xref: /freebsd/sys/vm/vm_mmap.c (revision 1670a1c2a47d10ecccd001970b859caf93cd3b6e)
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 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/sysproto.h>
55 #include <sys/filedesc.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/resource.h>
59 #include <sys/resourcevar.h>
60 #include <sys/vnode.h>
61 #include <sys/fcntl.h>
62 #include <sys/file.h>
63 #include <sys/mman.h>
64 #include <sys/mount.h>
65 #include <sys/conf.h>
66 #include <sys/stat.h>
67 #include <sys/sysent.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 #ifndef _SYS_SYSPROTO_H_
192 struct mmap_args {
193 	void *addr;
194 	size_t len;
195 	int prot;
196 	int flags;
197 	int fd;
198 	long pad;
199 	off_t pos;
200 };
201 #endif
202 
203 /*
204  * MPSAFE
205  */
206 int
207 mmap(td, uap)
208 	struct thread *td;
209 	struct mmap_args *uap;
210 {
211 #ifdef HWPMC_HOOKS
212 	struct pmckern_map_in pkm;
213 #endif
214 	struct file *fp;
215 	struct vnode *vp;
216 	vm_offset_t addr;
217 	vm_size_t size, pageoff;
218 	vm_prot_t prot, maxprot;
219 	void *handle;
220 	objtype_t handle_type;
221 	int flags, error;
222 	off_t pos;
223 	struct vmspace *vms = td->td_proc->p_vmspace;
224 
225 	addr = (vm_offset_t) uap->addr;
226 	size = uap->len;
227 	prot = uap->prot & VM_PROT_ALL;
228 	flags = uap->flags;
229 	pos = uap->pos;
230 
231 	fp = NULL;
232 
233 	/* Make sure mapping fits into numeric range, etc. */
234 	if ((uap->len == 0 && !SV_CURPROC_FLAG(SV_AOUT) &&
235 	     curproc->p_osrel >= 800104) ||
236 	    ((flags & MAP_ANON) && (uap->fd != -1 || pos != 0)))
237 		return (EINVAL);
238 
239 	if (flags & MAP_STACK) {
240 		if ((uap->fd != -1) ||
241 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
242 			return (EINVAL);
243 		flags |= MAP_ANON;
244 		pos = 0;
245 	}
246 
247 	/*
248 	 * Align the file position to a page boundary,
249 	 * and save its page offset component.
250 	 */
251 	pageoff = (pos & PAGE_MASK);
252 	pos -= pageoff;
253 
254 	/* Adjust size for rounding (on both ends). */
255 	size += pageoff;			/* low end... */
256 	size = (vm_size_t) round_page(size);	/* hi end */
257 
258 	/*
259 	 * Check for illegal addresses.  Watch out for address wrap... Note
260 	 * that VM_*_ADDRESS are not constants due to casts (argh).
261 	 */
262 	if (flags & MAP_FIXED) {
263 		/*
264 		 * The specified address must have the same remainder
265 		 * as the file offset taken modulo PAGE_SIZE, so it
266 		 * should be aligned after adjustment by pageoff.
267 		 */
268 		addr -= pageoff;
269 		if (addr & PAGE_MASK)
270 			return (EINVAL);
271 
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 	} else {
304 		/*
305 		 * Mapping file, get fp for validation and
306 		 * don't let the descriptor disappear on us if we block.
307 		 */
308 		if ((error = fget(td, uap->fd, &fp)) != 0)
309 			goto done;
310 		if (fp->f_type == DTYPE_SHM) {
311 			handle = fp->f_data;
312 			handle_type = OBJT_SWAP;
313 			maxprot = VM_PROT_NONE;
314 
315 			/* FREAD should always be set. */
316 			if (fp->f_flag & FREAD)
317 				maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
318 			if (fp->f_flag & FWRITE)
319 				maxprot |= VM_PROT_WRITE;
320 			goto map;
321 		}
322 		if (fp->f_type != DTYPE_VNODE) {
323 			error = ENODEV;
324 			goto done;
325 		}
326 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
327     defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
328 		/*
329 		 * POSIX shared-memory objects are defined to have
330 		 * kernel persistence, and are not defined to support
331 		 * read(2)/write(2) -- or even open(2).  Thus, we can
332 		 * use MAP_ASYNC to trade on-disk coherence for speed.
