xref: /freebsd/sys/vm/vm_mmap.c (revision 09e8dea79366f1e5b3a73e8a271b26e4b6bf2e6a)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
39  *
40  *	@(#)vm_mmap.c	8.4 (Berkeley) 1/12/94
41  * $FreeBSD$
42  */
43 
44 /*
45  * Mapped file (mmap) interface to VM
46  */
47 
48 #include "opt_compat.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/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/conf.h>
65 #include <sys/stat.h>
66 #include <sys/vmmeter.h>
67 #include <sys/sysctl.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_kern.h>
80 
81 #ifndef _SYS_SYSPROTO_H_
82 struct sbrk_args {
83 	int incr;
84 };
85 #endif
86 
87 static int max_proc_mmap;
88 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
89 
90 /*
91  * Set the maximum number of vm_map_entry structures per process.  Roughly
92  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
93  * of our KVM malloc space still results in generous limits.  We want a
94  * default that is good enough to prevent the kernel running out of resources
95  * if attacked from compromised user account but generous enough such that
96  * multi-threaded processes are not unduly inconvenienced.
97  */
98 static void vmmapentry_rsrc_init(void *);
99 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
100 
101 static void
102 vmmapentry_rsrc_init(dummy)
103         void *dummy;
104 {
105     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
106     max_proc_mmap /= 100;
107 }
108 
109 /*
110  * MPSAFE
111  */
112 /* ARGSUSED */
113 int
114 sbrk(td, uap)
115 	struct thread *td;
116 	struct sbrk_args *uap;
117 {
118 	/* Not yet implemented */
119 	/* mtx_lock(&Giant); */
120 	/* mtx_unlock(&Giant); */
121 	return (EOPNOTSUPP);
122 }
123 
124 #ifndef _SYS_SYSPROTO_H_
125 struct sstk_args {
126 	int incr;
127 };
128 #endif
129 
130 /*
131  * MPSAFE
132  */
133 /* ARGSUSED */
134 int
135 sstk(td, uap)
136 	struct thread *td;
137 	struct sstk_args *uap;
138 {
139 	/* Not yet implemented */
140 	/* mtx_lock(&Giant); */
141 	/* mtx_unlock(&Giant); */
142 	return (EOPNOTSUPP);
143 }
144 
145 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
146 #ifndef _SYS_SYSPROTO_H_
147 struct getpagesize_args {
148 	int dummy;
149 };
150 #endif
151 
152 /* ARGSUSED */
153 int
154 ogetpagesize(td, uap)
155 	struct thread *td;
156 	struct getpagesize_args *uap;
157 {
158 	/* MP SAFE */
159 	td->td_retval[0] = PAGE_SIZE;
160 	return (0);
161 }
162 #endif				/* COMPAT_43 || COMPAT_SUNOS */
163 
164 
165 /*
166  * Memory Map (mmap) system call.  Note that the file offset
167  * and address are allowed to be NOT page aligned, though if
168  * the MAP_FIXED flag it set, both must have the same remainder
169  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
170  * page-aligned, the actual mapping starts at trunc_page(addr)
171  * and the return value is adjusted up by the page offset.
172  *
173  * Generally speaking, only character devices which are themselves
174  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
175  * there would be no cache coherency between a descriptor and a VM mapping
176  * both to the same character device.
177  *
178  * Block devices can be mmap'd no matter what they represent.  Cache coherency
179  * is maintained as long as you do not write directly to the underlying
180  * character device.
181  */
182 #ifndef _SYS_SYSPROTO_H_
183 struct mmap_args {
184 	void *addr;
185 	size_t len;
186 	int prot;
187 	int flags;
188 	int fd;
189 	long pad;
190 	off_t pos;
191 };
192 #endif
193 
194 /*
195  * MPSAFE
196  */
197 int
198 mmap(td, uap)
199 	struct thread *td;
200 	struct mmap_args *uap;
201 {
202 	struct file *fp = NULL;
203 	struct vnode *vp;
204 	vm_offset_t addr;
205 	vm_size_t size, pageoff;
206 	vm_prot_t prot, maxprot;
207 	void *handle;
208 	int flags, error;
209 	int disablexworkaround;
210 	off_t pos;
211 	struct vmspace *vms = td->td_proc->p_vmspace;
212 	vm_object_t obj;
213 
214 	addr = (vm_offset_t) uap->addr;
215 	size = uap->len;
216 	prot = uap->prot & VM_PROT_ALL;
217 	flags = uap->flags;
218 	pos = uap->pos;
219 
220 	fp = NULL;
221 	/* make sure mapping fits into numeric range etc */
222 	if ((ssize_t) uap->len < 0 ||
223 	    ((flags & MAP_ANON) && uap->fd != -1))
224 		return (EINVAL);
225 
226 	if (flags & MAP_STACK) {
227 		if ((uap->fd != -1) ||
228 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
229 			return (EINVAL);
230 		flags |= MAP_ANON;
231 		pos = 0;
232 	}
233 
234 	/*
235 	 * Align the file position to a page boundary,
236 	 * and save its page offset component.
