xref: /freebsd/sys/vm/vm_mmap.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1988 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
37  *
38  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
39  */
40 
41 /*
42  * Mapped file (mmap) interface to VM
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_hwpmc_hooks.h"
49 #include "opt_vm.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
58 #include <sys/filedesc.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procctl.h>
62 #include <sys/racct.h>
63 #include <sys/resource.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/mman.h>
71 #include <sys/mount.h>
72 #include <sys/conf.h>
73 #include <sys/stat.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/vmmeter.h>
77 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
78 #include <machine/md_var.h>
79 #endif
80 
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_page.h>
94 #include <vm/vnode_pager.h>
95 
96 #ifdef HWPMC_HOOKS
97 #include <sys/pmckern.h>
98 #endif
99 
100 int old_mlock = 0;
101 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
102     "Do not apply RLIMIT_MEMLOCK on mlockall");
103 static int mincore_mapped = 1;
104 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
105     "mincore reports mappings, not residency");
106 
107 #ifdef MAP_32BIT
108 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
109 #endif
110 
111 #ifndef _SYS_SYSPROTO_H_
112 struct sbrk_args {
113 	int incr;
114 };
115 #endif
116 
117 int
118 sys_sbrk(struct thread *td, struct sbrk_args *uap)
119 {
120 	/* Not yet implemented */
121 	return (EOPNOTSUPP);
122 }
123 
124 #ifndef _SYS_SYSPROTO_H_
125 struct sstk_args {
126 	int incr;
127 };
128 #endif
129 
130 int
131 sys_sstk(struct thread *td, struct sstk_args *uap)
132 {
133 	/* Not yet implemented */
134 	return (EOPNOTSUPP);
135 }
136 
137 #if defined(COMPAT_43)
138 int
139 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
140 {
141 
142 	td->td_retval[0] = PAGE_SIZE;
143 	return (0);
144 }
145 #endif				/* COMPAT_43 */
146 
147 
148 /*
149  * Memory Map (mmap) system call.  Note that the file offset
150  * and address are allowed to be NOT page aligned, though if
151  * the MAP_FIXED flag it set, both must have the same remainder
152  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
153  * page-aligned, the actual mapping starts at trunc_page(addr)
154  * and the return value is adjusted up by the page offset.
155  *
156  * Generally speaking, only character devices which are themselves
157  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
158  * there would be no cache coherency between a descriptor and a VM mapping
159  * both to the same character device.
160  */
161 #ifndef _SYS_SYSPROTO_H_
162 struct mmap_args {
163 	void *addr;
164 	size_t len;
165 	int prot;
166 	int flags;
167 	int fd;
168 	long pad;
169 	off_t pos;
170 };
171 #endif
172 
173 int
174 sys_mmap(struct thread *td, struct mmap_args *uap)
175 {
176 
177 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
178 	    uap->flags, uap->fd, uap->pos));
179 }
180 
181 int
182 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
183     int fd, off_t pos)
184 {
185 	struct vmspace *vms;
186 	struct file *fp;
187 	vm_offset_t addr;
188 	vm_size_t pageoff;
189 	vm_prot_t cap_maxprot;
190 	int align, error;
191 	cap_rights_t rights;
192 
193 	vms = td->td_proc->p_vmspace;
194 	fp = NULL;
195 	AUDIT_ARG_FD(fd);
196 	addr = addr0;
197 
198 	/*
199 	 * Ignore old flags that used to be defined but did not do anything.
200 	 */
201 	flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
202 
203 	/*
204 	 * Enforce the constraints.
205 	 * Mapping of length 0 is only allowed for old binaries.
206 	 * Anonymous mapping shall specify -1 as filedescriptor and
207 	 * zero position for new code. Be nice to ancient a.out
208 	 * binaries and correct pos for anonymous mapping, since old
209 	 * ld.so sometimes issues anonymous map requests with non-zero
210 	 * pos.
211 	 */
212 	if (!SV_CURPROC_FLAG(SV_AOUT)) {
213 		if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
214 		    ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
215 			return (EINVAL);
216 	} else {
217 		if ((flags & MAP_ANON) != 0)
218 			pos = 0;
219 	}
220 
221 	if (flags & MAP_STACK) {
222 		if ((fd != -1) ||
223 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
224 			return (EINVAL);
225 		flags |= MAP_ANON;
226 		pos = 0;
227 	}
228 	if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
229 	    MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
230 	    MAP_PREFAULT_READ | MAP_GUARD |
231 #ifdef MAP_32BIT
232 	    MAP_32BIT |
233 #endif
234 	    MAP_ALIGNMENT_MASK)) != 0)
235 		return (EINVAL);
236 	if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
237 		return (EINVAL);
238 	if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
239 		return (EINVAL);
240 	if (prot != PROT_NONE &&
241 	    (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
242 		return (EINVAL);
243 	if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
244 	    pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
245 #ifdef MAP_32BIT
246 	    MAP_32BIT |
247 #endif
248 	    MAP_ALIGNMENT_MASK)) != 0))
249 		return (EINVAL);
250 
251 	/*
252 	 * Align the file position to a page boundary,
253 	 * and save its page offset component.
254 	 */
255 	pageoff = (pos & PAGE_MASK);
256 	pos -= pageoff;
257 
258 	/* Adjust size for rounding (on both ends). */
259 	size += pageoff;			/* low end... */
260 	size = (vm_size_t) round_page(size);	/* hi end */
261 
262 	/* Ensure alignment is at least a page and fits in a pointer. */
263 	align = flags & MAP_ALIGNMENT_MASK;
264 	if (align != 0 && align != MAP_ALIGNED_SUPER &&
265 	    (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
266 	    align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
267 		return (EINVAL);
268 
269 	/*
270 	 * Check for illegal addresses.  Watch out for address wrap... Note
271 	 * that VM_*_ADDRESS are not constants due to casts (argh).
