xref: /freebsd/sys/vm/vm_mmap.c (revision 721351876cd4d3a8a700f62d2061331fa951a488)
1 /*-
2  * Copyright (c) 1988 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  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.
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35  *
36  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
37  */
38 
39 /*
40  * Mapped file (mmap) interface to VM
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 #include "opt_mac.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/resource.h>
60 #include <sys/resourcevar.h>
61 #include <sys/vnode.h>
62 #include <sys/fcntl.h>
63 #include <sys/file.h>
64 #include <sys/mman.h>
65 #include <sys/mount.h>
66 #include <sys/conf.h>
67 #include <sys/stat.h>
68 #include <sys/vmmeter.h>
69 #include <sys/sysctl.h>
70 
71 #include <security/mac/mac_framework.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_pageout.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_kern.h>
84 
85 #ifdef HWPMC_HOOKS
86 #include <sys/pmckern.h>
87 #endif
88 
89 #ifndef _SYS_SYSPROTO_H_
90 struct sbrk_args {
91 	int incr;
92 };
93 #endif
94 
95 static int max_proc_mmap;
96 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
97 
98 /*
99  * Set the maximum number of vm_map_entry structures per process.  Roughly
100  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
101  * of our KVM malloc space still results in generous limits.  We want a
102  * default that is good enough to prevent the kernel running out of resources
103  * if attacked from compromised user account but generous enough such that
104  * multi-threaded processes are not unduly inconvenienced.
105  */
106 static void vmmapentry_rsrc_init(void *);
107 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init,
108     NULL);
109 
110 static void
111 vmmapentry_rsrc_init(dummy)
112         void *dummy;
113 {
114     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
115     max_proc_mmap /= 100;
116 }
117 
118 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
119     int *, struct vnode *, vm_ooffset_t, vm_object_t *);
120 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
121     int *, struct cdev *, vm_ooffset_t, vm_object_t *);
122 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
123     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
124 
125 /*
126  * MPSAFE
127  */
128 /* ARGSUSED */
129 int
130 sbrk(td, uap)
131 	struct thread *td;
132 	struct sbrk_args *uap;
133 {
134 	/* Not yet implemented */
135 	return (EOPNOTSUPP);
136 }
137 
138 #ifndef _SYS_SYSPROTO_H_
139 struct sstk_args {
140 	int incr;
141 };
142 #endif
143 
144 /*
145  * MPSAFE
146  */
147 /* ARGSUSED */
148 int
149 sstk(td, uap)
150 	struct thread *td;
151 	struct sstk_args *uap;
152 {
153 	/* Not yet implemented */
154 	return (EOPNOTSUPP);
155 }
156 
157 #if defined(COMPAT_43)
158 #ifndef _SYS_SYSPROTO_H_
159 struct getpagesize_args {
160 	int dummy;
161 };
162 #endif
163 
164 /* ARGSUSED */
165 int
166 ogetpagesize(td, uap)
167 	struct thread *td;
168 	struct getpagesize_args *uap;
169 {
170 	/* MP SAFE */
171 	td->td_retval[0] = PAGE_SIZE;
172 	return (0);
173 }
174 #endif				/* COMPAT_43 */
175 
176 
177 /*
178  * Memory Map (mmap) system call.  Note that the file offset
179  * and address are allowed to be NOT page aligned, though if
180  * the MAP_FIXED flag it set, both must have the same remainder
181  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
182  * page-aligned, the actual mapping starts at trunc_page(addr)
183  * and the return value is adjusted up by the page offset.
184  *
185  * Generally speaking, only character devices which are themselves
186  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
187  * there would be no cache coherency between a descriptor and a VM mapping
188  * both to the same character device.
189  *
190  * Block devices can be mmap'd no matter what they represent.  Cache coherency
191  * is maintained as long as you do not write directly to the underlying
192  * character device.
193  */
194 #ifndef _SYS_SYSPROTO_H_
195 struct mmap_args {
196 	void *addr;
197 	size_t len;
198 	int prot;
199 	int flags;
200 	int fd;
201 	long pad;
202 	off_t pos;
203 };
204 #endif
205 
206 /*
207  * MPSAFE
208  */
209 int
210 mmap(td, uap)
211 	struct thread *td;
212 	struct mmap_args *uap;
213 {
214 #ifdef HWPMC_HOOKS
215 	struct pmckern_map_in pkm;
216 #endif
217 	struct file *fp;
218 	struct vnode *vp;
219 	vm_offset_t addr;
220 	vm_size_t size, pageoff;
221 	vm_prot_t prot, maxprot;
222 	void *handle;
223 	objtype_t handle_type;
224 	int flags, error;
225 	off_t pos;
226 	struct vmspace *vms = td->td_proc->p_vmspace;
227 
228 	addr = (vm_offset_t) uap->addr;
229 	size = uap->len;
230 	prot = uap->prot & VM_PROT_ALL;
231 	flags = uap->flags;
232 	pos = uap->pos;
233 
234 	fp = NULL;
235 	/* make sure mapping fits into numeric range etc */
236 	if ((ssize_t) uap->len < 0 ||
237 	    ((flags & MAP_ANON) && uap->fd != -1))
238 		return (EINVAL);
239 
240 	if (flags & MAP_STACK) {
241 		if ((uap->fd != -1) ||
242 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
243 			return (EINVAL);
244 		flags |= MAP_ANON;
245 		pos = 0;
246 	}
247 
248 	/*
249 	 * Align the file position to a page boundary,
250 	 * and save its page offset component.