237 	 */
238 	pageoff = (pos & PAGE_MASK);
239 	pos -= pageoff;
240 
241 	/* Adjust size for rounding (on both ends). */
242 	size += pageoff;			/* low end... */
243 	size = (vm_size_t) round_page(size);	/* hi end */
244 
245 	/*
246 	 * Check for illegal addresses.  Watch out for address wrap... Note
247 	 * that VM_*_ADDRESS are not constants due to casts (argh).
248 	 */
249 	if (flags & MAP_FIXED) {
250 		/*
251 		 * The specified address must have the same remainder
252 		 * as the file offset taken modulo PAGE_SIZE, so it
253 		 * should be aligned after adjustment by pageoff.
254 		 */
255 		addr -= pageoff;
256 		if (addr & PAGE_MASK)
257 			return (EINVAL);
258 		/* Address range must be all in user VM space. */
259 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
260 			return (EINVAL);
261 #ifndef __i386__
262 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
263 			return (EINVAL);
264 #endif
265 		if (addr + size < addr)
266 			return (EINVAL);
267 	}
268 	/*
269 	 * XXX for non-fixed mappings where no hint is provided or
270 	 * the hint would fall in the potential heap space,
271 	 * place it after the end of the largest possible heap.
272 	 *
273 	 * There should really be a pmap call to determine a reasonable
274 	 * location.
275 	 */
276 	else if (addr == 0 ||
277 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
278 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
279 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
280 
281 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
282 	if (flags & MAP_ANON) {
283 		/*
284 		 * Mapping blank space is trivial.
285 		 */
286 		handle = NULL;
287 		maxprot = VM_PROT_ALL;
288 		pos = 0;
289 	} else {
290 		/*
291 		 * Mapping file, get fp for validation. Obtain vnode and make
292 		 * sure it is of appropriate type.
293 		 * don't let the descriptor disappear on us if we block
294 		 */
295 		if ((error = fget(td, uap->fd, &fp)) != 0)
296 			goto done;
297 		if (fp->f_type != DTYPE_VNODE) {
298 			error = EINVAL;
299 			goto done;
300 		}
301 
302 		/*
303 		 * POSIX shared-memory objects are defined to have
304 		 * kernel persistence, and are not defined to support
305 		 * read(2)/write(2) -- or even open(2).  Thus, we can
306 		 * use MAP_ASYNC to trade on-disk coherence for speed.
307 		 * The shm_open(3) library routine turns on the FPOSIXSHM
308 		 * flag to request this behavior.
309 		 */
310 		if (fp->f_flag & FPOSIXSHM)
311 			flags |= MAP_NOSYNC;
312 		vp = (struct vnode *) fp->f_data;
313 		if (vp->v_type != VREG && vp->v_type != VCHR) {
314 			error = EINVAL;
315 			goto done;
316 		}
317 		if (vp->v_type == VREG) {
318 			/*
319 			 * Get the proper underlying object
320 			 */
321 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
322 				error = EINVAL;
323 				goto done;
324 			}
325 			vp = (struct vnode*)obj->handle;
326 		}
327 		/*
328 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
329 		 * SunOS).
330 		 */
331 		if ((vp->v_type == VCHR) &&
332 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
333 			handle = NULL;
334 			maxprot = VM_PROT_ALL;
335 			flags |= MAP_ANON;
336 			pos = 0;
337 		} else {
338 			/*
339 			 * cdevs does not provide private mappings of any kind.
340 			 */
341 			/*
342 			 * However, for XIG X server to continue to work,
343 			 * we should allow the superuser to do it anyway.
344 			 * We only allow it at securelevel < 1.
345 			 * (Because the XIG X server writes directly to video
346 			 * memory via /dev/mem, it should never work at any
347 			 * other securelevel.