333 		 * The shm_open(3) library routine turns on the FPOSIXSHM
334 		 * flag to request this behavior.
335 		 */
336 		if (fp->f_flag & FPOSIXSHM)
337 			flags |= MAP_NOSYNC;
338 #endif
339 		vp = fp->f_vnode;
340 		/*
341 		 * Ensure that file and memory protections are
342 		 * compatible.  Note that we only worry about
343 		 * writability if mapping is shared; in this case,
344 		 * current and max prot are dictated by the open file.
345 		 * XXX use the vnode instead?  Problem is: what
346 		 * credentials do we use for determination? What if
347 		 * proc does a setuid?
348 		 */
349 		if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
350 			maxprot = VM_PROT_NONE;
351 		else
352 			maxprot = VM_PROT_EXECUTE;
353 		if (fp->f_flag & FREAD) {
354 			maxprot |= VM_PROT_READ;
355 		} else if (prot & PROT_READ) {
356 			error = EACCES;
357 			goto done;
358 		}
359 		/*
360 		 * If we are sharing potential changes (either via
361 		 * MAP_SHARED or via the implicit sharing of character
362 		 * device mappings), and we are trying to get write
363 		 * permission although we opened it without asking
364 		 * for it, bail out.
365 		 */
366 		if ((flags & MAP_SHARED) != 0) {
367 			if ((fp->f_flag & FWRITE) != 0) {
368 				maxprot |= VM_PROT_WRITE;
369 			} else if ((prot & PROT_WRITE) != 0) {
370 				error = EACCES;
371 				goto done;
372 			}
373 		} else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
374 			maxprot |= VM_PROT_WRITE;
375 		}
376 		handle = (void *)vp;
377 		handle_type = OBJT_VNODE;
378 	}
379 map:
380 
381 	/*
382 	 * Do not allow more then a certain number of vm_map_entry structures
383 	 * per process.  Scale with the number of rforks sharing the map
384 	 * to make the limit reasonable for threads.
385 	 */
386 	if (max_proc_mmap &&
387 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
388 		error = ENOMEM;
389 		goto done;
390 	}
391 
392 	td->td_fpop = fp;
393 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
394 	    flags, handle_type, handle, pos);
395 	td->td_fpop = NULL;
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 
626 	pageoff = (addr & PAGE_MASK);
627 	addr -= pageoff;
628 	size += pageoff;
629 	size = (vm_size_t) round_page(size);
630 	if (addr + size < addr)
631 		return (EINVAL);
632 
633 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
634 	    addr + size, prot, FALSE)) {
635 	case KERN_SUCCESS:
636 		return (0);
637 	case KERN_PROTECTION_FAILURE:
638 		return (EACCES);
639 	case KERN_RESOURCE_SHORTAGE:
640 		return (ENOMEM);
641 	}
642 	return (EINVAL);
643 }
644 
645 #ifndef _SYS_SYSPROTO_H_
646 struct minherit_args {
647 	void *addr;
648 	size_t len;
649 	int inherit;
650 };
651 #endif
652 /*
653  * MPSAFE
654  */
655 int
656 minherit(td, uap)
657 	struct thread *td;
658 	struct minherit_args *uap;
659 {
660 	vm_offset_t addr;
661 	vm_size_t size, pageoff;
662 	vm_inherit_t inherit;
663 
664 	addr = (vm_offset_t)uap->addr;
665 	size = uap->len;
666 	inherit = uap->inherit;
667 
668 	pageoff = (addr & PAGE_MASK);
669 	addr -= pageoff;
670 	size += pageoff;
671 	size = (vm_size_t) round_page(size);
672 	if (addr + size < addr)
673 		return (EINVAL);
674 
675 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
676 	    addr + size, inherit)) {
677 	case KERN_SUCCESS:
678 		return (0);
679 	case KERN_PROTECTION_FAILURE:
680 		return (EACCES);
681 	}
682 	return (EINVAL);
683 }
684 
685 #ifndef _SYS_SYSPROTO_H_
686 struct madvise_args {
687 	void *addr;
688 	size_t len;
689 	int behav;
690 };
691 #endif
692 
693 /*
694  * MPSAFE
695  */
696 /* ARGSUSED */
697 int
698 madvise(td, uap)
699 	struct thread *td;
700 	struct madvise_args *uap;
701 {
702 	vm_offset_t start, end;
703 	vm_map_t map;
704 	struct proc *p;
705 	int error;
706 
707 	/*
708 	 * Check for our special case, advising the swap pager we are
709 	 * "immortal."