272 	 */
273 	if (flags & MAP_FIXED) {
274 		/*
275 		 * The specified address must have the same remainder
276 		 * as the file offset taken modulo PAGE_SIZE, so it
277 		 * should be aligned after adjustment by pageoff.
278 		 */
279 		addr -= pageoff;
280 		if (addr & PAGE_MASK)
281 			return (EINVAL);
282 
283 		/* Address range must be all in user VM space. */
284 		if (addr < vm_map_min(&vms->vm_map) ||
285 		    addr + size > vm_map_max(&vms->vm_map))
286 			return (EINVAL);
287 		if (addr + size < addr)
288 			return (EINVAL);
289 #ifdef MAP_32BIT
290 		if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
291 			return (EINVAL);
292 	} else if (flags & MAP_32BIT) {
293 		/*
294 		 * For MAP_32BIT, override the hint if it is too high and
295 		 * do not bother moving the mapping past the heap (since
296 		 * the heap is usually above 2GB).
297 		 */
298 		if (addr + size > MAP_32BIT_MAX_ADDR)
299 			addr = 0;
300 #endif
301 	} else {
302 		/*
303 		 * XXX for non-fixed mappings where no hint is provided or
304 		 * the hint would fall in the potential heap space,
305 		 * place it after the end of the largest possible heap.
306 		 *
307 		 * There should really be a pmap call to determine a reasonable
308 		 * location.
309 		 */
310 		if (addr == 0 ||
311 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
312 		    addr < round_page((vm_offset_t)vms->vm_daddr +
313 		    lim_max(td, RLIMIT_DATA))))
314 			addr = round_page((vm_offset_t)vms->vm_daddr +
315 			    lim_max(td, RLIMIT_DATA));
316 	}
317 	if (size == 0) {
318 		/*
319 		 * Return success without mapping anything for old
320 		 * binaries that request a page-aligned mapping of
321 		 * length 0.  For modern binaries, this function
322 		 * returns an error earlier.
323 		 */
324 		error = 0;
325 	} else if ((flags & MAP_GUARD) != 0) {
326 		error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
327 		    VM_PROT_NONE, flags, NULL, pos, FALSE, td);
328 	} else if ((flags & MAP_ANON) != 0) {
329 		/*
330 		 * Mapping blank space is trivial.
331 		 *
332 		 * This relies on VM_PROT_* matching PROT_*.
333 		 */
334 		error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
335 		    VM_PROT_ALL, flags, NULL, pos, FALSE, td);
336 	} else {
337 		/*
338 		 * Mapping file, get fp for validation and don't let the
339 		 * descriptor disappear on us if we block. Check capability
340 		 * rights, but also return the maximum rights to be combined
341 		 * with maxprot later.
342 		 */
343 		cap_rights_init(&rights, CAP_MMAP);
344 		if (prot & PROT_READ)
345 			cap_rights_set(&rights, CAP_MMAP_R);
346 		if ((flags & MAP_SHARED) != 0) {
347 			if (prot & PROT_WRITE)
348 				cap_rights_set(&rights, CAP_MMAP_W);
349 		}
350 		if (prot & PROT_EXEC)
351 			cap_rights_set(&rights, CAP_MMAP_X);
352 		error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
353 		if (error != 0)
354 			goto done;
355 		if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
356 		    td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
357 			error = EINVAL;
358 			goto done;
359 		}
360 
361 		/* This relies on VM_PROT_* matching PROT_*. */
362 		error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
363 		    cap_maxprot, flags, pos, td);
364 	}
365 
366 	if (error == 0)
367 		td->td_retval[0] = (register_t) (addr + pageoff);
368 done:
369 	if (fp)
370 		fdrop(fp, td);
371 
372 	return (error);
373 }
374 
375 #if defined(COMPAT_FREEBSD6)
376 int
377 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
378 {
379 
380 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
381 	    uap->flags, uap->fd, uap->pos));
382 }
383 #endif
384 
385 #ifdef COMPAT_43
386 #ifndef _SYS_SYSPROTO_H_
387 struct ommap_args {
388 	caddr_t addr;
389 	int len;
390 	int prot;
391 	int flags;
392 	int fd;
393 	long pos;
394 };
395 #endif
396 int
397 ommap(struct thread *td, struct ommap_args *uap)
398 {
399 	static const char cvtbsdprot[8] = {
400 		0,
401 		PROT_EXEC,
402 		PROT_WRITE,
403 		PROT_EXEC | PROT_WRITE,
404 		PROT_READ,
405 		PROT_EXEC | PROT_READ,
406 		PROT_WRITE | PROT_READ,
407 		PROT_EXEC | PROT_WRITE | PROT_READ,
408 	};
409 	int flags, prot;
410 
411 #define	OMAP_ANON	0x0002
412 #define	OMAP_COPY	0x0020
413 #define	OMAP_SHARED	0x0010
414 #define	OMAP_FIXED	0x0100
415 
416 	prot = cvtbsdprot[uap->prot & 0x7];
417 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
418 	if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
419 	    prot != 0)
420 		prot |= PROT_EXEC;
421 #endif
422 	flags = 0;
423 	if (uap->flags & OMAP_ANON)
424 		flags |= MAP_ANON;
425 	if (uap->flags & OMAP_COPY)
426 		flags |= MAP_COPY;
427 	if (uap->flags & OMAP_SHARED)
428 		flags |= MAP_SHARED;
429 	else
430 		flags |= MAP_PRIVATE;
431 	if (uap->flags & OMAP_FIXED)
432 		flags |= MAP_FIXED;
433 	return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
434 	    uap->fd, uap->pos));
435 }
436 #endif				/* COMPAT_43 */
437 
438 
439 #ifndef _SYS_SYSPROTO_H_
440 struct msync_args {
441 	void *addr;
442 	size_t len;
443 	int flags;
444 };
445 #endif
446 int
447 sys_msync(struct thread *td, struct msync_args *uap)
448 {
449 
450 	return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
451 }
452 
453 int
454 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
455 {
456 	vm_offset_t addr;
457 	vm_size_t pageoff;
458 	vm_map_t map;
459 	int rv;
460 
461 	addr = addr0;
462 	pageoff = (addr & PAGE_MASK);
463 	addr -= pageoff;
464 	size += pageoff;
465 	size = (vm_size_t) round_page(size);
466 	if (addr + size < addr)
467 		return (EINVAL);
468 
469 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
470 		return (EINVAL);
471 
472 	map = &td->td_proc->p_vmspace->vm_map;
473 
474 	/*
475 	 * Clean the pages and interpret the return value.
