xref: /freebsd/sys/vm/vm_mmap.c (revision c4f6a2a9e1b1879b618c436ab4f56ff75c73a0f5)
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 	vp = NULL;
221 	fp = NULL;
222 	/* make sure mapping fits into numeric range etc */
223 	if ((ssize_t) uap->len < 0 ||
224 	    ((flags & MAP_ANON) && uap->fd != -1))
225 		return (EINVAL);
226 
227 	if (flags & MAP_STACK) {
228 		if ((uap->fd != -1) ||
229 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
230 			return (EINVAL);
231 		flags |= MAP_ANON;
232 		pos = 0;
233 	}
234 
235 	/*
236 	 * Align the file position to a page boundary,
237 	 * and save its page offset component.
238 	 */
239 	pageoff = (pos & PAGE_MASK);
240 	pos -= pageoff;
241 
242 	/* Adjust size for rounding (on both ends). */
243 	size += pageoff;			/* low end... */
244 	size = (vm_size_t) round_page(size);	/* hi end */
245 
246 	/*
247 	 * Check for illegal addresses.  Watch out for address wrap... Note
248 	 * that VM_*_ADDRESS are not constants due to casts (argh).
249 	 */
250 	if (flags & MAP_FIXED) {
251 		/*
252 		 * The specified address must have the same remainder
253 		 * as the file offset taken modulo PAGE_SIZE, so it
254 		 * should be aligned after adjustment by pageoff.
255 		 */
256 		addr -= pageoff;
257 		if (addr & PAGE_MASK)
258 			return (EINVAL);
259 		/* Address range must be all in user VM space. */
260 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
261 			return (EINVAL);
262 #ifndef __i386__
263 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
264 			return (EINVAL);
265 #endif
266 		if (addr + size < addr)
267 			return (EINVAL);
268 	}
269 	/*
270 	 * XXX for non-fixed mappings where no hint is provided or
271 	 * the hint would fall in the potential heap space,
272 	 * place it after the end of the largest possible heap.
273 	 *
274 	 * There should really be a pmap call to determine a reasonable
275 	 * location.
276 	 */
277 	else if (addr == 0 ||
278 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
279 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
280 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
281 
282 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
283 	if (flags & MAP_ANON) {
284 		/*
285 		 * Mapping blank space is trivial.
286 		 */
287 		handle = NULL;
288 		maxprot = VM_PROT_ALL;
289 		pos = 0;
290 	} else {
291 		/*
292 		 * Mapping file, get fp for validation. Obtain vnode and make
293 		 * sure it is of appropriate type.
294 		 * don't let the descriptor disappear on us if we block
295 		 */
296 		if ((error = fget(td, uap->fd, &fp)) != 0)
297 			goto done;
298 		if (fp->f_type != DTYPE_VNODE) {
299 			error = EINVAL;
300 			goto done;
301 		}
302 
303 		/*
304 		 * POSIX shared-memory objects are defined to have
305 		 * kernel persistence, and are not defined to support
306 		 * read(2)/write(2) -- or even open(2).  Thus, we can
307 		 * use MAP_ASYNC to trade on-disk coherence for speed.
308 		 * The shm_open(3) library routine turns on the FPOSIXSHM
309 		 * flag to request this behavior.
310 		 */
311 		if (fp->f_flag & FPOSIXSHM)
312 			flags |= MAP_NOSYNC;
313 		vp = (struct vnode *) fp->f_data;
314 		error = vget(vp, LK_EXCLUSIVE, td);
315 		if (error)
316 			goto done;
317 		if (vp->v_type != VREG && vp->v_type != VCHR) {
318 			error = EINVAL;
319 			goto done;
320 		}
321 		if (vp->v_type == VREG) {
322 			/*
323 			 * Get the proper underlying object
324 			 */
325 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
326 				error = EINVAL;
327 				goto done;
328 			}
329 			if (obj->handle != vp) {
330 				vput(vp);
331 				vp = (struct vnode*)obj->handle;
332 				vget(vp, LK_EXCLUSIVE, td);
333 			}
334 		}
335 		/*
336 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
337 		 * SunOS).