348 			 * XXX this will have to go
349 			 */
350 			if (securelevel_ge(td->td_ucred, 1))
351 				disablexworkaround = 1;
352 			else
353 				disablexworkaround = suser(td);
354 			if (vp->v_type == VCHR && disablexworkaround &&
355 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
356 				error = EINVAL;
357 				goto done;
358 			}
359 			/*
360 			 * Ensure that file and memory protections are
361 			 * compatible.  Note that we only worry about
362 			 * writability if mapping is shared; in this case,
363 			 * current and max prot are dictated by the open file.
364 			 * XXX use the vnode instead?  Problem is: what
365 			 * credentials do we use for determination? What if
366 			 * proc does a setuid?
367 			 */
368 			maxprot = VM_PROT_EXECUTE;	/* ??? */
369 			if (fp->f_flag & FREAD) {
370 				maxprot |= VM_PROT_READ;
371 			} else if (prot & PROT_READ) {
372 				error = EACCES;
373 				goto done;
374 			}
375 			/*
376 			 * If we are sharing potential changes (either via
377 			 * MAP_SHARED or via the implicit sharing of character
378 			 * device mappings), and we are trying to get write
379 			 * permission although we opened it without asking
380 			 * for it, bail out.  Check for superuser, only if
381 			 * we're at securelevel < 1, to allow the XIG X server
382 			 * to continue to work.
383 			 */
384 			if ((flags & MAP_SHARED) != 0 ||
385 			    (vp->v_type == VCHR && disablexworkaround)) {
386 				if ((fp->f_flag & FWRITE) != 0) {
387 					struct vattr va;
388 					if ((error =
389 					    VOP_GETATTR(vp, &va,
390 						        td->td_ucred, td))) {
391 						goto done;
392 					}
393 					if ((va.va_flags &
394 					   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
395 						maxprot |= VM_PROT_WRITE;
396 					} else if (prot & PROT_WRITE) {
397 						error = EPERM;
398 						goto done;
399 					}
400 				} else if ((prot & PROT_WRITE) != 0) {
401 					error = EACCES;
402 					goto done;
403 				}
404 			} else {
405 				maxprot |= VM_PROT_WRITE;
406 			}
407 
408 			handle = (void *)vp;
409 		}
410 	}
411 
412 	/*
413 	 * Do not allow more then a certain number of vm_map_entry structures
414 	 * per process.  Scale with the number of rforks sharing the map
415 	 * to make the limit reasonable for threads.
416 	 */
417 	if (max_proc_mmap &&
418 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
419 		error = ENOMEM;
420 		goto done;
421 	}
422 
423 	mtx_unlock(&Giant);
424 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
425 	    flags, handle, pos);
426 	if (error == 0)
427 		td->td_retval[0] = (register_t) (addr + pageoff);
428 	goto done2;
429 done:
430 	mtx_unlock(&Giant);
431 done2:
432 	if (fp)
433 		fdrop(fp, td);
434 	return (error);
435 }
436 
437 #ifdef COMPAT_43
438 #ifndef _SYS_SYSPROTO_H_
439 struct ommap_args {
440 	caddr_t addr;
441 	int len;
442 	int prot;
443 	int flags;
444 	int fd;
445 	long pos;
446 };
447 #endif
448 int
449 ommap(td, uap)
450 	struct thread *td;
451 	struct ommap_args *uap;
452 {
453 	struct mmap_args nargs;
454 	static const char cvtbsdprot[8] = {
455 		0,
456 		PROT_EXEC,
457 		PROT_WRITE,
458 		PROT_EXEC | PROT_WRITE,
459 		PROT_READ,
460 		PROT_EXEC | PROT_READ,
461 		PROT_WRITE | PROT_READ,
462 		PROT_EXEC | PROT_WRITE | PROT_READ,
463 	};
464 
465 #define	OMAP_ANON	0x0002
466 #define	OMAP_COPY	0x0020
467 #define	OMAP_SHARED	0x0010
468 #define	OMAP_FIXED	0x0100
469 
470 	nargs.addr = uap->addr;
471 	nargs.len = uap->len;
