xref: /dflybsd-src/sys/kern/kern_memio.c (revision 857fcb57d0e5e7f20e3d0ef58163c718377e363e)
1 /*-
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1982, 1986, 1990 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department, and code derived from software contributed to
9  * Berkeley by William Jolitz.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	from: Utah $Hdr: mem.c 1.13 89/10/08$
36  *	from: @(#)mem.c	7.2 (Berkeley) 5/9/91
37  * $FreeBSD: src/sys/i386/i386/mem.c,v 1.79.2.9 2003/01/04 22:58:01 njl Exp $
38  */
39 
40 /*
41  * Memory special file
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/buf.h>
47 #include <sys/conf.h>
48 #include <sys/fcntl.h>
49 #include <sys/filio.h>
50 #include <sys/interrupt.h>
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/memrange.h>
54 #include <sys/proc.h>
55 #include <sys/caps.h>
56 #include <sys/queue.h>
57 #include <sys/random.h>
58 #include <sys/signalvar.h>
59 #include <sys/uio.h>
60 #include <sys/vnode.h>
61 #include <sys/sysctl.h>
62 
63 #include <sys/signal2.h>
64 #include <sys/spinlock2.h>
65 
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_map.h>
69 #include <vm/vm_extern.h>
70 
71 
72 static	d_open_t	mmopen;
73 static	d_close_t	mmclose;
74 static	d_read_t	mmread;
75 static	d_write_t	mmwrite;
76 static	d_ioctl_t	mmioctl;
77 #if 0
78 static	d_mmap_t	memmmap;
79 #endif
80 static	d_kqfilter_t	mmkqfilter;
81 static int memuksmap(vm_map_backing_t ba, int op, cdev_t dev, vm_page_t fake);
82 
83 #define CDEV_MAJOR 2
84 static struct dev_ops mem_ops = {
85 	{ "mem", 0, D_MPSAFE | D_QUICK },
86 	.d_open =	mmopen,
87 	.d_close =	mmclose,
88 	.d_read =	mmread,
89 	.d_write =	mmwrite,
90 	.d_ioctl =	mmioctl,
91 	.d_kqfilter =	mmkqfilter,
92 #if 0
93 	.d_mmap =	memmmap,
94 #endif
95 	.d_uksmap =	memuksmap
96 };
97 
98 static struct dev_ops mem_ops_mem = {
99 	{ "mem", 0, D_MEM | D_MPSAFE | D_QUICK },
100 	.d_open =	mmopen,
101 	.d_close =	mmclose,
102 	.d_read =	mmread,
103 	.d_write =	mmwrite,
104 	.d_ioctl =	mmioctl,
105 	.d_kqfilter =	mmkqfilter,
106 #if 0
107 	.d_mmap =	memmmap,
108 #endif
109 	.d_uksmap =	memuksmap
110 };
111 
112 static struct dev_ops mem_ops_noq = {
113 	{ "mem", 0, D_MPSAFE },
114 	.d_open =	mmopen,
115 	.d_close =	mmclose,
116 	.d_read =	mmread,
117 	.d_write =	mmwrite,
118 	.d_ioctl =	mmioctl,
119 	.d_kqfilter =	mmkqfilter,
120 #if 0
121 	.d_mmap =	memmmap,
122 #endif
123 	.d_uksmap =	memuksmap
124 };
125 
126 static int rand_bolt;
127 static caddr_t	zbuf;
128 static cdev_t	zerodev = NULL;
129 static struct lock mem_lock = LOCK_INITIALIZER("memlk", 0, 0);
130 
131 MALLOC_DEFINE(M_MEMDESC, "memdesc", "memory range descriptors");
132 static int mem_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
133 static int random_ioctl (cdev_t, u_long, caddr_t, int, struct ucred *);
134 
135 struct mem_range_softc mem_range_softc;
136 
137 static int seedenable;
138 SYSCTL_INT(_kern, OID_AUTO, seedenable, CTLFLAG_RW, &seedenable, 0, "");
139 
140 static int
141 mmopen(struct dev_open_args *ap)
142 {
143 	cdev_t dev = ap->a_head.a_dev;
144 	int error;
145 
146 	switch (minor(dev)) {
147 	case 0:
148 	case 1:
149 		/*
150 		 * /dev/mem and /dev/kmem
151 		 */
152 		error = caps_priv_check(ap->a_cred, SYSCAP_RESTRICTEDROOT);
153 		if (error == 0) {
154 			if (ap->a_oflags & FWRITE) {
155 				if (securelevel > 0 || kernel_mem_readonly)
156 					error = EPERM;
157 			}
158 		}
159 		break;
160 	case 3:
161 	case 4:
162 		/*
163 		 * /dev/random
164 		 * /dev/urandom
165 		 *
166 		 * Cannot be written to from RESTRICTEDROOT environments.