476 	 */
477 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
478 	    (flags & MS_INVALIDATE) != 0);
479 	switch (rv) {
480 	case KERN_SUCCESS:
481 		return (0);
482 	case KERN_INVALID_ADDRESS:
483 		return (ENOMEM);
484 	case KERN_INVALID_ARGUMENT:
485 		return (EBUSY);
486 	case KERN_FAILURE:
487 		return (EIO);
488 	default:
489 		return (EINVAL);
490 	}
491 }
492 
493 #ifndef _SYS_SYSPROTO_H_
494 struct munmap_args {
495 	void *addr;
496 	size_t len;
497 };
498 #endif
499 int
500 sys_munmap(struct thread *td, struct munmap_args *uap)
501 {
502 
503 	return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
504 }
505 
506 int
507 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
508 {
509 #ifdef HWPMC_HOOKS
510 	struct pmckern_map_out pkm;
511 	vm_map_entry_t entry;
512 	bool pmc_handled;
513 #endif
514 	vm_offset_t addr;
515 	vm_size_t pageoff;
516 	vm_map_t map;
517 
518 	if (size == 0)
519 		return (EINVAL);
520 
521 	addr = addr0;
522 	pageoff = (addr & PAGE_MASK);
523 	addr -= pageoff;
524 	size += pageoff;
525 	size = (vm_size_t) round_page(size);
526 	if (addr + size < addr)
527 		return (EINVAL);
528 
529 	/*
530 	 * Check for illegal addresses.  Watch out for address wrap...
531 	 */
532 	map = &td->td_proc->p_vmspace->vm_map;
533 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
534 		return (EINVAL);
535 	vm_map_lock(map);
536 #ifdef HWPMC_HOOKS
537 	pmc_handled = false;
538 	if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
539 		pmc_handled = true;
540 		/*
541 		 * Inform hwpmc if the address range being unmapped contains
542 		 * an executable region.
543 		 */
544 		pkm.pm_address = (uintptr_t) NULL;
545 		if (vm_map_lookup_entry(map, addr, &entry)) {
546 			for (; entry->start < addr + size;
547 			    entry = entry->next) {
548 				if (vm_map_check_protection(map, entry->start,
549 					entry->end, VM_PROT_EXECUTE) == TRUE) {
550 					pkm.pm_address = (uintptr_t) addr;
551 					pkm.pm_size = (size_t) size;
552 					break;
553 				}
554 			}
555 		}
556 	}
557 #endif
558 	vm_map_delete(map, addr, addr + size);
559 
560 #ifdef HWPMC_HOOKS
561 	if (__predict_false(pmc_handled)) {
562 		/* downgrade the lock to prevent a LOR with the pmc-sx lock */
563 		vm_map_lock_downgrade(map);
564 		if (pkm.pm_address != (uintptr_t) NULL)
565 			PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
566 		vm_map_unlock_read(map);
567 	} else
568 #endif
569 		vm_map_unlock(map);
570 
571 	/* vm_map_delete returns nothing but KERN_SUCCESS anyway */
572 	return (0);
573 }
574 
575 #ifndef _SYS_SYSPROTO_H_
576 struct mprotect_args {
577 	const void *addr;
578 	size_t len;
579 	int prot;
580 };
581 #endif
582 int
583 sys_mprotect(struct thread *td, struct mprotect_args *uap)
584 {
585 
586 	return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
587 }
588 
589 int
590 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
591 {
592 	vm_offset_t addr;
593 	vm_size_t pageoff;
594 
595 	addr = addr0;
596 	prot = (prot & VM_PROT_ALL);
597 	pageoff = (addr & PAGE_MASK);
598 	addr -= pageoff;
599 	size += pageoff;
600 	size = (vm_size_t) round_page(size);
601 #ifdef COMPAT_FREEBSD32
602 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
603 		if (((addr + size) & 0xffffffff) < addr)
604 			return (EINVAL);
605 	} else
606 #endif
607 	if (addr + size < addr)
608 		return (EINVAL);
609 
610 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
611 	    addr + size, prot, FALSE)) {
612 	case KERN_SUCCESS:
613 		return (0);
614 	case KERN_PROTECTION_FAILURE:
615 		return (EACCES);
616 	case KERN_RESOURCE_SHORTAGE:
617 		return (ENOMEM);
618 	}
619 	return (EINVAL);
620 }
621 
622 #ifndef _SYS_SYSPROTO_H_
623 struct minherit_args {
624 	void *addr;
625 	size_t len;
626 	int inherit;
627 };
628 #endif
629 int
630 sys_minherit(struct thread *td, struct minherit_args *uap)
631 {
632 	vm_offset_t addr;
633 	vm_size_t size, pageoff;
634 	vm_inherit_t inherit;
635 
636 	addr = (vm_offset_t)uap->addr;
637 	size = uap->len;
638 	inherit = uap->inherit;
639 
640 	pageoff = (addr & PAGE_MASK);
641 	addr -= pageoff;
642 	size += pageoff;
643 	size = (vm_size_t) round_page(size);
644 	if (addr + size < addr)
645 		return (EINVAL);
646 
647 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
648 	    addr + size, inherit)) {
649 	case KERN_SUCCESS:
650 		return (0);
651 	case KERN_PROTECTION_FAILURE:
652 		return (EACCES);
653 	}
654 	return (EINVAL);
655 }
656 
657 #ifndef _SYS_SYSPROTO_H_
658 struct madvise_args {
659 	void *addr;
660 	size_t len;
661 	int behav;
662 };
663 #endif
664 
665 int
666 sys_madvise(struct thread *td, struct madvise_args *uap)
667 {
668 
669 	return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
670 }
671 
672 int
673 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
674 {
675 	vm_map_t map;
676 	vm_offset_t addr, end, start;
677 	int flags;
678 
679 	/*
680 	 * Check for our special case, advising the swap pager we are
681 	 * "immortal."