472 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
473 	nargs.flags = 0;
474 	if (uap->flags & OMAP_ANON)
475 		nargs.flags |= MAP_ANON;
476 	if (uap->flags & OMAP_COPY)
477 		nargs.flags |= MAP_COPY;
478 	if (uap->flags & OMAP_SHARED)
479 		nargs.flags |= MAP_SHARED;
480 	else
481 		nargs.flags |= MAP_PRIVATE;
482 	if (uap->flags & OMAP_FIXED)
483 		nargs.flags |= MAP_FIXED;
484 	nargs.fd = uap->fd;
485 	nargs.pos = uap->pos;
486 	return (mmap(td, &nargs));
487 }
488 #endif				/* COMPAT_43 */
489 
490 
491 #ifndef _SYS_SYSPROTO_H_
492 struct msync_args {
493 	void *addr;
494 	int len;
495 	int flags;
496 };
497 #endif
498 /*
499  * MPSAFE
500  */
501 int
502 msync(td, uap)
503 	struct thread *td;
504 	struct msync_args *uap;
505 {
506 	vm_offset_t addr;
507 	vm_size_t size, pageoff;
508 	int flags;
509 	vm_map_t map;
510 	int rv;
511 
512 	addr = (vm_offset_t) uap->addr;
513 	size = uap->len;
514 	flags = uap->flags;
515 
516 	pageoff = (addr & PAGE_MASK);
517 	addr -= pageoff;
518 	size += pageoff;
519 	size = (vm_size_t) round_page(size);
520 	if (addr + size < addr)
521 		return (EINVAL);
522 
523 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
524 		return (EINVAL);
525 
526 	mtx_lock(&Giant);
527 
528 	map = &td->td_proc->p_vmspace->vm_map;
529 
530 	/*
531 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
532 	 * pages with the region containing addr".  Unfortunately, we don't
533 	 * really keep track of individual mmaps so we approximate by flushing
534 	 * the range of the map entry containing addr. This can be incorrect
535 	 * if the region splits or is coalesced with a neighbor.
536 	 */
537 	if (size == 0) {
538 		vm_map_entry_t entry;
539 
540 		vm_map_lock_read(map);
541 		rv = vm_map_lookup_entry(map, addr, &entry);
542 		vm_map_unlock_read(map);
543 		if (rv == FALSE) {
544 			rv = -1;
545 			goto done2;
546 		}
547 		addr = entry->start;
548 		size = entry->end - entry->start;
549 	}
550 
551 	/*
552 	 * Clean the pages and interpret the return value.
553 	 */
554 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
555 	    (flags & MS_INVALIDATE) != 0);
556 
557 done2:
558 	mtx_unlock(&Giant);
559 
560 	switch (rv) {
561 	case KERN_SUCCESS:
562 		return (0);
563 	case KERN_INVALID_ADDRESS:
564 		return (EINVAL);	/* Sun returns ENOMEM? */
565 	case KERN_FAILURE:
566 		return (EIO);
567 	default:
568 		return (EINVAL);
569 	}
570 }
571 
572 #ifndef _SYS_SYSPROTO_H_
573 struct munmap_args {
574 	void *addr;
575 	size_t len;
576 };
577 #endif
578 /*
579  * MPSAFE
580  */
581 int
582 munmap(td, uap)
583 	struct thread *td;
584 	struct munmap_args *uap;
585 {
586 	vm_offset_t addr;
587 	vm_size_t size, pageoff;
588 	vm_map_t map;
589 
590 	addr = (vm_offset_t) uap->addr;
591 	size = uap->len;
592 
593 	pageoff = (addr & PAGE_MASK);
594 	addr -= pageoff;
595 	size += pageoff;
596 	size = (vm_size_t) round_page(size);
597 	if (addr + size < addr)
598 		return (EINVAL);
599 
600 	if (size == 0)
601 		return (0);
602 
603 	/*
604 	 * Check for illegal addresses.  Watch out for address wrap... Note
605 	 * that VM_*_ADDRESS are not constants due to casts (argh).
606 	 */
607 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
608 		return (EINVAL);
609 #ifndef __i386__
610 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
611 		return (EINVAL);
612 #endif
613 	map = &td->td_proc->p_vmspace->vm_map;
614 	/*
615 	 * Make sure entire range is allocated.