251 	 */
252 	pageoff = (pos & PAGE_MASK);
253 	pos -= pageoff;
254 
255 	/* Adjust size for rounding (on both ends). */
256 	size += pageoff;			/* low end... */
257 	size = (vm_size_t) round_page(size);	/* hi end */
258 
259 	/*
260 	 * Check for illegal addresses.  Watch out for address wrap... Note
261 	 * that VM_*_ADDRESS are not constants due to casts (argh).
262 	 */
263 	if (flags & MAP_FIXED) {
264 		/*
265 		 * The specified address must have the same remainder
266 		 * as the file offset taken modulo PAGE_SIZE, so it
267 		 * should be aligned after adjustment by pageoff.
268 		 */
269 		addr -= pageoff;
270 		if (addr & PAGE_MASK)
271 			return (EINVAL);
272 		/* Address range must be all in user VM space. */
273 		if (addr < vm_map_min(&vms->vm_map) ||
274 		    addr + size > vm_map_max(&vms->vm_map))
275 			return (EINVAL);
276 		if (addr + size < addr)
277 			return (EINVAL);
278 	} else {
279 	/*
280 	 * XXX for non-fixed mappings where no hint is provided or
281 	 * the hint would fall in the potential heap space,
282 	 * place it after the end of the largest possible heap.
283 	 *
284 	 * There should really be a pmap call to determine a reasonable
285 	 * location.
286 	 */
287 		PROC_LOCK(td->td_proc);
288 		if (addr == 0 ||
289 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
290 		    addr < round_page((vm_offset_t)vms->vm_daddr +
291 		    lim_max(td->td_proc, RLIMIT_DATA))))
292 			addr = round_page((vm_offset_t)vms->vm_daddr +
293 			    lim_max(td->td_proc, RLIMIT_DATA));
294 		PROC_UNLOCK(td->td_proc);
295 	}
296 	if (flags & MAP_ANON) {
297 		/*
298 		 * Mapping blank space is trivial.
299 		 */
300 		handle = NULL;
301 		handle_type = OBJT_DEFAULT;
302 		maxprot = VM_PROT_ALL;
303 		pos = 0;
304 	} else {
305 		/*
306 		 * Mapping file, get fp for validation and
307 		 * don't let the descriptor disappear on us if we block.
308 		 */
309 		if ((error = fget(td, uap->fd, &fp)) != 0)
310 			goto done;
311 		if (fp->f_type == DTYPE_SHM) {
312 			handle = fp->f_data;
313 			handle_type = OBJT_SWAP;
314 			maxprot = VM_PROT_NONE;
315 
316 			/* FREAD should always be set. */
317 			if (fp->f_flag & FREAD)
318 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
319 			if (fp->f_flag & FWRITE)
320 				maxprot |= VM_PROT_WRITE;
321 			goto map;
322 		}
323 		if (fp->f_type != DTYPE_VNODE) {
324 			error = ENODEV;
325 			goto done;
326 		}
327 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
328     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
329 		/*
330 		 * POSIX shared-memory objects are defined to have
331 		 * kernel persistence, and are not defined to support
332 		 * read(2)/write(2) -- or even open(2).  Thus, we can
333 		 * use MAP_ASYNC to trade on-disk coherence for speed.
334 		 * The shm_open(3) library routine turns on the FPOSIXSHM
335 		 * flag to request this behavior.
336 		 */
337 		if (fp->f_flag & FPOSIXSHM)
338 			flags |= MAP_NOSYNC;
339 #endif
340 		vp = fp->f_vnode;
341 		/*
342 		 * Ensure that file and memory protections are
343 		 * compatible.  Note that we only worry about
344 		 * writability if mapping is shared; in this case,
345 		 * current and max prot are dictated by the open file.
346 		 * XXX use the vnode instead?  Problem is: what
347 		 * credentials do we use for determination? What if
348 		 * proc does a setuid?
349 		 */
350 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
351 			maxprot = VM_PROT_NONE;
352 		else
353 			maxprot = VM_PROT_EXECUTE;
354 		if (fp->f_flag & FREAD) {
355 			maxprot |= VM_PROT_READ;
356 		} else if (prot & PROT_READ) {
357 			error = EACCES;
358 			goto done;
359 		}
360 		/*
361 		 * If we are sharing potential changes (either via
362 		 * MAP_SHARED or via the implicit sharing of character
363 		 * device mappings), and we are trying to get write
364 		 * permission although we opened it without asking
365 		 * for it, bail out.
366 		 */
367 		if ((flags & MAP_SHARED) != 0) {
368 			if ((fp->f_flag & FWRITE) != 0) {
369 				maxprot |= VM_PROT_WRITE;
370 			} else if ((prot & PROT_WRITE) != 0) {
371 				error = EACCES;
372 				goto done;
373 			}
374 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
375 			maxprot |= VM_PROT_WRITE;
376 		}
377 		handle = (void *)vp;
378 		handle_type = OBJT_VNODE;
379 	}
380 map:
381 
382 	/*
383 	 * Do not allow more then a certain number of vm_map_entry structures
384 	 * per process.  Scale with the number of rforks sharing the map
385 	 * to make the limit reasonable for threads.
