xref: /freebsd/sys/vm/vm_mmap.c (revision cacdd70cc751fb68dec4b86c5e5b8c969b6e26ef)
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     "Maximum number of memory-mapped files per process");
98 
99 /*
100  * Set the maximum number of vm_map_entry structures per process.  Roughly
101  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
102  * of our KVM malloc space still results in generous limits.  We want a
103  * default that is good enough to prevent the kernel running out of resources
104  * if attacked from compromised user account but generous enough such that
105  * multi-threaded processes are not unduly inconvenienced.
106  */
107 static void vmmapentry_rsrc_init(void *);
108 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init,
109     NULL);
110 
111 static void
112 vmmapentry_rsrc_init(dummy)
113         void *dummy;
114 {
115     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
116     max_proc_mmap /= 100;
117 }
118 
119 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
120     int *, struct vnode *, vm_ooffset_t, vm_object_t *);
121 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
122     int *, struct cdev *, vm_ooffset_t, vm_object_t *);
123 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
124     int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
125 
126 /*
127  * MPSAFE
128  */
129 /* ARGSUSED */
130 int
131 sbrk(td, uap)
132 	struct thread *td;
133 	struct sbrk_args *uap;
134 {
135 	/* Not yet implemented */
136 	return (EOPNOTSUPP);
137 }
138 
139 #ifndef _SYS_SYSPROTO_H_
140 struct sstk_args {
141 	int incr;
142 };
143 #endif
144 
145 /*
146  * MPSAFE
147  */
148 /* ARGSUSED */
149 int
150 sstk(td, uap)
151 	struct thread *td;
152 	struct sstk_args *uap;
153 {
154 	/* Not yet implemented */
155 	return (EOPNOTSUPP);
156 }
157 
158 #if defined(COMPAT_43)
159 #ifndef _SYS_SYSPROTO_H_
160 struct getpagesize_args {
161 	int dummy;
162 };
163 #endif
164 
165 /* ARGSUSED */
166 int
167 ogetpagesize(td, uap)
168 	struct thread *td;
169 	struct getpagesize_args *uap;
170 {
171 	/* MP SAFE */
172 	td->td_retval[0] = PAGE_SIZE;
173 	return (0);
174 }
175 #endif				/* COMPAT_43 */
176 
177 
178 /*
179  * Memory Map (mmap) system call.  Note that the file offset
180  * and address are allowed to be NOT page aligned, though if
181  * the MAP_FIXED flag it set, both must have the same remainder
182  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
183  * page-aligned, the actual mapping starts at trunc_page(addr)
184  * and the return value is adjusted up by the page offset.
185  *
186  * Generally speaking, only character devices which are themselves
187  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
188  * there would be no cache coherency between a descriptor and a VM mapping
189  * both to the same character device.
190  *
191  * Block devices can be mmap'd no matter what they represent.  Cache coherency
192  * is maintained as long as you do not write directly to the underlying
193  * character device.
194  */
195 #ifndef _SYS_SYSPROTO_H_
196 struct mmap_args {
197 	void *addr;
198 	size_t len;
199 	int prot;
200 	int flags;
201 	int fd;
202 	long pad;
203 	off_t pos;
204 };
205 #endif
206 
207 /*
208  * MPSAFE
209  */
210 int
211 mmap(td, uap)
212 	struct thread *td;
213 	struct mmap_args *uap;
214 {
215 #ifdef HWPMC_HOOKS
216 	struct pmckern_map_in pkm;
217 #endif
218 	struct file *fp;
219 	struct vnode *vp;
220 	vm_offset_t addr;
221 	vm_size_t size, pageoff;
222 	vm_prot_t prot, maxprot;
223 	void *handle;
224 	objtype_t handle_type;
225 	int flags, error;
226 	off_t pos;
227 	struct vmspace *vms = td->td_proc->p_vmspace;
228 
229 	addr = (vm_offset_t) uap->addr;
230 	size = uap->len;
231 	prot = uap->prot & VM_PROT_ALL;
232 	flags = uap->flags;
233 	pos = uap->pos;
234 
235 	fp = NULL;
236 	/* make sure mapping fits into numeric range etc */
237 	if ((ssize_t) uap->len < 0 ||
238 	    ((flags & MAP_ANON) && uap->fd != -1))
239 		return (EINVAL);
240 
241 	if (flags & MAP_STACK) {
242 		if ((uap->fd != -1) ||
243 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
244 			return (EINVAL);
245 		flags |= MAP_ANON;
246 		pos = 0;
247 	}
248 
249 	/*
250 	 * Align the file position to a page boundary,
251 	 * and save its page offset component.
252 	 */
253 	pageoff = (pos & PAGE_MASK);
254 	pos -= pageoff;
255 
256 	/* Adjust size for rounding (on both ends). */
257 	size += pageoff;			/* low end... */
258 	size = (vm_size_t) round_page(size);	/* hi end */
259 
260 	/*
261 	 * Check for illegal addresses.  Watch out for address wrap... Note
262 	 * that VM_*_ADDRESS are not constants due to casts (argh).
263 	 */
264 	if (flags & MAP_FIXED) {
265 		/*
266 		 * The specified address must have the same remainder
267 		 * as the file offset taken modulo PAGE_SIZE, so it
268 		 * should be aligned after adjustment by pageoff.