338 		 */
339 		if ((vp->v_type == VCHR) &&
340 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
341 			handle = NULL;
342 			maxprot = VM_PROT_ALL;
343 			flags |= MAP_ANON;
344 			pos = 0;
345 		} else {
346 			/*
347 			 * cdevs does not provide private mappings of any kind.
348 			 */
349 			/*
350 			 * However, for XIG X server to continue to work,
351 			 * we should allow the superuser to do it anyway.
352 			 * We only allow it at securelevel < 1.
353 			 * (Because the XIG X server writes directly to video
354 			 * memory via /dev/mem, it should never work at any
355 			 * other securelevel.
356 			 * XXX this will have to go
357 			 */
358 			if (securelevel_ge(td->td_ucred, 1))
359 				disablexworkaround = 1;
360 			else
361 				disablexworkaround = suser(td);
362 			if (vp->v_type == VCHR && disablexworkaround &&
363 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
364 				error = EINVAL;
365 				goto done;
366 			}
367 			/*
368 			 * Ensure that file and memory protections are
369 			 * compatible.  Note that we only worry about
370 			 * writability if mapping is shared; in this case,
371 			 * current and max prot are dictated by the open file.
372 			 * XXX use the vnode instead?  Problem is: what
373 			 * credentials do we use for determination? What if
374 			 * proc does a setuid?
375 			 */
376 			maxprot = VM_PROT_EXECUTE;	/* ??? */
377 			if (fp->f_flag & FREAD) {
378 				maxprot |= VM_PROT_READ;
379 			} else if (prot & PROT_READ) {
380 				error = EACCES;
381 				goto done;
382 			}
383 			/*
384 			 * If we are sharing potential changes (either via
385 			 * MAP_SHARED or via the implicit sharing of character
386 			 * device mappings), and we are trying to get write
387 			 * permission although we opened it without asking
388 			 * for it, bail out.  Check for superuser, only if
389 			 * we're at securelevel < 1, to allow the XIG X server
390 			 * to continue to work.
391 			 */
392 			if ((flags & MAP_SHARED) != 0 ||
393 			    (vp->v_type == VCHR && disablexworkaround)) {
394 				if ((fp->f_flag & FWRITE) != 0) {
395 					struct vattr va;
396 					if ((error =
397 					    VOP_GETATTR(vp, &va,
398 						        td->td_ucred, td))) {
399 						goto done;
400 					}
401 					if ((va.va_flags &
402 					   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
403 						maxprot |= VM_PROT_WRITE;
404 					} else if (prot & PROT_WRITE) {
405 						error = EPERM;
406 						goto done;
407 					}
408 				} else if ((prot & PROT_WRITE) != 0) {
409 					error = EACCES;
410 					goto done;
411 				}
412 			} else {
413 				maxprot |= VM_PROT_WRITE;
414 			}
415 
416 			handle = (void *)vp;
417 		}
418 	}
419 
420 	/*
421 	 * Do not allow more then a certain number of vm_map_entry structures
422 	 * per process.  Scale with the number of rforks sharing the map
423 	 * to make the limit reasonable for threads.
