nvmm.c revision 1.20 1 /* $NetBSD: nvmm.c,v 1.20 2019/05/01 09:20:21 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 1.20 2019/05/01 09:20:21 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 void *data;
342 int error;
343
344 if (__predict_false(args->op >= nvmm_impl->conf_max)) {
345 return EINVAL;
346 }
347
348 allocsz = nvmm_impl->conf_sizes[args->op];
349 data = kmem_alloc(allocsz, KM_SLEEP);
350
351 error = nvmm_machine_get(owner, args->machid, &mach, true);
352 if (error) {
353 kmem_free(data, allocsz);
354 return error;
355 }
356
357 error = copyin(args->conf, data, allocsz);
358 if (error) {
359 goto out;
360 }
361
362 error = (*nvmm_impl->machine_configure)(mach, args->op, data);
363
364 out:
365 nvmm_machine_put(mach);
366 kmem_free(data, allocsz);
367 return error;
368 }
369
370 static int
371 nvmm_vcpu_create(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_create *args)
372 {
373 struct nvmm_machine *mach;
374 struct nvmm_cpu *vcpu;
375 int error;
376
377 error = nvmm_machine_get(owner, args->machid, &mach, false);
378 if (error)
379 return error;
380
381 error = nvmm_vcpu_alloc(mach, args->cpuid, &vcpu);
382 if (error)
383 goto out;
384
385 /* Allocate the comm page. */
386 uao_reference(mach->commuobj);
387 error = uvm_map(kernel_map, (vaddr_t *)&vcpu->comm, PAGE_SIZE,
388 mach->commuobj, args->cpuid * PAGE_SIZE, 0, UVM_MAPFLAG(UVM_PROT_RW,
389 UVM_PROT_RW, UVM_INH_SHARE, UVM_ADV_RANDOM, 0));
390 if (error) {
391 uao_detach(mach->commuobj);
392 nvmm_vcpu_free(mach, vcpu);
393 nvmm_vcpu_put(vcpu);
394 goto out;
395 }
396 error = uvm_map_pageable(kernel_map, (vaddr_t)vcpu->comm,
397 (vaddr_t)vcpu->comm + PAGE_SIZE, false, 0);
398 if (error) {
399 nvmm_vcpu_free(mach, vcpu);
400 nvmm_vcpu_put(vcpu);
401 goto out;
402 }
403 memset(vcpu->comm, 0, PAGE_SIZE);
404
405 error = (*nvmm_impl->vcpu_create)(mach, vcpu);
406 if (error) {
407 nvmm_vcpu_free(mach, vcpu);
408 nvmm_vcpu_put(vcpu);
409 goto out;
410 }
411
412 nvmm_vcpu_put(vcpu);
413
414 out:
415 nvmm_machine_put(mach);
416 return error;
417 }
418
419 static int
420 nvmm_vcpu_destroy(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_destroy *args)
421 {
422 struct nvmm_machine *mach;
423 struct nvmm_cpu *vcpu;
424 int error;
425
426 error = nvmm_machine_get(owner, args->machid, &mach, false);
427 if (error)
428 return error;
429
430 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
431 if (error)
432 goto out;
433
434 (*nvmm_impl->vcpu_destroy)(mach, vcpu);
435 nvmm_vcpu_free(mach, vcpu);
436 nvmm_vcpu_put(vcpu);
437
438 out:
439 nvmm_machine_put(mach);
440 return error;
441 }
442
443 static int
444 nvmm_vcpu_setstate(struct nvmm_owner *owner,
445 struct nvmm_ioc_vcpu_setstate *args)
446 {
447 struct nvmm_machine *mach;
448 struct nvmm_cpu *vcpu;
449 int error;
450
451 error = nvmm_machine_get(owner, args->machid, &mach, false);
452 if (error)
453 return error;
454
455 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
456 if (error)
457 goto out;
458
459 (*nvmm_impl->vcpu_setstate)(vcpu);
460 nvmm_vcpu_put(vcpu);
461
462 out:
463 nvmm_machine_put(mach);
464 return error;
465 }
466
467 static int
468 nvmm_vcpu_getstate(struct nvmm_owner *owner,
