xref: /dflybsd-src/sys/vm/vm_mmap.c (revision 1f7ab7c9fc18f47a2f16dc45b13dee254c603ce7)
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: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
42  * $DragonFly: src/sys/vm/vm_mmap.c,v 1.35 2006/10/04 18:28:32 dillon Exp $
43  */
44 
45 /*
46  * Mapped file (mmap) interface to VM
47  */
48 
49 #include <sys/param.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/sysproto.h>
53 #include <sys/filedesc.h>
54 #include <sys/kern_syscall.h>
55 #include <sys/proc.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/vnode.h>
59 #include <sys/fcntl.h>
60 #include <sys/file.h>
61 #include <sys/mman.h>
62 #include <sys/conf.h>
63 #include <sys/stat.h>
64 #include <sys/vmmeter.h>
65 #include <sys/sysctl.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_param.h>
69 #include <sys/lock.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vm_pageout.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_kern.h>
79 
80 #include <sys/file2.h>
81 #include <sys/thread2.h>
82 
83 static int max_proc_mmap;
84 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
85 int vkernel_enable;
86 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
87 
88 /*
89  * Set the maximum number of vm_map_entry structures per process.  Roughly
90  * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
91  * of our KVM malloc space still results in generous limits.  We want a
92  * default that is good enough to prevent the kernel running out of resources
93  * if attacked from compromised user account but generous enough such that
94  * multi-threaded processes are not unduly inconvenienced.
95  */
96 
97 static void vmmapentry_rsrc_init (void *);
98 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
99 
100 static void
101 vmmapentry_rsrc_init(void *dummy)
102 {
103     max_proc_mmap = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
104 			sizeof(struct vm_map_entry);
105     max_proc_mmap /= 100;
106 }
107 
108 /* ARGSUSED */
109 int
110 sys_sbrk(struct sbrk_args *uap)
111 {
112 	/* Not yet implemented */
113 	return (EOPNOTSUPP);
114 }
115 
116 /*
117  * sstk_args(int incr)
118  */
119 /* ARGSUSED */
120 int
121 sys_sstk(struct sstk_args *uap)
122 {
123 	/* Not yet implemented */
124 	return (EOPNOTSUPP);
125 }
126 
127 /*
128  * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
129  *		long pad, off_t pos)
130  *
131  * Memory Map (mmap) system call.  Note that the file offset
132  * and address are allowed to be NOT page aligned, though if
133  * the MAP_FIXED flag it set, both must have the same remainder
134  * modulo the PAGE_SIZE (POSIX 1003.1b).  If the address is not
135  * page-aligned, the actual mapping starts at trunc_page(addr)
136  * and the return value is adjusted up by the page offset.
137  *
138  * Generally speaking, only character devices which are themselves
139  * memory-based, such as a video framebuffer, can be mmap'd.  Otherwise
140  * there would be no cache coherency between a descriptor and a VM mapping
141  * both to the same character device.
142  *
143  * Block devices can be mmap'd no matter what they represent.  Cache coherency
144  * is maintained as long as you do not write directly to the underlying
145  * character device.
146  */
147 
148 int
149 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
150 	  int uprot, int uflags, int fd, off_t upos, void **res)
151 {
152 	struct thread *td = curthread;
153  	struct proc *p = td->td_proc;
154 	struct file *fp = NULL;
155 	struct vnode *vp;
156 	vm_offset_t addr;
157 	vm_size_t size, pageoff;
158 	vm_prot_t prot, maxprot;
159 	void *handle;
160 	int flags, error;
161 	int disablexworkaround;
162 	off_t pos;
163 	vm_object_t obj;
164 
165 	KKASSERT(p);
166 
167 	addr = (vm_offset_t) uaddr;
168 	size = ulen;
169 	prot = uprot & VM_PROT_ALL;
170 	flags = uflags;
171 	pos = upos;
172 
173 	/* make sure mapping fits into numeric range etc */
174 	if ((ssize_t) ulen < 0 || ((flags & MAP_ANON) && fd != -1))
175 		return (EINVAL);
176 
177 	if (flags & MAP_STACK) {
178 		if ((fd != -1) ||
179 		    ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
180 			return (EINVAL);
181 		flags |= MAP_ANON;
182 		pos = 0;
183 	}
184 
185 	/*
186 	 * Virtual page tables cannot be used with MAP_STACK.  Apart from
187 	 * it not making any sense, the aux union is used by both
188 	 * types.