269 		 */
270 		addr -= pageoff;
271 		if (addr & PAGE_MASK)
272 			return (EINVAL);
273 		/* Address range must be all in user VM space. */
274 		if (addr < vm_map_min(&vms->vm_map) ||
275 		    addr + size > vm_map_max(&vms->vm_map))
276 			return (EINVAL);
277 		if (addr + size < addr)
278 			return (EINVAL);
279 	} else {
280 	/*
281 	 * XXX for non-fixed mappings where no hint is provided or
282 	 * the hint would fall in the potential heap space,
283 	 * place it after the end of the largest possible heap.
284 	 *
285 	 * There should really be a pmap call to determine a reasonable
286 	 * location.
287 	 */
288 		PROC_LOCK(td->td_proc);
289 		if (addr == 0 ||
290 		    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
291 		    addr < round_page((vm_offset_t)vms->vm_daddr +
292 		    lim_max(td->td_proc, RLIMIT_DATA))))
293 			addr = round_page((vm_offset_t)vms->vm_daddr +
294 			    lim_max(td->td_proc, RLIMIT_DATA));
295 		PROC_UNLOCK(td->td_proc);
296 	}
297 	if (flags & MAP_ANON) {
298 		/*
299 		 * Mapping blank space is trivial.
300 		 */
301 		handle = NULL;
302 		handle_type = OBJT_DEFAULT;
303 		maxprot = VM_PROT_ALL;
304 		pos = 0;
305 	} else {
306 		/*
307 		 * Mapping file, get fp for validation and
308 		 * don't let the descriptor disappear on us if we block.
309 		 */
310 		if ((error = fget(td, uap->fd, &fp)) != 0)
311 			goto done;
312 		if (fp->f_type == DTYPE_SHM) {
313 			handle = fp->f_data;
314 			handle_type = OBJT_SWAP;
315 			maxprot = VM_PROT_NONE;
316 
317 			/* FREAD should always be set. */
318 			if (fp->f_flag & FREAD)
319 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
320 			if (fp->f_flag & FWRITE)
321 				maxprot |= VM_PROT_WRITE;
322 			goto map;
323 		}
324 		if (fp->f_type != DTYPE_VNODE) {
325 			error = ENODEV;
326 			goto done;
327 		}
328 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
329     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
330 		/*
331 		 * POSIX shared-memory objects are defined to have
332 		 * kernel persistence, and are not defined to support
333 		 * read(2)/write(2) -- or even open(2).  Thus, we can
334 		 * use MAP_ASYNC to trade on-disk coherence for speed.
335 		 * The shm_open(3) library routine turns on the FPOSIXSHM
336 		 * flag to request this behavior.
337 		 */
338 		if (fp->f_flag & FPOSIXSHM)
339 			flags |= MAP_NOSYNC;
340 #endif
341 		vp = fp->f_vnode;
342 		/*
343 		 * Ensure that file and memory protections are
344 		 * compatible.  Note that we only worry about
345 		 * writability if mapping is shared; in this case,
346 		 * current and max prot are dictated by the open file.
347 		 * XXX use the vnode instead?  Problem is: what
348 		 * credentials do we use for determination? What if
349 		 * proc does a setuid?
350 		 */
351 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
352 			maxprot = VM_PROT_NONE;
353 		else
354 			maxprot = VM_PROT_EXECUTE;
355 		if (fp->f_flag & FREAD) {
356 			maxprot |= VM_PROT_READ;
357 		} else if (prot & PROT_READ) {
358 			error = EACCES;
359 			goto done;
360 		}
361 		/*
362 		 * If we are sharing potential changes (either via
363 		 * MAP_SHARED or via the implicit sharing of character
364 		 * device mappings), and we are trying to get write
365 		 * permission although we opened it without asking
366 		 * for it, bail out.
367 		 */
368 		if ((flags & MAP_SHARED) != 0) {
369 			if ((fp->f_flag & FWRITE) != 0) {
370 				maxprot |= VM_PROT_WRITE;
371 			} else if ((prot & PROT_WRITE) != 0) {
372 				error = EACCES;
373 				goto done;
374 			}
375 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
376 			maxprot |= VM_PROT_WRITE;
377 		}
378 		handle = (void *)vp;
379 		handle_type = OBJT_VNODE;
380 	}
381 map:
382 
383 	/*
384 	 * Do not allow more then a certain number of vm_map_entry structures
385 	 * per process.  Scale with the number of rforks sharing the map
386 	 * to make the limit reasonable for threads.