682 	 */
683 	if (behav == MADV_PROTECT) {
684 		flags = PPROT_SET;
685 		return (kern_procctl(td, P_PID, td->td_proc->p_pid,
686 		    PROC_SPROTECT, &flags));
687 	}
688 
689 	/*
690 	 * Check for illegal addresses.  Watch out for address wrap... Note
691 	 * that VM_*_ADDRESS are not constants due to casts (argh).
692 	 */
693 	map = &td->td_proc->p_vmspace->vm_map;
694 	addr = addr0;
695 	if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
696 		return (EINVAL);
697 	if ((addr + len) < addr)
698 		return (EINVAL);
699 
700 	/*
701 	 * Since this routine is only advisory, we default to conservative
702 	 * behavior.
703 	 */
704 	start = trunc_page(addr);
705 	end = round_page(addr + len);
706 
707 	/*
708 	 * vm_map_madvise() checks for illegal values of behav.
709 	 */
710 	return (vm_map_madvise(map, start, end, behav));
711 }
712 
713 #ifndef _SYS_SYSPROTO_H_
714 struct mincore_args {
715 	const void *addr;
716 	size_t len;
717 	char *vec;
718 };
719 #endif
720 
721 int
722 sys_mincore(struct thread *td, struct mincore_args *uap)
723 {
724 
725 	return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
726 }
727 
728 int
729 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
730 {
731 	vm_offset_t addr, first_addr;
732 	vm_offset_t end, cend;
733 	pmap_t pmap;
734 	vm_map_t map;
735 	int error = 0;
736 	int vecindex, lastvecindex;
737 	vm_map_entry_t current;
738 	vm_map_entry_t entry;
739 	vm_object_t object;
740 	vm_paddr_t locked_pa;
741 	vm_page_t m;
742 	vm_pindex_t pindex;
743 	int mincoreinfo;
744 	unsigned int timestamp;
745 	boolean_t locked;
746 
747 	/*
748 	 * Make sure that the addresses presented are valid for user
749 	 * mode.
750 	 */
751 	first_addr = addr = trunc_page(addr0);
752 	end = addr + (vm_size_t)round_page(len);
753 	map = &td->td_proc->p_vmspace->vm_map;
754 	if (end > vm_map_max(map) || end < addr)
755 		return (ENOMEM);
756 
757 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
758 
759 	vm_map_lock_read(map);
760 RestartScan:
761 	timestamp = map->timestamp;
762 
763 	if (!vm_map_lookup_entry(map, addr, &entry)) {
764 		vm_map_unlock_read(map);
765 		return (ENOMEM);
766 	}
767 
768 	/*
769 	 * Do this on a map entry basis so that if the pages are not
770 	 * in the current processes address space, we can easily look
771 	 * up the pages elsewhere.
772 	 */
773 	lastvecindex = -1;
774 	for (current = entry; current->start < end; current = current->next) {
775 
776 		/*
777 		 * check for contiguity
778 		 */
779 		if (current->end < end && current->next->start > current->end) {
780 			vm_map_unlock_read(map);
781 			return (ENOMEM);
782 		}
783 
784 		/*
785 		 * ignore submaps (for now) or null objects
786 		 */
787 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
788 			current->object.vm_object == NULL)
789 			continue;
790 
791 		/*
792 		 * limit this scan to the current map entry and the
793 		 * limits for the mincore call
794 		 */
795 		if (addr < current->start)
796 			addr = current->start;
797 		cend = current->end;
798 		if (cend > end)
799 			cend = end;
800 
801 		/*
802 		 * scan this entry one page at a time
803 		 */
804 		while (addr < cend) {
805 			/*
806 			 * Check pmap first, it is likely faster, also
807 			 * it can provide info as to whether we are the
808 			 * one referencing or modifying the page.
809 			 */
810 			object = NULL;
811 			locked_pa = 0;
812 		retry:
813 			m = NULL;
814 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
815 			if (mincore_mapped) {
816 				/*
817 				 * We only care about this pmap's
818 				 * mapping of the page, if any.
819 				 */
820 				if (locked_pa != 0) {
821 					vm_page_unlock(PHYS_TO_VM_PAGE(
822 					    locked_pa));
823 				}
824 			} else if (locked_pa != 0) {
825 				/*
826 				 * The page is mapped by this process but not
827 				 * both accessed and modified.  It is also
828 				 * managed.  Acquire the object lock so that
829 				 * other mappings might be examined.
830 				 */
831 				m = PHYS_TO_VM_PAGE(locked_pa);
832 				if (m->object != object) {
833 					if (object != NULL)
834 						VM_OBJECT_WUNLOCK(object);
835 					object = m->object;
836 					locked = VM_OBJECT_TRYWLOCK(object);
837 					vm_page_unlock(m);
838 					if (!locked) {
839 						VM_OBJECT_WLOCK(object);
840 						vm_page_lock(m);
841 						goto retry;
842 					}
843 				} else
844 					vm_page_unlock(m);
845 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
846 				    ("mincore: page %p is mapped but invalid",
847 				    m));
848 			} else if (mincoreinfo == 0) {
849 				/*
850 				 * The page is not mapped by this process.  If
851 				 * the object implements managed pages, then
852 				 * determine if the page is resident so that
853 				 * the mappings might be examined.