189 	 *
190 	 * Because the virtual page table is stored in the backing object
191 	 * and might be updated by the kernel, the mapping must be R+W.
192 	 */
193 	if (flags & MAP_VPAGETABLE) {
194 		if (vkernel_enable == 0)
195 			return (EOPNOTSUPP);
196 		if (flags & MAP_STACK)
197 			return (EINVAL);
198 		if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
199 			return (EINVAL);
200 	}
201 
202 	/*
203 	 * Align the file position to a page boundary,
204 	 * and save its page offset component.
205 	 */
206 	pageoff = (pos & PAGE_MASK);
207 	pos -= pageoff;
208 
209 	/* Adjust size for rounding (on both ends). */
210 	size += pageoff;			/* low end... */
211 	size = (vm_size_t) round_page(size);	/* hi end */
212 
213 	/*
214 	 * Check for illegal addresses.  Watch out for address wrap... Note
215 	 * that VM_*_ADDRESS are not constants due to casts (argh).
216 	 */
217 	if (flags & MAP_FIXED) {
218 		/*
219 		 * The specified address must have the same remainder
220 		 * as the file offset taken modulo PAGE_SIZE, so it
221 		 * should be aligned after adjustment by pageoff.
222 		 */
223 		addr -= pageoff;
224 		if (addr & PAGE_MASK)
225 			return (EINVAL);
226 		/* Address range must be all in user VM space. */
227 		if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
228 			return (EINVAL);
229 #ifndef i386
230 		if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
231 			return (EINVAL);
232 #endif
233 		if (addr + size < addr)
234 			return (EINVAL);
235 	}
236 	/*
237 	 * XXX for non-fixed mappings where no hint is provided or
238 	 * the hint would fall in the potential heap space,
239 	 * place it after the end of the largest possible heap.
240 	 *
241 	 * There should really be a pmap call to determine a reasonable
242 	 * location.
243 	 */
244 	else if (addr == 0 ||
245 	    (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
246 	     addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
247 		addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
248 
249 	if (flags & MAP_ANON) {
250 		/*
251 		 * Mapping blank space is trivial.
252 		 */
253 		handle = NULL;
254 		maxprot = VM_PROT_ALL;
255 		pos = 0;
256 	} else {
257 		/*
258 		 * Mapping file, get fp for validation. Obtain vnode and make
259 		 * sure it is of appropriate type.
260 		 */
261 		fp = holdfp(p->p_fd, fd, -1);
262 		if (fp == NULL)
263 			return (EBADF);
264 		if (fp->f_type != DTYPE_VNODE) {
265 			error = EINVAL;
266 			goto done;
267 		}
268 		/*
269 		 * POSIX shared-memory objects are defined to have
270 		 * kernel persistence, and are not defined to support
271 		 * read(2)/write(2) -- or even open(2).  Thus, we can
272 		 * use MAP_ASYNC to trade on-disk coherence for speed.
273 		 * The shm_open(3) library routine turns on the FPOSIXSHM
274 		 * flag to request this behavior.
275 		 */
276 		if (fp->f_flag & FPOSIXSHM)
277 			flags |= MAP_NOSYNC;
278 		vp = (struct vnode *) fp->f_data;
279 
280 		/*
281 		 * Validate the vnode for the operation.
282 		 */
283 		switch(vp->v_type) {
284 		case VREG:
285 			/*
286 			 * Get the proper underlying object
287 			 */
288 			if ((obj = vp->v_object) == NULL) {
289 				error = EINVAL;
290 				goto done;
291 			}
292 			KKASSERT((struct vnode *)obj->handle == vp);
293 			break;
294 		case VCHR:
295 			/*
296 			 * Make sure a device has not been revoked.
297 			 * Mappability is handled by the device layer.
298 			 */
299 			if (vp->v_rdev == NULL) {
300 				error = EBADF;
301 				goto done;
302 			}
303 			break;
304 		default:
305 			/*
306 			 * Nothing else is mappable.
307 			 */
308 			error = EINVAL;
309 			goto done;
310 		}
311 
312 		/*
313 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
314 		 * SunOS).
315 		 */
316 		if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
317 			handle = NULL;
318 			maxprot = VM_PROT_ALL;
319 			flags |= MAP_ANON;
320 			pos = 0;
321 		} else {
322 			/*
323 			 * cdevs does not provide private mappings of any kind.
