE_SIZE);
171 }
172 }
173
174 int
175 nvmm_vcpu_get(struct nvmm_machine *mach, nvmm_cpuid_t cpuid,
176 struct nvmm_cpu **ret)
177 {
178 struct nvmm_cpu *vcpu;
179
180 if (cpuid >= NVMM_MAX_VCPUS) {
181 return EINVAL;
182 }
183 vcpu = &mach->cpus[cpuid];
184
185 mutex_enter(&vcpu->lock);
186 if (!vcpu->present) {
187 mutex_exit(&vcpu->lock);
188 return ENOENT;
189 }
190 *ret = vcpu;
191
192 return 0;
193 }
194
195 void
196 nvmm_vcpu_put(struct nvmm_cpu *vcpu)
197 {
198 mutex_exit(&vcpu->lock);
199 }
200
201 /* -------------------------------------------------------------------------- */
202
203 static void
204 nvmm_kill_machines(struct nvmm_owner *owner)
205 {
206 struct nvmm_machine *mach;
207 struct nvmm_cpu *vcpu;
208 size_t i, j;
209 int error;
210
211 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
212 mach = &machines[i];
213
214 rw_enter(&mach->lock, RW_WRITER);
215 if (!mach->present || mach->owner != owner) {
216 rw_exit(&mach->lock);
217 continue;
218 }
219
220 /* Kill it. */
221 for (j = 0; j < NVMM_MAX_VCPUS; j++) {
222 error = nvmm_vcpu_get(mach, j, &vcpu);
223 if (error)
224 continue;
225 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
226 nvmm_vcpu_free(mach, vcpu);
227 nvmm_vcpu_put(vcpu);
228 }
229 (*nvmm_impl->machine_destroy)(mach);
230 uvmspace_free(mach->vm);
231
232 /* Drop the kernel UOBJ refs. */
233 for (j = 0; j < NVMM_MAX_HMAPPINGS; j++) {
234 if (!mach->hmap[j].present)
235 continue;
236 uao_detach(mach->hmap[j].uobj);
237 }
238
239 nvmm_machine_free(mach);
240
241 rw_exit(&mach->lock);
242 }
243 }
244
245 /* -------------------------------------------------------------------------- */
246
247 static int
248 nvmm_capability(struct nvmm_owner *owner, struct nvmm_ioc_capability *args)
249 {
250 args->cap.version = NVMM_CAPABILITY_VERSION;
251 args->cap.state_size = nvmm_impl->state_size;
252 args->cap.max_machines = NVMM_MAX_MACHINES;
253 args->cap.max_vcpus = NVMM_MAX_VCPUS;
254 args->cap.max_ram = NVMM_MAX_RAM;
255
256 (*nvmm_impl->capability)(&args->cap);
257
258 return 0;
259 }
260
261 static int
262 nvmm_machine_create(struct nvmm_owner *owner,
263 struct nvmm_ioc_machine_create *args)
264 {
265 struct nvmm_machine *mach;
266 int error;
267
268 error = nvmm_machine_alloc(&mach);
269 if (error)
270 return error;
271
272 /* Curproc owns the machine. */
273 mach->owner = owner;
274
275 /* Zero out the host mappings. */
276 memset(&mach->hmap, 0, sizeof(mach->hmap));
277
278 /* Create the machine vmspace. */
279 mach->gpa_begin = 0;
280 mach->gpa_end = NVMM_MAX_RAM;
281 mach->vm = uvmspace_alloc(0, mach->gpa_end - mach->gpa_begin, false);
282
283 /* Create the comm uobj. */
284 mach->commuobj = uao_create(NVMM_MAX_VCPUS * PAGE_SIZE, 0);
285
286 (*nvmm_impl->machine_create)(mach);
287
288 args->machid = mach->machid;
289 nvmm_machine_put(mach);
290
291 return 0;
292 }
293
294 static int
295 nvmm_machine_destroy(struct nvmm_owner *owner,
296 struct nvmm_ioc_machine_destroy *args)
297 {
298 struct nvmm_machine *mach;
299 struct nvmm_cpu *vcpu;
300 int error;
301 size_t i;
302
303 error = nvmm_machine_get(owner, args->machid, &mach, true);
304 if (error)
305 return error;
306
307 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
308 error = nvmm_vcpu_get(mach, i, &vcpu);
309 if (error)
310 continue;
311
