nvmm.c revision 1.17 1 /* $NetBSD: nvmm.c,v 1.17 2019/04/10 18:49:04 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 1.17 2019/04/10 18:49:04 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, struct nvmm_cpu **ret)
142 {
143 struct nvmm_cpu *vcpu;
144 size_t i;
145
146 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
147 vcpu = &mach->cpus[i];
148
149 mutex_enter(&vcpu->lock);
150 if (vcpu->present) {
151 mutex_exit(&vcpu->lock);
152 continue;
153 }
154
155 vcpu->present = true;
156 vcpu->cpuid = i;
157 vcpu->state = kmem_zalloc(nvmm_impl->state_size, KM_SLEEP);
158 *ret = vcpu;
159 return 0;
160 }
161
162 return ENOBUFS;
163 }
164
165 static void
166 nvmm_vcpu_free(struct nvmm_machine *mach, struct nvmm_cpu *vcpu)
167 {
168 KASSERT(mutex_owned(&vcpu->lock));
169 vcpu->present = false;
170 kmem_free(vcpu->state, nvmm_impl->state_size);
171 vcpu->hcpu_last = -1;
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 (*nvmm_impl->machine_create)(mach);
284
285 args->machid = mach->machid;
286 nvmm_machine_put(mach);
287
288 return 0;
289 }
290
291 static int
292 nvmm_machine_destroy(struct nvmm_owner *owner,
293 struct nvmm_ioc_machine_destroy *args)
294 {
295 struct nvmm_machine *mach;
296 struct nvmm_cpu *vcpu;
297 int error;
298 size_t i;
299
300 error = nvmm_machine_get(owner, args->machid, &mach, true);
301 if (error)
302 return error;
303
304 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
305 error = nvmm_vcpu_get(mach, i, &vcpu);
306 if (error)
307 continue;
308
309 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
310 nvmm_vcpu_free(mach, vcpu);
311 nvmm_vcpu_put(vcpu);
312 }
313
314 (*nvmm_impl->machine_destroy)(mach);
315
316 /* Free the machine vmspace. */
317 uvmspace_free(mach->vm);
318
319 /* Drop the kernel UOBJ refs. */
320 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
321 if (!mach->hmap[i].present)
322 continue;
323 uao_detach(mach->hmap[i].uobj);
324 }
325
326 nvmm_machine_free(mach);
327 nvmm_machine_put(mach);
328
329 return 0;
330 }
331
332 static int
333 nvmm_machine_configure(struct nvmm_owner *owner,
334 struct nvmm_ioc_machine_configure *args)
335 {
336 struct nvmm_machine *mach;
337 size_t allocsz;
338 void *data;
339 int error;
340
341 if (__predict_false(args->op >= nvmm_impl->conf_max)) {
342 return EINVAL;
343 }
344
345 allocsz = nvmm_impl->conf_sizes[args->op];
346 data = kmem_alloc(allocsz, KM_SLEEP);
347
348 error = nvmm_machine_get(owner, args->machid, &mach, true);
349 if (error) {
350 kmem_free(data, allocsz);
351 return error;
352 }
353
354 error = copyin(args->conf, data, allocsz);
355 if (error) {
356 goto out;
357 }
358
359 error = (*nvmm_impl->machine_configure)(mach, args->op, data);
360
361 out:
362 nvmm_machine_put(mach);
363 kmem_free(data, allocsz);
364 return error;
365 }
366
367 static int
368 nvmm_vcpu_create(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_create *args)
369 {
370 struct nvmm_machine *mach;
371 struct nvmm_cpu *vcpu;
372 int error;
373
374 error = nvmm_machine_get(owner, args->machid, &mach, false);
375 if (error)
376 return error;
377
378 error = nvmm_vcpu_alloc(mach, &vcpu);
379 if (error)
380 goto out;
381
382 error = (*nvmm_impl->vcpu_create)(mach, vcpu);
383 if (error) {