324 			 */
325 			/*
326 			 * However, for XIG X server to continue to work,
327 			 * we should allow the superuser to do it anyway.
328 			 * We only allow it at securelevel < 1.
329 			 * (Because the XIG X server writes directly to video
330 			 * memory via /dev/mem, it should never work at any
331 			 * other securelevel.
332 			 * XXX this will have to go
333 			 */
334 			if (securelevel >= 1)
335 				disablexworkaround = 1;
336 			else
337 				disablexworkaround = suser(td);
338 			if (vp->v_type == VCHR && disablexworkaround &&
339 			    (flags & (MAP_PRIVATE|MAP_COPY))) {
340 				error = EINVAL;
341 				goto done;
342 			}
343 			/*
344 			 * Ensure that file and memory protections are
345 			 * compatible.  Note that we only worry about
346 			 * writability if mapping is shared; in this case,
347 			 * current and max prot are dictated by the open file.
348 			 * XXX use the vnode instead?  Problem is: what
349 			 * credentials do we use for determination? What if
350 			 * proc does a setuid?
351 			 */
352 			maxprot = VM_PROT_EXECUTE;	/* ??? */
353 			if (fp->f_flag & FREAD) {
354 				maxprot |= VM_PROT_READ;
355 			} else if (prot & PROT_READ) {
356 				error = EACCES;
357 				goto done;
358 			}
359 			/*
360 			 * If we are sharing potential changes (either via
361 			 * MAP_SHARED or via the implicit sharing of character
362 			 * device mappings), and we are trying to get write
363 			 * permission although we opened it without asking
364 			 * for it, bail out.  Check for superuser, only if
365 			 * we're at securelevel < 1, to allow the XIG X server
366 			 * to continue to work.
367 			 */
368 
369 			if ((flags & MAP_SHARED) != 0 ||
370 			    (vp->v_type == VCHR && disablexworkaround)) {
371 				if ((fp->f_flag & FWRITE) != 0) {
372 					struct vattr va;
373 					if ((error = VOP_GETATTR(vp, &va))) {
374 						goto done;
375 					}
376 					if ((va.va_flags &
377 					    (IMMUTABLE|APPEND)) == 0) {
378 						maxprot |= VM_PROT_WRITE;
379 					} else if (prot & PROT_WRITE) {
380 						error = EPERM;
381 						goto done;
382 					}
383 				} else if ((prot & PROT_WRITE) != 0) {
384 					error = EACCES;
385 					goto done;
386 				}
387 			} else {
388 				maxprot |= VM_PROT_WRITE;
389 			}
390 			handle = (void *)vp;
391 		}
392 	}
393 
394 	/*
395 	 * Do not allow more then a certain number of vm_map_entry structures
396 	 * per process.  Scale with the number of rforks sharing the map
397 	 * to make the limit reasonable for threads.
398 	 */
399 	if (max_proc_mmap &&
400 	    vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
401 		error = ENOMEM;
402 		goto done;
403 	}
404 
405 	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
406 	    flags, handle, pos);
407 	if (error == 0)
408 		*res = (void *)(addr + pageoff);
409 done:
410 	if (fp)
411 		fdrop(fp);
412 	return (error);
413 }
414 
415 int
416 sys_mmap(struct mmap_args *uap)
417 {
418 	int error;
419 
420 	error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
421 			  uap->prot, uap->flags,
422 			  uap->fd, uap->pos, &uap->sysmsg_resultp);
423 
424 	return (error);
425 }
426 
427 /*
428  * msync_args(void *addr, int len, int flags)
429  */
430 int
431 sys_msync(struct msync_args *uap)
432 {
433 	struct proc *p = curproc;
434 	vm_offset_t addr;
435 	vm_size_t size, pageoff;
436 	int flags;
437 	vm_map_t map;
438 	int rv;
439 
440 	addr = (vm_offset_t) uap->addr;
441 	size = uap->len;
442 	flags = uap->flags;
443 
444 	pageoff = (addr & PAGE_MASK);
445 	addr -= pageoff;
446 	size += pageoff;
447 	size = (vm_size_t) round_page(size);
448 	if (addr + size < addr)
449 		return(EINVAL);
450 
451 	if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
452 		return (EINVAL);
453 
454 	map = &p->p_vmspace->vm_map;
455 
456 	/*
457 	 * XXX Gak!  If size is zero we are supposed to sync "all modified
458 	 * pages with the region containing addr".  Unfortunately, we don't
459 	 * really keep track of individual mmaps so we approximate by flushing
460 	 * the range of the map entry containing addr. This can be incorrect
461 	 * if the region splits or is coalesced with a neighbor.