312 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
313 nvmm_vcpu_free(mach, vcpu);
314 nvmm_vcpu_put(vcpu);
315 }
316
317 (*nvmm_impl->machine_destroy)(mach);
318
319 /* Free the machine vmspace. */
320 uvmspace_free(mach->vm);
321
322 /* Drop the kernel UOBJ refs. */
323 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
324 if (!mach->hmap[i].present)
325 continue;
326 uao_detach(mach->hmap[i].uobj);
327 }
328
329 nvmm_machine_free(mach);
330 nvmm_machine_put(mach);
331
332 return 0;
333 }
334
335 static int
336 nvmm_machine_configure(struct nvmm_owner *owner,
337 struct nvmm_ioc_machine_configure *args)
338 {
339 struct nvmm_machine *mach;
340 size_t allocsz;
341 uint64_t op;
342 void *data;
343 int error;
344
345 op = NVMM_MACH_CONF_MD(args->op);
346 if (__predict_false(op >= nvmm_impl->conf_max)) {
347 return EINVAL;
348 }
349
350 allocsz = nvmm_impl->conf_sizes[op];
351 data = kmem_alloc(allocsz, KM_SLEEP);
352
353 error = nvmm_machine_get(owner, args->machid, &mach, true);
354 if (error) {
355 kmem_free(data, allocsz);
356 return error;
357 }
358
359 error = copyin(args->conf, data, allocsz);
360 if (error) {
361 goto out;
362 }
363
364 error = (*nvmm_impl->machine_configure)(mach, op, data);
365
366 out:
367 nvmm_machine_put(mach);
368 kmem_free(data, allocsz);
369 return error;
370 }
371
372 static int
373 nvmm_vcpu_create(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_create *args)
374 {
375 struct nvmm_machine *mach;
376 struct nvmm_cpu *vcpu;
377 int error;
378
379 error = nvmm_machine_get(owner, args->machid, &mach, false);
380 if (error)
381 return error;
382
383 error = nvmm_vcpu_alloc(mach, args->cpuid, &vcpu);
384 if (error)
385 goto out;
386
387 /* Allocate the comm page. */
388 uao_reference(mach->commuobj);
389 error = uvm_map(kernel_map, (vaddr_t *)&vcpu->comm, PAGE_SIZE,
390 mach->commuobj, args->cpuid * PAGE_SIZE, 0, UVM_MAPFLAG(UVM_PROT_RW,
391 UVM_PROT_RW, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
392 if (error) {
393 uao_detach(mach->commuobj);
394 nvmm_vcpu_free(mach, vcpu);
395 nvmm_vcpu_put(vcpu);
396 goto out;
397 }
398 error = uvm_map_pageable(kernel_map, (vaddr_t)vcpu->comm,
399 (vaddr_t)vcpu->comm + PAGE_SIZE, false, 0);
400 if (error) {
401 nvmm_vcpu_free(mach, vcpu);
402 nvmm_vcpu_put(vcpu);
403 goto out;
404 }
405 memset(vcpu->comm, 0, PAGE_SIZE);
406
407 error = (*nvmm_impl->vcpu_create)(mach, vcpu);
408 if (error) {
409 nvmm_vcpu_free(mach, vcpu);
410 nvmm_vcpu_put(vcpu);
411 goto out;
412 }
413
414 nvmm_vcpu_put(vcpu);
415
416 out:
417 nvmm_machine_put(mach);
418 return error;
419 }
420
421 static int
422 nvmm_vcpu_destroy(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_destroy *args)
423 {
424 struct nvmm_machine *mach;
425 struct nvmm_cpu *vcpu;
426 int error;
427
428 error = nvmm_machine_get(owner, args->machid, &mach, false);
429 if (error)
430 return error;
431
432 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
433 if (error)
434 goto out;
435
436 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
437 nvmm_vcpu_free(mach, vcpu);
438 nvmm_vcpu_put(vcpu);
439
440 out:
441 nvmm_machine_put(mach);
442 return error;
443 }
444
445 static int
446 nvmm_vcpu_setstate(struct nvmm_owner *owner,
447 struct nvmm_ioc_vcpu_setstate *args)
448 {
449 struct nvmm_machine *mach;