387 	 */
388 	if (max_proc_mmap &&
389 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
390 		error = ENOMEM;
391 		goto done;
392 	}
393 
394 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
395 	    flags, handle_type, handle, pos);
396 #ifdef HWPMC_HOOKS
397 	/* inform hwpmc(4) if an executable is being mapped */
398 	if (error == 0 && handle_type == OBJT_VNODE &&
399 	    (prot & PROT_EXEC)) {
400 		pkm.pm_file = handle;
401 		pkm.pm_address = (uintptr_t) addr;
402 		PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
403 	}
404 #endif
405 	if (error == 0)
406 		td->td_retval[0] = (register_t) (addr + pageoff);
407 done:
408 	if (fp)
409 		fdrop(fp, td);
410 
411 	return (error);
412 }
413 
414 int
415 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
416 {
417 	struct mmap_args oargs;
418 
419 	oargs.addr = uap->addr;
420 	oargs.len = uap->len;
421 	oargs.prot = uap->prot;
422 	oargs.flags = uap->flags;
423 	oargs.fd = uap->fd;
424 	oargs.pos = uap->pos;
425 	return (mmap(td, &oargs));
426 }
427 
428 #ifdef COMPAT_43
429 #ifndef _SYS_SYSPROTO_H_
430 struct ommap_args {
431 	caddr_t addr;
432 	int len;
433 	int prot;
434 	int flags;
435 	int fd;
436 	long pos;
437 };
438 #endif
439 int
440 ommap(td, uap)
441 	struct thread *td;
442 	struct ommap_args *uap;
443 {
444 	struct mmap_args nargs;
445 	static const char cvtbsdprot[8] = {
446 		0,
447 		PROT_EXEC,
448 		PROT_WRITE,
449 		PROT_EXEC | PROT_WRITE,
450 		PROT_READ,
451 		PROT_EXEC | PROT_READ,
452 		PROT_WRITE | PROT_READ,
453 		PROT_EXEC | PROT_WRITE | PROT_READ,
454 	};
455 
456 #define	OMAP_ANON	0x0002
457 #define	OMAP_COPY	0x0020
458 #define	OMAP_SHARED	0x0010
459 #define	OMAP_FIXED	0x0100
460 
461 	nargs.addr = uap->addr;
462 	nargs.len = uap->len;
463 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
464 	nargs.flags = 0;
465 	if (uap->flags & OMAP_ANON)
466 		nargs.flags |= MAP_ANON;
467 	if (uap->flags & OMAP_COPY)
468 		nargs.flags |= MAP_COPY;
469 	if (uap->flags & OMAP_SHARED)
470 		nargs.flags |= MAP_SHARED;
471 	else
472 		nargs.flags |= MAP_PRIVATE;
473 	if (uap->flags & OMAP_FIXED)
474 		nargs.flags |= MAP_FIXED;
475 	nargs.fd = uap->fd;
476 	nargs.pos = uap->pos;
477 	return (mmap(td, &nargs));
478 }
479 #endif				/* COMPAT_43 */
480 
481 
482 #ifndef _SYS_SYSPROTO_H_
483 struct msync_args {
484 	void *addr;
485 	size_t len;
486 	int flags;
487 };
488 #endif
489 /*
490  * MPSAFE
491  */
492 int
493 msync(td, uap)
494 	struct thread *td;
495 	struct msync_args *uap;
496 {
497 	vm_offset_t addr;
498 	vm_size_t size, pageoff;
499 	int flags;
500 	vm_map_t map;
501 	int rv;
502 
503 	addr = (vm_offset_t) uap->addr;
504 	size = uap->len;
505 	flags = uap->flags;
506 
507 	pageoff = (addr & PAGE_MASK);
508 	addr -= pageoff;
509 	size += pageoff;
510 	size = (vm_size_t) round_page(size);
511 	if (addr + size < addr)
512 		return (EINVAL);
513 
514 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
515 		return (EINVAL);
516 
517 	map = &td->td_proc->p_vmspace->vm_map;
518 
519 	/*
520 	 * Clean the pages and interpret the return value.
521 	 */
522 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
523 	    (flags & MS_INVALIDATE) != 0);
524 	switch (rv) {
525 	case KERN_SUCCESS:
526 		return (0);
527 	case KERN_INVALID_ADDRESS:
528 		return (EINVAL);	/* Sun returns ENOMEM? */
529 	case KERN_INVALID_ARGUMENT:
530 		return (EBUSY);
531 	default:
532 		return (EINVAL);
533 	}
534 }
535 
536 #ifndef _SYS_SYSPROTO_H_
537 struct munmap_args {
538 	void *addr;
539 	size_t len;
540 };
541 #endif
542 /*
543  * MPSAFE
544  */
545 int
546 munmap(td, uap)
547 	struct thread *td;
548 	struct munmap_args *uap;
549 {
550 #ifdef HWPMC_HOOKS
551 	struct pmckern_map_out pkm;
552 	vm_map_entry_t entry;
553 #endif
554 	vm_offset_t addr;
555 	vm_size_t size, pageoff;
556 	vm_map_t map;
557 
558 	addr = (vm_offset_t) uap->addr;
559 	size = uap->len;
560 	if (size == 0)
561 		return (EINVAL);
562 
563 	pageoff = (addr & PAGE_MASK);
564 	addr -= pageoff;
565 	size += pageoff;
566 	size = (vm_size_t) round_page(size);
567 	if (addr + size < addr)
568 		return (EINVAL);
569 
570 	/*
571 	 * Check for illegal addresses.  Watch out for address wrap...
572 	 */
573 	map = &td->td_proc->p_vmspace->vm_map;
574 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
575 		return (EINVAL);
576 	vm_map_lock(map);
577 #ifdef HWPMC_HOOKS
578 	/*
579 	 * Inform hwpmc if the address range being unmapped contains
580 	 * an executable region.