424 	 */
425 	if (max_proc_mmap &&
426 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
427 		error = ENOMEM;
428 		goto done;
429 	}
430 
431 	mtx_unlock(&Giant);
432 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
433 	    flags, handle, pos);
434 	mtx_lock(&Giant);
435 	if (error == 0)
436 		td->td_retval[0] = (register_t) (addr + pageoff);
437 done:
438 	if (vp)
439 		vput(vp);
440 	mtx_unlock(&Giant);
441 	if (fp)
442 		fdrop(fp, td);
443 
444 	return (error);
445 }
446 
447 #ifdef COMPAT_43
448 #ifndef _SYS_SYSPROTO_H_
449 struct ommap_args {
450 	caddr_t addr;
451 	int len;
452 	int prot;
453 	int flags;
454 	int fd;
455 	long pos;
456 };
457 #endif
458 int
459 ommap(td, uap)
460 	struct thread *td;
461 	struct ommap_args *uap;
462 {
463 	struct mmap_args nargs;
464 	static const char cvtbsdprot[8] = {
465 		0,
466 		PROT_EXEC,
467 		PROT_WRITE,
468 		PROT_EXEC | PROT_WRITE,
469 		PROT_READ,
470 		PROT_EXEC | PROT_READ,
471 		PROT_WRITE | PROT_READ,
472 		PROT_EXEC | PROT_WRITE | PROT_READ,
473 	};
474 
475 #define	OMAP_ANON	0x0002
476 #define	OMAP_COPY	0x0020
477 #define	OMAP_SHARED	0x0010
478 #define	OMAP_FIXED	0x0100
479 
480 	nargs.addr = uap->addr;
481 	nargs.len = uap->len;
482 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
483 	nargs.flags = 0;
484 	if (uap->flags & OMAP_ANON)
485 		nargs.flags |= MAP_ANON;
486 	if (uap->flags & OMAP_COPY)
487 		nargs.flags |= MAP_COPY;
488 	if (uap->flags & OMAP_SHARED)
489 		nargs.flags |= MAP_SHARED;
490 	else
491 		nargs.flags |= MAP_PRIVATE;
492 	if (uap->flags & OMAP_FIXED)
493 		nargs.flags |= MAP_FIXED;
494 	nargs.fd = uap->fd;
495 	nargs.pos = uap->pos;
496 	return (mmap(td, &nargs));
497 }
498 #endif				/* COMPAT_43 */
499 
500 
501 #ifndef _SYS_SYSPROTO_H_
502 struct msync_args {
503 	void *addr;
504 	int len;
505 	int flags;
506 };
507 #endif
508 /*
509  * MPSAFE
510  */
511 int
512 msync(td, uap)
513 	struct thread *td;
514 	struct msync_args *uap;
515 {
516 	vm_offset_t addr;
517 	vm_size_t size, pageoff;
518 	int flags;
519 	vm_map_t map;
520 	int rv;
521 
522 	addr = (vm_offset_t) uap->addr;
523 	size = uap->len;
524 	flags = uap->flags;
525 
526 	pageoff = (addr & PAGE_MASK);
527 	addr -= pageoff;
528 	size += pageoff;
529 	size = (vm_size_t) round_page(size);
530 	if (addr + size < addr)
531 		return (EINVAL);
532 
533 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
534 		return (EINVAL);
535 
536 	mtx_lock(&Giant);
537 
538 	map = &td->td_proc->p_vmspace->vm_map;
539 
540 	/*
541 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
542 	 * pages with the region containing addr".  Unfortunately, we don't
543 	 * really keep track of individual mmaps so we approximate by flushing
544 	 * the range of the map entry containing addr. This can be incorrect
545 	 * if the region splits or is coalesced with a neighbor.
546 	 */
547 	if (size == 0) {
548 		vm_map_entry_t entry;
549 
550 		vm_map_lock_read(map);
551 		rv = vm_map_lookup_entry(map, addr, &entry);
552 		vm_map_unlock_read(map);
553 		if (rv == FALSE) {
554 			rv = -1;
555 			goto done2;
556 		}
557 		addr = entry->start;
558 		size = entry->end - entry->start;
559 	}
560 
561 	/*
562 	 * Clean the pages and interpret the return value.