469 struct nvmm_ioc_vcpu_getstate *args)
470 {
471 struct nvmm_machine *mach;
472 struct nvmm_cpu *vcpu;
473 int error;
474
475 error = nvmm_machine_get(owner, args->machid, &mach, false);
476 if (error)
477 return error;
478
479 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
480 if (error)
481 goto out;
482
483 (*nvmm_impl->vcpu_getstate)(vcpu);
484 nvmm_vcpu_put(vcpu);
485
486 out:
487 nvmm_machine_put(mach);
488 return error;
489 }
490
491 static int
492 nvmm_vcpu_inject(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_inject *args)
493 {
494 struct nvmm_machine *mach;
495 struct nvmm_cpu *vcpu;
496 int error;
497
498 error = nvmm_machine_get(owner, args->machid, &mach, false);
499 if (error)
500 return error;
501
502 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
503 if (error)
504 goto out;
505
506 error = (*nvmm_impl->vcpu_inject)(vcpu);
507 nvmm_vcpu_put(vcpu);
508
509 out:
510 nvmm_machine_put(mach);
511 return error;
512 }
513
514 static void
515 nvmm_do_vcpu_run(struct nvmm_machine *mach, struct nvmm_cpu *vcpu,
516 struct nvmm_exit *exit)
517 {
518 struct vmspace *vm = mach->vm;
519
520 while (1) {
521 (*nvmm_impl->vcpu_run)(mach, vcpu, exit);
522
523 if (__predict_true(exit->reason != NVMM_EXIT_MEMORY)) {
524 break;
525 }
526 if (exit->u.mem.gpa >= mach->gpa_end) {
527 break;
528 }
529 if (uvm_fault(&vm->vm_map, exit->u.mem.gpa, exit->u.mem.prot)) {
530 break;
531 }
532 }
533 }
534
535 static int
536 nvmm_vcpu_run(struct nvmm_owner *owner, struct nvmm_ioc_vcpu_run *args)
537 {
538 struct nvmm_machine *mach;
539 struct nvmm_cpu *vcpu;
540 int error;
541
542 error = nvmm_machine_get(owner, args->machid, &mach, false);
543 if (error)
544 return error;
545
546 error = nvmm_vcpu_get(mach, args->cpuid, &vcpu);
547 if (error)
548 goto out;
549
550 nvmm_do_vcpu_run(mach, vcpu, &args->exit);
551 nvmm_vcpu_put(vcpu);
552
553 out:
554 nvmm_machine_put(mach);
555 return error;
556 }
557
558 /* -------------------------------------------------------------------------- */
559
560 static struct uvm_object *
561 nvmm_hmapping_getuobj(struct nvmm_machine *mach, uintptr_t hva, size_t size,
562 size_t *off)
563 {
564 struct nvmm_hmapping *hmapping;
565 size_t i;
566
567 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
568 hmapping = &mach->hmap[i];
569 if (!hmapping->present) {
570 continue;
571 }
572 if (hva >= hmapping->hva &&
573 hva + size <= hmapping->hva + hmapping->size) {
574 *off = hva - hmapping->hva;
575 return hmapping->uobj;
576 }
577 }
578
579 return NULL;
580 }
581
582 static int
583 nvmm_hmapping_validate(struct nvmm_machine *mach, uintptr_t hva, size_t size)
584 {
585 struct nvmm_hmapping *hmapping;
586 size_t i;
587
588 if ((hva % PAGE_SIZE) != 0 || (size % PAGE_SIZE) != 0) {
589 return EINVAL;
590 }
591 if (hva == 0) {
592 return EINVAL;
593 }
594
595 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
596 hmapping = &mach->hmap[i];
597 if (!hmapping->present) {
598 continue;
599 }
600
601 if (hva >= hmapping->hva &&
602 hva + size <= hmapping->hva + hmapping->size) {
603 break;
604 }
605
606 if (hva >= hmapping->hva &&
607 hva < hmapping->hva + hmapping->size) {
608 return EEXIST;
609 }
610 if (hva + size > hmapping->hva &&
