xref: /freebsd/sys/vm/vm_mmap.c (revision 0fa02ea5f786ef02befd46f8f083f48c8cd9630b)
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  */
42 
43 /*
44  * Mapped file (mmap) interface to VM
45  */
46 
47 #include <sys/cdefs.h>
48 __FBSDID("$FreeBSD$");
49 
50 #include "opt_compat.h"
51 #include "opt_mac.h"
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
60 #include <sys/proc.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/vnode.h>
64 #include <sys/fcntl.h>
65 #include <sys/file.h>
66 #include <sys/mac.h>
67 #include <sys/mman.h>
68 #include <sys/conf.h>
69 #include <sys/stat.h>
70 #include <sys/vmmeter.h>
71 #include <sys/sysctl.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 #ifndef _SYS_SYSPROTO_H_
86 struct sbrk_args {
87 	int incr;
88 };
89 #endif
90 
91 static int max_proc_mmap;
92 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
93 
94 /*
95  * Set the maximum number of vm_map_entry structures per process.  Roughly
96  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
97  * of our KVM malloc space still results in generous limits.  We want a
98  * default that is good enough to prevent the kernel running out of resources
99  * if attacked from compromised user account but generous enough such that
100  * multi-threaded processes are not unduly inconvenienced.
101  */
102 static void vmmapentry_rsrc_init(void *);
103 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
104 
105 static void
106 vmmapentry_rsrc_init(dummy)
107         void *dummy;
108 {
109     max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
110     max_proc_mmap /= 100;
111 }
112 
113 /*
114  * MPSAFE
115  */
116 /* ARGSUSED */
117 int
118 sbrk(td, uap)
119 	struct thread *td;
120 	struct sbrk_args *uap;
121 {
122 	/* Not yet implemented */
123 	/* mtx_lock(&Giant); */
124 	/* mtx_unlock(&Giant); */
125 	return (EOPNOTSUPP);
126 }
127 
128 #ifndef _SYS_SYSPROTO_H_
129 struct sstk_args {
130 	int incr;
131 };
132 #endif
133 
134 /*
135  * MPSAFE
136  */
137 /* ARGSUSED */
138 int
139 sstk(td, uap)
140 	struct thread *td;
141 	struct sstk_args *uap;
142 {
143 	/* Not yet implemented */
144 	/* mtx_lock(&Giant); */
145 	/* mtx_unlock(&Giant); */
146 	return (EOPNOTSUPP);
147 }
148 
149 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
150 #ifndef _SYS_SYSPROTO_H_
151 struct getpagesize_args {
152 	int dummy;
153 };
154 #endif
155 
156 /* ARGSUSED */
157 int
158 ogetpagesize(td, uap)
159 	struct thread *td;
160 	struct getpagesize_args *uap;
161 {
162 	/* MP SAFE */
163 	td->td_retval[0] = PAGE_SIZE;
164 	return (0);
165 }
166 #endif				/* COMPAT_43 || COMPAT_SUNOS */
167 
168 
169 /*
170  * Memory Map (mmap) system call.  Note that the file offset
171  * and address are allowed to be NOT page aligned, though if
172  * the MAP_FIXED flag it set, both must have the same remainder
173  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
174  * page-aligned, the actual mapping starts at trunc_page(addr)
175  * and the return value is adjusted up by the page offset.
176  *
177  * Generally speaking, only character devices which are themselves
178  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
179  * there would be no cache coherency between a descriptor and a VM mapping
180  * both to the same character device.
181  *
182  * Block devices can be mmap'd no matter what they represent.  Cache coherency
183  * is maintained as long as you do not write directly to the underlying
184  * character device.
185  */
186 #ifndef _SYS_SYSPROTO_H_
187 struct mmap_args {
188 	void *addr;
189 	size_t len;
190 	int prot;
191 	int flags;
192 	int fd;
193 	long pad;
194 	off_t pos;
195 };
196 #endif
197 
198 /*
199  * MPSAFE
200  */
201 int
202 mmap(td, uap)
203 	struct thread *td;
204 	struct mmap_args *uap;
205 {
206 	struct file *fp = NULL;
207 	struct vnode *vp;
208 	vm_offset_t addr;
209 	vm_size_t size, pageoff;
210 	vm_prot_t prot, maxprot;
211 	void *handle;
212 	int flags, error;
213 	int disablexworkaround;
214 	off_t pos;
215 	struct vmspace *vms = td->td_proc->p_vmspace;
216 	vm_object_t obj;
217 
218 	addr = (vm_offset_t) uap->addr;
219 	size = uap->len;
220 	prot = uap->prot & VM_PROT_ALL;
221 	flags = uap->flags;
222 	pos = uap->pos;
223 
224 	vp = NULL;
225 	fp = NULL;
226 	/* make sure mapping fits into numeric range etc */
227 	if ((ssize_t) uap->len < 0 ||
228 	    ((flags & MAP_ANON) && uap->fd != -1))
229 		return (EINVAL);
230 
231 	if (flags & MAP_STACK) {
232 		if ((uap->fd != -1) ||
233 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
234 			return (EINVAL);
235 		flags |= MAP_ANON;
236 		pos = 0;
237 	}
238 
239 	/*
240 	 * Align the file position to a page boundary,
241 	 * and save its page offset component.