854 				 */
855 				if (current->object.vm_object != object) {
856 					if (object != NULL)
857 						VM_OBJECT_WUNLOCK(object);
858 					object = current->object.vm_object;
859 					VM_OBJECT_WLOCK(object);
860 				}
861 				if (object->type == OBJT_DEFAULT ||
862 				    object->type == OBJT_SWAP ||
863 				    object->type == OBJT_VNODE) {
864 					pindex = OFF_TO_IDX(current->offset +
865 					    (addr - current->start));
866 					m = vm_page_lookup(object, pindex);
867 					if (m != NULL && m->valid == 0)
868 						m = NULL;
869 					if (m != NULL)
870 						mincoreinfo = MINCORE_INCORE;
871 				}
872 			}
873 			if (m != NULL) {
874 				/* Examine other mappings to the page. */
875 				if (m->dirty == 0 && pmap_is_modified(m))
876 					vm_page_dirty(m);
877 				if (m->dirty != 0)
878 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
879 				/*
880 				 * The first test for PGA_REFERENCED is an
881 				 * optimization.  The second test is
882 				 * required because a concurrent pmap
883 				 * operation could clear the last reference
884 				 * and set PGA_REFERENCED before the call to
885 				 * pmap_is_referenced().
886 				 */
887 				if ((m->aflags & PGA_REFERENCED) != 0 ||
888 				    pmap_is_referenced(m) ||
889 				    (m->aflags & PGA_REFERENCED) != 0)
890 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
891 			}
892 			if (object != NULL)
893 				VM_OBJECT_WUNLOCK(object);
894 
895 			/*
896 			 * subyte may page fault.  In case it needs to modify
897 			 * the map, we release the lock.
898 			 */
899 			vm_map_unlock_read(map);
900 
901 			/*
902 			 * calculate index into user supplied byte vector
903 			 */
904 			vecindex = atop(addr - first_addr);
905 
906 			/*
907 			 * If we have skipped map entries, we need to make sure that
908 			 * the byte vector is zeroed for those skipped entries.
909 			 */
910 			while ((lastvecindex + 1) < vecindex) {
911 				++lastvecindex;
912 				error = subyte(vec + lastvecindex, 0);
913 				if (error) {
914 					error = EFAULT;
915 					goto done2;
916 				}
917 			}
918 
919 			/*
920 			 * Pass the page information to the user
921 			 */
922 			error = subyte(vec + vecindex, mincoreinfo);
923 			if (error) {
924 				error = EFAULT;
925 				goto done2;
926 			}
927 
928 			/*
929 			 * If the map has changed, due to the subyte, the previous
930 			 * output may be invalid.
931 			 */
932 			vm_map_lock_read(map);
933 			if (timestamp != map->timestamp)
934 				goto RestartScan;
935 
936 			lastvecindex = vecindex;
937 			addr += PAGE_SIZE;
938 		}
939 	}
940 
941 	/*
942 	 * subyte may page fault.  In case it needs to modify
943 	 * the map, we release the lock.
944 	 */
945 	vm_map_unlock_read(map);
946 
947 	/*
948 	 * Zero the last entries in the byte vector.
949 	 */
950 	vecindex = atop(end - first_addr);
951 	while ((lastvecindex + 1) < vecindex) {
952 		++lastvecindex;
953 		error = subyte(vec + lastvecindex, 0);
954 		if (error) {
955 			error = EFAULT;
956 			goto done2;
957 		}
958 	}
959 
960 	/*
961 	 * If the map has changed, due to the subyte, the previous
962 	 * output may be invalid.
963 	 */
964 	vm_map_lock_read(map);
965 	if (timestamp != map->timestamp)
966 		goto RestartScan;
967 	vm_map_unlock_read(map);
968 done2:
969 	return (error);
970 }
971 
972 #ifndef _SYS_SYSPROTO_H_
973 struct mlock_args {
974 	const void *addr;
975 	size_t len;
976 };
977 #endif
978 int
979 sys_mlock(struct thread *td, struct mlock_args *uap)
980 {
981 
982 	return (kern_mlock(td->td_proc, td->td_ucred,
983 	    __DECONST(uintptr_t, uap->addr), uap->len));
984 }
985 
986 int
987 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
988 {
989 	vm_offset_t addr, end, last, start;
990 	vm_size_t npages, size;
991 	vm_map_t map;
992 	unsigned long nsize;
993 	int error;
994 
995 	error = priv_check_cred(cred, PRIV_VM_MLOCK);
996 	if (error)
997 		return (error);
998 	addr = addr0;
999 	size = len;
1000 	last = addr + size;
1001 	start = trunc_page(addr);
1002 	end = round_page(last);
1003 	if (last < addr || end < addr)
1004 		return (EINVAL);
1005 	npages = atop(end - start);
1006 	if (npages > vm_page_max_wired)
1007 		return (ENOMEM);
1008 	map = &proc->p_vmspace->vm_map;
1009 	PROC_LOCK(proc);
1010 	nsize = ptoa(npages + pmap_wired_count(map->pmap));
1011 	if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1012 		PROC_UNLOCK(proc);
1013 		return (ENOMEM);
1014 	}
1015 	PROC_UNLOCK(proc);
1016 	if (npages + vm_wire_count() > vm_page_max_wired)
1017 		return (EAGAIN);
1018 #ifdef RACCT
1019 	if (racct_enable) {
1020 		PROC_LOCK(proc);
1021 		error = racct_set(proc, RACCT_MEMLOCK, nsize);
1022 		PROC_UNLOCK(proc);
1023 		if (error != 0)
1024 			return (ENOMEM);
1025 	}
1026 #endif
1027 	error = vm_map_wire(map, start, end,
1028 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1029 #ifdef RACCT
1030 	if (racct_enable && error != KERN_SUCCESS) {
1031 		PROC_LOCK(proc);
1032 		racct_set(proc, RACCT_MEMLOCK,
1033 		    ptoa(pmap_wired_count(map->pmap)));
1034 		PROC_UNLOCK(proc);
1035 	}
1036 #endif
1037 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1038 }
1039 
1040 #ifndef _SYS_SYSPROTO_H_
1041 struct mlockall_args {
1042 	int	how;
1043 };
1044 #endif
1045 
1046 int
1047 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1048 {
1049 	vm_map_t map;
1050 	int error;
1051 
1052 	map = &td->td_proc->p_vmspace->vm_map;
1053 	error = priv_check(td, PRIV_VM_MLOCK);
1054 	if (error)
1055 		return (error);
1056 
1057 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1058 		return (EINVAL);
1059 
1060 	/*
1061 	 * If wiring all pages in the process would cause it to exceed
1062 	 * a hard resource limit, return ENOMEM.