167 		 */
168 		error = 0;
169 		if (ap->a_oflags & FWRITE) {
170 			error = caps_priv_check(ap->a_cred,
171 						SYSCAP_RESTRICTEDROOT);
172 		}
173 		break;
174 	case 6:
175 		/*
176 		 * /dev/kpmap can only be opened for reading.
177 		 */
178 		error = 0;
179 		if (ap->a_oflags & FWRITE)
180 			error = EPERM;
181 		break;
182 	case 14:
183 		/*
184 		 * /dev/io
185 		 */
186 		error = caps_priv_check(ap->a_cred, SYSCAP_RESTRICTEDROOT);
187 		if (error == 0) {
188 			if (securelevel > 0 || kernel_mem_readonly)
189 				error = EPERM;
190 			else
191 				error = cpu_set_iopl();
192 		}
193 		break;
194 	default:
195 		error = 0;
196 		break;
197 	}
198 	return (error);
199 }
200 
201 static int
202 mmclose(struct dev_close_args *ap)
203 {
204 	cdev_t dev = ap->a_head.a_dev;
205 	int error;
206 
207 	switch (minor(dev)) {
208 	case 14:
209 		error = cpu_clr_iopl();
210 		break;
211 	default:
212 		error = 0;
213 		break;
214 	}
215 	return (error);
216 }
217 
218 
219 static int
220 mmrw(cdev_t dev, struct uio *uio, int flags)
221 {
222 	int o;
223 	u_int c;
224 	u_int poolsize;
225 	u_long v;
226 	struct iovec *iov;
227 	int error = 0;
228 	caddr_t buf = NULL;
229 
230 	while (uio->uio_resid > 0 && error == 0) {
231 		iov = uio->uio_iov;
232 		if (iov->iov_len == 0) {
233 			uio->uio_iov++;
234 			uio->uio_iovcnt--;
235 			if (uio->uio_iovcnt < 0)
236 				panic("mmrw");
237 			continue;
238 		}
239 		switch (minor(dev)) {
240 		case 0:
241 			/*
242 			 * minor device 0 is physical memory, /dev/mem
243 			 */
244 			v = uio->uio_offset;
245 			v &= ~(long)PAGE_MASK;
246 			pmap_kenter((vm_offset_t)ptvmmap, v);
247 			o = (int)uio->uio_offset & PAGE_MASK;
248 			c = (u_int)(PAGE_SIZE - ((uintptr_t)iov->iov_base & PAGE_MASK));
249 			c = min(c, (u_int)(PAGE_SIZE - o));
250 			c = min(c, (u_int)iov->iov_len);
251 			error = uiomove((caddr_t)&ptvmmap[o], (int)c, uio);
252 			pmap_kremove((vm_offset_t)ptvmmap);
253 			continue;
254 
255 		case 1: {
256 			/*
257 			 * minor device 1 is kernel memory, /dev/kmem
258 			 */
259 			vm_offset_t saddr, eaddr;
260 			int prot;
261 
262 			c = iov->iov_len;
263 
264 			/*
265 			 * Make sure that all of the pages are currently
266 			 * resident so that we don't create any zero-fill
267 			 * pages.