384 nvmm_vcpu_free(mach, vcpu);
385 nvmm_vcpu_put(vcpu);
386 goto out;
387 }
388
389 nvmm_vcpu_put(vcpu);
390
391 out:
392 nvmm_machine_put(mach);
393 return error;
394 }
395
396 static int
397 nvmm_vcpu_destroy(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_destroy *args)
398 {
399 struct nvmm_machine *mach;
400 struct nvmm_cpu *vcpu;
401 int error;
402
403 error = nvmm_machine_get(owner, args->machid, &mach, false);
404 if (error)
405 return error;
406
407 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
408 if (error)
409 goto out;
410
411 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
412 nvmm_vcpu_free(mach, vcpu);
413 nvmm_vcpu_put(vcpu);
414
415 out:
416 nvmm_machine_put(mach);
417 return error;
418 }
419
420 static int
421 nvmm_vcpu_setstate(struct nvmm_owner *owner,
422 struct nvmm_ioc_vcpu_setstate *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 error = copyin(args->state, vcpu->state, nvmm_impl->state_size);
437 if (error) {
438 nvmm_vcpu_put(vcpu);
439 goto out;
440 }
441
442 (*nvmm_impl->vcpu_setstate)(vcpu, vcpu->state, args->flags);
443 nvmm_vcpu_put(vcpu);
444
445 out:
446 nvmm_machine_put(mach);
447 return error;
448 }
449
450 static int
451 nvmm_vcpu_getstate(struct nvmm_owner *owner,
452 struct nvmm_ioc_vcpu_getstate *args)
453 {
454 struct nvmm_machine *mach;
455 struct nvmm_cpu *vcpu;
456 int error;
457
458 error = nvmm_machine_get(owner, args->machid, &mach, false);
459 if (error)
460 return error;
461
462 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
463 if (error)
464 goto out;
465
466 (*nvmm_impl->vcpu_getstate)(vcpu, vcpu->state, args->flags);
467 nvmm_vcpu_put(vcpu);
468 error = copyout(vcpu->state, args->state, nvmm_impl->state_size);
469
470 out:
471 nvmm_machine_put(mach);
472 return error;
473 }
474
475 static int
476 nvmm_vcpu_inject(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_inject *args)
477 {
478 struct nvmm_machine *mach;
479 struct nvmm_cpu *vcpu;
480 int error;
481
482 error = nvmm_machine_get(owner, args->machid, &mach, false);
483 if (error)
484 return error;
485
486 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
487 if (error)
488 goto out;
489
490 error = (*nvmm_impl->vcpu_inject)(mach, vcpu, &args->event);
491 nvmm_vcpu_put(vcpu);
492
493 out:
494 nvmm_machine_put(mach);
495 return error;
496 }
497
498 static void
499 nvmm_do_vcpu_run(struct nvmm_machine *mach, struct nvmm_cpu *vcpu,
500 struct nvmm_exit *exit)
501 {
502 struct vmspace *vm = mach->vm;
503
504 while (1) {
505 (*nvmm_impl->vcpu_run)(mach, vcpu, exit);
506
507 if (__predict_true(exit->reason != NVMM_EXIT_MEMORY)) {
508 break;
509 }
510 if (exit->u.mem.gpa >= mach->gpa_end) {
511 break;
512 }
513 if (uvm_fault(&vm->vm_map, exit->u.mem.gpa, exit->u.mem.prot)) {
514 break;
515 }
516 }
517 }
518
519 static int
520 nvmm_vcpu_run(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_run *args)
521 {
522 struct nvmm_machine *mach;
523 struct nvmm_cpu *vcpu;
524 int error;
525
526 error = nvmm_machine_get(owner, args->machid, &mach, false);
527 if (error)
528 return error;
529
530 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
531 if (error)
532 goto out;
533
534 nvmm_do_vcpu_run(mach, vcpu, &args->exit);
535 nvmm_vcpu_put(vcpu);
536
537 out:
538 nvmm_machine_put(mach);
539 return error;
540 }
541
542 /* -------------------------------------------------------------------------- */
543
544 static struct uvm_object *
545 nvmm_hmapping_getuobj(struct nvmm_machine *mach, uintptr_t hva, size_t size,
546 size_t *off)
547 {
548 struct nvmm_hmapping *hmapping;
549 size_t i;
550
551 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
552 hmapping = &mach->hmap[i];
553 if (!hmapping->present) {
554 continue;
555 }
556 if (hva >= hmapping->hva &&
557 hva + size <= hmapping->hva + hmapping->size) {
558 *off = hva - hmapping->hva;
559 return hmapping->uobj;
560 }
561 }
562
563 return NULL;
564 }
565
566 static int
567 nvmm_hmapping_validate(struct nvmm_machine *mach, uintptr_t hva, size_t size)
568 {
569 struct nvmm_hmapping *hmapping;
570 size_t i;
571
572 if ((hva % PAGE_SIZE) != 0 || (size % PAGE_SIZE) != 0) {
573 return EINVAL;
574 }
575 if (hva == 0) {
576 return EINVAL;
577 }
578
579 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
580 hmapping = &mach->hmap[i];
581 if (!hmapping->present) {
582 continue;
583 }
584
585 if (hva >= hmapping->hva &&
586 hva + size <= hmapping->hva + hmapping->size) {
587 break;
588 }
589
590 if (hva >= hmapping->hva &&
591 hva < hmapping->hva + hmapping->size) {
592 return EEXIST;
593 }
594 if (hva + size > hmapping->hva &&
595 hva + size <= hmapping->hva + hmapping->size) {
596 return EEXIST;
597 }
598 if (hva <= hmapping->hva &&
599 hva + size >= hmapping->hva + hmapping->size) {
600 return EEXIST;
601 }
602 }
603
604 return 0;
605 }
606
607 static struct nvmm_hmapping *
608 nvmm_hmapping_alloc(struct nvmm_machine *mach)
609 {
610 struct nvmm_hmapping *hmapping;
611 size_t i;
612
613 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
614 hmapping = &mach->hmap[i];
615 if (!hmapping->present) {
616 hmapping->present = true;
617 return hmapping;
618 }
619 }
620
621 return NULL;
622 }
623
624 static int
625 nvmm_hmapping_free(struct nvmm_machine *mach, uintptr_t hva, size_t size)
626 {
627 struct vmspace *vmspace = curproc->p_vmspace;
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 || hmapping->hva != hva ||
634 hmapping->size != size) {
635 continue;
636 }
637
638 uvm_unmap(&vmspace->vm_map, hmapping->hva,
639 hmapping->hva + hmapping->size);
640 uao_detach(hmapping->uobj);
641
642 hmapping->uobj = NULL;
643 hmapping->present = false;
644
645 return 0;
646 }
647
648 return ENOENT;
649 }
650
651 static int
652 nvmm_hva_map(struct nvmm_owner *owner, struct nvmm_ioc_hva_map *args)
653 {
654 struct vmspace *vmspace = curproc->p_vmspace;
655 struct nvmm_machine *mach;
656 struct nvmm_hmapping *hmapping;
657 vaddr_t uva;
658 int error;
659
660 error = nvmm_machine_get(owner, args->machid, &mach, true);
661 if (error)
662 return error;
663
664 error = nvmm_hmapping_validate(mach, args->hva, args->size);
665 if (error)
666 goto out;
667
668 hmapping = nvmm_hmapping_alloc(mach);
669 if (hmapping == NULL) {
670 error = ENOBUFS;
671 goto out;
672 }
673
674 hmapping->hva = args->hva;
675 hmapping->size = args->size;
676 hmapping->uobj = uao_create(hmapping->size, 0);
677 uva = hmapping->hva;
678
679 /* Take a reference for the user. */
680 uao_reference(hmapping->uobj);
681
682 /* Map the uobj into the user address space, as pageable. */