462 	 */
463 	if (size == 0) {
464 		vm_map_entry_t entry;
465 
466 		vm_map_lock_read(map);
467 		rv = vm_map_lookup_entry(map, addr, &entry);
468 		vm_map_unlock_read(map);
469 		if (rv == FALSE)
470 			return (EINVAL);
471 		addr = entry->start;
472 		size = entry->end - entry->start;
473 	}
474 
475 	/*
476 	 * Clean the pages and interpret the return value.
477 	 */
478 	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
479 	    (flags & MS_INVALIDATE) != 0);
480 
481 	switch (rv) {
482 	case KERN_SUCCESS:
483 		break;
484 	case KERN_INVALID_ADDRESS:
485 		return (EINVAL);	/* Sun returns ENOMEM? */
486 	case KERN_FAILURE:
487 		return (EIO);
488 	default:
489 		return (EINVAL);
490 	}
491 
492 	return (0);
493 }
494 
495 /*
496  * munmap_args(void *addr, size_t len)
497  */
498 int
499 sys_munmap(struct munmap_args *uap)
500 {
501 	struct proc *p = curproc;
502 	vm_offset_t addr;
503 	vm_size_t size, pageoff;
504 	vm_map_t map;
505 
506 	addr = (vm_offset_t) uap->addr;
507 	size = uap->len;
508 
509 	pageoff = (addr & PAGE_MASK);
510 	addr -= pageoff;
511 	size += pageoff;
512 	size = (vm_size_t) round_page(size);
513 	if (addr + size < addr)
514 		return(EINVAL);
515 
516 	if (size == 0)
517 		return (0);
518 
519 	/*
520 	 * Check for illegal addresses.  Watch out for address wrap... Note
521 	 * that VM_*_ADDRESS are not constants due to casts (argh).
522 	 */
523 	if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
524 		return (EINVAL);
525 #ifndef i386
526 	if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
527 		return (EINVAL);
528 #endif
529 	map = &p->p_vmspace->vm_map;
530 	/*
531 	 * Make sure entire range is allocated.
532 	 */
533 	if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
534 		return (EINVAL);
535 	/* returns nothing but KERN_SUCCESS anyway */
536 	vm_map_remove(map, addr, addr + size);
537 	return (0);
538 }
539 
540 /*
541  * mprotect_args(const void *addr, size_t len, int prot)
542  */
543 int
544 sys_mprotect(struct mprotect_args *uap)
545 {
546 	struct proc *p = curproc;
547 	vm_offset_t addr;
548 	vm_size_t size, pageoff;
549 	vm_prot_t prot;
550 
551 	addr = (vm_offset_t) uap->addr;
552 	size = uap->len;
553 	prot = uap->prot & VM_PROT_ALL;
554 #if defined(VM_PROT_READ_IS_EXEC)
555 	if (prot & VM_PROT_READ)
556 		prot |= VM_PROT_EXECUTE;
557 #endif
558 
559 	pageoff = (addr & PAGE_MASK);
560 	addr -= pageoff;
561 	size += pageoff;
562 	size = (vm_size_t) round_page(size);
563 	if (addr + size < addr)
564 		return(EINVAL);
565 
566 	switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
567 		FALSE)) {
568 	case KERN_SUCCESS:
569 		return (0);
570 	case KERN_PROTECTION_FAILURE:
571 		return (EACCES);
572 	}
573 	return (EINVAL);
574 }
575 
576 /*
577  * minherit_args(void *addr, size_t len, int inherit)
578  */
579 int
580 sys_minherit(struct minherit_args *uap)
581 {
582 	struct proc *p = curproc;
583 	vm_offset_t addr;
584 	vm_size_t size, pageoff;
585 	vm_inherit_t inherit;
586 
587 	addr = (vm_offset_t)uap->addr;
588 	size = uap->len;
589 	inherit = uap->inherit;
590 
591 	pageoff = (addr & PAGE_MASK);
592 	addr -= pageoff;
593 	size += pageoff;
594 	size = (vm_size_t) round_page(size);
595 	if (addr + size < addr)
596 		return(EINVAL);
597 
598 	switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
599 	    inherit)) {
600 	case KERN_SUCCESS:
601 		return (0);
602 	case KERN_PROTECTION_FAILURE:
603 		return (EACCES);
604 	}
605 	return (EINVAL);
606 }
607 
608 /*
609  * madvise_args(void *addr, size_t len, int behav)
610  */
611 /* ARGSUSED */
612 int
613 sys_madvise(struct madvise_args *uap)
614 {
615 	struct proc *p = curproc;
616 	vm_offset_t start, end;
617 
618 	/*
619 	 * Check for illegal behavior
620 	 */
621 	if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
622 		return (EINVAL);
623 	/*
624 	 * Check for illegal addresses.  Watch out for address wrap... Note
625 	 * that VM_*_ADDRESS are not constants due to casts (argh).