616 	 */
617 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
618 		return (EINVAL);
619 
620 	/* returns nothing but KERN_SUCCESS anyway */
621 	(void) vm_map_remove(map, addr, addr + size);
622 	return (0);
623 }
624 
625 #if 0
626 void
627 munmapfd(td, fd)
628 	struct thread *td;
629 	int fd;
630 {
631 	/*
632 	 * XXX should unmap any regions mapped to this file
633 	 */
634 	FILEDESC_LOCK(p->p_fd);
635 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
636 	FILEDESC_UNLOCK(p->p_fd);
637 }
638 #endif
639 
640 #ifndef _SYS_SYSPROTO_H_
641 struct mprotect_args {
642 	const void *addr;
643 	size_t len;
644 	int prot;
645 };
646 #endif
647 /*
648  * MPSAFE
649  */
650 int
651 mprotect(td, uap)
652 	struct thread *td;
653 	struct mprotect_args *uap;
654 {
655 	vm_offset_t addr;
656 	vm_size_t size, pageoff;
657 	vm_prot_t prot;
658 
659 	addr = (vm_offset_t) uap->addr;
660 	size = uap->len;
661 	prot = uap->prot & VM_PROT_ALL;
662 #if defined(VM_PROT_READ_IS_EXEC)
663 	if (prot & VM_PROT_READ)
664 		prot |= VM_PROT_EXECUTE;
665 #endif
666 
667 	pageoff = (addr & PAGE_MASK);
668 	addr -= pageoff;
669 	size += pageoff;
670 	size = (vm_size_t) round_page(size);
671 	if (addr + size < addr)
672 		return (EINVAL);
673 
674 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
675 	    addr + size, prot, FALSE)) {
676 	case KERN_SUCCESS:
677 		return (0);
678 	case KERN_PROTECTION_FAILURE:
679 		return (EACCES);
680 	}
681 	return (EINVAL);
682 }
683 
684 #ifndef _SYS_SYSPROTO_H_
685 struct minherit_args {
686 	void *addr;
687 	size_t len;
688 	int inherit;
689 };
690 #endif
691 /*
692  * MPSAFE
693  */
694 int
695 minherit(td, uap)
696 	struct thread *td;
697 	struct minherit_args *uap;
698 {
699 	vm_offset_t addr;
700 	vm_size_t size, pageoff;
701 	vm_inherit_t inherit;
702 
703 	addr = (vm_offset_t)uap->addr;
704 	size = uap->len;
705 	inherit = uap->inherit;
706 
707 	pageoff = (addr & PAGE_MASK);
708 	addr -= pageoff;
709 	size += pageoff;
710 	size = (vm_size_t) round_page(size);
711 	if (addr + size < addr)
712 		return (EINVAL);
713 
714 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
715 	    addr + size, inherit)) {
716 	case KERN_SUCCESS:
717 		return (0);
718 	case KERN_PROTECTION_FAILURE:
719 		return (EACCES);
720 	}
721 	return (EINVAL);
722 }
723 
724 #ifndef _SYS_SYSPROTO_H_
725 struct madvise_args {
726 	void *addr;
727 	size_t len;
728 	int behav;
729 };
730 #endif
731 
732 /*
733  * MPSAFE
734  */
735 /* ARGSUSED */
736 int
737 madvise(td, uap)
738 	struct thread *td;
739 	struct madvise_args *uap;
740 {
741 	vm_offset_t start, end;
742 
743 	/*
744 	 * Check for illegal behavior
745 	 */
746 	if (uap->behav < 0 || uap->behav > MADV_CORE)
747 		return (EINVAL);
748 	/*
749 	 * Check for illegal addresses.  Watch out for address wrap... Note
750 	 * that VM_*_ADDRESS are not constants due to casts (argh).
751 	 */
752 	if (VM_MAXUSER_ADDRESS > 0 &&
753 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
754 		return (EINVAL);
755 #ifndef __i386__
756 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
757 		return (EINVAL);
758 #endif
759 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
760 		return (EINVAL);
761 
762 	/*
763 	 * Since this routine is only advisory, we default to conservative
764 	 * behavior.
765 	 */
766 	start = trunc_page((vm_offset_t) uap->addr);
767 	end = round_page((vm_offset_t) uap->addr + uap->len);
768 
769 	if (vm_map_madvise(&td->td_proc->p_vmspace->vm_map, start, end,
770 	    uap->behav))
771 		return (EINVAL);
772 	return (0);
773 }
774 
775 #ifndef _SYS_SYSPROTO_H_
776 struct mincore_args {
777 	const void *addr;
778 	size_t len;
779 	char *vec;
780 };
781 #endif
782 
783 /*
784  * MPSAFE
785  */
786 /* ARGSUSED */
787 int
788 mincore(td, uap)
789 	struct thread *td;
790 	struct mincore_args *uap;
791 {
792 	vm_offset_t addr, first_addr;
793 	vm_offset_t end, cend;
794 	pmap_t pmap;
795 	vm_map_t map;
796 	char *vec;
797 	int error = 0;
798 	int vecindex, lastvecindex;
799 	vm_map_entry_t current;
800 	vm_map_entry_t entry;
801 	int mincoreinfo;
802 	unsigned int timestamp;
803 
804 	/*
805 	 * Make sure that the addresses presented are valid for user
806 	 * mode.