450 struct nvmm_cpu *vcpu;
451 int error;
452
453 error = nvmm_machine_get(owner, args->machid, &mach, false);
454 if (error)
455 return error;
456
457 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
458 if (error)
459 goto out;
460
461 (*nvmm_impl->vcpu_setstate)(vcpu);
462 nvmm_vcpu_put(vcpu);
463
464 out:
465 nvmm_machine_put(mach);
466 return error;
467 }
468
469 static int
470 nvmm_vcpu_getstate(struct nvmm_owner *owner,
471 struct nvmm_ioc_vcpu_getstate *args)
472 {
473 struct nvmm_machine *mach;
474 struct nvmm_cpu *vcpu;
475 int error;
476
477 error = nvmm_machine_get(owner, args->machid, &mach, false);
478 if (error)
479 return error;
480
481 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
482 if (error)
483 goto out;
484
485 (*nvmm_impl->vcpu_getstate)(vcpu);
486 nvmm_vcpu_put(vcpu);
487
488 out:
489 nvmm_machine_put(mach);
490 return error;
491 }
492
493 static int
494 nvmm_vcpu_inject(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_inject *args)
495 {
496 struct nvmm_machine *mach;
497 struct nvmm_cpu *vcpu;
498 int error;
499
500 error = nvmm_machine_get(owner, args->machid, &mach, false);
501 if (error)
502 return error;
503
504 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
505 if (error)
506 goto out;
507
508 error = (*nvmm_impl->vcpu_inject)(vcpu);
509 nvmm_vcpu_put(vcpu);
510
511 out:
512 nvmm_machine_put(mach);
513 return error;
514 }
515
516 static void
517 nvmm_do_vcpu_run(struct nvmm_machine *mach, struct nvmm_cpu *vcpu,
518 struct nvmm_exit *exit)
519 {
520 struct vmspace *vm = mach->vm;
521
522 while (1) {
523 (*nvmm_impl->vcpu_run)(mach, vcpu, exit);
524
525 if (__predict_true(exit->reason != NVMM_EXIT_MEMORY)) {
526 break;
527 }
528 if (exit->u.mem.gpa >= mach->gpa_end) {
529 break;
530 }
531 if (uvm_fault(&vm->vm_map, exit->u.mem.gpa, exit->u.mem.prot)) {
532 break;
533 }
534 }
535 }
536
537 static int
538 nvmm_vcpu_run(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_run *args)
539 {
540 struct nvmm_machine *mach;
541 struct nvmm_cpu *vcpu;
542 int error;
543
544 error = nvmm_machine_get(owner, args->machid, &mach, false);
545 if (error)
546 return error;
547
548 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
549 if (error)
550 goto out;
551
552 nvmm_do_vcpu_run(mach, vcpu, &args->exit);
553 nvmm_vcpu_put(vcpu);
554
555 out:
556 nvmm_machine_put(mach);
557 return error;
558 }
559
560 /* -------------------------------------------------------------------------- */
561
562 static struct uvm_object *
563 nvmm_hmapping_getuobj(struct nvmm_machine *mach, uintptr_t hva, size_t size,
564 size_t *off)
565 {
566 struct nvmm_hmapping *hmapping;
567 size_t i;
568
569 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
570 hmapping = &mach->hmap[i];
571 if (!hmapping->present) {
572 continue;
573 }
574 if (hva >= hmapping->hva &&
575 hva + size <= hmapping->hva + hmapping->size) {
576 *off = hva - hmapping->hva;
577 return hmapping->uobj;
578 }
579 }
580
581 return NULL;
582 }
583
584 static int
585 nvmm_hmapping_validate(struct nvmm_machine *mach, uintptr_t hva, size_t size)
586 {
587 struct nvmm_hmapping *hmapping;
588 size_t i;
589
590 if ((hva % PAGE_SIZE) != 0 || (size % PAGE_SIZE) != 0) {
591 return EINVAL;
592 }
593 if (hva == 0) {
594 return EINVAL;
595 }
596