710 	 */
711 	if (uap->behav == MADV_PROTECT) {
712 		error = priv_check(td, PRIV_VM_MADV_PROTECT);
713 		if (error == 0) {
714 			p = td->td_proc;
715 			PROC_LOCK(p);
716 			p->p_flag |= P_PROTECTED;
717 			PROC_UNLOCK(p);
718 		}
719 		return (error);
720 	}
721 	/*
722 	 * Check for illegal behavior
723 	 */
724 	if (uap->behav < 0 || uap->behav > MADV_CORE)
725 		return (EINVAL);
726 	/*
727 	 * Check for illegal addresses.  Watch out for address wrap... Note
728 	 * that VM_*_ADDRESS are not constants due to casts (argh).
729 	 */
730 	map = &td->td_proc->p_vmspace->vm_map;
731 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
732 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
733 		return (EINVAL);
734 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
735 		return (EINVAL);
736 
737 	/*
738 	 * Since this routine is only advisory, we default to conservative
739 	 * behavior.
740 	 */
741 	start = trunc_page((vm_offset_t) uap->addr);
742 	end = round_page((vm_offset_t) uap->addr + uap->len);
743 
744 	if (vm_map_madvise(map, start, end, uap->behav))
745 		return (EINVAL);
746 	return (0);
747 }
748 
749 #ifndef _SYS_SYSPROTO_H_
750 struct mincore_args {
751 	const void *addr;
752 	size_t len;
753 	char *vec;
754 };
755 #endif
756 
757 /*
758  * MPSAFE
759  */
760 /* ARGSUSED */
761 int
762 mincore(td, uap)
763 	struct thread *td;
764 	struct mincore_args *uap;
765 {
766 	vm_offset_t addr, first_addr;
767 	vm_offset_t end, cend;
768 	pmap_t pmap;
769 	vm_map_t map;
770 	char *vec;
771 	int error = 0;
772 	int vecindex, lastvecindex;
773 	vm_map_entry_t current;
774 	vm_map_entry_t entry;
775 	vm_object_t object;
776 	vm_paddr_t locked_pa;
777 	vm_page_t m;
778 	vm_pindex_t pindex;
779 	int mincoreinfo;
780 	unsigned int timestamp;
781 	boolean_t locked;
782 
783 	/*
784 	 * Make sure that the addresses presented are valid for user
785 	 * mode.
786 	 */
787 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
788 	end = addr + (vm_size_t)round_page(uap->len);
789 	map = &td->td_proc->p_vmspace->vm_map;
790 	if (end > vm_map_max(map) || end < addr)
791 		return (ENOMEM);
792 
793 	/*
794 	 * Address of byte vector
795 	 */
796 	vec = uap->vec;
797 
798 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
799 
800 	vm_map_lock_read(map);
801 RestartScan:
802 	timestamp = map->timestamp;
803 
804 	if (!vm_map_lookup_entry(map, addr, &entry)) {
805 		vm_map_unlock_read(map);
806 		return (ENOMEM);
807 	}
808 
809 	/*
810 	 * Do this on a map entry basis so that if the pages are not
811 	 * in the current processes address space, we can easily look
812 	 * up the pages elsewhere.
813 	 */
814 	lastvecindex = -1;
815 	for (current = entry;
816 	    (current != &map->header) && (current->start < end);
817 	    current = current->next) {
818 
819 		/*
820 		 * check for contiguity
821 		 */
822 		if (current->end < end &&
823 		    (entry->next == &map->header ||
824 		     current->next->start > current->end)) {
825 			vm_map_unlock_read(map);
826 			return (ENOMEM);
827 		}
828 
829 		/*
830 		 * ignore submaps (for now) or null objects
831 		 */
832 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
833 			current->object.vm_object == NULL)
834 			continue;
835 
836 		/*
837 		 * limit this scan to the current map entry and the
838 		 * limits for the mincore call
839 		 */
840 		if (addr < current->start)
841 			addr = current->start;
842 		cend = current->end;
843 		if (cend > end)
844 			cend = end;
845 
846 		/*
847 		 * scan this entry one page at a time
848 		 */
849 		while (addr < cend) {
850 			/*
851 			 * Check pmap first, it is likely faster, also
852 			 * it can provide info as to whether we are the
853 			 * one referencing or modifying the page.