581 	 */
582 	if (vm_map_lookup_entry(map, addr, &entry)) {
583 		for (;
584 		     entry != &map->header && entry->start < addr + size;
585 		     entry = entry->next) {
586 			if (vm_map_check_protection(map, entry->start,
587 				entry->end, VM_PROT_EXECUTE) == TRUE) {
588 				pkm.pm_address = (uintptr_t) addr;
589 				pkm.pm_size = (size_t) size;
590 				PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
591 				    (void *) &pkm);
592 				break;
593 			}
594 		}
595 	}
596 #endif
597 	/* returns nothing but KERN_SUCCESS anyway */
598 	vm_map_delete(map, addr, addr + size);
599 	vm_map_unlock(map);
600 	return (0);
601 }
602 
603 #ifndef _SYS_SYSPROTO_H_
604 struct mprotect_args {
605 	const void *addr;
606 	size_t len;
607 	int prot;
608 };
609 #endif
610 /*
611  * MPSAFE
612  */
613 int
614 mprotect(td, uap)
615 	struct thread *td;
616 	struct mprotect_args *uap;
617 {
618 	vm_offset_t addr;
619 	vm_size_t size, pageoff;
620 	vm_prot_t prot;
621 
622 	addr = (vm_offset_t) uap->addr;
623 	size = uap->len;
624 	prot = uap->prot & VM_PROT_ALL;
625 #if defined(VM_PROT_READ_IS_EXEC)
626 	if (prot & VM_PROT_READ)
627 		prot |= VM_PROT_EXECUTE;
628 #endif
629 
630 	pageoff = (addr & PAGE_MASK);
631 	addr -= pageoff;
632 	size += pageoff;
633 	size = (vm_size_t) round_page(size);
634 	if (addr + size < addr)
635 		return (EINVAL);
636 
637 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
638 	    addr + size, prot, FALSE)) {
639 	case KERN_SUCCESS:
640 		return (0);
641 	case KERN_PROTECTION_FAILURE:
642 		return (EACCES);
643 	}
644 	return (EINVAL);
645 }
646 
647 #ifndef _SYS_SYSPROTO_H_
648 struct minherit_args {
649 	void *addr;
650 	size_t len;
651 	int inherit;
652 };
653 #endif
654 /*
655  * MPSAFE
656  */
657 int
658 minherit(td, uap)
659 	struct thread *td;
660 	struct minherit_args *uap;
661 {
662 	vm_offset_t addr;
663 	vm_size_t size, pageoff;
664 	vm_inherit_t inherit;
665 
666 	addr = (vm_offset_t)uap->addr;
667 	size = uap->len;
668 	inherit = uap->inherit;
669 
670 	pageoff = (addr & PAGE_MASK);
671 	addr -= pageoff;
672 	size += pageoff;
673 	size = (vm_size_t) round_page(size);
674 	if (addr + size < addr)
675 		return (EINVAL);
676 
677 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
678 	    addr + size, inherit)) {
679 	case KERN_SUCCESS:
680 		return (0);
681 	case KERN_PROTECTION_FAILURE:
682 		return (EACCES);
683 	}
684 	return (EINVAL);
685 }
686 
687 #ifndef _SYS_SYSPROTO_H_
688 struct madvise_args {
689 	void *addr;
690 	size_t len;
691 	int behav;
692 };
693 #endif
694 
695 /*
696  * MPSAFE
697  */
698 /* ARGSUSED */
699 int
700 madvise(td, uap)
701 	struct thread *td;
702 	struct madvise_args *uap;
703 {
704 	vm_offset_t start, end;
705 	vm_map_t map;
706 	struct proc *p;
707 	int error;
708 
709 	/*
710 	 * Check for our special case, advising the swap pager we are
711 	 * "immortal."
712 	 */
713 	if (uap->behav == MADV_PROTECT) {
714 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
715 		if (error == 0) {
716 			p = td->td_proc;
717 			PROC_LOCK(p);
718 			p->p_flag |= P_PROTECTED;
719 			PROC_UNLOCK(p);
720 		}
721 		return (error);
722 	}
723 	/*
724 	 * Check for illegal behavior
725 	 */
726 	if (uap->behav < 0 || uap->behav > MADV_CORE)
727 		return (EINVAL);
728 	/*
729 	 * Check for illegal addresses.  Watch out for address wrap... Note
730 	 * that VM_*_ADDRESS are not constants due to casts (argh).
731 	 */
732 	map = &td->td_proc->p_vmspace->vm_map;
733 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
734 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
735 		return (EINVAL);
736 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
737 		return (EINVAL);
738 
739 	/*
740 	 * Since this routine is only advisory, we default to conservative
741 	 * behavior.
742 	 */
743 	start = trunc_page((vm_offset_t) uap->addr);
744 	end = round_page((vm_offset_t) uap->addr + uap->len);
745 
746 	if (vm_map_madvise(map, start, end, uap->behav))
747 		return (EINVAL);
748 	return (0);
749 }
750 
751 #ifndef _SYS_SYSPROTO_H_
752 struct mincore_args {
753 	const void *addr;
754 	size_t len;
755 	char *vec;
756 };
757 #endif
758 
759 /*
760  * MPSAFE
761  */
762 /* ARGSUSED */
763 int
764 mincore(td, uap)
765 	struct thread *td;
766 	struct mincore_args *uap;
767 {
768 	vm_offset_t addr, first_addr;
769 	vm_offset_t end, cend;
770 	pmap_t pmap;
771 	vm_map_t map;
772 	char *vec;
773 	int error = 0;
774 	int vecindex, lastvecindex;
775 	vm_map_entry_t current;
776 	vm_map_entry_t entry;
777 	int mincoreinfo;
778 	unsigned int timestamp;
779 
780 	/*
781 	 * Make sure that the addresses presented are valid for user
782 	 * mode.