611 hva + size <= hmapping->hva + hmapping->size) {
612 return EEXIST;
613 }
614 if (hva <= hmapping->hva &&
615 hva + size >= hmapping->hva + hmapping->size) {
616 return EEXIST;
617 }
618 }
619
620 return 0;
621 }
622
623 static struct nvmm_hmapping *
624 nvmm_hmapping_alloc(struct nvmm_machine *mach)
625 {
626 struct nvmm_hmapping *hmapping;
627 size_t i;
628
629 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
630 hmapping = &mach->hmap[i];
631 if (!hmapping->present) {
632 hmapping->present = true;
633 return hmapping;
634 }
635 }
636
637 return NULL;
638 }
639
640 static int
641 nvmm_hmapping_free(struct nvmm_machine *mach, uintptr_t hva, size_t size)
642 {
643 struct vmspace *vmspace = curproc->p_vmspace;
644 struct nvmm_hmapping *hmapping;
645 size_t i;
646
647 for (i = 0; i < NVMM_MAX_HMAPPINGS; i++) {
648 hmapping = &mach->hmap[i];
649 if (!hmapping->present || hmapping->hva != hva ||
650 hmapping->size != size) {
651 continue;
652 }
653
654 uvm_unmap(&vmspace->vm_map, hmapping->hva,
655 hmapping->hva + hmapping->size);
656 uao_detach(hmapping->uobj);
657
658 hmapping->uobj = NULL;
659 hmapping->present = false;
660
661 return 0;
662 }
663
664 return ENOENT;
665 }
666
667 static int
668 nvmm_hva_map(struct nvmm_owner *owner, struct nvmm_ioc_hva_map *args)
669 {
670 struct vmspace *vmspace = curproc->p_vmspace;
671 struct nvmm_machine *mach;
672 struct nvmm_hmapping *hmapping;
673 vaddr_t uva;
674 int error;
675
676 error = nvmm_machine_get(owner, args->machid, &mach, true);
677 if (error)
678 return error;
679
680 error = nvmm_hmapping_validate(mach, args->hva, args->size);
681 if (error)
682 goto out;
683
684 hmapping = nvmm_hmapping_alloc(mach);
685 if (hmapping == NULL) {
686 error = ENOBUFS;
687 goto out;
688 }
689
690 hmapping->hva = args->hva;
691 hmapping->size = args->size;
692 hmapping->uobj = uao_create(hmapping->size, 0);
693 uva = hmapping->hva;
694
695 /* Take a reference for the user. */
696 uao_reference(hmapping->uobj);
697
698 /* Map the uobj into the user address space, as pageable. */
699 error = uvm_map(&vmspace->vm_map, &uva, hmapping->size, hmapping->uobj,
700 0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, UVM_INH_SHARE,
701 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
702 if (error) {
703 uao_detach(hmapping->uobj);
704 }
705
706 out:
707 nvmm_machine_put(mach);
708 return error;
709 }
710
711 static int
712 nvmm_hva_unmap(struct nvmm_owner *owner, struct nvmm_ioc_hva_unmap *args)
713 {
714 struct nvmm_machine *mach;
715 int error;
716
717 error = nvmm_machine_get(owner, args->machid, &mach, true);
718 if (error)
719 return error;
720
721 error = nvmm_hmapping_free(mach, args->hva, args->size);
722
723 nvmm_machine_put(mach);
724 return error;
725 }
726
727 /* -------------------------------------------------------------------------- */
728
729 static int
730 nvmm_gpa_map(struct nvmm_owner *owner, struct nvmm_ioc_gpa_map *args)
731 {
732 struct nvmm_machine *mach;
733 struct uvm_object *uobj;
734 gpaddr_t gpa;
735 size_t off;
736 int error;
737
738 error = nvmm_machine_get(owner, args->machid, &mach, false);
739 if (error)
740 return error;
741
742 if ((args->prot & ~(PROT_READ|PROT_WRITE|PROT_EXEC)) != 0) {