563 	 */
564 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
565 	    (flags & MS_INVALIDATE) != 0);
566 
567 done2:
568 	mtx_unlock(&Giant);
569 
570 	switch (rv) {
571 	case KERN_SUCCESS:
572 		return (0);
573 	case KERN_INVALID_ADDRESS:
574 		return (EINVAL);	/* Sun returns ENOMEM? */
575 	case KERN_FAILURE:
576 		return (EIO);
577 	default:
578 		return (EINVAL);
579 	}
580 }
581 
582 #ifndef _SYS_SYSPROTO_H_
583 struct munmap_args {
584 	void *addr;
585 	size_t len;
586 };
587 #endif
588 /*
589  * MPSAFE
590  */
591 int
592 munmap(td, uap)
593 	struct thread *td;
594 	struct munmap_args *uap;
595 {
596 	vm_offset_t addr;
597 	vm_size_t size, pageoff;
598 	vm_map_t map;
599 
600 	addr = (vm_offset_t) uap->addr;
601 	size = uap->len;
602 
603 	pageoff = (addr & PAGE_MASK);
604 	addr -= pageoff;
605 	size += pageoff;
606 	size = (vm_size_t) round_page(size);
607 	if (addr + size < addr)
608 		return (EINVAL);
609 
610 	if (size == 0)
611 		return (0);
612 
613 	/*
614 	 * Check for illegal addresses.  Watch out for address wrap... Note
615 	 * that VM_*_ADDRESS are not constants due to casts (argh).
616 	 */
617 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
618 		return (EINVAL);
619 #ifndef __i386__
620 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
621 		return (EINVAL);
622 #endif
623 	map = &td->td_proc->p_vmspace->vm_map;
624 	/*
625 	 * Make sure entire range is allocated.
626 	 */
627 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
628 		return (EINVAL);
629 
630 	/* returns nothing but KERN_SUCCESS anyway */
631 	(void) vm_map_remove(map, addr, addr + size);
632 	return (0);
633 }
634 
635 #if 0
636 void
637 munmapfd(td, fd)
638 	struct thread *td;
639 	int fd;
640 {
641 	/*
642 	 * XXX should unmap any regions mapped to this file
643 	 */
644 	FILEDESC_LOCK(p->p_fd);
645 	td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED;
646 	FILEDESC_UNLOCK(p->p_fd);
647 }
648 #endif
649 
650 #ifndef _SYS_SYSPROTO_H_
651 struct mprotect_args {
652 	const void *addr;
653 	size_t len;
654 	int prot;
655 };
656 #endif
657 /*
658  * MPSAFE
659  */
660 int
661 mprotect(td, uap)
662 	struct thread *td;
663 	struct mprotect_args *uap;
664 {
665 	vm_offset_t addr;
666 	vm_size_t size, pageoff;
667 	vm_prot_t prot;
668 
669 	addr = (vm_offset_t) uap->addr;
670 	size = uap->len;
671 	prot = uap->prot & VM_PROT_ALL;
672 #if defined(VM_PROT_READ_IS_EXEC)
673 	if (prot & VM_PROT_READ)
674 		prot |= VM_PROT_EXECUTE;
675 #endif
676 
677 	pageoff = (addr & PAGE_MASK);
678 	addr -= pageoff;
679 	size += pageoff;
680 	size = (vm_size_t) round_page(size);
681 	if (addr + size < addr)
682 		return (EINVAL);
683 
684 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
685 	    addr + size, prot, FALSE)) {
686 	case KERN_SUCCESS:
687 		return (0);
688 	case KERN_PROTECTION_FAILURE:
689 		return (EACCES);
690 	}
691 	return (EINVAL);
692 }
693 
694 #ifndef _SYS_SYSPROTO_H_
695 struct minherit_args {
696 	void *addr;
697 	size_t len;
698 	int inherit;
699 };
700 #endif
701 /*
702  * MPSAFE
703  */
704 int
705 minherit(td, uap)
706 	struct thread *td;
707 	struct minherit_args *uap;
708 {
709 	vm_offset_t addr;
710 	vm_size_t size, pageoff;
711 	vm_inherit_t inherit;
712 
713 	addr = (vm_offset_t)uap->addr;
714 	size = uap->len;
715 	inherit = uap->inherit;
716 
717 	pageoff = (addr & PAGE_MASK);
718 	addr -= pageoff;
719 	size += pageoff;
720 	size = (vm_size_t) round_page(size);
721 	if (addr + size < addr)
722 		return (EINVAL);
723 
724 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
725 	    addr + size, inherit)) {
726 	case KERN_SUCCESS:
727 		return (0);
728 	case KERN_PROTECTION_FAILURE:
729 		return (EACCES);
730 	}
731 	return (EINVAL);
732 }
733 
734 #ifndef _SYS_SYSPROTO_H_
735 struct madvise_args {
736 	void *addr;
737 	size_t len;
738 	int behav;
739 };
740 #endif
741 
742 /*
743  * MPSAFE
744  */
745 /* ARGSUSED */
746 int
747 madvise(td, uap)
748 	struct thread *td;
749 	struct madvise_args *uap;
750 {
751 	vm_offset_t start, end;
752 
753 	/*
754 	 * Check for illegal behavior
755 	 */
756 	if (uap->behav < 0 || uap->behav > MADV_CORE)
757 		return (EINVAL);
758 	/*
759 	 * Check for illegal addresses.  Watch out for address wrap... Note
760 	 * that VM_*_ADDRESS are not constants due to casts (argh).