597 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
598 hmapping = &mach->hmap[i];
599 if (!hmapping->present) {
600 continue;
601 }
602
603 if (hva >= hmapping->hva &&
604 hva + size <= hmapping->hva + hmapping->size) {
605 break;
606 }
607
608 if (hva >= hmapping->hva &&
609 hva < hmapping->hva + hmapping->size) {
610 return EEXIST;
611 }
612 if (hva + size > hmapping->hva &&
613 hva + size <= hmapping->hva + hmapping->size) {
614 return EEXIST;
615 }
616 if (hva <= hmapping->hva &&
617 hva + size >= hmapping->hva + hmapping->size) {
618 return EEXIST;
619 }
620 }
621
622 return 0;
623 }
624
625 static struct nvmm_hmapping *
626 nvmm_hmapping_alloc(struct nvmm_machine *mach)
627 {
628 struct nvmm_hmapping *hmapping;
629 size_t i;
630
631 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
632 hmapping = &mach->hmap[i];
633 if (!hmapping->present) {
634 hmapping->present = true;
635 return hmapping;
636 }
637 }
638
639 return NULL;
640 }
641
642 static int
643 nvmm_hmapping_free(struct nvmm_machine *mach, uintptr_t hva, size_t size)
644 {
645 struct vmspace *vmspace = curproc->p_vmspace;
646 struct nvmm_hmapping *hmapping;
647 size_t i;
648
649 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
650 hmapping = &mach->hmap[i];
651 if (!hmapping->present || hmapping->hva != hva ||
652 hmapping->size != size) {
653 continue;
654 }
655
656 uvm_unmap(&vmspace->vm_map, hmapping->hva,
657 hmapping->hva + hmapping->size);
658 uao_detach(hmapping->uobj);
659
660 hmapping->uobj = NULL;
661 hmapping->present = false;
662
663 return 0;
664 }
665
666 return ENOENT;
667 }
668
669 static int
670 nvmm_hva_map(struct nvmm_owner *owner, struct nvmm_ioc_hva_map *args)
671 {
672 struct vmspace *vmspace = curproc->p_vmspace;
673 struct nvmm_machine *mach;
674 struct nvmm_hmapping *hmapping;
675 vaddr_t uva;
676 int error;
677
678 error = nvmm_machine_get(owner, args->machid, &mach, true);
679 if (error)
680 return error;
681
682 error = nvmm_hmapping_validate(mach, args->hva, args->size);
683 if (error)
684 goto out;
685
686 hmapping = nvmm_hmapping_alloc(mach);
687 if (hmapping == NULL) {
688 error = ENOBUFS;
689 goto out;
690 }
691
692 hmapping->hva = args->hva;
693 hmapping->size = args->size;
694 hmapping->uobj = uao_create(hmapping->size, 0);
695 uva = hmapping->hva;
696
697 /* Take a reference for the user. */
698 uao_reference(hmapping->uobj);
699
700 /* Map the uobj into the user address space, as pageable. */
701 error = uvm_map(&vmspace->vm_map, &uva, hmapping->size, hmapping->uobj,
702 0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, UVM_INH_SHARE,
703 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
704 if (error) {
705 uao_detach(hmapping->uobj);
706 }
707
708 out:
709 nvmm_machine_put(mach);
710 return error;
711 }
712
713 static int
714 nvmm_hva_unmap(struct nvmm_owner *owner, struct nvmm_ioc_hva_unmap *args)
715 {
716 struct nvmm_machine *mach;
717 int error;
718
719 error = nvmm_machine_get(owner, args->machid, &mach, true);
720 if (error)
721 return error;
722
723 error = nvmm_hmapping_free(mach, args->hva, args->size);
724
725 nvmm_machine_put(mach);
726 return error;
727 }
728
729 /* -------------------------------------------------------------------------- */
730
731 static int
732 nvmm_gpa_map(struct nvmm_owner *owner, struct nvmm_ioc_gpa_map *args)