626 	 */
627 	if (VM_MAXUSER_ADDRESS > 0 &&
628 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
629 		return (EINVAL);
630 #ifndef i386
631 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
632 		return (EINVAL);
633 #endif
634 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
635 		return (EINVAL);
636 
637 	/*
638 	 * Since this routine is only advisory, we default to conservative
639 	 * behavior.
640 	 */
641 	start = trunc_page((vm_offset_t) uap->addr);
642 	end = round_page((vm_offset_t) uap->addr + uap->len);
643 
644 	return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
645 		uap->behav, 0));
646 }
647 
648 /*
649  * mcontrol_args(void *addr, size_t len, int behav, off_t value)
650  */
651 /* ARGSUSED */
652 int
653 sys_mcontrol(struct mcontrol_args *uap)
654 {
655 	struct proc *p = curproc;
656 	vm_offset_t start, end;
657 
658 	/*
659 	 * Check for illegal behavior
660 	 */
661 	if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
662 		return (EINVAL);
663 	/*
664 	 * Check for illegal addresses.  Watch out for address wrap... Note
665 	 * that VM_*_ADDRESS are not constants due to casts (argh).
666 	 */
667 	if (VM_MAXUSER_ADDRESS > 0 &&
668 		((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
669 		return (EINVAL);
670 #ifndef i386
671 	if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
672 		return (EINVAL);
673 #endif
674 	if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
675 		return (EINVAL);
676 
677 	/*
678 	 * Since this routine is only advisory, we default to conservative
679 	 * behavior.
680 	 */
681 	start = trunc_page((vm_offset_t) uap->addr);
682 	end = round_page((vm_offset_t) uap->addr + uap->len);
683 
684 	return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
685 			      uap->behav, uap->value));
686 }
687 
688 
689 /*
690  * mincore_args(const void *addr, size_t len, char *vec)
691  */
692 /* ARGSUSED */
693 int
694 sys_mincore(struct mincore_args *uap)
695 {
696 	struct proc *p = curproc;
697 	vm_offset_t addr, first_addr;
698 	vm_offset_t end, cend;
699 	pmap_t pmap;
700 	vm_map_t map;
701 	char *vec;
702 	int error;
703 	int vecindex, lastvecindex;
704 	vm_map_entry_t current;
705 	vm_map_entry_t entry;
706 	int mincoreinfo;
707 	unsigned int timestamp;
708 
709 	/*
710 	 * Make sure that the addresses presented are valid for user
711 	 * mode.
712 	 */
713 	first_addr = addr = trunc_page((vm_offset_t) uap->addr);
714 	end = addr + (vm_size_t)round_page(uap->len);
715 	if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
716 		return (EINVAL);
717 	if (end < addr)
718 		return (EINVAL);
719 
720 	/*
721 	 * Address of byte vector
722 	 */
723 	vec = uap->vec;
724 
725 	map = &p->p_vmspace->vm_map;
726 	pmap = vmspace_pmap(p->p_vmspace);
727 
728 	vm_map_lock_read(map);
729 RestartScan:
730 	timestamp = map->timestamp;
731 
732 	if (!vm_map_lookup_entry(map, addr, &entry))
733 		entry = entry->next;
734 
735 	/*
736 	 * Do this on a map entry basis so that if the pages are not
737 	 * in the current processes address space, we can easily look
738 	 * up the pages elsewhere.