854 			 */
855 			object = NULL;
856 			locked_pa = 0;
857 		retry:
858 			m = NULL;
859 			mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
860 			if (locked_pa != 0) {
861 				/*
862 				 * The page is mapped by this process but not
863 				 * both accessed and modified.  It is also
864 				 * managed.  Acquire the object lock so that
865 				 * other mappings might be examined.
866 				 */
867 				m = PHYS_TO_VM_PAGE(locked_pa);
868 				if (m->object != object) {
869 					if (object != NULL)
870 						VM_OBJECT_UNLOCK(object);
871 					object = m->object;
872 					locked = VM_OBJECT_TRYLOCK(object);
873 					vm_page_unlock(m);
874 					if (!locked) {
875 						VM_OBJECT_LOCK(object);
876 						vm_page_lock(m);
877 						goto retry;
878 					}
879 				} else
880 					vm_page_unlock(m);
881 				KASSERT(m->valid == VM_PAGE_BITS_ALL,
882 				    ("mincore: page %p is mapped but invalid",
883 				    m));
884 			} else if (mincoreinfo == 0) {
885 				/*
886 				 * The page is not mapped by this process.  If
887 				 * the object implements managed pages, then
888 				 * determine if the page is resident so that
889 				 * the mappings might be examined.
890 				 */
891 				if (current->object.vm_object != object) {
892 					if (object != NULL)
893 						VM_OBJECT_UNLOCK(object);
894 					object = current->object.vm_object;
895 					VM_OBJECT_LOCK(object);
896 				}
897 				if (object->type == OBJT_DEFAULT ||
898 				    object->type == OBJT_SWAP ||
899 				    object->type == OBJT_VNODE) {
900 					pindex = OFF_TO_IDX(current->offset +
901 					    (addr - current->start));
902 					m = vm_page_lookup(object, pindex);
903 					if (m != NULL && m->valid == 0)
904 						m = NULL;
905 					if (m != NULL)
906 						mincoreinfo = MINCORE_INCORE;
907 				}
908 			}
909 			if (m != NULL) {
910 				/* Examine other mappings to the page. */
911 				if (m->dirty == 0 && pmap_is_modified(m))
912 					vm_page_dirty(m);
913 				if (m->dirty != 0)
914 					mincoreinfo |= MINCORE_MODIFIED_OTHER;
915 				/*
916 				 * The first test for PG_REFERENCED is an
917 				 * optimization.  The second test is
918 				 * required because a concurrent pmap
919 				 * operation could clear the last reference
920 				 * and set PG_REFERENCED before the call to
921 				 * pmap_is_referenced().
922 				 */
923 				if ((m->flags & PG_REFERENCED) != 0 ||
924 				    pmap_is_referenced(m) ||
925 				    (m->flags & PG_REFERENCED) != 0)
926 					mincoreinfo |= MINCORE_REFERENCED_OTHER;
927 			}
928 			if (object != NULL)
929 				VM_OBJECT_UNLOCK(object);
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 			 * calculate index into user supplied byte vector
939 			 */
940 			vecindex = OFF_TO_IDX(addr - first_addr);
941 
942 			/*
943 			 * If we have skipped map entries, we need to make sure that
944 			 * the byte vector is zeroed for those skipped entries.
945 			 */
946 			while ((lastvecindex + 1) < vecindex) {
947 				error = subyte(vec + lastvecindex, 0);
948 				if (error) {
949 					error = EFAULT;
950 					goto done2;
951 				}
952 				++lastvecindex;
953 			}
954 
955 			/*
956 			 * Pass the page information to the user
957 			 */
958 			error = subyte(vec + vecindex, mincoreinfo);
959 			if (error) {
960 				error = EFAULT;
961 				goto done2;
962 			}
963 
964 			/*
965 			 * If the map has changed, due to the subyte, the previous
966 			 * output may be invalid.