743 error = EINVAL;
744 goto out;
745 }
746
747 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0 ||
748 (args->hva % PAGE_SIZE) != 0) {
749 error = EINVAL;
750 goto out;
751 }
752 if (args->hva == 0) {
753 error = EINVAL;
754 goto out;
755 }
756 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
757 error = EINVAL;
758 goto out;
759 }
760 if (args->gpa + args->size <= args->gpa) {
761 error = EINVAL;
762 goto out;
763 }
764 if (args->gpa + args->size > mach->gpa_end) {
765 error = EINVAL;
766 goto out;
767 }
768 gpa = args->gpa;
769
770 uobj = nvmm_hmapping_getuobj(mach, args->hva, args->size, &off);
771 if (uobj == NULL) {
772 error = EINVAL;
773 goto out;
774 }
775
776 /* Take a reference for the machine. */
777 uao_reference(uobj);
778
779 /* Map the uobj into the machine address space, as pageable. */
780 error = uvm_map(&mach->vm->vm_map, &gpa, args->size, uobj, off, 0,
781 UVM_MAPFLAG(args->prot, UVM_PROT_RWX, UVM_INH_NONE,
782 UVM_ADV_RANDOM, UVM_FLAG_FIXED|UVM_FLAG_UNMAP));
783 if (error) {
784 uao_detach(uobj);
785 goto out;
786 }
787 if (gpa != args->gpa) {
788 uao_detach(uobj);
789 printf("[!] uvm_map problem\n");
790 error = EINVAL;
791 goto out;
792 }
793
794 out:
795 nvmm_machine_put(mach);
796 return error;
797 }
798
799 static int
800 nvmm_gpa_unmap(struct nvmm_owner *owner, struct nvmm_ioc_gpa_unmap *args)
801 {
802 struct nvmm_machine *mach;
803 gpaddr_t gpa;
804 int error;
805
806 error = nvmm_machine_get(owner, args->machid, &mach, false);
807 if (error)
808 return error;
809
810 if ((args->gpa % PAGE_SIZE) != 0 || (args->size % PAGE_SIZE) != 0) {
811 error = EINVAL;
812 goto out;
813 }
814 if (args->gpa < mach->gpa_begin || args->gpa >= mach->gpa_end) {
815 error = EINVAL;
816 goto out;
817 }
818 if (args->gpa + args->size <= args->gpa) {
819 error = EINVAL;
820 goto out;
821 }
822 if (args->gpa + args->size >= mach->gpa_end) {
823 error = EINVAL;
824 goto out;
825 }
826 gpa = args->gpa;
827
828 /* Unmap the memory from the machine. */
829 uvm_unmap(&mach->vm->vm_map, gpa, gpa + args->size);
830
831 out:
832 nvmm_machine_put(mach);
833 return error;
834 }
835
836 /* -------------------------------------------------------------------------- */
837
838 static int
839 nvmm_ctl_mach_info(struct nvmm_ioc_ctl *args)
840 {
841 struct nvmm_ctl_mach_info ctl;
842 struct nvmm_machine *mach;
843 struct nvmm_cpu *vcpu;
844 int error;
845 size_t i;
846
847 if (args->size != sizeof(ctl))
848 return EINVAL;
849 error = copyin(args->data, &ctl, sizeof(ctl));
850 if (error)
851 return error;
852
853 error = nvmm_machine_get(&root_owner, ctl.machid, &mach, true);
854 if (error)
855 return error;
856
857 ctl.nvcpus = 0;
858 for (i = 0; i < NVMM_MAX_VCPUS; i++) {
859 error = nvmm_vcpu_get(mach, i, &vcpu);
860 if (error)
861 continue;
862 ctl.nvcpus++;
863 nvmm_vcpu_put(vcpu);
864 }
865 ctl.pid = mach->owner->pid;
866 ctl.time = mach->time;
867
868 nvmm_machine_put(mach);
869
870 error = copyout(&ctl, args->data, sizeof(ctl));
871 if (error)
872 return error;
873
874 return 0;
875 }
876
877 static int
878 nvmm_ctl(struct nvmm_owner *owner, struct nvmm_ioc_ctl *args)
879 {
880 int error;
881