242 	 */
243 	pageoff = (pos & PAGE_MASK);
244 	pos -= pageoff;
245 
246 	/* Adjust size for rounding (on both ends). */
247 	size += pageoff;			/* low end... */
248 	size = (vm_size_t) round_page(size);	/* hi end */
249 
250 	/*
251 	 * Check for illegal addresses.  Watch out for address wrap... Note
252 	 * that VM_*_ADDRESS are not constants due to casts (argh).
253 	 */
254 	if (flags & MAP_FIXED) {
255 		/*
256 		 * The specified address must have the same remainder
257 		 * as the file offset taken modulo PAGE_SIZE, so it
258 		 * should be aligned after adjustment by pageoff.
259 		 */
260 		addr -= pageoff;
261 		if (addr & PAGE_MASK)
262 			return (EINVAL);
263 		/* Address range must be all in user VM space. */
264 		if (addr < vm_map_min(&vms->vm_map) ||
265 		    addr + size > vm_map_max(&vms->vm_map))
266 			return (EINVAL);
267 		if (addr + size < addr)
268 			return (EINVAL);
269 	}
270 	/*
271 	 * XXX for non-fixed mappings where no hint is provided or
272 	 * the hint would fall in the potential heap space,
273 	 * place it after the end of the largest possible heap.
274 	 *
275 	 * There should really be a pmap call to determine a reasonable
276 	 * location.
277 	 */
278 	else if (addr == 0 ||
279 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
280 	     addr < round_page((vm_offset_t)vms->vm_daddr +
281 	      td->td_proc->p_rlimit[RLIMIT_DATA].rlim_max)))
282 		addr = round_page((vm_offset_t)vms->vm_daddr +
283 		    td->td_proc->p_rlimit[RLIMIT_DATA].rlim_max);
284 
285 	mtx_lock(&Giant);	/* syscall marked mp-safe but isn't */
286 	do {
287 		if (flags & MAP_ANON) {
288 			/*
289 			 * Mapping blank space is trivial.
290 			 */
291 			handle = NULL;
292 			maxprot = VM_PROT_ALL;
293 			pos = 0;
294 			break;
295 		}
296 		/*
297 		 * Mapping file, get fp for validation. Obtain vnode and make
298 		 * sure it is of appropriate type.
299 		 * don't let the descriptor disappear on us if we block
300 		 */
301 		if ((error = fget(td, uap->fd, &fp)) != 0)
302 			goto done;
303 		if (fp->f_type != DTYPE_VNODE) {
304 			error = EINVAL;
305 			goto done;
306 		}
307 
308 		/*
309 		 * POSIX shared-memory objects are defined to have
310 		 * kernel persistence, and are not defined to support
311 		 * read(2)/write(2) -- or even open(2).  Thus, we can
312 		 * use MAP_ASYNC to trade on-disk coherence for speed.
313 		 * The shm_open(3) library routine turns on the FPOSIXSHM
314 		 * flag to request this behavior.
315 		 */
316 		if (fp->f_flag & FPOSIXSHM)
317 			flags |= MAP_NOSYNC;
318 		vp = fp->f_vnode;
319 		error = vget(vp, LK_EXCLUSIVE, td);
320 		if (error)
321 			goto done;
322 		if (vp->v_type != VREG && vp->v_type != VCHR) {
323 			error = EINVAL;
324 			goto done;
325 		}
326 		if (vp->v_type == VREG) {
327 			/*
328 			 * Get the proper underlying object
329 			 */
330 			if (VOP_GETVOBJECT(vp, &obj) != 0) {
331 				error = EINVAL;
332 				goto done;
333 			}
334 			if (obj->handle != vp) {
335 				vput(vp);
336 				vp = (struct vnode*)obj->handle;
337 				vget(vp, LK_EXCLUSIVE, td);
338 			}
339 		}
340 		/*
341 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
342 		 * SunOS).
343 		 */
344 		if ((vp->v_type == VCHR) &&
345 		    (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) {
346 			handle = NULL;
347 			maxprot = VM_PROT_ALL;
348 			flags |= MAP_ANON;
349 			pos = 0;
350 			break;
351 		}
352 		/*
353 		 * cdevs does not provide private mappings of any kind.
354 		 */
355 		/*
356 		 * However, for XIG X server to continue to work,
357 		 * we should allow the superuser to do it anyway.
358 		 * We only allow it at securelevel < 1.
359 		 * (Because the XIG X server writes directly to video
360 		 * memory via /dev/mem, it should never work at any
361 		 * other securelevel.