733 {
734 struct nvmm_machine *mach;
735 struct uvm_object *uobj;
736 gpaddr_t gpa;
737 size_t off;
738 int error;
739
740 error = nvmm_machine_get(owner, args->machid, &mach, false);
741 if (error)
742 return error;
743
744 if ((args->prot & ~(PROT_READ|PROT_WRITE|PROT_EXEC)) != 0) {
745 error = EINVAL;
746 goto out;
747 }
748
749 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0 ||
750 (args->hva % PAGE_SIZE) != 0) {
751 error = EINVAL;
752 goto out;
753 }
754 if (args->hva == 0) {
755 error = EINVAL;
756 goto out;
757 }
758 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
759 error = EINVAL;
760 goto out;
761 }
762 if (args->gpa + args->size <= args->gpa) {
763 error = EINVAL;
764 goto out;
765 }
766 if (args->gpa + args->size > mach->gpa_end) {
767 error = EINVAL;
768 goto out;
769 }
770 gpa = args->gpa;
771
772 uobj = nvmm_hmapping_getuobj(mach, args->hva, args->size, &off);
773 if (uobj == NULL) {
774 error = EINVAL;
775 goto out;
776 }
777
778 /* Take a reference for the machine. */
779 uao_reference(uobj);
780
781 /* Map the uobj into the machine address space, as pageable. */
782 error = uvm_map(&mach->vm->vm_map, &gpa, args->size, uobj, off, 0,
783 UVM_MAPFLAG(args->prot, UVM_PROT_RWX, UVM_INH_NONE,
784 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
785 if (error) {
786 uao_detach(uobj);
787 goto out;
788 }
789 if (gpa != args->gpa) {
790 uao_detach(uobj);
791 printf("[!] uvm_map problem\n");
792 error = EINVAL;
793 goto out;
794 }
795
796 out:
797 nvmm_machine_put(mach);
798 return error;
799 }
800
801 static int
802 nvmm_gpa_unmap(struct nvmm_owner *owner, struct nvmm_ioc_gpa_unmap *args)
803 {
804 struct nvmm_machine *mach;
805 gpaddr_t gpa;
806 int error;
807
808 error = nvmm_machine_get(owner, args->machid, &mach, false);
809 if (error)
810 return error;
811
812 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0) {
813 error = EINVAL;
814 goto out;
815 }
816 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
817 error = EINVAL;
818 goto out;
819 }
820 if (args->gpa + args->size <= args->gpa) {
821 error = EINVAL;
822 goto out;
823 }
824 if (args->gpa + args->size >= mach->gpa_end) {
825 error = EINVAL;
826 goto out;
827 }
828 gpa = args->gpa;
829
830 /* Unmap the memory from the machine. */
831 uvm_unmap(&mach->vm->vm_map, gpa, gpa + args->size);
832
833 out:
834 nvmm_machine_put(mach);
835 return error;
836 }
837
838 /* -------------------------------------------------------------------------- */
839
840 static int
841 nvmm_ctl_mach_info(struct nvmm_ioc_ctl *args)
842 {
843 struct nvmm_ctl_mach_info ctl;
844 struct nvmm_machine *mach;
845 struct nvmm_cpu *vcpu;
846 int error;
847 size_t i;
848
849 if (args->size != sizeof(ctl))
850 return EINVAL;
851 error = copyin(args->data, &ctl, sizeof(ctl));
852 if (error)
853 return error;
854
855 error = nvmm_machine_get(&root_owner, ctl.machid, &mach, true);
856 if (error)
857 return error;
858
859 ctl.nvcpus = 0;
860 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
861 error = nvmm_vcpu_get(mach, i, &vcpu);
862 if (error)
863 continue;
864 ctl.nvcpus++;
865 nvmm_vcpu_put(vcpu);
866 }
867 ctl.pid = mach->owner->pid;
868 ctl.time = mach->time;
869
870 nvmm_machine_put(mach);
871
872 error = copyout(&ctl, args->data, sizeof(ctl));
873 if (error)