761 	 */
762 	if (VM_MAXUSER_ADDRESS > 0 &&
763 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
764 		return (EINVAL);
765 #ifndef __i386__
766 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
767 		return (EINVAL);
768 #endif
769 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
770 		return (EINVAL);
771 
772 	/*
773 	 * Since this routine is only advisory, we default to conservative
774 	 * behavior.
775 	 */
776 	start = trunc_page((vm_offset_t) uap->addr);
777 	end = round_page((vm_offset_t) uap->addr + uap->len);
778 
779 	if (vm_map_madvise(&td->td_proc->p_vmspace->vm_map, start, end,
780 	    uap->behav))
781 		return (EINVAL);
782 	return (0);
783 }
784 
785 #ifndef _SYS_SYSPROTO_H_
786 struct mincore_args {
787 	const void *addr;
788 	size_t len;
789 	char *vec;
790 };
791 #endif
792 
793 /*
794  * MPSAFE
795  */
796 /* ARGSUSED */
797 int
798 mincore(td, uap)
799 	struct thread *td;
800 	struct mincore_args *uap;
801 {
802 	vm_offset_t addr, first_addr;
803 	vm_offset_t end, cend;
804 	pmap_t pmap;
805 	vm_map_t map;
806 	char *vec;
807 	int error = 0;
808 	int vecindex, lastvecindex;
809 	vm_map_entry_t current;
810 	vm_map_entry_t entry;
811 	int mincoreinfo;
812 	unsigned int timestamp;
813 
814 	/*
815 	 * Make sure that the addresses presented are valid for user
816 	 * mode.
817 	 */
818 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
819 	end = addr + (vm_size_t)round_page(uap->len);
820 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
821 		return (EINVAL);
822 	if (end < addr)
823 		return (EINVAL);
824 
825 	/*
826 	 * Address of byte vector
827 	 */
828 	vec = uap->vec;
829 
830 	mtx_lock(&Giant);
831 	map = &td->td_proc->p_vmspace->vm_map;
832 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
833 
834 	vm_map_lock_read(map);
835 RestartScan:
836 	timestamp = map->timestamp;
837 
838 	if (!vm_map_lookup_entry(map, addr, &entry))
839 		entry = entry->next;
840 
841 	/*
842 	 * Do this on a map entry basis so that if the pages are not
843 	 * in the current processes address space, we can easily look
844 	 * up the pages elsewhere.
845 	 */
846 	lastvecindex = -1;
847 	for (current = entry;
848 	    (current != &map->header) && (current->start < end);
849 	    current = current->next) {
850 
851 		/*
852 		 * ignore submaps (for now) or null objects
853 		 */
854 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
855 			current->object.vm_object == NULL)
856 			continue;
857 
858 		/*
859 		 * limit this scan to the current map entry and the
860 		 * limits for the mincore call
861 		 */
862 		if (addr < current->start)
863 			addr = current->start;
864 		cend = current->end;
865 		if (cend > end)
866 			cend = end;
867 
868 		/*
869 		 * scan this entry one page at a time
870 		 */
871 		while (addr < cend) {
872 			/*
873 			 * Check pmap first, it is likely faster, also
874 			 * it can provide info as to whether we are the
875 			 * one referencing or modifying the page.