362 		 * XXX this will have to go
363 		 */
364 		if (securelevel_ge(td->td_ucred, 1))
365 			disablexworkaround = 1;
366 		else
367 			disablexworkaround = suser(td);
368 		if (vp->v_type == VCHR && disablexworkaround &&
369 		    (flags & (MAP_PRIVATE|MAP_COPY))) {
370 			error = EINVAL;
371 			goto done;
372 		}
373 		/*
374 		 * Ensure that file and memory protections are
375 		 * compatible.  Note that we only worry about
376 		 * writability if mapping is shared; in this case,
377 		 * current and max prot are dictated by the open file.
378 		 * XXX use the vnode instead?  Problem is: what
379 		 * credentials do we use for determination? What if
380 		 * proc does a setuid?
381 		 */
382 		maxprot = VM_PROT_EXECUTE;	/* ??? */
383 		if (fp->f_flag & FREAD) {
384 			maxprot |= VM_PROT_READ;
385 		} else if (prot & PROT_READ) {
386 			error = EACCES;
387 			goto done;
388 		}
389 		/*
390 		 * If we are sharing potential changes (either via
391 		 * MAP_SHARED or via the implicit sharing of character
392 		 * device mappings), and we are trying to get write
393 		 * permission although we opened it without asking
394 		 * for it, bail out.  Check for superuser, only if
395 		 * we're at securelevel < 1, to allow the XIG X server
396 		 * to continue to work.
397 		 */
398 		if ((flags & MAP_SHARED) != 0 ||
399 		    (vp->v_type == VCHR && disablexworkaround)) {
400 			if ((fp->f_flag & FWRITE) != 0) {
401 				struct vattr va;
402 				if ((error =
403 				    VOP_GETATTR(vp, &va,
404 						td->td_ucred, td))) {
405 					goto done;
406 				}
407 				if ((va.va_flags &
408 				   (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) {
409 					maxprot |= VM_PROT_WRITE;
410 				} else if (prot & PROT_WRITE) {
411 					error = EPERM;
412 					goto done;
413 				}
414 			} else if ((prot & PROT_WRITE) != 0) {
415 				error = EACCES;
416 				goto done;
417 			}
418 		} else {
419 			maxprot |= VM_PROT_WRITE;
420 		}
421 
422 		handle = (void *)vp;
423 	} while (0);
424 
425 	/*
426 	 * Do not allow more then a certain number of vm_map_entry structures
427 	 * per process.  Scale with the number of rforks sharing the map
428 	 * to make the limit reasonable for threads.
429 	 */
430 	if (max_proc_mmap &&
431 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
432 		error = ENOMEM;
433 		goto done;
434 	}
435 
436 	error = 0;
437 #ifdef MAC
438 	if (handle != NULL && (flags & MAP_SHARED) != 0) {
439 		error = mac_check_vnode_mmap(td->td_ucred,
440 		    (struct vnode *)handle, prot);
441 	}
442 #endif
443 	if (vp != NULL) {
444 		vput(vp);
445 		vp = NULL;
446 	}
447 	mtx_unlock(&Giant);
448 	if (error == 0)
449 		error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
450 		    flags, handle, pos);
451 	mtx_lock(&Giant);
452 	if (error == 0)
453 		td->td_retval[0] = (register_t) (addr + pageoff);
454 done:
455 	if (vp)
456 		vput(vp);
457 	mtx_unlock(&Giant);
458 	if (fp)
459 		fdrop(fp, td);
460 
461 	return (error);
462 }
463 
464 #ifdef COMPAT_43
465 #ifndef _SYS_SYSPROTO_H_
466 struct ommap_args {
467 	caddr_t addr;
468 	int len;
469 	int prot;
470 	int flags;
471 	int fd;
472 	long pos;
473 };
474 #endif
475 int
476 ommap(td, uap)
477 	struct thread *td;
478 	struct ommap_args *uap;
479 {
480 	struct mmap_args nargs;
481 	static const char cvtbsdprot[8] = {
482 		0,
483 		PROT_EXEC,
484 		PROT_WRITE,
485 		PROT_EXEC | PROT_WRITE,
486 		PROT_READ,
487 		PROT_EXEC | PROT_READ,
488 		PROT_WRITE | PROT_READ,
489 		PROT_EXEC | PROT_WRITE | PROT_READ,
490 	};
491 
492 #define	OMAP_ANON	0x0002
493 #define	OMAP_COPY	0x0020
494 #define	OMAP_SHARED	0x0010
495 #define	OMAP_FIXED	0x0100
496 
497 	nargs.addr = uap->addr;
498 	nargs.len = uap->len;
499 	nargs.prot = cvtbsdprot[uap->prot & 0x7];
500 	nargs.flags = 0;
501 	if (uap->flags & OMAP_ANON)
502 		nargs.flags |= MAP_ANON;
503 	if (uap->flags & OMAP_COPY)
504 		nargs.flags |= MAP_COPY;
505 	if (uap->flags & OMAP_SHARED)
506 		nargs.flags |= MAP_SHARED;
507 	else
508 		nargs.flags |= MAP_PRIVATE;
509 	if (uap->flags & OMAP_FIXED)
510 		nargs.flags |= MAP_FIXED;
511 	nargs.fd = uap->fd;
512 	nargs.pos = uap->pos;
513 	return (mmap(td, &nargs));
514 }
515 #endif				/* COMPAT_43 */