683 error = uvm_map(&vmspace->vm_map, &uva, hmapping->size, hmapping->uobj,
684 0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, UVM_INH_SHARE,
685 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
686 if (error) {
687 uao_detach(hmapping->uobj);
688 }
689
690 out:
691 nvmm_machine_put(mach);
692 return error;
693 }
694
695 static int
696 nvmm_hva_unmap(struct nvmm_owner *owner, struct nvmm_ioc_hva_unmap *args)
697 {
698 struct nvmm_machine *mach;
699 int error;
700
701 error = nvmm_machine_get(owner, args->machid, &mach, true);
702 if (error)
703 return error;
704
705 error = nvmm_hmapping_free(mach, args->hva, args->size);
706
707 nvmm_machine_put(mach);
708 return error;
709 }
710
711 /* -------------------------------------------------------------------------- */
712
713 static int
714 nvmm_gpa_map(struct nvmm_owner *owner, struct nvmm_ioc_gpa_map *args)
715 {
716 struct nvmm_machine *mach;
717 struct uvm_object *uobj;
718 gpaddr_t gpa;
719 size_t off;
720 int error;
721
722 error = nvmm_machine_get(owner, args->machid, &mach, false);
723 if (error)
724 return error;
725
726 if ((args->prot & ~(PROT_READ|PROT_WRITE|PROT_EXEC)) != 0) {
727 error = EINVAL;
728 goto out;
729 }
730
731 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0 ||
732 (args->hva % PAGE_SIZE) != 0) {
733 error = EINVAL;
734 goto out;
735 }
736 if (args->hva == 0) {
737 error = EINVAL;
738 goto out;
739 }
740 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
741 error = EINVAL;
742 goto out;
743 }
744 if (args->gpa + args->size <= args->gpa) {
745 error = EINVAL;
746 goto out;
747 }
748 if (args->gpa + args->size > mach->gpa_end) {
749 error = EINVAL;
750 goto out;
751 }
752 gpa = args->gpa;
753
754 uobj = nvmm_hmapping_getuobj(mach, args->hva, args->size, &off);
755 if (uobj == NULL) {
756 error = EINVAL;
757 goto out;
758 }
759
760 /* Take a reference for the machine. */
761 uao_reference(uobj);
762
763 /* Map the uobj into the machine address space, as pageable. */
764 error = uvm_map(&mach->vm->vm_map, &gpa, args->size, uobj, off, 0,
765 UVM_MAPFLAG(args->prot, UVM_PROT_RWX, UVM_INH_NONE,
766 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
767 if (error) {
768 uao_detach(uobj);
769 goto out;
770 }
771 if (gpa != args->gpa) {
772 uao_detach(uobj);
773 printf("[!] uvm_map problem\n");
774 error = EINVAL;
775 goto out;
776 }
777
778 out:
779 nvmm_machine_put(mach);
780 return error;
781 }
782
783 static int
784 nvmm_gpa_unmap(struct nvmm_owner *owner, struct nvmm_ioc_gpa_unmap *args)
785 {
786 struct nvmm_machine *mach;
787 gpaddr_t gpa;
788 int error;
789
790 error = nvmm_machine_get(owner, args->machid, &mach, false);
791 if (error)
792 return error;
793
794 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0) {
795 error = EINVAL;
796 goto out;
797 }
798 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
799 error = EINVAL;
800 goto out;
801 }
802 if (args->gpa + args->size <= args->gpa) {
803 error = EINVAL;
804 goto out;
805 }
806 if (args->gpa + args->size >= mach->gpa_end) {
807 error = EINVAL;
808 goto out;
809 }
810 gpa = args->gpa;
811
812 /* Unmap the memory from the machine. */
813 uvm_unmap(&mach->vm->vm_map, gpa, gpa + args->size);
814
815 out:
816 nvmm_machine_put(mach);
817 return error;
818 }
819
820 /* -------------------------------------------------------------------------- */
821
822 static int
823 nvmm_ctl_mach_info(struct nvmm_ioc_ctl *args)
824 {
825 struct nvmm_ctl_mach_info ctl;
826 struct nvmm_machine *mach;
827 struct nvmm_cpu *vcpu;
828 int error;
829 size_t i;
830
831 if (args->size != sizeof(ctl))
832 return EINVAL;
833 error = copyin(args->data, &ctl, sizeof(ctl));
834 if (error)
835 return error;
836
837 error = nvmm_machine_get(&root_owner, ctl.machid, &mach, true);
838 if (error)
839 return error;
840
841 ctl.nvcpus = 0;
842 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
843 error = nvmm_vcpu_get(mach, i, &vcpu);
844 if (error)
845 continue;
846 ctl.nvcpus++;
847 nvmm_vcpu_put(vcpu);
848 }
849 ctl.pid = mach->owner->pid;
850 ctl.time = mach->time;
851
852 nvmm_machine_put(mach);
853
854 error = copyout(&ctl, args->data, sizeof(ctl));
855 if (error)
856 return error;
857
858 return 0;
859 }
860
861 static int
862 nvmm_ctl(struct nvmm_owner *owner, struct nvmm_ioc_ctl *args)
863 {
864 int error;
865
866 error = kauth_authorize_device(curlwp->l_cred, KAUTH_DEVICE_NVMM_CTL,
867 NULL, NULL, NULL, NULL);
868 if (error)
869 return error;
870
871 switch (args->op) {
872 case NVMM_CTL_MACH_INFO:
873 return nvmm_ctl_mach_info(args);
874 default:
875 return EINVAL;
876 }
877 }
878
879 /* -------------------------------------------------------------------------- */
880
881 static int
882 nvmm_init(void)
883 {
884 size_t i, n;
885
886 for (i = 0; i < __arraycount(nvmm_impl_list); i++) {
887 if (!(*nvmm_impl_list[i]->ident)()) {
888 continue;
889 }
890 nvmm_impl = nvmm_impl_list[i];
891 break;
892 }
893 if (nvmm_impl == NULL) {
894 printf("[!] No implementation found\n");
895 return ENOTSUP;
896 }
897
898 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
899 machines[i].machid = i;
900 rw_init(&machines[i].lock);
901 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
902 mutex_init(&machines[i].cpus[n].lock, MUTEX_DEFAULT,
903 IPL_NONE);
904 machines[i].cpus[n].hcpu_last = -1;
905 }
906 }
907
908 (*nvmm_impl->init)();
909
910 return 0;
911 }
912
913 static void
914 nvmm_fini(void)
915 {
916 size_t i, n;
917
918 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
919 rw_destroy(&machines[i].lock);
920 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
921 mutex_destroy(&machines[i].cpus[n].lock);
922 }
923 }
924
925 (*nvmm_impl->fini)();
926 }
927
928 /* -------------------------------------------------------------------------- */
929
930 static dev_type_open(nvmm_open);
931
932 const struct cdevsw nvmm_cdevsw = {
933 .d_open = nvmm_open,
934 .d_close = noclose,
935 .d_read = noread,
936 .d_write = nowrite,
937 .d_ioctl = noioctl,
938 .d_stop = nostop,
939 .d_tty = notty,
940 .d_poll = nopoll,
941 .d_mmap = nommap,
942 .d_kqfilter = nokqfilter,
943 .d_discard = nodiscard,
944 .d_flag = D_OTHER | D_MPSAFE
945 };
946
947 static int nvmm_ioctl(file_t *, u_long, void *);
948 static int nvmm_close(file_t *);
949
950 const struct fileops nvmm_fileops = {
951 .fo_read = fbadop_read,
952 .fo_write = fbadop_write,
953 .fo_ioctl = nvmm_ioctl,
954 .fo_fcntl = fnullop_fcntl,
955 .fo_poll = fnullop_poll,
956 .fo_stat = fbadop_stat,