386 	 */
387 	if (max_proc_mmap &&
388 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
389 		error = ENOMEM;
390 		goto done;
391 	}
392 
393 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
394 	    flags, handle_type, handle, pos);
395 #ifdef HWPMC_HOOKS
396 	/* inform hwpmc(4) if an executable is being mapped */
397 	if (error == 0 && handle_type == OBJT_VNODE &&
398 	    (prot & PROT_EXEC)) {
399 		pkm.pm_file = handle;
400 		pkm.pm_address = (uintptr_t) addr;
401 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
402 	}
403 #endif
404 	if (error == 0)
405 		td->td_retval[0] = (register_t) (addr + pageoff);
406 done:
407 	if (fp)
408 		fdrop(fp, td);
409 
410 	return (error);
411 }
412 
413 int
414 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
415 {
416 	struct mmap_args oargs;
417 
418 	oargs.addr = uap->addr;
419 	oargs.len = uap->len;
420 	oargs.prot = uap->prot;
421 	oargs.flags = uap->flags;
422 	oargs.fd = uap->fd;
423 	oargs.pos = uap->pos;
424 	return (mmap(td, &oargs));
425 }
426 
427 #ifdef COMPAT_43
428 #ifndef _SYS_SYSPROTO_H_
429 struct ommap_args {
430 	caddr_t addr;
431 	int len;
432 	int prot;
433 	int flags;
434 	int fd;
435 	long pos;
436 };
437 #endif
438 int
439 ommap(td, uap)
440 	struct thread *td;
441 	struct ommap_args *uap;
442 {
443 	struct mmap_args nargs;
444 	static const char cvtbsdprot[8] = {
445 		0,
446 		PROT_EXEC,
447 		PROT_WRITE,
448 		PROT_EXEC | PROT_WRITE,
449 		PROT_READ,
450 		PROT_EXEC | PROT_READ,
451 		PROT_WRITE | PROT_READ,
452 		PROT_EXEC | PROT_WRITE | PROT_READ,
453 	};
454 
455 #define	OMAP_ANON	0x0002
456 #define	OMAP_COPY	0x0020
457 #define	OMAP_SHARED	0x0010
458 #define	OMAP_FIXED	0x0100
459 
460 	nargs.addr = uap->addr;
461 	nargs.len = uap->len;
462 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
463 	nargs.flags = 0;
464 	if (uap->flags & OMAP_ANON)
465 		nargs.flags |= MAP_ANON;
466 	if (uap->flags & OMAP_COPY)
467 		nargs.flags |= MAP_COPY;
468 	if (uap->flags & OMAP_SHARED)
469 		nargs.flags |= MAP_SHARED;
470 	else
471 		nargs.flags |= MAP_PRIVATE;
472 	if (uap->flags & OMAP_FIXED)
473 		nargs.flags |= MAP_FIXED;
474 	nargs.fd = uap->fd;
475 	nargs.pos = uap->pos;
476 	return (mmap(td, &nargs));
477 }
478 #endif				/* COMPAT_43 */
479 
480 
481 #ifndef _SYS_SYSPROTO_H_
482 struct msync_args {
483 	void *addr;
484 	size_t len;
485 	int flags;
486 };
487 #endif
488 /*
489  * MPSAFE
490  */
491 int
492 msync(td, uap)
493 	struct thread *td;
494 	struct msync_args *uap;
495 {
496 	vm_offset_t addr;
497 	vm_size_t size, pageoff;
498 	int flags;
499 	vm_map_t map;
500 	int rv;
501 
502 	addr = (vm_offset_t) uap->addr;
503 	size = uap->len;
504 	flags = uap->flags;
505 
506 	pageoff = (addr & PAGE_MASK);
507 	addr -= pageoff;
508 	size += pageoff;
509 	size = (vm_size_t) round_page(size);
510 	if (addr + size < addr)
511 		return (EINVAL);
512 
513 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
514 		return (EINVAL);
515 
516 	map = &td->td_proc->p_vmspace->vm_map;
517 
518 	/*
519 	 * Clean the pages and interpret the return value.
520 	 */
521 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
522 	    (flags & MS_INVALIDATE) != 0);
523 	switch (rv) {
524 	case KERN_SUCCESS:
525 		return (0);
526 	case KERN_INVALID_ADDRESS:
527 		return (EINVAL);	/* Sun returns ENOMEM? */
528 	case KERN_INVALID_ARGUMENT:
529 		return (EBUSY);
530 	default:
531 		return (EINVAL);
532 	}
533 }
534 
535 #ifndef _SYS_SYSPROTO_H_
536 struct munmap_args {
537 	void *addr;
538 	size_t len;
539 };
540 #endif
541 /*
542  * MPSAFE
543  */
544 int
545 munmap(td, uap)
546 	struct thread *td;
547 	struct munmap_args *uap;
548 {
549 #ifdef HWPMC_HOOKS
550 	struct pmckern_map_out pkm;
551 	vm_map_entry_t entry;
552 #endif
553 	vm_offset_t addr;
554 	vm_size_t size, pageoff;
555 	vm_map_t map;
556 
557 	addr = (vm_offset_t) uap->addr;
558 	size = uap->len;
559 	if (size == 0)
560 		return (EINVAL);
561 
562 	pageoff = (addr & PAGE_MASK);
563 	addr -= pageoff;
564 	size += pageoff;
565 	size = (vm_size_t) round_page(size);
566 	if (addr + size < addr)
567 		return (EINVAL);
568 
569 	/*
570 	 * Check for illegal addresses.  Watch out for address wrap...
571 	 */
572 	map = &td->td_proc->p_vmspace->vm_map;
573 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
574 		return (EINVAL);
575 	vm_map_lock(map);
576 #ifdef HWPMC_HOOKS
577 	/*
578 	 * Inform hwpmc if the address range being unmapped contains
579 	 * an executable region.