739 	 */
740 	lastvecindex = -1;
741 	for(current = entry;
742 		(current != &map->header) && (current->start < end);
743 		current = current->next) {
744 
745 		/*
746 		 * ignore submaps (for now) or null objects
747 		 */
748 		if (current->maptype != VM_MAPTYPE_NORMAL &&
749 		    current->maptype != VM_MAPTYPE_VPAGETABLE) {
750 			continue;
751 		}
752 		if (current->object.vm_object == NULL)
753 			continue;
754 
755 		/*
756 		 * limit this scan to the current map entry and the
757 		 * limits for the mincore call
758 		 */
759 		if (addr < current->start)
760 			addr = current->start;
761 		cend = current->end;
762 		if (cend > end)
763 			cend = end;
764 
765 		/*
766 		 * scan this entry one page at a time
767 		 */
768 		while (addr < cend) {
769 			/*
770 			 * Check pmap first, it is likely faster, also
771 			 * it can provide info as to whether we are the
772 			 * one referencing or modifying the page.
773 			 *
774 			 * If we have to check the VM object, only mess
775 			 * around with normal maps.  Do not mess around
776 			 * with virtual page tables (XXX).
777 			 */
778 			mincoreinfo = pmap_mincore(pmap, addr);
779 			if (mincoreinfo == 0 &&
780 			    current->maptype == VM_MAPTYPE_NORMAL) {
781 				vm_pindex_t pindex;
782 				vm_ooffset_t offset;
783 				vm_page_t m;
784 
785 				/*
786 				 * calculate the page index into the object
787 				 */
788 				offset = current->offset + (addr - current->start);
789 				pindex = OFF_TO_IDX(offset);
790 
791 				/*
792 				 * if the page is resident, then gather
793 				 * information about it.  spl protection is
794 				 * required to maintain the object
795 				 * association.  And XXX what if the page is
796 				 * busy?  What's the deal with that?
797 				 */
798 				crit_enter();
799 				m = vm_page_lookup(current->object.vm_object,
800 						    pindex);
801 				if (m && m->valid) {
802 					mincoreinfo = MINCORE_INCORE;
803 					if (m->dirty ||
804 						pmap_is_modified(m))
805 						mincoreinfo |= MINCORE_MODIFIED_OTHER;
806 					if ((m->flags & PG_REFERENCED) ||
807 						pmap_ts_referenced(m)) {
808 						vm_page_flag_set(m, PG_REFERENCED);
809 						mincoreinfo |= MINCORE_REFERENCED_OTHER;
810 					}
811 				}
812 				crit_exit();
813 			}
814 
815 			/*
816 			 * subyte may page fault.  In case it needs to modify
817 			 * the map, we release the lock.
818 			 */
819 			vm_map_unlock_read(map);
820 
821 			/*
822 			 * calculate index into user supplied byte vector
823 			 */
824 			vecindex = OFF_TO_IDX(addr - first_addr);
825 
826 			/*
827 			 * If we have skipped map entries, we need to make sure that
828 			 * the byte vector is zeroed for those skipped entries.
829 			 */
830 			while((lastvecindex + 1) < vecindex) {
831 				error = subyte( vec + lastvecindex, 0);
832 				if (error) {
833 					return (EFAULT);
834 				}
835 				++lastvecindex;
836 			}
837 
838 			/*
839 			 * Pass the page information to the user
840 			 */
841 			error = subyte( vec + vecindex, mincoreinfo);
842 			if (error) {
843 				return (EFAULT);
844 			}
845 
846 			/*
847 			 * If the map has changed, due to the subyte, the previous
848 			 * output may be invalid.
849 			 */
850 			vm_map_lock_read(map);
851 			if (timestamp != map->timestamp)
852 				goto RestartScan;
853 
854 			lastvecindex = vecindex;
855 			addr += PAGE_SIZE;
856 		}
857 	}
858 
859 	/*
860 	 * subyte may page fault.  In case it needs to modify
861 	 * the map, we release the lock.
862 	 */
863 	vm_map_unlock_read(map);
864 
865 	/*
866 	 * Zero the last entries in the byte vector.
867 	 */
868 	vecindex = OFF_TO_IDX(end - first_addr);
869 	while((lastvecindex + 1) < vecindex) {
870 		error = subyte( vec + lastvecindex, 0);
871 		if (error) {
872 			return (EFAULT);
873 		}
874 		++lastvecindex;
875 	}
876 
877 	/*
878 	 * If the map has changed, due to the subyte, the previous
879 	 * output may be invalid.