783 	 */
784 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
785 	end = addr + (vm_size_t)round_page(uap->len);
786 	map = &td->td_proc->p_vmspace->vm_map;
787 	if (end > vm_map_max(map) || end < addr)
788 		return (ENOMEM);
789 
790 	/*
791 	 * Address of byte vector
792 	 */
793 	vec = uap->vec;
794 
795 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
796 
797 	vm_map_lock_read(map);
798 RestartScan:
799 	timestamp = map->timestamp;
800 
801 	if (!vm_map_lookup_entry(map, addr, &entry)) {
802 		vm_map_unlock_read(map);
803 		return (ENOMEM);
804 	}
805 
806 	/*
807 	 * Do this on a map entry basis so that if the pages are not
808 	 * in the current processes address space, we can easily look
809 	 * up the pages elsewhere.
810 	 */
811 	lastvecindex = -1;
812 	for (current = entry;
813 	    (current != &map->header) && (current->start < end);
814 	    current = current->next) {
815 
816 		/*
817 		 * check for contiguity
818 		 */
819 		if (current->end < end &&
820 		    (entry->next == &map->header ||
821 		     current->next->start > current->end)) {
822 			vm_map_unlock_read(map);
823 			return (ENOMEM);
824 		}
825 
826 		/*
827 		 * ignore submaps (for now) or null objects
828 		 */
829 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
830 			current->object.vm_object == NULL)
831 			continue;
832 
833 		/*
834 		 * limit this scan to the current map entry and the
835 		 * limits for the mincore call
836 		 */
837 		if (addr < current->start)
838 			addr = current->start;
839 		cend = current->end;
840 		if (cend > end)
841 			cend = end;
842 
843 		/*
844 		 * scan this entry one page at a time
845 		 */
846 		while (addr < cend) {
847 			/*
848 			 * Check pmap first, it is likely faster, also
849 			 * it can provide info as to whether we are the
850 			 * one referencing or modifying the page.
851 			 */
852 			mincoreinfo = pmap_mincore(pmap, addr);
853 			if (!mincoreinfo) {
854 				vm_pindex_t pindex;
855 				vm_ooffset_t offset;
856 				vm_page_t m;
857 				/*
858 				 * calculate the page index into the object
859 				 */
860 				offset = current->offset + (addr - current->start);
861 				pindex = OFF_TO_IDX(offset);
862 				VM_OBJECT_LOCK(current->object.vm_object);
863 				m = vm_page_lookup(current->object.vm_object,
864 					pindex);
865 				/*
866 				 * if the page is resident, then gather information about
867 				 * it.
868 				 */
869 				if (m != NULL && m->valid != 0) {
870 					mincoreinfo = MINCORE_INCORE;
871 					vm_page_lock_queues();
872 					if (m->dirty ||
873 						pmap_is_modified(m))
874 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
875 					if ((m->flags & PG_REFERENCED) ||
876 						pmap_ts_referenced(m)) {
877 						vm_page_flag_set(m, PG_REFERENCED);
878 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
879 					}
880 					vm_page_unlock_queues();
881 				}
882 				VM_OBJECT_UNLOCK(current->object.vm_object);
883 			}
884 
885 			/*
886 			 * subyte may page fault.  In case it needs to modify
887 			 * the map, we release the lock.
888 			 */
889 			vm_map_unlock_read(map);
890 
891 			/*
892 			 * calculate index into user supplied byte vector
893 			 */
894 			vecindex = OFF_TO_IDX(addr - first_addr);
895 
896 			/*
897 			 * If we have skipped map entries, we need to make sure that
898 			 * the byte vector is zeroed for those skipped entries.
899 			 */
900 			while ((lastvecindex + 1) < vecindex) {
901 				error = subyte(vec + lastvecindex, 0);
902 				if (error) {
903 					error = EFAULT;
904 					goto done2;
905 				}
906 				++lastvecindex;
907 			}
908 
909 			/*
910 			 * Pass the page information to the user
911 			 */
912 			error = subyte(vec + vecindex, mincoreinfo);
913 			if (error) {
914 				error = EFAULT;
915 				goto done2;
916 			}
917 
918 			/*
919 			 * If the map has changed, due to the subyte, the previous
920 			 * output may be invalid.
921 			 */
922 			vm_map_lock_read(map);
923 			if (timestamp != map->timestamp)
924 				goto RestartScan;
925 
926 			lastvecindex = vecindex;
927 			addr += PAGE_SIZE;
928 		}
929 	}
930 
931 	/*
932 	 * subyte may page fault.  In case it needs to modify
933 	 * the map, we release the lock.
934 	 */
935 	vm_map_unlock_read(map);
936 
937 	/*
938 	 * Zero the last entries in the byte vector.
939 	 */
940 	vecindex = OFF_TO_IDX(end - first_addr);
941 	while ((lastvecindex + 1) < vecindex) {
942 		error = subyte(vec + lastvecindex, 0);
943 		if (error) {
944 			error = EFAULT;
945 			goto done2;
946 		}
947 		++lastvecindex;
948 	}
949 
950 	/*
951 	 * If the map has changed, due to the subyte, the previous
952 	 * output may be invalid.