1063 	 */
1064 	if (!old_mlock && uap->how & MCL_CURRENT) {
1065 		if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1066 			return (ENOMEM);
1067 	}
1068 #ifdef RACCT
1069 	if (racct_enable) {
1070 		PROC_LOCK(td->td_proc);
1071 		error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1072 		PROC_UNLOCK(td->td_proc);
1073 		if (error != 0)
1074 			return (ENOMEM);
1075 	}
1076 #endif
1077 
1078 	if (uap->how & MCL_FUTURE) {
1079 		vm_map_lock(map);
1080 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1081 		vm_map_unlock(map);
1082 		error = 0;
1083 	}
1084 
1085 	if (uap->how & MCL_CURRENT) {
1086 		/*
1087 		 * P1003.1-2001 mandates that all currently mapped pages
1088 		 * will be memory resident and locked (wired) upon return
1089 		 * from mlockall(). vm_map_wire() will wire pages, by
1090 		 * calling vm_fault_wire() for each page in the region.
1091 		 */
1092 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1093 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1094 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1095 	}
1096 #ifdef RACCT
1097 	if (racct_enable && error != KERN_SUCCESS) {
1098 		PROC_LOCK(td->td_proc);
1099 		racct_set(td->td_proc, RACCT_MEMLOCK,
1100 		    ptoa(pmap_wired_count(map->pmap)));
1101 		PROC_UNLOCK(td->td_proc);
1102 	}
1103 #endif
1104 
1105 	return (error);
1106 }
1107 
1108 #ifndef _SYS_SYSPROTO_H_
1109 struct munlockall_args {
1110 	register_t dummy;
1111 };
1112 #endif
1113 
1114 int
1115 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1116 {
1117 	vm_map_t map;
1118 	int error;
1119 
1120 	map = &td->td_proc->p_vmspace->vm_map;
1121 	error = priv_check(td, PRIV_VM_MUNLOCK);
1122 	if (error)
1123 		return (error);
1124 
1125 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1126 	vm_map_lock(map);
1127 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1128 	vm_map_unlock(map);
1129 
1130 	/* Forcibly unwire all pages. */
1131 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1132 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1133 #ifdef RACCT
1134 	if (racct_enable && error == KERN_SUCCESS) {
1135 		PROC_LOCK(td->td_proc);
1136 		racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1137 		PROC_UNLOCK(td->td_proc);
1138 	}
1139 #endif
1140 
1141 	return (error);
1142 }
1143 
1144 #ifndef _SYS_SYSPROTO_H_
1145 struct munlock_args {
1146 	const void *addr;
1147 	size_t len;
1148 };
1149 #endif
1150 int
1151 sys_munlock(struct thread *td, struct munlock_args *uap)
1152 {
1153 
1154 	return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1155 }
1156 
1157 int
1158 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1159 {
1160 	vm_offset_t addr, end, last, start;
1161 #ifdef RACCT
1162 	vm_map_t map;
1163 #endif
1164 	int error;
1165 
1166 	error = priv_check(td, PRIV_VM_MUNLOCK);
1167 	if (error)
1168 		return (error);
1169 	addr = addr0;
1170 	last = addr + size;
1171 	start = trunc_page(addr);
1172 	end = round_page(last);
1173 	if (last < addr || end < addr)
1174 		return (EINVAL);
1175 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1176 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1177 #ifdef RACCT
1178 	if (racct_enable && error == KERN_SUCCESS) {
1179 		PROC_LOCK(td->td_proc);
1180 		map = &td->td_proc->p_vmspace->vm_map;
1181 		racct_set(td->td_proc, RACCT_MEMLOCK,
1182 		    ptoa(pmap_wired_count(map->pmap)));
1183 		PROC_UNLOCK(td->td_proc);
1184 	}
1185 #endif
1186 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1187 }
1188 
1189 /*
1190  * vm_mmap_vnode()
1191  *
1192  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1193  * operations on vnodes.
1194  */
1195 int
1196 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1197     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1198     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1199     boolean_t *writecounted)
1200 {
1201 	struct vattr va;
1202 	vm_object_t obj;
1203 	vm_ooffset_t foff;
1204 	struct ucred *cred;
1205 	int error, flags, locktype;
1206 
1207 	cred = td->td_ucred;
1208 	if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1209 		locktype = LK_EXCLUSIVE;
1210 	else
1211 		locktype = LK_SHARED;
1212 	if ((error = vget(vp, locktype, td)) != 0)
1213 		return (error);
1214 	AUDIT_ARG_VNODE1(vp);
1215 	foff = *foffp;
1216 	flags = *flagsp;
1217 	obj = vp->v_object;
1218 	if (vp->v_type == VREG) {
1219 		/*
1220 		 * Get the proper underlying object
1221 		 */
1222 		if (obj == NULL) {
1223 			error = EINVAL;
1224 			goto done;
1225 		}
1226 		if (obj->type == OBJT_VNODE && obj->handle != vp) {
1227 			vput(vp);
1228 			vp = (struct vnode *)obj->handle;
1229 			/*
1230 			 * Bypass filesystems obey the mpsafety of the
1231 			 * underlying fs.  Tmpfs never bypasses.