882 error = kauth_authorize_device(curlwp->l_cred, KAUTH_DEVICE_NVMM_CTL,
883 NULL, NULL, NULL, NULL);
884 if (error)
885 return error;
886
887 switch (args->op) {
888 case NVMM_CTL_MACH_INFO:
889 return nvmm_ctl_mach_info(args);
890 default:
891 return EINVAL;
892 }
893 }
894
895 /* -------------------------------------------------------------------------- */
896
897 static int
898 nvmm_init(void)
899 {
900 size_t i, n;
901
902 for (i = 0; i < __arraycount(nvmm_impl_list); i++) {
903 if (!(*nvmm_impl_list[i]->ident)()) {
904 continue;
905 }
906 nvmm_impl = nvmm_impl_list[i];
907 break;
908 }
909 if (nvmm_impl == NULL) {
910 printf("[!] No implementation found\n");
911 return ENOTSUP;
912 }
913
914 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
915 machines[i].machid = i;
916 rw_init(&machines[i].lock);
917 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
918 machines[i].cpus[n].present = false;
919 machines[i].cpus[n].cpuid = n;
920 mutex_init(&machines[i].cpus[n].lock, MUTEX_DEFAULT,
921 IPL_NONE);
922 }
923 }
924
925 (*nvmm_impl->init)();
926
927 return 0;
928 }
929
930 static void
931 nvmm_fini(void)
932 {
933 size_t i, n;
934
935 for (i = 0; i < NVMM_MAX_MACHINES; i++) {
936 rw_destroy(&machines[i].lock);
937 for (n = 0; n < NVMM_MAX_VCPUS; n++) {
938 mutex_destroy(&machines[i].cpus[n].lock);
939 }
940 }
941
942 (*nvmm_impl->fini)();
943 }
944
945 /* -------------------------------------------------------------------------- */
946
947 static dev_type_open(nvmm_open);
948
949 const struct cdevsw nvmm_cdevsw = {
950 .d_open = nvmm_open,
951 .d_close = noclose,
952 .d_read = noread,
953 .d_write = nowrite,
954 .d_ioctl = noioctl,
955 .d_stop = nostop,
956 .d_tty = notty,
957 .d_poll = nopoll,
958 .d_mmap = nommap,
959 .d_kqfilter = nokqfilter,
960 .d_discard = nodiscard,
961 .d_flag = D_OTHER | D_MPSAFE
962 };
963
964 static int nvmm_ioctl(file_t *, u_long, void *);
965 static int nvmm_close(file_t *);
966 static int nvmm_mmap(file_t *, off_t *, size_t, int, int *, int *,
967 struct uvm_object **, int *);
968
969 const struct fileops nvmm_fileops = {
970 .fo_read = fbadop_read,
971 .fo_write = fbadop_write,
972 .fo_ioctl = nvmm_ioctl,
973 .fo_fcntl = fnullop_fcntl,
974 .fo_poll = fnullop_poll,
975 .fo_stat = fbadop_stat,
976 .fo_close = nvmm_close,
977 .fo_kqfilter = fnullop_kqfilter,
978 .fo_restart = fnullop_restart,
979 .fo_mmap = nvmm_mmap,
980 };
981
982 static int
983 nvmm_open(dev_t dev, int flags, int type, struct lwp *l)
984 {
985 struct nvmm_owner *owner;
986 struct file *fp;
987 int error, fd;
988
989 if (minor(dev) != 0)
990 return EXDEV;
991 error = fd_allocfile(&fp, &fd);
992 if (error)
993 return error;
994
995 owner = kmem_alloc(sizeof(*owner), KM_SLEEP);
996 owner->pid = l->l_proc->p_pid;
997
998 return fd_clone(fp, fd, flags, &nvmm_fileops, owner);
999 }
1000
1001 static int
1002 nvmm_close(file_t *fp)
1003 {
1004 struct nvmm_owner *owner = fp->f_data;
1005
1006 KASSERT(owner != NULL);
1007 nvmm_kill_machines(owner);
1008 kmem_free(owner, sizeof(*owner));
1009 fp->f_data = NULL;
1010
1011 return 0;
1012 }
1013
1014 static int
1015 nvmm_mmap(file_t *fp, off_t *offp, size_t size, int prot, int *flagsp,