807 	 */
808 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
809 	end = addr + (vm_size_t)round_page(uap->len);
810 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
811 		return (EINVAL);
812 	if (end < addr)
813 		return (EINVAL);
814 
815 	/*
816 	 * Address of byte vector
817 	 */
818 	vec = uap->vec;
819 
820 	mtx_lock(&Giant);
821 	map = &td->td_proc->p_vmspace->vm_map;
822 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
823 
824 	vm_map_lock_read(map);
825 RestartScan:
826 	timestamp = map->timestamp;
827 
828 	if (!vm_map_lookup_entry(map, addr, &entry))
829 		entry = entry->next;
830 
831 	/*
832 	 * Do this on a map entry basis so that if the pages are not
833 	 * in the current processes address space, we can easily look
834 	 * up the pages elsewhere.
835 	 */
836 	lastvecindex = -1;
837 	for (current = entry;
838 	    (current != &map->header) && (current->start < end);
839 	    current = current->next) {
840 
841 		/*
842 		 * ignore submaps (for now) or null objects
843 		 */
844 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
845 			current->object.vm_object == NULL)
846 			continue;
847 
848 		/*
849 		 * limit this scan to the current map entry and the
850 		 * limits for the mincore call
851 		 */
852 		if (addr < current->start)
853 			addr = current->start;
854 		cend = current->end;
855 		if (cend > end)
856 			cend = end;
857 
858 		/*
859 		 * scan this entry one page at a time
860 		 */
861 		while (addr < cend) {
862 			/*
863 			 * Check pmap first, it is likely faster, also
864 			 * it can provide info as to whether we are the
865 			 * one referencing or modifying the page.
866 			 */
867 			mincoreinfo = pmap_mincore(pmap, addr);
868 			if (!mincoreinfo) {
869 				vm_pindex_t pindex;
870 				vm_ooffset_t offset;
871 				vm_page_t m;
872 				/*
873 				 * calculate the page index into the object
874 				 */
875 				offset = current->offset + (addr - current->start);
876 				pindex = OFF_TO_IDX(offset);
877 				m = vm_page_lookup(current->object.vm_object,
878 					pindex);
879 				/*
880 				 * if the page is resident, then gather information about
881 				 * it.
882 				 */
883 				if (m) {
884 					mincoreinfo = MINCORE_INCORE;
885 					if (m->dirty ||
886 						pmap_is_modified(m))
887 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
888 					if ((m->flags & PG_REFERENCED) ||
889 						pmap_ts_referenced(m)) {
890 						vm_page_flag_set(m, PG_REFERENCED);
891 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
892 					}
893 				}
894 			}
895 
896 			/*
897 			 * subyte may page fault.  In case it needs to modify
898 			 * the map, we release the lock.
899 			 */
900 			vm_map_unlock_read(map);
901 
902 			/*
903 			 * calculate index into user supplied byte vector
904 			 */
905 			vecindex = OFF_TO_IDX(addr - first_addr);
906 
907 			/*
908 			 * If we have skipped map entries, we need to make sure that
909 			 * the byte vector is zeroed for those skipped entries.
910 			 */
911 			while ((lastvecindex + 1) < vecindex) {
912 				error = subyte(vec + lastvecindex, 0);
913 				if (error) {
914 					error = EFAULT;
915 					goto done2;
916 				}
917 				++lastvecindex;
918 			}
919 
920 			/*
921 			 * Pass the page information to the user
922 			 */
923 			error = subyte(vec + vecindex, mincoreinfo);
924 			if (error) {
925 				error = EFAULT;
926 				goto done2;
927 			}
928 
929 			/*
930 			 * If the map has changed, due to the subyte, the previous
931 			 * output may be invalid.
932 			 */
933 			vm_map_lock_read(map);
934 			if (timestamp != map->timestamp)
935 				goto RestartScan;
936 
937 			lastvecindex = vecindex;
938 			addr += PAGE_SIZE;
939 		}
940 	}
941 
942 	/*
943 	 * subyte may page fault.  In case it needs to modify
944 	 * the map, we release the lock.
945 	 */
946 	vm_map_unlock_read(map);
947 
948 	/*
949 	 * Zero the last entries in the byte vector.
950 	 */
951 	vecindex = OFF_TO_IDX(end - first_addr);
952 	while ((lastvecindex + 1) < vecindex) {
953 		error = subyte(vec + lastvecindex, 0);
954 		if (error) {
955 			error = EFAULT;
956 			goto done2;
957 		}
958 		++lastvecindex;
959 	}
960 
961 	/*
962 	 * If the map has changed, due to the subyte, the previous
963 	 * output may be invalid.