580 	 */
581 	if (vm_map_lookup_entry(map, addr, &entry)) {
582 		for (;
583 		     entry != &map->header && entry->start < addr + size;
584 		     entry = entry->next) {
585 			if (vm_map_check_protection(map, entry->start,
586 				entry->end, VM_PROT_EXECUTE) == TRUE) {
587 				pkm.pm_address = (uintptr_t) addr;
588 				pkm.pm_size = (size_t) size;
589 				PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
590 				    (void *) &pkm);
591 				break;
592 			}
593 		}
594 	}
595 #endif
596 	/* returns nothing but KERN_SUCCESS anyway */
597 	vm_map_delete(map, addr, addr + size);
598 	vm_map_unlock(map);
599 	return (0);
600 }
601 
602 #ifndef _SYS_SYSPROTO_H_
603 struct mprotect_args {
604 	const void *addr;
605 	size_t len;
606 	int prot;
607 };
608 #endif
609 /*
610  * MPSAFE
611  */
612 int
613 mprotect(td, uap)
614 	struct thread *td;
615 	struct mprotect_args *uap;
616 {
617 	vm_offset_t addr;
618 	vm_size_t size, pageoff;
619 	vm_prot_t prot;
620 
621 	addr = (vm_offset_t) uap->addr;
622 	size = uap->len;
623 	prot = uap->prot & VM_PROT_ALL;
624 #if defined(VM_PROT_READ_IS_EXEC)
625 	if (prot & VM_PROT_READ)
626 		prot |= VM_PROT_EXECUTE;
627 #endif
628 
629 	pageoff = (addr & PAGE_MASK);
630 	addr -= pageoff;
631 	size += pageoff;
632 	size = (vm_size_t) round_page(size);
633 	if (addr + size < addr)
634 		return (EINVAL);
635 
636 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
637 	    addr + size, prot, FALSE)) {
638 	case KERN_SUCCESS:
639 		return (0);
640 	case KERN_PROTECTION_FAILURE:
641 		return (EACCES);
642 	}
643 	return (EINVAL);
644 }
645 
646 #ifndef _SYS_SYSPROTO_H_
647 struct minherit_args {
648 	void *addr;
649 	size_t len;
650 	int inherit;
651 };
652 #endif
653 /*
654  * MPSAFE
655  */
656 int
657 minherit(td, uap)
658 	struct thread *td;
659 	struct minherit_args *uap;
660 {
661 	vm_offset_t addr;
662 	vm_size_t size, pageoff;
663 	vm_inherit_t inherit;
664 
665 	addr = (vm_offset_t)uap->addr;
666 	size = uap->len;
667 	inherit = uap->inherit;
668 
669 	pageoff = (addr & PAGE_MASK);
670 	addr -= pageoff;
671 	size += pageoff;
672 	size = (vm_size_t) round_page(size);
673 	if (addr + size < addr)
674 		return (EINVAL);
675 
676 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
677 	    addr + size, inherit)) {
678 	case KERN_SUCCESS:
679 		return (0);
680 	case KERN_PROTECTION_FAILURE:
681 		return (EACCES);
682 	}
683 	return (EINVAL);
684 }
685 
686 #ifndef _SYS_SYSPROTO_H_
687 struct madvise_args {
688 	void *addr;
689 	size_t len;
690 	int behav;
691 };
692 #endif
693 
694 /*
695  * MPSAFE
696  */
697 /* ARGSUSED */
698 int
699 madvise(td, uap)
700 	struct thread *td;
701 	struct madvise_args *uap;
702 {
703 	vm_offset_t start, end;
704 	vm_map_t map;
705 	struct proc *p;
706 	int error;
707 
708 	/*
709 	 * Check for our special case, advising the swap pager we are
710 	 * "immortal."
711 	 */
712 	if (uap->behav == MADV_PROTECT) {
713 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
714 		if (error == 0) {
715 			p = td->td_proc;
716 			PROC_LOCK(p);
717 			p->p_flag |= P_PROTECTED;
718 			PROC_UNLOCK(p);
719 		}
720 		return (error);
721 	}
722 	/*
723 	 * Check for illegal behavior
724 	 */
725 	if (uap->behav < 0 || uap->behav > MADV_CORE)
726 		return (EINVAL);
727 	/*
728 	 * Check for illegal addresses.  Watch out for address wrap... Note
729 	 * that VM_*_ADDRESS are not constants due to casts (argh).
730 	 */
731 	map = &td->td_proc->p_vmspace->vm_map;
732 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
733 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
734 		return (EINVAL);
735 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
736 		return (EINVAL);
737 
738 	/*
739 	 * Since this routine is only advisory, we default to conservative
740 	 * behavior.
741 	 */
742 	start = trunc_page((vm_offset_t) uap->addr);
743 	end = round_page((vm_offset_t) uap->addr + uap->len);
744 
745 	if (vm_map_madvise(map, start, end, uap->behav))
746 		return (EINVAL);
747 	return (0);
748 }
749 
750 #ifndef _SYS_SYSPROTO_H_
751 struct mincore_args {
752 	const void *addr;
753 	size_t len;
754 	char *vec;
755 };
756 #endif
757 
758 /*
759  * MPSAFE
760  */
761 /* ARGSUSED */
762 int
763 mincore(td, uap)
764 	struct thread *td;
765 	struct mincore_args *uap;
766 {
767 	vm_offset_t addr, first_addr;
768 	vm_offset_t end, cend;
769 	pmap_t pmap;
770 	vm_map_t map;
771 	char *vec;
772 	int error = 0;
773 	int vecindex, lastvecindex;
774 	vm_map_entry_t current;
775 	vm_map_entry_t entry;
776 	int mincoreinfo;
777 	unsigned int timestamp;
778 
779 	/*
780 	 * Make sure that the addresses presented are valid for user
781 	 * mode.