967 			 */
968 			vm_map_lock_read(map);
969 			if (timestamp != map->timestamp)
970 				goto RestartScan;
971 
972 			lastvecindex = vecindex;
973 			addr += PAGE_SIZE;
974 		}
975 	}
976 
977 	/*
978 	 * subyte may page fault.  In case it needs to modify
979 	 * the map, we release the lock.
980 	 */
981 	vm_map_unlock_read(map);
982 
983 	/*
984 	 * Zero the last entries in the byte vector.
985 	 */
986 	vecindex = OFF_TO_IDX(end - first_addr);
987 	while ((lastvecindex + 1) < vecindex) {
988 		error = subyte(vec + lastvecindex, 0);
989 		if (error) {
990 			error = EFAULT;
991 			goto done2;
992 		}
993 		++lastvecindex;
994 	}
995 
996 	/*
997 	 * If the map has changed, due to the subyte, the previous
998 	 * output may be invalid.
999 	 */
1000 	vm_map_lock_read(map);
1001 	if (timestamp != map->timestamp)
1002 		goto RestartScan;
1003 	vm_map_unlock_read(map);
1004 done2:
1005 	return (error);
1006 }
1007 
1008 #ifndef _SYS_SYSPROTO_H_
1009 struct mlock_args {
1010 	const void *addr;
1011 	size_t len;
1012 };
1013 #endif
1014 /*
1015  * MPSAFE
1016  */
1017 int
1018 mlock(td, uap)
1019 	struct thread *td;
1020 	struct mlock_args *uap;
1021 {
1022 	struct proc *proc;
1023 	vm_offset_t addr, end, last, start;
1024 	vm_size_t npages, size;
1025 	int error;
1026 
1027 	error = priv_check(td, PRIV_VM_MLOCK);
1028 	if (error)
1029 		return (error);
1030 	addr = (vm_offset_t)uap->addr;
1031 	size = uap->len;
1032 	last = addr + size;
1033 	start = trunc_page(addr);
1034 	end = round_page(last);
1035 	if (last < addr || end < addr)
1036 		return (EINVAL);
1037 	npages = atop(end - start);
1038 	if (npages > vm_page_max_wired)
1039 		return (ENOMEM);
1040 	proc = td->td_proc;
1041 	PROC_LOCK(proc);
1042 	if (ptoa(npages +
1043 	    pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
1044 	    lim_cur(proc, RLIMIT_MEMLOCK)) {
1045 		PROC_UNLOCK(proc);
1046 		return (ENOMEM);
1047 	}
1048 	PROC_UNLOCK(proc);
1049 	if (npages + cnt.v_wire_count > vm_page_max_wired)
1050 		return (EAGAIN);
1051 	error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1052 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1053 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1054 }
1055 
1056 #ifndef _SYS_SYSPROTO_H_
1057 struct mlockall_args {
1058 	int	how;
1059 };
1060 #endif
1061 
1062 /*
1063  * MPSAFE
1064  */
1065 int
1066 mlockall(td, uap)
1067 	struct thread *td;
1068 	struct mlockall_args *uap;
1069 {
1070 	vm_map_t map;
1071 	int error;
1072 
1073 	map = &td->td_proc->p_vmspace->vm_map;
1074 	error = 0;
1075 
1076 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1077 		return (EINVAL);
1078 
1079 #if 0
1080 	/*
1081 	 * If wiring all pages in the process would cause it to exceed
1082 	 * a hard resource limit, return ENOMEM.
1083 	 */
1084 	PROC_LOCK(td->td_proc);
1085 	if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1086 		PROC_UNLOCK(td->td_proc);
1087 		return (ENOMEM);
1088 	}
1089 	PROC_UNLOCK(td->td_proc);
1090 #else
1091 	error = priv_check(td, PRIV_VM_MLOCK);
1092 	if (error)
1093 		return (error);
1094 #endif
1095 
1096 	if (uap->how & MCL_FUTURE) {
1097 		vm_map_lock(map);
1098 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1099 		vm_map_unlock(map);
1100 		error = 0;
1101 	}
1102 
1103 	if (uap->how & MCL_CURRENT) {
1104 		/*
1105 		 * P1003.1-2001 mandates that all currently mapped pages
1106 		 * will be memory resident and locked (wired) upon return
1107 		 * from mlockall(). vm_map_wire() will wire pages, by
1108 		 * calling vm_fault_wire() for each page in the region.