874 return error;
875
876 return 0;
877 }
878
879 static int
880 nvmm_ctl(struct nvmm_owner *owner, struct nvmm_ioc_ctl *args)
881 {
882 int error;
883
884 error = kauth_authorize_device(curlwp->l_cred, KAUTH_DEVICE_NVMM_CTL,
885 NULL, NULL, NULL, NULL);
886 if (error)
887 return error;
888
889 switch (args->op) {
890 case NVMM_CTL_MACH_INFO:
891 return nvmm_ctl_mach_info(args);
892 default:
893 return EINVAL;
894 }
895 }
896
897 /* -------------------------------------------------------------------------- */
898
899 static int
900 nvmm_init(void)
901 {
902 size_t i, n;
903
904 for (i = 0; i < __arraycount(nvmm_impl_list); i++) {
905 if (!(*nvmm_impl_list[i]->ident)()) {
906 continue;
907 }
908 nvmm_impl = nvmm_impl_list[i];
909 break;
910 }
911 if (nvmm_impl == NULL) {
912 printf("[!] No implementation found\n");
913 return ENOTSUP;
914 }
915
916 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
917 machines[i].machid = i;
918 rw_init(&machines[i].lock);
919 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
920 machines[i].cpus[n].present = false;
921 machines[i].cpus[n].cpuid = n;
922 mutex_init(&machines[i].cpus[n].lock, MUTEX_DEFAULT,
923 IPL_NONE);
924 }
925 }
926
927 (*nvmm_impl->init)();
928
929 return 0;
930 }
931
932 static void
933 nvmm_fini(void)
934 {
935 size_t i, n;
936
937 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
938 rw_destroy(&machines[i].lock);
939 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
940 mutex_destroy(&machines[i].cpus[n].lock);
941 }
942 }
943
944 (*nvmm_impl->fini)();
945 }
946
947 /* -------------------------------------------------------------------------- */
948
949 static dev_type_open(nvmm_open);
950
951 const struct cdevsw nvmm_cdevsw = {
952 .d_open = nvmm_open,
953 .d_close = noclose,
954 .d_read = noread,
955 .d_write = nowrite,
956 .d_ioctl = noioctl,
957 .d_stop = nostop,
958 .d_tty = notty,
959 .d_poll = nopoll,
960 .d_mmap = nommap,
961 .d_kqfilter = nokqfilter,
962 .d_discard = nodiscard,
963 .d_flag = D_OTHER | D_MPSAFE
964 };
965
966 static int nvmm_ioctl(file_t *, u_long, void *);
967 static int nvmm_close(file_t *);
968 static int nvmm_mmap(file_t *, off_t *, size_t, int, int *, int *,
969 struct uvm_object **, int *);
970
971 const struct fileops nvmm_fileops = {
972 .fo_read = fbadop_read,
973 .fo_write = fbadop_write,
974 .fo_ioctl = nvmm_ioctl,
975 .fo_fcntl = fnullop_fcntl,
976 .fo_poll = fnullop_poll,
977 .fo_stat = fbadop_stat,
978 .fo_close = nvmm_close,
979 .fo_kqfilter = fnullop_kqfilter,
980 .fo_restart = fnullop_restart,
981 .fo_mmap = nvmm_mmap,
982 };
983
984 static int
985 nvmm_open(dev_t dev, int flags, int type, struct lwp *l)
986 {
987 struct nvmm_owner *owner;
988 struct file *fp;
989 int error, fd;
990
991 if (minor(dev) != 0)
992 return EXDEV;
993 error = fd_allocfile(&fp, &fd);
994 if (error)
995 return error;
996
997 owner = kmem_alloc(sizeof(*owner), KM_SLEEP);
998 owner->pid = l->l_proc->p_pid;
999
1000 return fd_clone(fp, fd, flags, &nvmm_fileops, owner);
1001 }
1002
1003 static int
1004 nvmm_close(file_t *fp)
1005 {
1006 struct nvmm_owner *owner = fp->f_data;
1007
1008 KASSERT(owner != NULL);
1009 nvmm_kill_machines(owner);
1010 kmem_free(owner, sizeof(*owner));
1011 fp->f_data = NULL;
1012
1013 return 0;
1014 }
1015
1016 static int