1016 int *advicep, struct uvm_object **uobjp, int *maxprotp)
1017 {
1018 struct nvmm_owner *owner = fp->f_data;
1019 struct nvmm_machine *mach;
1020 nvmm_machid_t machid;
1021 nvmm_cpuid_t cpuid;
1022 int error;
1023
1024 if (prot & PROT_EXEC)
1025 return EACCES;
1026 if (size != PAGE_SIZE)
1027 return EINVAL;
1028
1029 cpuid = NVMM_COMM_CPUID(*offp);
1030 if (__predict_false(cpuid >= NVMM_MAX_VCPUS))
1031 return EINVAL;
1032
1033 machid = NVMM_COMM_MACHID(*offp);
1034 error = nvmm_machine_get(owner, machid, &mach, false);
1035 if (error)
1036 return error;
1037
1038 uao_reference(mach->commuobj);
1039 *uobjp = mach->commuobj;
1040 *offp = cpuid * PAGE_SIZE;
1041 *maxprotp = prot;
1042 *advicep = UVM_ADV_RANDOM;
1043
1044 nvmm_machine_put(mach);
1045 return 0;
1046 }
1047
1048 static int
1049 nvmm_ioctl(file_t *fp, u_long cmd, void *data)
1050 {
1051 struct nvmm_owner *owner = fp->f_data;
1052
1053 KASSERT(owner != NULL);
1054
1055 switch (cmd) {
1056 case NVMM_IOC_CAPABILITY:
1057 return nvmm_capability(owner, data);
1058 case NVMM_IOC_MACHINE_CREATE:
1059 return nvmm_machine_create(owner, data);
1060 case NVMM_IOC_MACHINE_DESTROY:
1061 return nvmm_machine_destroy(owner, data);
1062 case NVMM_IOC_MACHINE_CONFIGURE:
1063 return nvmm_machine_configure(owner, data);
1064 case NVMM_IOC_VCPU_CREATE:
1065 return nvmm_vcpu_create(owner, data);
1066 case NVMM_IOC_VCPU_DESTROY:
1067 return nvmm_vcpu_destroy(owner, data);
1068 case NVMM_IOC_VCPU_SETSTATE:
1069 return nvmm_vcpu_setstate(owner, data);
1070 case NVMM_IOC_VCPU_GETSTATE:
1071 return nvmm_vcpu_getstate(owner, data);
1072 case NVMM_IOC_VCPU_INJECT:
1073 return nvmm_vcpu_inject(owner, data);
1074 case NVMM_IOC_VCPU_RUN:
1075 return nvmm_vcpu_run(owner, data);
1076 case NVMM_IOC_GPA_MAP:
1077 return nvmm_gpa_map(owner, data);
1078 case NVMM_IOC_GPA_UNMAP:
1079 return nvmm_gpa_unmap(owner, data);
1080 case NVMM_IOC_HVA_MAP:
1081 return nvmm_hva_map(owner, data);
1082 case NVMM_IOC_HVA_UNMAP:
1083 return nvmm_hva_unmap(owner, data);
1084 case NVMM_IOC_CTL:
1085 return nvmm_ctl(owner, data);
1086 default:
1087 return EINVAL;
1088 }
1089 }
1090
1091 /* -------------------------------------------------------------------------- */
1092
1093 void
1094 nvmmattach(int nunits)
1095 {
1096 /* nothing */
1097 }
1098
1099 MODULE(MODULE_CLASS_MISC, nvmm, NULL);
1100
1101 static int
1102 nvmm_modcmd(modcmd_t cmd, void *arg)
1103 {
1104 int error;
1105
1106 switch (cmd) {
1107 case MODULE_CMD_INIT:
1108 error = nvmm_init();
1109 if (error)
1110 return error;
1111
1112 #if defined(_MODULE)
1113 {
1114 devmajor_t bmajor = NODEVMAJOR;
1115 devmajor_t cmajor = 345;
1116
1117 /* mknod /dev/nvmm c 345 0 */
1118 error = devsw_attach("nvmm", NULL, &bmajor,
1119 &nvmm_cdevsw, &cmajor);
1120 if (error) {
1121 nvmm_fini();
1122 return error;
1123 }
1124 }
1125 #endif
1126 return 0;
1127
1128 case MODULE_CMD_FINI:
1129 if (nmachines > 0) {
1130 return EBUSY;
1131 }
1132 #if defined(_MODULE)
1133 {
1134 error = devsw_detach(NULL, &nvmm_cdevsw);
1135 if (error) {
1136 return error;
1137 }
1138 }
1139 #endif
1140 nvmm_fini();
1141 return 0;
1142
1143 case MODULE_CMD_AUTOUNLOAD:
1144 return EBUSY;
1145
1146 default:
1147 return ENOTTY;
1148 }
1149 }
1150