1232 			 */
1233 			error = vget(vp, locktype, td);
1234 			if (error != 0)
1235 				return (error);
1236 		}
1237 		if (locktype == LK_EXCLUSIVE) {
1238 			*writecounted = TRUE;
1239 			vnode_pager_update_writecount(obj, 0, objsize);
1240 		}
1241 	} else {
1242 		error = EINVAL;
1243 		goto done;
1244 	}
1245 	if ((error = VOP_GETATTR(vp, &va, cred)))
1246 		goto done;
1247 #ifdef MAC
1248 	/* This relies on VM_PROT_* matching PROT_*. */
1249 	error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1250 	if (error != 0)
1251 		goto done;
1252 #endif
1253 	if ((flags & MAP_SHARED) != 0) {
1254 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1255 			if (prot & VM_PROT_WRITE) {
1256 				error = EPERM;
1257 				goto done;
1258 			}
1259 			*maxprotp &= ~VM_PROT_WRITE;
1260 		}
1261 	}
1262 	/*
1263 	 * If it is a regular file without any references
1264 	 * we do not need to sync it.
1265 	 * Adjust object size to be the size of actual file.
1266 	 */
1267 	objsize = round_page(va.va_size);
1268 	if (va.va_nlink == 0)
1269 		flags |= MAP_NOSYNC;
1270 	if (obj->type == OBJT_VNODE) {
1271 		obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1272 		    cred);
1273 		if (obj == NULL) {
1274 			error = ENOMEM;
1275 			goto done;
1276 		}
1277 	} else {
1278 		KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1279 		    ("wrong object type"));
1280 		VM_OBJECT_WLOCK(obj);
1281 		vm_object_reference_locked(obj);
1282 #if VM_NRESERVLEVEL > 0
1283 		vm_object_color(obj, 0);
1284 #endif
1285 		VM_OBJECT_WUNLOCK(obj);
1286 	}
1287 	*objp = obj;
1288 	*flagsp = flags;
1289 
1290 	vfs_mark_atime(vp, cred);
1291 
1292 done:
1293 	if (error != 0 && *writecounted) {
1294 		*writecounted = FALSE;
1295 		vnode_pager_update_writecount(obj, objsize, 0);
1296 	}
1297 	vput(vp);
1298 	return (error);
1299 }
1300 
1301 /*
1302  * vm_mmap_cdev()
1303  *
1304  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1305  * operations on cdevs.
1306  */
1307 int
1308 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1309     vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1310     vm_ooffset_t *foff, vm_object_t *objp)
1311 {
1312 	vm_object_t obj;
1313 	int error, flags;
1314 
1315 	flags = *flagsp;
1316 
1317 	if (dsw->d_flags & D_MMAP_ANON) {
1318 		*objp = NULL;
1319 		*foff = 0;
1320 		*maxprotp = VM_PROT_ALL;
1321 		*flagsp |= MAP_ANON;
1322 		return (0);
1323 	}
1324 	/*
1325 	 * cdevs do not provide private mappings of any kind.
1326 	 */
1327 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1328 	    (prot & VM_PROT_WRITE) != 0)
1329 		return (EACCES);
1330 	if (flags & (MAP_PRIVATE|MAP_COPY))
1331 		return (EINVAL);
1332 	/*
1333 	 * Force device mappings to be shared.
1334 	 */
1335 	flags |= MAP_SHARED;
1336 #ifdef MAC_XXX
1337 	error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1338 	if (error != 0)
1339 		return (error);
1340 #endif
1341 	/*
1342 	 * First, try d_mmap_single().  If that is not implemented
1343 	 * (returns ENODEV), fall back to using the device pager.
1344 	 * Note that d_mmap_single() must return a reference to the
1345 	 * object (it needs to bump the reference count of the object
1346 	 * it returns somehow).
1347 	 *
1348 	 * XXX assumes VM_PROT_* == PROT_*
1349 	 */
1350 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1351 	if (error != ENODEV)
1352 		return (error);
1353 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1354 	    td->td_ucred);
1355 	if (obj == NULL)
1356 		return (EINVAL);
1357 	*objp = obj;
1358 	*flagsp = flags;
1359 	return (0);
1360 }
1361 
1362 /*
1363  * vm_mmap()
1364  *
1365  * Internal version of mmap used by exec, sys5 shared memory, and
1366  * various device drivers.  Handle is either a vnode pointer, a
1367  * character device, or NULL for MAP_ANON.
1368  */
1369 int
1370 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1371 	vm_prot_t maxprot, int flags,
1372 	objtype_t handle_type, void *handle,
1373 	vm_ooffset_t foff)
1374 {
1375 	vm_object_t object;
1376 	struct thread *td = curthread;
1377 	int error;
1378 	boolean_t writecounted;
1379 
1380 	if (size == 0)
1381 		return (EINVAL);
1382 
1383 	size = round_page(size);
1384 	object = NULL;
1385 	writecounted = FALSE;
1386 
1387 	/*
1388 	 * Lookup/allocate object.
1389 	 */
1390 	switch (handle_type) {
1391 	case OBJT_DEVICE: {
1392 		struct cdevsw *dsw;
1393 		struct cdev *cdev;
1394 		int ref;
1395 
1396 		cdev = handle;
1397 		dsw = dev_refthread(cdev, &ref);
1398 		if (dsw == NULL)
1399 			return (ENXIO);
1400 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1401 		    dsw, &foff, &object);
1402 		dev_relthread(cdev, ref);
1403 		break;
1404 	}
1405 	case OBJT_VNODE:
1406 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1407 		    handle, &foff, &object, &writecounted);
1408 		break;
1409 	case OBJT_DEFAULT:
1410 		if (handle == NULL) {
1411 			error = 0;
1412 			break;
1413 		}
1414 		/* FALLTHROUGH */
1415 	default:
1416 		error = EINVAL;
1417 		break;
1418 	}
1419 	if (error)
1420 		return (error);
1421 
1422 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1423 	    foff, writecounted, td);
1424 	if (error != 0 && object != NULL) {
1425 		/*
1426 		 * If this mapping was accounted for in the vnode's
1427 		 * writecount, then undo that now.