268 			 */
269 			saddr = trunc_page(uio->uio_offset);
270 			eaddr = round_page(uio->uio_offset + c);
271 			if (saddr > eaddr)
272 				return EFAULT;
273 
274 			/*
275 			 * Make sure the kernel addresses are mapped.
276 			 * platform_direct_mapped() can be used to bypass
277 			 * default mapping via the page table (virtual kernels
278 			 * contain a lot of out-of-band data).
279 			 */
280 			prot = VM_PROT_READ;
281 			if (uio->uio_rw != UIO_READ)
282 				prot |= VM_PROT_WRITE;
283 			error = kvm_access_check(saddr, eaddr, prot);
284 			if (error)
285 				return (error);
286 			error = uiomove((caddr_t)(vm_offset_t)uio->uio_offset,
287 					(int)c, uio);
288 			continue;
289 		}
290 		case 2:
291 			/*
292 			 * minor device 2 (/dev/null) is EOF/RATHOLE
293 			 */
294 			if (uio->uio_rw == UIO_READ)
295 				return (0);
296 			c = iov->iov_len;
297 			break;
298 		case 3:
299 			/*
300 			 * minor device 3 (/dev/random) is source of filth
301 			 * on read, seeder on write
302 			 */
303 			if (buf == NULL)
304 				buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
305 			c = min(iov->iov_len, PAGE_SIZE);
306 			if (uio->uio_rw == UIO_WRITE) {
307 				error = uiomove(buf, (int)c, uio);
308 				if (error == 0 &&
309 				    seedenable &&
310 				    securelevel <= 0)
311 				{
312 					error = add_buffer_randomness_src(buf, c, RAND_SRC_SEEDING);
313 				} else if (error == 0) {
314 					error = EPERM;
315 				}
316 			} else {
317 				poolsize = read_random(buf, c, 0);
318 				if (poolsize == 0) {
319 					if (buf)
320 						kfree(buf, M_TEMP);
321 					if ((flags & IO_NDELAY) != 0)
322 						return (EWOULDBLOCK);
323 					return (0);
324 				}
325 				c = min(c, poolsize);
326 				error = uiomove(buf, (int)c, uio);
327 			}
328 			continue;
329 		case 4:
330 			/*
331 			 * minor device 4 (/dev/urandom) is source of muck
332 			 * on read, writes are disallowed.
333 			 */
334 			c = min(iov->iov_len, PAGE_SIZE);
335 			if (uio->uio_rw == UIO_WRITE) {
336 				error = EPERM;
337 				break;
338 			}
339 			if (CURSIG(curthread->td_lwp) != 0) {
340 				/*
341 				 * Use tsleep() to get the error code right.
342 				 * It should return immediately.
343 				 */
344 				error = tsleep(&rand_bolt, PCATCH, "urand", 1);
345 				if (error != 0 && error != EWOULDBLOCK)
346 					continue;
347 			}
348 			if (buf == NULL)
349 				buf = kmalloc(PAGE_SIZE, M_TEMP, M_WAITOK);
350 			poolsize = read_random(buf, c, 1);
351 			c = min(c, poolsize);
352 			error = uiomove(buf, (int)c, uio);
353 			continue;
354 		/* case 5: read/write not supported, mmap only */
355 		/* case 6: read/write not supported, mmap only */
356 		case 12:
357 			/*
358 			 * minor device 12 (/dev/zero) is source of nulls
359 			 * on read, write are disallowed.