782 	 */
783 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
784 	end = addr + (vm_size_t)round_page(uap->len);
785 	map = &td->td_proc->p_vmspace->vm_map;
786 	if (end > vm_map_max(map) || end < addr)
787 		return (ENOMEM);
788 
789 	/*
790 	 * Address of byte vector
791 	 */
792 	vec = uap->vec;
793 
794 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
795 
796 	vm_map_lock_read(map);
797 RestartScan:
798 	timestamp = map->timestamp;
799 
800 	if (!vm_map_lookup_entry(map, addr, &entry)) {
801 		vm_map_unlock_read(map);
802 		return (ENOMEM);
803 	}
804 
805 	/*
806 	 * Do this on a map entry basis so that if the pages are not
807 	 * in the current processes address space, we can easily look
808 	 * up the pages elsewhere.
809 	 */
810 	lastvecindex = -1;
811 	for (current = entry;
812 	    (current != &map->header) && (current->start < end);
813 	    current = current->next) {
814 
815 		/*
816 		 * check for contiguity
817 		 */
818 		if (current->end < end &&
819 		    (entry->next == &map->header ||
820 		     current->next->start > current->end)) {
821 			vm_map_unlock_read(map);
822 			return (ENOMEM);
823 		}
824 
825 		/*
826 		 * ignore submaps (for now) or null objects
827 		 */
828 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
829 			current->object.vm_object == NULL)
830 			continue;
831 
832 		/*
833 		 * limit this scan to the current map entry and the
834 		 * limits for the mincore call
835 		 */
836 		if (addr < current->start)
837 			addr = current->start;
838 		cend = current->end;
839 		if (cend > end)
840 			cend = end;
841 
842 		/*
843 		 * scan this entry one page at a time
844 		 */
845 		while (addr < cend) {
846 			/*
847 			 * Check pmap first, it is likely faster, also
848 			 * it can provide info as to whether we are the
849 			 * one referencing or modifying the page.
850 			 */
851 			mincoreinfo = pmap_mincore(pmap, addr);
852 			if (!mincoreinfo) {
853 				vm_pindex_t pindex;
854 				vm_ooffset_t offset;
855 				vm_page_t m;
856 				/*
857 				 * calculate the page index into the object
858 				 */
859 				offset = current->offset + (addr - current->start);
860 				pindex = OFF_TO_IDX(offset);
861 				VM_OBJECT_LOCK(current->object.vm_object);
862 				m = vm_page_lookup(current->object.vm_object,
863 					pindex);
864 				/*
865 				 * if the page is resident, then gather information about
866 				 * it.
867 				 */
868 				if (m != NULL && m->valid != 0) {
869 					mincoreinfo = MINCORE_INCORE;
870 					vm_page_lock_queues();
871 					if (m->dirty ||
872 						pmap_is_modified(m))
873 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
874 					if ((m->flags & PG_REFERENCED) ||
875 						pmap_ts_referenced(m)) {
876 						vm_page_flag_set(m, PG_REFERENCED);
877 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
878 					}
879 					vm_page_unlock_queues();
880 				}
881 				VM_OBJECT_UNLOCK(current->object.vm_object);
882 			}
883 
884 			/*
885 			 * subyte may page fault.  In case it needs to modify
886 			 * the map, we release the lock.
887 			 */
888 			vm_map_unlock_read(map);
889 
890 			/*
891 			 * calculate index into user supplied byte vector
892 			 */
893 			vecindex = OFF_TO_IDX(addr - first_addr);
894 
895 			/*
896 			 * If we have skipped map entries, we need to make sure that
897 			 * the byte vector is zeroed for those skipped entries.
898 			 */
899 			while ((lastvecindex + 1) < vecindex) {
900 				error = subyte(vec + lastvecindex, 0);
901 				if (error) {
902 					error = EFAULT;
903 					goto done2;
904 				}
905 				++lastvecindex;
906 			}
907 
908 			/*
909 			 * Pass the page information to the user
910 			 */
911 			error = subyte(vec + vecindex, mincoreinfo);
912 			if (error) {
913 				error = EFAULT;
914 				goto done2;
915 			}
916 
917 			/*
918 			 * If the map has changed, due to the subyte, the previous
919 			 * output may be invalid.
920 			 */
921 			vm_map_lock_read(map);
922 			if (timestamp != map->timestamp)
923 				goto RestartScan;
924 
925 			lastvecindex = vecindex;
926 			addr += PAGE_SIZE;
927 		}
928 	}
929 
930 	/*
931 	 * subyte may page fault.  In case it needs to modify
932 	 * the map, we release the lock.
933 	 */
934 	vm_map_unlock_read(map);
935 
936 	/*
937 	 * Zero the last entries in the byte vector.
938 	 */
939 	vecindex = OFF_TO_IDX(end - first_addr);
940 	while ((lastvecindex + 1) < vecindex) {
941 		error = subyte(vec + lastvecindex, 0);
942 		if (error) {
943 			error = EFAULT;
944 			goto done2;
945 		}
946 		++lastvecindex;
947 	}
948 
949 	/*
950 	 * If the map has changed, due to the subyte, the previous
951 	 * output may be invalid.