516 
517 
518 #ifndef _SYS_SYSPROTO_H_
519 struct msync_args {
520 	void *addr;
521 	int len;
522 	int flags;
523 };
524 #endif
525 /*
526  * MPSAFE
527  */
528 int
529 msync(td, uap)
530 	struct thread *td;
531 	struct msync_args *uap;
532 {
533 	vm_offset_t addr;
534 	vm_size_t size, pageoff;
535 	int flags;
536 	vm_map_t map;
537 	int rv;
538 
539 	addr = (vm_offset_t) uap->addr;
540 	size = uap->len;
541 	flags = uap->flags;
542 
543 	pageoff = (addr & PAGE_MASK);
544 	addr -= pageoff;
545 	size += pageoff;
546 	size = (vm_size_t) round_page(size);
547 	if (addr + size < addr)
548 		return (EINVAL);
549 
550 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
551 		return (EINVAL);
552 
553 	map = &td->td_proc->p_vmspace->vm_map;
554 
555 	/*
556 	 * Clean the pages and interpret the return value.
557 	 */
558 	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
559 	    (flags & MS_INVALIDATE) != 0);
560 	switch (rv) {
561 	case KERN_SUCCESS:
562 		return (0);
563 	case KERN_INVALID_ADDRESS:
564 		return (EINVAL);	/* Sun returns ENOMEM? */
565 	case KERN_INVALID_ARGUMENT:
566 		return (EBUSY);
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 	if (size == 0)
593 		return (EINVAL);
594 
595 	pageoff = (addr & PAGE_MASK);
596 	addr -= pageoff;
597 	size += pageoff;
598 	size = (vm_size_t) round_page(size);
599 	if (addr + size < addr)
600 		return (EINVAL);
601 
602 	/*
603 	 * Check for illegal addresses.  Watch out for address wrap...
604 	 */
605 	map = &td->td_proc->p_vmspace->vm_map;
606 	if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
607 		return (EINVAL);
608 	vm_map_lock(map);
609 	/*
610 	 * Make sure entire range is allocated.
611 	 */
612 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) {
613 		vm_map_unlock(map);
614 		return (EINVAL);
615 	}
616 	/* returns nothing but KERN_SUCCESS anyway */
617 	vm_map_delete(map, addr, addr + size);
618 	vm_map_unlock(map);
619 	return (0);
620 }
621 
622 #ifndef _SYS_SYSPROTO_H_
623 struct mprotect_args {
624 	const void *addr;
625 	size_t len;
626 	int prot;
627 };
628 #endif
629 /*
630  * MPSAFE
631  */
632 int
633 mprotect(td, uap)
634 	struct thread *td;
635 	struct mprotect_args *uap;
636 {
637 	vm_offset_t addr;
638 	vm_size_t size, pageoff;
639 	vm_prot_t prot;
640 
641 	addr = (vm_offset_t) uap->addr;
642 	size = uap->len;
643 	prot = uap->prot & VM_PROT_ALL;
644 #if defined(VM_PROT_READ_IS_EXEC)
645 	if (prot & VM_PROT_READ)
646 		prot |= VM_PROT_EXECUTE;
647 #endif
648 
649 	pageoff = (addr & PAGE_MASK);
650 	addr -= pageoff;
651 	size += pageoff;
652 	size = (vm_size_t) round_page(size);
653 	if (addr + size < addr)
654 		return (EINVAL);
655 
656 	switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
657 	    addr + size, prot, FALSE)) {
658 	case KERN_SUCCESS:
659 		return (0);
660 	case KERN_PROTECTION_FAILURE:
661 		return (EACCES);
662 	}
663 	return (EINVAL);
664 }
665 
666 #ifndef _SYS_SYSPROTO_H_
667 struct minherit_args {
668 	void *addr;
669 	size_t len;
670 	int inherit;
671 };
672 #endif
673 /*
674  * MPSAFE
675  */
676 int
677 minherit(td, uap)
678 	struct thread *td;
679 	struct minherit_args *uap;
680 {
681 	vm_offset_t addr;
682 	vm_size_t size, pageoff;
683 	vm_inherit_t inherit;
684 
685 	addr = (vm_offset_t)uap->addr;
686 	size = uap->len;
687 	inherit = uap->inherit;
688 
689 	pageoff = (addr & PAGE_MASK);
690 	addr -= pageoff;
691 	size += pageoff;
692 	size = (vm_size_t) round_page(size);
693 	if (addr + size < addr)
694 		return (EINVAL);
695 
696 	switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
697 	    addr + size, inherit)) {
698 	case KERN_SUCCESS:
699 		return (0);
700 	case KERN_PROTECTION_FAILURE:
701 		return (EACCES);
702 	}
703 	return (EINVAL);
704 }
705 
706 #ifndef _SYS_SYSPROTO_H_
707 struct madvise_args {
708 	void *addr;
709 	size_t len;
710 	int behav;
711 };
712 #endif
713 
714 /*
715  * MPSAFE
716  */
717 /* ARGSUSED */
718 int
719 madvise(td, uap)
720 	struct thread *td;
721 	struct madvise_args *uap;
722 {
723 	vm_offset_t start, end;
724 	vm_map_t map;
725 	struct proc *p;
726 	int error;
727 
728 	/*
729 	 * Check for our special case, advising the swap pager we are
730 	 * "immortal."