1428 		 */
1429 		if (writecounted)
1430 			vnode_pager_release_writecount(object, 0, size);
1431 		vm_object_deallocate(object);
1432 	}
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Internal version of mmap that maps a specific VM object into an
1438  * map.  Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1439  */
1440 int
1441 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1442     vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1443     boolean_t writecounted, struct thread *td)
1444 {
1445 	boolean_t curmap, fitit;
1446 	vm_offset_t max_addr;
1447 	int docow, error, findspace, rv;
1448 
1449 	curmap = map == &td->td_proc->p_vmspace->vm_map;
1450 	if (curmap) {
1451 		RACCT_PROC_LOCK(td->td_proc);
1452 		if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1453 			RACCT_PROC_UNLOCK(td->td_proc);
1454 			return (ENOMEM);
1455 		}
1456 		if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1457 			RACCT_PROC_UNLOCK(td->td_proc);
1458 			return (ENOMEM);
1459 		}
1460 		if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1461 			if (ptoa(pmap_wired_count(map->pmap)) + size >
1462 			    lim_cur(td, RLIMIT_MEMLOCK)) {
1463 				racct_set_force(td->td_proc, RACCT_VMEM,
1464 				    map->size);
1465 				RACCT_PROC_UNLOCK(td->td_proc);
1466 				return (ENOMEM);
1467 			}
1468 			error = racct_set(td->td_proc, RACCT_MEMLOCK,
1469 			    ptoa(pmap_wired_count(map->pmap)) + size);
1470 			if (error != 0) {
1471 				racct_set_force(td->td_proc, RACCT_VMEM,
1472 				    map->size);
1473 				RACCT_PROC_UNLOCK(td->td_proc);
1474 				return (error);
1475 			}
1476 		}
1477 		RACCT_PROC_UNLOCK(td->td_proc);
1478 	}
1479 
1480 	/*
1481 	 * We currently can only deal with page aligned file offsets.
1482 	 * The mmap() system call already enforces this by subtracting
1483 	 * the page offset from the file offset, but checking here
1484 	 * catches errors in device drivers (e.g. d_single_mmap()
1485 	 * callbacks) and other internal mapping requests (such as in
1486 	 * exec).
1487 	 */
1488 	if (foff & PAGE_MASK)
1489 		return (EINVAL);
1490 
1491 	if ((flags & MAP_FIXED) == 0) {
1492 		fitit = TRUE;
1493 		*addr = round_page(*addr);
1494 	} else {
1495 		if (*addr != trunc_page(*addr))
1496 			return (EINVAL);
1497 		fitit = FALSE;
1498 	}
1499 
1500 	if (flags & MAP_ANON) {
1501 		if (object != NULL || foff != 0)
1502 			return (EINVAL);
1503 		docow = 0;
1504 	} else if (flags & MAP_PREFAULT_READ)
1505 		docow = MAP_PREFAULT;
1506 	else
1507 		docow = MAP_PREFAULT_PARTIAL;
1508 
1509 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1510 		docow |= MAP_COPY_ON_WRITE;
1511 	if (flags & MAP_NOSYNC)
1512 		docow |= MAP_DISABLE_SYNCER;
1513 	if (flags & MAP_NOCORE)
1514 		docow |= MAP_DISABLE_COREDUMP;
1515 	/* Shared memory is also shared with children. */
1516 	if (flags & MAP_SHARED)
1517 		docow |= MAP_INHERIT_SHARE;
1518 	if (writecounted)
1519 		docow |= MAP_VN_WRITECOUNT;
1520 	if (flags & MAP_STACK) {
1521 		if (object != NULL)
1522 			return (EINVAL);
1523 		docow |= MAP_STACK_GROWS_DOWN;
1524 	}
1525 	if ((flags & MAP_EXCL) != 0)
1526 		docow |= MAP_CHECK_EXCL;
1527 	if ((flags & MAP_GUARD) != 0)
1528 		docow |= MAP_CREATE_GUARD;
1529 
1530 	if (fitit) {
1531 		if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1532 			findspace = VMFS_SUPER_SPACE;
1533 		else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1534 			findspace = VMFS_ALIGNED_SPACE(flags >>
1535 			    MAP_ALIGNMENT_SHIFT);
1536 		else
1537 			findspace = VMFS_OPTIMAL_SPACE;
1538 		max_addr = 0;
1539 #ifdef MAP_32BIT
1540 		if ((flags & MAP_32BIT) != 0)
1541 			max_addr = MAP_32BIT_MAX_ADDR;
1542 #endif
1543 		if (curmap) {
1544 			rv = vm_map_find_min(map, object, foff, addr, size,
1545 			    round_page((vm_offset_t)td->td_proc->p_vmspace->
1546 			    vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1547 			    findspace, prot, maxprot, docow);
1548 		} else {
1549 			rv = vm_map_find(map, object, foff, addr, size,
1550 			    max_addr, findspace, prot, maxprot, docow);
1551 		}
1552 	} else {
1553 		rv = vm_map_fixed(map, object, foff, *addr, size,
1554 		    prot, maxprot, docow);
1555 	}
1556 
1557 	if (rv == KERN_SUCCESS) {
1558 		/*
1559 		 * If the process has requested that all future mappings
1560 		 * be wired, then heed this.
1561 		 */
1562 		if (map->flags & MAP_WIREFUTURE) {
1563 			vm_map_wire(map, *addr, *addr + size,
1564 			    VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1565 			    VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1566 		}
1567 	}
1568 	return (vm_mmap_to_errno(rv));
1569 }
1570 
1571 /*
1572  * Translate a Mach VM return code to zero on success or the appropriate errno
1573  * on failure.
1574  */
1575 int
1576 vm_mmap_to_errno(int rv)
1577 {
1578 
1579 	switch (rv) {
1580 	case KERN_SUCCESS:
1581 		return (0);
1582 	case KERN_INVALID_ADDRESS:
1583 	case KERN_NO_SPACE:
1584 		return (ENOMEM);
1585 	case KERN_PROTECTION_FAILURE:
1586 		return (EACCES);
1587 	default:
1588 		return (EINVAL);
1589 	}
1590 }
1591