952 	 */
953 	vm_map_lock_read(map);
954 	if (timestamp != map->timestamp)
955 		goto RestartScan;
956 	vm_map_unlock_read(map);
957 done2:
958 	return (error);
959 }
960 
961 #ifndef _SYS_SYSPROTO_H_
962 struct mlock_args {
963 	const void *addr;
964 	size_t len;
965 };
966 #endif
967 /*
968  * MPSAFE
969  */
970 int
971 mlock(td, uap)
972 	struct thread *td;
973 	struct mlock_args *uap;
974 {
975 	struct proc *proc;
976 	vm_offset_t addr, end, last, start;
977 	vm_size_t npages, size;
978 	int error;
979 
980 	error = priv_check(td, PRIV_VM_MLOCK);
981 	if (error)
982 		return (error);
983 	addr = (vm_offset_t)uap->addr;
984 	size = uap->len;
985 	last = addr + size;
986 	start = trunc_page(addr);
987 	end = round_page(last);
988 	if (last < addr || end < addr)
989 		return (EINVAL);
990 	npages = atop(end - start);
991 	if (npages > vm_page_max_wired)
992 		return (ENOMEM);
993 	proc = td->td_proc;
994 	PROC_LOCK(proc);
995 	if (ptoa(npages +
996 	    pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
997 	    lim_cur(proc, RLIMIT_MEMLOCK)) {
998 		PROC_UNLOCK(proc);
999 		return (ENOMEM);
1000 	}
1001 	PROC_UNLOCK(proc);
1002 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1003 		return (EAGAIN);
1004 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1005 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1006 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1007 }
1008 
1009 #ifndef _SYS_SYSPROTO_H_
1010 struct mlockall_args {
1011 	int	how;
1012 };
1013 #endif
1014 
1015 /*
1016  * MPSAFE
1017  */
1018 int
1019 mlockall(td, uap)
1020 	struct thread *td;
1021 	struct mlockall_args *uap;
1022 {
1023 	vm_map_t map;
1024 	int error;
1025 
1026 	map = &td->td_proc->p_vmspace->vm_map;
1027 	error = 0;
1028 
1029 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1030 		return (EINVAL);
1031 
1032 #if 0
1033 	/*
1034 	 * If wiring all pages in the process would cause it to exceed
1035 	 * a hard resource limit, return ENOMEM.
1036 	 */
1037 	PROC_LOCK(td->td_proc);
1038 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1039 		lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1040 		PROC_UNLOCK(td->td_proc);
1041 		return (ENOMEM);
1042 	}
1043 	PROC_UNLOCK(td->td_proc);
1044 #else
1045 	error = priv_check(td, PRIV_VM_MLOCK);
1046 	if (error)
1047 		return (error);
1048 #endif
1049 
1050 	if (uap->how & MCL_FUTURE) {
1051 		vm_map_lock(map);
1052 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1053 		vm_map_unlock(map);
1054 		error = 0;
1055 	}
1056 
1057 	if (uap->how & MCL_CURRENT) {
1058 		/*
1059 		 * P1003.1-2001 mandates that all currently mapped pages
1060 		 * will be memory resident and locked (wired) upon return
1061 		 * from mlockall(). vm_map_wire() will wire pages, by
1062 		 * calling vm_fault_wire() for each page in the region.
1063 		 */
1064 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1065 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1066 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1067 	}
1068 
1069 	return (error);
1070 }
1071 
1072 #ifndef _SYS_SYSPROTO_H_
1073 struct munlockall_args {
1074 	register_t dummy;
1075 };
1076 #endif
1077 
1078 /*
1079  * MPSAFE
1080  */
1081 int
1082 munlockall(td, uap)
1083 	struct thread *td;
1084 	struct munlockall_args *uap;
1085 {
1086 	vm_map_t map;
1087 	int error;
1088 
1089 	map = &td->td_proc->p_vmspace->vm_map;
1090 	error = priv_check(td, PRIV_VM_MUNLOCK);
1091 	if (error)
1092 		return (error);
1093 
1094 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1095 	vm_map_lock(map);
1096 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1097 	vm_map_unlock(map);
1098 
1099 	/* Forcibly unwire all pages. */
1100 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1101 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1102 
1103 	return (error);
1104 }
1105 
1106 #ifndef _SYS_SYSPROTO_H_
1107 struct munlock_args {
1108 	const void *addr;
1109 	size_t len;
1110 };
1111 #endif
1112 /*
1113  * MPSAFE
1114  */
1115 int
1116 munlock(td, uap)
1117 	struct thread *td;
1118 	struct munlock_args *uap;
1119 {
1120 	vm_offset_t addr, end, last, start;
1121 	vm_size_t size;
1122 	int error;
1123 
1124 	error = priv_check(td, PRIV_VM_MUNLOCK);
1125 	if (error)
1126 		return (error);
1127 	addr = (vm_offset_t)uap->addr;
1128 	size = uap->len;
1129 	last = addr + size;
1130 	start = trunc_page(addr);
1131 	end = round_page(last);
1132 	if (last < addr || end < addr)
1133 		return (EINVAL);
1134 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1135 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1136 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1137 }
1138 
1139 /*
1140  * vm_mmap_vnode()
1141  *
1142  * MPSAFE
1143  *
1144  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1145  * operations on vnodes.