731 	 */
732 	if (uap->behav == MADV_PROTECT) {
733 		error = suser(td);
734 		if (error == 0) {
735 			p = td->td_proc;
736 			PROC_LOCK(p);
737 			p->p_flag |= P_PROTECTED;
738 			PROC_UNLOCK(p);
739 		}
740 		return (error);
741 	}
742 	/*
743 	 * Check for illegal behavior
744 	 */
745 	if (uap->behav < 0 || uap->behav > MADV_CORE)
746 		return (EINVAL);
747 	/*
748 	 * Check for illegal addresses.  Watch out for address wrap... Note
749 	 * that VM_*_ADDRESS are not constants due to casts (argh).
750 	 */
751 	map = &td->td_proc->p_vmspace->vm_map;
752 	if ((vm_offset_t)uap->addr < vm_map_min(map) ||
753 	    (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
754 		return (EINVAL);
755 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
756 		return (EINVAL);
757 
758 	/*
759 	 * Since this routine is only advisory, we default to conservative
760 	 * behavior.
761 	 */
762 	start = trunc_page((vm_offset_t) uap->addr);
763 	end = round_page((vm_offset_t) uap->addr + uap->len);
764 
765 	if (vm_map_madvise(map, start, end, uap->behav))
766 		return (EINVAL);
767 	return (0);
768 }
769 
770 #ifndef _SYS_SYSPROTO_H_
771 struct mincore_args {
772 	const void *addr;
773 	size_t len;
774 	char *vec;
775 };
776 #endif
777 
778 /*
779  * MPSAFE
780  */
781 /* ARGSUSED */
782 int
783 mincore(td, uap)
784 	struct thread *td;
785 	struct mincore_args *uap;
786 {
787 	vm_offset_t addr, first_addr;
788 	vm_offset_t end, cend;
789 	pmap_t pmap;
790 	vm_map_t map;
791 	char *vec;
792 	int error = 0;
793 	int vecindex, lastvecindex;
794 	vm_map_entry_t current;
795 	vm_map_entry_t entry;
796 	int mincoreinfo;
797 	unsigned int timestamp;
798 
799 	/*
800 	 * Make sure that the addresses presented are valid for user
801 	 * mode.
802 	 */
803 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
804 	end = addr + (vm_size_t)round_page(uap->len);
805 	map = &td->td_proc->p_vmspace->vm_map;
806 	if (end > vm_map_max(map) || end < addr)
807 		return (EINVAL);
808 
809 	/*
810 	 * Address of byte vector
811 	 */
812 	vec = uap->vec;
813 
814 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
815 
816 	vm_map_lock_read(map);
817 RestartScan:
818 	timestamp = map->timestamp;
819 
820 	if (!vm_map_lookup_entry(map, addr, &entry))
821 		entry = entry->next;
822 
823 	/*
824 	 * Do this on a map entry basis so that if the pages are not
825 	 * in the current processes address space, we can easily look
826 	 * up the pages elsewhere.
827 	 */
828 	lastvecindex = -1;
829 	for (current = entry;
830 	    (current != &map->header) && (current->start < end);
831 	    current = current->next) {
832 
833 		/*
834 		 * ignore submaps (for now) or null objects
835 		 */
836 		if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
837 			current->object.vm_object == NULL)
838 			continue;
839 
840 		/*
841 		 * limit this scan to the current map entry and the
842 		 * limits for the mincore call
843 		 */
844 		if (addr < current->start)
845 			addr = current->start;
846 		cend = current->end;
847 		if (cend > end)
848 			cend = end;
849 
850 		/*
851 		 * scan this entry one page at a time
852 		 */
853 		while (addr < cend) {
854 			/*
855 			 * Check pmap first, it is likely faster, also
856 			 * it can provide info as to whether we are the
857 			 * one referencing or modifying the page.