1109 		 */
1110 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1111 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1112 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1113 	}
1114 
1115 	return (error);
1116 }
1117 
1118 #ifndef _SYS_SYSPROTO_H_
1119 struct munlockall_args {
1120 	register_t dummy;
1121 };
1122 #endif
1123 
1124 /*
1125  * MPSAFE
1126  */
1127 int
1128 munlockall(td, uap)
1129 	struct thread *td;
1130 	struct munlockall_args *uap;
1131 {
1132 	vm_map_t map;
1133 	int error;
1134 
1135 	map = &td->td_proc->p_vmspace->vm_map;
1136 	error = priv_check(td, PRIV_VM_MUNLOCK);
1137 	if (error)
1138 		return (error);
1139 
1140 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1141 	vm_map_lock(map);
1142 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1143 	vm_map_unlock(map);
1144 
1145 	/* Forcibly unwire all pages. */
1146 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1147 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1148 
1149 	return (error);
1150 }
1151 
1152 #ifndef _SYS_SYSPROTO_H_
1153 struct munlock_args {
1154 	const void *addr;
1155 	size_t len;
1156 };
1157 #endif
1158 /*
1159  * MPSAFE
1160  */
1161 int
1162 munlock(td, uap)
1163 	struct thread *td;
1164 	struct munlock_args *uap;
1165 {
1166 	vm_offset_t addr, end, last, start;
1167 	vm_size_t size;
1168 	int error;
1169 
1170 	error = priv_check(td, PRIV_VM_MUNLOCK);
1171 	if (error)
1172 		return (error);
1173 	addr = (vm_offset_t)uap->addr;
1174 	size = uap->len;
1175 	last = addr + size;
1176 	start = trunc_page(addr);
1177 	end = round_page(last);
1178 	if (last < addr || end < addr)
1179 		return (EINVAL);
1180 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1181 	    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1182 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1183 }
1184 
1185 /*
1186  * vm_mmap_vnode()
1187  *
1188  * MPSAFE
1189  *
1190  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1191  * operations on vnodes.
1192  */
1193 int
1194 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1195     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1196     struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp)
1197 {
1198 	struct vattr va;
1199 	vm_object_t obj;
1200 	vm_offset_t foff;
1201 	struct mount *mp;
1202 	struct ucred *cred;
1203 	int error, flags;
1204 	int vfslocked;
1205 
1206 	mp = vp->v_mount;
1207 	cred = td->td_ucred;
1208 	vfslocked = VFS_LOCK_GIANT(mp);
1209 	if ((error = vget(vp, LK_SHARED, td)) != 0) {
1210 		VFS_UNLOCK_GIANT(vfslocked);
1211 		return (error);
1212 	}
1213 	foff = *foffp;
1214 	flags = *flagsp;
1215 	obj = vp->v_object;
1216 	if (vp->v_type == VREG) {
1217 		/*
1218 		 * Get the proper underlying object
1219 		 */
1220 		if (obj == NULL) {
1221 			error = EINVAL;
1222 			goto done;
1223 		}
1224 		if (obj->handle != vp) {
1225 			vput(vp);
1226 			vp = (struct vnode*)obj->handle;
1227 			vget(vp, LK_SHARED, td);
1228 		}
1229 	} else if (vp->v_type == VCHR) {
1230 		error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1231 		    vp->v_rdev, foffp, objp);
1232 		if (error == 0)
1233 			goto mark_atime;
1234 		goto done;
1235 	} else {
1236 		error = EINVAL;
1237 		goto done;
1238 	}
1239 	if ((error = VOP_GETATTR(vp, &va, cred)))
1240 		goto done;
1241 #ifdef MAC
1242 	error = mac_vnode_check_mmap(cred, vp, prot, flags);
1243 	if (error != 0)
1244 		goto done;
1245 #endif
1246 	if ((flags & MAP_SHARED) != 0) {
1247 		if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1248 			if (prot & PROT_WRITE) {
1249 				error = EPERM;
1250 				goto done;
1251 			}
1252 			*maxprotp &= ~VM_PROT_WRITE;
1253 		}
1254 	}
1255 	/*
1256 	 * If it is a regular file without any references
1257 	 * we do not need to sync it.