876 			 */
877 			mincoreinfo = pmap_mincore(pmap, addr);
878 			if (!mincoreinfo) {
879 				vm_pindex_t pindex;
880 				vm_ooffset_t offset;
881 				vm_page_t m;
882 				/*
883 				 * calculate the page index into the object
884 				 */
885 				offset = current->offset + (addr - current->start);
886 				pindex = OFF_TO_IDX(offset);
887 				m = vm_page_lookup(current->object.vm_object,
888 					pindex);
889 				/*
890 				 * if the page is resident, then gather information about
891 				 * it.
892 				 */
893 				if (m) {
894 					mincoreinfo = MINCORE_INCORE;
895 					if (m->dirty ||
896 						pmap_is_modified(m))
897 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
898 					if ((m->flags & PG_REFERENCED) ||
899 						pmap_ts_referenced(m)) {
900 						vm_page_flag_set(m, PG_REFERENCED);
901 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
902 					}
903 				}
904 			}
905 
906 			/*
907 			 * subyte may page fault.  In case it needs to modify
908 			 * the map, we release the lock.
909 			 */
910 			vm_map_unlock_read(map);
911 
912 			/*
913 			 * calculate index into user supplied byte vector
914 			 */
915 			vecindex = OFF_TO_IDX(addr - first_addr);
916 
917 			/*
918 			 * If we have skipped map entries, we need to make sure that
919 			 * the byte vector is zeroed for those skipped entries.
920 			 */
921 			while ((lastvecindex + 1) < vecindex) {
922 				error = subyte(vec + lastvecindex, 0);
923 				if (error) {
924 					error = EFAULT;
925 					goto done2;
926 				}
927 				++lastvecindex;
928 			}
929 
930 			/*
931 			 * Pass the page information to the user
932 			 */
933 			error = subyte(vec + vecindex, mincoreinfo);
934 			if (error) {
935 				error = EFAULT;
936 				goto done2;
937 			}
938 
939 			/*
940 			 * If the map has changed, due to the subyte, the previous
941 			 * output may be invalid.
942 			 */
943 			vm_map_lock_read(map);
944 			if (timestamp != map->timestamp)
945 				goto RestartScan;
946 
947 			lastvecindex = vecindex;
948 			addr += PAGE_SIZE;
949 		}
950 	}
951 
952 	/*
953 	 * subyte may page fault.  In case it needs to modify
954 	 * the map, we release the lock.
955 	 */
956 	vm_map_unlock_read(map);
957 
958 	/*
959 	 * Zero the last entries in the byte vector.
960 	 */
961 	vecindex = OFF_TO_IDX(end - first_addr);
962 	while ((lastvecindex + 1) < vecindex) {
963 		error = subyte(vec + lastvecindex, 0);
964 		if (error) {
965 			error = EFAULT;
966 			goto done2;
967 		}
968 		++lastvecindex;
969 	}
970 
971 	/*
972 	 * If the map has changed, due to the subyte, the previous
973 	 * output may be invalid.