964 	 */
965 	vm_map_lock_read(map);
966 	if (timestamp != map->timestamp)
967 		goto RestartScan;
968 	vm_map_unlock_read(map);
969 done2:
970 	mtx_unlock(&Giant);
971 	return (error);
972 }
973 
974 #ifndef _SYS_SYSPROTO_H_
975 struct mlock_args {
976 	const void *addr;
977 	size_t len;
978 };
979 #endif
980 /*
981  * MPSAFE
982  */
983 int
984 mlock(td, uap)
985 	struct thread *td;
986 	struct mlock_args *uap;
987 {
988 	vm_offset_t addr;
989 	vm_size_t size, pageoff;
990 	int error;
991 
992 	addr = (vm_offset_t) uap->addr;
993 	size = uap->len;
994 
995 	pageoff = (addr & PAGE_MASK);
996 	addr -= pageoff;
997 	size += pageoff;
998 	size = (vm_size_t) round_page(size);
999 
1000 	/* disable wrap around */
1001 	if (addr + size < addr)
1002 		return (EINVAL);
1003 
1004 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1005 		return (EAGAIN);
1006 
1007 #ifdef pmap_wired_count
1008 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1009 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1010 		return (ENOMEM);
1011 #else
1012 	error = suser(td);
1013 	if (error)
1014 		return (error);
1015 #endif
1016 
1017 	error = vm_map_wire(&td->td_proc->p_vmspace->vm_map, addr,
1018 		     addr + size, TRUE);
1019 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1020 }
1021 
1022 #ifndef _SYS_SYSPROTO_H_
1023 struct mlockall_args {
1024 	int	how;
1025 };
1026 #endif
1027 
1028 /*
1029  * MPSAFE
1030  */
1031 int
1032 mlockall(td, uap)
1033 	struct thread *td;
1034 	struct mlockall_args *uap;
1035 {
1036 	/* mtx_lock(&Giant); */
1037 	/* mtx_unlock(&Giant); */
1038 	return 0;
1039 }
1040 
1041 #ifndef _SYS_SYSPROTO_H_
1042 struct munlockall_args {
1043 	int	how;
1044 };
1045 #endif
1046 
1047 /*
1048  * MPSAFE
1049  */
1050 int
1051 munlockall(td, uap)
1052 	struct thread *td;
1053 	struct munlockall_args *uap;
1054 {
1055 	/* mtx_lock(&Giant); */
1056 	/* mtx_unlock(&Giant); */
1057 	return 0;
1058 }
1059 
1060 #ifndef _SYS_SYSPROTO_H_
1061 struct munlock_args {
1062 	const void *addr;
1063 	size_t len;
1064 };
1065 #endif
1066 /*
1067  * MPSAFE
1068  */
1069 int
1070 munlock(td, uap)
1071 	struct thread *td;
1072 	struct munlock_args *uap;
1073 {
1074 	vm_offset_t addr;
1075 	vm_size_t size, pageoff;
1076 	int error;
1077 
1078 	addr = (vm_offset_t) uap->addr;
1079 	size = uap->len;
1080 
1081 	pageoff = (addr & PAGE_MASK);
1082 	addr -= pageoff;
1083 	size += pageoff;
1084 	size = (vm_size_t) round_page(size);
1085 
1086 	/* disable wrap around */
1087 	if (addr + size < addr)
1088 		return (EINVAL);
1089 
1090 #ifndef pmap_wired_count
1091 	error = suser(td);
1092 	if (error)
1093 		return (error);
1094 #endif
1095 
1096 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, addr,
1097 		     addr + size, TRUE);
1098 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1099 }
1100 
1101 /*
1102  * vm_mmap()
1103  *
1104  * MPSAFE
1105  *
1106  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1107  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1108  */
1109 int
1110 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1111 	vm_prot_t maxprot, int flags,
1112 	void *handle,
1113 	vm_ooffset_t foff)
1114 {
1115 	boolean_t fitit;
1116 	vm_object_t object;
1117 	struct vnode *vp = NULL;
1118 	objtype_t type;
1119 	int rv = KERN_SUCCESS;
1120 	vm_ooffset_t objsize;
1121 	int docow;
1122 	struct thread *td = curthread;
1123 
1124 	if (size == 0)
1125 		return (0);
1126 
1127 	objsize = size = round_page(size);
1128 
1129 	if (td->td_proc->p_vmspace->vm_map.size + size >
1130 	    td->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1131 		return(ENOMEM);
1132 	}
1133 
1134 	/*
1135 	 * We currently can only deal with page aligned file offsets.