858 			 */
859 			mtx_lock(&Giant);
860 			mincoreinfo = pmap_mincore(pmap, addr);
861 			mtx_unlock(&Giant);
862 			if (!mincoreinfo) {
863 				vm_pindex_t pindex;
864 				vm_ooffset_t offset;
865 				vm_page_t m;
866 				/*
867 				 * calculate the page index into the object
868 				 */
869 				offset = current->offset + (addr - current->start);
870 				pindex = OFF_TO_IDX(offset);
871 				VM_OBJECT_LOCK(current->object.vm_object);
872 				m = vm_page_lookup(current->object.vm_object,
873 					pindex);
874 				/*
875 				 * if the page is resident, then gather information about
876 				 * it.
877 				 */
878 				if (m != NULL && m->valid != 0) {
879 					mincoreinfo = MINCORE_INCORE;
880 					vm_page_lock_queues();
881 					if (m->dirty ||
882 						pmap_is_modified(m))
883 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
884 					if ((m->flags & PG_REFERENCED) ||
885 						pmap_ts_referenced(m)) {
886 						vm_page_flag_set(m, PG_REFERENCED);
887 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
888 					}
889 					vm_page_unlock_queues();
890 				}
891 				VM_OBJECT_UNLOCK(current->object.vm_object);
892 			}
893 
894 			/*
895 			 * subyte may page fault.  In case it needs to modify
896 			 * the map, we release the lock.
897 			 */
898 			vm_map_unlock_read(map);
899 
900 			/*
901 			 * calculate index into user supplied byte vector
902 			 */
903 			vecindex = OFF_TO_IDX(addr - first_addr);
904 
905 			/*
906 			 * If we have skipped map entries, we need to make sure that
907 			 * the byte vector is zeroed for those skipped entries.
908 			 */
909 			while ((lastvecindex + 1) < vecindex) {
910 				error = subyte(vec + lastvecindex, 0);
911 				if (error) {
912 					error = EFAULT;
913 					goto done2;
914 				}
915 				++lastvecindex;
916 			}
917 
918 			/*
919 			 * Pass the page information to the user
920 			 */
921 			error = subyte(vec + vecindex, mincoreinfo);
922 			if (error) {
923 				error = EFAULT;
924 				goto done2;
925 			}
926 
927 			/*
928 			 * If the map has changed, due to the subyte, the previous
929 			 * output may be invalid.
930 			 */
931 			vm_map_lock_read(map);
932 			if (timestamp != map->timestamp)
933 				goto RestartScan;
934 
935 			lastvecindex = vecindex;
936 			addr += PAGE_SIZE;
937 		}
938 	}
939 
940 	/*
941 	 * subyte may page fault.  In case it needs to modify
942 	 * the map, we release the lock.
943 	 */
944 	vm_map_unlock_read(map);
945 
946 	/*
947 	 * Zero the last entries in the byte vector.
948 	 */
949 	vecindex = OFF_TO_IDX(end - first_addr);
950 	while ((lastvecindex + 1) < vecindex) {
951 		error = subyte(vec + lastvecindex, 0);
952 		if (error) {
953 			error = EFAULT;
954 			goto done2;
955 		}
956 		++lastvecindex;
957 	}
958 
959 	/*
960 	 * If the map has changed, due to the subyte, the previous
961 	 * output may be invalid.
962 	 */
963 	vm_map_lock_read(map);
964 	if (timestamp != map->timestamp)
965 		goto RestartScan;
966 	vm_map_unlock_read(map);
967 done2:
968 	return (error);
969 }
970 
971 #ifndef _SYS_SYSPROTO_H_
972 struct mlock_args {
973 	const void *addr;
974 	size_t len;
975 };
976 #endif
977 /*
978  * MPSAFE
979  */
980 int
981 mlock(td, uap)
982 	struct thread *td;
983 	struct mlock_args *uap;
984 {
985 	vm_offset_t addr;
986 	vm_size_t size, pageoff;
987 	int error;
988 
989 	addr = (vm_offset_t) uap->addr;
990 	size = uap->len;
991 
992 	pageoff = (addr & PAGE_MASK);
993 	addr -= pageoff;
994 	size += pageoff;
995 	size = (vm_size_t) round_page(size);
996 
997 	/* disable wrap around */
998 	if (addr + size < addr)
999 		return (EINVAL);
1000 
1001 	if (atop(size) + cnt.v_wire_count > vm_page_max_wired)
1002 		return (EAGAIN);
1003 
1004 #if 0
1005 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) >
1006 	    td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
1007 		return (ENOMEM);
1008 #else
1009 	error = suser(td);
1010 	if (error)
1011 		return (error);
1012 #endif
1013 
1014 	error = vm_map_wire(&td->td_proc->p_vmspace->vm_map, addr,
1015 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1016 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1017 }
1018 
1019 #ifndef _SYS_SYSPROTO_H_
1020 struct mlockall_args {
1021 	int	how;
1022 };
1023 #endif
1024 
1025 /*
1026  * MPSAFE
1027  */
1028 int
1029 mlockall(td, uap)
1030 	struct thread *td;
1031 	struct mlockall_args *uap;
1032 {
1033 	vm_map_t map;
1034 	int error;
1035 
1036 	map = &td->td_proc->p_vmspace->vm_map;
1037 	error = 0;
1038 
1039 	if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1040 		return (EINVAL);
1041 
1042 #if 0
1043 	/*
1044 	 * If wiring all pages in the process would cause it to exceed
1045 	 * a hard resource limit, return ENOMEM.