974 	 */
975 	vm_map_lock_read(map);
976 	if (timestamp != map->timestamp)
977 		goto RestartScan;
978 	vm_map_unlock_read(map);
979 done2:
980 	mtx_unlock(&Giant);
981 	return (error);
982 }
983 
984 #ifndef _SYS_SYSPROTO_H_
985 struct mlock_args {
986 	const void *addr;
987 	size_t len;
988 };
989 #endif
990 /*
991  * MPSAFE
992  */
993 int
994 mlock(td, uap)
995 	struct thread *td;
996 	struct mlock_args *uap;
997 {
998 	vm_offset_t addr;
999 	vm_size_t size, pageoff;
1000 	int error;
1001 
1002 	addr = (vm_offset_t) uap->addr;
1003 	size = uap->len;
1004 
1005 	pageoff = (addr & PAGE_MASK);
1006 	addr -= pageoff;
1007 	size += pageoff;
1008 	size = (vm_size_t) round_page(size);
1009 
1010 	/* disable wrap around */
1011 	if (addr + size < addr)
1012 		return (EINVAL);
1013 
1014 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1015 		return (EAGAIN);
1016 
1017 #ifdef pmap_wired_count
1018 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1019 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1020 		return (ENOMEM);
1021 #else
1022 	error = suser(td);
1023 	if (error)
1024 		return (error);
1025 #endif
1026 
1027 	error = vm_map_wire(&td->td_proc->p_vmspace->vm_map, addr,
1028 		     addr + size, TRUE);
1029 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1030 }
1031 
1032 #ifndef _SYS_SYSPROTO_H_
1033 struct mlockall_args {
1034 	int	how;
1035 };
1036 #endif
1037 
1038 /*
1039  * MPSAFE
1040  */
1041 int
1042 mlockall(td, uap)
1043 	struct thread *td;
1044 	struct mlockall_args *uap;
1045 {
1046 	/* mtx_lock(&Giant); */
1047 	/* mtx_unlock(&Giant); */
1048 	return 0;
1049 }
1050 
1051 #ifndef _SYS_SYSPROTO_H_
1052 struct munlockall_args {
1053 	int	how;
1054 };
1055 #endif
1056 
1057 /*
1058  * MPSAFE
1059  */
1060 int
1061 munlockall(td, uap)
1062 	struct thread *td;
1063 	struct munlockall_args *uap;
1064 {
1065 	/* mtx_lock(&Giant); */
1066 	/* mtx_unlock(&Giant); */
1067 	return 0;
1068 }
1069 
1070 #ifndef _SYS_SYSPROTO_H_
1071 struct munlock_args {
1072 	const void *addr;
1073 	size_t len;
1074 };
1075 #endif
1076 /*
1077  * MPSAFE
1078  */
1079 int
1080 munlock(td, uap)
1081 	struct thread *td;
1082 	struct munlock_args *uap;
1083 {
1084 	vm_offset_t addr;
1085 	vm_size_t size, pageoff;
1086 	int error;
1087 
1088 	addr = (vm_offset_t) uap->addr;
1089 	size = uap->len;
1090 
1091 	pageoff = (addr & PAGE_MASK);
1092 	addr -= pageoff;
1093 	size += pageoff;
1094 	size = (vm_size_t) round_page(size);
1095 
1096 	/* disable wrap around */
1097 	if (addr + size < addr)
1098 		return (EINVAL);
1099 
1100 #ifndef pmap_wired_count
1101 	error = suser(td);
1102 	if (error)
1103 		return (error);
1104 #endif
1105 
1106 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, addr,
1107 		     addr + size, TRUE);
1108 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1109 }
1110 
1111 /*
1112  * vm_mmap()
1113  *
1114  * MPSAFE
1115  *
1116  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1117  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1118  */
1119 int
1120 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1121 	vm_prot_t maxprot, int flags,
1122 	void *handle,
1123 	vm_ooffset_t foff)
1124 {
1125 	boolean_t fitit;
1126 	vm_object_t object;
1127 	struct vnode *vp = NULL;
1128 	objtype_t type;
1129 	int rv = KERN_SUCCESS;
1130 	vm_ooffset_t objsize;
1131 	int docow;
1132 	struct thread *td = curthread;
1133 
1134 	if (size == 0)
1135 		return (0);
1136 
1137 	objsize = size = round_page(size);
1138 
1139 	if (td->td_proc->p_vmspace->vm_map.size + size >
1140 	    td->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1141 		return(ENOMEM);
1142 	}
1143 
1144 	/*
1145 	 * We currently can only deal with page aligned file offsets.
1146 	 * The check is here rather than in the syscall because the
1147 	 * kernel calls this function internally for other mmaping
1148 	 * operations (such as in exec) and non-aligned offsets will
1149 	 * cause pmap inconsistencies...so we want to be sure to
1150 	 * disallow this in all cases.