880 	 */
881 	vm_map_lock_read(map);
882 	if (timestamp != map->timestamp)
883 		goto RestartScan;
884 	vm_map_unlock_read(map);
885 
886 	return (0);
887 }
888 
889 /*
890  * mlock_args(const void *addr, size_t len)
891  */
892 int
893 sys_mlock(struct mlock_args *uap)
894 {
895 	vm_offset_t addr;
896 	vm_size_t size, pageoff;
897 	int error;
898 	struct proc *p = curproc;
899 
900 	addr = (vm_offset_t) uap->addr;
901 	size = uap->len;
902 
903 	pageoff = (addr & PAGE_MASK);
904 	addr -= pageoff;
905 	size += pageoff;
906 	size = (vm_size_t) round_page(size);
907 
908 	/* disable wrap around */
909 	if (addr + size < addr)
910 		return (EINVAL);
911 
912 	if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
913 		return (EAGAIN);
914 
915 #ifdef pmap_wired_count
916 	if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
917 	    p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
918 		return (ENOMEM);
919 #else
920 	error = suser_cred(p->p_ucred, 0);
921 	if (error)
922 		return (error);
923 #endif
924 
925 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
926 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
927 }
928 
929 /*
930  * mlockall_args(int how)
931  */
932 int
933 sys_mlockall(struct mlockall_args *uap)
934 {
935 	return 0;
936 }
937 
938 /*
939  * munlockall_args(void)
940  */
941 int
942 sys_munlockall(struct munlockall_args *uap)
943 {
944 	return 0;
945 }
946 
947 /*
948  * munlock_args(const void *addr, size_t len)
949  */
950 int
951 sys_munlock(struct munlock_args *uap)
952 {
953 	struct thread *td = curthread;
954 	struct proc *p = td->td_proc;
955 	vm_offset_t addr;
956 	vm_size_t size, pageoff;
957 	int error;
958 
959 	addr = (vm_offset_t) uap->addr;
960 	size = uap->len;
961 
962 	pageoff = (addr & PAGE_MASK);
963 	addr -= pageoff;
964 	size += pageoff;
965 	size = (vm_size_t) round_page(size);
966 
967 	/* disable wrap around */
968 	if (addr + size < addr)
969 		return (EINVAL);
970 
971 #ifndef pmap_wired_count
972 	error = suser(td);
973 	if (error)
974 		return (error);
975 #endif
976 
977 	error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
978 	return (error == KERN_SUCCESS ? 0 : ENOMEM);
979 }
980 
981 /*
982  * Internal version of mmap.
983  * Currently used by mmap, exec, and sys5 shared memory.
984  * Handle is either a vnode pointer or NULL for MAP_ANON.
985  */
986 int
987 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
988 	vm_prot_t maxprot, int flags,
989 	void *handle,
990 	vm_ooffset_t foff)
991 {
992 	boolean_t fitit;
993 	vm_object_t object;
994 	struct vnode *vp = NULL;
995 	struct proc *p;
996 	objtype_t type;
997 	int rv = KERN_SUCCESS;
998 	off_t objsize;
999 	int docow;
1000 
1001 	if (size == 0)
1002 		return (0);
1003 
1004 	objsize = size = round_page(size);
1005 
1006 	/*
1007 	 * XXX messy code, fixme
1008 	 */
1009 	if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1010 		if (map->size + size > p->p_rlimit[RLIMIT_VMEM].rlim_cur)
1011 			return(ENOMEM);
1012 	}
1013 
1014 	/*
1015 	 * We currently can only deal with page aligned file offsets.
1016 	 * The check is here rather than in the syscall because the
1017 	 * kernel calls this function internally for other mmaping
1018 	 * operations (such as in exec) and non-aligned offsets will
1019 	 * cause pmap inconsistencies...so we want to be sure to
1020 	 * disallow this in all cases.
1021 	 */
1022 	if (foff & PAGE_MASK)
1023 		return (EINVAL);
1024 
1025 	if ((flags & MAP_FIXED) == 0) {
1026 		fitit = TRUE;
1027 		*addr = round_page(*addr);
1028 	} else {
1029 		if (*addr != trunc_page(*addr))
1030 			return (EINVAL);
1031 		fitit = FALSE;
1032 		vm_map_remove(map, *addr, *addr + size);
1033 	}
1034 
1035 	/*
1036 	 * Lookup/allocate object.