360 			 */
361 			if (uio->uio_rw == UIO_WRITE) {
362 				c = iov->iov_len;
363 				break;
364 			}
365 			if (zbuf == NULL) {
366 				zbuf = (caddr_t)kmalloc(PAGE_SIZE, M_TEMP,
367 				    M_WAITOK | M_ZERO);
368 			}
369 			c = min(iov->iov_len, PAGE_SIZE);
370 			error = uiomove(zbuf, (int)c, uio);
371 			continue;
372 		default:
373 			return (ENODEV);
374 		}
375 		if (error)
376 			break;
377 		iov->iov_base = (char *)iov->iov_base + c;
378 		iov->iov_len -= c;
379 		uio->uio_offset += c;
380 		uio->uio_resid -= c;
381 	}
382 	if (buf)
383 		kfree(buf, M_TEMP);
384 	return (error);
385 }
386 
387 static int
388 mmread(struct dev_read_args *ap)
389 {
390 	return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
391 }
392 
393 static int
394 mmwrite(struct dev_write_args *ap)
395 {
396 	return(mmrw(ap->a_head.a_dev, ap->a_uio, ap->a_ioflag));
397 }
398 
399 /*******************************************************\
400 * allow user processes to MMAP some memory sections	*
401 * instead of going through read/write			*
402 \*******************************************************/
403 
404 static int user_kernel_mapping(vm_map_backing_t ba, int num,
405 			vm_ooffset_t offset, vm_ooffset_t *resultp);
406 
407 static int
408 memuksmap(vm_map_backing_t ba, int op, cdev_t dev, vm_page_t fake)
409 {
410 	vm_ooffset_t result;
411 	int error;
412 	struct lwp *lp;
413 
414 	error = 0;
415 
416 	switch(op) {
417 	case UKSMAPOP_ADD:
418 		/*
419 		 * We only need to track mappings for /dev/lpmap, all process
420 		 * mappings will be deleted when the process exits and we
421 		 * do not need to track kernel mappings.
422 		 */
423 		if (minor(dev) == 7) {
424 			lp = ba->aux_info;
425 			spin_lock(&lp->lwp_spin);
426 			TAILQ_INSERT_TAIL(&lp->lwp_lpmap_backing_list,
427 					  ba, entry);
428 			spin_unlock(&lp->lwp_spin);
429 		}
430 		break;
431 	case UKSMAPOP_REM:
432 		/*
433 		 * We only need to track mappings for /dev/lpmap, all process
434 		 * mappings will be deleted when the process exits and we
435 		 * do not need to track kernel mappings.
436 		 */
437 		if (minor(dev) == 7) {
438 			lp = ba->aux_info;
439 			spin_lock(&lp->lwp_spin);
440 			TAILQ_REMOVE(&lp->lwp_lpmap_backing_list, ba, entry);
441 			spin_unlock(&lp->lwp_spin);
442 		}
443 		break;
444 	case UKSMAPOP_FAULT:
445 		switch (minor(dev)) {
446 		case 0:
447 			/*
448 			 * minor device 0 is physical memory
449 			 */
450 			fake->phys_addr = ptoa(fake->pindex);
451 			break;
452 		case 1:
453 			/*
454 			 * minor device 1 is kernel memory
455 			 */
456 			fake->phys_addr = vtophys(ptoa(fake->pindex));
457 			break;
458 		case 5:
459 		case 6:
460 		case 7:
461 			/*
462 			 * minor device 5 is /dev/upmap (see sys/upmap.h)
463 			 * minor device 6 is /dev/kpmap (see sys/upmap.h)
464 			 * minor device 7 is /dev/lpmap (see sys/upmap.h)
465 			 */
466 			result = 0;
467 			error = user_kernel_mapping(ba,
468 						    minor(dev),
469 						    ptoa(fake->pindex),
470 						    &result);
471 			fake->phys_addr = result;
472 			break;
473 		default:
474 			error = EINVAL;
475 			break;
476 		}
477 		break;
478 	default:
479 		error = EINVAL;
480 		break;
481 	}
482 	return error;
483 }
484 
485 static int
486 mmioctl(struct dev_ioctl_args *ap)
487 {
488 	cdev_t dev = ap->a_head.a_dev;
489 	int error;
490 
491 	lockmgr(&mem_lock, LK_EXCLUSIVE);
492 
493 	switch (minor(dev)) {
494 	case 0:
495 		error = mem_ioctl(dev, ap->a_cmd, ap->a_data,
496 				  ap->a_fflag, ap->a_cred);
497 		break;
498 	case 3:
499 	case 4:
500 		error = random_ioctl(dev, ap->a_cmd, ap->a_data,
501 				     ap->a_fflag, ap->a_cred);
502 		break;
503 	default:
504 		error = ENODEV;
505 		break;
506 	}
507 
508 	lockmgr(&mem_lock, LK_RELEASE);
509 
510 	return (error);
511 }
512 
513 /*
514  * Operations for changing memory attributes.