1258 	 * Adjust object size to be the size of actual file.
1259 	 */
1260 	objsize = round_page(va.va_size);
1261 	if (va.va_nlink == 0)
1262 		flags |= MAP_NOSYNC;
1263 	obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, td->td_ucred);
1264 	if (obj == NULL) {
1265 		error = ENOMEM;
1266 		goto done;
1267 	}
1268 	*objp = obj;
1269 	*flagsp = flags;
1270 
1271 mark_atime:
1272 	vfs_mark_atime(vp, cred);
1273 
1274 done:
1275 	vput(vp);
1276 	VFS_UNLOCK_GIANT(vfslocked);
1277 	return (error);
1278 }
1279 
1280 /*
1281  * vm_mmap_cdev()
1282  *
1283  * MPSAFE
1284  *
1285  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1286  * operations on cdevs.
1287  */
1288 int
1289 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1290     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1291     struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1292 {
1293 	vm_object_t obj;
1294 	struct cdevsw *dsw;
1295 	int error, flags;
1296 
1297 	flags = *flagsp;
1298 
1299 	dsw = dev_refthread(cdev);
1300 	if (dsw == NULL)
1301 		return (ENXIO);
1302 	if (dsw->d_flags & D_MMAP_ANON) {
1303 		dev_relthread(cdev);
1304 		*maxprotp = VM_PROT_ALL;
1305 		*flagsp |= MAP_ANON;
1306 		return (0);
1307 	}
1308 	/*
1309 	 * cdevs do not provide private mappings of any kind.
1310 	 */
1311 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1312 	    (prot & PROT_WRITE) != 0) {
1313 		dev_relthread(cdev);
1314 		return (EACCES);
1315 	}
1316 	if (flags & (MAP_PRIVATE|MAP_COPY)) {
1317 		dev_relthread(cdev);
1318 		return (EINVAL);
1319 	}
1320 	/*
1321 	 * Force device mappings to be shared.
1322 	 */
1323 	flags |= MAP_SHARED;
1324 #ifdef MAC_XXX
1325 	error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1326 	if (error != 0) {
1327 		dev_relthread(cdev);
1328 		return (error);
1329 	}
1330 #endif
1331 	/*
1332 	 * First, try d_mmap_single().  If that is not implemented
1333 	 * (returns ENODEV), fall back to using the device pager.
1334 	 * Note that d_mmap_single() must return a reference to the
1335 	 * object (it needs to bump the reference count of the object
1336 	 * it returns somehow).
1337 	 *
1338 	 * XXX assumes VM_PROT_* == PROT_*
1339 	 */
1340 	error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1341 	dev_relthread(cdev);
1342 	if (error != ENODEV)
1343 		return (error);
1344 	obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1345 	    td->td_ucred);
1346 	if (obj == NULL)
1347 		return (EINVAL);
1348 	*objp = obj;
1349 	*flagsp = flags;
1350 	return (0);
1351 }
1352 
1353 /*
1354  * vm_mmap_shm()
1355  *
1356  * MPSAFE
1357  *
1358  * Helper function for vm_mmap.  Perform sanity check specific for mmap
1359  * operations on shm file descriptors.
1360  */
1361 int
1362 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1363     vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1364     struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1365 {
1366 	int error;
1367 
1368 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1369 	    (prot & PROT_WRITE) != 0)
1370 		return (EACCES);
1371 #ifdef MAC
1372 	error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1373 	if (error != 0)
1374 		return (error);
1375 #endif
1376 	error = shm_mmap(shmfd, objsize, foff, objp);
1377 	if (error)
1378 		return (error);
1379 	return (0);
1380 }
1381 
1382 /*
1383  * vm_mmap()
1384  *
1385  * MPSAFE
1386  *
1387  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1388  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1389  */
1390 int
1391 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1392 	vm_prot_t maxprot, int flags,
1393 	objtype_t handle_type, void *handle,
1394 	vm_ooffset_t foff)
1395 {
1396 	boolean_t fitit;
1397 	vm_object_t object = NULL;
1398 	int rv = KERN_SUCCESS;
1399 	int docow, error;
1400 	struct thread *td = curthread;
1401 
1402 	if (size == 0)
1403 		return (0);
1404 
1405 	size = round_page(size);
1406 
1407 	PROC_LOCK(td->td_proc);
1408 	if (td->td_proc->p_vmspace->vm_map.size + size >
1409 	    lim_cur(td->td_proc, RLIMIT_VMEM)) {
1410 		PROC_UNLOCK(td->td_proc);
1411 		return(ENOMEM);
1412 	}
1413 	PROC_UNLOCK(td->td_proc);
1414 
1415 	/*
1416 	 * We currently can only deal with page aligned file offsets.