1146  */
1147 int
1148 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1149     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1150     struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1151 {
1152 	struct vattr va;
1153 	void *handle;
1154 	vm_object_t obj;
1155 	struct mount *mp;
1156 	struct cdevsw *dsw;
1157 	int error, flags, type;
1158 	int vfslocked;
1159 
1160 	mp = vp->v_mount;
1161 	vfslocked = VFS_LOCK_GIANT(mp);
1162 	if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1163 		VFS_UNLOCK_GIANT(vfslocked);
1164 		return (error);
1165 	}
1166 	flags = *flagsp;
1167 	obj = vp->v_object;
1168 	if (vp->v_type == VREG) {
1169 		/*
1170 		 * Get the proper underlying object
1171 		 */
1172 		if (obj == NULL) {
1173 			error = EINVAL;
1174 			goto done;
1175 		}
1176 		if (obj->handle != vp) {
1177 			vput(vp);
1178 			vp = (struct vnode*)obj->handle;
1179 			vget(vp, LK_EXCLUSIVE, td);
1180 		}
1181 		type = OBJT_VNODE;
1182 		handle = vp;
1183 	} else if (vp->v_type == VCHR) {
1184 		type = OBJT_DEVICE;
1185 		handle = vp->v_rdev;
1186 
1187 		dsw = dev_refthread(handle);
1188 		if (dsw == NULL) {
1189 			error = ENXIO;
1190 			goto done;
1191 		}
1192 		if (dsw->d_flags & D_MMAP_ANON) {
1193 			dev_relthread(handle);
1194 			*maxprotp = VM_PROT_ALL;
1195 			*flagsp |= MAP_ANON;
1196 			error = 0;
1197 			goto done;
1198 		}
1199 		dev_relthread(handle);
1200 		/*
1201 		 * cdevs does not provide private mappings of any kind.
1202 		 */
1203 		if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1204 		    (prot & PROT_WRITE) != 0) {
1205 			error = EACCES;
1206 			goto done;
1207 		}
1208 		if (flags & (MAP_PRIVATE|MAP_COPY)) {
1209 			error = EINVAL;
1210 			goto done;
1211 		}
1212 		/*
1213 		 * Force device mappings to be shared.
1214 		 */
1215 		flags |= MAP_SHARED;
1216 	} else {
1217 		error = EINVAL;
1218 		goto done;
1219 	}
1220 	if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1221 		goto done;
1222 	}
1223 #ifdef MAC
1224 	error = mac_vnode_check_mmap(td->td_ucred, vp, prot, flags);
1225 	if (error != 0)
1226 		goto done;
1227 #endif
1228 	if ((flags & MAP_SHARED) != 0) {
1229 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1230 			if (prot & PROT_WRITE) {
1231 				error = EPERM;
1232 				goto done;
1233 			}
1234 			*maxprotp &= ~VM_PROT_WRITE;
1235 		}
1236 	}
1237 	/*
1238 	 * If it is a regular file without any references
1239 	 * we do not need to sync it.
1240 	 * Adjust object size to be the size of actual file.
1241 	 */
1242 	if (vp->v_type == VREG) {
1243 		objsize = round_page(va.va_size);
1244 		if (va.va_nlink == 0)
1245 			flags |= MAP_NOSYNC;
1246 	}
1247 	obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1248 	if (obj == NULL) {
1249 		error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1250 		goto done;
1251 	}
1252 	*objp = obj;
1253 	*flagsp = flags;
1254 	vfs_mark_atime(vp, td);
1255 
1256 done:
1257 	vput(vp);
1258 	VFS_UNLOCK_GIANT(vfslocked);
1259 	return (error);
1260 }
1261 
1262 /*
1263  * vm_mmap_cdev()
1264  *
1265  * MPSAFE
1266  *
1267  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1268  * operations on cdevs.
1269  */
1270 int
1271 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1272     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1273     struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1274 {
1275 	vm_object_t obj;
1276 	struct cdevsw *dsw;
1277 	int flags;
1278 
1279 	flags = *flagsp;
1280 
1281 	dsw = dev_refthread(cdev);
1282 	if (dsw == NULL)
1283 		return (ENXIO);
1284 	if (dsw->d_flags & D_MMAP_ANON) {
1285 		dev_relthread(cdev);
1286 		*maxprotp = VM_PROT_ALL;
1287 		*flagsp |= MAP_ANON;
1288 		return (0);
1289 	}
1290 	dev_relthread(cdev);
1291 	/*
1292 	 * cdevs does not provide private mappings of any kind.
1293 	 */
1294 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1295 	    (prot & PROT_WRITE) != 0)
1296 		return (EACCES);
1297 	if (flags & (MAP_PRIVATE|MAP_COPY))
1298 		return (EINVAL);
1299 	/*
1300 	 * Force device mappings to be shared.
1301 	 */
1302 	flags |= MAP_SHARED;
1303 #ifdef MAC_XXX
1304 	error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1305 	if (error != 0)
1306 		return (error);
1307 #endif
1308 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1309 	if (obj == NULL)
1310 		return (EINVAL);
1311 	*objp = obj;
1312 	*flagsp = flags;
1313 	return (0);
1314 }
1315 
1316 /*
1317  * vm_mmap_shm()
1318  *
1319  * MPSAFE
1320  *
1321  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1322  * operations on shm file descriptors.