953 	 */
954 	vm_map_lock_read(map);
955 	if (timestamp != map->timestamp)
956 		goto RestartScan;
957 	vm_map_unlock_read(map);
958 done2:
959 	return (error);
960 }
961 
962 #ifndef _SYS_SYSPROTO_H_
963 struct mlock_args {
964 	const void *addr;
965 	size_t len;
966 };
967 #endif
968 /*
969  * MPSAFE
970  */
971 int
972 mlock(td, uap)
973 	struct thread *td;
974 	struct mlock_args *uap;
975 {
976 	struct proc *proc;
977 	vm_offset_t addr, end, last, start;
978 	vm_size_t npages, size;
979 	int error;
980 
981 	error = priv_check(td, PRIV_VM_MLOCK);
982 	if (error)
983 		return (error);
984 	addr = (vm_offset_t)uap->addr;
985 	size = uap->len;
986 	last = addr + size;
987 	start = trunc_page(addr);
988 	end = round_page(last);
989 	if (last < addr || end < addr)
990 		return (EINVAL);
991 	npages = atop(end - start);
992 	if (npages > vm_page_max_wired)
993 		return (ENOMEM);
994 	proc = td->td_proc;
995 	PROC_LOCK(proc);
996 	if (ptoa(npages +
997 	    pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
998 	    lim_cur(proc, RLIMIT_MEMLOCK)) {
999 		PROC_UNLOCK(proc);
1000 		return (ENOMEM);
1001 	}
1002 	PROC_UNLOCK(proc);
1003 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1004 		return (EAGAIN);
1005 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1006 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1007 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1008 }
1009 
1010 #ifndef _SYS_SYSPROTO_H_
1011 struct mlockall_args {
1012 	int	how;
1013 };
1014 #endif
1015 
1016 /*
1017  * MPSAFE
1018  */
1019 int
1020 mlockall(td, uap)
1021 	struct thread *td;
1022 	struct mlockall_args *uap;
1023 {
1024 	vm_map_t map;
1025 	int error;
1026 
1027 	map = &td->td_proc->p_vmspace->vm_map;
1028 	error = 0;
1029 
1030 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1031 		return (EINVAL);
1032 
1033 #if 0
1034 	/*
1035 	 * If wiring all pages in the process would cause it to exceed
1036 	 * a hard resource limit, return ENOMEM.
1037 	 */
1038 	PROC_LOCK(td->td_proc);
1039 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1040 		lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1041 		PROC_UNLOCK(td->td_proc);
1042 		return (ENOMEM);
1043 	}
1044 	PROC_UNLOCK(td->td_proc);
1045 #else
1046 	error = priv_check(td, PRIV_VM_MLOCK);
1047 	if (error)
1048 		return (error);
1049 #endif
1050 
1051 	if (uap->how & MCL_FUTURE) {
1052 		vm_map_lock(map);
1053 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1054 		vm_map_unlock(map);
1055 		error = 0;
1056 	}
1057 
1058 	if (uap->how & MCL_CURRENT) {
1059 		/*
1060 		 * P1003.1-2001 mandates that all currently mapped pages
1061 		 * will be memory resident and locked (wired) upon return
1062 		 * from mlockall(). vm_map_wire() will wire pages, by
1063 		 * calling vm_fault_wire() for each page in the region.
1064 		 */
1065 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1066 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1067 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1068 	}
1069 
1070 	return (error);
1071 }
1072 
1073 #ifndef _SYS_SYSPROTO_H_
1074 struct munlockall_args {
1075 	register_t dummy;
1076 };
1077 #endif
1078 
1079 /*
1080  * MPSAFE
1081  */
1082 int
1083 munlockall(td, uap)
1084 	struct thread *td;
1085 	struct munlockall_args *uap;
1086 {
1087 	vm_map_t map;
1088 	int error;
1089 
1090 	map = &td->td_proc->p_vmspace->vm_map;
1091 	error = priv_check(td, PRIV_VM_MUNLOCK);
1092 	if (error)
1093 		return (error);
1094 
1095 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1096 	vm_map_lock(map);
1097 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1098 	vm_map_unlock(map);
1099 
1100 	/* Forcibly unwire all pages. */
1101 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1102 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1103 
1104 	return (error);
1105 }
1106 
1107 #ifndef _SYS_SYSPROTO_H_
1108 struct munlock_args {
1109 	const void *addr;
1110 	size_t len;
1111 };
1112 #endif
1113 /*
1114  * MPSAFE
1115  */
1116 int
1117 munlock(td, uap)
1118 	struct thread *td;
1119 	struct munlock_args *uap;
1120 {
1121 	vm_offset_t addr, end, last, start;
1122 	vm_size_t size;
1123 	int error;
1124 
1125 	error = priv_check(td, PRIV_VM_MUNLOCK);
1126 	if (error)
1127 		return (error);
1128 	addr = (vm_offset_t)uap->addr;
1129 	size = uap->len;
1130 	last = addr + size;
1131 	start = trunc_page(addr);
1132 	end = round_page(last);
1133 	if (last < addr || end < addr)
1134 		return (EINVAL);
1135 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1136 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1137 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1138 }
1139 
1140 /*
1141  * vm_mmap_vnode()
1142  *
1143  * MPSAFE
1144  *
1145  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1146  * operations on vnodes.