1136 	 * The check is here rather than in the syscall because the
1137 	 * kernel calls this function internally for other mmaping
1138 	 * operations (such as in exec) and non-aligned offsets will
1139 	 * cause pmap inconsistencies...so we want to be sure to
1140 	 * disallow this in all cases.
1141 	 */
1142 	if (foff & PAGE_MASK)
1143 		return (EINVAL);
1144 
1145 	if ((flags & MAP_FIXED) == 0) {
1146 		fitit = TRUE;
1147 		*addr = round_page(*addr);
1148 	} else {
1149 		if (*addr != trunc_page(*addr))
1150 			return (EINVAL);
1151 		fitit = FALSE;
1152 		(void) vm_map_remove(map, *addr, *addr + size);
1153 	}
1154 
1155 	/*
1156 	 * Lookup/allocate object.
1157 	 */
1158 	if (flags & MAP_ANON) {
1159 		type = OBJT_DEFAULT;
1160 		/*
1161 		 * Unnamed anonymous regions always start at 0.
1162 		 */
1163 		if (handle == 0)
1164 			foff = 0;
1165 	} else {
1166 		vp = (struct vnode *) handle;
1167 		mtx_lock(&Giant);
1168 		if (vp->v_type == VCHR) {
1169 			type = OBJT_DEVICE;
1170 			handle = (void *)(intptr_t)vp->v_rdev;
1171 		} else {
1172 			struct vattr vat;
1173 			int error;
1174 
1175 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1176 			if (error) {
1177 				mtx_unlock(&Giant);
1178 				return (error);
1179 			}
1180 			objsize = round_page(vat.va_size);
1181 			type = OBJT_VNODE;
1182 			/*
1183 			 * if it is a regular file without any references
1184 			 * we do not need to sync it.
1185 			 */
1186 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1187 				flags |= MAP_NOSYNC;
1188 			}
1189 		}
1190 		mtx_unlock(&Giant);
1191 	}
1192 
1193 	if (handle == NULL) {
1194 		object = NULL;
1195 		docow = 0;
1196 	} else {
1197 		object = vm_pager_allocate(type,
1198 			handle, objsize, prot, foff);
1199 		if (object == NULL) {
1200 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1201 		}
1202 		docow = MAP_PREFAULT_PARTIAL;
1203 	}
1204 
1205 	/*
1206 	 * Force device mappings to be shared.
1207 	 */
1208 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1209 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1210 		flags |= MAP_SHARED;
1211 	}
1212 
1213 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1214 		docow |= MAP_COPY_ON_WRITE;
1215 	if (flags & MAP_NOSYNC)
1216 		docow |= MAP_DISABLE_SYNCER;
1217 	if (flags & MAP_NOCORE)
1218 		docow |= MAP_DISABLE_COREDUMP;
1219 
1220 #if defined(VM_PROT_READ_IS_EXEC)
1221 	if (prot & VM_PROT_READ)
1222 		prot |= VM_PROT_EXECUTE;
1223 
1224 	if (maxprot & VM_PROT_READ)
1225 		maxprot |= VM_PROT_EXECUTE;
1226 #endif
1227 
1228 	if (fitit)
1229 		*addr = pmap_addr_hint(object, *addr, size);
1230 
1231 	if (flags & MAP_STACK)
1232 		rv = vm_map_stack (map, *addr, size, prot,
1233 				   maxprot, docow);
1234 	else
1235 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1236 				 prot, maxprot, docow);
1237 
1238 	if (rv != KERN_SUCCESS) {
1239 		/*
1240 		 * Lose the object reference. Will destroy the
1241 		 * object if it's an unnamed anonymous mapping
1242 		 * or named anonymous without other references.
1243 		 */
1244 		vm_object_deallocate(object);
1245 	} else if (flags & MAP_SHARED) {
1246 		/*
1247 		 * Shared memory is also shared with children.
1248 		 */
1249 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1250 		if (rv != KERN_SUCCESS)
1251 			(void) vm_map_remove(map, *addr, *addr + size);
1252 	}
1253 	switch (rv) {
1254 	case KERN_SUCCESS:
1255 		return (0);
1256 	case KERN_INVALID_ADDRESS:
1257 	case KERN_NO_SPACE:
1258 		return (ENOMEM);
1259 	case KERN_PROTECTION_FAILURE:
1260 		return (EACCES);
1261 	default:
1262 		return (EINVAL);
1263 	}
1264 }
1265