1046 	 */
1047 	if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1048 		td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur))
1049 		return (ENOMEM);
1050 #else
1051 	error = suser(td);
1052 	if (error)
1053 		return (error);
1054 #endif
1055 
1056 	if (uap->how & MCL_FUTURE) {
1057 		vm_map_lock(map);
1058 		vm_map_modflags(map, MAP_WIREFUTURE, 0);
1059 		vm_map_unlock(map);
1060 		error = 0;
1061 	}
1062 
1063 	if (uap->how & MCL_CURRENT) {
1064 		/*
1065 		 * P1003.1-2001 mandates that all currently mapped pages
1066 		 * will be memory resident and locked (wired) upon return
1067 		 * from mlockall(). vm_map_wire() will wire pages, by
1068 		 * calling vm_fault_wire() for each page in the region.
1069 		 */
1070 		error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1071 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1072 		error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1073 	}
1074 
1075 	return (error);
1076 }
1077 
1078 #ifndef _SYS_SYSPROTO_H_
1079 struct munlockall_args {
1080 	register_t dummy;
1081 };
1082 #endif
1083 
1084 /*
1085  * MPSAFE
1086  */
1087 int
1088 munlockall(td, uap)
1089 	struct thread *td;
1090 	struct munlockall_args *uap;
1091 {
1092 	vm_map_t map;
1093 	int error;
1094 
1095 	map = &td->td_proc->p_vmspace->vm_map;
1096 	error = suser(td);
1097 	if (error)
1098 		return (error);
1099 
1100 	/* Clear the MAP_WIREFUTURE flag from this vm_map. */
1101 	vm_map_lock(map);
1102 	vm_map_modflags(map, 0, MAP_WIREFUTURE);
1103 	vm_map_unlock(map);
1104 
1105 	/* Forcibly unwire all pages. */
1106 	error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1107 	    VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1108 
1109 	return (error);
1110 }
1111 
1112 #ifndef _SYS_SYSPROTO_H_
1113 struct munlock_args {
1114 	const void *addr;
1115 	size_t len;
1116 };
1117 #endif
1118 /*
1119  * MPSAFE
1120  */
1121 int
1122 munlock(td, uap)
1123 	struct thread *td;
1124 	struct munlock_args *uap;
1125 {
1126 	vm_offset_t addr;
1127 	vm_size_t size, pageoff;
1128 	int error;
1129 
1130 	addr = (vm_offset_t) uap->addr;
1131 	size = uap->len;
1132 
1133 	pageoff = (addr & PAGE_MASK);
1134 	addr -= pageoff;
1135 	size += pageoff;
1136 	size = (vm_size_t) round_page(size);
1137 
1138 	/* disable wrap around */
1139 	if (addr + size < addr)
1140 		return (EINVAL);
1141 
1142 	error = suser(td);
1143 	if (error)
1144 		return (error);
1145 
1146 	error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, addr,
1147 		     addr + size, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1148 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
1149 }
1150 
1151 /*
1152  * vm_mmap()
1153  *
1154  * MPSAFE
1155  *
1156  * Internal version of mmap.  Currently used by mmap, exec, and sys5
1157  * shared memory.  Handle is either a vnode pointer or NULL for MAP_ANON.
1158  */
1159 int
1160 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1161 	vm_prot_t maxprot, int flags,
1162 	void *handle,
1163 	vm_ooffset_t foff)
1164 {
1165 	boolean_t fitit;
1166 	vm_object_t object;
1167 	struct vnode *vp = NULL;
1168 	objtype_t type;
1169 	int rv = KERN_SUCCESS;
1170 	vm_ooffset_t objsize;
1171 	int docow, error;
1172 	struct thread *td = curthread;
1173 
1174 	if (size == 0)
1175 		return (0);
1176 
1177 	objsize = size = round_page(size);
1178 
1179 	if (td->td_proc->p_vmspace->vm_map.size + size >
1180 	    td->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1181 		return(ENOMEM);
1182 	}
1183 
1184 	/*
1185 	 * We currently can only deal with page aligned file offsets.
1186 	 * The check is here rather than in the syscall because the
1187 	 * kernel calls this function internally for other mmaping
1188 	 * operations (such as in exec) and non-aligned offsets will
1189 	 * cause pmap inconsistencies...so we want to be sure to
1190 	 * disallow this in all cases.