1147  */
1148 int
1149 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1150     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1151     struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1152 {
1153 	struct vattr va;
1154 	void *handle;
1155 	vm_object_t obj;
1156 	struct mount *mp;
1157 	struct cdevsw *dsw;
1158 	int error, flags, type;
1159 	int vfslocked;
1160 
1161 	mp = vp->v_mount;
1162 	vfslocked = VFS_LOCK_GIANT(mp);
1163 	if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1164 		VFS_UNLOCK_GIANT(vfslocked);
1165 		return (error);
1166 	}
1167 	flags = *flagsp;
1168 	obj = vp->v_object;
1169 	if (vp->v_type == VREG) {
1170 		/*
1171 		 * Get the proper underlying object
1172 		 */
1173 		if (obj == NULL) {
1174 			error = EINVAL;
1175 			goto done;
1176 		}
1177 		if (obj->handle != vp) {
1178 			vput(vp);
1179 			vp = (struct vnode*)obj->handle;
1180 			vget(vp, LK_EXCLUSIVE, td);
1181 		}
1182 		type = OBJT_VNODE;
1183 		handle = vp;
1184 	} else if (vp->v_type == VCHR) {
1185 		type = OBJT_DEVICE;
1186 		handle = vp->v_rdev;
1187 
1188 		dsw = dev_refthread(handle);
1189 		if (dsw == NULL) {
1190 			error = ENXIO;
1191 			goto done;
1192 		}
1193 		if (dsw->d_flags & D_MMAP_ANON) {
1194 			dev_relthread(handle);
1195 			*maxprotp = VM_PROT_ALL;
1196 			*flagsp |= MAP_ANON;
1197 			error = 0;
1198 			goto done;
1199 		}
1200 		dev_relthread(handle);
1201 		/*
1202 		 * cdevs does not provide private mappings of any kind.
1203 		 */
1204 		if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1205 		    (prot & PROT_WRITE) != 0) {
1206 			error = EACCES;
1207 			goto done;
1208 		}
1209 		if (flags & (MAP_PRIVATE|MAP_COPY)) {
1210 			error = EINVAL;
1211 			goto done;
1212 		}
1213 		/*
1214 		 * Force device mappings to be shared.
1215 		 */
1216 		flags |= MAP_SHARED;
1217 	} else {
1218 		error = EINVAL;
1219 		goto done;
1220 	}
1221 	if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1222 		goto done;
1223 	}
1224 #ifdef MAC
1225 	error = mac_vnode_check_mmap(td->td_ucred, vp, prot, flags);
1226 	if (error != 0)
1227 		goto done;
1228 #endif
1229 	if ((flags & MAP_SHARED) != 0) {
1230 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1231 			if (prot & PROT_WRITE) {
1232 				error = EPERM;
1233 				goto done;
1234 			}
1235 			*maxprotp &= ~VM_PROT_WRITE;
1236 		}
1237 	}
1238 	/*
1239 	 * If it is a regular file without any references
1240 	 * we do not need to sync it.
1241 	 * Adjust object size to be the size of actual file.
1242 	 */
1243 	if (vp->v_type == VREG) {
1244 		objsize = round_page(va.va_size);
1245 		if (va.va_nlink == 0)
1246 			flags |= MAP_NOSYNC;
1247 	}
1248 	obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1249 	if (obj == NULL) {
1250 		error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1251 		goto done;
1252 	}
1253 	*objp = obj;
1254 	*flagsp = flags;
1255 	vfs_mark_atime(vp, td);
1256 
1257 done:
1258 	vput(vp);
1259 	VFS_UNLOCK_GIANT(vfslocked);
1260 	return (error);
1261 }
1262 
1263 /*
1264  * vm_mmap_cdev()
1265  *
1266  * MPSAFE
1267  *
1268  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1269  * operations on cdevs.
1270  */
1271 int
1272 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1273     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1274     struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1275 {
1276 	vm_object_t obj;
1277 	struct cdevsw *dsw;
1278 	int flags;
1279 
1280 	flags = *flagsp;
1281 
1282 	dsw = dev_refthread(cdev);
1283 	if (dsw == NULL)
1284 		return (ENXIO);
1285 	if (dsw->d_flags & D_MMAP_ANON) {
1286 		dev_relthread(cdev);
1287 		*maxprotp = VM_PROT_ALL;
1288 		*flagsp |= MAP_ANON;
1289 		return (0);
1290 	}
1291 	dev_relthread(cdev);
1292 	/*
1293 	 * cdevs does not provide private mappings of any kind.
1294 	 */
1295 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1296 	    (prot & PROT_WRITE) != 0)
1297 		return (EACCES);
1298 	if (flags & (MAP_PRIVATE|MAP_COPY))
1299 		return (EINVAL);
1300 	/*
1301 	 * Force device mappings to be shared.
1302 	 */
1303 	flags |= MAP_SHARED;
1304 #ifdef MAC_XXX
1305 	error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1306 	if (error != 0)
1307 		return (error);
1308 #endif
1309 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1310 	if (obj == NULL)
1311 		return (EINVAL);
1312 	*objp = obj;
1313 	*flagsp = flags;
1314 	return (0);
1315 }
1316 
1317 /*
1318  * vm_mmap_shm()
1319  *
1320  * MPSAFE
1321  *
1322  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1323  * operations on shm file descriptors.