1323  */
1324 int
1325 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1326     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1327     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1328 {
1329 	int error;
1330 
1331 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1332 	    (prot & PROT_WRITE) != 0)
1333 		return (EACCES);
1334 #ifdef MAC
1335 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1336 	if (error != 0)
1337 		return (error);
1338 #endif
1339 	error = shm_mmap(shmfd, objsize, foff, objp);
1340 	if (error)
1341 		return (error);
1342 	return (0);
1343 }
1344 
1345 /*
1346  * vm_mmap()
1347  *
1348  * MPSAFE
1349  *
1350  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1351  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1352  */
1353 int
1354 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1355 	vm_prot_t maxprot, int flags,
1356 	objtype_t handle_type, void *handle,
1357 	vm_ooffset_t foff)
1358 {
1359 	boolean_t fitit;
1360 	vm_object_t object = NULL;
1361 	int rv = KERN_SUCCESS;
1362 	int docow, error;
1363 	struct thread *td = curthread;
1364 
1365 	if (size == 0)
1366 		return (0);
1367 
1368 	size = round_page(size);
1369 
1370 	PROC_LOCK(td->td_proc);
1371 	if (td->td_proc->p_vmspace->vm_map.size + size >
1372 	    lim_cur(td->td_proc, RLIMIT_VMEM)) {
1373 		PROC_UNLOCK(td->td_proc);
1374 		return(ENOMEM);
1375 	}
1376 	PROC_UNLOCK(td->td_proc);
1377 
1378 	/*
1379 	 * We currently can only deal with page aligned file offsets.
1380 	 * The check is here rather than in the syscall because the
1381 	 * kernel calls this function internally for other mmaping
1382 	 * operations (such as in exec) and non-aligned offsets will
1383 	 * cause pmap inconsistencies...so we want to be sure to
1384 	 * disallow this in all cases.
1385 	 */
1386 	if (foff & PAGE_MASK)
1387 		return (EINVAL);
1388 
1389 	if ((flags & MAP_FIXED) == 0) {
1390 		fitit = TRUE;
1391 		*addr = round_page(*addr);
1392 	} else {
1393 		if (*addr != trunc_page(*addr))
1394 			return (EINVAL);
1395 		fitit = FALSE;
1396 	}
1397 	/*
1398 	 * Lookup/allocate object.
1399 	 */
1400 	switch (handle_type) {
1401 	case OBJT_DEVICE:
1402 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1403 		    handle, foff, &object);
1404 		break;
1405 	case OBJT_VNODE:
1406 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1407 		    handle, foff, &object);
1408 		break;
1409 	case OBJT_SWAP:
1410 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1411 		    handle, foff, &object);
1412 		break;
1413 	case OBJT_DEFAULT:
1414 		if (handle == NULL) {
1415 			error = 0;
1416 			break;
1417 		}
1418 		/* FALLTHROUGH */
1419 	default:
1420 		error = EINVAL;
1421 		break;
1422 	}
1423 	if (error)
1424 		return (error);
1425 	if (flags & MAP_ANON) {
1426 		object = NULL;
1427 		docow = 0;
1428 		/*
1429 		 * Unnamed anonymous regions always start at 0.
1430 		 */
1431 		if (handle == 0)
1432 			foff = 0;
1433 	} else {
1434 		docow = MAP_PREFAULT_PARTIAL;
1435 	}
1436 
1437 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1438 		docow |= MAP_COPY_ON_WRITE;
1439 	if (flags & MAP_NOSYNC)
1440 		docow |= MAP_DISABLE_SYNCER;
1441 	if (flags & MAP_NOCORE)
1442 		docow |= MAP_DISABLE_COREDUMP;
1443 
1444 #if defined(VM_PROT_READ_IS_EXEC)
1445 	if (prot & VM_PROT_READ)
1446 		prot |= VM_PROT_EXECUTE;
1447 
1448 	if (maxprot & VM_PROT_READ)
1449 		maxprot |= VM_PROT_EXECUTE;
1450 #endif
1451 
1452 	if (flags & MAP_STACK)
1453 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1454 		    docow | MAP_STACK_GROWS_DOWN);
1455 	else if (fitit)
1456 		rv = vm_map_find(map, object, foff, addr, size,
1457 		    object != NULL && object->type == OBJT_DEVICE ?
1458 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1459 	else
1460 		rv = vm_map_fixed(map, object, foff, *addr, size,
1461 				 prot, maxprot, docow);
1462 
1463 	if (rv != KERN_SUCCESS) {
1464 		/*
1465 		 * Lose the object reference. Will destroy the
1466 		 * object if it's an unnamed anonymous mapping
1467 		 * or named anonymous without other references.
1468 		 */
1469 		vm_object_deallocate(object);
1470 	} else if (flags & MAP_SHARED) {
1471 		/*
1472 		 * Shared memory is also shared with children.
1473 		 */
1474 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1475 		if (rv != KERN_SUCCESS)
1476 			(void) vm_map_remove(map, *addr, *addr + size);
1477 	}
1478 
1479 	/*
1480 	 * If the process has requested that all future mappings
1481 	 * be wired, then heed this.
1482 	 */
1483 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1484 		vm_map_wire(map, *addr, *addr + size,
1485 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1486 
1487 	switch (rv) {
1488 	case KERN_SUCCESS:
1489 		return (0);
1490 	case KERN_INVALID_ADDRESS:
1491 	case KERN_NO_SPACE:
1492 		return (ENOMEM);
1493 	case KERN_PROTECTION_FAILURE:
1494 		return (EACCES);
1495 	default:
1496 		return (EINVAL);
1497 	}
1498 }
1499