515  *
516  * This is basically just an ioctl shim for mem_range_attr_get
517  * and mem_range_attr_set.
518  */
519 static int
520 mem_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
521 {
522 	int nd, error = 0;
523 	struct mem_range_op *mo = (struct mem_range_op *)data;
524 	struct mem_range_desc *md;
525 
526 	/* is this for us? */
527 	if ((cmd != MEMRANGE_GET) &&
528 	    (cmd != MEMRANGE_SET))
529 		return (ENOTTY);
530 
531 	/* any chance we can handle this? */
532 	if (mem_range_softc.mr_op == NULL)
533 		return (EOPNOTSUPP);
534 
535 	/* do we have any descriptors? */
536 	if (mem_range_softc.mr_ndesc == 0)
537 		return (ENXIO);
538 
539 	switch (cmd) {
540 	case MEMRANGE_GET:
541 		nd = imin(mo->mo_arg[0], mem_range_softc.mr_ndesc);
542 		if (nd > 0) {
543 			md = (struct mem_range_desc *)
544 				kmalloc(nd * sizeof(struct mem_range_desc),
545 				       M_MEMDESC, M_WAITOK);
546 			error = mem_range_attr_get(md, &nd);
547 			if (!error)
548 				error = copyout(md, mo->mo_desc,
549 					nd * sizeof(struct mem_range_desc));
550 			kfree(md, M_MEMDESC);
551 		} else {
552 			nd = mem_range_softc.mr_ndesc;
553 		}
554 		mo->mo_arg[0] = nd;
555 		break;
556 
557 	case MEMRANGE_SET:
558 		md = (struct mem_range_desc *)kmalloc(sizeof(struct mem_range_desc),
559 						    M_MEMDESC, M_WAITOK);
560 		error = copyin(mo->mo_desc, md, sizeof(struct mem_range_desc));
561 		/* clamp description string */
562 		md->mr_owner[sizeof(md->mr_owner) - 1] = 0;
563 		if (error == 0)
564 			error = mem_range_attr_set(md, &mo->mo_arg[0]);
565 		kfree(md, M_MEMDESC);
566 		break;
567 	}
568 	return (error);
569 }
570 
571 /*
572  * Implementation-neutral, kernel-callable functions for manipulating
573  * memory range attributes.
574  */
575 int
576 mem_range_attr_get(struct mem_range_desc *mrd, int *arg)
577 {
578 	/* can we handle this? */
579 	if (mem_range_softc.mr_op == NULL)
580 		return (EOPNOTSUPP);
581 
582 	if (*arg == 0) {
583 		*arg = mem_range_softc.mr_ndesc;
584 	} else {
585 		bcopy(mem_range_softc.mr_desc, mrd, (*arg) * sizeof(struct mem_range_desc));
586 	}
587 	return (0);
588 }
589 
590 int
591 mem_range_attr_set(struct mem_range_desc *mrd, int *arg)
592 {
593 	/* can we handle this? */
594 	if (mem_range_softc.mr_op == NULL)
595 		return (EOPNOTSUPP);
596 
597 	return (mem_range_softc.mr_op->set(&mem_range_softc, mrd, arg));
598 }
599 
600 void
601 mem_range_AP_init(void)
602 {
603 	if (mem_range_softc.mr_op && mem_range_softc.mr_op->initAP)
604 		mem_range_softc.mr_op->initAP(&mem_range_softc);
605 }
606 
607 static int
608 random_ioctl(cdev_t dev, u_long cmd, caddr_t data, int flags, struct ucred *cred)
609 {
610 	int error;
611 	int intr;
612 
613 	/*
614 	 * Even inspecting the state is privileged, since it gives a hint
615 	 * about how easily the randomness might be guessed.