957 .fo_close = nvmm_close,
958 .fo_kqfilter = fnullop_kqfilter,
959 .fo_restart = fnullop_restart,
960 .fo_mmap = NULL,
961 };
962
963 static int
964 nvmm_open(dev_t dev, int flags, int type, struct lwp *l)
965 {
966 struct nvmm_owner *owner;
967 struct file *fp;
968 int error, fd;
969
970 if (minor(dev) != 0)
971 return EXDEV;
972 error = fd_allocfile(&fp, &fd);
973 if (error)
974 return error;
975
976 owner = kmem_alloc(sizeof(*owner), KM_SLEEP);
977 owner->pid = l->l_proc->p_pid;
978
979 return fd_clone(fp, fd, flags, &nvmm_fileops, owner);
980 }
981
982 static int
983 nvmm_close(file_t *fp)
984 {
985 struct nvmm_owner *owner = fp->f_data;
986
987 KASSERT(owner != NULL);
988 nvmm_kill_machines(owner);
989 kmem_free(owner, sizeof(*owner));
990 fp->f_data = NULL;
991
992 return 0;
993 }
994
995 static int
996 nvmm_ioctl(file_t *fp, u_long cmd, void *data)
997 {
998 struct nvmm_owner *owner = fp->f_data;
999
1000 KASSERT(owner != NULL);
1001
1002 switch (cmd) {
1003 case NVMM_IOC_CAPABILITY:
1004 return nvmm_capability(owner, data);
1005 case NVMM_IOC_MACHINE_CREATE:
1006 return nvmm_machine_create(owner, data);
1007 case NVMM_IOC_MACHINE_DESTROY:
1008 return nvmm_machine_destroy(owner, data);
1009 case NVMM_IOC_MACHINE_CONFIGURE:
1010 return nvmm_machine_configure(owner, data);
1011 case NVMM_IOC_VCPU_CREATE:
1012 return nvmm_vcpu_create(owner, data);
1013 case NVMM_IOC_VCPU_DESTROY:
1014 return nvmm_vcpu_destroy(owner, data);
1015 case NVMM_IOC_VCPU_SETSTATE:
1016 return nvmm_vcpu_setstate(owner, data);
1017 case NVMM_IOC_VCPU_GETSTATE:
1018 return nvmm_vcpu_getstate(owner, data);
1019 case NVMM_IOC_VCPU_INJECT:
1020 return nvmm_vcpu_inject(owner, data);
1021 case NVMM_IOC_VCPU_RUN:
1022 return nvmm_vcpu_run(owner, data);
1023 case NVMM_IOC_GPA_MAP:
1024 return nvmm_gpa_map(owner, data);
1025 case NVMM_IOC_GPA_UNMAP:
1026 return nvmm_gpa_unmap(owner, data);
1027 case NVMM_IOC_HVA_MAP:
1028 return nvmm_hva_map(owner, data);
1029 case NVMM_IOC_HVA_UNMAP:
1030 return nvmm_hva_unmap(owner, data);
1031 case NVMM_IOC_CTL:
1032 return nvmm_ctl(owner, data);
1033 default:
1034 return EINVAL;
1035 }
1036 }
1037
1038 /* -------------------------------------------------------------------------- */
1039
1040 void
1041 nvmmattach(int nunits)
1042 {
1043 /* nothing */
1044 }
1045
1046 MODULE(MODULE_CLASS_MISC, nvmm, NULL);
1047
1048 static int
1049 nvmm_modcmd(modcmd_t cmd, void *arg)
1050 {
1051 int error;
1052
1053 switch (cmd) {
1054 case MODULE_CMD_INIT:
1055 error = nvmm_init();
1056 if (error)
1057 return error;
1058
1059 #if defined(_MODULE)
1060 {
1061 devmajor_t bmajor = NODEVMAJOR;
1062 devmajor_t cmajor = 345;
1063
1064 /* mknod /dev/nvmm c 345 0 */
1065 error = devsw_attach("nvmm", NULL, &bmajor,
1066 &nvmm_cdevsw, &cmajor);
1067 if (error) {
1068 nvmm_fini();
1069 return error;
1070 }
1071 }
1072 #endif
1073 return 0;
1074
1075 case MODULE_CMD_FINI:
1076 if (nmachines > 0) {
1077 return EBUSY;
1078 }
1079 #if defined(_MODULE)
1080 {
1081 error = devsw_detach(NULL, &nvmm_cdevsw);
1082 if (error) {
1083 return error;
1084 }
1085 }
1086 #endif
1087 nvmm_fini();
1088 return 0;
1089
1090 case MODULE_CMD_AUTOUNLOAD:
1091 return EBUSY;
1092
1093 default:
1094 return ENOTTY;
1095 }
1096 }
1097