1324  */
1325 int
1326 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1327     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1328     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1329 {
1330 	int error;
1331 
1332 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1333 	    (prot & PROT_WRITE) != 0)
1334 		return (EACCES);
1335 #ifdef MAC
1336 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1337 	if (error != 0)
1338 		return (error);
1339 #endif
1340 	error = shm_mmap(shmfd, objsize, foff, objp);
1341 	if (error)
1342 		return (error);
1343 	return (0);
1344 }
1345 
1346 /*
1347  * vm_mmap()
1348  *
1349  * MPSAFE
1350  *
1351  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1352  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1353  */
1354 int
1355 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1356 	vm_prot_t maxprot, int flags,
1357 	objtype_t handle_type, void *handle,
1358 	vm_ooffset_t foff)
1359 {
1360 	boolean_t fitit;
1361 	vm_object_t object = NULL;
1362 	int rv = KERN_SUCCESS;
1363 	int docow, error;
1364 	struct thread *td = curthread;
1365 
1366 	if (size == 0)
1367 		return (0);
1368 
1369 	size = round_page(size);
1370 
1371 	PROC_LOCK(td->td_proc);
1372 	if (td->td_proc->p_vmspace->vm_map.size + size >
1373 	    lim_cur(td->td_proc, RLIMIT_VMEM)) {
1374 		PROC_UNLOCK(td->td_proc);
1375 		return(ENOMEM);
1376 	}
1377 	PROC_UNLOCK(td->td_proc);
1378 
1379 	/*
1380 	 * We currently can only deal with page aligned file offsets.
1381 	 * The check is here rather than in the syscall because the
1382 	 * kernel calls this function internally for other mmaping
1383 	 * operations (such as in exec) and non-aligned offsets will
1384 	 * cause pmap inconsistencies...so we want to be sure to
1385 	 * disallow this in all cases.
1386 	 */
1387 	if (foff & PAGE_MASK)
1388 		return (EINVAL);
1389 
1390 	if ((flags & MAP_FIXED) == 0) {
1391 		fitit = TRUE;
1392 		*addr = round_page(*addr);
1393 	} else {
1394 		if (*addr != trunc_page(*addr))
1395 			return (EINVAL);
1396 		fitit = FALSE;
1397 	}
1398 	/*
1399 	 * Lookup/allocate object.
1400 	 */
1401 	switch (handle_type) {
1402 	case OBJT_DEVICE:
1403 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1404 		    handle, foff, &object);
1405 		break;
1406 	case OBJT_VNODE:
1407 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1408 		    handle, foff, &object);
1409 		break;
1410 	case OBJT_SWAP:
1411 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1412 		    handle, foff, &object);
1413 		break;
1414 	case OBJT_DEFAULT:
1415 		if (handle == NULL) {
1416 			error = 0;
1417 			break;
1418 		}
1419 		/* FALLTHROUGH */
1420 	default:
1421 		error = EINVAL;
1422 		break;
1423 	}
1424 	if (error)
1425 		return (error);
1426 	if (flags & MAP_ANON) {
1427 		object = NULL;
1428 		docow = 0;
1429 		/*
1430 		 * Unnamed anonymous regions always start at 0.
1431 		 */
1432 		if (handle == 0)
1433 			foff = 0;
1434 	} else {
1435 		docow = MAP_PREFAULT_PARTIAL;
1436 	}
1437 
1438 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1439 		docow |= MAP_COPY_ON_WRITE;
1440 	if (flags & MAP_NOSYNC)
1441 		docow |= MAP_DISABLE_SYNCER;
1442 	if (flags & MAP_NOCORE)
1443 		docow |= MAP_DISABLE_COREDUMP;
1444 
1445 #if defined(VM_PROT_READ_IS_EXEC)
1446 	if (prot & VM_PROT_READ)
1447 		prot |= VM_PROT_EXECUTE;
1448 
1449 	if (maxprot & VM_PROT_READ)
1450 		maxprot |= VM_PROT_EXECUTE;
1451 #endif
1452 
1453 	if (flags & MAP_STACK)
1454 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1455 		    docow | MAP_STACK_GROWS_DOWN);
1456 	else if (fitit)
1457 		rv = vm_map_find(map, object, foff, addr, size,
1458 		    object != NULL && object->type == OBJT_DEVICE ?
1459 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1460 	else
1461 		rv = vm_map_fixed(map, object, foff, *addr, size,
1462 				 prot, maxprot, docow);
1463 
1464 	if (rv != KERN_SUCCESS) {
1465 		/*
1466 		 * Lose the object reference. Will destroy the
1467 		 * object if it's an unnamed anonymous mapping
1468 		 * or named anonymous without other references.
1469 		 */
1470 		vm_object_deallocate(object);
1471 	} else if (flags & MAP_SHARED) {
1472 		/*
1473 		 * Shared memory is also shared with children.
1474 		 */
1475 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1476 		if (rv != KERN_SUCCESS)
1477 			(void) vm_map_remove(map, *addr, *addr + size);
1478 	}
1479 
1480 	/*
1481 	 * If the process has requested that all future mappings
1482 	 * be wired, then heed this.
1483 	 */
1484 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1485 		vm_map_wire(map, *addr, *addr + size,
1486 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1487 
1488 	switch (rv) {
1489 	case KERN_SUCCESS:
1490 		return (0);
1491 	case KERN_INVALID_ADDRESS:
1492 	case KERN_NO_SPACE:
1493 		return (ENOMEM);
1494 	case KERN_PROTECTION_FAILURE:
1495 		return (EACCES);
1496 	default:
1497 		return (EINVAL);
1498 	}
1499 }
1500