1017 nvmm_mmap(file_t *fp, off_t *offp, size_t size, int prot, int *flagsp,
1018 int *advicep, struct uvm_object **uobjp, int *maxprotp)
1019 {
1020 struct nvmm_owner *owner = fp->f_data;
1021 struct nvmm_machine *mach;
1022 nvmm_machid_t machid;
1023 nvmm_cpuid_t cpuid;
1024 int error;
1025
1026 if (prot & PROT_EXEC)
1027 return EACCES;
1028 if (size != PAGE_SIZE)
1029 return EINVAL;
1030
1031 cpuid = NVMM_COMM_CPUID(*offp);
1032 if (__predict_false(cpuid >= NVMM_MAX_VCPUS))
1033 return EINVAL;
1034
1035 machid = NVMM_COMM_MACHID(*offp);
1036 error = nvmm_machine_get(owner, machid, &mach, false);
1037 if (error)
1038 return error;
1039
1040 uao_reference(mach->commuobj);
1041 *uobjp = mach->commuobj;
1042 *offp = cpuid * PAGE_SIZE;
1043 *maxprotp = prot;
1044 *advicep = UVM_ADV_RANDOM;
1045
1046 nvmm_machine_put(mach);
1047 return 0;
1048 }
1049
1050 static int
1051 nvmm_ioctl(file_t *fp, u_long cmd, void *data)
1052 {
1053 struct nvmm_owner *owner = fp->f_data;
1054
1055 KASSERT(owner != NULL);
1056
1057 switch (cmd) {
1058 case NVMM_IOC_CAPABILITY:
1059 return nvmm_capability(owner, data);
1060 case NVMM_IOC_MACHINE_CREATE:
1061 return nvmm_machine_create(owner, data);
1062 case NVMM_IOC_MACHINE_DESTROY:
1063 return nvmm_machine_destroy(owner, data);
1064 case NVMM_IOC_MACHINE_CONFIGURE:
1065 return nvmm_machine_configure(owner, data);
1066 case NVMM_IOC_VCPU_CREATE:
1067 return nvmm_vcpu_create(owner, data);
1068 case NVMM_IOC_VCPU_DESTROY:
1069 return nvmm_vcpu_destroy(owner, data);
1070 case NVMM_IOC_VCPU_SETSTATE:
1071 return nvmm_vcpu_setstate(owner, data);
1072 case NVMM_IOC_VCPU_GETSTATE:
1073 return nvmm_vcpu_getstate(owner, data);
1074 case NVMM_IOC_VCPU_INJECT:
1075 return nvmm_vcpu_inject(owner, data);
1076 case NVMM_IOC_VCPU_RUN:
1077 return nvmm_vcpu_run(owner, data);
1078 case NVMM_IOC_GPA_MAP:
1079 return nvmm_gpa_map(owner, data);
1080 case NVMM_IOC_GPA_UNMAP:
1081 return nvmm_gpa_unmap(owner, data);
1082 case NVMM_IOC_HVA_MAP:
1083 return nvmm_hva_map(owner, data);
1084 case NVMM_IOC_HVA_UNMAP:
1085 return nvmm_hva_unmap(owner, data);
1086 case NVMM_IOC_CTL:
1087 return nvmm_ctl(owner, data);
1088 default:
1089 return EINVAL;
1090 }
1091 }
1092
1093 /* -------------------------------------------------------------------------- */
1094
1095 void
1096 nvmmattach(int nunits)
1097 {
1098 /* nothing */
1099 }
1100
1101 MODULE(MODULE_CLASS_MISC, nvmm, NULL);
1102
1103 static int
1104 nvmm_modcmd(modcmd_t cmd, void *arg)
1105 {
1106 int error;
1107
1108 switch (cmd) {
1109 case MODULE_CMD_INIT:
1110 error = nvmm_init();
1111 if (error)
1112 return error;
1113
1114 #if defined(_MODULE)
1115 {
1116 devmajor_t bmajor = NODEVMAJOR;
1117 devmajor_t cmajor = 345;
1118
1119 /* mknod /dev/nvmm c 345 0 */
1120 error = devsw_attach("nvmm", NULL, &bmajor,
1121 &nvmm_cdevsw, &cmajor);
1122 if (error) {
1123 nvmm_fini();
1124 return error;
1125 }
1126 }
1127 #endif
1128 return 0;
1129
1130 case MODULE_CMD_FINI:
1131 if (nmachines > 0) {
1132 return EBUSY;
1133 }
1134 #if defined(_MODULE)
1135 {
1136 error = devsw_detach(NULL, &nvmm_cdevsw);
1137 if (error) {
1138 return error;
1139 }
1140 }
1141 #endif
1142 nvmm_fini();
1143 return 0;
1144
1145 case MODULE_CMD_AUTOUNLOAD:
1146 return EBUSY;
1147
1148 default:
1149 return ENOTTY;
1150 }
1151 }
1152