1191 	 */
1192 	if (foff & PAGE_MASK)
1193 		return (EINVAL);
1194 
1195 	if ((flags & MAP_FIXED) == 0) {
1196 		fitit = TRUE;
1197 		*addr = round_page(*addr);
1198 	} else {
1199 		if (*addr != trunc_page(*addr))
1200 			return (EINVAL);
1201 		fitit = FALSE;
1202 		(void) vm_map_remove(map, *addr, *addr + size);
1203 	}
1204 
1205 	/*
1206 	 * Lookup/allocate object.
1207 	 */
1208 	if (flags & MAP_ANON) {
1209 		type = OBJT_DEFAULT;
1210 		/*
1211 		 * Unnamed anonymous regions always start at 0.
1212 		 */
1213 		if (handle == 0)
1214 			foff = 0;
1215 	} else {
1216 		vp = (struct vnode *) handle;
1217 		mtx_lock(&Giant);
1218 		error = vget(vp, LK_EXCLUSIVE, td);
1219 		if (error) {
1220 			mtx_unlock(&Giant);
1221 			return (error);
1222 		}
1223 		if (vp->v_type == VCHR) {
1224 			type = OBJT_DEVICE;
1225 			handle = vp->v_rdev;
1226 			vput(vp);
1227 			mtx_unlock(&Giant);
1228 		} else {
1229 			struct vattr vat;
1230 
1231 			error = VOP_GETATTR(vp, &vat, td->td_ucred, td);
1232 			if (error) {
1233 				vput(vp);
1234 				mtx_unlock(&Giant);
1235 				return (error);
1236 			}
1237 			objsize = round_page(vat.va_size);
1238 			type = OBJT_VNODE;
1239 			/*
1240 			 * if it is a regular file without any references
1241 			 * we do not need to sync it.
1242 			 */
1243 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1244 				flags |= MAP_NOSYNC;
1245 			}
1246 		}
1247 	}
1248 
1249 	if (handle == NULL) {
1250 		object = NULL;
1251 		docow = 0;
1252 	} else {
1253 		object = vm_pager_allocate(type,
1254 			handle, objsize, prot, foff);
1255 		if (type == OBJT_VNODE) {
1256 			vput(vp);
1257 			mtx_unlock(&Giant);
1258 		}
1259 		if (object == NULL) {
1260 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1261 		}
1262 		docow = MAP_PREFAULT_PARTIAL;
1263 	}
1264 
1265 	/*
1266 	 * Force device mappings to be shared.
1267 	 */
1268 	if (type == OBJT_DEVICE) {
1269 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1270 		flags |= MAP_SHARED;
1271 	}
1272 
1273 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1274 		docow |= MAP_COPY_ON_WRITE;
1275 	if (flags & MAP_NOSYNC)
1276 		docow |= MAP_DISABLE_SYNCER;
1277 	if (flags & MAP_NOCORE)
1278 		docow |= MAP_DISABLE_COREDUMP;
1279 
1280 #if defined(VM_PROT_READ_IS_EXEC)
1281 	if (prot & VM_PROT_READ)
1282 		prot |= VM_PROT_EXECUTE;
1283 
1284 	if (maxprot & VM_PROT_READ)
1285 		maxprot |= VM_PROT_EXECUTE;
1286 #endif
1287 
1288 	if (fitit)
1289 		*addr = pmap_addr_hint(object, *addr, size);
1290 
1291 	if (flags & MAP_STACK)
1292 		rv = vm_map_stack(map, *addr, size, prot, maxprot,
1293 		    docow | MAP_STACK_GROWS_DOWN);
1294 	else
1295 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1296 				 prot, maxprot, docow);
1297 
1298 	if (rv != KERN_SUCCESS) {
1299 		/*
1300 		 * Lose the object reference. Will destroy the
1301 		 * object if it's an unnamed anonymous mapping
1302 		 * or named anonymous without other references.
1303 		 */
1304 		vm_object_deallocate(object);
1305 	} else if (flags & MAP_SHARED) {
1306 		/*
1307 		 * Shared memory is also shared with children.
1308 		 */
1309 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1310 		if (rv != KERN_SUCCESS)
1311 			(void) vm_map_remove(map, *addr, *addr + size);
1312 	}
1313 
1314 	/*
1315 	 * If the process has requested that all future mappings
1316 	 * be wired, then heed this.
1317 	 */
1318 	if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1319 		vm_map_wire(map, *addr, *addr + size,
1320 		    VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1321 
1322 	switch (rv) {
1323 	case KERN_SUCCESS:
1324 		return (0);
1325 	case KERN_INVALID_ADDRESS:
1326 	case KERN_NO_SPACE:
1327 		return (ENOMEM);
1328 	case KERN_PROTECTION_FAILURE:
1329 		return (EACCES);
1330 	default:
1331 		return (EINVAL);
1332 	}
1333 }
1334