616 	 */
617 	error = 0;
618 
619 	switch (cmd) {
620 	/* Really handled in upper layer */
621 	case FIOASYNC:
622 		break;
623 	case MEM_SETIRQ:
624 		intr = *(int16_t *)data;
625 		if ((error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT)) != 0)
626 			break;
627 		if (intr < 0 || intr >= MAX_INTS)
628 			return (EINVAL);
629 		register_randintr(intr);
630 		break;
631 	case MEM_CLEARIRQ:
632 		intr = *(int16_t *)data;
633 		if ((error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT)) != 0)
634 			break;
635 		if (intr < 0 || intr >= MAX_INTS)
636 			return (EINVAL);
637 		unregister_randintr(intr);
638 		break;
639 	case MEM_RETURNIRQ:
640 		error = ENOTSUP;
641 		break;
642 	case MEM_FINDIRQ:
643 		intr = *(int16_t *)data;
644 		if ((error = caps_priv_check(cred, SYSCAP_RESTRICTEDROOT)) != 0)
645 			break;
646 		if (intr < 0 || intr >= MAX_INTS)
647 			return (EINVAL);
648 		intr = next_registered_randintr(intr);
649 		if (intr == MAX_INTS)
650 			return (ENOENT);
651 		*(u_int16_t *)data = intr;
652 		break;
653 	default:
654 		error = ENOTSUP;
655 		break;
656 	}
657 	return (error);
658 }
659 
660 static int
661 mm_filter_read(struct knote *kn, long hint)
662 {
663 	return (1);
664 }
665 
666 static int
667 mm_filter_write(struct knote *kn, long hint)
668 {
669 	return (1);
670 }
671 
672 static void
673 dummy_filter_detach(struct knote *kn) {}
674 
675 /* Implemented in kern_nrandom.c */
676 static struct filterops random_read_filtops =
677         { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, dummy_filter_detach, random_filter_read };
678 
679 static struct filterops mm_read_filtops =
680         { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, dummy_filter_detach, mm_filter_read };
681 
682 static struct filterops mm_write_filtops =
683         { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, dummy_filter_detach, mm_filter_write };
684 
685 static int
686 mmkqfilter(struct dev_kqfilter_args *ap)
687 {
688 	struct knote *kn = ap->a_kn;
689 	cdev_t dev = ap->a_head.a_dev;
690 
691 	ap->a_result = 0;
692 	switch (kn->kn_filter) {
693 	case EVFILT_READ:
694 		switch (minor(dev)) {
695 		case 3:
696 			kn->kn_fop = &random_read_filtops;
697 			break;
698 		default:
699 			kn->kn_fop = &mm_read_filtops;
700 			break;
701 		}
702 		break;
703 	case EVFILT_WRITE:
704 		kn->kn_fop = &mm_write_filtops;
705 		break;
706 	default:
707 		ap->a_result = EOPNOTSUPP;
708 		return (0);
709 	}
710 
711 	return (0);
712 }
713 
714 int
715 iszerodev(cdev_t dev)
716 {
717 	return (zerodev == dev);
718 }
719 
720 /*
721  * /dev/lpmap, /dev/upmap, /dev/kpmap.
722  */
723 static int
724 user_kernel_mapping(vm_map_backing_t ba, int num, vm_ooffset_t offset,
725 		    vm_ooffset_t *resultp)
726 {
727 	struct proc *p;
728 	struct lwp *lp;
729 	int error;
730 	int invfork;
731 
732 	if (offset < 0)
733 		return (EINVAL);
734 
735 	error = EINVAL;
736 
737 	switch(num) {
738 	case 5:
739 		/*
740 		 * /dev/upmap - maps RW per-process shared user-kernel area.
741 		 */
742 
743 		/*
744 		 * If this is a child currently in vfork the pmap is shared
745 		 * with the parent!  We need to actually set-up the parent's
746 		 * p_upmap, not the child's, and we need to set the invfork
747 		 * flag.  Userland will probably adjust its static state so
748 		 * it must be consistent with the parent or userland will be
749 		 * really badly confused.