1417 	 * The check is here rather than in the syscall because the
1418 	 * kernel calls this function internally for other mmaping
1419 	 * operations (such as in exec) and non-aligned offsets will
1420 	 * cause pmap inconsistencies...so we want to be sure to
1421 	 * disallow this in all cases.
1422 	 */
1423 	if (foff & PAGE_MASK)
1424 		return (EINVAL);
1425 
1426 	if ((flags & MAP_FIXED) == 0) {
1427 		fitit = TRUE;
1428 		*addr = round_page(*addr);
1429 	} else {
1430 		if (*addr != trunc_page(*addr))
1431 			return (EINVAL);
1432 		fitit = FALSE;
1433 	}
1434 	/*
1435 	 * Lookup/allocate object.
1436 	 */
1437 	switch (handle_type) {
1438 	case OBJT_DEVICE:
1439 		error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1440 		    handle, &foff, &object);
1441 		break;
1442 	case OBJT_VNODE:
1443 		error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1444 		    handle, &foff, &object);
1445 		break;
1446 	case OBJT_SWAP:
1447 		error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1448 		    handle, foff, &object);
1449 		break;
1450 	case OBJT_DEFAULT:
1451 		if (handle == NULL) {
1452 			error = 0;
1453 			break;
1454 		}
1455 		/* FALLTHROUGH */
1456 	default:
1457 		error = EINVAL;
1458 		break;
1459 	}
1460 	if (error)
1461 		return (error);
1462 	if (flags & MAP_ANON) {
1463 		object = NULL;
1464 		docow = 0;
1465 		/*
1466 		 * Unnamed anonymous regions always start at 0.
1467 		 */
1468 		if (handle == 0)
1469 			foff = 0;
1470 	} else {
1471 		docow = MAP_PREFAULT_PARTIAL;
1472 	}
1473 
1474 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1475 		docow |= MAP_COPY_ON_WRITE;
1476 	if (flags & MAP_NOSYNC)
1477 		docow |= MAP_DISABLE_SYNCER;
1478 	if (flags & MAP_NOCORE)
1479 		docow |= MAP_DISABLE_COREDUMP;
1480 
1481 	if (flags & MAP_STACK)
1482 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1483 		    docow | MAP_STACK_GROWS_DOWN);
1484 	else if (fitit)
1485 		rv = vm_map_find(map, object, foff, addr, size,
1486 		    object != NULL && object->type == OBJT_DEVICE ?
1487 		    VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1488 	else
1489 		rv = vm_map_fixed(map, object, foff, *addr, size,
1490 				 prot, maxprot, docow);
1491 
1492 	if (rv != KERN_SUCCESS) {
1493 		/*
1494 		 * Lose the object reference. Will destroy the
1495 		 * object if it's an unnamed anonymous mapping
1496 		 * or named anonymous without other references.
1497 		 */
1498 		vm_object_deallocate(object);
1499 	} else if (flags & MAP_SHARED) {
1500 		/*
1501 		 * Shared memory is also shared with children.
1502 		 */
1503 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1504 		if (rv != KERN_SUCCESS)
1505 			(void) vm_map_remove(map, *addr, *addr + size);
1506 	}
1507 
1508 	/*
1509 	 * If the process has requested that all future mappings
1510 	 * be wired, then heed this.
1511 	 */
1512 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1513 		vm_map_wire(map, *addr, *addr + size,
1514 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1515 
1516 	switch (rv) {
1517 	case KERN_SUCCESS:
1518 		return (0);
1519 	case KERN_INVALID_ADDRESS:
1520 	case KERN_NO_SPACE:
1521 		return (ENOMEM);
1522 	case KERN_PROTECTION_FAILURE:
1523 		return (EACCES);
1524 	default:
1525 		return (EINVAL);
1526 	}
1527 }
1528