1151 	 */
1152 	if (foff & PAGE_MASK)
1153 		return (EINVAL);
1154 
1155 	if ((flags & MAP_FIXED) == 0) {
1156 		fitit = TRUE;
1157 		*addr = round_page(*addr);
1158 	} else {
1159 		if (*addr != trunc_page(*addr))
1160 			return (EINVAL);
1161 		fitit = FALSE;
1162 		(void) vm_map_remove(map, *addr, *addr + size);
1163 	}
1164 
1165 	/*
1166 	 * Lookup/allocate object.
1167 	 */
1168 	if (flags & MAP_ANON) {
1169 		type = OBJT_DEFAULT;
1170 		/*
1171 		 * Unnamed anonymous regions always start at 0.
1172 		 */
1173 		if (handle == 0)
1174 			foff = 0;
1175 	} else {
1176 		vp = (struct vnode *) handle;
1177 		mtx_lock(&Giant);
1178 		ASSERT_VOP_LOCKED(vp, "vm_mmap");
1179 		if (vp->v_type == VCHR) {
1180 			type = OBJT_DEVICE;
1181 			handle = (void *)(intptr_t)vp->v_rdev;
1182 		} else {
1183 			struct vattr vat;
1184 			int error;
1185 
1186 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1187 			if (error) {
1188 				mtx_unlock(&Giant);
1189 				return (error);
1190 			}
1191 			objsize = round_page(vat.va_size);
1192 			type = OBJT_VNODE;
1193 			/*
1194 			 * if it is a regular file without any references
1195 			 * we do not need to sync it.
1196 			 */
1197 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1198 				flags |= MAP_NOSYNC;
1199 			}
1200 		}
1201 		mtx_unlock(&Giant);
1202 	}
1203 
1204 	if (handle == NULL) {
1205 		object = NULL;
1206 		docow = 0;
1207 	} else {
1208 		object = vm_pager_allocate(type,
1209 			handle, objsize, prot, foff);
1210 		if (object == NULL) {
1211 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1212 		}
1213 		docow = MAP_PREFAULT_PARTIAL;
1214 	}
1215 
1216 	/*
1217 	 * Force device mappings to be shared.
1218 	 */
1219 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1220 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1221 		flags |= MAP_SHARED;
1222 	}
1223 
1224 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1225 		docow |= MAP_COPY_ON_WRITE;
1226 	if (flags & MAP_NOSYNC)
1227 		docow |= MAP_DISABLE_SYNCER;
1228 	if (flags & MAP_NOCORE)
1229 		docow |= MAP_DISABLE_COREDUMP;
1230 
1231 #if defined(VM_PROT_READ_IS_EXEC)
1232 	if (prot & VM_PROT_READ)
1233 		prot |= VM_PROT_EXECUTE;
1234 
1235 	if (maxprot & VM_PROT_READ)
1236 		maxprot |= VM_PROT_EXECUTE;
1237 #endif
1238 
1239 	if (fitit)
1240 		*addr = pmap_addr_hint(object, *addr, size);
1241 
1242 	if (flags & MAP_STACK)
1243 		rv = vm_map_stack (map, *addr, size, prot,
1244 				   maxprot, docow);
1245 	else
1246 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1247 				 prot, maxprot, docow);
1248 
1249 	if (rv != KERN_SUCCESS) {
1250 		/*
1251 		 * Lose the object reference. Will destroy the
1252 		 * object if it's an unnamed anonymous mapping
1253 		 * or named anonymous without other references.
1254 		 */
1255 		vm_object_deallocate(object);
1256 	} else if (flags & MAP_SHARED) {
1257 		/*
1258 		 * Shared memory is also shared with children.
1259 		 */
1260 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1261 		if (rv != KERN_SUCCESS)
1262 			(void) vm_map_remove(map, *addr, *addr + size);
1263 	}
1264 	switch (rv) {
1265 	case KERN_SUCCESS:
1266 		return (0);
1267 	case KERN_INVALID_ADDRESS:
1268 	case KERN_NO_SPACE:
1269 		return (ENOMEM);
1270 	case KERN_PROTECTION_FAILURE:
1271 		return (EACCES);
1272 	default:
1273 		return (EINVAL);
1274 	}
1275 }
1276