1037 	 */
1038 	if (flags & MAP_ANON) {
1039 		type = OBJT_DEFAULT;
1040 		/*
1041 		 * Unnamed anonymous regions always start at 0.
1042 		 */
1043 		if (handle == 0)
1044 			foff = 0;
1045 	} else {
1046 		vp = (struct vnode *) handle;
1047 		if (vp->v_type == VCHR) {
1048 			type = OBJT_DEVICE;
1049 			handle = (void *)(intptr_t)vp->v_rdev;
1050 		} else {
1051 			struct vattr vat;
1052 			int error;
1053 
1054 			error = VOP_GETATTR(vp, &vat);
1055 			if (error)
1056 				return (error);
1057 			objsize = vat.va_size;
1058 			type = OBJT_VNODE;
1059 			/*
1060 			 * if it is a regular file without any references
1061 			 * we do not need to sync it.
1062 			 */
1063 			if (vp->v_type == VREG && vat.va_nlink == 0) {
1064 				flags |= MAP_NOSYNC;
1065 			}
1066 		}
1067 	}
1068 
1069 	if (handle == NULL) {
1070 		object = NULL;
1071 		docow = 0;
1072 	} else {
1073 		object = vm_pager_allocate(type, handle, objsize, prot, foff);
1074 		if (object == NULL)
1075 			return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1076 		docow = MAP_PREFAULT_PARTIAL;
1077 	}
1078 
1079 	/*
1080 	 * Force device mappings to be shared.
1081 	 */
1082 	if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1083 		flags &= ~(MAP_PRIVATE|MAP_COPY);
1084 		flags |= MAP_SHARED;
1085 	}
1086 
1087 	if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1088 		docow |= MAP_COPY_ON_WRITE;
1089 	if (flags & MAP_NOSYNC)
1090 		docow |= MAP_DISABLE_SYNCER;
1091 	if (flags & MAP_NOCORE)
1092 		docow |= MAP_DISABLE_COREDUMP;
1093 
1094 #if defined(VM_PROT_READ_IS_EXEC)
1095 	if (prot & VM_PROT_READ)
1096 		prot |= VM_PROT_EXECUTE;
1097 
1098 	if (maxprot & VM_PROT_READ)
1099 		maxprot |= VM_PROT_EXECUTE;
1100 #endif
1101 
1102 	if (fitit) {
1103 		*addr = pmap_addr_hint(object, *addr, size);
1104 	}
1105 
1106 	/*
1107 	 * Stack mappings need special attention.  Mappings that use virtual
1108 	 * page tables will default to storing the page table at offset 0.
1109 	 */
1110 	if (flags & MAP_STACK) {
1111 		rv = vm_map_stack (map, *addr, size, prot, maxprot, docow);
1112 	} else if (flags & MAP_VPAGETABLE) {
1113 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1114 				 VM_MAPTYPE_VPAGETABLE, prot, maxprot, docow);
1115 	} else {
1116 		rv = vm_map_find(map, object, foff, addr, size, fitit,
1117 				 VM_MAPTYPE_NORMAL, prot, maxprot, docow);
1118 	}
1119 
1120 	if (rv != KERN_SUCCESS) {
1121 		/*
1122 		 * Lose the object reference. Will destroy the
1123 		 * object if it's an unnamed anonymous mapping
1124 		 * or named anonymous without other references.
1125 		 */
1126 		vm_object_deallocate(object);
1127 		goto out;
1128 	}
1129 
1130 	/*
1131 	 * Shared memory is also shared with children.
1132 	 */
1133 	if (flags & (MAP_SHARED|MAP_INHERIT)) {
1134 		rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1135 		if (rv != KERN_SUCCESS) {
1136 			vm_map_remove(map, *addr, *addr + size);
1137 			goto out;
1138 		}
1139 	}
1140 out:
1141 	switch (rv) {
1142 	case KERN_SUCCESS:
1143 		return (0);
1144 	case KERN_INVALID_ADDRESS:
1145 	case KERN_NO_SPACE:
1146 		return (ENOMEM);
1147 	case KERN_PROTECTION_FAILURE:
1148 		return (EACCES);
1149 	default:
1150 		return (EINVAL);
1151 	}
1152 }
1153