750 		 *
751 		 * (this situation can happen when user code in vfork() calls
752 		 *  libc's getpid() or some other function which then decides
753 		 *  it wants the upmap).
754 		 */
755 		p = ba->aux_info;
756 		if (p == NULL)
757 			break;
758 		if (p->p_flags & P_PPWAIT) {
759 			p = p->p_pptr;
760 			if (p == NULL)
761 				return (EINVAL);
762 			invfork = 1;
763 		} else {
764 			invfork = 0;
765 		}
766 
767 		/*
768 		 * Create the kernel structure as required, set the invfork
769 		 * flag if we are faulting in on a vfork().
770 		 */
771 		if (p->p_upmap == NULL)
772 			proc_usermap(p, invfork);
773 		if (p->p_upmap && invfork)
774 			p->p_upmap->invfork = invfork;
775 
776 		/*
777 		 * Extract address for pmap
778 		 */
779 		if (p->p_upmap &&
780 		    offset < roundup2(sizeof(*p->p_upmap), PAGE_SIZE)) {
781 			/* only good for current process */
782 			*resultp = pmap_kextract((vm_offset_t)p->p_upmap +
783 						 offset);
784 			error = 0;
785 		}
786 		break;
787 	case 6:
788 		/*
789 		 * /dev/kpmap - maps RO shared kernel global page
790 		 *
791 		 * Extract address for pmap
792 		 */
793 		if (kpmap &&
794 		    offset < roundup2(sizeof(*kpmap), PAGE_SIZE)) {
795 			*resultp = pmap_kextract((vm_offset_t)kpmap + offset);
796 			error = 0;
797 		}
798 		break;
799 	case 7:
800 		/*
801 		 * /dev/lpmap - maps RW per-thread shared user-kernel area.
802 		 */
803 		lp = ba->aux_info;
804 		if (lp == NULL)
805 			break;
806 
807 		/*
808 		 * Create the kernel structure as required
809 		 */
810 		if (lp->lwp_lpmap == NULL)
811 			lwp_usermap(lp, -1);	/* second arg not yet XXX */
812 
813 		/*
814 		 * Extract address for pmap
815 		 */
816 		if (lp->lwp_lpmap &&
817 		    offset < roundup2(sizeof(*lp->lwp_lpmap), PAGE_SIZE)) {
818 			/* only good for current process */
819 			*resultp = pmap_kextract((vm_offset_t)lp->lwp_lpmap +
820 						 offset);
821 			error = 0;
822 		}
823 		break;
824 	default:
825 		break;
826 	}
827 	return error;
828 }
829 
830 static void
831 mem_drvinit(void *unused)
832 {
833 
834 	/* Initialise memory range handling */
835 	if (mem_range_softc.mr_op != NULL)
836 		mem_range_softc.mr_op->init(&mem_range_softc);
837 
838 	make_dev(&mem_ops_mem, 0, UID_ROOT, GID_KMEM, 0640, "mem");
839 	make_dev(&mem_ops_mem, 1, UID_ROOT, GID_KMEM, 0640, "kmem");
840 	make_dev(&mem_ops, 2, UID_ROOT, GID_WHEEL, 0666, "null");
841 	make_dev(&mem_ops, 3, UID_ROOT, GID_WHEEL, 0644, "random");
842 	make_dev(&mem_ops, 4, UID_ROOT, GID_WHEEL, 0644, "urandom");
843 	make_dev(&mem_ops, 5, UID_ROOT, GID_WHEEL, 0666, "upmap");
844 	make_dev(&mem_ops, 6, UID_ROOT, GID_WHEEL, 0444, "kpmap");
845 	make_dev(&mem_ops, 7, UID_ROOT, GID_WHEEL, 0666, "lpmap");
846 	zerodev = make_dev(&mem_ops, 12, UID_ROOT, GID_WHEEL, 0666, "zero");
847 	make_dev(&mem_ops_noq, 14, UID_ROOT, GID_WHEEL, 0600, "io");
848 }
849 
850 SYSINIT(memdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE + CDEV_MAJOR, mem_drvinit,
851     NULL);
852 
853